## ADVANCES IN BIOLOGICAL UNDERSTANDING OF TUMOR RADIATION RESISTANCE

EDITED BY : Ira Ida Skvortsova and Paul N. Span PUBLISHED IN : Frontiers in Oncology

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88966-053-7 DOI 10.3389/978-2-88966-053-7

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

## ADVANCES IN BIOLOGICAL UNDERSTANDING OF TUMOR RADIATION RESISTANCE

Topic Editors: Ira Ida Skvortsova, Innsbruck Medical University, Austria Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands

Citation: Skvortsova, I. I., Span, P. N., eds. (2020). Advances in Biological Understanding of Tumor Radiation Resistance. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-053-7

# Table of Contents

*05 Editorial: Advances in Biological Understanding of Tumor Radiation Resistance*

Ira Ida Skvortsova and Paul N. Span

*07 Combining the DNA Repair Inhibitor Dbait With Radiotherapy for the Treatment of High Grade Glioma: Efficacy and Protein Biomarkers of Resistance in Preclinical Models*

Julian Biau, Emmanuel Chautard, Nathalie Berthault, Leanne de Koning, Frank Court, Bruno Pereira, Pierre Verrelle and Marie Dutreix

*17 Flavonoid Derivative of* Cannabis *Demonstrates Therapeutic Potential in Preclinical Models of Metastatic Pancreatic Cancer*

Michele Moreau, Udoka Ibeh, Kaylie Decosmo, Noella Bih, Sayeda Yasmin-Karim, Ngeh Toyang, Henry Lowe and Wilfred Ngwa

*26 Corrigendum: Flavonoid Derivative of* Cannabis *Demonstrates Therapeutic Potential in Preclinical Models of Metastatic Pancreatic Cancer*

Michele Moreau, Udoka Ibeh, Kaylie Decosmo, Noella Bih, Sayeda Yasmin-Karim, Ngeh Toyang, Henry Lowe and Wilfred Ngwa


Jie Ni, Joseph Bucci, David Malouf, Matthew Knox, Peter Graham and Yong Li


Julian Biau, Emmanuel Chautard, Pierre Verrelle and Marie Dutreix


Katsutoshi Sato, Takashi Shimokawa and Takashi Imai


Gro Elise Rødland, Katrine Melhus, Roman Generalov, Sania Gilani, Francesco Bertoni, Jostein Dahle, Randi G. Syljuåsen and Sebastian Patzke

*149 The Extracellular, Cellular, and Nuclear Stiffness, a Trinity in the Cancer Resistome—A Review*

Sara Sofia Deville and Nils Cordes

*163 Biological Determinants of Chemo-Radiotherapy Response in HPV-Negative Head and Neck Cancer: A Multicentric External Validation*

Martijn van der Heijden, Paul B. M. Essers, Monique C. de Jong, Reinout H. de Roest, Sebastian Sanduleanu, Caroline V. M. Verhagen, Olga Hamming-Vrieze, Frank Hoebers, Philippe Lambin, Harry Bartelink, C. René Leemans, Marcel Verheij, Ruud H. Brakenhoff, Michiel W. M. van den Brekel and Conchita Vens


# Editorial: Advances in Biological Understanding of Tumor Radiation Resistance

Ira Ida Skvortsova1,2 and Paul N. Span<sup>3</sup> \*

*<sup>1</sup> Laboratory for Experimental and Translational Research on Radiation Oncology (EXTRO-Lab), Department of Therapeutic Radiology and Oncology, Innsbruck Medical University, Innsbruck, Austria, <sup>2</sup> Tyrolean Cancer Research Institute, Innsbruck, Austria, <sup>3</sup> Radiotherapy and OncoImmunology Laboratory, Department of Radiation Oncology, Radboud University Medical Center, Nijmegen, Netherlands*

Keywords: radiotherapy, treatment resistance, DNA damage repair, hypoxia, cancer stem cells

**The Editorial on the Research Topic**

#### **Advances in Biological Understanding of Tumor Radiation Resistance**

Radiation therapy (RT) is a frequently-applied powerful treatment approach generally leading to enhanced local tumor control. However, clinical outcome may be compromised by ineffective eradication of cancer cells that exhibit intrinsic or acquired radiation resistance. Although with increased doses of irradiation these radioresistant carcinoma cells can successfully be killed, it is usually impossible to reach these doses without pronounced damage to healthy tissue. To enhance the efficacy of RT, it is important to elucidate the molecular mechanisms regulating radiation resistance of tumor cells. The development of novel chemo- or targeting therapeutics targeting these mechanisms may thereby improve RT efficacy.

A number of mechanisms are known to be involved in cancer cell insensitivity to irradiation. Among them are mutations in genes related to DNA damage and repair, activation of intracellular pro-survival signaling pathways, affected cell cycle regulation, compromised cell death machinery, etc. Furthermore, microenvironmental factors are also very much involved in tumor cell radioresistance. For example, hypoxia, tumor-associated fibroblasts, immune cells, etc. could diminish tumor responses to ionizing radiation. Additionally, cancer stem cells (CSC) encapsulate in a single concept many of the above-mentioned explanations of tumor insensitivity to cytotoxic radiotherapy. Therefore, the role of CSCs in tumor formation, development and response to anti-tumor therapies is at present under intense investigation.

In this Special Issue, several reviews and original research papers address these different mechanisms of radioresistance.

For example, targeting of the repair of RT-induced Double Stranded Breaks in DNA may enhance RT efficacy (Biau, Verrelle et al.), for example via the DNA repair inhibitor Dbait (Biau, Berthault et al.).

Considering the association of intracellular signaling pathways and immunity with radioresistance, (Liu and Sidi) discuss how specific Innate Immune Kinase IRAK1 inhibitors might attenuate tumor radioresistance, while enhancing innate anti-tumor immune responses. Furthermore, Rødland et al. found that the dual-specific CDK1/2 and AURA/B kinase inhibitor JNJ-7706621 inhibits resistance to radioimmunotherapy in Diffuse Large B Cell Lymphoma. A role for the gene Schlafen11 (SLFN11) in determining cancer cell sensitivity to DNA-damaging chemotherapeutic agents and patient outcomes for several cancers has been described. Kaur et al. shows that CD47 is involved in this SLFN11-associated radioresistance.

The tissue microenvironment (TME) influences radiosensitivity. In the multicentric validation study of van der Heijden et al., (acute and chronic) hypoxia, stem cell-ness, tumor growth, epithelial-to-mesenchymal transition (EMT) and DNA repair were found to be related to locoregional control in chemoradiotherapy treated HNSCC patients. Radiosensitivity is also

Edited and reviewed by:

*Timothy James Kinsella, Warren Alpert Medical School of Brown University, United States*

#### \*Correspondence:

*Paul N. Span paul.span@radboudumc.nl*

#### Specialty section:

*This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology*

Received: *16 April 2020* Accepted: *20 April 2020* Published: *12 May 2020*

#### Citation:

*Skvortsova II and Span PN (2020) Editorial: Advances in Biological Understanding of Tumor Radiation Resistance. Front. Oncol. 10:754. doi: 10.3389/fonc.2020.00754* modified by mechanical cues from the TME as reviewed by Deville and Cordes, and cells communicate radioresistance via exosomes as described in Ni et al.. Reciprocally, RT induces extensive changes in the TME that subsequently contribute to radioresistance, as found in glioblastoma (Seo et al.). This latter effect seems mediated via glioblastoma stem cells. This role of CSCs in radioresistance is discussed by Arnold et al.. Herein, the authors indicate that CSC targeting therapy is relevant for the efficacy of RT, but that correct identification of CSCs and reliable distinction from healthy cells is necessary. Furthermore, Terraneo et al. highlight mechanisms of CSC therapy resistance such as EMT and stemness, and describe novel therapeutic strategies for ovarian CSC. Indeed, Neuropilin-2 (NRP2) is associated with radioresistance in bladder cancer, and in Schulz et al. this is reportedly mediated via effects on EMT.

Lindell Jonsson et al. used liquid chromatography-mass spectrometry (LC-MS) to compare the metabolism between 2 HNSCC cell lines with differing radiosensitivity before and after irradiation, and found important differences that may account for their radiosensivity. Notably, Dadgar and Rajaram review different approaches to assess cellular metabolism, such as two-photon microscopy, diffuse reflectance, and Raman spectroscopy, which yield functional and molecular differences between radiation-resistant and sensitive tumors in response to radiation.

Other reviews consider alternative modes of irradiation as a way of alleviating radioresistance. For example, differences in photon vs. particle irradiation exist, as reviewed by Sato et al.. Furthermore, Ultra-High Dose Rate irradiation (FLASH) appears to have differential effects on normal tissues vs. tumors, making -to a certain extent- higher, more effective doses feasible (Wilson et al.). Moreau et al. finds that FBL-03G, a flavonoid cannabis derivative, radiosensitizes metastatic pancreatic cancer. On the other hand, Bettoni et al. report how in rectal cancer, neoadjuvant chemoradiation can increase intratumoral genetic heterogeneity, thereby leading to an increased risk in more aggressive residual tumors.

Overall, this Special Issue has addressed a multitude of potential mechanisms associated with radioresistance, and report on new targets for sensitizing treatment options. Knowledge on how these different mechanisms are induced and interact, how these occur in different cancers, and how these may be countered, will aid in assessing, preventing and/or targeting radioresistance. As radiotherapy is still one of the most effective and widely applied cancer treatment options, countering radioresistance may have far reaching effects on clinical outcome of many cancer patients.

### AUTHOR CONTRIBUTIONS

PS wrote the first draft. IS contributed and finalized the manuscript.

**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Skvortsova and Span. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Combining the DNA Repair Inhibitor Dbait With Radiotherapy for the Treatment of High Grade Glioma: Efficacy and Protein Biomarkers of Resistance in Preclinical Models

Julian Biau1,2,3,4,5,6 \*, Emmanuel Chautard5,7, Nathalie Berthault 1,2,3,4, Leanne de Koning8,9 , Frank Court <sup>10</sup>, Bruno Pereira<sup>11</sup>, Pierre Verrelle1,6,12,13 and Marie Dutreix 1,2,3,4

<sup>1</sup> Centre de Recherche, Institut Curie, PSL Research University, Paris, France, <sup>2</sup> UMR3347, CNRS, Orsay, France, <sup>3</sup> U1021, INSERM, Orsay, France, <sup>4</sup> Research Department, Université Paris Sud, Orsay, France, <sup>5</sup> INSERM, U1240 IMoST, Université Clermont Auvergne, Clermont Ferrand, France, <sup>6</sup> Radiotherapy Department, Centre Jean Perrin, Université Clermont Auvergne, Clermont-Ferrand, France, <sup>7</sup> Pathology Department, Centre Jean Perrin, Université Clermont Auvergne, Clermont-Ferrand, France, <sup>8</sup> Laboratory of Proteomic Mass Spectrometry, Centre de Recherche, Institut Curie, Paris, France, <sup>9</sup> Department of Translational Research, Institut Curie, PSL Research University, Paris, France, <sup>10</sup> GReD Laboratory, CNRS UMR 6293, INSERM U1103, Université Clermont Auvergne, Clermont-Ferrand, France, <sup>11</sup> Biostatistics Department, DRCI, Clermont-Ferrand Hospital, Clermont-Ferrand, France, <sup>12</sup> U1196, INSERM, UMR9187, CNRS, Orsay, France, <sup>13</sup> Radiotherapy Department, Institut Curie Hospital, Paris, France

#### Edited by:

Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands

#### Reviewed by:

Ranjit Bindra, Yale Medicine, United States Dalong Pang, Georgetown University, United States

> \*Correspondence: Julian Biau

Julian.biau@clermont.unicancer.fr

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 27 March 2019 Accepted: 05 June 2019 Published: 19 June 2019

#### Citation:

Biau J, Chautard E, Berthault N, de Koning L, Court F, Pereira B, Verrelle P and Dutreix M (2019) Combining the DNA Repair Inhibitor Dbait With Radiotherapy for the Treatment of High Grade Glioma: Efficacy and Protein Biomarkers of Resistance in Preclinical Models. Front. Oncol. 9:549. doi: 10.3389/fonc.2019.00549 High grade glioma relapses occur often within the irradiated volume mostly due to a high resistance to radiation therapy (RT). Dbait (which stands for DNA strand break bait) molecules mimic DSBs and trap DNA repair proteins, thereby inhibiting repair of DNA damage induced by RT. Here we evaluate the potential of Dbait to sensitize high grade glioma to RT. First, we demonstrated the radiosensitizer properties of Dbait in 6/9 tested cell lines. Then, we performed animal studies using six cell derived xenograft and five patient derived xenograft models, to show the clinical potential and applicability of combined Dbait+RT treatment for human high grade glioma. Using a RPPA approach, we showed that Phospho-H2AX/H2AX and Phospho-NBS1/NBS1 were predictive of Dbait efficacy in xenograft models. Our results provide the preclinical proof of concept that combining RT with Dbait inhibition of DNA repair could be of benefit to patients with high grade glioma.

Keywords: radiation therapy, high grade glioma, Dbait, preclinical study, double-strand break, single-strand break, radioresistance

## INTRODUCTION

High grade gliomas are the most frequent primary brain tumors in adults (1, 2). They represent an important source of morbidity and mortality and are a public health care challenge (3, 4). Maximal possible surgery is generally the first step of the management of high grade gliomas. Radiotherapy (RT) (+/- chemotherapy), is a major adjuvant therapy that improves survival (5, 6). Despite these treatments, median survival remains very low (1, 4). Early recurrence often occurs in the irradiated volume due to a high radioresistance of glioblastoma cells (7–10). These recurrences emphasize the

**7**

need to overcome tumor radioresistance with new molecules that target pathways underlying the mechanisms of such resistance (10–12).

The cytotoxicity of RT is mostly due to DNA damage (13). About 10,000 damaged bases, 1,000 single-strand breaks (SSB) and 40 double-strand breaks (DSB) are produced per gray, and per cell (13, 14). The most severe RT-induced DNA damages are DSB that are lethal to the cell if not repaired (15). The capacity of cancer cells to recognize DNA damages and initiate repair plays a major role in radioresistance (16–18). DNA repair inhibition could make cancer cells particularly sensitive to the DNA damaging treatments like RT (18, 19). Therefore, to inhibit DNA repair, we designed innovative molecules called Dbait (for DNA strand break bait). Dbait are 32 base-pair deoxyribonucleotides forming intramolecular DNA double helix mimicking DNA damages (18, 20–22). They act as a bait for DNA damage signaling enzymes inducing a "false" DNA damage signal that prevents repair enzyme recruitment at damage site and ultimately inhibits DSB and SSB repair pathways (**Figure 1**) (18, 20–22). Dbait was tested in combination with RT in first-inhuman phase 1 clinical trial for the treatment of skin metastases of melanoma with encouraging results (23).

To decipher the mechanisms sustaining resistance to anticancer treatments is one of the most current challenges to avoid treatment escape. High-throughput screening strategies are widely used for the identification of predictive and prognostic biomarkers (24, 25). The most currently used analyzed the RNA content (transcriptome) or DNA modification (genome). However, in mammalian, it is widely accepted that regulatory modifications occur at the protein levels (25–27). Therefore, to explore in vivo predictive biomarkers of RT efficacy we used reverse-phase protein array (RPPA), a technology using high-throughput antibody-based detection. It requires just a few micrograms of protein lysate and allows measuring protein expression and their main modification in a highly quantitative manner (25, 27, 28). Hundreds of samples can be analyzed simultaneously and thus generate large datasets to identify potential biomarkers (25, 29).

In this preclinical study, we analyzed the potential of Dbait to sensitize high grade glioma to RT. First, we demonstrated the radiosensitizer properties of Dbait in vitro. Secondly, animal studies were performed to test the clinical potential of the combination of Dbait and RT for the treatment of high grade glioma. We identified potential protein biomarkers of resistance using RPPA. For that purpose, we assayed a selection of proteins and modifications involved in different RT signaling pathways.

#### MATERIALS AND METHODS

#### Cell Culture and Dbait Molecules

Nine human high grade glioma cell lines were used (CB193, MO59J, MO59K, SF763, SF767, SNB19, T98G, U87MG, and U118MG) and were grown using a 10% Fetal Calf Serum DMEM medium in a humidified incubator containing 5% CO2 at 37◦C as previously described (25).

As already described (18): Dbait molecules consist in 32 base-pair oligonucleotides (5'- GCTGTGCCCACAACCCAGCAAACAAGCCTAGA-(H)- TCTAGGCTTGTTTGCTGG GTTGTGGGCACAGC-3', Eurogentec, Seraing, Belgium). A short inactive molecule, Dbait-8H (5'-ACGCACGG-(H)-CCGTGCGT-3') was used as control in the in vitro experiments. H is a hexaethyleneglycol linker and the letters underlined indicate the phosphorodiamidate nucleosides.

#### In vitro Dbait and Irradiation Treatments and Cell Survival Assay

Dbait (1.25 mg.L−<sup>1</sup> ) or transfection control, complexed with 11 kDa polyethylenimine (PEI) (Polypus-transfection, Illkirch, France) were used to treat the cells as previously published (18, 21). Cells were incubated during 5 h for transfection in serum-free RPMI medium (in twenty-four-well plates). After transfection, the medium was removed and replaced with complete DMEM (Gibco, Cergy Pontoise, France) (18). Cells were then subjected to 2.5-Gy irradiation, using a <sup>137</sup>Cs unit (0.5 Gy/min). Nine days later, cell fixation (paraformaldehyde 4%) and permeabilization (Triton X100 0.5%) were done, and the number of nuclei was estimated following staining with TO-PRO3 for 10 min. Nuclear staining signals were determined by imaging with an infrared scanner (LI-COR Odyssey).

#### Western Blot

Cells were harvested and boiled 10 min in Laemmli buffer and subjected to SDSPAGE. Proteins were transferred to nitrocellulose membranes, blocked by incubation (1 h) with Odyssey buffer (LI-COR Biosciences, Lincoln, NE, USA). Membranes were incubated overnight at 4◦C with primary antibody diluted in Odyssey buffer. Depending on primary antibodies, the membranes were then probed with goat secondary antibodies (anti-mouse or anti-rabbit) conjugated to Alexa Fluor 680 (Invitrogen) or IRdye 800 (Rockland Immunochemicals, Gilbertsville, PA, USA). Anti-γ-H2AX (Upstate, Millipore, Molsheim, France) and anti-β-actin clone AC-15 mouse monoclonal antibodies (Sigma-Aldrich, Saint-Quentin-Fallavier, France) were used. The obtained signals were analyzed with the Odyssey Infrared Imaging System (LI-COR Biosciences) and Odyssey software.

#### Dbait and Irradiation Treatments in Mice

Xenografts derived from cell lines (CDX) and patient derived xenograft (PDX; ODA-17GIR, GBM-1-HAM, GBM-14-RAV, GBM-14-CHA, ODA-4-GEN) were, respectively obtained by injecting 4 × 10<sup>6</sup> cells of each cell line into the flank, and by successive grafting into scapular area of adult female nude mice (Swiss nu/nu, 6–8 weeks, Janvier, Le Genest Saint Isle, France) (10). Small fragments of PDX tumors were grafted subcutaneously into the flank of nude mice before experiments. When the tumor volume were around 125 mm<sup>3</sup> , mice were divided into uniform groups (n = 6 to 12) (18): no treatment (NT), RT alone for 2 weeks (RT2w: 6x5Gy), Dbait alone for 2 weeks (Dbait: 6x3nmol) and RT + Dbait for 2 weeks (RT2w+Dbait 6x5Gy+6x3nmol). We had previously checked

that mock treated animals did not show any change in tumor growth or survival as compared to animals only treated with or without RT (21). In the same way as beforehand (18), Dbait molecules with in vivo-jet polyethylenimine (PEI) reagent (Polyplus Transfection) at the N/P ratio 6 were diluted in 100 µL of 5% glucose. Dbait was combined with PEI to facilitate cellular delivery (21). Prior to injection, the Dbait-PEI mixture was incubated for 15 min at room temperature. Dbait intratumoral injections were realized 5 h before each RT session. To deliver RT by a <sup>137</sup>Cs unit (0.5 Gy/min), a shield was conceived to spare about two-thirds of the animal's body. Doses were measured by thermoluminescence dosimetry. Tumors were monitored for all experiments with a digital caliper every 2–3 days. The formula (length × width × width/2) was used to calculate the tumor volumes. Mice weight was determined every week and followed up for 200 days. When tumors attained 2000 mm3, animals were sacrificed according to ethical recommendations. All animals were housed in our animal facility, and all experiments were approved by the Local Committee on Ethics of Animal Experimentation.

### Immunofluorescence Staining and Dog MRI

The MRI of a boxer dog having spontaneously developed a brain tumor was performed at the Veterinary School of Maisont-Alfort (94-France) by Dr. P. Devauchelle and tumor samples were obtained with the consent of the dog owner. For immunofluorescence staining, cells were processed as previously described (20, 30). Microscopy was performed at room temperature with the Leica SP5 confocal system, attached to a DMI6000 stand, with a 636/1.4 oil immersion objective. Images were processed with the freely available ImageJ software (http:// rsb.info.nih.gov.gate1.inist.fr/ij/) and the Leica SP5 confocal system.

### Antibodies and Validation for RPPA

We explored 39 total proteins, 26 phosphoproteins and then calculated 23 ratios of phosphoproteins on total proteins giving a total of 88 protein biomarkers (**Table S1**) involved in 10 different signaling pathways: tyrosine kinase signaling, SAPK/JNK signaling, stress signaling, DNA repair, PI3K pathway, apoptosis, cell cycle, adhesion/cytoskeleton, MAPK/ERK signaling and NFκB signaling. As reported earlier, before being used in RPPA, the antibodies quality and specificity were confirmed by Western blotting on a large panel of cell lines (25).

## Reverse Phase Protein Array (RPPA)

Proteins from 11 subcutaneous xenograft models were analyzed (6 replicates with 2 different locations in three different tumors per model). Tumors were mechanically dissociated (10) and protein concentration was determined using the Reducing Agent Compatible BCA kit (Pierce, Rockford, USA). The samples were then processed using previously reported method (10). Briefly, serial dilutions of samples (from 2 to 0.125 mg/ml) were placed on nitrocellulose-covered slides (2470 Arrayer, Aushon Biosystems, Billerica, MA) before incubation overnight at 4 ◦C with specific antibodies. Slides were then probed with horseradish peroxidase-coupled secondary antibodies (Jackson ImmunoResearch, Newmarket, UK) for 1 h at room temperature. After an amplification step, the arrays were probed with Cy5-streptavidin (Jackson ImmunoResearch) for 1 h at room temperature. Finally, the processed slides were scanned with a GenePix 4000B microarray scanner (Molecular Devices, Sunnyvale, CA) and Spot intensity was evaluated with MicroVigene 4.0.0.0 software (VigeneTech Inc., Carlisle, MA). Quantification of the data was done with SuperCurve (31), and the data were normalized against negative control slides and Sypro Ruby slides.

#### Statistical Analysis

Data analysis was realized with R v2.15.1 (http://www.cran.rproject.org). The tests were two-sided, with a Type I error set at α = 0.05. To explore variations between groups, Mann-Whitney tests were done according to sample size, and if assumptions of parametric test are not met (normality and homoscedasticity). The Kaplan–Meier method was used to draw the survival curves. The log-rank test was used to compare the survival fraction of groups (NT: not treated, RT or RT+Dbait). P ≤ 0.05 was considered to be a significant difference.

#### RESULTS

#### Dbait Disorganizes Repair of Radio-Induced DNA Damage in High Grade Glioma Cell Lines and Leads Proliferation Inhibition

In a previous study, we have shown that Dbait lead to activation of the DNA-dependent protein kinase (DNA-PK) (20). This hyperactivation triggers phosphorylation of H2AX and other markers such as RPA32, CHK2 and HSP90 (20, 30), prevents detection of the radio-induced DSBs and further recruitment of DNA repair enzymes at damage site (**Figure 1**). First, we tested the potential of Dbait to induce DNA-PK activation in human glioblastoma cell lines by assaying phosphorylated H2AX proteins by Western blot in the 9 high grade glioma cell lines. Western blot analysis showed that in all the glioma cell lines except in the DNA-PK deficient MO59J cell line, Dbait induced phosphorylation of H2AX (**Figure 2A**). As already published (20) Dbait induced phosphorylation of H2AX is strictly dependent of DNA-PKcs kinase activity. In contrast, irradiation induced γ-H2AX foci that are mainly due to ATM activation, in all cell lines including MO59J. Combining Dbait and irradiation induced equal to superior level of γ-H2AX. The level of H2AX was not significantly affected by the various treatments (**Supplementary Figure 1**). As we had access to samples from a dog that spontaneously developed a glioblastoma (**Figure 2B**), we confirmed that Dbait induced phosphorylation of both H2AX and HSP90 in dissociated cells from the brain tumor (**Figure 2C**). As previously observed, γ-H2AX formed foci after irradiation, at location of radio-induced DNA DSB in irradiated cells whereas it distributed all over the chromatin after Dbait treatment (in at least 65% of the cells), showing DNA-PK activation in absence of chromosome damage (20).

The consequences of DNA-PK hyperactivation for cell survival after irradiation were investigated. Nine high-grade glioma cell lines were treated with Dbait or control (Dbait-8h) 5 h before RT to allow DNA-PK activation before inducing damage (**Figure 3**). As we have previously reported (18), without RT, Dbait treatment itself was able to decrease cell survival. For 6/9 cell lines (MO59K, SF763, SNB19, U118MG, U87MG, and T98G), the combination of Dbait and RT led to a significant radiosensitization (p < 0.05). SF763 was sensitized only at the highest dose of Dbait. For 3/9 cell lines (CB193, MO59J and SF767), radiosensitization was not statistically significant. γ-H2AX increase after Dbait treatment was observed in SF767 (**Figure 2**) eliminating the possibility that the lack of sensitization could be due to defect in transfection or incapacity to activate DNA-PKcs as it is the case of MO59J cells.

### Radiosensitizing Effect of Intratumoral Injections of Dbait in vivo

We have recently shown that protein status are well conserved between cell lines and tumors formed by xenografting of these cell lines (25). However, micro-environment plays an important role in tumor cell response to treatment and could modify therapy response. Therefore, we reproduced our survival analysis in vivo using athymic nude mice bearing glioma CDX obtained by grafting the cell lines characterized in vitro. Among the 9 cell lines tested in vitro, only six models form tumors with enough efficacy and homogeneity to allow in vivo treatment efficiency study. Consistent with one of the currently used stereotactic RT protocols for the reirradiation of high grade glioma (32), 6 fractions of 5 Gy were given locally over a 2 weeks period. Dbait was administered locally 5 h prior to RT, every other day (for a total of 6 sessions; **Figure 4A**). The combined treatment (RT2w+Dbait) significantly decreased tumor growth and enhanced survival of 3/6 models (**Figure 4B**). The survival enhancement by addition of Dbait to radiotherapy in U118MG, SF763 and T98G was, respectively of 129, 136 and 234%. SF763 which was sensitized only at the highest dose of Dbait in vitro appeared to be sensitive to Dbait addition to radiotherapy in vivo. The CB193 and SF767 models were not radiosensitized consistently with in vitro results. Dbait effect did not depend upon the tumor growth speed. While U87MG cells were radiosensitized in vitro (**Figure 3**), addition of Dbait to radiotherapy had no impact on survival of U87-MG in vivo models.

In order to confirm that the Dbait effect is not a specificity of CDX models we performed in parallel a similar analysis on five PDX directly derived from patient samples (**Figure 4C**). Three were radiosensitized by Dbait with increase in survival compared to RT alone of 125% for GBM-14-RAV, 128% for GBM-14-CHA and 188% for ODA-4-GEN. The two other models (ODA-17-GIR and GBM-1-HAM) were not radiosensitized by such a treatment. Interestingly, in all the treated animal models, no significant skin toxicity was observed in irradiated and Dbait-treated healthy tissue. Depending on in vivo model, tumor growth after Dbait

FIGURE 2 | gH2AX induction in gliomas cells after Dbait treatment. (A) gH2AX induction in human gliomas cell lines after Dbait treatment. The nine glioma cell lines were untreated (NT), irradiated (IR, 10Gy), treated with Dbait (5h) or treated with Dbait and irradiated. One hour after treatment completion, total proteins were electrophoresed followed by immunobloting. The blots were analyzed using the Odyssey Infrared Imaging System (LI-COR Biosciences) and Odyssey software. The induction of gH2AX (gH2AX on H2AX ratio) is presented. Mann-Whitney test was performed (\*p < 0.05). (B) MRI of a boxer dog with a spontaneously brain tumor. (C) Activation of DNA damage response in dog glioblastoma. Dissociated cells of dog glioblastoma were untreated (NT), irradiated or treated with Dbait (5 h). Cells were fixed and permeabilized after treatment before the use of anti- gH2AX, anti-HSP90 antibodies and DAPI.

performed (\*p < 0.05).

treatment alone, was at the best very similar to those observed following RT alone, making the combination a better option in most of the cases.

#### Predictive Biomarkers of Dbait Efficacy

We then used a RPPA approach to identify protein biomarkers predictive of Dbait response of the 6 CDX and 5 PDX to

RT+Dbait. Eighty-eight protein markers were analyzed: 39 total proteins, 26 phosphoproteins and 23 ratios of phosphoproteins/total proteins were analyzed. A Mann-Whitney test was performed between radiosensitized and not radiosensitized xenografts. We identified 2/88 protein biomarkers predictive of Dbait efficacy: the two most significant biomarkers were the ratio of phosphorylated forms on native forms of the two repair proteins NBS1 and H2AX. Actually Phospho-H2AX/H2AX (p = 0.05, fold change = 2.2) and Phospho-NBS1/NBS1 (p <0.01, fold change = 1.6) were significantly higher in the xenografts that were not radiosensitized (**Figure 5**). Interestingly, Phospho-H2AX was not sufficient (p = 0.66) to predict sensitivity to Dbait radiosensitizing effect. The total amount of H2AX, which has been shown to vary extensively between cell lines (**Figure 1**) and tumors was also not predictive of the Dbait radiosensitization (p = 0.66) however their ratio became highly indicative (p = 0.05).

#### DISCUSSION

The resistance of cancer cells to RT is increased by efficient DNA repair activity (18, 33, 34). In the past years, many DNA repair inhibitors have been developed (18, 35–39). These strategies are mainly based on specific target inhibition. They may be overpassed by target mutation or activation of another repair pathway. On the other hand, Dbait is not a specific enzyme inhibitor. It represents a new drug strategy targeting the whole DNA DSB repair system via perturbation of DNA repair signaling (18, 20, 21). On the one side, the DNA DSB signaling system induced by Dbait is dispersed all over the chromatin and inhibits the recruitment of the DSB repair proteins at damage site. On the other side, Dbait molecules can also be recognized by PARP [major protein involved in base excision repair (BER) and single strand break repair (SSBR)]. This leads to its autoPARylation which allows the recruitment of various BER and SSBR proteins on Dbait molecules inducing BER/SSBR inhibition (18, 40).

In the present study, Dbait molecules were used to radiosensitize human high grade glioma. The experimental design was planned to assay the clinical relevance of a current hypofractionated stereotactic RT protocol used for high grade glioma reirradiation (32) and local administration of Dbait. Hypofractionated stereotactic RT is particularly interesting due to its ability to precisely deliver high doses of RT to a specific target volume in a low numbers of fractions and to spare surrounding organs at risk. Hypofractionated stereotactic RT appeared to be associated with acceptable toxicity if certain limits were observed in terms of treated volume and radiation dose

(41–45). In these series, median survival was low (about 7–13 months from time of salvage treatment) suggesting a therapeutic effect in selected patients. Despite modern developments in spatial targeting, long term control of diseases is not achieved, emphasizing the need to overcome tumor radioresistance with innovative agents (10). Combined Dbait and hypofractionated stereotactic RT treatment is addressing this major challenge, and is particularly attractive to treat recurrent high grade glioma as it provides a double targeting through molecular pathway by Dbait and highly focalized ionizing radiation beam by hypofractionated stereotactic RT. This should achieve a better local control which is the main clinical objective for high grade glioma.

In this study, we used local administration of Dbait, which might limit clinical transfer in this specific indication. However, local administration to high grade glioma of different molecules has already been studied. For example, Gliadel wafer containing carmustine (BCNU) as an interstitial chemotherapy treatment is already approved for malignant glioma (46). Other modalities of local delivery such as convection-enhanced delivery have also shown preclinical and clinical promising results (22, 47, 48). Dbait distribution to the brain has already been evaluated in an RG2 rat glioma model and showed promising results (22).

One of the drawbacks of our preclinical study is that we chose to use flank xenografts rather than intracranial orthotopic xenografts (49–51). We preferred flank xenografts in this preclinical study for different reasons: the number of models and conditions tested (over 300 mice); the use of a 137Cs unit (0.5 Gy/min) which did not allow focal cerebral irradiation (necessary for 6x5Gy delivery); the need of repeated Dbait intratumoral injections; and difficulties in rigorous tumor monitoring with orthotopic xenografts by repeated imaging with high number of animals. Despite the above-cited advantages of flank xenograft models, the drawbacks include: a different microenvironment as it would be within the brain; and a lack of blood brain barrier that can alter the pharmaceutical kinetics (49–51). If the lack of blood brain barrier is not a major issue in our setting as we studied direct intratumoral injections of Dbait, the different microenvironment might significantly influence the results (52). In the past years, most documented resistance mechanisms involve secondary pathway mutations or bypass mechanisms within the tumor cells. However, the recent identification of mechanisms of therapeutic resistance that were conferred largely by alterations, not only in the tumor cells, but also in their microenvironment, indicates the importance of taking into account the tumor cell extrinsic compartments. The nature of the vasculature, the presence of cancer associated fibroblasts, the presence/absence of immune cells, the signaling network between tumor cells and stromal cells are the most studied components that could influence treatment response (52).

As previously shown, Dbait administration to mice did not increase the sensitivity of healthy tissue around the tumor to RT (18). In a previous study, we showed that Dbait does not induce cell cycle arrest (18, 20). Hence the specificity of action of Dbait in tumor cells could be due to an impairment in cell cycle checkpoints that is frequently reported in tumors. Tumors cells would be able to divide despite Dbait induced unrepaired breaks and therefore enter mitotic catastrophe. p53 mutations are often associated with this deficient cell cycle controls (18, 53). At the contrary, non-tumor cells with proficient cell cycle control stop dividing until repair is completed, that can take place when Dbait molecules have disappeared (18, 54). Therefore, Dbait, which does not make new lesions on chromosomes but prevents DNA repair of RT induced damage, is toxic for dividing tumor cells but not for healthy tissues. Dbait toxicology studies were realized in wistar rats and cynomolgus monkeys. They showed that the only side effect was a slight to moderate, dose-dependent and reversible inflammatory response at injection sites (18, 55). The tolerance of the clinical form of Dbait (called AsiDNA) in association with RT has been tested in first-in-man phase 1 trial (DRIIM) for patients with in-transit metastases of melanoma (23). No dose-limiting toxicity was observed and the maximum-tolerated dose was not reached.

Oncology has entered an era of personalized medicine in which the selection of treatments for each cancer patient becomes more individualized (56). Identifying predictive biomarkers of treatment sensitivity or resistance is becoming a major challenge. In this study, we have chosen the RPPA technology to search for potential protein biomarkers of Dbait resistance. This technology presents many advantages: it requires only a few micrograms of protein lysate to study activation of cell signaling pathways and allows the comparison of hundreds of samples in the same experiment (10). Thus, we included replicate samples for the cell lines, and different tumor regions of multiple mice for the xenografts. We were able to obtain robust data and assess heterogeneity within and among tumors (10). Here we showed that basal Phospho-H2AX/H2AX and Phospho-NBS1/NBS1 activations [two major actors of DNA damage signaling and cell cycle control (57, 58)] were significantly correlated with Dbait resistance. Interestingly, whereas the amount of phosphorylated H2AX and NBS1 was moderately indicative of sensitivity to Dbait, the frequency of phosphorylated molecules became highly indicative suggesting that resistance is linked to the intensity of chromatin modification. These constitutive activations may reflect that tumor cells are used to survive despite a basal disturbance of DNA damage signaling and cell cycle control and thus are resistant to Dbait. We have previously reported that Dbait molecules disorganize the downstream DNA damage response notably through H2AX and NBS1 disruption (20). In models with a constitutive high level of NBS1 and H2AX activations, low-dose Dbait failed to enhance these disturbances once more and higher concentrations were required. If we compare the results obtained from Dbaitresistant CDX (CB193, SF767 and U87MG) with the ones of the corresponding cell lines, we can note that CB193 and SF797 were also Dbait-resistant in vitro while U87MG was Dbait-sensitive. Interestingly, concerning CB193 and SF767 cell lines, they were also the two cell lines harboring the highest level of Phospho-NBS1/NBS1 activation and among the highest level of Phospho-H2AX/H2AX activation. Concerning U87MG, the difference in Dbait response could, at least partly, be explained by the differences in the proteic profiles that can exist between a cell line and its corresponding xenograft. We have previously showed (25) that U87MG had a relatively important difference between in vitro and in vivo proteic profiles with 37/89 (42%) proteins differentially expressed with a fold change >1.5.

In this study, we provide the preclinical proof of concept that a combination of RT with Dbait (an inhibitor of DNA repair) could be of interest in the treatment of high grade glioma. A first-in-human phase I trial has evaluated the therapeutic potential of local Dbait injections in combination with RT to treat patients with in-transit metastases of melanoma and provided encouraging results (18, 23). The preclinical data we report suggest that a clinical trial combining HSRT and Dbait could be considered in the treatment of recurrence high grade glioma.

#### REFERENCES


### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

### ETHICS STATEMENT

The Local Committee of Institut Curie on Ethics of Animal Experimentation approved all experiments.

### AUTHOR CONTRIBUTIONS

JB and EC designed the study, collected, and analyzed the data. They also wrote the manuscript. MD designed the study and carefully revised the manuscript. NB, LdK, and FC have contributed to all experiments. They also gave important intellectual input and carefully revised the manuscript. BP have done statistical analysis and carefully revised the manuscript. PV supervised the study and contributed to data interpretation. All authors approved the final manuscript for submission.

### FUNDING

This work benefited from the technical support of the Institut Curie animal facility, microscopy, experimental irradiation RADEXp and RPPA platforms. This study was supported by the Institut Curie, the Centre National de la Recherche Scientifique, and the Agence Nationale de la Recherche (ANR -08-BiotecS-009).

### ACKNOWLEDGMENTS

The PDX models were provided by Dr. Marie-France Poupon and the laboratory of preclinical investigation (LIP) of Institut Curie. The dog model was provided by Patrick Devauchelle (MICEN, Creteil). We thank Marie-Christine Lienafa and Mano Sayarath (DNA Therapeutics) for technical assistance. The Institut Curie animal facility, microscopy, experimental irradiation RADEXp and RPPA platforms were very convenient to achieve this study. The Institut Curie, the Centre National de la Recherche Scientifique, and the Agence Nationale de la Recherche (ANR−08-BiotecS-009) sustained this work.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00549/full#supplementary-material

tumors of the central nervous system: a summary. Acta Neuropathol. (2016) 131:803–20. doi: 10.1007/s00401-016-1545-1


delivery by biodegradable polymers of chemotherapy for recurrent gliomas. The Polymer-brain Tumor Treatment Group. Lancet. (1995) 345:1008–12.


double-stranded DNA oligonucleotide conjugated to cholesterol. Mol Ther Nucleic Acids. (2012) 1:e33. doi: 10.1038/mtna.2012.27


**Conflict of Interest Statement:** MD is cofounder of DNA Therapeutics, Onxeo.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Biau, Chautard, Berthault, de Koning, Court, Pereira, Verrelle and Dutreix. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Flavonoid Derivative of Cannabis Demonstrates Therapeutic Potential in Preclinical Models of Metastatic Pancreatic Cancer

Michele Moreau1,2, Udoka Ibeh1,3, Kaylie Decosmo1,4, Noella Bih<sup>1</sup> , Sayeda Yasmin-Karim<sup>1</sup> , Ngeh Toyang<sup>5</sup> , Henry Lowe<sup>5</sup> and Wilfred Ngwa1,2 \*

*<sup>1</sup> Brigham and Women's Hospital, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, United States, <sup>2</sup> Department of Physics, University of Massachusetts Lowell, Lowell, MA, United States, <sup>3</sup> Department of Biology, University of Massachusetts Boston, Boston, MA, United States, <sup>4</sup> Department of CaNCURE Program, Northeastern University, Boston, MA, United States, <sup>5</sup> Flavocure Biotech Inc., Baltimore, MD, United States*

#### Edited by:

*Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands*

#### Reviewed by:

*James William Jacobberger, Case Western Reserve University, United States Hyuk-Jin Cha, Seoul National University, South Korea*

> \*Correspondence: *Wilfred Ngwa wngwa@lroc.harvard.edu*

#### Specialty section:

*This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology*

Received: *28 March 2019* Accepted: *05 July 2019* Published: *23 July 2019*

#### Citation:

*Moreau M, Ibeh U, Decosmo K, Bih N, Yasmin-Karim S, Toyang N, Lowe H and Ngwa W (2019) Flavonoid Derivative of Cannabis Demonstrates Therapeutic Potential in Preclinical Models of Metastatic Pancreatic Cancer. Front. Oncol. 9:660. doi: 10.3389/fonc.2019.00660* Pancreatic cancer is particularly refractory to modern therapies, with a 5-year survival rate for patients at a dismal 8%. One of the significant barriers to effective treatment is the immunosuppressive pancreatic tumor microenvironment and development of resistance to treatment. New treatment options to increase both the survival and quality of life of patients are urgently needed. This study reports on a new non-cannabinoid, non-psychoactive derivative of cannabis, termed FBL-03G, with the potential to treat pancreatic cancer. *In vitro* results show major increase in apoptosis and consequential decrease in survival for two pancreatic cancer models- Panc-02 and KPC pancreatic cancer cells treated with varying concentrations of FBL-03G and radiotherapy. Meanwhile, *in vivo* results demonstrate therapeutic efficacy in delaying both local and metastatic tumor progression in animal models with pancreatic cancer when using FBL-03G sustainably delivered from smart radiotherapy biomaterials. Repeated experiments also showed significant (*P* < 0.0001) increase in survival for animals with pancreatic cancer compared to control cohorts. The findings demonstrate the potential for this new cannabis derivative in the treatment of both localized and advanced pancreatic cancer, providing impetus for further studies toward clinical translation.

Keywords: pancreatic cancer, flavonoids, cannabis, metastasis, radiotherapy, smart biomaterials

## INTRODUCTION

Pancreatic ductal adenocarcinoma is an antagonistic internecine ailment of the exocrine pancreas with < 8% of patients surviving within a 5-year period (1, 2). A major challenge shared by pancreatic cancers is its aggressiveness, which often metastasizes to other organs before the patient is even diagnosed (3, 4).

Current treatment options for pancreatic cancer include: surgery, chemotherapy, targeted therapy, immunotherapy, and radiation therapy. Curative treatment is available only if the tumor is found early and can be removed by surgery before metastasis. If the cancer has metastasized, the standard of care is chemotherapy, or radiotherapy. However, pancreatic cancer is notoriously defiant to current therapies including chemotherapy, radiotherapy and immunotherapy (1, 5).

Cannabinoids, which are the bioactive components of Cannabis sativa and their derivatives, have been investigated as both anti-cancer agents and for managing the side effects of conventional cancer treatments like chemotherapy and radiotherapy (6). Previous studies have indicated that medical cannabis derivatives could enhance survival in pancreatic cancer animal models, when used in synergy with radiotherapy (7). Smart materials for drug delivery like the smart radiotherapy biomaterials (SRBs) system have also been investigated for delivering cannabinoids into tumors to enhance radiotherapy treatment for pancreatic cancer (8). A major benefit of the SRB approach is their ability to be employed in place of currently used inert radiotherapy biomaterials (e.g., spacers, or fiducial markers) and hence their use could come at no additional inconvenience to patients.

In this study, we investigate a new non-cannabinoid, nonpsychoactive derivative of cannabis, called FBL-03G, to assess its potential for the treatment of pancreatic cancer. We hypothesize that the use of FBL-03G will have therapeutic potential and can enhance radiotherapy during the treatment of pancreatic cancer. To investigate this hypothesis, in vitro studies were first carried out with and without radiotherapy (RT). In vitro studies, in vivo studies were also conducted in small animals employing FBL-03G sustainably delivered from smart radiotherapy biomaterials, allowing continual exposure of the tumor to the cannabis derivative payloads over time.

Apart from the antineoplastic properties of cannabis derivatives, the immune system modulative properties of these extracts have been well documented (8–12). Different volumes and concentrations of FBL-03G payloads were also investigated for their potential to generate systemic tumor responses. In particular, we investigated the abscopal effect, whereby radiotherapy (RT) at one site may lead to regression of metastatic cancer at distant sites that are not irradiated (13). The abscopal effect has been connected to mechanisms involving the immune system (14). However, the abscopal effect is rare because at the time of treatment, established immune-tolerance mechanisms may hamper the development of sufficiently robust abscopal responses. Today, the growing consensus is that combining radiotherapy with immunoadjuvants provides an opportunity to boost abscopal response rates, extending the use of radiotherapy to treatment of both local and metastatic disease (15). With in this context, the cannabis derivative FBL-03G was also investigated as a potential immunoadjuvant to radiotherapy.

#### MATERIALS AND METHODS

#### Materials and Antibody

Acetone, Dimethyl sulfoxide (DMSO), Poly (lactic-co-glycolic) acid (PLGA) (M.W.: 50–50 kDa), and Crystal Violet dye were acquired from Sigma-Aldrich. The Harvard apparatus was obtained from Harvard Bioscience (Holliston, MA, USA), and silicone tubing (ID 1/32′′) was purchased from Saint-Gobain Performance Plastics Laboratory Division (USA). Brachytherapy needles were purchased from IZI Medical Products (MD, USA). All cell culture products (DMEM, RPMI, Trypsin, Fetal Bovine Serum, MEM non-essential amino acids, sodium pyruvate, β-mercaptoethanol, penicillin/streptomycin, and PBS pH 7.4) were obtained from Gibco, Thermo Fisher, and Life Technologies (Waltham, MA, USA). Flavocure Biotech Inc. (Baltimore, MD, USA) supplied the test molecule, FBL-03G with a purity of 98.7% determined by High Performance Liquid Chromatography (HPLC).

### FBL-03G Synthesis

FBL-03G, a flavonoid derived from Cannabis sativa L., is the unnatural isomer of Cannflavin B, a metabolite of Cannabis. Through a bioactivity guided isolation approach, 11 flavonoids were isolated using flash chromatography and characterized by nuclear magnetic resonance (NMR) and mass spectrometry (MS) methods (2, 16, 17). Generated spectroscopic data for FBL-03G were similar to those of the following 11 previously isolated and characterized compounds of the Cannabis plant; apigenin (1), Chrysoeriol (2), kaempferol (3), luteolin (4), quercetin (5), vitexin (6), isovitexin (7), orientin (8) and prenylated flavonoids including Cannflavin A (9), Cannflavin B (10) and Cannflavin C (11) (**Figure S1**) (16–21). The molecules were further screened for kinase inhibition, and chrysoeriol (Cresorol) demonstrated significant activity against FLT3, FLT3-ITD, and FLT3-D835Y and moderate activity against CSF1R (2). FBL-03G demonstrated significant activity against CSF1R kinase and moderate activity against FLT3, FLT3-ITD, FLT3-D835Y, CK2a, CK2a2, Aurora A, Aurora B, Aurora C, and Pim2 (2).

### Fabrication of Smart Radiotherapy Biomaterials (SRB)

This study used combination treatment of FBL-03G (Mw = 368.38 g/mol) as an immunoadjuvant, delivered from smart radiotherapy biomaterial (SRB), and radiotherapy (RT). SRBs were developed following previously reported procedures for fabricating and loading drugs into SRBs. Briefly, 300 mg of Poly (lactic-co-glycolic) acid (PLGA) polymer added to 3.5 mL of acetone was homogenously mixed into a hydrogel (8, 22). The Harvard apparatus was used to provide a constant flow rate of the mixture into the silicon tubing with inner diameter of 1/32′ . The PLGA hydrogel loaded in silicon tubing was allowed to cure under 50◦C for 48 h. After curing, the silicon tubing was cut to 5 mm length and the SRBs were extracted. Three different concentrations of FBL-03G, respectively, were used as payloads in the SRBs. A small animal radiation research platform (SARRP, Xtrahl, Inc., Suwanee, GA, USA) was used for radiotherapy using 220 kVp, 13 mA, (10 × 10) mm nozzle, and 0.15 mm copper (Cu) filter to deliver 6 gray (Gy) dose. In addition, computed tomography (CT) images of the mice were taken using at 65-kVp and 0.8-mA. Mice were anesthetized with isoflurane and imageguided radiotherapy was used to specifically irradiate tumors on one site as needed.

### Cell Culture

Pancreatic cancer cell line, Panc-02, was obtained from the National Cancer Institute and cultured with Dulbecco's Modified Eagle's Medium (DMEM) with 10% FBS and 1% penicillin/streptomycin. Another pancreatic cancer cell line, Ptf1/p48-Cre (KPC) cells were a gift from Dr. Anirban Maitra (MD Anderson Cancer center). KPC cell line was derived from an LSL-Kras; p53+/floxed, Pdx-cre mouse. KPC cells were cultured in RPMI media supplemented with 10% FBS, 2 mmol/L Lglutamine, 1% penicillin/streptomycin, 1% MEM non-essential amino acids, 1 mmol/L sodium pyruvate, and 0.1 mmol/L βmercaptoethanol. All cells were cultured at 37◦C in a humidified incubator with 5% CO2.

#### Clonogenic Survival Assay

Actively growing monolayers of KPC and Panc-02 cancer cells were trypsinized and 300 cells per well were seeded in 6-well plates. 24 hours later, seeded cells were treated with 0, 1, 2, or 4µM of FBL-03G concentrations per well. The cells were irradiated at 0, 2, or 4 Gy using 220 kVp energy, 13 mA, 24 h after the FBL-03G treatment. A small animal radiation research platform (SARRP) was used to deliver external beam radiation. The growing colonies (≥ 50 cells/colony) were fixed with 75% ethanol and stained with 1% crystal violet 9–12 days after treatment. Colonies were counted using ImageJ software and a percent survival was calculated following standard protocol.

#### Mice and Generation of Syngeneic Pancreatic Cancer Models

Wild-type C57BL/6 strain male and female mice were acquired from Taconic Biosciences, Inc. at 8-weeks old. Animal experiments followed the guidelines and regulations set by the Dana-Farber Cancer Institute Institutional Animal Care and Use Committee (IACUC). Mice maintenance in Dana-Farber Cancer Institute animal facility was in accordance with the Institutional Animal Care and Use Committee approved guidelines. All treatments were given directly to one tumor either by direct intra-tumoral injection or by intra-tumoral implantation of loaded SRB for sustain release. For cohorts treated in conjunction with radiotherapy, a Small Animal Radiation Research platform (SARRP) was used for imageguided radiation therapy at 220 kVp and 13 mA. The study design included a randomization process of the mice followed by assortment into the following cohorts of: no treatment, RT dose of 6 Gy, FBL-03G with/without 6Gy, and SRB loaded with FBL- 03G and with/without 6Gy. All mice that received FBL-03G treatment in the first and second trials were treated with 100-µg of FBL-03G immunoadjuvant. The same amount of payload was used in SRBs as with other administration routes. SRBs were administered in the right tumors using a 17-Gauge clinical brachytherapy needle. Dimethyl sulfoxide (DMSO) was used as a solvent to dissolve FBL-03G powder. To investigate the potential of FBL-03G as an immunotherapy, different concentrations for FBL-03G in the amount of 100, 200, and 300 µg were considered. Same procedure of drug loading into SRBs were followed as the first and second trials. Tumor volumes were measured for both tumors on day of treatment and at least 1–2 times/week post-treatment. A survival study was also performed. Mice were euthanized when either tumor exceeded 20 mm in diameter collectively and/or when tumors were ulcerated or ruptured. A control cohort was created with no treatment of FBL-03G but an inoculation of DMSO.

In 3 independent animal studies, the mice were randomized and divided into some the following cohorts of: no treatment, 6 Gy, FBL-03G with/without 6Gy, and SRB loaded with/without FBL-03G and with/without 6Gy. Mice inoculated with FBL-03G

treatment received either 100, 200, or 300 ug of FBL-03G. Payload of the same amount of FBL-03G was used in SRBs as with other administration routes.

### Tumor Volume Assessment

Directly after treatment, a digital Vernier caliper was used to measure the length and width of the dermal tumors. Tumor volume formula used: (length × width<sup>2</sup> )/2. Measurement imaginary longitude to the leg was chosen as length and the vertical was for width. The tumors were restrained between the skin surface layers. The tumor volume was plotted against time. Animal survival was performed for treatments following IACUCapproved protocol, which was predetermined based on published evidence justifying such a study design. Tumor attainment > 1 cm in diameter on both flanks or tumor burst were determined as excessive tumor burden and mouse was euthanized following the protocol.

#### Statistical Analysis

The in-vitro experiments were conducted in triplicate, and data were presented as mean ± standard error or in the form quantified otherwise. Mice tumor volume were scrutinized using

FIGURE 2 | *In-vitro* anti-cancer effect of FBL-03G. FBL-03G drug from flavonoids shows anti-cell proliferation effects in combination with radiotherapy. Average results of normalized clonogenic assays are shown, respectively, for Panc-02 and KPC cells. (A,B) Results of synergistic outcomes when combining radiotherapy at 4Gy with different FBL-03G doses. Statistics is shown for cohorts treated at 4Gy at different doses of FBL-03G. Statistical Analyses using Student's *T*-Test for the Cell % survival at different concentrations of FBL-03G graphs (*n* = 3 independent trials), (\**P* < 0.05; \*\**P* < 0.01; \*\*\**P* < 0.001; \*\*\*\**P* < 0.0001).

post treatment.

effect in pancreatic cancer slowing down tumor growth for both treated and untreated tumors. Two experiments were conducted simultaneously: Study 1 results are shown in graphs (A–D) and combined survival results for study 1 and study 2 results are displayed in (E). (A) Volumes of non-treated tumors over time without SRB (*n* = 3 for each cohort). (B) Volumes of treated tumors over time (*n* = 3 for each cohort). (C) Volume of non-treated tumors over time with SRB and FBL-03G (*n* = 3 for control and 6Gy cohorts respectively; *n* = 4 for SRB loaded with FBL-03G with/without radiotherapy cohorts respectively). (D) Volume of treated tumors over time for cohorts treated with SRB and FBL-03G (*n* = 3 for control and 6 Gy cohorts respectively; *n* = 4 for SRB loaded with FBL-03G with/without radiotherapy cohorts respectively). (E) Survival results show significant increase in survival for cohorts treated with SRB loaded with FBL-03G (each *n* = 9) compared to control (*n* = 6), 6Gy/FBL-03G/FBL-03G\_6Gy (each *n* = 3). For Statistical Analyses (\**P* < 0.05; \*\**P* < 0.01) Student's *T*-Test was used for comparing the volumes of tumors for each treatment group versus those of the control group with no additional corrections, and Log-rank (Mantel-Cox) was used for the survival graphs.

standard Student's two-tailed t-test. Mice survival were analyzed using GraphPad Prism 8.0. A p-value of <sup>∗</sup>P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001, and ∗∗∗∗P < 0.0001 were deemed as statistically significant difference.

### RESULTS

**Figure 1** illustrates the therapy approach using FBL-03G loaded in smart radiotherapy biomaterials (SRBs) for sustained delivery

to tumor cells. Before in-vivo studies with the FBL-03G, in-vitro studies were carried out with sustained exposure of cancer cells with FBL-03G. Clonogenic assay was performed to identify the anti-cancer effect of the FBL-03G drug with and without radiotherapy on 2 pancreatic cancer cell lines, KPC and Panc-02. **Figure 2** highlights enhanced tumor cell death for the combination treatment of FBL-03G and radiation compared to individual treatments alone. The clonogenic survival results show that the use of 1µM of FBL-03G has synergistic effect on pancreatic cells with exposure to 4 Gy of radiotherapy in terms of decreasing pancreatic cancer cell proliferation. These findings were observed for both Panc-02 and KPC pancreatic cancer cell lines. This demonstrates therapy potential for the FBL-03G. Moreover, the use of 4µM of FBL-03G was apparently more effective in killing pancreatic cancer cells than 4 Gy of radiotherapy. This suggests that FBL-03G can induce apoptosis and inhibit cancer cell proliferation with optimized drug concentrations. The FBL-03G effect on cancer cells combined with DNA damage from radiotherapy could account for the observed synergistic outcomes.

Using smart radiotherapy biomaterials for prolonged delivery of FBL-03G into tumors can also boost malignant cell death in-vivo with FBL-03G. The in-vivo study design is illustrated in **Figure 3**. The results are shown in **Figure 4**. **Figures 4A–D** shows a distinction between direct intra-tumor injection of FBL-03G vs. using the same concentration of FBL-03G with the smart radiotherapy biomaterial platform. **Figures 4C,D** shows reduction of tumor growth in animal cohorts treated with FBL-03G loaded in SRB compared to cohorts treated with direct administration of the same dose of FBL-03G shown in **Figures 4A,B** vs. control and irradiated cohorts. Remarkably, the results in **Figure 4D** revealed that nearby non-treated (abscopal) tumors, representing metastasis, were also significantly affected with slowed tumor growth. Repeated experiments showed significant increase in mice survival (**Figures 4E,F**) compared to control cohorts. The findings provide a basis for further studies to optimize different parameters for maximal outcomes via this approach.

In another study we evaluated the effect of FBL-03G using 3 different concentrations (100, 200, and 300 ug) loaded in smart radiotherapy biomaterials with/without radiotherapy. The findings in **Figure 5** show no significant difference in tumor volume between using smart radiotherapy biomaterials with FBL-03G alone vs. using SRBs with FBL-03G payloads in combination with radiotherapy. However, a significant difference between 6Gy and control cohorts vs. combination of FBL-03G in SRB treated groups is observed in **Figures 5A,B**, where tumor growth of both treated and non-treated tumors is inhibited compared to control and 6Gy cohorts. Overall, the data demonstrates significant therapeutic potential for using FBL-03G in the treatment of both local and metastatic disease, significantly increasing survival (**Figure 5C**).

#### DISCUSSION

From the results of this study, the key findings include, observation that a non-cannabinoid derivative of cannabis can enhance radiotherapy treatment outcomes in-vitro and in-vivo as highlighted in **Figures 2**, **4**. Secondly, the sustained delivery of the cannabis derivative FBL-03G from smart radiotherapy biomaterials (SRBs) results in tumor growth inhibition of both locally treated and distant untreated tumors, with and without radiotherapy. The use of smart radiotherapy biomaterials (SRBs) (8, 23) was recently proposed as a novel approach to deliver cannabinoids, allowing for prolonged exposure of tumor cells to these cannabis derivatives, which is expected to be more effective (10). The FBL-03G payload used in this study is a flavonoid noncannabinoid derivative of cannabis, and the potential to inhibit both local and metastatic tumor progression is remarkable, especially for pancreatic cancer, with a dismal 5-year survival rate of 8% (1).

While ongoing studies are in progress to address the specific mechanism for this immunotherapy potential of this cannabis derivative, the possibility of leveraging such a therapy approach to treat metastasis or increase survival is significant, given that most pancreatic cancer patients are diagnosed already with metastatic disease, with limited treatment options. The results highlight the potential of using non-cannabinoid/non-psychoactive derivatives of cannabis for such treatment. Further work to optimize therapeutic efficacy for such cannabis derivatives and evaluate toxicity could set the stage for clinical translation. An advantage of the SRB approach here is also that this could minimize any toxicity due to in-situ delivery and use of multifold less immunoadjuvant. Furthermore, the use of a single dose of RT as done in this project would minimize normal tissue toxicity.

Although **Figure 4** shows no significant difference between using SRBs alone vs. SRBs with RT, SRBs could simply be used like fiducial markers (23) (**Figure 1**) offering a viable pathway to clinical translation at no additional inconvenience to patients. Another advantage of using SRBs for sustained in-situ delivery of payloads is the relative convenience in delivering the immunoadjuvants, compared to repeated injections. Using only one fraction of RT would also be more convenient for cancer patients who usually must come in repeatedly over many weeks to be treated with several fractions of radiotherapy. This should significantly reduce treatment time and costs. It would be a benefit in resource-poor-settings where access to RT services is limited, reducing cancer health disparities, with major impact in global health.

While the results indicate that sustained exposure of tumor cells to FBL-03G can boost both local and metastatic tumor cell kill, the mechanism of such action needs to be further investigated. One hypothesis is that, FBL-03G can serve as an immunotherapy agent, inhibiting growth of locally treated and untreated tumors, representing metastasis. Metastasis accounts for most of all cancer associated suffering and death, and questionably presents the most daunting challenge in cancer management. Henceforth, the observed significant increase in survival is promising, especially for pancreatic cancer which is often recalcitrant to treatments. Another hypothesis is that sustained delivery allows FBL-03G to reach the untreated tumor over a prolonged period as well. Either way, the FBL-03G results reveal a new potential non-cannabinoid cannabis derivative with major potential for consideration in further investigations in the treatment of pancreatic cancer, where new therapy options are urgently needed.

### CONCLUSION

In this study, a flavonoid derivative of cannabis demonstrates significant therapy potential in the treatment of pancreatic cancer, including radio-sensitizing and cancer metastasis treatment potential. The results justify further studies to optimize therapy outcomes toward clinical translation.

#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

### ETHICS STATEMENT

Animal experiments and protocol followed the guidelines and regulations set by the Dana-Farber Cancer Institute Institutional Animal Care and Use Committee (IACUC). Mice maintenance in Dana Farber Cancer Institute animal facility was in accordance with the Institutional Animal Care and Use Committee approved guidelines.

### AUTHOR CONTRIBUTIONS

MM provided intellectual contributions to the design of the mice study, generated all the results in this study, designed the smart radiotherapy biomaterial (SRB) to implant in mice, and wrote most of the manuscript. SY-K reviewed the manuscript. UI, KD, and NB helped in tumor measurements for mice and reviewed the manuscript. NT and HL provided the FBL-03G drug, contributed to study design, made input in manuscript including reviewing the manuscript. WN is the principal investigator who designed the study and wrote a portion of the manuscript.

#### FUNDING

Funding support for this work is acknowledged from the USA National Institutes of Health (NIH), grant number R21CA205094 and Flavocure Biotech Inc.

#### REFERENCES


#### ACKNOWLEDGMENTS

We thank all the members at Ngwa's lab at Dana Farber Cancer Institute and Harvard Medical School, Boston, Massachusetts for their support, and Servicebio Inc., Woburn, Massachusetts for technical support. Bio-Tech R&D Institute, University of the West Indies, Mona, Jamaica is also acknowledged for its technical support.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00660/full#supplementary-material

Figure S1 | Structures of Cannabis flavonoids and the unnatural molecule FBL-03G. Through a bioactivity guided isolation approach, Cannflavin B (10) was isolated from a flavonoid rich fraction of Cannabis using flash chromatography along with 10 other compounds and characterized by NMR and MS spectroscopic methods. The spectroscopic data were similar to the data of the following 11 compounds previously isolated and characterized from the Cannabis plant; apigenin (1), Chrysoeriol (2), kaempferol (3), luteolin (4), quercetin (5), vitexin (6), isovitexin (7), orientin (8) and prenylated flavonoids including Cannflavin A (9), Cannflavin B (10), and Cannflavin C (11).


23. Ngwa W, Kumar R, Moreau M, Dabney R, Herman A. Nanoparticle drones to target lung cancer with radiosensitizers and cannabinoids. Front Oncol. (2017) 7:208. doi: 10.3389/fonc.2017.00208

**Conflict of Interest Statement:** NT and HL work for Flavocure Biotech Inc., a for-profit company.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Moreau, Ibeh, Decosmo, Bih, Yasmin-Karim, Toyang, Lowe and Ngwa. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Corrigendum: Flavonoid Derivative of Cannabis Demonstrates Therapeutic Potential in Preclinical Models of Metastatic Pancreatic Cancer

Michele Moreau1,2, Udoka Ibeh1,3, Kaylie Decosmo1,4, Noella Bih<sup>1</sup> , Sayeda Yasmin-Karim<sup>1</sup> , Ngeh Toyang<sup>5</sup> , Henry Lowe<sup>5</sup> and Wilfred Ngwa1,2 \*

*<sup>1</sup> Brigham and Women's Hospital, Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, United States, <sup>2</sup> Department of Physics, University of Massachusetts Lowell, Lowell, MA, United States, <sup>3</sup> Department of Biology, University of Massachusetts Boston, Boston, MA, United States, <sup>4</sup> Department of CaNCURE Program, Northeastern University, Boston, MA, United States, <sup>5</sup> Flavocure Biotech Inc., Baltimore, MD, United States*

Keywords: pancreatic cancer, flavonoids, cannabis, metastasis, radiotherapy, smart biomaterials

#### Edited by:

**A Corrigendum on**

*Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands*

Reviewed by: *Otto Kalliokoski, University of Copenhagen, Denmark*

#### \*Correspondence:

*Wilfred Ngwa wngwa@lroc.harvard.edu*

#### Specialty section:

*This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology*

Received: *30 April 2020* Accepted: *07 July 2020* Published: *21 August 2020*

#### Citation:

*Moreau M, Ibeh U, Decosmo K, Bih N, Yasmin-Karim S, Toyang N, Lowe H and Ngwa W (2020) Corrigendum: Flavonoid Derivative of Cannabis Demonstrates Therapeutic Potential in Preclinical Models of Metastatic Pancreatic Cancer. Front. Oncol. 10:1434. doi: 10.3389/fonc.2020.01434*

#### **Flavonoid Derivative of Cannabis Demonstrates Therapeutic Potential in Preclinical Models of Metastatic Pancreatic Cancer**

by Moreau, M., Ibeh, U., Decosmo, K., Bih, N., Yasmin-Karim, S., Toyang, N., et al. (2019). Front. Oncol. 9:660. doi: 10.3389/fonc.2019.00660

In the original article, there was a mistake in **Figures 4E, 4F**, and **5C** as published. This was due to errors during use of analysis software. The survival data in **Figures 4E,F** has been combined into one **Figure 4E**. The figure legend of **Figure 4** has been updated to reflect the correction made in the figure. The corrected **Figures 4** and **5** appear below.

The data for the tumor volume and survival results has also now been published as **Supplementary Material**.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/ articles/10.3389/fonc.2020.01434/full#supplementary-material

The authors apologize for this error and state that these do not change the scientific conclusions of the article in any way. The original article has been updated.

Copyright © 2020 Moreau, Ibeh, Decosmo, Bih, Yasmin-Karim, Toyang, Lowe and Ngwa. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

effect in pancreatic cancer slowing down tumor growth for both treated and untreated tumors. Two experiments were conducted simultaneously: Study 1 results are shown in graphs (A–D) and combined survival results for study 1 and study 2 results are displayed in (E). (A) Volumes of non-treated tumors over time without SRB (*n* = 3 for each cohort). (B) Volumes of treated tumors over time (*n* = 3 for each cohort). (C) Volume of non-treated tumors over time with SRB and FBL-03G (*n* = 3 for control and 6Gy cohorts respectively; *n* = 4 for SRB loaded with FBL-03G with/without radiotherapy cohorts respectively). (D) Volume of treated tumors over time for cohorts treated with SRB and FBL-03G (*n* = 3 for control and 6 Gy cohorts respectively; *n* = 4 for SRB loaded with FBL-03G with/without radiotherapy cohorts respectively). (E) Survival results show significant increase in survival for cohorts treated with SRB loaded with FBL-03G (each *n* = 9) compared to control (*n* = 6), 6Gy/FBL-03G/FBL-03G\_6Gy (each *n* = 3). For Statistical Analyses (\**P* < 0.05; \*\**P* < 0.01) Student's *T*-Test was used for comparing the volumes of tumors for each treatment group versus those of the control group with no additional corrections, and Log-rank (Mantel-Cox) was used for the survival graphs.

10 for each cohort) were assessed. (A) Volumes of non-treated tumors 2-weeks post treatment (*n* = 10 for each cohort); (B) volumes of treated tumors 2-weeks post treatment (*n* = 10 for each cohort). This study investigated using different concentrations of FBL-03G with/without 6Gy to determine its potential effect on mice survival over time. (C) Represents a Log-rank (Mantel-Cox) survival graph (*n* = 10) (\*\*\*\**p* < 0.0001). (C) Survival results show no difference in survival for cohorts treated with different concentrations of SRB loaded with FBL-03G.

# Exploring Radiation Response in Two Head and Neck Squamous Carcinoma Cell Lines Through Metabolic Profiling

Eva Lindell Jonsson1†, Ida Erngren2†, Mikael Engskog<sup>2</sup> , Jakob Haglöf <sup>2</sup> , Torbjörn Arvidsson2,3, Mikael Hedeland<sup>2</sup> , Curt Petterson<sup>2</sup> , Göran Laurell <sup>1</sup> and Marika Nestor <sup>4</sup> \*

<sup>1</sup> Department of Surgical Sciences, Uppsala University, Uppsala, Sweden, <sup>2</sup> Department of Medicinal Chemistry, Uppsala University, Uppsala, Sweden, <sup>3</sup> Medical Product Agency, Uppsala, Sweden, <sup>4</sup> Department of Immunology, Genetics and Pathology, Uppsala University, Uppsala, Sweden

#### Edited by:

Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands

#### Reviewed by:

Mansi Babbar, National Institute on Aging (NIA), United States Kerstin Borgmann, University Medical Center Hamburg-Eppendorf, Germany

> \*Correspondence: Marika Nestor marika.nestor@igp.uu.se

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 17 April 2019 Accepted: 12 August 2019 Published: 30 August 2019

#### Citation:

Lindell Jonsson E, Erngren I, Engskog M, Haglöf J, Arvidsson T, Hedeland M, Petterson C, Laurell G and Nestor M (2019) Exploring Radiation Response in Two Head and Neck Squamous Carcinoma Cell Lines Through Metabolic Profiling. Front. Oncol. 9:825. doi: 10.3389/fonc.2019.00825 Head and neck squamous cell carcinoma (HNSCC) is the sixth most common form of cancer worldwide. Radiotherapy, with or without surgery, represents the major approach to curative treatment. However, not all tumors are equally sensitive to irradiation. It is therefore of interest to apply newer system biology approaches (e.g., metabolic profiling) in squamous cancer cells with different radiosensitivities in order to provide new insights on the mechanisms of radiation response. In this study, two cultured HNSCC cell lines from the same donor, UM-SCC-74A and UM-SCC-74B, were first genotyped using Short Tandem Repeat (STR), and assessed for radiation response by the means of clonogenic survival and growth inhibition assays. Thereafter, cells were cultured, irradiated and collected for subsequent metabolic profiling analyses using liquid chromatography-mass spectrometry (LC-MS). STR verified the similarity of UM-SCC-74A and UM-SCC-74B cells, and three independent assays proved UM-SCC-74B to be clearly more radioresistant than UM-SCC-74A. The LC-MS metabolic profiling demonstrated significant differences in the intracellular metabolome of the two cell lines before irradiation, as well as significant alterations after irradiation. The most important differences between the two cell lines before irradiation were connected to nicotinic acid and nicotinamide metabolism and purine metabolism. In the more radiosensitive UM-SCC-74A cells, the most significant alterations after irradiation were linked to tryptophan metabolism. In the more radioresistant UM-SCC-74B cells, the major alterations after irradiation were connected to nicotinic acid and nicotinamide metabolism, purine metabolism, the methionine cycle as well as the serine, and glycine metabolism. The data suggest that the more radioresistant cell line UM-SCC-74B altered the metabolism to control redox-status, manage DNA-repair, and change DNA methylation after irradiation. This provides new insights on the mechanisms of radiation response, which may aid future identification of biomarkers associated with radioresistance of cancer cells.

Keywords: radioresistance, radiosensitivity, metabolomics, mass spectrometry, redox status

## INTRODUCTION

Every year more than half a million new cases of squamous cell carcinoma of the head and neck (HNSCC) are reported (1), which makes it the 6th most common type of cancer worldwide (2). HNSCC is a heterogeneous disease, including epithelial cancers of the oral cavity, lip, nasal cavity, paranasal sinuses, larynx, pharynx, and salivary glands. Despite the high frequency of HNSCC worldwide, it has one of the lowest survival rates among the more common cancer types (3), and almost 50% of all patients with HNSCC will die from their disease (4). Treatment challenges include complex anatomy, functional preservation of substantial organs, and minimization of side effects (5).

Radiotherapy (RT), the clinical application of ionizing radiation, is one of the most effective tools in therapy of cancer today (6). Even though advances in treatment methods have been made during recent decades, RT with or without surgery remains the major approach to curative treatment of HNSCC (7). The efficacy of RT is however still limited by different technological, biological, and clinical constraints. HNSCC is on average only moderately radiosensitive, which means that radiotherapy often must be given to such an extent that it approaches the maximum tolerated dose for the surrounding normal tissue. This may cause substantial acute and late toxicities, resulting in significant morbidity and altered quality of life (8). Moreover, the individual heterogeneity in terms of HNSCC radiosensitivity is immense, and therapeutic responses to RT have been shown to vary from complete to no response (9, 10). Consequently, there is a great need for individualized radiotherapy treatment approaches in HNSCC, to aid in predicting and monitoring tumor response to radiotherapy before, during, and after treatment. This requires new insights on the mechanisms of radiation response, novel markers to predict tumor response to radiotherapy, as well as potential treatment targets to enhance radiation-sensitivity.

The metabolism of cancer cells differs from normal differentiated cells (11). Tumor progression, the development of increasingly aggressive and resistant tumor cells, has increasingly been understood to be associated with perturbations in cellular metabolism, such as increased glutaminolysis and fatty acid oxidation, the Warburg effect, as well as altered patterns of macromolecule synthesis and storage (11). Cancer cells are able to adapt metabolically to many types of cellular stress, for example, hypoxia, nutrient depletion, and radiation, and studies have shown several mechanistic links between cellular metabolism and growth control (11). No doubt, the plasticity of cancer cell metabolism can be vital for causing many patients to relapse into disseminated and treatment-resistant disease.

Cancer metabolism is altered by ionizing radiation. Radiation exposure induces different types of genome damage, including DNA single- and double-strand breaks and bulky lesions. Thereby multiple signaling pathways are activated, involved in e.g., DNA damage response, signal transduction pathways, and regulation of survival (9, 12, 13). Enzymatic pathways quickly repair many of these lesions, but lesions that are not repaired correctly lead to chromosomal abnormalities and a possible change of cell phenotype, culminating in cell cycle arrest and/or cell death. In addition to rapid proliferation, many cancer cells are also deficient in repair proteins and cell cycle checkpoints, making them more sensitive to radiation (12). Cellular exposure to ionizing radiation also triggers a complex series of molecular responses that can affect metabolism, either directly or indirectly, by altering cell growth (14). Ionizing radiation also impacts multiple cellular compartments even at relatively modest doses, which can also trigger a variety of signaling pathways (15). These molecular events are not only important for the therapeutic response, but may also influence the inflammatory response at a local and systemic host level. Subsequently, ionizing radiation will result in different alterations in the metabolome, depending on what pathways and processes that the specific cell alters in response to radiation.

Several studies have identified different molecular entities associated with radioresistance, nevertheless the underlying mechanisms are still inconclusive (16). Suggested mechanisms include hypoxia, alterations of the DNA damage response, activation of pathways involved in pro-survival or cell cycle regulation, as well as vascular, stromal, and immunological changes (17–26). However, a majority of these studies are based on genome, transcriptome, and proteome data. Consequently, it would be of major interest to conduct studies in this field closer to the phenotype, such as metabolite profiling (27). Metabolomics in radiation biology have previously been used for two main purposes (i) metabolic profiling for utilization in biodosimetry or for biomarker discovery of radiation exposure (14, 28–39) and (ii) metabolic profiling for a more mechanistic understating of the radiation response of the metabolome (14, 34, 40–43) However, to date no study has focused on investigating the different metabolic responses of genetically similar cells with divergent radiation sensitivities.

Metabolic profiling is the comprehensive and quantitative analysis of small endogenous metabolites that are the downstream products in biological systems. It can be a powerful approach to study the phenotype of cancer cells, since it analyses the biochemical outcome of the activities of the proteome (27). The promise of metabolomics as a scientific tool has been fueled largely by the advancement in nuclear magnetic resonance (NMR) and mass spectrometry (MS), and could be a well-suited and cost-effective complement to current genomic and proteomic data in the field (14).

Even though our current understanding of the mechanisms in play during radiation of tumor cells, and how they are related to radiosensitivity, is incomplete (40), it has been shown to at least in part be due to the different metabolic alterations that the tumor can make (44). Ionizing radiation induces complex biological responses that interfere with gene and protein expression, which disrupts normal metabolic processes in cells and organs. As a result, metabolites related to classical pathways of radiation damage, including oxidative stress and subsequent DNA breakdown have been shown to be affected. Additionally, polyunsaturated fatty acids (PUFA) metabolism can be disrupted as an inflammatory effect of radiation exposure (14). Changes in nicotinate and nicotinamide metabolism and cofactor biosynthesis have also been reported in radiation related research, suggested to be linked to DNA repair (14, 45, 46).

Consequently, metabolomics may provide insights into the mechanism behind a reduced sensitivity to radiotherapy by identifying differences in metabolites produced in response of irradiation in cancer cells with different sensitivity to radiotherapy. This may provide significant mechanistic understanding related to cellular response due to perturbations caused by radiation treatment (14), which might be a possible way to find a pharmacologically alterable pathway that is altered in the less sensitive cells, or to predict response or non-response to radiation therapy.

In the present study, we investigate the relationship between radiation response and the metabolome of HNSCC in a unique model system, using two HNSCC cell lines from the same donor but with different radiosensitivity. This enables the study of their metabolic response to radiation in an exceptionally controlled setting, and has to our knowledge not been performed previously. The aim was to assess the metabolic differences between the two cell lines, and how this was affected by radiation. Cells were first genotyped, and assessed for radiosensitivity using clonogenic survival and long-term growth inhibition assays at several radiation doses. Cell-based metabolic profiling using liquid chromatography hyphenated to high resolution mass spectrometry (LC-HRMS) was then performed to investigate the influence of early and intermediate radiation responses on metabolites, and to assess the potential correlation to the different radiosensitivities of the cells. In the long-term, the study may contribute to provide new insights on the mechanisms of radiation response, discover possible biomarkers for radiationsensitivity, and possibly present potential treatment targets in order to enhance radiation-sensitivity of HNSCC.

### MATERIALS AND METHODS

This study has been conducted in accordance with Frontiers guidelines on study ethics. It does not involve any animal or human subjects or identifiable human data, thus does not require ethical permission.

#### Cell Lines

The squamous cell carcinoma cell lines UM-SCC-74A and UM-SCC-74B were kindly provided by Professor TE Carey (University of Michigan, USA), and have previously been described by Brenner et al. (47). The cell lines were taken from the same male patient with oral squamous cell carcinoma after radiochemotherapy (UM-SCC-74A) and then again at surgery for persistent cancer (UM-SCC-74B) (47). Cells were cultured at 37◦C, in 5% CO<sup>2</sup> in DMEM medium containing 2 mM l-glutamine (Biochrom GmbH, Germany), supplemented with 5% fetal bovine serum (Sigma-Aldrich, Germany), MEM non-essential amino acids (Sigma-Aldrich AB, Germany), and antibiotics (100 IU penicillin and 100µg/ml streptomycin) (Biochrome GmbH, Germany).

### Genotyping

UM-SCC-74A and UM-SCC-74B cells were Genotyped using Short Tandem Repeat (STR) Analysis in order to verify the origin and similarity of the cell lines. DNA was extracted from frozen cell cultures and analyzed using the AmpFLSTR <sup>R</sup> Identifiler <sup>R</sup> PCR Amplification Kit. The Identifiler kit amplifies 15 loci and Amelogenin in a single tube and provides loci consistent with major worldwide STR databasing standards.

### Clonogenic Survival

Clonogenic survival assays were performed as described previously (48), in order to assess the radiosensitivity of UM-SCC-74A and UM-SCC-74B cells. In short, cells were preplated into 25 cm<sup>2</sup> culture flasks with 8 ml complete medium. After 48 h, cells were exposed to external beam radiation using <sup>137</sup>Cs gamma-ray photons at a dose-rate of ∼1 Gy/min (Best Theratronics Gammacell <sup>R</sup> 40 Exactor, Springfield, USA), corresponding to a dose of 0, 2, 4, 6, or 8 Gy. Colonies were allowed to form for 10–14 days. Cells were then fixated with 95% ethanol and stained with crystal violet. The colonies were inspected under a microscope, and only cells giving rise to colonies consisting of 50 or more cells were considered clonogenic survivors. Plating efficiency, PE (number of colonies formed/number of cells seeded in the control), and the survival fraction (number of colonies formed after treatment/number of cells seeded × PE) were calculated in Microsoft Office Excel 2016 for Mac version 14.6.1 (Microsoft, Redmond, WA, USA). Statistical analyses were performed using GraphPad Prism 6 (GraphPad Software, San Diego, CA, USA). Differences in normalized survival fractions of 2 Gy irradiated UM-SCC-74A cells vs. UM-SCC-74B cells were assessed using an unpaired t-test and were considered statistically significant if P < 0.05.

### Radiation Induced Long-Term Growth Inhibition

As a complement to the clonogenic- and 3D cell culture assays, the long-term growth inhibitory effects of radiation were evaluated using a growth inhibition assay as described earlier (49). In short, UM-SCC-74A or UM-SCC-74B cells were pre-plated into 25 cm<sup>2</sup> culture flasks with complete medium. After 48 h, cells were exposed to external beam radiation corresponding to a dose of 0, 2, 4, 6, or 8 Gy. Cells were then counted and reseeded about once a week, and the corresponding total cell numbers were calculated. The increase in cell number was followed for 4 weeks. Cell doubling times were calculated using the least square fitting method. In order to determine any statistically significant differences from the untreated group at the last data point, total cell numbers were analyzed using oneway ANOVA followed by Dunnett's multiple comparisons test in GraphPad Prism and were considered statistically significant if P < 0.05.

#### Radiation Response in 3D Cell Culture

For liquid overlay 3D multicellular tumor spheroid formation, 96-well plates were coated with 0.15% agarose dissolved in PBS with 1% penicillin/streptomycin. One thousand UM-SCC-74B cells/well or 1500 UM-SCC-74A cells/well were seeded and incubated at 37◦C in supplemented media for 3 days prior to irradiation with 2 Gy or mock radiation (0 Gy) using <sup>137</sup>Cs gamma-ray photons as described above. Spheroid images were obtained at start of treatment and 10 days after treatment using a Canon EOS 700D camera mounted on an inverted Nikon Diaphot-TMD microscope. The images were analyzed using ImageJ version 1.48 (NIH, Bethesda, MD, USA), by measuring the surface area of each spheroid and calculating the volume, assuming each spheroid retained a spherical form. Each spheroid was normalized to its own starting volume at the start of treatment (Day 0, growth ratio = 1). Statistical analyses were performed using GraphPad Prism 6 (GraphPad Software, San Diego, CA, USA). Differences in normalized spheroid growth ratios of UM-SCC-74A cells vs. UM-SCC-74B cells were assessed using an unpaired t-test and were considered statistically significant if P < 0.05.

#### Measurement of Cleaved Poly ADP Ribose Polymerase (PARP)

Levels of cleaved PARP1 in cell lines were measured using ELISA. The assay detects the presence of the 89 kDa PARP1 fragment containing the catalytic domain. The proteolysis of PARP1 by activated caspase-3 renders the enzyme inactive, which further facilitates apoptotic cell death. Thus, the presence of the 89 kDa PARP1 fragment is considered to be a reliable biomarker of apoptosis. Cells were incubated for 48 h prior to irradiation (2 Gy) or mock radiation (0 Gy) using <sup>137</sup>Cs gamma-ray photons as described above. Whole-cell lysates were prepared 12 h after irradiation according to standard protocols. Cell lysates were diluted 1:1,000. The Cleaved PARP1 Human SimpleStep ELISA <sup>R</sup> Kit (Abcam, Cambridge, UK) was used according to the manufacturer's protocol. The OD was then measured at 450 nm using a microtiter plate reader (BioRad, USA). Statistical analyses were performed using GraphPad Prism 6.

Differences in cleaved PARP1 levels were assessed using an unpaired t-test and were considered statistically significant if P < 0.05.

#### Irradiation of Cells for Metabolic Profiling

Two days before irradiation (18–25) × 10<sup>6</sup> UM-SCC-74A or (10–25) × 10<sup>6</sup> UM-SCC-74B cells were cultured in cell culture plates (n = 46 and n = 52, respectively, Nunclon Surface, 15 cm diameter, Cat No. 168 381, 145 cm<sup>2</sup> ) at 37◦C, in 5% CO<sup>2</sup> in supplemented DMEM medium. At the time of irradiation, cells were exposed to external beam radiation corresponding to a dose of 0 or 2 Gy. Cells were subsequently harvested at ∼75% confluence at 4 h (n = 20 and 22 for UM-SCC-74A and UM-SCC-74B, respectively) and 24 h (n = 26 and 30 for UM-SCC-74A and UM-SCC-74B, respectively) after irradiation as according to Engskog et al. (50). The time points were chosen in order to detect IR-induced perturbations in the cell metabolome in the most optimal settings possible. These time points have been shown to be relevant for IRinduced early- and intermediate cellular responses in previous cell-based radiation metabolomic assessments (40, 41, 43). Moreover, while the cellular responses have occurred at these time points, they have not yet resulted in apoptosis in the majority of cells. Consequently, in these experimental settings a majority of the irradiated cells can be harvested, reducing any risk of selection-biased analyses. All cell sample harvesting was performed on ice. Cell medium was removed and cells were rapidly washed three times with cold, sterile phosphate buffered saline (PBS, Medicago, Uppsala, Sweden), followed by detachment of cells using a 23 cm long rubber-tipped Nunc cell scraper (Thermo Scientific). The detached cells were collected in 3.5 ml cold MilliQ water, transferred to 15 mL polypropylene brown tubes (Greiner bio-one GmbH, Germany) and snapfrozen in liquid N<sup>2</sup> followed by thawing at 37◦C for 10 min. The freeze/thaw cycle was then repeated twice with subsequent sonication on ice for 30 s. Samples were stored at −80◦C until metabolite extraction.

### Metabolite Extraction

Prior to extraction of the intracellular metabolites, the samples were randomized into five separate sample batches comprised of 20 samples each. Each sample batch was prepared and analyzed separately. The samples were thawed at room temperature and centrifuged at 2000 RCF for 10 min at 4◦C. A quality control (QC) sample was created for each batch by pooling an equal volume from all samples within each batch. The QC sample was extracted as described below. The five QC samples were pooled after extraction of all five batches. The aqueous supernatants were transferred to fresh extraction tubes followed by addition of chloroform and methanol for the final proportion 2.85:4:4 water:methanol:chloroform (51–53). The samples were vortexed gently and stored at 8◦C for 20 min. to the samples were the centrifuged for 20 min at 2000 RCF and 4◦C. The aqueous phases were recovered and evaporated to dryness at 40◦C under a gentle stream of N2(g).The samples were stored at −80◦C after evaporation. Prior to analysis the samples were reconstituted in acetonitrile:Milli-Q water 76:24.

#### Metabolite Profiling With LC-MS

All metabolite profiling analyses were performed on an Acquity UPLC I-class system from Waters (Manchester, UK) hyphenated to a G2S Synapt Q-TOF equipped with an electrospray ionization (ESI) source (Waters). All systems were controlled using Masslynx version 4.1 (Waters). For chromatographic sample separation prior to detection a Acquity BEH amide column (1.7µm, i.d. 2.1 × 50 mm) from Waters was used. The column temperature was kept at 40◦C for all analyses and the injection volume was 5 µl. Mobile phase A consisted of 95:5 acetonitrile/water with 10 mM ammonium formate and 0.1% FA and mobile phase B consisted of 50:50 acetonitrile/water with 10 mM ammonium formate and 0.1% FA. A non-linear elution gradient from 100% A to 100% B was used, the flow rate was set to 0.3 ml/min. In detail: 100% A was kept for 0.5 min then decreased non-linearly (slope-factor 8 in MassLynx) over 12.5 min to 100% B, 100% B was held for 4 min followed by 6 min at 100% A to re-equilibrate the column for a total run-time of 23 min. Detection was performed in both positive and negative ionization mode in resolution MS<sup>E</sup> -mode within the scan-range m/z 50–800. All samples were analyzed in negative ionization mode first, followed by positive ionization mode. The capillary voltage was 1 and −2 kV and the cone voltage was set to 30 and 25 V in positive and negative ionization mode, respectively. In both ionization modes the source temperature was 120◦C. Nitrogen was used as desolvation gas at the flow-rate 800 l/h and the desolvation temperature was 500◦C and 450◦C in positive and negative mode, respectively. Nitrogen was used as cone gas as well at a flow-rate of 50 l/h. and. A collision energy ramp from 20 to 45 eV was used for MS<sup>E</sup> acquisition with argon as collision gas. Lock-mass correction for accurate m/z measurements was applied using a solution of leucine-enkephalin (m/z 556.2766). Each sample batch was analyzed separately, i.e., one sample batch per day. Prior to each sample batch analysis the instrument was mass calibrated and the sample cone was cleaned. A reference mix (30µM of hypoxanthine, cytidine, phenylalanine, tryptophan, and glutamine, respectively) was analyzed before and after each batch to check the system suitability with regard to mass accuracy, instrument sensitivity, and column performance. The QC sample was analyzed repeatedly prior to the study samples for system conditioning, to ensure stable analytical conditions, as well as, between the randomized study samples in regular intervals to monitor the analytical stability throughout the analysis (54).

### Chemicals

Formic acid, FA(LC-MS grade), ammonium formate (LC-MS grade), methanol (LC-MS grade), Cytidine (99%), hypoxanthine (≥99%), and tryptophan (≥99.5%) were purchased from Sigma Aldrich (Steinheim, Germany). Phenylalanine (>99%) was purchased from MERCK (Kenilwoth, N.J., USA) while glutamine (>99%) was purchased from Fluka (Buchs, Switzerland). Acetonitrile (LC-MS grade) was obtained from Fischer Scientific (Zurich, Switzerland) and chloroform (analytical grade) was purchased from BDH Laboratory Supplies (Poole, England, UK). Leucine-encephalin was prepared and certified by ERA (Golden, CO, USA). The water was purified using a Milli-QTM water system from MilliPore (Bedford, MA, USA).

#### Data Processing

Data quality was assessed through in-depth examinations of five representative metabolites spread out in the obtained chromatograms; hypoxanthine, cytidine, phenylalanine, tryptophan, and glutamine. The evaluation was performed by univariate data analysis based on all QC sample injections, in total 25 injections (20% of all sample injections), prior to data pre-processing and multivariate data analysis. Mass accuracy, retention time, and peak area was monitored to check system stability throughout the analysis. DataBridge (Masslynx version 4.1, Waters) was used to convert the raw data files to NetCDF files. The R-based software XCMS was used for peak detection, retention time alignment, and peak grouping (55). The centWave function was used for peak detection with the following function parameters; the maximal deviation in m/z between scans was set to 10 ppm, the boundaries for peak width was set between 5 and 45 s and the signal to noise ratio cut-off was set to 5. The "obiwarp" function was used for retention time alignment. The processed data was subjected to adduct, isotope and fragment annotation using the R-Package, CAMERA (56). The resulting dataset was exported to Microsoft Excel and prior to normalization all features with a retention time <50 s were removed. The data was normalized using LocalMean correction where all features were normalized to the feature mean of the QC:s in the respective batches (57). After normalization, all features with coefficient of variance (CV) >30% in the QC samples were removed (54, 58, 59).

### Multivariate and Univariate Data Analysis

The reduced and filtered data sets from the data processing were analyzed by multivariate data analysis using SIMCA-P+ (version 14, Umetrics, Umeå, Sweden). All data was pareto scaled prior to multivariate data analysis. Principal Component Analysis (PCA) was used to find sample clustering, identify possible sample outliers, and systematic trends in the data. Orthogonal Projection to Latent Structures- Discriminant Analysis (OPLS-DA) in combination with S-Plots were used analyse differences between sample groups and to identify differentiating features between sample groups (60, 61).

Comparisons were made between the two cell line controls as well as between irradiated cells and controls of the respective SCC cell lines. The irradiated cells were further divided into subgroups of rapid response and intermediate response depending on the time between cell irradiation and cell sample harvesting. Rapid response subgroups were harvested 4 h after irradiation and intermediate response subgroups were harvested 24 h after irradiation.

Features with p-corr values >0.4 were selected and annotated. Molecular weight, isotopic patterns, fragmentation and, when possible, retention time comparison to an in-house database were utilized for feature annotation. The Human Metabolome Database (HMDB), METLIN and in-house databases was utilized to search for the experimental m/z values with a molecular weight difference tolerance of 30 ppm. The raw data signal isotopic pattern, fragmentation (when reference spectra was available) as well as related adducts present at the same retention time in the raw data were all matched against the plausible hits from the data base search. All annotated metabolites should be considered putatively annotated (level 2) according to the Metabolomics Standards Initiative nomenclature (62). All annotated metabolites were subjected to pathway analysis using MetaboAnalyst 3.0 and the highest score pathways were subjected to further data analysis. One-Way Analysis of Variance (ANOVA) and post-hoc Tukey tests using Origin 2015 (OriginLab corporation, Northampton, MA, USA) was used for univariate significance testing of all annotated features and p-values <0.05 was considered significant. The significantly altered metabolite are presented as fold changes with 95% confidence intervals. The confidence intervals of fold changes were calculated using Fieller's theorem.

### RESULTS

#### Genotyping

STR results demonstrated identical results for unirradiated UM-SCC-74A and UM-SCC-74B (**Supplemental Table 1**), verifying the origin and similarity of UM-SCC-74A and UM-SCC-74B cells.

### Clonogenic Survival, Cell Growth, and Apoptosis Assays

The effect of radiation on UM-SCC-74A and UM-SCC-74B cell viability was studied using clonogenic survival assays, 3D cell growth assays, and long term growth inhibition assays, and can be seen in **Figures 1**, **2**. In all three assays, UM-SCC-74A cells proved more sensitive to radiation than UM-SCC-74B cells.

In the clonogenic survival assay (**Figures 1A,B**), UM-SCC-74A cells demonstrated a lower survival fraction than UM-SCC74B at all radiation doses assessed (**Figure 1A**). Accordingly, a radiation dose of 2 Gy to UM-SCC-74A resulted in a survival fraction of 24 ± 5 (SEM)% of the unirradiated controls, whereas the survival fraction of irradiated UM-SCC-74B cells was significantly higher, 52 ± 6% (**Figure 1B**).

In the three-dimensional cell growth assay (**Figure 1C**), spheroid growth was followed for 10 days after irradiation, reflecting effects of both cell death and growth inhibition in a more in vivo-like environment (63). Also here, UM-SCC-74B cells were significantly less affected by radiation, where a radiation dose of 2 Gy resulted in a normalized spheroid size of 84 ± 5 (SEM)% of unirradiated controls, compared to 44 ± 3% for UM-SCC-74A spheroids.

In the long-term cell growth assay, cell growth was followed for 4 weeks' time, reflecting long-term effects of both cell death and growth inhibition. Doubling times of unirradiated UM-SCC-74A and UM-SCC-74B cells were 1.48 and 1.28 days, respectively. Also in this assay, UM-SCC-74A cells were more affected by radiation than UM-SCC-74B cells (**Figures 2A,B**). At the last assessed time point (27 days), UM-SCC-74A cells exposed to 2 Gy of irradiation were 52 ± 19% (SEM) of unirradiated controls. For 4, 6, and 8 Gy, the corresponding numbers were 5.9 ± 2.0, 1.0 ± 0.1, and 0.01 ± 0.01% (**Figure 2C**). For UM-SCC-74B cells, 2 Gy irradiation resulted in a cell number of 65 ± 12% of unirradiated controls at day 27, and for 4, 6, and 8 Gy the corresponding numbers were 37 ± 16, 4.2 ± 0.9, and 0.11 ± 0.05% (**Figure 2D**).

Moreover, apoptosis was also studied in the cells using a cleaved PARP1 assay (**Supplemental Figure 1**). Levels of cleaved PARP1 were significantly increased in 2 Gy irradiated UM-SCC-74A cells compared to unirradiated cells, whereas levels did not significantly differ for UM-SCC-74B cells.

Consequently, as all these independent assays demonstrated UM-SCC-74A cells to be clearly more affected by radiation in terms of cell viability and growth than UM-SCC-74B cells, UM-SCC-74A are referred to as "radiosensitive" and UM-SCC-74B cells as "radioresistant" in the present study. These are to be viewed as relative terms, where UM-SCC-74A cells are "radiosensitive" in relation to UM-SCC-74B cells and vice versa.

#### Metabolic Profiling Data Quality Control

Mass accuracy, retention time, and peak area of five metabolites; hypoxanthine, cytidine, phenylalanine, tryptophan, and glutamine were monitored in the QC samples throughout the analysis to verify system stability. The mass error was found to be within 10 and 12 ppm in positive and negative ionization mode, respectively, and the variation in retention time displayed a CV-value below 1.5% throughout the analysis. The peak areas over the five analytical batches evidenced, as expected, some batch variations; this was however corrected for after the normalization by local mean correction as no sample separation due to batch effects were found in the multivariate data analysis. The peak areas of hypoxanthine, cytidine, phenylalanine and tryptophan varied from 26 to 48% (CV) in the raw data while glutamine showed huge variations of up to 128%.

### Metabolic Profiling

Multivariate data analysis was performed on the pre-processed and filtered data using PCA and OPLS-DA. All samples were analyzed with the unsupervised model PCA to examine possible sample group separations and sample clustering behavior. The PCA scores plot revealed clear discrimination between the intracellular metabolome of the SCC cell lines in the second component (PC2) (**Figure 3**). There was some separation between irradiated cells and controls along the first component (PC1), the separation was most pronounced for the less radiosensitive UM-SCC-74B cell line. However, for both cell lines the controls cluster at the left hand side of the scores plot with the irradiated cells on the right hand side.

The supervised model OPLS-DA was used to analyse the metabolic changes due to irradiation in the two SCC cell lines as well as the metabolic differences between the SCC cell line controls (unirradiated controls). The metabolic changes after irradiation in each SCC cell line was separated into rapid response (cells were harvested 4 h after irradiation) and intermediate response (cells were harvested 24 h after irradiation), respectively.

The metabolic profiling demonstrated a number of significant (p < 0.05) metabolic differences between the two non-irradiated cell line controls (**Table 1**). Several metabolites connected to nicotinate and nicotinamide metabolism such as nicotinic acid, nicotinamide, and nicotinic acid mononucleotide were found altered in the UM-SCC-74B cell line as compared to UM-SCC-74A (p = 0.005). Moreover, guanosine, inosine, xanthine, and the purine intermediate 5-phosphoribosylamine were found at significantly different levels in the two cell lines indicating changes in purine metabolism and biosynthesis between UM-SCC-74A and UM-SCC-74B (p = 0.006).

In the radiosensitive cell line UM-SCC-74A, very few metabolic alterations were observed after irradiation (**Table 2**). The main metabolic alterations were linked to tryptophan metabolism through tryptophan and 5-hydroxyindoleacetic acid (5-HIAA) (p = 0.0006). In contrast, the less radiosensitive cell line UM-SCC-74B showed numerous metabolic alterations after irradiation, where the main metabolic alterations were connected to nicotinate and nicotinamide metabolism (p = 0.003), the methionine cycle (p = 0.05), and purine metabolism (p = 0.001) (**Table 3**). A number of metabolites involved in the nicotinate and nicotinamide metabolism such as 1-methylnicotinamide, niacinamide, beta-nicotinamide D-ribonucleotide, nicotinic acid mononucleotide, and nicotinamide adenine dinucleotide (NAD) were all down-regulated in the UM-SCC-74B 24 h after irradiation. Several metabolites involved in the methionine cycle as well as the glycine and serine biosynthesis and

FIGURE 1 | (A) Clonogenic survival assay of UM-SCC-74A and UM-SCC-74B cells treated with a radiation dose of 0, 2, 4, 6, and 8 Gy. N > 3. Groups are normalized to the plating efficiency of the non-irradiated controls. Error bars represent the standard error of mean. (B) Comparison of clonogenic survival of UM-SCC-74A and UM-SCC-74B cells treated with a radiation dose of 2 Gy. N > 6. Difference in survival fraction was assessed using an unpaired t-test. \*\*p < 0.01. (C) Assessment of UM-SCC-74A and UM-SCC-74B spheroid growth 10 days after 2 Gy irradiation. Groups are normalized to the growth ratio of the non-irradiated controls. N > 5. Difference in spheroid growth was assessed using an unpaired t-test. \*\*\*\*p < 0.0001.

metabolism were found altered in the irradiated UM-SCC-74B cells. Increased levels of methionine were observed 4 h after irradiation, while the levels of S-adenosylmethionine (SAM) was increased both 4 h and 24 h after irradiation. Metabolites involved in purine metabolism such as adenosine, guanosine, and guanine were upregulated 24 h after irradiation while guanosine mono phosphate (GMP) was found in significantly higher levels 4 h after irradiation as compared to 24 h after irradiation.

#### DISCUSSION

A current clinical problem in HNSCC is the varying therapeutic response to irradiation, from complete to no response, moreover relapse and resistance is common (64). Individualized dosimetry, e.g., by predicting or monitoring tumor radiation response before, during, or after treatment, could help optimize radiotherapy, improve therapeutic outcome, and reduce normal tissue complication after radiotherapy. The plasticity of cancer

(PC2). Along the first component (PC1) there is some separation between the controls and irradiated cells, however, no clear clustering. UM-SCC-74A Control (), UM-SCC-74A Rapid response (4 h after 2 Gy irradiation) (N), UM-SCC-74A Intermediate response (24 h after 2 Gy irradiation) (•), UM-SCC-74B Control (), UM-SCC-74B Rapid response (△), and UM-SCC-74B Intermediate response (◦).

cell metabolism plays a major role in cancer cell survival and treatment-resistant disease. Cancer cells are able to adapt metabolically to many types of cellular stress, such as radiation, and previous studies have demonstrated alterations in cellular metabolites after irradiation of the cells (40, 41, 43, 65–68). Consequently, a feasible way to provide significant understanding on the mechanisms of radiation response may be through metabolic profiling. In the present study, we have assessed metabolic profiling as a mean to investigate the relationship between radiation response and the metabolome. This was done by utilizing a unique in vitro model, in which two HNSCC cell lines from the same donor were employed. These cells exhibited the same STR genetic profile (**Supplemental Table 1**) but were shown to display different radiosensitivities, where UM-SCC-74A was shown to be clearly more sensitive to radiation than UM-SCC-74B in four independent assays (**Figures 1**, **2**, **Supplemental Figure 1**). These assays reflect different thresholds, parameters, and time-frames for assessment of cell viability and growth. Consequently, they are to be seen as important complements to each other in order to assess the full spectrum of radiation response.

### UM-SCC-74B Cells Were More Radioresistant Than UM-SCC-74A Cells

In the clonogenic survival assay, considered as a gold standard in radiation research (48, 69), UM-SCC-74A displayed a lower survival fraction at all radiation doses assessed, at 2 Gy approximately half of that of UM-SCC-74B (**Figures 1A,B**). The assay combines contribution of all types of cell death, encompassing both early and late events. However, the impact of cell-to-cell communication is disregarded. Moreover, quiescent cells and cells growing at a slower rate may be counted as non-surviving clones in the clonogenic survival assay.

In contrast, the long-term growth inhibition assay, and in particular the 3D cell culture assay, include cell-to-cell communication, and both assays reflect both cell death and cell growth inhibition. The long-term growth inhibition assay is an especially suitable complement to clonogenic survival assays at higher doses, where the plating efficiency for clonogenic survival may be too low to ensure reliable results. The longer time-frame also enables visualization of delayed or reduced growth. This was demonstrated in the present study, e.g., for the 8 Gy irradiated samples, where UM-SCC-74A cells resumed growth after 3 weeks, whereas, UM-SCC-74B growth TABLE 1 | Significantly different metabolites (p < 0.05) between non-irradiated cell lines UM-SCC-74A and UM-SCC-74B are presented as fold changes (95% confidence).


The p-values were calculated using ANOVA followed by a Tukey test and the fold changes were calculated using Fieller's theorem for a 95% confidence interval. A positive fold change indicates a higher level in 74A UM-SCC-74A as compared to 74B UM-SCC-74B

was resumed already after 2 weeks (**Figures 2A,B**). Results in the long-term growth inhibition assay (**Figure 2**) verified the clear differences in radiosensitivity at higher doses observed in the clonogenic survival assays (**Figure 1A**). Also at 2 Gy, UM-SCC-74A cells demonstrated lower cell growth and viability than UM-SCC-74B cells, although not as pronounced as in the clonogenic survival assay, potentially reflecting the different time-frames, culture conditions, and assessed parameters in the different assays. The long-term growth assay also visualized the different doubling times of the cell lines, where UM-SCC-74B demonstrated a shorter doubling time than UM-SCC-74A. This is consistent with clinical experiences, where the recurrence of a tumor is often faster growing and more aggressive. The differences in growth rates were also in line with the clearly separated PCA scores plot (**Figure 3**), and reflected in some of the metabolic differences between the unirradiated cell lines (**Table 1**), discussed more in detail below. The fact that the STR profiles were the same, whereas factors such as growth rates, radiation resistance, and metabolic profiles were not, also demonstrates the advantage of complementing genetic data with cell assays and metabolomic profiling.

The 3D cell culture assay visualizes radiation effects on viability and cell growth in the same time frame as clonogenic survival. However, in the 3D assay, cells are cultured in a more TABLE 2 | Significantly altered metabolites (p < 0.05) in the cell line UM-SCC-74A between cells irradiated with 2 Gy [assessed 4 h (rapid) or 24 h (intermediate) after irradiation] vs. non-irradiated cells (control) from the analysis in negative ionization mode are presented as fold changes (95% confidence intervals).


The p-values were calculated using ANOVA followed by a Tukey test and the fold changes were calculated using Fieller's theorem for a 95% confidence interval. A positive fold change indicates a higher level in the radiated cells as compared to the control.

in vivo like situation. In addition to cell-to-cell communication, important factors such as oxygen and nutritient gradients are mimicking the environment of small non-vascularized metastases (63). Results from the 3D cell culture assay were in line with the clonogenic survival assay, where both assays demonstrated approximately twice as many surviving UM-SCC-74B cells than UM-SCC-74A cells 10–14 days after 2 Gy irradiation (**Figures 1B,C**), further validating the difference in radiosensitivity between the cell lines. Consequently, we conclude that in all three independent assays UM-SCC-74A cells were clearly more affected by radiation in terms of cell viability and growth than UM-SCC-74B cells. This was also in line with the apoptosis assay, demonstrating increased levels of cleaved PARP1 in irradiated UM-SCC74A cells, but not in UM-SCC-74B cells (**Supplemental Figure 1**). Thus, we concluded that the cell lines were a suitable model system for subsequent complex metabolic evaluations of radiation response.

#### Unirradiated Cells Differed in Nicotinamide and Nicotinic Acid Metabolism and Purine Metabolism Pathways

Even though unirradiated UM-SCC-74A and UM-SCC-74B cells were identical according to STR genotyping, metabolic profiles differed clearly between the two unirradiated cell lines, as they were distinctly separated in the PCA scores plot (**Figure 3**). The nicotinamide and nicotinic acid metabolism pathways were indicated as important differences in the pathway analysis between the two cell lines, with metabolites such as nicotinamide, nicotinic acid, and nicotinic acid mononucleotide significantly different between the two control groups (**Figure 4**, **Table 1**). Nicotinamide and nicotinic acid mononucleotide were found at significantly higher levels in unirradiated UM-SCC-74B cells compared to UM-SCC-74A cells, while nicotinic acid was found in higher levels in unirradiated UM-SCC-74A cells compared TABLE 3 | Significantly altered metabolites (p < 0.05) in the cell line UM-SCC-74B between cells irradiated with 2 Gy [assessed 4 h (rapid) or 24 h (intermediate) after irradiation] vs. non-irradiated cells (control) from the analysis in positive ionization mode are presented as fold changes (95% confidence).


(Continued)

TABLE 3 | Continued


The p-values were calculated using ANOVA followed by a Tukey test and the fold changes were calculated using Fieller's theorem for a 95% confidence interval. A positive fold change indicates a higher level in the radiated cells as compared to the control.

to UM-SCC-74B cells (**Figure 4**). This indicates a lower rate of biosynthesis of nicotinic acid mononucleotide and ultimately in nicotinamide adenosine dinucleotide (NAD+) in the UM-SCC-74A cell line. The other main pathway that was found differentiating between the two non-irradiated cell lines was purine metabolism. Xanthine, inosine, and guanosine were all found at significantly higher levels in the UM-SCC-74A cells than in the UM-SCC-74B cells. The higher levels of inosine and xanthine in UM-SCC-74A may indicate a higher purine degradation in the UM-SCC-74A cell line. In UM-SCC-74B, 5 phosphoribosylamine was found in higher concentration, which is an intermediate in de novo purine synthesis, and might indicate a higher rate of de novo synthesis of purine nucleotides in the UM-SCC-74B cells. Higher rates of purine de novo synthesis have previously been linked to increased growth rates (70), and is in line with the results from the growth inhibition assay, where unirradiated UM-SCC-74B cells demonstrated a shorter doubling time than UM-SCC-74A cells (**Figure 2**).

#### Altered Tryptophan and Serotonin Metabolism in Irradiated UM-SCC-74A Cells

In general, very few metabolic alterations were observed after irradiation in UM-SCC-74A cells compared to UM-SCC-74B cells. In the UM-SCC-74A cell line the most important radiation induced metabolic perturbation was the alteration in tryptophan metabolism. Tryptophan was found at higher levels as compared to the control 4 h after radiation but was reduced to the same levels as the control 24 h after radiation. Tryptophan has two main metabolic fates, it can be used in the biosynthesis of quinolinic acid, which is a precursor to nicotinic acid mononucleotide, or it can be converted to serotonin via 5 hydroxytryptophan. The increase in tryptophan levels 4 h after irradiation could be an indication of a shift in the metabolism toward an increase in quinolinic acid and nicotinic acid mononucleotide biosynthesis in the UM-SCC-74A cell line. However, this assumption could not be verified, as the nicotinic acid mononucleotide levels were not altered neither 4, nor 24 h after irradiation in UM-SCC-74A, and quinolinic acid was not detected in this analysis. Moreover, the serotonin metabolite 5-hydroxyindoleacetic acid (5-HIIA) was found increased 24 h after irradiation the UM-SCC-74A cell line, which might suggest an increase in serotonin degradation to 5-HIIA. Serotonin has been shown to exhibit growth stimulatory effects on several types of carcinoma and other tumor cells, and to play a role in radiation-induced bystander effect (71). In previous studies, serotonin concentrations in culture media have been shown to be depleted after exposure of cells to radiation, suggesting that serotonin may be bound by membrane receptors after irradiation, thus facilitating calcium entry into cells, production of ROS and activation of apoptosis pathways (71, 72). Serotonin can also act as both vasoconstrictor and promote angiogenesis in solid tumors, and could therefore have important effects on the tumor hypoxia which have been linked to radiation sensitivity previously (73).

#### Decreased Levels of NAD+ and Increased NAD+ Turnover Suggest Initiated DNA Repair Mechanisms in Irradiated UM-SCC-74B Cells

In the UM-SCC-74A cell line, no investigated metabolites in the nicotinamide and nicotinic acid metabolism demonstrated significantly altered levels after irradiation. This is in contrast to the UM-SCC-74B cell line, where the metabolites nicotinamide, 1-methylnicotinamide, nicotinamide ribotide, nicotinic acid mononucleotide, NADH as well as NAD+ were all almost depleted 24 h after irradiation (**Figure 4**). This demonstrates inherent differences in nicotinamide and nicotinate metabolism in the two cell lines both before and after irradiation. While the decrease in NAD+ could be due to an increase in turnover from NAD+ to NADP/NADPH to maintain the redox status in the cells, regenerate glutathione, and an increased biosynthesis (45), the fact that NADP+ and NADPH were not detected in the analysis contradicts this explanation. Thus, a more likely explanation is that the decreased levels of NAD+ and increased NAD+ turnover in UM-SCC-74B cells after irradiation is due to an increased ADP-ribosylation by poly(ADP-ribose)polymerases (PARPs) to initiate DNA repair mechanisms (74–77). The PARP proteins are the main consumers of NAD+ during genotoxic stress, and the levels of NAD+ can be depleted following ionizing radiation to meet the demands for DNArepair signaling by ADP-ribosyl (74–76). PARP1 is a member of the PARP family of enzymes. Its primary function is to detect and repair DNA damage, where amplified PARP1 activity results in high NAD+ consumption. This process is blocked by rapid cleavage and inactivation of PARP1 by the action of caspases. In the present study, the cleaved PARP1 assay demonstrated that the levels of inactivated PARP1 were increased in irradiated UM-SCC-74A cells, whereas levels in UM-SCC-74B cells were unchanged after radiation (**Supplemental Figure 1**), which could further support this hypothesis. Moreover, previous studies have demonstrated that inhibition of PARP or PARP silencing increase radiosensitivity (78–85). Van Vuurden et al. (81) observed an overexpression of PARP1 as well as a radiosensitizing effect by the PARP1 inhibitor olaparib in pediatric medulloblastoma, ependymoma, and high grade glioma cell lines. Similarly, Owonikoko et al. (84) investigated the PARP1 inhibitor veliparib in combination with DNA-damaging treatments including radiation in small cell lung cancer cells, and found that veliparib sensitized some cells to DNA damaging treatment. Both Godon et al. (80) and Noël et al. (82) found that the radiosensitizing effect of PARP1 inhibition by 4-amino-1,8-naphthalimide (ANI) was cell cycle dependent, and that rapidly growing cells with high fraction of cells in the S-phase were more sensitive to PARP-inhibition in combination with radiation. Consequently, our data suggest that PARP inhibition may be especially suitable to overcome the radioresistance in the radioresistant UM-SCC-74B cells, and should be evaluated in future studies.

### Increased Levels of Guanosine and Adenosine Indicate a Functioning Purine Salvage Pathway and More Efficient ROS Protection in Irradiated UM-SCC-74B Cells

In UM-SCC-74B, the levels of adenosine, guanosine and guanine were increased 24 h after irradiation, while the levels of adenosine monophosphate (AMP) and guanosine monophosphate (GMP) were decreased (not significantly) 24 h after irradiation. This indicates a functioning purine salvage pathway with an increased degradation of the purine nucleotides to the corresponding nucleosides and nucleobases. This is in line with previous studies, where guanosine have been shown to protect DNA in vitro from oxidative damage induced by reactive oxygen species (ROS), and to serve as a radioprotector (86). Adenosine has demonstrated the same effect as guanosine however not as strong, while the pyrimidine nucleobases had the opposite effect (86). In contrast, this pattern was not observed in the UM-SCC-74A cell line, where altered (although not significant) levels of inosine, hypoxanthine, and xanthine instead might indicate an increased purine nucleotide catabolism through the transformation of inosine to hypoxanthine to xanthine. This data suggests that the UM-SCC-74B cells may have enabled a more efficient ROS protection through increased levels of guanosine and adenosine after irradiation.

### Increased Levels of SAM Indicate Alterations in the DNA Methylation in Irradiated UM-SCC-74B Cells

The pathway of cysteine and methionine metabolism was altered in the UM-SCC-74B cell line after irradiation, in contrast to UM-SCC-74A cells (**Figure 5**). S-adenosylmethionine (SAM) is a methyl donor, involved in almost all methylation reactions in the cells, such as DNA and histone methylation but also methyl transfer reactions to proteins, lipids, and secondary metabolites (3, 4). SAM is also an important component in many trans-sulfuration reactions and aminopropylation reactions (87). After irradiation, there was an increase in both methionine and SAM in UM-SCC-74B cells, suggesting an increased turnover from homocysteine to methionine, which can be driven by either the folate cycle or methyl-group transfer by betaine (87). As the levels of glycine and serine were not found altered after irradiation and neither were the levels of 5 methyltetrahydrofolate (data not shown), data suggest that the turnover from homocysteine to methionine was not driven by the folate cycle but rather betaine (88). This was supported by the decreased levels on choline, the main precursor of betaine, 24 h after irradiation. Moreover, the levels of glutathione in UM-SCC-74B cells were lower after irradiation, indicating that homocysteine is not converted to cystathionine, but mainly reconverted to methionine, since cysteine is the rate-limiting precursor for biosynthesis of glutathione (87). Consequently, our data suggest that irradiated UM-SCC-74B cells mobilized the homocysteine-methionine cycle, thereby increasing the synthesis of SAM to avoid radiation induced DNA-hypomethylation (89). This is in line with previous studies, linking global changes in DNA methylation to ionizing radiation (90–93), and to the development of radioresistance in oral squamous cell carcinoma (93) and lung cancer (92). Kim et al. (92) found that several key regulators in radiosensitivity in lung cancer were epigenetically controlled by CpG methylation. Batra et al. (94) demonstrated that methyl donor deficient diets increased the irradiation induced metabolic stress in mice and decreased DNA methyl transferase activity, indicating decreased DNA methylation (94, 95). Moreover, Batra et al. (96) demonstrated that L-methionine supplementation might help to alleviate radiation induced loss of genomic DNA methylation in murine liver tissue. This could open up for new possibilities of sensitizing tumors to radiation treatment and in the future avoid radio-resistance in radiation treatment.

## CONCLUSION

Our data strongly implicates that the radioresistant cells changed their metabolism to control the redox status, DNA repair as well as DNA methylation. A lot of preclinical efforts have over time been devoted to the development of strategies to sensitize cancer cells to radiation therapy. This study was a first step in the understanding of which metabolic pathways in SCC that were important for the differences in radiosensitivity between the two cell lines UM-SCC-74A and UM-SCC-74B. The elucidation of the mechanisms behind radioresistance could lead to better prediction of radiation treatment outcome or possibilities to sensitize tumors to radiation. However, all

metabolites and metabolic pathways investigated in this study all require further investigation as to whether they will be able to pose as targets for prediction of radiation response or to enhance radiation sensitivity.

### DATA AVAILABILITY

The datasets for this study are available on request.

### AUTHOR CONTRIBUTIONS

EL, IE, ME, JH, TA, CP, GL, and MN designed the study. EL, IE, ME, JH, TA, MH, CP, GL, and MN contributed to data analysis and interpretation, and revised the manuscript. EL and IE contributed to experimental studies and drafted the manuscript.

#### FUNDING

This study was supported by grants from the Swedish Cancer Society (CAN 2018/494, CAN 2015/1080, and CAN 2015/385), and the Swedish Research Council (2013- 30876-104113-30). Targeted funding for development of metabolomics research was obtained from the disciplinary domain of medicine and pharmacy at Uppsala University, and ALF-grants were obtained from Uppsala University Hospital.

### ACKNOWLEDGMENTS

The authors would like to acknowledge Sara Häggblad Sahlberg and Jörgen Carlsson for kind manuscript feedback, and Christina Atterby, Anja Mortensen, and Tabassom Mohajer Shojai for help with cell culturing, cell assays, and harvesting.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00825/full#supplementary-material

Supplemental Figure 1 | Levels of cleaved PARP1 in UM-SCC-74A (black bars) and UM-SCC-74B (gray bars) cells 12 h after 0 or 2 Gy irradiation, measured by ELISA. Error bars represent the standard error of mean N = 3.

Supplemental Table 1 | STR analysis of non-irradiated UM-SCC-74A and UM-SCC-74B.

## REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Lindell Jonsson, Erngren, Engskog, Haglöf, Arvidsson, Hedeland, Petterson, Laurell and Nestor. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Exosomes in Cancer Radioresistance

Jie Ni 1,2, Joseph Bucci 1,2, David Malouf 1,3, Matthew Knox 1,2, Peter Graham1,2 and Yong Li 1,2,4 \*

*<sup>1</sup> Cancer Care Centre, St. George Hospital, Sydney, NSW, Australia, <sup>2</sup> St. George and Sutherland Clinical School, Faculty of Medicine, University of New South Wales (UNSW) Sydney, Kensington, NSW, Australia, <sup>3</sup> Department of Urology, St. George Hospital, Sydney, NSW, Australia, <sup>4</sup> School of Basic Medical Sciences, Zhengzhou University, Zhengzhou, China*

Radiation is a mainstay of cancer therapy. Radioresistance is a significant challenge in the treatment of locally advanced, recurrent and metastatic cancers. The mechanisms of radioresistance are complicated and still not completely understood. Exosomes are 40–150 nm vesicles released by cancer cells that contain pathogenic components, such as proteins, mRNAs, DNA fragments, non-coding RNAs, and lipids. Exosomes play a critical role in cancer progression, including cell-cell communication, tumor-stromal interactions, activation of signaling pathways, and immunomodulation. Emerging data indicate that radiation-derived exosomes increase tumor burden, decrease survival, cause radiation-induced bystander effects and promote radioresistance. In addition, radiation can change the contents of exosomes, which allows exosomes to be used as a prognostic and predictive biomarker to monitor radiation response. Therefore, understanding the roles and mechanisms of exosomes in radiation response may shed light on how exosomes play a role in radioresistance and open a new way in radiotherapy and translational medicine. In this review, we discuss recent advances in radiation-induced exosome changes in components, focus on the roles of exosome in radiation-induced bystander effect in cancer and emphasize the importance of exosomes in cancer progression and radioresistance for developing novel therapy.

*James William Jacobberger, Case Western Reserve University, United States Evagelia C. Laiakis, Georgetown University, United States*

*Radboud University Nijmegen Medical*

\*Correspondence: *Yong Li y.li@unsw.edu.au*

Edited by: *Paul N. Span,*

*Centre, Netherlands* Reviewed by:

#### Specialty section:

*This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology*

Received: *10 July 2019* Accepted: *21 August 2019* Published: *06 September 2019*

#### Citation:

*Ni J, Bucci J, Malouf D, Knox M, Graham P and Li Y (2019) Exosomes in Cancer Radioresistance. Front. Oncol. 9:869. doi: 10.3389/fonc.2019.00869* Keywords: cancer, exosomes, radiotherapy, bystander, radioresistance

#### INTRODUCTION

Cancer is a major health burden. Radiotherapy (RT) is widely used in more than 50% of localized cancer patients (1), and is a critical and inseparable component of comprehensive cancer treatment and care (2). In addition, RT is often combined with surgery, chemotherapy, and more recently, immunotherapy (3). Despite progress made in radiation delivery approaches and precision medicine, tumor therapeutic resistance and recurrence frequently occur in clinical settings.

Radioresistance is a complicated biological process associated with abnormal DNA damage response (DDR), apoptosis, autophagy, gene mutations, cell cycle checkpoint, and deregulated signaling pathways (4). It leads to poor prognosis in cancer patients and represents a major clinical obstacle for RT, which ultimately leads to tumor relapse and metastasis (5). Tumor microenvironment is an important factor affecting tumor progression and therapeutic response (6, 7). It was reported radioresistance is highly associated with tumor microenvironment (8, 9).

Exosomes are important components and regulators of the tumor microenvironment. Exosomes are small extracellular vesicles (40–150 nm) secreted by different cells. Exosomes are representative of the original cells and can reflect a regulated sorting mechanism (10). These vesicles are composed of proteins (receptors, transcription factors, enzymes), nucleic acids [nuclear DNA, mitochondrial DNA (mtDNA), mRNA, microRNA (miRNA), long non-coding RNA (lncRNA), and circular RNA (circRNA)], and lipids (11–14). Exosome cargos such as nucleic acid and proteins from originating tumor cells communicate with neighbor cells or recipient cells, resulting in cancer progression and recurrence (15, 16).

Stressful conditions affect exosome secretion, composition, abundance, and potential binding on recipient cells. It was reported that different physiological and environmental conditions could alter the composition of exosomes shed from cells (17). Accumulating evidence indicates an increased release of exosomes after exposure to RT and the altered contents from donor cells were more oncogenic (18–21). Several recent studies have confirmed irradiated cells were involved in radiation-related communication between cells (20, 22, 23).

Despite the progress that has been made in exosome-mediated functions in in vitro models, there are still many challenges to be faced in exosomes and radiation oncology research. Here, we review recent advances in the radiation-induced exosome changes, discuss the roles of exosome in radiation-induced bystander effect (RIBE) in cancer and emphasize the importance of exosomes in cancer radioresistance and progression for the development of novel therapeutic strategies.

#### RADIATION-INDUCED EXOSOME CHANGES IN CANCER RADIOTHERAPY

Cellular stress affects the composition and abundance of exosomes, as well as their potential impact on the recipient cells. Exosomes are a major environmental factor for cellular stress, and radiation can enhance the release of exosomes and affect exosome-based intercellular communication, which has been observed in various types of normal and tumor cell lines (18, 19, 24).

An increased level of exosomal CD276 was observed in the irradiated and senescent 22RV1 prostate cancer cell line, suggesting that this marker may provide a noninvasive way to monitor the efficacy of RT for prostate cancer patients (25). It was reported that circulating Hsp72 level was increased after radiation exposure in prostate cancer mouse xenografts and clinical sample; and that the exosomes containing Hsp72 are a possible contributor, leading to pro-inflammatory cytokine production and immune modulation (26). Khan et al. found that the level of exosomal survivin was increased after proton irradiation in HeLa cells and the rate of exosome secretion was not influenced (27). These findings suggest exosomal survivin may be associated with cancer recurrence after RT and could be a potential therapeutic target for preventing cervical cancer progression.

Radiation also alters the molecular composition within the exosomes. In one study, it was found radiation increased levels of exosomal connective tissue growth factor (CTGF) and Insulin Like Growth Factor Binding Protein 2 (IGFBP2) proteins, both of which are important in cell migration. By transferring CTGF mRNA, exosomes from irradiated cells were found to upregulate the migration-related signaling molecules including neurotrophic tyrosine kinase receptor type 1, focal adhesion kinase, Paxillin, and proto-oncogene tyrosine-protein kinase Src in recipient cells (18). In another study, exosomes released from the irradiated head and neck squamous carcinoma cell (HNSCC) FaDu cells demonstrated a distinctive protein expression profile compared to those from non-irradiated cells (20). Interestingly, most of the proteins that are specifically overexpressed in exosomes are those involved in transcription, translocation, and cell division, indicating that exosomal cargo is reflective of radiation-induced changes in cells (20). Similarly, exosomes derived from irradiated HNSCC BHY cells were found to be associated with not only immunity but also cell adhesion and motility, the underlying molecular mechanism being enhanced AKT-signaling triggered by the exosomal proteins (28). Using a shotgun liquid chromatography–tandem mass spectrometry (LC-MS/MS) approach, Abramowicz et al. found 472 exosomal proteins that are significantly affected by ionizing radiation (IR) in HNSCC UM-SCC6 cells and identified their role in mediating the cellular response to IR (29). Zhao et al. recently identified 63 upregulated and 48 downregulated circular RNAs from the exosomes of radioresistant glioma cells compared with those in control cells. Using qRT-PCR, they found circATP8B4 from radioresistant exosomes of glioma cells may be transferred to radiation-naïve cells and promoted cell radioresistance by acting as a microRNA(miR)-766 sponge (30).

The role of exosomes in radiation has garnered increasing attention in recent years. One study screened 752 exosomederived miRNAs of locally advanced non-small cell lung cancer (NSCLC) patients and demonstrated that increased radiation dosage reduced miR29a-3p and miR150-5p expression (31), indicating that circulating exosomal miRNAs could help predict RT toxicity. In another study evaluating the outcome of HNSCC patients treated with chemoradiation therapy (CRT), exosomes from pooled plasma samples of patients who had complete or incomplete responses to CRT were screened. They identified a distinctive expression pattern of proteins between patients who had a complete response and who did not (32). In a recent Phase I clinical trial, 18 HNSCC patients receiving a combination of cetuximab, ipilimumab and RT were serially monitored for tumor cell-derived exosomes and T cell-derived exosomes. The results suggested tumor cell-derived exosomes and T cell-derived circulating exosomes instead of immune cells were suitable for monitoring of patients' responses to oncological therapy, supporting the potential role of exosomes as a non-invasive tumor and immune cell biomarkers in cancer (33).

The exosomes in RT research is stating to move from bench to bedside. However, the main limitation is the sample numbers tested from patiensts are still low and the preliminary results obtained need to be further validated in a large cohort of patients. Another challenge is the methods used for detecting exosome contents (biomarkers) such as proteomics and next generation sequencing need to be further optimimised to obtain the maximum number of interesting pontential biomarkers for verification. Although advances in exosome-based biomarkers such as proteins and miRNAs highlight an optimistic outlook for RT, the great progress has not been achieved due to limited reports in clinical research. Future direction in this area should move on the translational research and clinical trials.

In summary, the studies so far on radiation-induced changes in exosome composition have mainly been confined to their proteome contents using in vitro cancer cell line models (Summarized in **Table 1**), and there is a big gap in understanding how these changes are regulated, with the hope of translating into their functional importance. Future studies in this area should focus on (1) investigating the mechanisms of how the components of exosomes from donor cells after radiation are transferred to receipt cells; (2) establishing in vivo animal models for further studying the changes of exosomal components on the effect of RT; (3) investigating whether the changes of exosomal components could be used as biomarkers to evaluate the efficacy of RT; and (4) investigating whether the changes of exosomal components could be used as useful therapeutic targets to overcome radioresistance and improve the current RT.

### THE BYSTANDER EFFECTS BY RADIATION-INDUCED EXOSOMES

Radiation affects not only its direct targeted cells but also non-irradiated neighbors. This is evidenced by RIBE that cells that were not exposed to radiation exhibit effects as a result of intercellular communication. RIBE can also lead to biological changes in bystander cells and tissues, including chromosomal rearrangement, genomic instability, DNA damage, gene expression alteration, and apoptosis (34).

Numerous studies have demonstrated a large and complex interconnected web of mechanisms that contribute to the generation of RIBE, including reactive oxygen species (ROS), cytokines, free radicals, immune system, and epigenetic modulators (35–37). As previously reviewed more than a decade ago, it was believed that RIBE was mediated by both "a soluble secreted factor" and the cell-to-cell gap junction (38). It was not until recently that the important role of exosomes mediating RIBE had been recognized (39). Exosomes shed by irradiated cells are putatively involved in different aspects of the systemic response to IR, including the RIBE (40–42). Jella et al. showed


*CaP, prostate cancer; CTGF, connective tissue growth factor; HNSCC, head and neck squamous cell carcinoma; GBM, glioblastoma; IGFBP2, insulin-like growth factor binding protein 2, LC-MS/MS, liquid chromatography-tandem mass spectrometry; N/A, not applicable; SEC, size-exclusion chromatography; UC, ultracentrifugation.*

that exosomes released from gamma-irradiated keratinocyte HaCaT cells induced increased cell death and ROS production in non-irradiated cells (24).

Mechanistically, IR elicits a set of dysregulated proteins and nucleic acids within the cell. These effectors, such as proteins, miRNAs and mtDNAs, are packaged into exosomes during their formation, which are then released to the extracellular environment. These cargos within the radiation-targeted cellreleased exosomes subsequently get access to the adjacent cells as a result of exosome migration and internalization, and prompt RIBE in the non-targeted distant cells (21, 43–45).

These components of exosomes have different functions in RIBE, such as regulation of inflammation and modulation of DDR (46).

One report demonstrated that the production and release of exosomes following radiation-induced DNA damage were regulated by the p53 pathway (47). Tian et al. demonstrated that miR-21, a well-established DDR-related miRNA, played a mediating role in bystander DNA damage since it elevated ROS levels and increased the double-strand break (DSB) marker p53 binding protein 1 (53BP1) foci in non-irradiated cells (48). The role of miR-21 in RIBE was also validated by Yin et al. and Xu et al. using different experiment models (21, 49). Exosomal miR-1246 was also found to act as a messenger and contribute to DNA damage by directly repressing the DNA Ligase 4 (LIG4) gene (22). In combination, Yentrapalli et al. found that proteins such as afamin and serpine peptidase F1 along with miRNAs including miR-204-5p, miR-92a-3p, and miR-31-5p, play an important role in inducing and regulating RIBE (50). Another group also demonstrated that exosomal proteins and RNAs could mediate short- and long-term RIBE in human MCF-7 breast cancer cells (19, 51), implying that proteins and miRNAs may work synergistically during this process.

Moreover, exosomal miRNAs are able to travel remotely to influence cellular functions and regulate the niche-host reaction in targeted or non-targeted cells. It is worth mentioning that the role of a miRNA as either a positive or negative regulator in the RIBE varies from cell type to cell type. Interestingly, cells affected by RIBE and their progeny also showed the ability to secrete exosomes, and this cascade could potentially lead to a delayed RIBE-related inflammatory response (19). We have summarized exosome-mediated RIBE studies in cancer RT in **Table 2**. The mechanistic diagram demonstrating effects of radiation-inducible exosomal miRNAs and proteins in mediating RIBE is shown in **Figure 1**.

In the therapeutic arena, on the one hand, RIBE is harmful to normal tissues, but on the other hand may be beneficial to induce non-irradiated cancer cell death during the treatment. RIBEs have critical implications in cancer RT. As direct effects of RT and RIBE are mechanistically distinctive and therefore it is important to develop different types of drugs to specifically target each mechanism, such as novel radiosensitisers to upregulate RIBE to kill more adjacent tumor cells; or adjuvant inhibitors to minimize the RIBE-induced systemic toxicity after RT.


*h, hours; HNSCC, head and neck squamous cell carcinoma; UC, ultracentrifugation.*

Moreover, although in vitro studies bring promise for using exosomes and their cargo to regulate DDR and RIBE, it remains unclear whether these findings can be translated into the clinical setting. In future studies, the mechanisms of RIBE need to be deeply investigated, and in vivo animal models and clinical samples should be applied.

#### EXOSOMES IN CANCER PROGRESSION AFTER RADIOTHERAPY

RT itself may increase the motility of surviving cancer cells, evidenced in glioblastoma (GBM), lung cancer and HNSCC, thus facilitating the spread of the tumor to local and distant sites (53–55). Radiation-induced exosomes have recently found to be an accomplice in promoting tumor cell motility and assisting in the pre-metastatic niche formation, the effectors again being the exosomal cargo incorporated by the recipient cells. Arscott et al. showed that radiation-derived exosomes enhanced U87MG GBM cell migration in co-culture (18). Using a wound healing assay, Mutschelknaus et al. found a pro-migratory role of exosomes in boosting the migratory capacity of BHY and FaDu HNSCC cells, in a dose-dependent and AKT-dependent manner (28). Apart from cell motility, angiogenesis also plays a crucial role in RT and tumor metastasis. Zheng et al. recently demonstrated that in lung cancer, the exosome-induced proangiogenesis effect was enhanced when the A549 and H1299 lung cancer cells were exposed to IR, and the miR23-mediated phosphatase and tensin homolog (PTEN) downregulation played an important role in this process (56). These findings indicate that radiation-induced exosomes function as a driver of cancer progression and metastasis during RT, and may represent a putative target to improve RT strategies.

The recent striking responses to immune-checkpoint inhibitors in the treatment of melanoma and solid tumors are paradigm-shifting and stirring up much research interest in the combination of immune-checkpoint inhibitors with RT. There is a close association between radiation response and immunity (57). RT plays either an immune-suppressive role (due to the sensitivity of leukocytes) or an immune-stimulatory role, evidenced by enhancing several antigen processing and presentation pathways (58, 59). Conventionally, the immune stimulation is believed to be T cell-mediated, but it is not until recently have we found that T cell-derived exosomes also plays a role in promoting esophageal cancer metastasis, via activation of epithelial-mesenchymal transition (EMT), β-catenin, and NF-κB/SNAIL pathways (60). This study provides a rationale for targeting exosomes during the synergy between RT and immunotherapy, but there are still quite a few unsolved riddles in this synergy, such as the sequencing, dose, and fractionation. Clearer mechanistic understandings of RT's immune-stimulatory role and further investigations on exosomes' functions in immune modulation are needed to expand this research field.

#### EXOSOMES IN RADIORESISTANCE

Despite the recent advances in RT, many cancer patients, especially the locally advanced ones, failed radiation treatment (radioresistance), leading to a local recurrence or even distant metastasis. As previously reviewed (4), radioresistance can arise either from genetic or phenotypic changes within the tumor or as a result of the tumor stromal and microenvironment protecting the tumor against IR. Exosomes are one of the key components of the tumor microenvironment and increasing evidence suggests that they play a significant role in facilitating the development of radioresistance.

The key players in exosome-mediated radiosensitivity are found to be exosomal non-coding RNAs, proteins, and the crosstalk with survival and apoptotic pathways. Mutschelknaus et al. recently demonstrated that exosomes derived from irradiated HNSCC cells transmitted pro-survival signals to recipient cells via exosome cargos (52). In breast cancer, RNAs within exosomes were found to regulate radioresistance via an antiviral STAT1/NOTCH3 pathway (61). Radiationinduced exosomal miR-208a increased the proliferation and radioresistance via targeting p21 with activation of the AKT/mTOR pathway in lung cancer (62). In GBM, Mrowczynski et al. recently discovered that exosomes could enhance cell survival to radiation exposure by increasing levels of oncogenic miRNAs, mRNAs and pro-survival pathway proteins and at the same time decreasing levels of tumor-suppressive miRNAs and mRNAs (23). Recently, other exosomal non-coding RNAs were also found to be involved in the promotion of radioresistance, such as long non-coding RNA AHIF in glioblastoma (63), and circATP8B4 in glioma (30).

While exosomes were reported to play an important role in the promotion of cancer radioresistance, other reports were controversial. Wang et al. recently reported that autocrine secretions enhance the radioresistance of H460 NSCLC cell line in an exosome-independent manner and that these secretions mainly affect the DNA repair process (64). In another study, it was shown that exosomes derived form mesenchymal stem cells (MSCs), combined with RT, enhance RT-induced cell death in tumor and metastatic tumor foci in a melanoma mouse model. The finding provides a rationale to use MSCs-derived exosomes as an adjuvant to support and complement RT (65). These data indicate the role of exosomes in cancer radioresistance is complicated and could be affected by many factors such as tumor type, tumor microenvironment, experimental methods or different combinations of therapies.

Cancer is highly heterogeneous and includes a small subset of cells that possess the capacity of self-renewal and differentiation, referred to as cancer stem cells (CSCs) (66). CSCs are inherently more resistant to radiation than ordinary cancer cells, and more likely to survive after being exposed to RT. On one hand, these surviving CSCs may release exosomes to transfer resistant or refractory phenotypes to recipient cells, limiting the treatment efficacy. It was found that lncRNA H19 in exosomes derived from CSCs induced angiogenesis in hepatocellular carcinoma (67). Exosomes released from prostate and breast CSCs were also capable to induce autophagy (68), which has been shown to modulate sensitivity of cancer to RT (69). On the other, CSCs can also be eliminated via being reprogrammed into nontumorigenic cells, using exosomes derived from adipose-derived stem cells (70). It not only re-justifies that the CSCs have to be eradicated during cancer therapy, but also preludes the usage of exosomes with modified surface or cargos to target CSCs.

In summary, all the data indicate that radiation-derived exosomes play important roles in cancer radioresistance through re-programmed cargos and intercellular communication. mRNAs, non-coding RNAs and signaling pathway proteins

are closely related in exosome-associated radioresistance. CSC-associated exosomes are also a potential player in radioresistance and should be deeply investigated in the future study. The potential mechanisms of exosomes in radioresistance are shown in **Figure 2**.

#### CONCLUSIONS

Exploiting the biological functions of exosomes is intriguing, as they provide a snapshot of the entire tumor, transfer molecules intercellularly and can be used as a therapeutic target. In order to achieve these goals, isolation of exosomes should be standardized and optimized.

Exosomes in radiation research is a new and developing area. Radiation affects not only the production of exosomes but also the composition within, which makes exosomes an ideal prognostic and/or predictive biomarker to monitor the radiation response.

The radiation-altered exosomal cargos can be taken up by recipient cells, thus exerting various biological functions to impact radiosensitivity, and it is also worthwhile to note that this impact can be the results of synergistic or opposing effects of different exosomes for the sake of their heterogeneity. Due to the unique physical and biological features of exosomes, using functional exosomes to target CSCs and facilitate immunotherapy, is a promising avenue to be explored in RT, since exosomes are more stable, endogenous, and can be easily engineered or labeled.

There are still many challenges existing for exosome study. The future development of methodologies for exosome isolation

#### REFERENCES


and purification should be universal, precise, and suitable for clinical settings. Our experience is the combination of two or more approaches is a better choice for exosome isolation. In addition, the optimal sample source should be determined based on the cancer types and the preservation conditions of samples should be standardized. Furthermore, specific biomarkers of tumor exosomes should be screened and exploited for investigating the mechanisms of radioresistance.

Currently, knowledge of exosomes in cancer RT is in its infancy and mostly limited in in vitro studies. Further in vivo and clinical studies are warranted. Increasing knowledge of the biology of exosome and its cargo, along with standardized methods for exosome isolation and characterization will greatly contribute to a better understanding of mechanisms of exosome-mediated RT response and a good harnessing of exosomes as a therapeutic target, which might ultimately lead to the development of novel treatment strategies.

#### AUTHOR CONTRIBUTIONS

JN, JB, DM, MK, PG, and YL reviewed the literature, developed the structures, wrote the review, and approved the final manuscript.

### FUNDING

This work is mainly supported by the St. George Hospital Cancer Research Trust Fund and Prostate & Breast Cancer Foundation.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Ni, Bucci, Malouf, Knox, Graham and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Effects of Neoadjuvant Chemoradiation in Locally Advanced Rectal Cancer—The Impact in Intratumoral Heterogeneity

Fabiana Bettoni 1†, Cibele Masotti 1†, Bruna R. Corrêa<sup>1</sup> , Elisa Donnard<sup>1</sup> , Filipe F. dos Santos <sup>1</sup> , Guilherme P. São Julião<sup>2</sup> , Bruna B. Vailati <sup>2</sup> , Angelita Habr-Gama<sup>2</sup> , Pedro A. F. Galante<sup>1</sup> , Rodrigo O. Perez <sup>2</sup> and Anamaria A. Camargo1,3 \*

<sup>1</sup> Hospital Sírio Libanês, São Paulo, Brazil, <sup>2</sup> Instituto Angelita and Joaquim Gama, São Paulo, Brazil, <sup>3</sup> Ludwig Institute for Cancer Research, São Paulo, Brazil

#### Edited by:

Sunil Krishnan, University of Texas MD Anderson Cancer Center, United States

#### Reviewed by:

Edmund Mroz, The Ohio State University, United States Ira Ida Skvortsova, Innsbruck Medical University, Austria

#### \*Correspondence:

Anamaria A. Camargo aacamargo@mochsl.org.br

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 01 March 2019 Accepted: 13 September 2019 Published: 27 September 2019

#### Citation:

Bettoni F, Masotti C, Corrêa BR, Donnard E, dos Santos FF, São Julião GP, Vailati BB, Habr-Gama A, Galante PAF, Perez RO and Camargo AA (2019) The Effects of Neoadjuvant Chemoradiation in Locally Advanced Rectal Cancer—The Impact in Intratumoral Heterogeneity. Front. Oncol. 9:974. doi: 10.3389/fonc.2019.00974 Purpose: Intratumoral genetic heterogeneity (ITGH) is a common feature of solid tumors. However, little is known about the effect of neoadjuvant chemoradiation (nCRT) in ITGH of rectal tumors that exhibit poor response to nCRT. Here, we examined the impact of nCRT in the mutational profile and ITGH of rectal tumors and its adjacent irradiated normal mucosa in the setting of incomplete response to nCRT.

Methods and Materials: To evaluate ITGH in rectal tumors, we analyzed whole-exome sequencing (WES) data from 79 tumors obtained from The Cancer Genome Atlas (TCGA). We also compared matched peripheral blood cells, irradiated normal rectal mucosa and pre and post-treatment tumor samples (PRE-T and POS-T) from one individual to examine the iatrogenic effects of nCRT. Finally, we performed WES of 7 PRE-T/POST-T matched samples to examine how nCRT affects ITGH. ITGH was assessed by quantifying subclonal mutations within individual tumors using the Mutant-Allele Tumor Heterogeneity score (MATH score).

Results: Rectal tumors exhibit remarkable ITGH that is ultimately associated with disease stage (MATH score stage I/II 35.54 vs. stage III/IV 44.39, p = 0.047) and lymph node metastasis (MATH score N0 35.87 vs. N+ 45.79, p = 0.026). We also showed that nCRT does not seem to introduce detectable somatic mutations in the irradiated mucosa. Comparison of PRE-T and POST-T matched samples revealed a significant increase in ITGH in 5 out 7 patients and MATH scores were significantly higher after nCRT (median 41.7 vs. 28.8, p = 0.04). Finally, we were able to identify a subset of "enriched mutations" with significant changes in MAFs between PRE-T and POST-T samples. These "enriched mutations" were significantly more frequent in POST-T compared to PRE-T samples (92.9% vs. 7.1% p < 0.00001) and include mutations in genes associated with genetic instability and drug resistance in colorectal cancer, indicating the expansion of tumor cell subpopulations more prone to resist to nCRT.

Conclusions: nCRT increases ITGH and may result in the expansion of resistant tumor cell populations in residual tumors. The risk of introducing relevant somatic mutations

**54**

in the adjacent mucosa is minimal but non-responsive tumors may have potentially worse biological behavior when compared to their untreated counterparts. This was an exploratory study, and due to the limited number of samples analyzed, our results need to be validated in larger cohorts.

Keywords: neoadjuvant therapy, rectal cancer, intratumoral heterogeneity, clonal evolution, therapy resistance

#### INTRODUCTION

Neoadjuvant chemoradiotherapy (nCRT) is one of the preferred treatment strategies for locally advanced rectal cancer (1, 2). In addition to providing improved local disease control (particularly for patients with high-risk features for local recurrence), nCRT may allow the opportunity for organ-preservation among patients with complete clinical response (cCR). However, treatment of patients with low-risk for local recurrence with nCRT, for the sole purpose of organ-preservation, may result in significant detrimental functional and biological consequences among patients who do not achieve a cCR and still need radical surgery.

Intratumoral genetic heterogeneity (ITGH) was first described in the early 1980's (3). However, only recently, the full extent and the functional implications of ITGH have been appreciated (4). ITGH increases phenotypic variation and is currently seen as a critical mechanism underlying disease progression and therapeutic failure (5, 6). We and others have recently characterized the clonal architecture of locally advanced rectal tumors through multi-region whole-exome sequencing (WES). We demonstrated that non-treated rectal tumors exhibit a complex clonal architecture and significant, ITGH with 27–97% of exonic somatic mutations shared among all regions of an individual's tumor and with a mutant allele frequency (MAF) correlation between disparate tumor regions ranging from R <sup>2</sup> = 0.69–0.96 (7, 8). However, in these studies ITGH, was determined using a small number of tumors, and the effect of nCRT in shaping the mutational landscape and clonal architecture of rectal cancer was not addressed. Ultimately, tumors that do not respond completely to nCRT may acquire novel mutations and/or harbor selected tumor cells subpopulations compared to their baseline counterparts, leading to increased ITGH. Here, we evaluated ITGH in untreated rectal tumors and examined its association with disease stage and presence of lymph node metastasis. Also, we analyzed the impact of nCRT in the mutational landscape and ITGH of rectal tumors with incomplete response to nCRT and searched for somatic mutations introduced by nCRT in the adjacent normal irradiated mucosa.

#### MATERIALS AND METHODS

#### TCGA Data

To evaluate ITGH in rectal tumors and determine its association with disease stage and presence of lymph node metastasis, we analyzed whole-exome sequencing (WES) data from 79 rectal tumors from The Cancer Genome Atlas (TCGA) colorectal cohort (9, 10). Clinical, pathological and mutational data for all 79 rectal tumors are provided in **Supplementary Table 1**.

#### Rectal Cancer Patients and nCRT

Consecutive patients with rectal cancer (adenocarcinoma biopsyproven), located no more than 7 cm from the anal verge, and treated at the Angelita & Joaquim Gama Institute between 2007 and 2010, were eligible for the study. Only patients undergoing neoadjuvant chemoradiation were recruited for the study. Inclusion criteria included tumors with cT3/T4 or cN+ disease by radiological staging using magnetic resonance (MR) or endorectal ultrasound. Additionally, patients with cT2N0 otherwise considered for abdominal perineal excision or ultralow anterior resections were also referred for neoadjuvant chemoradiation and included in the study. Patients with metastatic disease were excluded from the study. Patients with clinical and radiological findings consistent with cCR were also excluded from the present study. Only patients with ≥10% residual cancer cells in the final pathological assessment were included in an attempt to avoid contamination of "incomplete responders" with "near-complete responders" that could eventually develop complete response if longer resting intervals had been used (**Table 1**). Macrodissection of tumor regions was performed whenever necessary prior to DNA extraction to increase sample purity. Tumor sections were required to contain at least 80% tumor cell nuclei with <20% necrosis for inclusion in the study. We have randomly selected one of the patients for the analysis of the nCRT effect on the normal rectal mucosa. Baseline staging and assessment of patients included digital rectal examination (DRE), proctoscopy and high-resolution MR. All patients underwent long-course chemoradiation therapy as described previously (11).

#### Assessment of Tumor Response

All patients were assessed for tumor response after at least 12 weeks from the last day of nCRT completion. Assessment of tumor response was performed with DRE, proctoscopy and MR. Patients with incomplete clinical response (clinical or radiological) were referred to immediate radical surgery.

#### Tumor and Blood Samples

Tumor samples were collected at diagnosis (PRE-T samples) and during surgical removal of the residual tumor (POST-T samples). Tumor-adjacent normal colonic mucosa exposed to nCRT (Nrx) was also collected from one patient immediately after tumor resection. Peripheral blood cells (BC) were collected from all patients before nCRT. This study was approved by the Ethics Committee of Hospital Alemão Oswaldo Cruz, São Paulo, Brazil (reference number 19/08) and was conducted in


TABLE 1 | Clinical and pathological data of rectal cancer patients submitted to nCRT.

LAR, low anterior resection; APR, abdominal perineal resection; TRG, tumor regression grade.

accordance with the Declaration of Helsinki. Patients provided written informed consent for tumor sample collection and study participation. Samples were processed as described in **Supplementary Material**.

#### Whole Exome Sequencing (WES)

Whole-exome libraries were prepared using SureSelect Human All Exon Target Enrichment kit (Agilent Technologies, Santa Clara, CA) and sequences were generated on a 5500xl SOLiD sequencing platform (Thermo-Fisher Scientific, Waltham, MA). Sequencing and coverage information are provided in **Supplementary Table 2**.

### SNV Calling and Somatic Mutation Detection

SNVs were identified using a combination of published and local pipelines (8, 12, 13) as described in **Supplementary Material**. Somatic point mutations were annotated using ANNOVAR (14).

#### ITGH and MATH Score

Significant changes in allele frequencies were used as a surrogate for changes in clonal structure and were detected using exact binomial tests. False discovery rate (FDR) was calculated using the p.adjust R function to correct for multiple-testing (15). ITGH was measured using the mutant allele tumor heterogeneity (MATH) score (16, 17). The MATH score is a quantitative measure of ITGH based on the Mutant Allele Frequency (MAF) distribution. MATH scores were calculated as the width ratio to the center of MAFs' distribution for somatic point mutations present within individual tumors. Due to the presence of genetically distinct cellular populations, heterogeneous tumors exhibit a broader allele frequency distribution compared to homogeneous tumors and higher scores. The MATH score is the most cost-effective method to compare ITGH among different tumors and to monitor global changes in ITGH (16–21).

#### Mutational Spectrum and Signatures

Mutational spectrum and signature analyses were performed, according to a previously published pipeline (22) and are detailed in **Supplementary Material**.

### Gene Set Enrichment Analysis

Gene set enrichment analysis (GSEA) was performed using the Molecular Signatures Database v5.0 (MSigDB) as detailed in **Supplementary Material** (23).

### RESULTS

#### ITGH in Rectal Tumors Is Associated With Disease Stage and Progression

To expand the characterization of ITGH in rectal tumors, we used WES data from 79 non-treated rectal tumors obtained from TCGA (**Supplementary Figure 1**). Since multi-region WES data was not available for TCGA samples, we used the MATH score to measure ITGH in these samples (16–21). Rectal tumors exhibit remarkable variability in ITGH, with MATH scores ranging from 18.2 to 66.7 (median = 40.1; mean = 41; first quartile = 31.1; third quartile = 49.8; **Figure 1A**). We also observed a significant positive association between MATH values, disease stage (Stage I/II median of 35.54 vs. stage III/IV median of 44.39, p = 0.047, Wilcoxon test, **Figure 1B**) and lymph node metastases (N0 median of 35.87 vs. N1+N2 median of 45.79, p = 0.026, Wilcoxon paired test, **Figure 1C**).

The MATH score is a simple and cost-effective method to measure ITGH that shows little influence of copy number variations (CNVs) and provides a first-order correction for the presence of contaminating normal tissue in tumor samples (16–21). To determine the influence of CNVs in our analysis we examined the correlation between MATH scores and total number of CNVs in all 79 individual tumors. As shown in **Supplementary Figure 2**, there is no significant correlation between higher MATH scores and aberrant CNV profiles in rectal tumors (cor = −0.03, p = 0.8, Pearson correlation). Likewise, we did not observe a significant association between the total number of CNVs, disease stage (Stage I/II median of 13.5 vs. stage III/IV median of 17.0, Wilcoxon test, p = 0.53, **Supplementary Figure 2**) or lymph node metastases (N0 median of 13 vs. N1+N2 median of 17, Wilcoxon paired test p = 0.42, **Supplementary Figure 2**). We also evaluated the impact of tumor sample purity in our results by analyzing the correlation between MATH scores and sample purity information provided for all 79 TGCA samples. As shown in **Supplementary Figure 3**, there is no significant correlation between MATH scores and tumor sample purity (cor = −0.037,

p = 0.75, Pearson correlation). Therefore, non-treated rectal tumors exhibit a remarkable variability in ITGH, which is not significantly influenced by underlying somatic CNVs and tumor sample purity and is significantly associated with disease stage and lymph node metastases.

### Effects of nCRT in Normal Adjacent Mucosa

To evaluate the iatrogenic effect of nCRT, we compared the mutational landscape of a matched set of peripheral blood cells collected before nCRT (BC), tumor adjacent colonic mucosa exposed to nCRT (Nrx) and pre (PRE-T) and posttreatment (POST-T) tumor samples derived from a single rectal cancer patient.

We first compared the total set of single nucleotide variants (SNVs) detected in BC and Nrx to determine if nCRT could introduce novel somatic mutations in the irradiated rectal colonic mucosa. As expected, since both samples are derived from the same patient, the majority of the SNVs (99.87%, 9,698/9,711) was shared between both samples and only a very small fraction of SNVs is exclusively detected in BC (0.06%, 6/9,711) or in Nrx sample (0.07%, 7/9,705) (**Figure 2A**). We next searched for significant variations in allele frequencies of SNVs shared between BC and NrX samples. PRE-T and POST-T samples were used as positive controls for allele frequency variations. Significant variations in allele frequencies were observed in the comparison between BC and POST-T (**Figure 2D**) and, to a lesser extent, in the comparison between BC and PRE-T (**Figure 2C**). In contrast, only 10 variants presented significant variations in allele frequencies between BC and Nrx (**Figure 2B**). None of these variants occurred in regions involved in V(D)J recombination but they are dispersed throughout the genome. Also, we observed a significant direct correlation of allele frequencies for SNVs present in both BC and NRX samples (R <sup>2</sup> = 0.91; p < 2 × 10 – 16, Pearson Correlation test, **Figure 2B**). Although these results need to be validated using a larger number of matched samples, they indicate that nCRT per se does not seem to introduce detectable novel somatic mutations or copy number variations in the irradiated mucosa.

### ITGH Increases After nCRT

To address the effect of nCRT on the clonal structure of rectal tumors, we generated WES data from 7 matched PRE-T and POST-T samples (**Supplementary Table 2**). PRE-T and POST-T samples presented a median of 133 (min. 42, max. 341) and 83 (min. 50, max. 676) somatic point mutations, respectively (**Supplementary Table 3**). No significant difference in the number of mutations between PRE-T and POST-T tumors was observed (Wilcoxon test, p = 0.9). We also did not observe significant alterations in the spectrum of DNA base changes between PRE-T and POST-T samples (**Supplementary Figure 4**) and we were unable to detect a DNA damage mutational signature in POST-T samples (**Supplementary Figure 5**). Overall, the most frequent mutations observed in our cohort are also consistent with results reported by TCGA (**Supplementary Table 4**) (24). On average, only 20% (min. 10%—max. 32%) of the somatic mutations were shared between PRE-T and POST-T samples, with MAF correlations between matched samples ranging from R <sup>2</sup> = 0.025–0.393 (**Supplementary Table 3**).

To quantify the effect of nCRT on the clonal structure of rectal tumors, we next determined MAF distributions and calculated MATH scores for the 7 matched PRE-T and POST-T samples (**Figures 3A,B**). Median MAF and MATH scores varied from 0.13 to 0.33 and from 23 to 60.1 among all 14 samples, respectively (**Supplementary Table 3**). Overall MATH scores were significantly higher in POST-T samples compared to PRE-T samples (median 41.7 vs. 28.8, p =

vs. POST-T. Significant variations in allele frequencies are highlighted in black, blue, and red, respectively (p < 0.05; Binomial test, Bonferroni adjusted).

0.04, Wilcoxon paired test, **Figure 3C**). Noteworthy, five out of the seven tumors with incomplete response to nCRT presented an increase in MATH values in POST-T sample (**Figure 3B** and **Supplementary Table 3**). This suggests that nCRT can significantly alter clonal structure in residual tumors, increasing ITGH.

#### nCRT Acts as a Strong Selective Pressure

To examine if nCRT can select pre-existing tumor cell subpopulations more prone to resist to nCRT, we monitored tumor cell subpopulation dynamics before and after nCRT. Since MATH score does not allow direct enumeration of distinct tumor cell subpopulations, we monitored their dynamics by applying binomial tests to identify somatic mutations with significant changes in MAFs between PRE-T and POST-T samples (named as enriched mutations). For this analysis, we focused on 401 coding and splice site somatic mutations shared between PRE-T and POST-T samples, since they are more likely to have a deleterious impact on protein function.

We identified a total of 210 somatic mutations (52.4%) enriched in PRE-T or POST-T samples (**Figure 4**). Mutation enrichment was validated using Sanger Sequencing (**Supplementary Figure 6**). Enriched mutations were significantly more frequent in POST-T compared to PRE-T samples [195/210 (92.9%) vs. 15/210 (7.1%), p < 0.00001<sup>7</sup> , Fisher's exact test, **Supplementary Table 5**]. Noteworthy, we observed an excess of deleterious non-synonymous mutations over neutral synonymous mutations in POST-T [136 nonsynonymous (N) and 46 synonymous (S), N/S = 2.96] compared to PRE-T-enriched mutations [8 non-synonymous (N) and 5 synonymous (S), N/S = 1.6]. The observed difference, however, was not statistically significant, probably due to the small number of enriched mutations in the PRE-T samples (p = 0.23, Fisher's exact test).

Finally, we used GSEA to verify if nCRT could select tumor cell subpopulations more prone to resist to nCRT. In addition to the 210 enriched mutations, we found a total of 634 somatic mutations that were exclusively detected in POST-T samples. More than 59% of these POST-T specific

mutations were observed in a single patient (PT02 377/634) with high tumor mutational burden. Although the presence of some of these POST-T specific mutations could be attributed to tumor topographic heterogeneity, not contemplated in the samples used for sequencing, some of these mutations could indeed result from tumor genetic instability and clonal selection during neoadjuvant therapy and were, therefore, used for GSEA. We observed that POST-T specific and POST-T enriched mutations frequently occurred in genes associated with cell cycle regulation and proliferation (mitotic spindle assembly and mitotic checkpoint gene sets) as well as with cell survival and differentiation (K-Ras, TNF-alpha, and Hedgehog signaling gene sets) (**Supplementary Table 6**). Among POST-T enriched mutations, we found non-synonymous mutations in genes associated with genetic instability and drug resistance in colorectal cancer, including mutations in the ATM (25, 26), DIDO1 (27, 28), and AKAP9 (29) (**Supplementary Figure 7**). Among POST-T specific mutations, we also found nonsynonymous mutations in genes involved in DNA repair and apoptosis, including ERCC6 previously associated with resistance

to 5-Fluorouracil and poor prognosis (30). This suggests that nCRT may act as a strong selective pressure resulting in the selection of tumor cell subpopulations in the residual tumor that are more prone to resist to nCRT.

### DISCUSSION

nCRT may result in significant primary tumor regression. Ultimately, tumors that achieve complete or even nearcomplete response may allow for organ-preserving strategies (31, 32). In this setting, even patients with early rectal cancer (cT2N0), otherwise considered for abdominal perineal resections or ultra-low anterior resections, have been considered for nCRT in order to avoid a definitive colostomy or poor anorectal function (11). However, even though patients with early stage disease are more likely to achieve a cCR, many patients will still harbor residual disease requiring surgical resection (33). In these patients, nCRT may contribute to significant increases in postoperative complications and worsening of functional anorectal outcomes (34, 35). Here, we demonstrated that nCRT may also have significant biological consequences to the residual cancer in the setting of incomplete tumor response.

In the present study, we showed that non-treated rectal tumors exhibited a remarkable variability in ITGH that is directly associated with disease stage and lymph node metastases. Studies using large tumor collections have shown that ITGH impacts clinical outcome (36, 37) and may contribute to drug resistance in different tumors (38). ITGH has been previously reported in colon (39–43) and rectal cancers (7, 8), however, for most of these studies, ITGH was determined only for a small number of tumors and the prognostic and predictive significance of ITGH for these cancers remains to be determined. Recently, the MATH score was used in two independent studies to quantify ITGH in colorectal tumors. Zhang et al. analyzed WES data from 284 colorectal tumors obtained from TCGA and 187 colorectal tumors obtained from the International Cancer Genome Consortium (ICGC) (43). The mean MATH value was 41.58 and 46.1 for the TCGA and ICGC cohorts, respectively. Similarly to our results, higher MATH scores were associated with disease stage and lymph node metastasis. Although the authors observed a significant difference in MATH scores between rectal and colon tumors (MATH = 45.9 vs. 39.96 p = 0.004), a separate analysis for rectal cancer was not performed and samples previously treated with nCRT were not excluded. Hardiman et al. have used the MATH score to analyze ITGH in 7 stage II/III rectal tumors, MATH scores in these tumors varied from 9.08 to 25.24 and were significantly higher in stage III tumors (7). Although further studies will be necessary to determine the prognostic significance of ITGH in rectal cancers, the strong association between ITGH and disease stage and lymph node involvement both known predictors of survival after surgical treatment among these patients—supports a possible role for ITGH as a prognostic biomarker in rectal cancer.

Another relevant finding was that nCRT, per se, does not introduce detectable somatic mutations in the irradiated colonic mucosa. To the best of our knowledge, our study was the first to directly address the potentially iatrogenic effect of nCRT. The use of treatment-exposed normal colonic mucosa was critical to distinguish treatment-induced mutations from those arising from tumor genetic instability and positive clonal selection after treatment exposure. Until present, few studies have indirectly addressed the impact of radiation and chemotherapy by comparing mutational landscapes of matched PRE-T and POST-T samples (44–47). Although post-treatment mutation spectrum shifts have been reported for esophageal adenocarcinoma following platinum-based neoadjuvant chemotherapy (45), WES of matched anal squamous cell carcinomas before and after chemoradiation revealed a similar number of somatic mutations and a similar pattern of DNA substitutions in pre and posttreatment tumors (47).

The most relevant finding of the present study is that MATH scores are significantly higher in POST-T compared to PRE-T samples. This suggests that nCRT can significantly affect ITGH in residual tumors. Significant alterations in ITGH have been reported for esophageal adenocarcinoma after exposure to neoadjuvant chemotherapy. Murugaesu et al. found that mutations in post-chemotherapy samples were rarely clonal (3%), while 50% of the somatic mutations identified prior to chemotherapy were clonal (45). Findlay et al. observed a variety of clonal behaviors in esophageal tumors after chemotherapy, including samples showing little changes in clonal composition and samples with marked differences in the clonal architecture after therapy (44). Most importantly, in both studies, there was a significant association between ITGH and response to neoadjuvant chemotherapy. More recently, marked clonal landscape remodeling has also been described for hormone-positive breast cancer exposed to neoadjuvant aromatase inhibitor treatment, but no associations between ITGH and treatment response were established (48). By comparing PRE-T and POST-T samples, we observed a significant overall increase in ITGH. Five out 7 patients presented a significant increase in ITGH. Interestingly, tumor regression in these 5 samples was minimal (20–30% tumor regression). In contrast one of the 2 patients showing minimal changes in ITGH presented the most significant tumor response (70% tumor regression). Even though these results could indicate a possible association between ITGH and tumor response to nCRT, the limited size of our cohort, and the fact that we have only analyzed tumors in the setting of incomplete response to nCRT, did not allow us to fully explore this association in the present work.

Finally, we monitored tumor cell subpopulation dynamics during nCRT by identifying enriched somatic mutations with significant changes in allele frequencies between PRE-T and POST-T samples. Enriched mutations were more frequently found in POST-T samples. We also observed higher proportion of potentially deleterious mutations in these samples. Enriched mutations in POST-T samples were frequently present among genes involved in DNA damage repair, genetic instability, cell cycle regulation, proliferation, survival, and differentiation (25– 29). All these molecular pathways have been shown to contribute to chemoradiotherapy resistance to colorectal tumor cells, suggesting that nCRT may result in tumors more aggressive than their baseline counterparts in the setting of incomplete response. Clonal evolution in response to neoadjuvant therapy has been previously studied in breast, esophageal, and anal squamous cell carcinomas (44, 45, 47, 48). Together with our study, these studies indicate that neoadjuvant therapy can profoundly affect tumor clonal architecture by promoting significant changes in the frequency of somatic mutations owing to the outgrowth of subclones with selective growth advantages in the residual tumor.

There are some limitations to this study that should be considered for the interpretation of our results and prevent the immediate application of our findings in clinical practice. First, we used a single set of matched samples in the analysis of the iatrogenic effect of nCRT in the normal colonic mucosa. Confirmation of our findings in larger cohorts is definitively necessary. Also, we cannot exclude the possibility that our sequencing strategy was not sensitive enough to detect novel somatic alterations, present in individual cells in the normal adjacent irradiated mucosa, which have not expanded significantly in the sampled population. Second, although we observed a significant variation in ITGH between PRE-T and POS-T samples, these observations were also based on a limited number of matched samples and in a single tumor region. We and others have described significant topographical intratumor heterogeneity in rectal cancer, and therefore, the impact of nCRT in ITGH and clonal selection in rectal tumors needs further evaluation. Most importantly, in the present work, we monitored tumor cell population dynamics by identifying "enriched mutations" with significant changes in MAFs between PRE-T and POST-T samples. Apart from a subclonal distribution and the presence of local somatic CNVs, the observed MAF at a specific locus is also directly influenced by tumor sample purity. Therefore, variations in sample purity, rather than in subclonal composition, could result in significant MAF differences between PRE-T and POST-T samples and consequently influence our analysis of tumor cell subpopulation dynamics before and after nCRT. In the present work, all tumor samples were microdissected by an experienced pathologist to enrich for tumor purity and minimize this possibility. Tumor sections were required to contain at least 80% tumor cell nuclei with <20% necrosis for inclusion in the study.

In conclusion, nCRT per se does not seem to introduce novel somatic mutations in the irradiated normal rectal mucosa. Instead, nCRT may drive a marked clonal selection in residual rectal tumors. This results in frequent increases in ITGH in residual cancers when compared to their baseline counterparts, which are driven by significant alterations in the frequency of biologically relevant mutations in genes associated with response to nCRT. The risk of more heterogeneous residual tumors leading to more biologically aggressive cancers may constitute a potential disadvantage of nCRT among incomplete responders. This may be particularly relevant among patients with early stage disease considering nCRT solely for the purpose of achieving cCR and organ-preservation. Future studies should address the oncological impact of significant ITGH increase after nCRT and incomplete response in rectal cancer.

### DATA AVAILABILITY STATEMENT

Sequence data has been deposited at the European Genomephenome Archive (EGA, http://www.ebi.ac.uk/ega/), which is hosted by the EBI, under accession number EGAS00001003250.

## ETHICS STATEMENT

This study was approved by the Ethics Committee of Hospital Alemão Oswaldo Cruz, São Paulo, Brazil (reference number 19/08) and was conducted in accordance with the Declaration of Helsinki. Patients provided written informed consent for tumor sample collection and study participation.

## AUTHOR CONTRIBUTIONS

AC and RP designed the study. FB performed experiments and acquired data. AC, BC, CM, ED, FS, FB, PG, and RP analyzed and interpreted data. RP, AH-G, GS, and BV recruited patients and acquired samples for analysis. AC, RP, and PG supervised the study. AC, AH-G, PG, and RP acquired funding for the study. AC, CM, FB, and RP prepared the manuscript. AC, AH-G, BC, CM, ED, FB, PG, and RP performed critical revision of the manuscript. All authors read and approved the final manuscript.

## FUNDING

This work was supported by The Ludwig Institute for Cancer Research (LICR) and Fundação de Amparo a Pesquisa do Estado de São Paulo (FAPESP 2015/17208-9 to CM; 2013/07159-5 to BC; 2010/12658-2 to ED, and 2011/51130-6 to RP).

#### ACKNOWLEDGMENTS

We thank Drs. Romualdo Barroso and Marcelo Negrão for reading the manuscript and Natália Felicio and Bruna Hessel

#### REFERENCES


for technical assistance. We are also indebted to with patients enrolled in this study.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00974/full#supplementary-material

analysis of data from the cancer genome atlas. PLoS Med. (2015) 12:e1001786. doi: 10.1371/journal.pmed.1001786


Wait Database (IWWD): an international multicentre registry study. Lancet. (2018) 391:2537–45. doi: 10.1016/S0140-6736(18)31078-X


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Bettoni, Masotti, Corrêa, Donnard, dos Santos, São Julião, Vailati, Habr-Gama, Galante, Perez and Camargo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Identification of Schlafen-11 as a Target of CD47 Signaling That Regulates Sensitivity to Ionizing Radiation and Topoisomerase Inhibitors

Sukhbir Kaur 1†, Anthony L. Schwartz 1†‡, David G. Jordan<sup>1</sup> , David R. Soto-Pantoja1‡ , Bethany Kuo<sup>1</sup> , Abdel G. Elkahloun<sup>2</sup> , Lesley Mathews Griner <sup>3</sup> , Craig J. Thomas <sup>3</sup> , Marc Ferrer <sup>3</sup> , Anish Thomas <sup>4</sup> , Sai-Wen Tang<sup>4</sup> , Vinodh N. Rajapakse<sup>4</sup> , Yves Pommier <sup>4</sup> and David D. Roberts <sup>1</sup> \*

<sup>1</sup> Laboratory of Pathology, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, MD, United States, <sup>2</sup> Cancer Genetics Branch, National Human Genome Research Institute, National Institutes of Health, Bethesda, MD, United States, <sup>3</sup> National Center for Advancing Translational Sciences, National Institutes of Health, Bethesda, MD, United States, <sup>4</sup> Developmental Therapeutics Branch and Laboratory of Molecular Pharmacology, Center for Cancer Research, National Cancer Institute, National Institutes of Health, Bethesda, MD, United States

Knockdown or gene disruption of the ubiquitously expressed cell surface receptor CD47 protects non-malignant cells from genotoxic stress caused by ionizing radiation or cytotoxic chemotherapy but sensitizes tumors in an immune competent host to genotoxic stress. The selective radioprotection of non-malignant cells is mediated in part by enhanced autophagy and protection of anabolic metabolism pathways, but differential H2AX activation kinetics suggested that the DNA damage response is also CD47-dependent. A high throughput screen of drug sensitivities indicated that CD47 expression selectively sensitizes Jurkat T cells to inhibitors of topoisomerases, which are known targets of Schlafen-11 (SLFN11). CD47 mRNA expression positively correlated with schlafen-11 mRNA expression in a subset of human cancers but not the corresponding non-malignant tissues. CD47 mRNA expression was also negatively correlated with SLFN11 promoter methylation in some cancers. CD47 knockdown, gene disruption, or treatment with a CD47 function-blocking antibody decreased SLFN11 expression in Jurkat cells. The CD47 signaling ligand thrombospondin-1 also suppressed schlafen-11 expression in wild type but not CD47-deficient T cells. Re-expressing SLFN11 restored radiosensitivity to a CD47-deficient Jurkat cells. Disruption of CD47 in PC3 prostate cancer cells similarly decreased schlafen-11 expression and was associated with a CD47-dependent decrease in acetylation and increased methylation of histone H3 in the SLFN11 promoter region. The ability of histone deacetylase or topoisomerase inhibitors to induce SLFN11 expression in PC3 cells was lost when CD47 was targeted in these cells. Disrupting CD47 in PC3 cells increased resistance

#### Edited by:

Ira Ida Skvortsova, Innsbruck Medical University, Austria

#### Reviewed by:

Jih-Hwa Guh, National Taiwan University, Taiwan Katerina Smesny Trtkova, Palacký University, Czechia

#### \*Correspondence:

David D. Roberts droberts@mail.nih.gov

†These authors have contributed equally to this work

#### ‡Present address:

Anthony L. Schwartz, Morphiex Biotherapeutics, Boston, MA, United States David R. Soto-Pantoja, Wake Forest School of Medicine, Winston-Salem, NC, United States

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 15 August 2019 Accepted: 16 September 2019 Published: 01 October 2019

#### Citation:

Kaur S, Schwartz AL, Jordan DG, Soto-Pantoja DR, Kuo B, Elkahloun AG, Mathews Griner L, Thomas CJ, Ferrer M, Thomas A, Tang S-W, Rajapakse VN, Pommier Y and Roberts DD (2019) Identification of Schlafen-11 as a Target of CD47 Signaling That Regulates Sensitivity to Ionizing Radiation and Topoisomerase Inhibitors. Front. Oncol. 9:994. doi: 10.3389/fonc.2019.00994

**64**

to etoposide but, in contrast to Jurkat cells, not to ionizing radiation. These data identify CD47 as a context-dependent regulator of SLFN11 expression and suggest an approach to improve radiotherapy and chemotherapy responses by combining with CD47-targeted therapeutics.

Keywords: radioresistance, epigenetics, CD47, thrombospondin-1, DNA damage response, schlafen-11, prostate cancer

### INTRODUCTION

CD47 is a widely expressed cell surface molecule in higher vertebrates (1, 2). CD47 plays a physiological role in recognition of self by serving as a counter-receptor for the inhibitory receptor SIRPα on macrophages and dendritic cells (3). CD47 like proteins acquired by Poxviridae also bind SIRPα and may have similar roles in protecting infected cells from host innate immunity (4, 5). Correspondingly, over-expression of CD47 in some cancers can protect tumors from innate immune surveillance (3, 6, 7). This has led to the development of therapeutic antibodies and decoy molecules that inhibit the CD47-SIRPα interaction and their entry into multiple clinical trials for cancer patients as potential innate immune checkpoint inhibitors (8–10).

In addition to the passive role of CD47 in self-recognition, cell-intrinsic signaling functions of CD47 have been identified in some tumor cells as well as in vascular and immune cells in the tumor microenvironment (11–13). CD47 signaling is induced by binding of its secreted ligand thrombospondin-1 (TSP1 encoded by THBS1), which modulates CD47 association with heterotrimeric G-proteins as well as lateral interactions of CD47 with specific integrins and tyrosine kinase receptors (1). In vascular cells, ligation of CD47 modulates calcium, nitric oxide, cAMP, and cGMP signaling (13). TSP1 also inhibits NK cell activation (14) and T cell receptor signaling in a CD47 dependent manner (15, 16). Genetic disruption or antisense suppression of CD47 enhances cytotoxic T cell killing of target tumor cells in vitro and suppresses tumor growth in vivo when combined with local tumor irradiation or cytotoxic chemotherapy (17, 18). In addition to enhancing their antitumor efficacy, blockade of CD47 signaling protects non-malignant tissues from the off-target effects of these genotoxic therapies by enhancing autophagy pathways, stem cell self-renewal, and broadly enhancing metabolic pathways to repair cell damage caused by ionizing radiation (19–21).

Here we utilized a high throughput screen of drug sensitivity to identify pathways that contribute to the radioresistance and chemoresistance of CD47-deficient cells. CD47-deficient cells exhibited significant resistance to topoisomerase and class I histone deacetylase (HDAC) inhibitors. Global differences in gene expression in WT Jurkat T cells and a CD47-deficient mutant and following siRNA knockdown of CD47 were examined to identify specific genes through which therapeutic targeting of CD47 could modulate radioresistance and chemoresistance. One of the genes that showed consistent down-regulation in CD47-deficient cells was schlafen-11 (SLFN11), which in human cancers is positively correlated with sensitivity of cytotoxic agents including topoisomerase inhibitors (22–28). Loss of SLFN11 expression in cancer cells involves both hypermethylation of its promoter and epigenetic changes in histone modification (29, 30). Correspondingly, expression of SLFN11 in some resistant cancer cell lines can be induced by class I HDAC inhibitors and restores their sensitivity, whereas knockdown of SLFN11 confers resistance (29). The mechanism by which SLFN11 regulates sensitivity to DNA damaging agents includes limiting expression of the kinases ATM and ATR (31). Other evidence indicates that SLFN11 blocks DNA replication in stressed cells upon recruitment to the replication fork independent of ATR (32). Parallels between the effects of SLFN11 and CD47 on resistance to genotoxic stress suggested that SLFN11 may be an effector mediating the selective cytoprotective effects of CD47 knockdown, prompting us to examine the regulation of SLFN11 and its orthologs by CD47 and the potential implications for combining CD47-targeted therapeutics with genotoxic cancer therapies.

### MATERIALS AND METHODS

#### Reagents and Cell Culture

Entinostat and rocilinostat were obtained from the NCI Division of Cancer Treatment and Diagnosis. Etoposide was from Bedford Laboratories. Doxorubicin was from Sigma-Aldrich.

PC3 and Jurkat T cells were purchased from the American Type Culture Collection and maintained at 37◦C with 5% CO<sup>2</sup> using RPMI 1640 medium supplemented with 10% FBS, glutamine, penicillin and streptomycin (Thermo Fisher Scientific, USA). The CD47-deficient Jurkat T cell mutant (clone JinB8) was from (33) and cultured as described previously (34). WT and CD47-deficient Jurkat cells were maintained at 2–5 × 10<sup>5</sup> cells per ml to prevent activation.

For transient SLFN11 over-expression, 1 × 10<sup>6</sup> JinB8 cells were transfected with 2 µg of SLFN11 expression vector (29) or control plasmid using an Amaxa nucleofection kit (Lonza) 48 h before irradiation. To assess cell viability Jurkat and JinB8 cells were plated at 2 × 10<sup>4</sup> cells/well and irradiated with a single dose of 20 Gy radiation (operating at kV/10 mA with 2-mm aluminum filter, Precision X-Ray, East Haven, CT) or treated with etoposide. Cells were incubated for an additional 48–72 h at 37◦C. Cell viability was determined by 3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxymethoxyphenyl)-2- (4-sulfophenyl)-2H-tetrazolium, inner salt (MTS) reduction using the CellTiter 96 Aqueous One Solution proliferation assay (Promega). Absorbance was read at 490 nm on a microplate spectrophotometer.

PC3 cells and CD47-null CRISPR edited cells (2,000/well) were either unlabeled or labeled with Rapid Red Dye (Sartorius) and were plated in triplicate in 96-well plates (Corning, USA) and cultured overnight. The cells were treated as indicated in the figure legends, and cell proliferation was measured by Phase object Confluence (%) analysis using the IncuCyte instrument. Similarly, Jurkat and JinB8 T cells (5,000/well) were plated on 96 well plates for 1 h. The cells were treated with anti-human CD47 (B6H12, 1µg/ml) as indicated in the figure legends.

400,000 WT and CD47 null PC3 cells were plated using 6 well plates in 2 ml of complete RPMI medium. The plates were irradiated with 20 Gy at a dose rate of 0.6 Gy/min using a GammaCell 40 Irradiator. PC3 cells were treated with entinostat, etoposide, doxorubicin, and rocilinostat at a concentration of 300 nM for 24 h for IncuCyte assays or for 72 h plated at 2,000 cells/well for MTS proliferation assays. Absorbance of untreated WT and CD47-null PC3 cells was normalized to 100%, and IC<sup>50</sup> values were calculated using IC50 Calculator | AAT Bioquest software. Control and treated cells were also harvested for RNA extraction and real-time PCR.

#### H2AX Assay

Jurkat and JinB8 cells or Jurkat cells pretreated with B6H12 antibody, were irradiated at 10 Gy, and then incubated for 0– 6 h before fixing with paraformaldehyde for 15 min and washing 2 times with PBS. Cells were permeabilized using 0.14% Triton X-100 in PBS and 3% BSA for 5 min and washed three times for 5 min each. Then, cells were stained with H2AX primary antibody 1:300 for 60 min and secondary antibody 1:600 Alexafluor 488 (ebiosciences) for 60 min. Cells were washed three times with PBS and mounted using DAPI VECTASHIELD <sup>R</sup> Mounting Medium (Vector Laboratories, Inc, Burlingame, CA). Images were acquired using Zeiss 710 or Zeiss 780 microscopes with a 63x objective. High throughput antibody screening or quantification of FITC and DAPI was acquired on a Mirrorball instrument (TTP Labtech).

### Comet Assay

DNA fragmentation in WT and CD47<sup>−</sup> Jurkat cells 24 h after irradiation at 10 Gy was assessed using a single cell gel electrophoresis (Comet) assay essentially as described (35).

#### Real-Time PCR

Total RNA was extracted using the Trizol method (11, 34) or NucleoSpin RNA isolation kit (Clonetech), and the concentration and quality of RNA was measured using Nanodrop. First strand cDNA was generated using the Maxima First Strand cDNA Synthesis Kit for RT-qPCR, with dsDNase. All RNA samples were subjected to treatment with DNAse-1 prior to first strand cDNA synthesis according to the manufacturer's instructions (Thermo Fisher Scientific). Real-time PCR was performed using SYBR Green detection on a Bio-Rad CFX instrument as described previously (36).

For assessing responses to irradiation, total RNA was extracted as described above using TriPure isolation reagent (Roche). The concentration of RNA was measured using Nanodrop. Hundred nanogram and 1 µg of RNA was used to generate First strand cDNA using the Maxima First Strand cDNA Synthesis Kit for RT-qPCR, with dsDNase. mRNA expression of SLNF11 was amplified using SLFN11-F 5′ -GGCCCAGACCAAGCCTTAAT-3 ′ and SLFN11-R, 5′ -CACTGAAAGCCAGGGCAAAC-3′ primers with 1 µg of RNA template, while 18S and actin were amplified using 100 ng RNA. The relative expression is normalized to control untreated samples.

### Confocal Microscopy

PC3 cells (WT, low CD47, and CD47-null cells) were plated on 4 well ibidi chambers overnight. The next day, the cells were treated with Rocilinostat or Entinostat for 24 h. The cells were fixed with 4% paraformaldehyde (Sigma Aldrich), and immunostaining was performed using antibodies against CD47 (Proteintech Group, Inc) and SLFN11 (Santa Cruz Biotechnologies) as described previously (36). The images were captured using a Zeiss 710 microscope with an oil immersion 63x objective. Cells were treated with 300 nM of Rocilinostat, or Entinostat, or etoposide for 24 h, and the cells were immunostained using anti-SLFN11 as described above. The images were captured using a Zeiss 780 microscope with an oil immersion 63X objective. All the images were captured with 5 and 10µm scale bars as indicated in the legends.

#### CD47 Knockdown and Microarray Analysis

CD47 knockdown was performed using Jurkat T cells with CD47-siRNA as described earlier (34). Oligofectamine transfection reagent alone was used as mock control. Total RNA was extracted using the Trizol method. The quality of RNA was checked using a RNA Bioanalyzer (Agilent Inc). Global expression analysis was performed using Affymetrix microarray protocols as described previously (34, 36). Disruption of CD47 in PC3 cells was performed using a human CD47 gRNA targeting the first exon, 5′ -CAGCAACAGCGCCGCTACCAGGG (37) using Cas9-GFP plasmid from Addgene (Cambridge, MA). The CRISPR plasmid was transfected using Oligofectamine (Thermo Fisher Scientific) according to the manufacturer's instructions. The cells were sorted based on CD47 expression to isolate CD47-low and CD47-null populations using CD47-PE antibody (Biolegend). The cells were expanded, and CD47 expression was re-validated by flow cytometry analysis using CD47-APC (Biolegend). CD47 Human siRNA Oligo Duplex (Locus ID 961) was purchased from OriGene and transfected using Oligofectamine (Thermo Fisher Scientific) into PC3 cells for 24 h using a 15 nanomolar concentration of the pooled CD47-siRNAs. After 24 h, the medium was removed, the cells were washed with PBS, and the cells were lysed using Tri Pure Isolation reagents. The knockdown of CD47 siRNA was assessed using real time PCR with the following primers: CD47-F (GGTTTGAGTATCTTAGCTCTAGCA), Long CD47-R (TCTACAGCTTTCCTAGGA) and short CD47-R (CCATCACTTCACTTCAGTCAGTTATTC).

#### Human Tumor Expression Data

The Cancer Genome Atlas (TCGA) data was analyzed using cBioPortal tools to determine correlations between SLFN11 and CD47 mRNA expression in human tumors with sufficient RNAseq data (38, 39). Additional TCGA SLFN11 vs. CD47 mRNA expression plots and correlations were derived using log2(x + 1) transformed RSEM normalized count data obtained from the TCGA data portal (prostate adenocarcinoma and normal prostate tissue, invasive breast carcinoma and normal breast tissue, lung squamous cell carcinoma tissue). For some cancer types, correlations between SLFN11 or CD47 mRNA expression and SLFN11 promoter DNA methylation were evaluated using TCGA methylation data derived using the Illumina HumanMethylation450 (HM450) BeadChip.

#### Quantitative High-Throughput Drug Sensitivity Screen

Wild type Jurkat T cells and the CD47-deficient mutant JinB8 cells were seeded into 1,536-well plates at 500 cells per well, in 5 µL of medium. A library of FDA approved small molecule drugs and late preclinical stage compounds were added at multiple doses ranging from 0.8 nM to 46µM. Cell viability was assessed after 48-h incubation at 37◦C by adding 3µL of CellTiter-Glo reagent (Promega) and measuring luminescence (RLU) after a 15 min incubation at 25◦C, with ViewLux (PerkinElmer). Data from the high throughput screening assays was analyzed as previously described (18). Differential activity of each compound between the wild type and CD47 deficient cell lines was determined by calculating a difference in the maximum response or logIC<sup>50</sup> (concentration giving 50% of maximal inhibition) for each compound.

#### Chromatin Immunoprecipitation (ChIP)

WT CD47-low and CD47-null PC3 cells were plated overnight in 6-Well plates. The cells were fixed using Paraformaldehyde solution (Sigma), and chromatin was extracted using Chromatin Isolation kit (Abcam). Genomic DNA was sheared using 34 pulses of 10–12 s each at level by sonication with Disruptor sonication System from (Diagenode). ChIP was performed following instructions from ChIP Kit—One Step (Abcam). The Anti-Histone H3 (tri methyl K27) antibody, Anti-Histone H3 (di methyl K4) antibody and Anti-Histone H3 (acetyl K18) antibodies (Abcam) was incubated overnight for 4◦C. The genomic SLFN11 primers were designed by using genomic DNA region of hg38\_dna range=chr17:35373531-35374940 using UCSC Genome Browser. The following primers were designed using the Primer 3 program: SLFN-838 (CCGTCACGCTGCTAGTGATA), SLFN-968 (GAGTTGGCCAAAGACAGGAG), SLFN-949 (CTCCTGTCTTTGGCCAACTC), SLFN-1076 (CTCCGCATCAGTGAGAAGTG). SLFN11 level of eluted CHIP-DNA was measured using real-time genomic SLFN11 primers with control GAPDH (Abcam). Enrichment in the ChIP assay was calculated by normalizing to the input.

#### Statistical Analysis

Two-sample t-test assuming equal variances was used for cell viability assays to quantify statistical significance (<sup>∗</sup> for p-value < 0.05 and ∗∗∗ for p < 0.001). Two-factor with replication ANOVA was used for real-time PCR analysis (<sup>∗</sup> for p-value < 0.05 and ∗∗∗ for p < 0.001).

#### RESULTS

#### CD47 Mutation or Antibody Engagement Modulates the DNA Damage Response

We previously reported that non-malignant cells and Jurkat T cells lacking CD47 are protected from genotoxic stress induced by ionizing radiation (19, 21). This protection is mediated in part by an enhanced protective autophagy response in cells lacking CD47 or with reduced CD47 expression (19). Radioresistance in the CD47-deficient mutant is associated with global metabolic stabilization, including induction of anabolic metabolites that mediate repair of DNA damage induced by ionizing radiation (21). To evaluate whether CD47 also regulates the repair of genomic DNA damage caused by ionizing radiation, we assessed nuclear H2AX foci in WT and CD47<sup>−</sup> Jurkat T cells 1 h after irradiation at 10 Gy (**Figure 1A**). Notably, the CD47<sup>−</sup> cells showed a stronger H2AX response at this time. Quantitative analysis of the kinetics of foci formation in WT cells showed a maximal response at 2 h and subsequent decline by 6 h (**Figure 1B**). Previous studies of WT Jurkat cells subjected to this dose of radiation demonstrated metabolic collapse at 8 h followed by cell death (21), but treating WT cells with the CD47 functionblocking antibody B6H12 protected cells from radiation-induced death (40). Consistent with these results, subjecting the WT T cells to 10 Gy irradiation in the presence of B6H12 resulted in accelerated but less intense H2AX foci formation that resolved by 6 h (**Figure 1B**). A comet assay to assess DNA fragmentation in the CD47<sup>−</sup> T cells 24 h after irradiation at 10 Gy showed no detectable DNA fragments, whereas DNA fragmentation at 24 h remained extensive in the irradiated WT cells (**Figure 1C**). Therefore, loss of CD47 or blocking its function improves the ability of these cells to restore genomic integrity after damage caused by ionizing radiation.

#### Lack of CD47 Protects T Cells From Topoisomerase and HDAC Inhibitors

Non-malignant cells lacking CD47 are also protected from genotoxic stress induced by the anthracycline doxorubicin (18), which causes DNA damage by multiple mechanisms including redox stress and inhibition of topoisomerase activity (41). A quantitative high-throughput screen of drug sensitivity was performed using the WT and CD47<sup>−</sup> Jurkat T cell lines to identify additional drugs that may exhibit CD47-dependent cytotoxic activities and the resistance pathways they target. In addition to increased resistance to anthracyclines in the CD47-deficient cells, analysis of the 72 drugs that exhibited significantly decreased potencies (>3-fold) in CD47-deficient cells identified significant enrichments of topoisomerase I (TOP1), topoisomerase II (TOP2), and HDAC1 inhibitors (**Figure 2A**). Topoisomerase I inhibitors exhibited 5- to 100 fold increases in their IC<sup>50</sup> values in the CD47-deficient cells (**Figures 2B–E**, **Supplementary Data File 1**). These included camptothecin, its therapeutic analogs topotecan and irinotecan, and the highly active irinotecan metabolite SN-38. Resistance of the CD47<sup>−</sup> T cells extended to several drugs in the anthracycline family including idarubicin and mitoxantrone (**Figures 2F,G**).

CD47<sup>−</sup> cells exhibited enhanced resistance to 16 HDAC1 inhibitors in the screen (**Supplementary Data File 1**). One of these, the class I HDAC inhibitor entinostat (HDAC1>HDAC3), was previously shown to restore SLFN11 expression and sensitivity to DNA damage in resistant cancer cell lines (29). The IC<sup>50</sup> value for entinostat was 3.2-fold higher for the CD47<sup>−</sup> cells compared to WT cells in the CellTiter Glo assay (**Figure 3A**) and was confirmed to be less potent for inhibiting proliferation of CD47<sup>−</sup> cells (**Figure 3C**). However, the class I HDAC inhibitor Romidepsin, which was shown to similarly induce SLFN11 (29), did not show differential activity in CD47<sup>−</sup> vs. WT cells (**Supplementary Data File 1**). Conversely, the selective HDAC6 inhibitor Rocilinostat, which did not induce SLFN11 in K562 chronic myelogenous leukemia or HT1080 fibrosarcoma cells (29), was 1.51-fold less potent for CD47<sup>−</sup> vs. WT cells in the CellTiter Glo assay (**Figure 3B**) and was less potent for inhibiting proliferation of CD47<sup>−</sup> cells (**Figure 3D**). Therefore, some but not all of the differences in drug sensitivities between WT and CD47<sup>−</sup> Jurkat cells are consistent with the previously reported effects of these drugs on SLFN11 expression.

### CD47 Correlation With SLFN11 Gene Expression

Two independent microarray analyses of the same WT and CD47-deficient T cell lines identified 8.7- and 10-fold decreased expression of SLFN11 mRNA in the CD47-deficient Jurkat

FIGURE 2 | Loss of CD47 confers selective cytoprotection against cytotoxic drugs. (A) High throughput screen of FDA-approved and late stage development drugs. Cumulative data is presented as log(IC50) values comparing treated cells lacking or expressing CD47. Yellow points indicate compounds where the IC50 values differ significantly, and those with positive values indicate compounds where the absence of CD47 protects cell viability. The table lists the identified significant target classes for 72 compounds that exhibited >3-fold decreased potency (1IC<sup>50</sup> for compounds with Curve Response Class −1.n and −2.n) in CD47-deficient Jurkat T cells compared to WT cells. (B–G) Representative dose response curves for WT Jurkat T cells (red) and CD47- mutant cells (blue). CellTiter Glo signal assessing cellular ATP levels is plotted as a function of Log(concentration) for topotecan (B), camptothecin (C), SN38 (D), Irinotecan (E), mitoxantrone (F), and Idarubicin (G).

mutant (p<0.05, **Figure 4A**). No other Schlafen gene family members showed a significant difference in expression between WT and CD47-deficient cells using a 2-fold cutoff. Decreased SLFN11 mRNA in the CD47<sup>−</sup> mutant was confirmed using real-time qPCR (**Figure 4B**).

Reexamination of our published microarray data comparing primary lung endothelial cells from WT and cd47−/<sup>−</sup> mice [GSE43133, (20)] identified a 3.6-fold decrease in mRNA expression of Slfn9, a presumed murine ortholog of the SLFN11 gene (42), in cd47−/<sup>−</sup> cells, suggesting that CD47 regulation

of mRNA expression for SLFN11 orthologs is conserved across species (**Figure 4A**). However, RNAseq analysis of unstimulated mouse NK cells (GSE113980) revealed increased Slfn4 and Slfn14 in cd47−/<sup>−</sup> NK cells but no significant difference in Slfn9 mRNA expression comparing sorted Lin−NK1.1+NKp46<sup>+</sup> cells isolated from naïve cd47−/<sup>−</sup> and cd47+/<sup>+</sup> mouse spleens (43) (**Figure 4A**), suggesting that CD47 regulation of murine Slfn9 expression is cell type-specific.

### CD47 Positively Regulates SLFN11 Expression

The decreased expression of SLFN11 in the CD47<sup>−</sup> mutant T cells suggested that CD47 signaling regulates the expression of SLFN11. To establish causality and exclude the potential for secondary mutations in the Jurkat somatic mutant suppressing SLFN11 expression, we examined whether acute CD47 knockdown decreases schalfen-11 expression (**Figures 4A,C**, **Supplementary Data File 2**). siRNA knockdown of CD47 in the WT cells resulted in a 1.4-fold decrease in SLFN11 (p = 0.046). If loss of SLFN11 contributes to the radioresistance of the CD47<sup>−</sup> cells, forced expression of SLFN11 should restore sensitivity. Transient transfection of the CD47<sup>−</sup> cells with a SLFN11 expression vector increased SLFN mRNA expression (**Figure 4D**) and decreased the viability of CD47<sup>−</sup> cells subjected to 20 Gy irradiation relative to untreated cells or cells transfected with the control plasmid (**Figure 4E**).

## CD47 Ligands Alter SLFN11 Expression

TSP1 signaling in T cells can be mediated by several cell surface receptors (44, 45), but at concentrations < 5 nM signaling is primarily CD47-dependent (15, 16). Correspondingly, treatment of WT but not CD47-deficient Jurkat T cells with 2.2 nM TSP1 time-dependently suppressed SLFN11 mRNA expression (**Figure 5A**). Treatment of WT Jurkat T cells with TSP1 decreased SLFN11 mRNA as early as 30 min. after addition. Inhibition was maximal by 1–3 h and decreased thereafter. This time-dependence is consistent with the known uptake and degradation of TSP1 by T cells (45). No inhibition of SLFN11 mRNA expression by TSP1 was observed in the CD47<sup>−</sup> mutant, indicating that the inhibitory effect of TSP1 on SLFN11 expression is CD47-dependent.

The function modifying CD47 antibody B6H12 has been used extensively as an antagonist of CD47 signaling through SIRPα in preclinical studies and demonstrated tumor suppressing activities in many xenograft models (3). However, some effects of B6H12 on cancer cells may be independent of inhibiting binding of CD47 ligands (11). B6H12 at 1µg/ml rapidly suppressed SLFN11 mRNA expression in Jurkat cells, with variable recovery at later time points in repeated experiments (**Figure 5A**). This is consistent with the attenuated DNA damage response following irradiation of Jurkat cells in the presence of B6H12 in **Figure 1B** and the known ability of B6H12 treatment to preserve proliferative capacity in irradiated Jurkat cells (40).

long and short CD47 mRNA transcripts. (D) RT-qPCR confirmation of transient over-expression of SLFN11 by plasmid transfection. (E) Radioresistance was assessed by MTS assay after irradiation at 20 Gy of CD47<sup>−</sup> JinB8 cells transfected as indicated in (D). \*p < 0.05, \*\*\*p < 0.001.

A proliferation assay was used to examine whether the suppression of SLFN11 by B6H12 also alters the sensitivity of Jurkat cells to doxorubicin. A concentration of doxorubicin was selected that is suboptimal for directly inhibiting growth of the cells (**Figures 5B,C**). B6H12 alone at 1µg/ml significantly inhibited cell proliferation, whereas an isotypematched control antibody was inactive. Combining B6H12 with 100 nM doxorubicin resulted in more inhibition of

the IncuCyte live cell analysis system. Significance: \* = vs. untreated, # = vs. B6H12 alone.

cell growth, consistent with an additive effect (p = 0.001, **Figure 5C**). Therefore, this CD47 antibody can protect Jurkat cells from genotoxic stress induced by ionizing radiation or cytotoxic chemotherapy.

TABLE 1 | Co-expression of SLFN11 and CD47 mRNA in human cancers.


RNAseq data from the indicated TCGA provisional datasets (except TARGET datasets for pediatric ALL and pediatric neuroblastoma) were analyzed using cBioPortal tools. q-values were derived from the Benjamini-Hochberg FDR correction procedure. Significant q-values are indicated in bold font.

#### Correlation Between CD47 and SLFN11 Expression in Human Cancers

An initial survey of TCGA cancer datasets with sufficient RNAseq data indicated that the positive correlation between CD47 and SLFN11 mRNA expression observed in Jurkat T cells extends to a subset of human cancers (**Table 1**). The most significant positive correlations were found for bladder urothelial carcinoma (r = 0.44, p = 8.75 × 10−21) and lung squamous cell carcinoma (r = 0.41, p = 2.2 × 10−<sup>21</sup> , **Supplementary Figure 1A**). Additional cancers with significant correlation included pediatric acute lymphoid leukemia, cutaneous melanoma, prostate adenocarcinoma, glioblastoma multiforme, hepatocellular carcinoma, esophageal carcinoma, invasive breast carcinoma, acute myeloid leukemia, soft tissue

TABLE 2 | Drug sensitivities of WT and CD47-null PC3 cells.


sarcoma, and renal clear cell carcinoma. The strong positive Spearman's correlation for lung squamous carcinoma contrasted with a lack of correlation for lung adenocarcinoma (r = −0.01, p = 0.89), further suggesting that the relationship between CD47 and SLFN11 expression is cancer type specific. Eight cancer types with adequate RNAseq expression data showed no significant correlation, and a significant negative correlation between CD47 and SLFN11 mRNA was observed for papillary thyroid carcinoma (r = −0.22, p = 3.4 × 10−<sup>7</sup> ).

Further analysis of the TCGA data for prostate cancers showed a positive correlation between CD47 and SLFN11 mRNA expression in the tumors but not in normal prostate tissues (**Figures 6A,B**). Normal breast tissue similarly lacked the positive correlation between CD47 and SLFN11 observed in invasive breast carcinomas (**Supplementary Figures 1B,C**). These data further indicate that CD47 regulation of SLFN11 mRNA expression is cell type-specific and differs between normal and malignant tissues.

A positive correlation between CD47 and SLFN11 was also found for the cell lines in the Cancer Cell Line Encyclopedia (Spearman's correlation 0.193, p = 1.6 × 10−<sup>9</sup> , q = 2.6 × 10−<sup>8</sup> , **Figure 6C**). These data suggest that the underlying mechanism for these positive correlations is at least partially intrinsic to the cancer cells. SLFN11 mRNA expression in the CCLE was bimodal. Segregating high vs. low expressing cell lines with a mean cutoff showed higher CD47 in the SLFN11 high cell lines (log ratio 0.35, p = 3.1 × 10−<sup>8</sup> ). Of the 7 prostate cancer cell lines in the CCLE, LNCAP and 22RV1 were high SLFN11 expressers, PC3 was moderate, and the remaining 4 were low expressers.

### Loss of CD47 Regulates SLFN11 Expression in Prostate Cancer Cells

We chose the PC3 line to examine whether CD47 also regulates SLFN11 expression and sensitivity to genotoxic stress in prostate cancer cells. CD47 was targeted using CRISPR/Cas9, and pools of mutant PC3 cells with low residual CD47 or completely lacking CD47 were isolated by fluorescence activated cell sorting. Lack of or decreased CD47 expression was confirmed by flow cytometry and visualized by immunofluorescent staining (**Figure 7A**). The CD47-null PC3 cells proliferated at a somewhat slower rate that the WT PC3 cells (**Supplementary Figure 2**). Loss or absence of CD47 expression in PC3 cells was accompanied by decreased SLFN11protein expression (**Figures 7B,C**). SLFN11 mRNA expression was also reduced in the CD47-null PC3 cells (**Figure 7D**).

### Loss of CD47 Differentially Regulates Drug and Radiation Sensitivities in Prostate Cancer Cells

Although loss of CD47 in Jurkat cells consistently protects these cells from ionizing radiation [present results and (19, 21, 40)], this was not the case when CD47 was disrupted in PC3 cells (**Figure 7E**). The initial loss and recovery of WT and CD47-null PC3 cells after irradiation at 20 Gy were similar. SLFN11 mRNA remained lower 2 h post-irradiation in the CD47-null cells but rose above that in irradiated WT PC3 cells at 24 h (**Figure 7D**).

Consistent with their lower SLFN11 and with the Jurkat cell results, CD47-null PC3 cells were less sensitive to rocilinostat and etoposide than were the WT PC3 cells (**Table 2**, **Supplementary Figure 3**). In contrast, CD47-null PC3 cells were moderately more sensitive than WT cells to entinostat and doxorubicin.

Treating WT PC3 cells with a sublethal concentrations of rocilinostat, entinostat, or etoposide for 24 h increased SLFN11 protein levels as detected by immunofluorescence (**Figure 8A**). In contrast, SLFN11 expression in CD47-null PC3 cells was not significantly induced by the same treatments. Induction of SLFN11 protein by rocilinostat, entinostat was paralleled by increased SLFN11 mRNA at 24 h in WT PC3 cells (**Figure 8B**). No elevation in SLFN11 mRNA was observed at the same time point in WT PC3 cells treated with etoposide. However, a time course for treatment with 300 nM etoposide indicated acute induction of SLFN11 mRNA at 2 h following the initial decrease (**Figure 8C**), which may account for the elevation of SLFN11 protein seen at 24 h in **Figure 8A**. Notably, SLFN11 mRNA was up-regulated at 24 h in the CD47-null cells treated with rocilinostat or etoposide. These data suggest a CD47 contextdependent effect of HDAC inhibition on SLFN11 expression in PC3 cells.

### Tumor Type-Specific Correlation of CD47 Expression With SLFN11 Promoter Methylation

Previous studies have identified roles for promoter methylation and epigenetic regulation in the loss of SLFN11 expression in various cancers (24, 29, 30). We further analyzed TCGA prostate cancer data to evaluate a potential role of CD47 in these two mechanisms for regulating SLFN11 transcription. Consistent with the data in **Figure 6A**, prostate tumors with low SLFN11 mRNA (z-score <0) were enriched in the quadrant with low CD47 mRNA (34% CD47 z-score <0 vs. 26% z-score >0, **Figure 9A**). As reported previously for a broad collection of cancer cell lines (29), SLFN11 mRNA in prostate tumors was negatively correlated with methylation of the SLFN11 promoter (p = 4.5 × 10−<sup>31</sup> , **Figure 9B**). A weaker negative correlation between CD47 mRNA expression and SLFN11 promoter methylation (p = 4.6 × 10−19) suggested that the regulation of SLFN11 expression in human prostate cancers by CD47 is mediated in part by this mechanism (**Figure 9C**). However, another subset of the prostate cancers with low SLFN11 expression had low promoter methylation (**Figure 9B**),

which was previously demonstrated using a diverse panel of cancer cell lines to predict epigenetic regulation of SLFN11 (29).

TCGA data for additional cancer types in **Table 1** were examined to determine the specificity of the correlation between CD47 expression and SLFN11 promoter methylation. In all the cancer types where adequate methylation data was available, SLFN11 mRNA expression was negative correlated with SLFN11 promoter methylation (**Supplementary Figure 4**, left panels). Consistent with the mRNA correlations in **Table 1**, CD47 mRNA expression in breast carcinomas was negatively correlated

expression was calculated using 18S RNA primer control and normalized to untreated controls for each time point (\* for p-value < 0.05 and \*\*\* for p < 0.001).

FIGURE 9 | using cBioPortal tools. (A) Positive correlation between z-scores for SLFN11 and CD47 mRNA expression determined by RNAseq analysis. (B) z-scores for SLFN11 mRNA expression are negatively correlated with β-values for SLFN11 promoter methylation data from the Illumina HumanMethylation450 (HM450) BeadChip. The indicated subsets with low SLFN11 mRNA (z < 0) are predicted to be promoter methylation-dependent or -independent based on the previous in vitro analysis of tumor cell lines (29). (C) SLFN11 promoter methylation is negatively correlated with CD47 mRNA expression.

with SLFN11 methylation (p = 3.6 × 10−14), and colorectal carcinoma lacked a significant correlation (p = 0.62). In contrast, the positive correlation between CD47 and SLFN11 expression for melanomas in **Table 1** (p = 4.4 × 10−14) diverged from the weak negative correlation between CD47 expression and SLFN11 methylation in these tumors (p = 3.1 × 10−<sup>3</sup> ). Furthermore, soft tissue sarcomas exhibited low levels of SLFN11 methylation that were independent of CD47 expression (p = 0.27), but had a significant positive correlation between CD47 and SLFN11 expression (p = 4.4 × 10−<sup>4</sup> ), consistent with epigenetic crossregulation independent of SLFN11 promoter methylation in these tumors.

#### Epigenetic Regulation of SLFN11 by CD47

The ability of selective HDAC1 inhibitors to restore SLFN11 expression in some resistant cancer cell lines (29) and the subset of prostate cancers in the TCGA data with low SLFN11 expression despite low promoter methylation suggested a potential epigenetic mechanism by which CD47 signaling could alter SLFN11 expression in prostate cancer. To examine potential epigenetic mechanisms for regulation of SLFN11 gene expression by CD47 in prostate cancer cells, we performed chromosome immunoprecipitation in WT and CD47-null PC3 cells using acetylated H3K18, trimethylated H3K4, and trimethylated H3K27 antibodies and analyzed their enrichment in a region upstream from SLFN11 that was identified based on ENCODE data to contain a high abundance of histone H3K27Me3 modification (**Supplementary Figure 5**). Consistent with the low SLFN11 expression in the CD47-null PC3 cells, H3K18Ac enrichment was markedly decreased at 838–968 and 949–1,076 (**Figures 10A,B**). However, H3K18Ac enrichment did not show a corresponding decrease consistent with the decrease in SLFN11 mRNA expression in the CD47-low pool. In contrast, enrichment of trimethylated H3K4 and H3K27 was dose-dependent with decreasing CD47 expression (**Figures 10C–F**).

### DISCUSSION

Previous studies have identified SLFN11 expression as a major determinant of cancer cell sensitivity to DNA-damaging chemotherapeutic agents and patient outcomes for several cancers (22–26, 29, 46). The present data extends this role to regulating the sensitivity of cells to ionizing radiation. We further identify a role for SLFN11 in the regulation by CD47 of the sensitivity of cells to radiotherapy and chemotherapy. Decreased expression of CD47 or its engagement by physiological or pharmacological ligands suppresses SLFN11 expression, and

using primers to amplify 838–968 (A,C,E) or 949–1,076 bp (B,D,F) 5 ′ from hg38\_dna range=chr17:35373531-35374940 5'pad in the SLFN11 promoter. Results are expressed as fold enrichment relative to input.

re-expression of SLFN11 in some cells with low CD47 expression is sufficient to restore their sensitivity to DNA damage. Previous studies have identified other cytoprotective pathways regulated by CD47 that are probably independent of SLFN11 including upregulation of autophagy (19, 47), anabolic metabolites (21), and transcription factors that support asymmetric stem cell self-renewal (20), but SLFN11 regulation by CD47 provides a complementary mechanism to more proximally regulate the DNA damage response.

Loss or blockade of CD47 in non-transformed cells and tissues and in the Jurkat T cell line consistently protects cells from genotoxic and ischemic stresses (1). However, some of the underlying protective mechanisms are lost or lead to different outcomes in cancer cells. For example, the protective autophagy response in non-transformed CD47-deficient cells exposed to ionizing radiation manifests as a non-protective mitophagy response in breast cancer cells (18). Similarly, blockade of CD47 signaling that preserves non-transformed stem cells results in differentiation of breast and hepatocellular carcinoma and stem cells (11, 12, 48). The present data demonstrate a similar divergence in regulation of the SLFN11 pathway in different cell lines. Loss of CD47 coincides with loss of expression for SLFN11 or its presumed murine ortholog Slfn9 in non-transformed cells, and at least in human Jurkat cells this contributes to protection from genotoxic stress induced by ionizing radiation or cytotoxic chemotherapy. Transient over-expression of SLFN11 is sufficient to resensitize CD47-deficient Jurkat cells to ionizing radiation. Conversely, ligation of CD47 by a CD47 antibody, which is known to confer cytoprotection in WT Jurkat cells (40), rapidly decreases SLFN11 expression in these cells. The physiological CD47 signaling ligand TSP1 similarly induces a decrease in SLFN11 expression.

The positive regulation of SLFN11 expression by CD47 extends to prostate cancer cells, and correlative data in human tumors extends this relationship to a subset of human cancers. Our PC3 cell data indicates that CD47 regulation of SLFN11 and responses to stress is more restricted in this cancer cell line. The decreased SLFN11 in CD47-null PC3 cells was not sufficient to protect these cells from ionizing radiation, but resistance to the anti-proliferative effects of etoposide and rocilinostat were observed. The latter resistance is consistent with the inability of rocilinostat to induce SLFN11 protein expression in the CD47 null PC3 cells.

To interpret the differences in SLFN11 regulation in WT and CD47-deficient cells following exposure to ionizing radiation, DNA damaging agents, or HDAC inhibitors, it is important to recognize the temporal differences in their action. DNA strand breaks are induced rapidly by radiation and locally produced ROS, and all damage occurs over a few minutes. In contrast, doxorubicin causes cumulative DNA damage by several mechanisms including intercalation, ROS-induced strand breaks, and inhibition of topoisomerase activity. Etoposide is a more specific inhibitor of topoisomerase activity, but this activity is similarly sustained for the duration of treatment. Our data show that sustained exposure to doxorubicin or etoposide results in a CD47-dependent accumulation of SLFN11 over 24 h. The HDAC inhibitors similarly induce SLFN11 in a CD47-dependent manner. Looking only at 24 h, radiation appears to induce the opposite response. SLFN11 mRNA was induced in CD47-null PC3 cells but not in irradiated WT cells. Considering the noted temporal differences in the respective genotoxic stresses, the observed CD47-dependence for SLFN11 regulation by these stresses may be consistent.

The detailed molecular mechanism by which CD47 signaling regulates SLFN11 mRNA and protein levels remains to be determined. The high throughput drug screen identified a significant resistance of CD47<sup>−</sup> Jurkat cells to class I HDAC inhibitors including the selective HDAC1 entinostat. On the other hand, lack of differential activity for romidepsin and differential inhibition of the CD47<sup>−</sup> and WT cells by the selective HDAC6 inhibitor rocilinostat suggested that the regulation of SLFN11 by CD47 may not exclusively involve HDAC1. Consistent with the drug screening data, ChIP data identified CD47-dependent regulation of histone modification in the SLFN11 promoter in prostate cancer cells. Loss of CD47 in PC3 cells was associated with increased histone H3 K4 methylation and K27 methylation and decreased H3K18 acetylation at this locus. The observed epigenetic effects of CD47 signaling on SLFN11 could also account for the differential resistance of CD47-deficient Jurkat T cells to several HDAC inhibitors in the drug screening. However, the ability of etoposide and rocilinostat to induce SLFN11 mRNA without a corresponding increase in protein expression in CD47-null cells suggests that CD47 positively controls SLFN11 expression by a posttranscriptional mechanism.

Analysis of tumor data in TCGA also implicated CD47 regulation of SLFN11 promoter methylation in a subset of cancers that includes prostate adenocarcinoma. However, SLFN11 promoter methylation is independent of CD47 expression in some cancer types that exhibit a positive correlation between SLFN11 and CD47 mRNA expression. Further studies will be required to define the relative importance of these two mechanisms in the cross talk between CD47 and SLFN11 in each cancer type. These data could guide the design of clinical trials combining CD47-targeted therapeutics with anticancer drugs that target DNA methylation or histone modification to maximize therapeutic responses in each cancer.

### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the manuscript/**Supplementary Files**.

## ETHICS STATEMENT

Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. The patients/participants provided their written informed consent to participate in this study.

### AUTHOR CONTRIBUTIONS

SK, AS AT, S-WT, YP, and DR contributed to conception and design of the study. SK, AS, DJ, DS-P, BK, AE, LM, CT, MF, VR, YP, and DR performed acquisition, analysis, or interpretation of data. SK, AS, and AE performed the statistical analysis. SK, AS, and DR wrote the first draft of the manuscript. DS-P wrote sections of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.

### FUNDING

This work was supported by the Intramural Research Programs of the NIH/NCI (DR: ZIA SC 009172, YP: Z01-BC-006150) and NHGRI (AE: Z01 HG200365-09), the CCR Drug Development Collaborative (DR), the National Center for Advancing Translational Sciences (NCATS) (MF, CT), and a NCI Career Transition Award K22 (1K22CA181274-01A1) and the Metavivor Foundation Young Investigator Award (DS-P).

### ACKNOWLEDGMENTS

We thank Dr. Kun Dong for assisting BK with ligation of gCD47 into the CAS9GFP vector.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.00994/full#supplementary-material

### REFERENCES


differentially modulate T cell behavior. J Cell Biol. (2002) 157:509–19. doi: 10.1083/jcb.200109098


**Conflict of Interest:** AS is Chief Executive Officer and a shareholder of Morphiex Biotherapeutics.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Kaur, Schwartz, Jordan, Soto-Pantoja, Kuo, Elkahloun, Mathews Griner, Thomas, Ferrer, Thomas, Tang, Rajapakse, Pommier and Roberts. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Altering DNA Repair to Improve Radiation Therapy: Specific and Multiple Pathway Targeting

Julian Biau1,2,3,4,5,6 \*, Emmanuel Chautard5,7, Pierre Verrelle1,6,8,9 and Marie Dutreix 1,2,3,4

1 Institut Curie, PSL Research University, Centre de Recherche, Paris, France, <sup>2</sup> UMR3347, CNRS, Orsay, France, <sup>3</sup> U1021, INSERM, Orsay, France, <sup>4</sup> Université Paris Sud, Orsay, France, <sup>5</sup> Université Clermont Auvergne, INSERM, U1240 IMoST, Clermont Ferrand, France, <sup>6</sup> Radiotherapy Department, Université Clermont Auvergne, Centre Jean Perrin, Clermont-Ferrand, France, <sup>7</sup> Pathology Department, Université Clermont Auvergne, Centre Jean Perrin, Clermont-Ferrand, France, <sup>8</sup> U1196, INSERM, UMR9187, CNRS, Orsay, France, <sup>9</sup> Radiotherapy Department, Institut Curie Hospital, Paris, France

Radiation therapy (RT) is widely used in cancer care strategies. Its effectiveness relies mainly on its ability to cause lethal damage to the DNA of cancer cells. However, some cancers have shown to be particularly radioresistant partly because of efficient and redundant DNA repair capacities. Therefore, RT efficacy might be enhanced by using drugs that can disrupt cancer cells' DNA repair machinery. Here we review the recent advances in the development of novel inhibitors of DNA repair pathways in combination with RT. A large number of these compounds are the subject of preclinical/clinical studies and target key enzymes involved in one or more DNA repair pathways. A totally different strategy consists of mimicking DNA double-strand breaks via small interfering DNA (siDNA) to bait the whole DNA repair machinery, leading to its global inhibition.

#### Edited by:

Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands

#### Reviewed by:

Michael Wayne Epperly, University of Pittsburgh, United States Bevin P. Engelward, Massachusetts Institute of Technology, United States

> \*Correspondence: Julian Biau Julian.biau@clermont.unicancer.fr

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 15 July 2019 Accepted: 19 September 2019 Published: 10 October 2019

#### Citation:

Biau J, Chautard E, Verrelle P and Dutreix M (2019) Altering DNA Repair to Improve Radiation Therapy: Specific and Multiple Pathway Targeting. Front. Oncol. 9:1009. doi: 10.3389/fonc.2019.01009

Keywords: DNA damage, repair systems, radiotherapy, radioresistance, inhibition

## INTRODUCTION

Radiation therapy (RT), in conjunction with surgery and systemic therapies (chemotherapy, targeted therapies, immunotherapy. . . ), is a cornerstone of cancer care. About 50% of cancer patients receive RT (1). The primary objective of RT is to increase the amount of radiation delivered to the tumor to ensure local control and reduce the amount of radiation in adjacent healthy tissues. Advanced developments such as image-guided RT (IGRT) or intensity-modulated RT (IMRT) have led to the enhancement of this therapeutic ratio (2). Despite such improvements, many patients still experience local recurrence of the disease after RT. Clinical factors such as tumor stage, frequently associated with increased hypoxia, can explain some of the failures, but it is clear that biological characteristics play a key part in successful treatment (3–5). RT-induced cell death is mostly due to DNA damage, especially to double-strand breaks (DSBs) (6, 7). Consequently, tumor cells with highly efficient DNA repair are radioresistant (8), whereas deficiencies in pathways that repair DSBs are particularly detrimental to the cells (9). Therefore, therapies that inhibit the DNA repair machinery have the potential to enhance RT efficacy (10, 11). Inhibiting DNA repair offers an opportunity to target genetic differences between tumor and normal cells, as DNA repair is often dysregulated in tumor cells (10, 12–14). Tumor cells divide rapidly because of unregulated cell cycle control. Thus, they have less time to repair DNA damage as compared to normal cells that are not dividing or will stop dividing after activation of key checkpoints induced by RT (15, 16). Beside altered cell cycle control, cancer cells may also present defects in their DNA repair system, inducing dependence on specific repair pathways and/or overexpression of alternative pathways

**82**

(16, 17). Furthermore, cancer cells often develop under stress conditions, thus raising the frequency of endogenous DNA damage (18, 19). This review will firstly focus on the distinct categories of DNA lesions induced by RT and the DNA repair pathways required for their repair. Subsequently, it will present the approaches that are currently being developed to enhance RT efficacy by modulating DNA repair.

#### RT-INDUCED DNA DAMAGE RESPONSE

DNA lesions induced by RT activate the DNA damage response (DDR), which essentially involves post-translational modifications of proteins to activate downstream signaling pathways (20). DDR is based on an intricate network of proteins that work together to manage DNA repair and cell cycle coordination. DDR interrupts the cell cycle, thereby inhibiting the spread of DNA damage to daughter cells and facilitating repair. Cell division arrest is mainly controlled by the checkpoint kinases CHK1 and CHK2, which are activated by the phosphatidylinsositol-3-kinases (PI3K) of the DDR machinery (15). DDR signaling is also essential for triggering apoptosis when repair is unsuccessful, notably through modifications to the p53 protein (20).

RT induces a variety of DNA lesions. Approximately 10,000 damaged bases, 1,000 single strand breaks (SSBs) and 40 DSBs are produced per gray per cell (21, 22). Such lesions, if not corrected, can lead to cell death by mitotic catastrophe and apoptosis. DSBs are the most lethal to the cells despite their low proportion, as one single unrepaired DSB can trigger cell death (7). DSBs are produced directly and indirectly by RT. Indirect DSBs most often occur during replication if the initial damage is unrepaired. As an example, when a replication fork encounters an unrepaired SSB, the fork is blocked and leads to the conversion of this SSB into a DSB (10, 23, 24). The resulting DSB can directly trigger cell death or activate DDR, which induces cell cycle arrest and promotes DNA repair. This repair is usually error-free, allowing the cell to survive without genetic consequences. It can also be error-prone, leading either to cell death if the error is not viable or mutation and chromosomic aberrations (25).

#### DNA REPAIR OF RT-INDUCED DAMAGE

Following RT, damaged bases induced by oxidative stress are repaired by the base excision repair pathway (BER) (26–31). In BER, damaged bases are excised by DNA glycosylases, resulting in apurinic (AP) sites. Subsequently; these AP sites are cleaved by apurinic endonuclease 1 (APE1) or an APlyase activity, leading to SSBs. SSBs are repaired by the part of the BER pathway called single strand break repair (SSBR) (**Figure 1A**) (12, 32). Either short-patch or long-patch SSBR can then proceed, depending on several factors such as type of lesion and cell cycle state. Single nucleotide insertion by DNA polymerase (Pol) β and ligation by DNA ligase III are described as short-patch SSBR, and interact with the protein Xray repair cross-complementing 1 (XRCC1). Long-patch SSBR involves the removal of a larger DNA segment, which requires several DNA replication factors such as proliferating cell nuclear antigen (PCNA), Pol δ/ǫ, flap endonuclease 1 (FEN1), and DNA ligase I. Concerning SSBs detection, poly-ADP-ribose polymerase (PARP1 or PARP2) are required (28, 33–37). The binding of PARP to SSB activates its auto-PARylation, and leads to the recruitment of BER/SSBR proteins. PARP-1 was also reported as a regulator of DNA repair gene expression through the E2F1 pathway (38). Unrepaired SSB or a damaged base can block the replication forks, resulting in fork collapse and DSB (23). The great majority of oxidative damage induced by ionizing radiation is corrected by BER. However, under hypoxic conditions, IR causes the formation of cyclodeoxynucleosides that can be only removed by nucleotide excision repair (NER). Several results suggest that NER may be involved in the repair of oxidized DNA damage. In addition, ionizing-radiation breast cancer risk has been related to polymorphism in ERCC2 (one of the main NER enzymes) (39).

Two major pathways repair DSBs: homologous recombination (HR) and non-homologous end joining (NHEJ) (40). However, both mismatch repair (MMR) and NER pathways have been reported to affect both HR- and NHEJ-mediated DSB repair efficacy to a lesser extent (41). The formation of DSBs triggers the activation of three key enzymes from the PIKK family: ataxia telangiectasia mutated kinase (ATM), ATM-related kinase (ATR), and DNA-dependent protein kinase (DNA-PK). This leads to the phosphorylation of many proteins, signaling damage and initiating DNA repair. One of the early steps is the phosphorylation of histone H2AX (γ-H2AX), which signals the presence of DSB to repair proteins where they aggregate in ionizing radiation-induced foci (IRIF) (42). Besides their crucial roles in DDR signaling, the kinases ATR and ATM are also involved in maintaining replication fork stability (14) and fork reversal in case of fork-stalling lesions, notably through SMARCAL1 (43).

In mammalian cells, c-NHEJ (classical NHEJ, **Figure 1B**) is the most efficient DSB repair mechanism. It acts by directly ligating the broken DNA ends (44). c-NHEJ can occur during the entire cell cycle. It is frequently accompanied by small deletions at the repair break site and is considered to be the main cause of DSB error-prone repair. The first step of NHEJ is the binding of the heterodimer Ku70/Ku80 at the end of the DSB (45), allowing the recruitment of catalytic subunit DNA-PKcs forming the protein complex DNA-PK (46). DNA-PK, bounded to DNA, is activated and phosphorylates numerous proteins including H2AX (47), Artemis (48), X-ray repair crosscomplementing 4 (XRCC4) and ligase IV complex (49), and XLF (XRCC4-like factor) (50) that are recruited on the site of the DSB and participate in its repair. When c-NHEJ is impaired, an alternative pathway called a-EJ (alternative EJ) or MHEJ (Micro Homology End Joining) (**Figure 1C**) is activated (51). At the initial breaking site, a deletion of 5–25 nucleotides is necessary to reveal micro-homologies to realize a-EJ (52), while a maximum of 4 deleted nucleotides is necessary for c-NHEJ (44). The microhomologies that are slightly longer in the case of a-EJ could partly explain the higher number of large deletions and other genomic rearrangements that occur (53, 54). The a-EJ pathway is differentiated from c-NHEJ by the fact that it is independent

leading to the formation of Holliday junctions. DNA polymerases can then synthetize across the missing regions. The Holliday junctions are finally resolved by cleavage

and followed by ligation of adjacent ends. Represents inhibitors of DNA repair in preclinical or clinical development. of Ku proteins (52). a-EJ involves mainly PARP1, XRCC1, ligase III (LIGIII), and the MRE11/RAD50/NBS1 (MRN) complex (55, 56). DNA polymerase theta (Pol θ or PolQ) is specifically involved

in nucleotide incorporation in the a-EJ mechanism through the TMEJ (theta-mediated end joining) pathway (57). HR is an alternative pathway for repairing DSBs that uses

the sister chromatid as a model, restricting this mechanism to the S and G2 cell cycle phases (**Figure 1D**) (40). HR is the most conservative and least error-prone repair mechanism. It necessitates the presence of BRCA proteins, defects of which increase susceptibility to breast or ovarian cancer. The DSB site is bounded by several factors such as the MRN complex, EXO1 (exonuclease 1), DNA2-BLM (Bloom syndrome), BRCA1 and CTIP (CtBP-interacting protein) that contribute to DNA resection and formation of a 3′ single-strand DNA (58–60), which is then coated by proteins of replication A (RPA). After the RPA protein's displacement by RAD51, BRCA2 together with the localizer of BRCA2 (PALB2), RAD54, and BARD1 (BRCA1-associated RING domain protein 1) mediates the nucleoprotein filament invasion of the homologous strand of the sister chromatid and creates the "D-loop" (61). DNA polymerases can then synthetize across the missing regions. The resulting Holliday junctions are finally resolved by cleavage and followed by ligation of adjacent ends (62). However, HR can sometimes be error-prone, especially if template switching occurs, e.g., in repeat sequences (63).

The choice between the two major mechanisms for DSB repair (NHEJ and HR pathways) seems to be linked to several factors such as cell-cycle phase, chromatin context, or availability of key actors such as the Ku complex, 53BP1 or RAD51 (64, 65).

### CURRENT STRATEGIES INVOLVED IN DDR INHIBITION IN COMBINATION WITH RT

#### Targeting Key Enzymes Involved in a Specific DNA Repair Pathway Inhibiting BER/SSBR

BER and SSBR pathways repair damaged bases and SSBs. Inhibiting BER/SSBR may lead to unrepaired damages that are converted to DSBs when encountering a replication fork (23). Therefore, in cells already defective for HR, such as BRCA−/<sup>−</sup> breast or ovarian cancer tumors, the inhibition of BER by PARP inhibitors leads to unrepaired DSBs and cell death (14). This effect, called synthetic lethality, has been extensively described (17, 66, 67) and studied in numerous clinical trials (68–70). Since the majority of RT-induced damages are repaired by BER/SSBR, inhibition of this pathway should highly sensitize cells to RT even in HR-proficient cells (29). The preclinical evaluation of PARP inhibitors has shown enhanced RT efficacy both in vitro and in vivo (71–73). Several PARP inhibitors have already been tested in or entered into numerous clinical trials in association with RT for brain metastases, ovarian cancer, breast cancer, rectal cancer, or glioblastoma, among others (**Table 1**). However, early data did not demonstrate convincing and coherent proof of synergy, although neither did they demonstrate unexpected toxic effects (74, 75). Another strategy for the inhibition of BER/SSBR is the development of APE1 inhibitors. APE1 is crucial for BER/SSBR and is commonly overexpressed in cancer cells (80, 81), giving to this strategy some tumor specificity. APE1 inhibitors have shown efficacy in combination with RT in preclinical studies (82). Lucanthone, an APE1 inhibitor, combined with temozolomide, has recently been tested in a phase 2 clinical trial in glioblastoma (**Table 1**). The results are not yet published.

#### Inhibiting NHEJ

DNA-PK, a key enzyme in NHEJ, is a member of the PI3K family that performs a central role in many cellular functions (83). Selective DNA-PK inhibitors have led to radiosensitization in preclinical studies (84–86). Three phase 1 trials are currently testing the safety and tolerability of a DNA-PK inhibitor (M3814) in combination with palliative RT +/- immunotherapy in advanced solid tumors (NCT02516813 and NCT03724890) and curative-intent radiotherapy in locally advanced rectal cancer (NCT03770689) (**Table 1**). Such strategies, which are not based on a selective effect on the tumor, are considered promising by some (14) though they have been criticized by others (87). Early reports of the clinical combination of M3814 and palliative RT showed enhanced normal tissue reactions including dysphagia, prolonged mucosal inflammation/stomatitis, and skin injury (87, 88). Inhibition of Ku subunits could also result in reduced DNA-PK activity and NHEJ. This is consistent with the existing data reporting that shRNA depletion of Ku70 or Ku80 showed cytotoxicity and radiosensitization in pancreatic cancer cells (89, 90). CC-115, a dual inhibitor of DNA-PK and mammalian target of rapamycin (mTOR), is being tested; preliminary anti-tumor activity has been reported, although whether these responses are attributable to activity against DNA-PK or mTOR is unclear (14, 91). A phase 1 trial testing CC-115 in combination with RT and temozolomide in the treatment of glioblastoma is ongoing (NCT02977780). NHEJ can also be indirectly inhibited via the EGFR pathway (see below).

#### Inhibiting HR

Cancer cells are known to be more proliferative than normal cells (92). Inhibitors of replication-associated processes such as HR exploit this specificity to enhance the therapeutic ratio. Nevertheless, there are few specific inhibitors of HR. It has been shown that RAD51 expression and functional HR can be reduced using imatinib during experimental RT, leading to increased radiosensitization (13, 93). Indirect inhibition of HR can also be obtained via cell cycle checkpoint targeting (see below).

### Targeting Key Enzymes Involved in Multiple Repair Pathways

DNA damage detection and signaling is the first step common to all DNA repair pathways. Acting on this step will alter several pathways. Therefore, several approaches have been tested to disable part or all of the DNA damage recognition/signaling steps.

#### Inhibiting ATM

ATM is one of the key enzymes in DNA damage signaling of DSBs for HR but also NHEJ (94). Defective cells in ATM are extremely radiosensitive, independent of their p53 status (95). ATM inhibitors have shown radiosensitization in preclinical studies (96–98). A single 15Gy RT dose suppressed tumor growth in a preclinical model when ATM was deleted in cancer cells vs. when deleted in endothelial cells (99), underlining the interest in testing ATM inhibitors in combination with highly conformal RT. Like DNA-PK, ATM is part of the PI3K family and has many cellular functions. A phase 1 study is currently testing an ATM inhibitor (AZD1390) in combination with RT in brain tumors including glioblastoma and brain metastases (NCT03423628). Indirectly, inhibition of the TGFβ-signaling or mitogen-activated protein kinase (MAPK) pathway can lead to reduced ATM activation and increased tumor cell radiosensitivity through reduced DSB repair (100–102).

#### Inhibiting ATR

ATR is a critical kinase that is activated in reaction to replication stress and blocked replication forks. ATR is one of the key enzymes in DNA damage signaling of DSBs (103). Cancer cells, which exhibit relatively elevated levels of replication stress, are more susceptible to dependence on ATR signaling for survival (104). An ATR inhibitor (AZD6738) has given encouraging preclinical results (67, 105) and is currently in phase I trials as monotherapy or in combination with olaparib, RT (NCT02223923), carboplatin and immunotherapy agents. Another ATR inhibitor (M6620) is being tested in three phase 1 trials with radiotherapy in esophageal cancer (NCT03641547), TABLE 1 | Examples of clinical trials of inhibitors of the DNA damage response in combination with radiation therapy.


locally advanced head and neck squamous cell carcinoma (NCT02567422) and brain metastases (NCT02589522).

#### Inhibiting MRN Complex

Mirin is an inhibitor of MRE11 endonuclease and thus of HR function. However, MRE11 is also upstream of NHEJ, and so mirin inhibits NHEJ and its effects are not specific to HR (16, 106, 107). It might be of particular interest in combination with RT.

#### Baiting DNA Breaks Signaling

This approach developed recently is represented by the molecules called Dbait/AsiDNATM. Dbait/AsiDNATM consist of doublestrand oligonucleotides that mimic DNA strand breaks and consequently bind and trap the signaling and repair proteins DNA-PK (24, 108, 109) and PARP (110), leading to inhibition of both SSB and DSB repair. In preclinical studies, the proof of concept that a RT-Dbait association could be used in treating melanoma (24) and high-grade glioma (111) has been established. A first-in-man phase 1 trial was conducted combining Dbait/AsiDNATM in combination with palliative RT in in-transit metastases of melanoma (79) (**Table 1**). In this trial, no dose-limiting toxicity was reported and the maximum tolerated dose was not met.

#### Targeting Chromatin Dynamics via Epigenetic Modifications

Epigenetics is an emerging field in cancer biology (112). It focuses on functionally relevant genome modifications that do not modify the nucleotide sequence. Such modifications include DNA methylation or histone modifications that may regulate gene expression but do not alter the associated DNA sequence. These modifications could also affect DNA repair ability. The loss of ARID1A, a piece of the SWI/SNF chromatin remodeling complex, was recently reported to induce a selective vulnerability to combined RT and PARP inhibitor therapy (113).

#### Inhibiting Histone Deacetylases (HDACs)

HDAC inhibitors are epigenetic therapeutics. They have the capacity to lower RT-induced damage repair both at the level of damage signaling, via inhibition of the ATM or MRN complex, and by directly affecting proteins involved in NHEJ and HR (112, 114–117). Several clinical trials have been carried out for various cancer types (77, 78) (**Table 1**).

### Inhibition of Kinases Involved in DDR-Related Survival Pathways

Sorafenib, a multi-kinase inhibitor, is currently utilized in the clinic for the treatment of hepatocellular and renal cancers. It inhibits MAPK signaling together with additional intracellular Ser/Thr kinases, leading to both NHEJ and HR inhibition. Sorafenib has shown a radiosensitization effect in preclinical studies (118). Dasatinib is another multi-kinase inhibitor inhibiting ABL and SRC tyrosine kinases. In preclinical studies, it has shown a radiosensitization effect partly due to blocking of DNA repair pathways involved in DSB repair (119). Sorafenib and dasatinib are clinically evaluated in association with RT. Because of their large spectrum of targets, most of these inhibitors may show high toxicity, which prevents them from being used at the dosage required to be efficient in combination with RT in the management of aggressive tumors that overexpress some of their targets (120, 121).

After RT, EGFR has been found to translocate into the nucleus and modulate DNA repair (especially NHEJ) through association with DNA-PKcs (122, 123). A current in-clinic approach is using the monoclonal antibody cetuximab to inhibit this nuclear translocation of EGFR. Cetuximab combined with RT has improved patients' overall survival in a phase III trial in head and neck cancer (124). Furthermore, EGFR signaling may be interrupted by small-molecule tyrosine kinase inhibitors such as erlotinib or gefitinib, especially in the case of specific EGFR mutation; these are currently being tested in combination with RT (125, 126).

#### Targeting Cell Cycle Checkpoints

Checkpoint dysfunction represents a common molecular defect acquired during tumorigenesis (15, 127), underlying the importance of its regulation in cancer development. Interfering with cell cycle checkpoint signaling is an alternative approach to modulating DNA repair activity and potentially improving the therapeutic ratio. The induction of DNA lesions by RT in normal cells stops their progression in the cell cycle, thereby avoiding the accumulation of other lesions and their damaging effects (20). This cell cycle arrest is subtly correlated with DNA repair to fine-tune cell cycle restart with the disappearance of damage. In cancer cells with an altered G1 checkpoint, cell cycle progression goes on relentlessly and, as a result, the removal of the G2 block increases unrepaired damage and its transfer to the daughter cells. This finally causes the loss of essential genetic material and cell death, a process that strengthens checkpoint inhibition strategies. Combination of RT with a dual CHK1 and CHK2 inhibitors (AZD7762 and prexsertib) showed a radiosensitization effect with an increase in mitotic catastrophe in different cancer cell lines and xenografts (128–130). A phase 1b trial was completed that combined prexsertib with RT and cisplatin or cetuximab in locally advanced head and neck cancer (NCT02555644), the results of which are not yet published. However, in addition to checkpoint activation, CHK1 is also involved in HR (131, 132) and it is uncertain if this is only a result of checkpoint inhibition or if it is partly due to HR inhibition (133).

Another target is the WEE1 kinase, which has been shown to be a major regulator of the G2-M checkpoint (134). This tyrosine kinase inhibits the entrance in mitosis by adding an inhibitory phosphorylation to Cdc2 (the human homolog of tyrosine kinase 1[Cdk1]) to tyrosine 15 (Y15). As a consequence, the Cdc2/cyclin B complex becomes inactivated, which stops the cells in G2- M and allows DNA repair. Preclinical studies have shown the potential use of WEE 1 inhibitors as radiosentizers (135, 136). Several ongoing clinical trials are testing WEE1 inhibitors with RT. In addition, several phase 1 trials are currently testing the WEE1 inhibitor adavosertib (AZD1775) in combination with RT and temozolomide in the treatment of glioblastoma (NCT01849146), with RT and cisplatin in cervical, vaginal or uterine cancer (NCT03345784), or in combination with RT and cisplatin in advanced head and neck cancer (NCT03028766 and NCT02585973).

### Combining DNA Repair Targeting, Immunotherapy, and Radiotherapy

DNA repair proteins preserve the integrity of the genome; therefore, DNA repair targeting may enhance the tumor mutational burden, which may lead to the production of neoantigens and increased activity of anti-cancer T cells. Some clinical trials have been set to investigate the use of immune checkpoint inhibitors, notably by combining Dravulumab (anti-PD-L1) with PARP (NCT02484404), ATR (NCT02264678) or WEE1 inhibition (NCT02617277). To date, the interplay between radiation and the immune system is far from being completely deciphered, but several interesting facts have been reported. The cytotoxic action of radiotherapy on tumor cells provides T lymphocytes with tumor neoantigens and releases pro-inflammatory cytokines, thus promoting the immune response (137). The cell death inducing this type of immune response is called immunogenic cell death. Combining immunotherapy with radiotherapy (several recent trials: NCT02707588, NCT02952586, NCT02999087) could increase the ability to cause immunogenic cell death by removing locks that block the immune system (138). The non-overlapping toxicities of DNA repair targeting and immune checkpoint inhibitors render the use of combinations of these agents with radiotherapy appealing (14, 139).

## CONCLUSION

Targeted therapies are beginning to demonstrate activity across a number of tumor types. The most promising approach toward improving the efficiency of a treatment and gaining a reliable response is to develop therapy combinations that decrease the chance of resistance and to treat resistance when it emerges. There has been a considerable renewed emphasis on new targeted treatments such as radiosensitizers that do not cause overlapping dose-limiting toxicities. Selection of appropriate targeted agents represents a challenge. As predicted, during preclinical and clinical trials, particular attention was paid to proteins involved in DNA repair pathways. Various strategies have been explored, ranging from specific protein targeting to global inhibition, and many DNA repair inhibitors have been developed. Up until now, only a few of them have reached the clinical stage, while even fewer have been tested in combination with RT. The several clinical trials currently underway will tell whether these new compounds can be tolerable and efficient.

RT-induced lesions can be corrected by various DNA repair pathways. The intricacy of crosstalk between DNA repair pathways suggests that biomarker assays to determine the status of multiple DNA repair pathways could provide essential information on the sensitivity and resistance of cancer cells to

#### REFERENCES


repair inhibitors. Understanding these DNA repair pathways and identifying effective stratification biomarkers from the various DNA repair pathways that are specifically altered in some tumors would be required to characterize patients' responses to specific DNA repair inhibitors.

#### AUTHOR CONTRIBUTIONS

JB and EC analyzed the literature and wrote the manuscript. PV and MD gave important intellectual input and carefully revised the manuscript. All authors approved the final manuscript for submission.


DNA-damaging agents in preclinical tumor models. Clin Cancer Res. (2007) 13:2728–37. doi: 10.1158/1078-0432.CCR-06-3039


The past or future of cancer treatment? Pharmacol Ther. (2016) 160:65–83. doi: 10.1016/j.pharmthera.2016.02.003


molecules disorganizes DNA damage response. PLoS ONE. (2009) 4:e6298. doi: 10.1371/journal.pone.0006298


and neck: a randomized phase II trial. J Clin Oncol. (2013) 31:1415– 21. doi: 10.1200/JCO.2012.46.3299


**Conflict of Interest:** MD is the cofounder of DNA Therapeutics/Onxeo, which is involved in DT01 development.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Biau, Chautard, Verrelle and Dutreix. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Optical Imaging Approaches to Investigating Radiation Resistance

#### Sina Dadgar and Narasimhan Rajaram\*

*Department of Biomedical Engineering, University of Arkansas, Fayetteville, AR, United States*

Radiation therapy is frequently the first line of treatment for over 50% of cancer patients. While great advances have been made in improving treatment response rates and reducing damage to normal tissue, radiation resistance remains a persistent clinical problem. While hypoxia or a lack of tumor oxygenation has long been considered a key factor in causing treatment failure, recent evidence points to metabolic reprogramming under well-oxygenated conditions as a potential route to promoting radiation resistance. In this review, we present recent studies from our lab and others that use high-resolution optical imaging as well as clinical translational optical spectroscopy to shine light on the biological basis of radiation resistance. Two-photon microscopy of endogenous cellular metabolism has identified key changes in both mitochondrial structure and function that are specific to radiation-resistant cells and help promote cell survival in response to radiation. Optical spectroscopic approaches, such as diffuse reflectance and Raman spectroscopy have demonstrated functional and molecular differences between radiation-resistant and sensitive tumors in response to radiation. These studies have uncovered key changes in metabolic pathways and present a viable route to clinical translation of optical technologies to determine radiation resistance at a very early stage

in the clinic. Keywords: raman spectroscopy, diffuse reflectance spectroscopy, optical metabolic imaging, nonlinear optical

# INTRODUCTION

microscopy, mitochondrial organization, radiation resistance

About half of cancer patients from all cancer types are treated with radiation therapy either followed by or concurrently with surgery, chemotherapy, or other forms (1). However, despite the recent advances in targeted radiation therapy, several patients subsequently experience loco-regional recurrence due to acquired or intrinsic radiation resistance. The current standard of care to determine radiation response is an anatomical assessment of tumor volume shrinkage. This evaluation is typically performed 6–8 weeks after completion of treatment using X-ray Computed Tomography (CT) or Magnetic Resonance Imaging (MRI). There are currently no methods to determine radiation response either during or immediately after treatment. An early determination of radiation resistance could help physicians modify the radiation dosage to improve response rates and hence quality of life. The development of methods to identify radiation-resistant tumors early requires a better understanding of the biological mechanisms promoting radiation resistance.

Ionizing radiation functions by producing free radicals in cancer cells either directly in the DNA or indirectly in other molecules, primarily water (H2O). These radiation-induced free radicals, in the presence of O2, can generate peroxy radicals (DNA-OO·) capable of breaking chemical bonds

#### Edited by:

*Ira Ida Skvortsova, Innsbruck Medical University, Austria*

#### Reviewed by:

*Nikolaos Patsoukis, Harvard Medical School, United States Pavithra Viswanath, University of California, San Francisco, United States*

> \*Correspondence: *Narasimhan Rajaram nrajaram@uark.edu*

#### Specialty section:

*This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology*

Received: *29 July 2019* Accepted: *16 October 2019* Published: *05 November 2019*

#### Citation:

*Dadgar S and Rajaram N (2019) Optical Imaging Approaches to Investigating Radiation Resistance. Front. Oncol. 9:1152. doi: 10.3389/fonc.2019.01152*

**92**

and initiating a series of events which lead to DNA modification, and cell death (damage fixation). In contrast, lack of O<sup>2</sup> leads to the reduction of free radicals in DNA and restoration of the original form of DNA (DNA-H) leading to cancer cell survival (2–4). Landmark studies in clinical head and neck cancer and soft-tissue sarcoma found that pre-treatment oxygenation levels were predictive of treatment response and disease-free survival (5–7). This important role of oxygen is the rationale for fractionated radiation therapy (2 Gy/day for 6–7 weeks), which is believed to re-oxygenate and radio-sensitize former hypoxic cells and hence, cause cell death via damage fixation (8–10). However, recent work has started to uncover a possible role for radiationinduced reoxygenation in also promoting radiation resistance through hypoxia-inducible factor (HIF).

Hypoxia leads to stabilization of HIF-1 (11). While HIF-1 expression is inhibited under oxygenated conditions via prolyl hydroxylases (PHDs), its transcription is significantly upregulated under hypoxic conditions (3, 12, 13). However, radiation-induced tumor reoxygenation can lead to activation of HIF-1 as well-through accumulation of reactive oxygen species (ROS), which is necessary and sufficient to stabilize HIF-1 (14). Nuclear accumulation of HIF-1 in response to ROS has been shown to promote endothelial cell survival and hence promote radiation resistance (15, 16). In a tumor bearing window chamber model, Moeller et al. demonstrated an increase in ROS during radiation-induced reoxygenation. Additionally, they showed that injecting hydrogen peroxide (H2O2) into the window chamber lead to an increase in HIF-1 expression (15). HIF-1 directly targets several glycolytic genes and leads to increased glucose catabolism under oxygenated conditions (17–20). HIF-1 transactivates pyruvate dehydrogenase kinase (PDK), which inhibits pyruvate dehydrogenase and shunts pyruvate away from the mitochondria resulting in glucose catabolism to lactate even under oxygenated conditions (17, 18). Inhibition of HIF-1 and subsequent inhibition of PDK-1 restores glucose flux toward mitochondria and increases O<sup>2</sup> consumption (21). Other studies have shown that HIF-1 and pyruvate kinase M2 exist in a positive feedback loop that enhances glycolysis under aerobic conditions (19, 20).

Zhong et al. demonstrated that scavenging ROS resulted in a reduction in post-radiation aerobic glycolysis without reducing the magnitude of reoxygenation (22).

The switch to increased glucose catabolism can promote radiation resistance through utilization of the pentose phosphate shunt (PPP) to maintain the NADPH-glutathione buffer and hence scavenge radiation-induced ROS. Inhibition of glucose flux through the PPP in combination with 2 Gy of radiation treatment significantly decreased cancer cell proliferation, especially in radiation-resistant cells (23). Increased glucose catabolism can also lead to increased production of lactate, an important ROS scavenger, leading to decreased radiation sensitivity (24, 25). Thus, in addition to being key hallmarks in the development of cancer, tumor oxygenation (or hypoxia) and metabolism play a significant role in the development of radiation resistance. Technologies that are sensitive to these key hallmarks and that can measure them both at the "bench" and "bedside" can provide powerful tools to shed light on radiation resistance.

Optical imaging can provide non-destructive and quantitative methods to reveal morphological and biochemical changes within cells and tissue across length scales in response to radiation therapy. Due to its non-destructive nature, optical imaging can be used to longitudinally monitor dynamic biological changes with high resolution to investigate the underlying mechanisms that promote radiation resistance. Given the low cost and non-ionizing nature of the light used, optical techniques are also well-positioned for clinical translation, especially for accessible tumors of breast, skin, oral cavity, and uterine-cervix. In addition, same instrumentation and quantitative models are frequently used to extract meaningful information from pre-clinical animal models. This review highlights recent work that used non-linear optical microscopy and diffuse optical spectroscopy to shed light on differences between radiation-resistant and sensitive cancer cells. Specifically, we highlight studies that identified differences in oxygenation or reoxygenation trends post-radiation therapy as well as those that investigate metabolic and molecular changes in the post-radiation tumor milieu. These studies encompass models ranging from in vitro cell culture to in vivo animal studies and indicate the great potential of optical imaging in the sphere of biological investigations of radiation resistance and the development of clinically translational optical technologies to benefit patients receiving radiation therapy.

### OPTICAL MICROSCOPY

Non-linear microscopy approaches, such as two-photon microscopy present numerous advantages over conventional single-photon microscopy (26). Because autofluorescence is generated through simultaneous absorption of two photons, the excitation wavelengths used are at twice the single-photon excitation wavelength and half the energy. Doubling the singlephoton excitation wavelength usually places the non-linear excitation wavelength in the near-infrared range, which allows light to penetrate deeper within tissue (27). Additionally, the localization of autofluorescence to just the focal point of the objective provides an efficient method for rejecting out-of-focus light and minimizing photodamage to the sample. In this review, we discuss how two-photon excited fluorescence (TPEF) from two key metabolic cofactors—nicotinamide and flavin adenine dinucleotides (NADH and FAD, respectively), can provide a non-destructive metabolic profile of cells and how these approaches have been utilized to study the metabolic response to therapy in radiation-resistant and sensitive cancer cells.

#### Cellular Metabolism

Non-linear optical microscopy is well-suited to provide noninvasive high-resolution 3D images of mitochondrial structure and function within live cells, tissues, and animals (27, 28). Through two-photon excited fluorescence (TPEF), the intrinsic fluorescence of nicotinamide and flavin adenine dinucleotides (NADH and FAD, respectively) can be detected without the aid of exogenous dyes (26, 29). Based on the autofluorescence of NADH and FAD, the optical reduction-oxidation (or redox) state of cells can be quantified as FAD/(NADH+FAD). This optical

A549-RR cells at baseline and 24 h after radiation, indicating reoxygenation-induced HIF-1 expression in the A549-RR cells. Asterisks placed above bars indicate statistical significance. Error bars in (B,C), and (D) represent standard deviation of the mean plate value. Adapted with permission from Alhallak et al. (38).

redox ratio (ORR) has been shown to be significantly correlated with mass spectrometry-based measurements of NAD+/(NAD<sup>+</sup> + NADH), and can thus reveal the specific metabolic pathways engaged within a cell (30). Specifically, an increase in ORR has been attributed to increased oxidative phosphorylation because of the oxidation of NADH to non-fluorescent NAD<sup>+</sup> and FADH<sup>2</sup> to fluorescent FAD. On the other hand, hypoxia-like conditions that drive a buildup of NADH due to the inability to convert to NAD<sup>+</sup> and increased glucose catabolism has been shown to decrease the ORR (30, 31). Recent work from separate groups has demonstrated that the optical redox ratio is sensitive to dynamic changes in oxygen consumption and can provide metabolic assessments comparable to those of the Seahorse metabolic flux analyzer (32, 33). The optical redox ratio has been used to create metabolic image maps of key organs (34), such as the heart and brain, identify metabolic changes associated with cancer progression (35, 36), determine cellular response to therapy (37– 39), and discover a relationship between metastatic potential and cellular metabolism (32, 40, 41).

Alhallak et al. determined the early metabolic alterations in response to radiation in human A549 lung cancer cells and an isogenic radiation-resistant clone (38). This clone was obtained by repeated exposure of parental radiation-sensitive human lung cancer cell line (A549) to ionizing radiation (25 fractions of 2.2 Gy every 3 days). Although there was no significant difference in ORR of radiation-resistant and -sensitive cells prior to administration of radiation, there was a significant decrease in ORR of radiation-resistant cells 24 h after radiation, which was consistent with Seahorse-based quantification of the normalized oxygen consumption rate (n-OCR) (**Figure 1**). The observed results indicate that the radiation-resistant cancer cells have decreased levels of oxygen consumption both at baseline and post-radiation and resort to increased glucose catabolism after radiation to potentially reduce ROS-induced toxicity. Interestingly, this radiation-induced decrease in the optical redox ratio was also associated with a large increase in the HIF-1 expression in the radiation-resistant A549 clone.

A subsequent by Lee et al. investigated metabolic changes in response to HIF-1 inhibition to determine if the changes in optical redox ratio post-radiation were indeed mediated by HIF-1 and a mechanism to avoid ROS-induced toxicity (39). They used multiphoton microscopy to determine the ORR of A549-RR prior to and post-treatment with YC-1, an established HIF-1 inhibitor. Treatment with YC-1 for 24 h resulted in a significant increase in the ORR compared with baseline, with a concomitant increase in mitochondrial ROS (**Figure 2**), a decrease in reduced glutathione and a decrease in glucose uptake (39). These results support the conclusion also reached by Furdui and colleagues who found increased glucose uptake that was utilized within the pentose phosphate pathway (PPP) to maintain the NADPH-glutathione buffer. This buffer helps scavenge radiation-induced ROS and hence promote radiation resistance (23). These results demonstrate the enormous potential of autofluorescence microscopy to not only provide clinically translational biomarkers of cellular response to therapy but also create opportunities for investigating radiation biology in live cells and animals at very high resolution.

#### Lifetime Imaging

Fluorescent lifetime imaging microscopy (FLIM) measures the average time that a molecule spends in an excited state prior to emission. One significant advantage of FLIM over measurements

of endogenous autofluorescence is that lifetime is independent of the fluorophore concentration. The lifetime of fluorophores, such as NADH and FAD depend on whether they are free or bound to a protein complex. For instance, the lifetime of NADH autofluorescence is shorter (∼0.4 ns) when free and longer (∼1 ns) when bound to protein complexes, such as malate dehydrogenase and lactate dehydrogenase while the lifetime of FAD autofluorescence is longer when free and shorter when bound to protein complexes, such as alpha-lipoamide dehydrogenase (42–46). By quantifying the ratio of free to protein-bound NADH and their respective lifetimes, FLIM can be used to identify the metabolic state of cells and tissue (42, 47–49). A recent study investigated the application of FLIM in radiation research (50). Campos et al. first treated human cancer cells and normal oral keratinocytes (NOK) with 10 Gy of radiation and recorded the resultant metabolic changes using FLIM. As early as 30 min post treatment, there was a significant decrease in NADH lifetime of cancer cells while there was no change in NADH lifetime of the NOK cells.

### Mitochondrial Organization

In addition to being the powerhouse of the cell, mitochondria are also critical to cell death pathways. The energy demands of a cell are maintained by a delicate balance between the rate of oxidative phosphorylation, tricarboxylic acid (TCA) cycle activity, structural changes to the mitochondrial network, and mitochondrial biogenesis. Mitochondria are continuously changing their organization through fission and fusion allowing for adaptation to different functional demands (51, 52). This dynamic mitochondrial network is sensitive to cell differentiation as well as oxygen and nutrient availability (30, 53–55). Fission is critical for mitochondrial biogenesis, cell division, and mitochondrial autophagy and manifests as numerous mitochondrial fragments. Fusion helps to maintain functionality through the sharing of proteins, genetic material, and metabolites and leads to the generation of interconnected mitochondria (56). Alterations to fusion-fission dynamics and hence the mitochondrial organization have been shown to be associated with several pathological conditions, including hypoxia-reoxygenation injury (57–59). Hypoxia-reoxygenation has been shown to result in a decrease in mitochondrial fusion and subsequent changes in length and shape of mitochondria (60). Targeting the changes in mitochondrial fusion and fission has been shown to protect cells from the effects of hypoxia-reoxygenation injury (61, 62). These studies of changes to mitochondrial structure in response to hypoxia-reoxygenation injury are highly relevant to radiation therapy due to the similarity in mechanisms generating oxidative stress. Radiation therapy leads to reoxygenation of previously hypoxic cells, thereby triggering a large production of mitochondrial ROS. The NADH autofluorescence images, which are used to calculate the optical redox ratio, can also be used to evaluate mitochondrial organization and specifically, fission and fusion. Specifically, Fourier-based power spectral density analysis of NADH autofluorescence images has been used to compute a metric termed mitochondrial clustering to quantify mitochondrial organization (30, 63). An increase in mitochondrial clustering was found during periods of increased glucose catabolism, such as hypoxia, resulting in more fragmented, or fissioned mitochondria. On the other hand, glutaminolysis was found to be associated with a decrease in mitochondrial clustering or more networked mitochondria (fusion). The same method was used to investigate mitochondrial structural dynamics in human skin in vivo (53). A recent study used an improved image processing method in the spatial domain to rapidly quantify the local fractal dimension (FD) within individual cells in response to radiation therapy (64). This analysis found a significant decrease in FD (or an increase in mitochondrial clustering) of radiation-resistant lung cancer cells between 12- and 24-h post-radiation compared with preradiation measurements. There were no significant changes in the radiation-sensitive cell population in response to radiation at any time point (**Figure 3**). The increased mitochondrial clustering observed here is consistent with the decreased optical redox ratio and increased glucose catabolism observed by Alhallak et al. using the same cancer cells (**Figure 1**) (38, 39).

### OPTICAL SPECTROSCOPY

Optical spectroscopy is a fiber-based approach using nonionizing radiation to non-destructively and non-invasively examine tissue of interest. Their low cost and small footprint make "optical spectroscopy methods" an excellent tool for conducting pilot studies in animal models of cancer and in humans. Since optical measurements using the fiber optic probe are non-invasive or minimally invasive (depending on the tissue site), the same subject can be monitored multiple times a day or over weeks to evaluate response to treatment. In addition to its obvious benefits as a clinical adjunct to existing clinical imaging modalities that cannot be used every day on patients, optical spectroscopy obviates the need for sacrificing large numbers of animals at several time points in longitudinal studies to evaluate treatment response. Here, we describe two specific techniques—diffuse reflectance and Raman spectroscopy—that have demonstrated potential for monitoring radiation response in tumors and studying the differences between resistant and sensitive tumors.

#### Diffuse Reflectance Spectroscopy

Diffuse reflectance or elastic scattering spectroscopy is an optical fiber- based technique for non-invasive interrogation of tissue. DRS uses optical fibers to deliver low-power non-ionizing light from a broad-band light source (400– 650 nm) to tissue surface. The incident weak light undergoes multiple scattering and absorption events and is remitted back to the tissue surface as diffusely reflected light. Since the collected light has interacted non-destructively with the tissue, it provides a wealth of quantitative information about absorption and scattering, a combination of which is used for tissue pathology. Using models of light-tissue interaction that simulate the travel of photons within a scattering and absorbing medium, it is possible to quantify the diffusely reflected light and extract meaningful information related to tissue scattering as well as prominent tissue absorbers, such as oxygenated and deoxygenated hemoglobin (65–70). By exploiting the differences in light absorption spectra of oxygenated and deoxygenated hemoglobin, we can quantify the vascular oxygen content in tissue and obtain volumeaveraged estimates of hemoglobin concentration. Measurements of vascular oxygenation have been shown to be concordant with microelectrode-based determinations of tissue oxygenation (71, 72) and immunohistochemical measurements of tumor hypoxia (73). Cell nuclei, mitochondria, and collagen are among the major contributors to light scattering in tissue and are known to undergo significant changes during disease progression (74). Taking advantage of these non-invasive and quantitative measurements, DRS has been used in several studies, with an eye toward clinical translation, for early cancer detection (75– 77), prediction of response to therapy (78–80), and evaluation of tumor surgical margins (81). Given the importance of tumor oxygenation in radiation therapy, DRS can provide a non-invasive approach to quantify the biological response to radiation. Vishwanath et al. used DRS to longitudinally monitor tumor oxygenation and determine whether vascular oxygenation can identify treatment outcome earlier than tumor growth assays in a murine model of head and neck cancer treated with single dose of 39 Gy radiation. As early as 5 days post-radiation, radiation-responsive tumors exhibited faster and greater increase in vascular oxygenation compared with nonresponding tumors (82). A more recent study from the same group found similar large increases in vascular oxygenation in both locally controlled and locally recurring tumors when the radiation dosage was split into five daily doses instead of a single

dose. Additionally, the study also found that within the locally recurring group of tumors, a faster increase in reoxygenation during therapy was negatively correlated with recurrence time (83). Diaz et al. recently used DRS to study short-term changes in vascular oxygen saturation and hemoglobin concentration in radiation-sensitive and resistant A549 tumors treated with 4 dose fractions of 2 Gy (84), and also found significantly higher reoxygenation in radiation-resistant tumors 24 and 48 h after treatment (**Figure 4**).

This study was the first to report changes in reoxygenation kinetics measured in tumors which were established from a matched model of radiation-resistance. A matched model of radiation-resistance allows direct comparison of resistancerelated features due to similar genetic background. While further studies are necessary to fully understand the mechanism of reoxygenation in the radiation-resistant tumors, results from other studies conducted using the same matched model of radiation resistance hint at the possibility of reduced oxygen consumption as a possible reason for the appearance of increased vascular oxygenation within the radiation-resistant tumors. Although the studies by Hu et al. (83) and Diaz et al. (84) used different cell lines in formation of tumor xenografts and treated them with different doses of radiation, they both showed that radiation-resistant tumors reoxygenate in response to radiation. These results are in agreement with a clinical study by Dietz et al. that used oxygen-sensing microelectrodes to measure pO<sup>2</sup> in the cervical lymph nodes of head and neck cancer patients and found that increased reoxygenation correlated with poor radiation response (85). This suggests that DRS is a sensitive detector of reoxygenation and can provide valuable information about radiation response.

#### Raman Spectroscopy

Raman spectroscopy offers the ability to probe biomolecular changes and visualize the complex molecular heterogeneity directly from cells and tissues (86, 87). Spontaneous Raman spectroscopy relies on the inelastic scattering of light, arising from its interactions with the biological specimen, to quantify the unique vibrational modes of molecules within its native context (88, 89). This exquisite chemical specificity of Raman spectroscopy has been exploited primarily within the realm of early detection of cancers of the oral cavity (90, 91), breast (92–98), cervix (99–101), and the brain (88, 102).

Recent studies have shown the presence of radiation-induced alterations in Raman spectral features and biochemical changes in cell lines with varying radiosensitivity (103, 104). The radiation response of single living cells has been studied to demonstrate dose-dependent changes in spectral features using principle component analysis (105, 106). In a series of human cancer cell lines treated with clinically relevant doses of radiation (<10 Gy), Matthews et al. found radiationinduced accumulation of intracellular glycogen in relatively radiation-resistant breast and lung cancer cell lines (107). Recent Raman spectroscopic studies on ex vivo lung and breast tumor xenografts have also identified elevated levels of glycogen in tumors exposed to a single, high radiation dose of 15 Gy (108, 109). These findings are of interest because separate nonimaging studies have identified a critical role for glycogen synthase kinase (GSK-3β) in the development of radiation resistance (110).

Radiation-induced changes in Raman spectra of excised cervical tumors have been shown to differentiate radiation responders from non-responders while pretreatment Raman spectra were incapable of predicting radiation response (111). In a recent study, Paidi et al. investigated whether radiation induced biomolecular changes detected by Raman spectroscopy could differentiate between radiation-resistant and sensitive tumors (112). They treated radiation-resistant and sensitive human head and neck (HN) and lung tumor xenografts with 2 Gy of radiation twice weekly for 2 weeks and conducted chemometric

showing radiation-induced differences in sensitive lung tumors (A549-NT vs. A549-XT) (NT: not treated, XT: X-ray treated) (A) and radiation-induced differences in resistant lung tumors (rA549-NT vs. rA549-XT) (B). (C,D), Boxplots of normalized scores of lipid-rich and collagen-rich MCR-ALS loadings showing radiation-induced differences in sensitive head and neck tumors (UM-SCC-22B-NT vs. UM-SCC-22B-XT) (C) and radiation-induced differences in resistant head and neck tumors (UM-SCC-47-NT vs. UM-SCC-47-XT) (D). The effect size (*r*), characterizing magnitude of differences between groups, is provided for each comparison. \**p* < 0.001. Adapted with permission from Paidi et al. (112).

analysis using multivariate curve resolution-alternating least squares (MCR-ALS) to uncover biomolecular changes in the tumor microenvironment. MCR-ALS recovers the pure spectral profiles of the chemical constituents of the tissue specimen without a priori information of the composition of the specimen (113). Paidi et al. found an increase in lipid, collagen, and glycogen (lung only) levels for both sensitive and resistant lung and head neck tumors that were treated with radiation, with a much larger increase in the lipidrich and collagen-rich signatures in the radiation-sensitive tumors (**Figure 5**) (112). Comparison of the treated tumors alone (RS-XT vs. RR-XT) pointed to a significantly higher collagen content in the sensitive tumors compared to their resistant counterparts in both lung and HN models, which could be attributed to radiation-induced fibrosis (114, 115). The lipid results are intriguing due to other studies that have found elevated levels of fatty acid synthase (FASN) in radiation-resistant cells (23). These findings demonstrate that clear spectral distinctions exist between radiation-resistant and sensitive tumors, and that these distinctions are consistent with recent work seeking to uncover the molecular mechanisms of radiation resistance.

### DISCUSSION AND FUTURE DIRECTION

The use of optical microscopy and diffuse optical spectroscopy presents exciting avenues for exploring radiation-induced changes across different length scales in cells and tissue. The technologies discussed in this review paper (summarized in **Table 1**)—although limited to superficial layers—are sensitive to two key hallmarks of tumors that play a critical role in radiation resistance—tumor hypoxia and metabolic reprogramming. While two-photon excited fluorescence from NADH and FAD can provide valuable information about specific metabolic pathways preferred by cells in response to radiation and the effect of such preferences on radiation resistance, Raman spectroscopy (or microscopy) can shed light on hitherto unknown biomolecular species in the tumor microenvironment that play a role in radiation resistance. Such studies have the potential to lead to new technologies


TABLE 1 | Comparison of optical microscopy and spectroscopy techniques for investigating radiation biology.

centered on specific biomarkers for continuous monitoring during radiation treatment. Additionally, these studies can lead to the identification of novel therapeutic targets that can be exploited to possibly reverse radiation resistance. While optical spectroscopy has been at the forefront of optical technologies attempting to break into the clinical workflow, more work is required to establish baseline optical endpoints and the accuracy and reproducibility of these measurements. In addition, it will be necessary to associate these changes with specific outcomes corresponding to treatment response or failure. Optical spectroscopy has faced challenges with clinical translation, with attempts at early detection of cancer, discrimination between benign and malignant cancer, and demarcation of surgical margins not acquiring enough traction. The principal concerns in these clinical workflows was the perception that optical spectroscopy could never replace pathology, which is currently standard-of-care for these clinical problems. A possible advantage of advancing optical spectroscopy for measuring tumor response to therapy is the complete lack of any imaging technology or treatment biopsies that currently evaluate treatment response during the treatment regimen. While other imaging modalities such as optoacoustic imaging (OAI) can measure tumor oxygenation (116, 117), they have not yet been used in the context of radiation resistance. If decisions to escalate or de-escalate treatment for exceptional treatment responders or non-responders are to be made based on endpoints provided by optical techniques,

#### REFERENCES


near-perfect identification of treatment response within the first 1–2 weeks will be necessary to effect meaningful change. Tromberg et al. have demonstrated the ability of optical spectroscopy to provide an early indicator of chemotherapy response in breast cancer (78–80). Recent work has also significantly advanced the translation of non-linear optical microscopy from a laboratory-only method to the clinic for imaging the skin (118). The ability to translate two-photon excited autofluorescence from NADH and FAD to clinically compatible technologies, such as fiber optic probes could allow simultaneous determination of cellular redox state and mitochondrial fractal dimension in vivo. When combined with other information from DRS and RS, such as vascular oxygenation and biomolecular content, optical techniques could provide a powerful addition to a clinical workflow that could greatly benefit patients by improving response rates and quality of life.

#### AUTHOR CONTRIBUTIONS

SD and NR wrote the manuscript.

#### FUNDING

This work was supported by grants from the Arkansas Biosciences Institute and the National Science Foundation (1847347).


consumption in precancerous epithelial tissues. Cancer Res. (2014) 4:3067–75. doi: 10.1158/0008-5472.CAN-13-2713


fluorescence lifetime imaging of the coenzyme NADH. Cancer Res. (2005) 65:8766–73. doi: 10.1158/0008-5472.CAN-04-3922


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Dadgar and Rajaram. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Difference in Acquired Radioresistance Induction Between Repeated Photon and Particle Irradiation

#### Katsutoshi Sato<sup>1</sup> , Takashi Shimokawa<sup>2</sup> and Takashi Imai <sup>3</sup> \*

*<sup>1</sup> Division of Hematology and Medical Oncology, Icahn School of Medicine at Mount Sinai, The Tisch Cancer Institute, New York, NY, United States, <sup>2</sup> Department of Accelerator and Medical Physics, National Institute of Radiological Sciences, National Institutes for Quantum and Radiological Sciences and Technology, Chiba, Japan, <sup>3</sup> Medical Databank, Department of Radiation Medicine, QST Hospital, National Institutes for Quantum and Radiological Science and Technology, Chiba, Japan*

In recent years, advanced radiation therapy techniques, including stereotactic body radiotherapy and carbon–ion radiotherapy, have progressed to such an extent that certain types of cancer can be treated with radiotherapy alone. The therapeutic outcomes are particularly promising for early stage lung cancer, with results matching those of surgical resection. Nevertheless, patients may still experience local tumor recurrence, which might be exacerbated by the acquisition of radioresistance after primary radiotherapy. Notwithstanding the risk of tumors acquiring radioresistance, secondary radiotherapy is increasingly used to treat recurrent tumors. In this context, it appears essential to comprehend the radiobiological effects of repeated photon and particle irradiation and their underlying cellular and molecular mechanisms in order to achieve the most favorable therapeutic outcome. However, to date, the mechanisms of acquisition of radioresistance in cancer cells have mainly been studied after repeated *in vitro* X-ray irradiation. By contrast, other critical aspects of radioresistance remain mostly unexplored, including the response to carbon-ion irradiation of X-ray radioresistant cancer cells, the mechanisms of acquisition of carbon-ion resistance, and the consequences of repeated *in vivo* X-ray or carbon-ion irradiation. In this review, we discuss the underlying mechanisms of acquisition of X-ray and carbon-ion resistance in cancer cells, as well as the phenotypic differences between X-ray and carbon-ion-resistant cancer cells, the biological implications of repeated *in vivo* X-ray or carbon-ion irradiation, and the main open questions in the field.

Keywords: cancer, radioresistance, acquisition, X-ray radiation, carbon-ion radiation, repeated irradiation, DNA repair, aggressiveness

#### INTRODUCTION

The previous decade has seen significant developments in the techniques used in radiotherapy, and advanced radiotherapy has become increasingly adopted. Among advanced radiotherapy techniques, stereotactic body radiation therapy (SBRT) relies on a small irradiation field to precisely deliver high doses of radiation, typically above 10 Gy per fraction, to local tumors. SBRT has been applied to the treatment of various cancers, including lung (1), liver (2), and prostate cancer (3).

#### Edited by:

*Ira I. Skvortsova, Innsbruck Medical University, Austria*

#### Reviewed by:

*Zhongxing Liao, University of Texas MD Anderson Cancer Center, United States Michael W. Epperly, University of Pittsburgh, United States*

> \*Correspondence: *Takashi Imai imai.takashi@qst.go.jp*

#### Specialty section:

*This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology*

Received: *14 August 2019* Accepted: *23 October 2019* Published: *12 November 2019*

#### Citation:

*Sato K, Shimokawa T and Imai T (2019) Difference in Acquired Radioresistance Induction Between Repeated Photon and Particle Irradiation. Front. Oncol. 9:1213. doi: 10.3389/fonc.2019.01213*

**103**

The therapeutic outcomes of SBRT are particularly promising for early stage lung cancers, with local control rates exceeding 80% (1, 4) and clinical outcomes comparable to those of surgical resection (5, 6).

In addition to SBRT, particle-beam therapy, such as carbonion (C-ion) radiotherapy (CIRT), has demonstrated excellent therapeutic outcomes in various types of cancer (7). CIRT has both physical and biological advantages compared with Xray therapy. With CIRT, tumors are irradiated with C-ions of relativistic energy, which means that C-ions penetrate the body with lower ionization that significantly increases toward the end of the beam path. This radiophysical feature is commonly referred to as the "Bragg-peak" and contributes to increasing the radiation dose delivered to the tumor while minimizing the radiation dose delivered to the skin and surrounding healthy tissues. Furthermore, CIRT has a relative biological effectiveness, which is defined as the ratio of a dose of radiation to the dose of X-ray radiation producing the same biological effects that is >2. Another important feature of CIRT is its effectiveness against conventional X-ray radiotherapy resistant cancers, including melanoma (8) and bone and soft-tissue sarcoma (9–11). Furthermore, CIRT reportedly works as an alternative ablative treatment for early stage lung cancer, in particular for elderly and inoperable patients (12).

Nevertheless, recent studies show that local recurrence can still occur after advanced radiotherapy. For example, an incidence of local recurrence ranging from 4.9 to 19% in patients who received SBRT for lung cancer treatment was reportedly dependent on treatment regimen, tumor stage, and follow-up periods (13–18). Furthermore, 23.3% of patients who received CIRT for the treatment of stage I non-small cell lung cancer also exhibited local recurrence (19).

In cases of tumor recurrence after primary radiotherapy, patients can rarely be treated again with the same radiation regimen, because the tumor might acquire radioresistance, and it is possible that healthy surrounding tissues will not tolerate additional irradiation. Nevertheless, recent studies report that SBRT and CIRT can be used for re-irradiation of recurrent tumors, taking into account both the dose tolerance of healthy tissues and location of the recurrent tumor (20–23). However, several issues related to re-irradiation with SBRT or CIRT still need to be considered. First, only a few studies have reported the clinical outcomes of repeated irradiation, and second, the characteristics of recurrent tumor after primary radiotherapy are largely unknown. In this review, we focus on the biological aspects of acquired X-ray and C-ion resistance in cancer cells and discuss the differences between the consequences of in vitro and in vivo repeated irradiation, the possible mechanisms of acquired resistance in cancer cells, and issues that must be addressed in this research field.

### ACQUISITION OF PHOTON RADIORESISTANCE IN VITRO

Radioresistance acquisition in cancer cells and its underlying mechanisms have been mainly studied using radioresistant cell lines established through repeated in vitro photon (e.g., X-ray or γ-ray) irradiation. Because conventional radiotherapy usually relies on a total dose of ∼60 Gy applied in 2-Gy fractions, many studies adopted similar radiation regimens in order to establish radioresistant cell lines (**Table 1**) (24–41). Importantly, most of these studies showed that the survival of repeatedly irradiated cells was significantly higher than that of the parental cells, which indicated that in vitro, various cell lines could acquire radioresistance following multiple rounds of X-ray irradiation.

### Cellular Processes Involved in the Acquisition of Radioresistance Following Repeated Photon Irradiation

The mechanisms of acquisition of radioresistance in cancer cells have been associated with a variety of biological processes (**Table 1**). However, in many cases, the acquisition of radioresistance can be reasonably explained by the induction of epithelial-to-mesenchymal transition (EMT), which is defined as phenotypic and molecular alterations that result in the loss of epithelial-cell characteristics and the gain of mesenchymal-cell characteristics. As cancer cells undergo EMT, epithelial markers, such as E-cadherin, ZO-1, and cytokeratin, are downregulated, whereas mesenchymal markers, such as N-cadherin, vimentin, snail, and twist, are upregulated, and in some cases, morphological changes lead to the appearance of spindle-shaped cells (42). The most prominent characteristics acquired by cancer cells after EMT are migratory and invasive properties, conferring them a significant metastatic potential, and resistance to chemotherapeutic drugs and ionizing radiation. Indeed, several studies report that EMT in cancer cells, which was defined by reduced E-cadherin protein levels, increased Ncadherin protein levels, and enhanced migration potential, could be induced by single X-ray irradiation (43, 44). Furthermore, Shintani et al. (39) showed that repeated X-ray irradiation of A549 cells (2 Gy weekly for >6 months) induced significant radioresistance and typical EMT (i.e., decreased E-cadherin and increased N-cadherin mRNA and protein levels). Collectively, these data support the notion that EMT induction following X-ray irradiation is a contributing factor in the acquisition of radioresistance.

Another factor involved in the acquisition of radioresistance following repeated X-ray irradiation is an enrichment in cancer stem cells (CSCs), which are known to exhibit higher DNArepair potential (45) and resistance to reactive oxygen speciesinduced cytotoxicity (46, 47). CSCs are also often found to be in the G0 phase (48), a quiescent state outside the normal cell cycle and associated with reduced cell proliferation. All of these characteristics are recognized for their role in cellular radioresistance. Additionally, CSCs are important in radioresistance acquisition because of their impact on the heterogeneity of the cell population within a tumor. Indeed, in the hierarchy model, CSCs produce a more differentiated non-CSC progeny exhibiting significant cell-proliferation potential but lacking stem cell properties. Notwithstanding their reduced proliferation rate as compared with their non-CSC progeny, CSCs can self-renew, and maintaining their stemness. Notably, TABLE 1 | Repeated photon irradiation regimen for the establishment of radioresistant cancer cells.


*(*\**1) There is no description about treatment regimen.*

*(*\**2) There is no significant finding other than radioresistance acquisition.*

*(*\**3) 5 Gy and 30 Gy of single and total dose is that of C-ion irradiation.*

the non-CSC population displays higher radiosensitivity than the CSC population. Consequently, radiation treatment can increase the relative abundance of CSCs in the tumor, which promotes asymmetric cell proliferation and, therefore, an enrichment in CSCs.

Lagadec et al. (49) reported enrichment in CD44high/CD24low breast CSCs following repeated X-ray irradiation of human breast cancer MCF7 and T47D cells, with irradiated cells displaying increased sphere-formation potential. Furthermore, they found that repeated irradiations led CSCs in the G0 phase to reenter the cell cycle, thereby promoting their proliferation, whereas the non-CSC population underwent apoptosis according to the increased fraction of cells in the sub-G1 phase (49). Additionally, Ghisolfi et al. (50) showed that single X-ray irradiation of cancer cells with a dose of 2 Gy to 10 Gy increased the expression of the pluripotency markers OCT3/4 and SOX2 and promoted the enrichment of a CSC subpopulation. Moreover, Mani et al. (51) established a link between EMT and CSCs by demonstrating that TGF-β-induced EMT generated a subpopulation with CSC properties, including characteristic CSC markers, such as CD44high/CD24low and elevated sphere- and mammosphereformation potential.

To the best of our knowledge, a definitive mechanism responsible for the induction of CSCs remains unclear; however, DNA damage or chromosomal aberration can enhance CSC induction along with increased oncogene activity. Liang et al. (52) showed that DNA damage from UV irradiation and the chromosomal aberrations induced by Mad2 overexpression also increased by Myc and SOX2 expression in human nasopharyngeal carcinoma CNE cell lines and promoted cell dyeexclusion, colony formation, and sphere-formation capacities. These data suggest that the accumulation of DNA damage by repeated X-ray irradiation induces not only EMT but also enrichment of CSCs with increasing oncogenic activity, whereas secondary induction of a CSC subpopulation by EMT (known as cancer plasticity) further contributes to the development of radioresistance.

#### Molecular Processes Involved in the Acquisition of Radioresistance Following Repeated Photon Irradiation

We and others have independently reported that repeated X-ray irradiation can result in enhanced DNA-repair capacity (24, 29, 33, 34). In our study, the mouse squamous cell carcinoma NR-S1 cell line was irradiated with a total dose of 60 Gy of X-ray radiation applied in 10-Gy fractions in order to establish the X60 radioresistant cancer cell line (**Figure 1**). Notably, the D10 value (i.e., the radiation dose required to decrease the survival to 10% of the non-irradiated condition) and cell survival after 10 Gy of X-ray radiation were 1.6- and 3.8-fold higher, respectively, for X60 cells than for parental NR-S1 cells (34). Furthermore, 24 h after exposure to 10 Gy X-ray radiation, the number of S139 phosphorylated-H2AX (γ-H2AX) foci, a marker of DNA doublestrand breaks (DSBs), was 2.5-fold lower in X60 cells than in NR-S1 cells, indicating that DSBs were repaired more efficiently in X60 cells than in NR-S1 cells (34). Indeed, the collected results of numerousstudies (**Table 1**) further demonstrate that enhanced DNA-repair capacity is a common feature of radioresistant cancer cells arising from repeated X-ray irradiation.

As part of the investigation of the molecular mechanisms underlying the acquisition of radioresistance, several groups have highlighted a relationship between DNA repair and prosurvival signaling pathways, such as the Akt and mechanistic target of rapamycin (mTOR) pathways. Shimura et al. (36) suggested a potential molecular mechanism for the acquisition of radioresistance induced by repeated X-ray irradiation, showing that cyclin D1 expression and Akt phosphorylation levels were increased in X-ray-resistant derivatives of HeLa and HepG2 cells established following repeated irradiation. These radioresistant cancer cell lines also displayed constitutively elevated levels of DSBs, as measured by H2AX and ataxia telangiectasia mutated (ATM) phosphorylation, relative to those in parental cell lines. Strikingly, downregulation of cyclin D1 in radioresistant HeLa and HepG2 derivatives decreased H2AX-, ATM-, and Aktphosphorylation levels, as well as cell survival, after further Xray irradiation. Therefore, they proposed that repeated X-ray irradiation triggered cyclin D1 overexpression and forced cell cycle progression, which in turn caused further DNA damage and led to the activation of both Akt signaling and DNAdependent protein kinase activity, a central component in the non-homologous end joining DSB-repair pathway. Eventually, these signals promoted further cyclin D1 overexpression as part of a positive-feedback loop that likely resulted in the acquisition of radioresistance (36, 38, 53, 54).

In addition to Akt signaling, mTOR signaling has been associated with the acquisition of X-ray resistance in cancer cells. Chang et al. (55) established radioresistant derivatives of PC-3, DU145, and LNCaP cells following repeated Xray irradiation with a dose of 2 Gy/day for 5 consecutive days (55) and showed that these radioresistant cancer cell lines exhibited both mesenchymal and CSC phenotypic traits. Interestingly, they found that radioresistant cells treated with BEZ235, a specific inhibitor of the phosphoinositide 3 (PI3) and mTOR kinases, displayed decreased expression of mesenchymal (N-cadherin, vimentin, and snail) and CSC (OCT3/4, SOX2, and CD44) markers, increased expression of the epithelial marker E-cadherin, and reduced cell survival. They further showed that BEZ235, the PI3-kinase inhibitor BKM120, and the mTOR-kinase inhibitor rapamycin, suppressed the expression

cells were irradiated six times at 2-week intervals with 10 Gy of X-ray radiation (left) or 5 Gy of C-ion radiation (left). The radioresistant derivative cell lines exposed to total doses of 60 Gy of X-ray radiation and 30 Gy of C-ion radiation were denoted as X60 and C30 cells, respectively (34, 35).

of DNA-repair proteins induced by X-ray irradiation, including Ku80, BCRA1, and BRCA2 (56). Moreover, we independently demonstrated that mTOR signaling was enhanced in X60 radioresistant cancer cells as compared with parental NR-S1 cells, whereas rapamycin treatment decreased their radioresistant phenotype (35). Importantly, rapamycin also suppresses both non-homologous end joining and homologous recombination (HR) DSB-repair pathways (57). Collectively, these results indicate that activation of the pro-survival Akt and mTOR signaling pathways can eventually increase the DNA-repair capacity of repeatedly irradiated cancer cells and thereby promote the acquisition of radioresistance.

#### ACQUISITION OF C-ion RADIORESISTANCE IN VITRO

#### Acquisition of C-Ion Resistance Following Repeated X-Ray Irradiation

In clinical practice, CIRT is an effective treatment for locally recurrent tumors after primary radiotherapy, likely because cell killing by C-ion radiation is independent of various cellular or tumor characteristics, including p53 status (58), cell cycle phase (59, 60), and hypoxia (61–63). Although these findings suggest that C-ion radiation should be effective against radioresistant cancer cells arising from repeated X-ray irradiation, no available experimental data supported this hypothesis. Therefore, in a recent study, we determined whether C-ion radiation could efficiently kill X60 radioresistant cancer cells. Contrary to our expectations, we found that compared with parental NR-S1 cells, X60 cells exhibited significant levels of resistance against C-ion radiation (34). Furthermore, 24 h after C-ion irradiation, the number of γ-H2AX foci was 2.5-fold lower in X60 cells than in NR-S1 cells. These observations indicated that repeated X-ray irradiation of cancer cells with a relatively high dose of 10 Gy per fraction could induce not only X-ray resistance but also Cion resistance (34). We believe that further investigations using such radioresistant cells will likely lead to the discovery of novel mechanisms contributing to C-ion resistance in cancer cells.

To gain further insight into the underlying mechanisms of radioresistance in X60 cells, in a recent study, we compared several biological and morphological traits of X60 cells and parental NR-S1 cells, including cell shape and size, number of heterochromatin domains in the nucleus, and DNA content (34). Additionally, we analyzed the correlation between these factors and X-ray or C-ion resistance. Interestingly, we found that the number of heterochromatin domains was strongly correlated with both X-ray and C-ion resistance, which suggested that heterochromatin components or the dynamics of heterochromatin were also involved in the acquisition of radioresistance.

Indeed, previous studies show that heterochromatin proteins, such as HP1α (64) and CAF1 (65), are directly involved in DNA repair, and recent studies report that DNA damage in heterochromatin domains is mainly repaired by the HR machinery (66, 67). The damaged heterochromatin first move to the periphery of the heterochromatin domain to prevent abnormal recombination or deleterious expansion at satellite or repetitive DNA sequences, after which Rad51, the core component of the HR machinery, accumulates at DNA-damage sites located at the periphery of the heterochromatin domain (66, 67). Although these studies were conducted in Drosophila and yeast cells, Jakob et al. (68) observed similar dynamics of damaged heterochromatin and DNA-repair following heavy ionbeam irradiation of mammalian cells, finding that immediately after irradiation, DSBs were formed in the heterochromatin along the uranium ion-beam track. Within 30 min, the DSB sites relocated to the periphery of the heterochromatin domain, and replication protein A, a marker of DNA-end resection during HR repair, accumulated at these DSB sites (68). In agreement with these findings, our preliminary results showed that 1 h after Xray or C-ion irradiation, X60 cells displayed an increased number of Rad51 foci as compared with NR-S1 cells (**Figure 2**). Because Rad51 is a central factor in the HR machinery, Rad51 foci are commonly considered to represent sites of ongoing DSB repair by HR.

Furthermore, several studies report that HR repair can remove complex DNA lesions, including clustered DNA damage (69) and DNA-protein crosslinks (70, 71). Given that X60 cells exhibit increased numbers of heterochromatin domains and radiation-induced Rad51 foci as compared with NR-S1 cells (**Figure 2**), it appears conceivable that enhanced HRrepair capacity and heterochromatin dynamics contribute to X-ray and C-ion resistance in X60 cells. Whether such a mechanism of acquisition of radioresistance could be specific to X60 cells or shared by other radioresistant cancer cell lines remains to be explored. Nevertheless, future studies focusing on further our understanding of the underlying mechanisms of DSB repair in heterochromatin domains could, therefore, represent a significant breakthrough toward elucidating the mechanisms of acquisition of radioresistance in cancer cells and identifying novel therapeutic strategies for the treatment of radioresistant tumors.

### Acquisition of C-Ion Resistance Following Repeated C-Ion Irradiation

To date, only a limited number of studies have investigated the acquisition of C-ion resistance, and it remains unclear whether repeated C-ion irradiation can lead to radioresistance in cancer cells. To address this question, in a recent study, we irradiated NR-S1 cells with a total dose of 30 Gy of C-ion radiation applied in 5-Gy fractions in order to establish a C30 radioresistant cancer cell line (**Figure 1**). This C-ion irradiation regimen is biologically equivalent to the X-ray irradiation regimen used to establish the X60 cell line, because NR-S1 cells exhibit comparable survival following exposure to 5 Gy and 10 Gy of C-ion and X-ray radiation, respectively. Interestingly, when we assessed the Xray and C-ion radiation sensitivity of C30 cells using a colony formation assay, we found that C30 cells displayed moderate resistance to C-ion radiation but not to X-ray radiation (35).

Although it remains unclear why repeated X-ray irradiation but not repeated C-ion irradiation conferred significant C-ionresistance to NR-S1 derivative-cells, we believe that enrichment

in CSCs might be a contributing factor. Indeed, it is widely recognized that CSCs are more resistant to X-ray radiation and anticancer drugs than more differentiated cancer cells. Furthermore, as noted, several studies report that repeated Xray irradiation can increase the CSC fraction within a cancer cell population (49, 50, 72–74). Conversely, Cui et al. (75) showed that C-ion radiation could efficiently kill CSCs both in vitro and in vivo. Therefore, it is conceivable that repeated X-ray irradiation but not repeated C-ion irradiation could contribute to enriching a radioresistant subpopulation with CSC-like characteristics. However, there is one element in contradiction to this hypothesis. Because most CSCs are in a G0 quiescent state (45), the non-CSC subpopulation might proliferate faster; therefore, if the primary factor in the acquisition of X-ray and Cion resistance is enrichment in CSCs, the level of radioresistance of a growing X60 cell population, for example, could gradually decrease over time, which was not observed (34, 35).

Collectively, these findings indicate that CSC enrichment and other mechanisms, such as genetic alterations and mutations (76), might jointly contribute to the acquisition of both X-ray and C-ion resistance in cancer cells. Although further investigations are required to elucidate these mechanisms, the current body of evidence suggests that C-ion irradiation does not induce radioresistance and can be used for the treatment of locally recurrent tumors arising after primary CIRT.

### EFFECTS OF IN VIVO REPEATED PHOTON OR C-ion IRRADIATION

Numerous studies are focused on translating results obtained with in vitro models of radioresistant cancer cells into clinical practice. Therefore, it appears essential to determine whether phenotypic changes resulting from repeated photon or particle irradiation also occur in vivo. However, to the best of our knowledge, the effects of in vivo repeated photon or particle irradiation on the acquisition of radioresistance in cancer cells has not been reported.

To address this question and examine whether the characteristics of repeatedly irradiated tumors differ from those of parental tumors, we recently established in vivo models of regrown irradiated tumors (77). To this end, NR-S1-derived tumors engrafted into C3H/He mice were irradiated with single doses of γ-ray (30 Gy) or C-ion (15 Gy) radiation, which have comparable effects on tumor growth. Two weeks after irradiation, we harvested the irradiated tumors and transplanted them into healthy mice, and 2 weeks later, the regrown tumors were irradiated again, with this irradiation/regrowth/transplant process repeated six times in total. The resulting repeatedly irradiated tumors were exposed to total doses of 180 Gy of γ-ray radiation and 90 Gy of C-ion radiation and denoted as G180 and C90 in vivo regrown tumor models, respectively

(**Figure 3**). We then examined differences in tumor-growth potential, spontaneous metastasis from the primary site to the lung surface, tumor-grafted mouse survival, and radiosensitivity between non-irradiated NR-S1-derived tumors and G180 and C90 tumors.

Notably, G180 tumors displayed drastically increased tumorgrowth rates and metastatic potential compared with those of non-irradiated tumors, and mice grafted with G180 tumors displayed significantly shorter survival than those grafted with non-irradiated tumors. By contrast, the characteristics of the C90 tumors remained comparable to those of non-irradiated tumors. Importantly, X-ray and C-ion irradiation of G180 and C90 tumors did not affect the relative tumor-growth rates, spontaneous lung metastasis, and survival of tumor-grafted mice as compared with non-irradiated tumors. Furthermore, colony formation assays performed using cells isolated from non-irradiated, G180, and C90 tumors showed that they all added similar sensitivity to X-ray and C-ion radiation (77). Moreover, compared with non-irradiated and C90 tumors, G180 tumors harbored numerous microvessels and expressed genes associated with angiogenesis and metastasis, including VEGFA, HIF1A, FN1, MMP2, MMP9, PAI1, and PLAU. Together, these data indicated that, contrary to repeated in vitro irradiation, repeated in vivo γ-ray and C-ion irradiation did not lead to the acquisition of radioresistance in regrown tumors. However, repeated photon irradiation but not particle irradiation appeared to enhance tumor growth and metastasis, resulting in an increased aggressiveness of regrown tumors.

These findings could suggest that repeated irradiation affects the tumor microenvironment rather than the tumor itself or its CSC subpopulation. Although our investigations could not determine whether repeated photon irradiation enriched the CSC subpopulation in regrown tumors, our data showed that G180 tumor cells in suspension culture had a significantly higher sphere-formation potential than their non-irradiated and C90 counterparts (77). Because CSCs are characterized by significant resistance to cytotoxic agents, including radiation and anticancer drugs, we believe that G180 tumors are likely enriched in tumor-initiating cells (TICs) that differ from a typical CSC subpopulation. To date, such phenomena have not been reported, and further studies will be required to determine which cells among cancer and stromal cells are mostly affected by repeated in vivo irradiation and what mechanisms lead to increased aggressiveness in repeatedly irradiated tumors.

Crucially, these findings also demonstrated that repeated C-ion irradiation was far less prone to induce acquisition of radioresistance and enhance tumor aggressiveness, as assessed by tumor growth, metastatic potential, and prognosis of tumor-grafted mice. Although it remains necessary to ascertain why C-ion radiation effectively suppressed tumor aggressiveness and TIC or CSC subpopulations, we believe that the accumulated evidence supports CIRT as a promising treatment for local recurrent tumors.

### RELATIONSHIP BETWEEN THE 4Rs OF RADIOTHERAPY AND RADIORESISTANCE ACQUISITION

Tumor shrinkage by fractionated radiotherapy has been explained by the "4Rs" of radiotherapy, where each "R" represents "Repair," "Repopulation," "Redistribution," and "Reoxygenation" (78). "Repair" denotes the difference in cell survival of tumor cells and normal tissue between single or fractionated irradiation at the same radiation dose (79) and is basically measured by colony formation assay, followed by calculation of α and β values using a linear-quadratic model to quantify the radiosensitivity of each cell (80). "Repopulation" describes regeneration of normal tissue, such as skin and mucosal tissue (81). This concept relies on experimental results showing that the recovery of skin and mucosal tissue occurs faster than regrowth of gross tumor mass, and that each fractionation regime of radiotherapy can be determined based on these differences. Because α and β values and difference in recovery between tumor and normal tissue are used for treatment planning of radiotherapy, they are recognized as important therapeutic components. "Redistribution" indicates synchronization of the cell cycle in cells exposed to radiation. The tumor harbors multiple cell types exhibiting various cell cycle phases (82). Upon treatment of the tumor with radiotherapy, the relatively radiosensitive cell fractions, such as those in the G2/M phase, will die first, whereas the relatively radioresistant cell fractions, such as those in the G1 and S phases, will survive. However, cells surviving the first round of irradiation will enter a radiosensitive phase of the cell cycle during subsequent rounds and eventually will subsequently be efficiently killed. "Reoxygenation" describes changes in well-oxygenated areas of an irradiated tumor (83). Partial oxygen pressure in the peripheral tumor is higher than that in the center, because nutrition and oxygen at the periphery is well-supplied by tumor blood vessels (78). The partial oxygen pressure enhances the cellkilling effect, because radiation-induced reactive oxygen species, such as OH radical, initiated DNA breakage (84). Therefore, well-oxygenated areas of a tumor (i.e., the periphery) are killed first, followed by vascularization of the central tumor along with tumor shrinkage. Repetition of this process enhances the efficacy of radiotherapy. The 4Rs reasonably describe the process of tumor shrinkage during radiotherapy and are useful for recognizing tumor and environmental conditions the determine radioresistance or radiosensitivity.

On the other hand, our previous results suggest the possibility that use of the 4Rs might not be appropriate for planning secondary radiotherapy. At the very least, using the same definitions as those used to determine primary radiotherapy might not be suitable for secondary radiotherapy. If the tumor targeted for secondary radiotherapy has acquired radioresistance via EMT, the total dose required to control the tumor should be increased. This suggests that the α and β values of the tumor cells and the dose fractionation used to prevent normaltissue complications should be changed. Therefore, this suggests that "Repair" and "Repopulation" should be properly adjusted in the planning of secondary radiotherapy. In cases where primary radiotherapy fails to control tumor growth, followed by tumor regrowth within the irradiation field, this suggests that radioresistant cancer cells, such as CSCs, likely exist in the target area. If these tumors are treated with another round of irradiation, "Reoxygenation" might not be suitable for interpreting tumor radiosensitivity, because the CSCs might be in a quiescent state and capable of surviving within the hypoxic area. In addition to the induction of the radioresistant cancer cells, our data showed that repeated photon irradiation in vivo promoted acquisition of a more aggressive phenotype in the tumors. These characteristic changes do not fit the classical 4Rs of radiotherapy. Although our results were obtained by experiments using mouse cancer cell lines rather that human specimens, and the results in vitro did not match those obtained in vivo, they indicated that other hallmarks are required to interpret possible radioresistant or aggressive fractions in target tumors for planning secondary radiotherapy targeting regrown tumors. Given that hypoxic areas are primary niches of CSCs (85), and the tumor vasculature clearly changes after irradiation (86), imaging techniques used to identify hypoxic areas and well-vascularized areas in target tumors will likely be useful for planning secondary radiotherapy. Indeed, drugs targeting hypoxic areas have been developed (87, 88), and tumor blood vessels can be visualized by contrast-enhanced magnetic resonance imaging (89).

#### WHAT ARE THE MAIN OPEN QUESTIONS IN THE FIELD?

Although numerous studies show that repeated X-ray irradiation can lead to increased radioresistance in various cancer cells, the primary source of radioresistant cancer cells remains elusive. Therefore, it is imperative to determine whether the selection of inherently radioresistant cells or emergence of radioresistant cells due to de novo genetic alterations is the primary cause of the acquisition of radioresistance in order to prevent the appearance of radioresistance and improve patient care. With the recent development of genetic barcoding techniques (90–93), we can now label a large number of cells within a given population. Combined with high-throughput DNA sequencing, genetic barcoding could be used to track and identify the type(s) of cells that can survive and proliferate following repeated irradiation.

Although extensive efforts have been made to investigate the mechanisms of acquisition of radioresistance in cancer cells using in vitro models, the radiobiological effects of repeated in vivo irradiation remain poorly understood. Tumors are complex ecosystems comprising various cell types, including cancer, stromal, and immune cells. Furthermore, the tumor environment is partly heterogeneous, with hypoxic or nutrientdeprived areas. In this regard, our in vivo data suggest that changes in the tumor microenvironment, including angiogenesis, might be critical for the prognosis of mice bearing regrown tumors after repeated irradiation.

In addition to the limited amount of data available concerning the effects of repeated in vivo irradiation, the differences between photon and particle irradiation remain largely unknown. Indeed, photon and particle radiation have distinct physical properties, and their resulting biological effects might be very different. For example, particle radiation produces a high density of reactive oxygen species and clustered DNA damage along the particlebeam track (94). Nevertheless, there is, to date, no sensible theory linking the physical characteristics of radiation to their biological effects, including high relative biological effectiveness in cellsurvival assays and suppression of metastasis both in vitro and in vivo.

Answers to these questions would definitely promote the understand of why repeated C-ion irradiation does not appear to induce significant radioresistance in cancer cells, whereas repeated X-ray irradiation leads to significant resistance to both X-ray and C-ion radiation in NR-S1 cells. The identification of potential targets to enhance or elicit radiosensitization could facilitate the development of novel therapeutic strategies for the treatment of radioresistant tumors and recurrent tumors after primary radiotherapy.

### CONCLUDING REMARKS

To date, few reports have been published describing the acquisition of radioresistance in repeatedly irradiated tumor cells, particularly after particle irradiation. A series of experiments using models of tumors repeatedly irradiated with either photon or particle radiation show that repeated in vitro irradiation with relatively high doses of X-ray radiation can induce significant

#### REFERENCES


resistance to both X-ray and C-ion radiation. By contrast, repeated in vitro irradiation with relative biological effectiveness doses of C-ion radiation does not contribute to the acquisition of X-ray or C-ion resistance in tumor cells. Somewhat surprisingly, repeated X-ray or C-ion irradiation of in vivo regrown tumor models does not increase their radioresistance; however, repeated photon irradiation but not C-ion irradiation increased tumor aggressiveness. Because the evidence was limited to a single tumor cell type, further studies are required to conclusively determine the effects of repeated irradiation on the acquisition of radioresistance in tumors.

### AUTHOR CONTRIBUTIONS

KS contributed to the design and wrote the manuscript. TS and TI rewrote and made edits. TI conducted the final review.

### FUNDING

This work was supported by JPSP KAKENHI Grant Nos. 15K15467 and 24591857, and by the Center of Innovation Program (COI stream from JST) and P-CREATE grant from AMED, and was performed as a research project with heavy ions at NIRS-HIMAC (No. 15J183).

### ACKNOWLEDGMENTS

The authors would like to thank their colleagues in ex-Research Center for Charged Particle Therapy at the National Institute of Radiological Sciences. This research was supported in part by the Research Project Heavy Ions at NIRS-HIMAC. We also thank Editage (www.editage.com) for English language editing.


using a gadolinium-dendron modified nano-liposomal contrast agent with magnetic resonance micro-imaging. Nanomedicine. (2018) 14:1315–24. doi: 10.1016/j.nano.2018.03.006


94. Tobias F, Durante M, Taucher-Scholz G, Jakob B. Spatiotemporal analysis of DNA repair using charged particle radiation. Mutat Res. (2010) 704:54–60. doi: 10.1016/j.mrrev.2009.11.004

**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Sato, Shimokawa and Imai. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

## Targeting the Innate Immune Kinase IRAK1 in Radioresistant Cancer: Double-Edged Sword or One-Two Punch?

#### Peter H. Liu1,2 and Samuel Sidi 1,2,3 \*

*<sup>1</sup> Division of Hematology and Medical Oncology, Icahn School of Medicine at Mount Sinai, Tisch Cancer Institute, New York, NY, United States, <sup>2</sup> Department of Cell, Developmental and Regenerative Biology, The Graduate School of Biomedical Sciences, Icahn School of Medicine at Mount Sinai, New York, NY, United States, <sup>3</sup> Department of Oncological Sciences, Icahn School of Medicine at Mount Sinai, New York, NY, United States*

#### Edited by:

*Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands*

#### Reviewed by:

*James W. Jacobberger, Case Western Reserve University, United States Marleen Ansems, Radboud Institute for Molecular Life Sciences, Netherlands*

> \*Correspondence: *Samuel Sidi samuel.sidi@mssm.edu*

#### Specialty section:

*This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology*

Received: *17 July 2019* Accepted: *18 October 2019* Published: *13 November 2019*

#### Citation:

*Liu PH and Sidi S (2019) Targeting the Innate Immune Kinase IRAK1 in Radioresistant Cancer: Double-Edged Sword or One-Two Punch? Front. Oncol. 9:1174. doi: 10.3389/fonc.2019.01174* Antitumor immunity has emerged as a favorable byproduct of radiation therapy (RT), whereby tumor-associated antigens released from irradiated cells unleash innate and adaptive attacks on tumors located both within and outside the radiation field. RT-induced immune responses further provide actionable targets for overcoming tumor resistance to RT (R-RT); immunotherapy (IT) with checkpoint inhibitors or Toll-like receptor (TLR) agonists can markedly improve, if not synergize with, RT in preclinical models, and several of these drugs are currently investigated as radiosensitizers in patients. In an unbiased chemical-genetic screen in a zebrafish model of tumor R-RT, we unexpectedly found that Interleukin 1 Receptor-Associated Kinase 1 (IRAK1), a core effector of TLR-mediated innate immunity, also functions in live fish and human cancer models to counter RT-induced cell death mediated by the PIDDosome complex (PIDD-RAIDD-caspase-2). IRAK1 acting both as a driver of intrinsic tumor R-RT and as an effector of RT-induced antitumor immunity would, at first glance, pose obvious therapeutic conundrums. IRAK1 inhibitors would be expected to sensitize the irradiated tumor to RT but simultaneously thwart RT-induced antitumor immunity as initiated by stromal dendritic cells. Conversely, TLR agonist-based immunotherapy would be expected to intensify RT-induced antitumor immunity but at the expense of fueling IRAK1-mediated cell survival in the irradiated tumor. We discuss how IRAK1's differential reliance on catalytic activity in the radiation vs. TLR responses might help overcome these hurdles, as well as the crucial importance of developing IRAK1 inhibitors that lack activity against IRAK4, the kinase activity of which is essential for IRAK1 activation in both pathways.

Keywords: IRAK1, irak4, radiation therapy, immunotherapy, radiosensitiser

#### IRAK1: A CORE EFFECTOR IN IL-1R/TLR INNATE IMMUNE SIGNALING

IRAK1 is a conserved death domain (DD)-containing protein kinase whose Drosophila homolog, pelle, transduces dorsoventral patterning and microbial cues recognized by the transmembrane receptor, Toll (1–6). The discovery of a Toll-like receptor (TLR) family of proteins in humans (3), composed of 10 TLRs, was soon followed by the finding that, as in flies, TLRs are responsible for the innate response to microbial infection through binding to pathogen- and damage-associated molecular patterns (PAMPs and DAMPs) and viral/bacterial nucleic acids in the intracellular space (endosomal TLRs). These discoveries were awarded the 2011 Nobel Prize in Physiology or Medicine (3).

Upon ligation, TLRs and IL-1R receptor (IL-1R/TLR) signal proinflammatory and cell survival responses, the majority of which through IRAK1/4 kinases and attendant downstream signaling cascades such as NF-kB, p38/MAPK, and JNK (3, 7) (**Figure 1A**). IRAK1 and IRAK4 are recruited to the ligated receptor by the Toll/IL-1R homology (TIR) and DD-containing adaptor protein, Myeloid Differentiation Primary Response 88 (MyD88) (8). MyD88 engages in homotypic TIR:TIR and DD:DD interactions with IL-1R/TLR and IRAK1/4, respectively, mobilizing the kinases to the receptor and resulting in the formation of the "MyDDosome" (9) complex (MyD88-IRAK4-IRAK1) (10) (**Figure 1A**). Only once in the MyDDosome, comprising six MyD88, four IRAK4, and four IRAK1 subunits (11), can IRAK4 dimerize. This proximityinduced dimerization of IRAK4 is the key initiating step in IRAK1 activation, with most (10, 12–16) but not all (17) models involving trans autophosphorylation of IRAK4 and ultimately phosphorylation of T209 on IRAK1 by fully active IRAK4. Once primed for activation by T209 phosphorylation, IRAK1 autophosphorylates on T387 in its activation loop, resulting in full activation, dissociation from the complex, and activation of downstream pathways (**Figure 1A**) (10, 13). IRAK1 activation also notably involves the peptidyl prolyl cis/trans isomerase PIN1, whose binding to IRAK1 is required for activation within, and dissociation from, the MyDDosome, and is overall essential for TLR signaling (**Figure 1A**) (18). Surprisingly, whether the catalytic activity of IRAK1 is required at any step for its function remains unclear (5, 17, 19), with genetic studies involving kinase-dead variants questioning reliance on catalytic activity (4, 6, 19–23). Consistent with this, engagement of three major signaling branches downstream of IRAK1, namely NF-κB, p38/MAPK, and JNK, relies on physical contact between activated IRAK1 and TNF receptorassociated factor 6 (TRAF6), independently of IRAK1 catalytic activity (**Figure 1A**) (3, 4, 21, 24). The relative importance of catalytic vs. structural functionalities of IRAK1 is an important consideration for the development of IRAK1 inhibitors for clinical use, particularly in radioresistant cancer, and will be discussed in detail in the closing sections of this review.

FIGURE 1 | IRAK1 kinase drives distinct prosurvival responses to microbial infection and ionizing radiation. (A) Diagram of the TLR signaling cascade which stimulates immune cell survival and inflammation in response to pathogen sensing. Ligated TLRs recruit MyD88 to trigger Myddosome (MyD88-IRAK4-IRAK1) formation, resulting in the activation of IRAK1 and release of the kinase from the complex. In turn, the activated form of IRAK1 binds TRAF6 to enable TRAF6-mediated activation of multiple pathways involved in anti-apoptotic and pro-inflammatory signaling. (B) Diagram of the newly identified IRAK1 signaling pathway triggered by IR, which involves IRAK4 but not MyD88 and antagonizes apoptosis through a different route involving inhibition of PIDDosome formation. Note that while IRAK1 catalytic activity is required in the radiation response (as symbolized by a green glare), it is dispensable for microbial responses relying on TRAF6 as signaling intermediate downstream of IRAK1.

### IL-1R/TLR SIGNALING CONTRIBUTES TO RT-INDUCED ANTITUMOR IMMUNITY AND DEFINES A TARGET FOR RT+IT-BASED RADIOSENSITIZATION STRATEGIES

While predominantly activated by microbes, IL-1R/TLR signaling is also notably engaged by stromal dendritic cells (DCs) and macrophages located in the vicinity of irradiated tumors (**Figures 2A,B**). Indeed, many of the molecules released by irradiated cancer cells (i.e., damaged/apoptotic/necrotic cancer cells) are bona fide ligands for IL-1R/TLR, including IL-1β itself and a number of DAMPs such as heat shock proteins, high mobility group protein 1 (HMGB1) and tumor DNA/RNA fragments (25–34). In response to IL-1R/TLR ligation, DCs

FIGURE 2 | "One-two punch" vs. "double-edged sword" scenarios for tumor radiosensitization strategies exploiting IRAK1 inhibitors. (A) Simplified view of RT-induced antitumor immunity. DAMPs and cytokines (i.e., IL-1β) released by irradiated tumor cells are recognized by cell surface IL-1R/TLRs on surrounding stromal DCs and macrophages, stimulating their activation, maturation, and antigen presentation activity toward T-cells in lymph nodes, and ultimately unleashing tumor-specific T-cells against the irradiated tumor (as well as distant tumors not pictured here). TAA, tumor-associated antigen; DAMPs, damage-associated molecular patterns; IT, immunotherapy; TLRa, toll-like receptor agonist; DC, dendritic cell; ag pres., antigen presentation. \*IT (with TLRa) is optional and acts as a boost for the immune events otherwise described in the figure. (B) Simplified views of the IRAK1-mediated response to RT (left; tumor cell-intrinsic antiapoptotic response) and DAMP-bound TLRs (right, innate immune response). Note that while IRAK1 catalytic activity is required for the tumor response to RT (illustrated by green glare), it is largely dispensable for immune IRAK1 signaling. (C) "One-two punch" scenario, as afforded by a highly specific IRAK1 inhibitor with no activity against IRAK4. Such drugs would be expected to both blunt intrinsic tumor radioresistance (which depends on IRAK1 kinase activity) and spare IRAK1 mediated-antitumor immunity (which is less reliant on IRAK1 catalytic activity), resulting in a "one-two punch" on the tumor. The double-punch is illustrated by two red dart target symbols on the tumor. (D) "Double-edged sword" scenario, as afforded by a less specific IRAK1i with similar activity against IRAK4. Such IRAK1/4i would be expected to block both the tumor and immune responses to RT (each of which depends on IRAK4 catalytic activity; see text). Thus, in this scenario, intrinsic tumor radiosensitization activity would be retained but at the expense of blunting the immune component. A small, residual "punch" from the immune system on the tumor is indicated to further emphasize the detrimental effects of IRAK1/4i relative to the "one-two punch" effects of specific IRAK1i [compare with (C)]. Figure design by Ni-Ka Ford, printed with permission from with permission from ©Mount Sinai Health System.

engage in increased proliferation, maturation, and antigen presentation activity, ultimately triggering T-cell-mediated attacks of tumors located within and outside the radiation field (immune attacks of distant tumors are responsible for the "abscopal" effect of RT long observed in a small subset of patients). The molecules, immune cell types and mechanisms believed to underlie RT-induced, IL-1R/TLR-mediated antitumor immunity are briefly summarized in **Figures 2A,B** but have been extensively investigated and reviewed by expert colleagues (27–29, 31, 34–41).

The notion that RT acts as a trigger for IL-1R/TLR signaling is at the root of emerging RT+IT combination strategies making use of TLR agonists (TLRa) as adjuvant or neoadjuvant therapies (**Figure 2A**). TLRa such as CpG oligodeoxynucleotides (CpG-ODN, TLR9a) and various imidazoquinolines and nucleoside analogs (TLR7a; e.g., imiquimod/Aldara/R-837, resiquimod/R-848, DSR-6434, DSR-29133, 3M-011/854A) have demonstrated substantial efficacy, if not outright synergy, when combined with RT in mouse spontaneous or xenograft models of fibrosarcoma (38, 39), lymphoma (37), colorectal cancer (35, 36, 40), sarcoma (35), breast cancer (42), renal cell carcinoma (36), lung adenocarcinoma (43), pancreatic cancer (40), and metastatic osteosarcoma (36). Success with these preclinical studies has spurred a number of clinical trials of CpG-ODNs in combination with diverse chemo-RT treatment regimens (34, 44–46). Such trials initiated between 2015 and 2018 include NCT03410901, NCT01745354, NCT02254772, and NCT02266147 for the treatment lymphoma; NCT02927964 for the treatment of follicular lymphoma; NCT03322384 for the treatment of advanced solid tumors and lymphoma; and NCT03007732 for the treatment of prostate carcinoma [reviewed in (44)]. Despite mixed results so far, favorable clinical responses observed in patient subsets warrant further testing (34, 44–46).

### IRAK1 ALSO ANCHORS AN ANTIAPOPTOTIC RESPONSE TO RT DISTINCT FROM IRAK1 IMMUNE SIGNALING

As discussed in Introduction, while mammalian IRAK1 is a genuine protein kinase and is a central transducer in IL-1R/TLR signaling, its catalytic activity appears largely dispensable for innate immunity. TLR/IL-1R-independent roles for IRAK1 might explain this paradox, yet until recently no such nonimmune IRAK1 function had been reported in vertebrates. In a screen for small molecules that restore RT-induced cell death in otherwise radioresistant p53 mutant zebrafish (47, 48), we identified the microtubule inhibitor, oxfendazole (47). Surprisingly, target discovery identified IRAK1, and not tubulin, as the key target whose inhibition by oxfendazole was responsible for cell death recovery in irradiated fish (47). The requirement of IRAK1 for cell survival after RT was conserved in multiple human cancer cell lines tested in vitro or as tumor xenografts in vivo, regardless of p53 genotype. Overexpression of IRAK1 was sufficient to force cell survival after RT in otherwise radiosensitive cells, in a manner that completely relied on its catalytic activity. Likewise, kinase-dead IRAK1 failed to complement IRAK1 deficiency in both human and fish models (47). Rather than promoting survival through NF-κB and other attendant pathways, we found that IRAK1 acts to deny RTinduced apoptosis mediated by the PIDDosome complex (PIDD-RAIDD-caspase-2) (47, 49, 50). These observations identified an essential role for IRAK1 outside of innate immunity as a gene required for the survival of irradiated vertebrate cells. IRAK1's reliance on its catalytic activity and engagement of a distinct antiapoptotic cascade were first clues that it might function in a pathway distinct from the canonical IL-1R/TLR axis (**Figure 1B**) (47).

Further evidence for IRAK1 functioning in a novel pathway came when we asked whether its known upstream proximal regulators, MyD88, IRAK4, and PIN1, were also required for the survival of irradiated cells. While IRAK4 and PIN1 clearly were, MyD88 clearly was not, whether in human cells or zebrafish embryos (47). Likewise, while IRAK4 and PIN1 were required for IRAK1 activation after RT, as assessed by T209 phosphorylation, MyD88 was not (47). In summary, RT-induced IRAK1 signaling differs from its canonical counterpart in three fundamental ways: (1) It fully relies on its kinase activity; (2) it acts through distinct downstream antiapoptotic mechanisms; and (3) it does not require MyD88 for activation by IRAK4 and PIN1 (**Figure 1B** vs. **Figure 1A**).

### IR-INDUCED IRAK1 SIGNALING AS A DRIVER OF INTRINSIC TUMOR R-RT

Thus far, the case for IR-induced IRAK1 signaling acting as a driver of intrinsic tumor R-RT is four-fold. (i) IRAK1 and PIN1 are both sufficient to force R-RT in otherwise radiosensitive tumor cells (47). (ii) IRAK1 and PIN1 enzymatic activities are required for R-RT in cancer cell lines derived from multiple tumor types including HNSCC, breast cancer, colorectal cancer, and glioblastoma. These requirements for R-RT were verified in vivo in a mouse xenograft model of radioresistant HNSCC (47). (iii) IR-induced activation of IRAK1, as assessed by T209 phosphorylation, systematically correlated with tumor cell line sensitivity to RT+IRAK1i (47). (iv) Patients with highrisk HNSCC (HPVneg, mutant TP53) whose tumors resisted post-operative RT (51) show evidence of pathway activation, whereby elevated PIN1 expression levels strongly associate with locoregional recurrence (LRR; P = 0.006) and reduced overall survival (OS; P = 0.007) (47). Notably, PIN1 overexpression did not otherwise correlate with metastatic potential, arguing against the notion that PIN1 levels merely reflected an aggressive tumor subtype. While upregulation of IRAK1 itself failed to correlate with R-RT in this cohort, this is not unexpected given the upstream role played by PIN1 in the pathway (see above; **Figure 1B**). Upregulation of PIN1 would in fact be expected to alleviate selective pressure to overexpress IRAK1 in this context. Deregulation/amplification at the IRAK1 locus might also not be a mechanism of choice via which tumors upregulate IRAK1 activity, though IRAK1 overexpression has been detected in several tumor types (4, 19), with particularly convincing evidence for causality in triple-negative breast cancer (52). Alternative routes to IRAK1 activation include upregulation of upstream positive regulators, such as seen with PIN1 (see above) as well as S100A-7/9 proteins in breast cancers with 1q21.3 amplification (53); mutational inactivation or downregulation of negative regulators such as miR-146a, as seen in del(5q) acute myeloid leukemia (54); and likely additional mechanisms [reviewed in (4, 19)]. Complementing our microarray analyses with that of exome sequence datasets from radioresistant tumors across tumor spectra will further clarify the extent to which IR-induced IRAK1 signaling drives R-RT in human cancer.

### TARGETING IRAK1 IN RADIORESISTANT CANCER

As discussed earlier, IRAK1 inhibitors (IRAK1i) were highly effective at suppressing R-RT in live p53 mutant zebrafish and human cancer cell lines assayed in vitro or as mouse xenografts in vivo (47). Remarkably, effective doses of IRAK1i in these models caused little to no cell death in non-irradiated controls. This was in stark contrast with the traditional radiosensitizer cisplatin, which failed to overcome R-RT at maximal tolerable doses (47). This data, combined with the previously established viability of Irak1−/<sup>−</sup> mice (55), suggests that systemic IRAK1i could restore RT sensitivity in patients without affecting healthy tissues outside of the radiation field.

While our work thus outlines a strong rationale for targeting IRAK1 in radioresistant tumors, as based on the projected efficacy and safety of such treatments, the strategy poses an immediate conundrum. Wouldn't systemic inhibition of the kinase simultaneously thwart the patient's immune attack on the irradiated tumor or the enhancement thereof by means of TLRa-based IT? Our tumor xenograft experiments, which were performed in immunodeficient mice, left this key question unanswered. Neoadjuvant administration of the TLRa (i.e., prior to RT+IRAK1i) or post-treatment delivery thereof might help circumvent the issue. However, our studies indicate that the window for IRAK1i radiosensitizing efficacy is limited to within a few hours of RT (47), and such treatments would thus be expected to come at the cost of blunting any acute immune contribution to the overall tumor response to RT.

However, such a "double-edged sword"-like tradeoff in efficacy is likely to be avoided by virtue of a critical, differential reliance of IRAK1 on catalytic activity when operating in response to IL-1R/TLR vs. when operating in response to RT (**Figures 1A,B**). As outlined earlier, kinase activity is essential for IRAK1 signaling in response to RT in all settings tested, both in zebrafish embryos and human cancer cells (47). In contrast, similar experiments making use of kinase dead IRAK1 variants in human cells (D340N, K239A) or knock-in mice (D359A) have indicated that catalytic activity is largely dispensable for IRAK1 function in IL-1R/TLR signaling (21–23, 56). Specifically, kinase dead IRAK1 retained full NF-κB inducing activity in all tested settings, presumably reflecting the protein's strict structural role when engaging TRAF6 (4, 24, 57). IL-1R/TLR-induced secretion of IL-6, TNFα, and IL-10 were likewise unaffected in bone marrow-derived macrophages from Irak1D359<sup>A</sup> knockin mice (22). Thus, RT+IRAK1i-based radiosensitization strategies, whether alone or in combination with TLRa-based IT, would be expected to largely spare IL-1R/TLR-initiated immune attacks on the tumor, leading to an effective "one-two punch" both from within and outside the irradiated tumor (**Figure 2C**). It should be noted, however, that IRAK1 catalytic activity might not be entirely dispensable for all forms of IL-1R/TLR signaling. In the TLR7/9-IRF7 signaling branch, for instance, an intact IRAK1 kinase domain appears required for the transcriptional activation of IRF7 as well as for the timely induction of interferons by TLR7/9 (56), as further evidenced by a significant delay in IFNβ production by plasmacytoid DCs derived from Irak1D359<sup>A</sup> mice (22). The relative contributions of the IL-1R/TLR-NF-κB (kinaseindependent branch) vs. IL-1R/TLR-NF-α/β (partially kinasedependent branch) to RT-induced antitumor immunity have not been rigorously explored to date and is an important topic for future studies.

The "one-two punch" hypothesis that IRAK1i will both intrinsically sensitize tumor cells to RT while also allowing for RT-induced antitumor immunity to proceed (**Figure 2C**) is further contingent on the use of IRAK1i that are highly specific to IRAK1. Indeed, unlike IRAK1, the catalytic activity of the sister kinase IRAK4 is essential for IRAK1 signaling in both the RT and

#### REFERENCES

1. Jain A, Kaczanowska S, Davila E. IL-1 receptor-associated kinase signaling and its role in inflammation, cancer progression, and therapy resistance. Front Immunol. (2014) 5:553. doi: 10.3389/fimmu.2014. 00553

IL-1R/TLR response pathways, in which IRAK4 acts to activate IRAK1 via direct phosphorylation on T209 (4, 13, 47). Thus, any IRAK1i with significant off-target activity against IRAK4 would be expected to radiosensitize the tumor proper but at the expense of affecting its immunogenic attack (**Figure 2D**). We recently confirmed the essential role of IRAK4 in RTinduced IRAK1 signaling in vivo, whereby (i) irak4-depleted p53MK/MK zebrafish embryos recover RT-induced cell death as efficiently as irak1-depleted embryos (Liu and Sidi, unpublished observations); and (ii) irak1-depleted embryos reconstituted with T209A human IRAK1 mRNA fail to resist RT-induced cell death, as opposed to embryos complemented with WT IRAK1 mRNA (Li and Sidi, unpublished observations). Thus, IRAK1i used for radiosentization purposes should, at the very least, demonstrate marked selectivity for IRAK1 over IRAK4 (**Figures 2C,D**).

Of the many IRAK1i developed so far [reviewed in (19)], only one, pacritinib (58), combines clinical efficacy, acceptable safety, and selectivity for IRAK1 over IRAK4. This selectivity is only moderate, however, with IC50s of 6 and 177 nM vs. IRAK1 and IRAK4, respectively (19). In spite of IRAK1 and IRAK4 kinase domains sharing >90% amino-acid sequence identity within the ATP binding pocket as well as identical gatekeeper tyrosine residues, the selectivity—albeit moderate—of pacritinib for IRAK1 indicates that developing a highly specific IRAK1i is feasible in principle. The crystal structure of the human IRAK1 kinase domain bound to a small molecule was recently reported (10), which together with the known structure of the IRAK4 kinase domain (15) should help develop such selective IRAK1i. A very first example of such a compound was recently reported by Buhrlage, Treon, Gray and colleagues (59). The drug, Jh-X-119-01, labels IRAK1 at C302 and shows irreversible inhibition with an IC50 of 9.3 nM against IRAK1 vs. >10µM vs. IRAK4. Disclosure of the structure of Jh-X-199- 01 should spur future efforts to develop IRAK1i suited for use as radiosensitizers.

#### AUTHOR CONTRIBUTIONS

SS conceived the review and figures. SS and PL wrote the paper.

#### ACKNOWLEDGMENTS

We thank Julio Aguirre-Ghiso and Jian Jin for helpful discussions and Ni-Ka Ford for the design of **Figure 2**. Our study (47) discussed in this perspective was supported by the National Institutes of Health (RO1CA178162 to SS; F30CA186448 to PL), and awards from the JJR Foundation, Pershing Square Sohn Cancer Research Alliance, New York Community Trust, and Searle Scholars Program to SS.


entities. Oncoimmunology. (2015) 4:e1042201. doi: 10.1080/2162402X.2015.1 042201


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Liu and Sidi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Radiation-Induced Changes in Tumor Vessels and Microenvironment Contribute to Therapeutic Resistance in Glioblastoma

Yun-Soo Seo1,2†, In Ok Ko3†, Hyejin Park <sup>1</sup> , Ye Ji Jeong<sup>1</sup> , Ji-Ae Park <sup>3</sup> , Kwang Seok Kim<sup>1</sup> , Myung-Jin Park <sup>1</sup> \* and Hae-June Lee<sup>1</sup> \*

*<sup>1</sup> Division of Radiation Biomedical Research, Korea Institute of Radiological & Medical Sciences, Naju, South Korea, <sup>2</sup> Herbal Medicine Resources Research Center, Korea Institute of Oriental Medicine, Naju, South Korea, <sup>3</sup> Division of Applied RI, Korea Institute of Radiological & Medical Science, Seoul, South Korea*

#### Edited by:

*Ira Ida Skvortsova, Innsbruck Medical University, Austria*

#### Reviewed by:

*Yidong Yang, University of Science and Technology of China, China Sara Pedron, University of Illinois at Urbana-Champaign, United States*

#### \*Correspondence:

*Myung-Jin Park mjpark@kirams.re.kr Hae-June Lee hjlee@kirams.re.kr*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology*

Received: *18 July 2019* Accepted: *31 October 2019* Published: *15 November 2019*

#### Citation:

*Seo Y-S, Ko IO, Park H, Jeong YJ, Park J-A, Kim KS, Park M-J and Lee H-J (2019) Radiation-Induced Changes in Tumor Vessels and Microenvironment Contribute to Therapeutic Resistance in Glioblastoma. Front. Oncol. 9:1259. doi: 10.3389/fonc.2019.01259* Glioblastoma (GBM) is a largely fatal and highly angiogenic malignancy with a median patient survival of just over 1 year with radiotherapy (RT). The effects of RT on GBM remain unclear, although increasing evidence suggests that RT-induced alterations in the brain microenvironment affect the recurrence and aggressiveness of GBM. Glioma stem cells (GSCs) in GBM are resistant to conventional therapies, including RT. This study aimed to investigate the effect of radiation on tumor growth and the GSC microenvironment in a mouse model of glioma. To evaluate the growth-inhibitory effects of ionizing radiation on GSCs, tumor volume was measured via anatomical magnetic resonance imaging (MRI) after the intracranial injection of 1 × 10<sup>4</sup> human patient-derived GSCs (83NS cells), which exhibit marked radioresistance. When a tumor mass of ∼5 mm<sup>3</sup> was detected in each animal, 10 Gy of cranial irradiation was administered. Tumor progression was observed in the orthotopic xenografted GSC tumor (primary tumor) from a detectable tumor mass (5 mm<sup>3</sup> ) to a lethal tumor mass (78 mm<sup>3</sup> ) in ∼7 d in the non-irradiated group. In the RT group, tumor growth was halted for almost 2 weeks after administering 10 Gy cranial irradiation, with tumor growth resuming thereafter and eventually approaching a lethal mass (56 mm<sup>3</sup> ) 21 d after radiation. Radiation therapy yielded good therapeutic effects, with a 2-fold increase in GSC glioma survival; however, tumor relapse after RT resulted in higher mortality for the mice with a smaller tumor volume (*p* = 0.029) than the non-irradiated tumor-bearing mice. Moreover, tumor regrowth after IR resulted in different phenotypes associated with glioma aggressiveness compared with the non-irradiated mice; the apparent diffusion coefficient by diffusion MRI decreased significantly (*p* < 0.05, 0 Gy vs. 10 Gy) alongside decreased angiogenesis, abnormal vascular dilatation, and upregulated CD34, VWF, AQP1, and AQP4 expression in the tumor. These findings demonstrate that radiation affects GSCs in GBM, potentially resulting in therapeutic resistance by changing the tumor microenvironment. Thus, the results of this study suggest potential therapeutic targets for overcoming the resistance of GBMs to RT.

Keywords: glioblastoma, glioma stem cell, radiotherapy, tumor microenvironment, tumor vessel

#### INTRODUCTION

Glioblastoma (GBM) is the most malignant and highly angiogenic tumor in the central nervous system (CNS). Due to its aggressiveness, GBM has a very poor prognosis, with a median survival of 14.6 months from diagnosis (1–3). Radiotherapy (RT) is currently used to treat GBM alongside surgical resection and chemotherapy. Highly proliferative cells, including tumor cells, are sensitive to ionizing radiation-induced DNA damage via reactive oxygen species, which causes apoptosis and reduces cell proliferation. Although RT is a locoregional treatment for GBM, RT-induced alterations in the brain microenvironment have been shown to contribute to GBM recurrence and aggressiveness (4). Understanding of the origin of therapeutic resistance in GBM may therefore help to improve patient outcome and survival.

A specific subpopulation of glioma cells, known as glioma stem cells (GSCs), contribute to tumor recurrence during aggressive multimodal therapies. After RT, GBM frequently recurs as focal masses (5), indicating that GSCs are radioresistant and responsible for relapse (6, 7). GSCs can inherently resist conventional therapy due to their enhanced self-renewal and differentiation potential (8). Recent studies reporting that GSCs maintain GBM (9) have indicated that GSCs interact closely with the vascular niche and promote neovasculogenesis by releasing angiogenic factors (10); however, the underlying mechanisms remain unclear.

GBM patients have poor prognosis due to tumor cells that survive initial treatment and cause relapse or recurrence; thus, failure to inhibit tumor growth at the primary site is a major cause of mortality (11). GBM tumors originate in the brain and interact closely with their unique microenvironment (12). Furthermore, highly aggressive tumors rarely metastasize outside the brain (13), indicating a preference for the brain microenvironment. Cell-cell interactions, tissue dynamics, and cytokines and growth factors constitute a complex microenvironment that is altered in the presence of GBM tumors, promoting tumor invasion, and therapeutic resistance. Therefore, it is important to identify and characterize treatment-resistant tumor cells and whether they influence their microenvironment.

In this study, we investigated the extent of GSC glioma progression after brain irradiation for a specific duration via magnetic resonance imaging (MRI) with an orthotopic xenograft mouse model of GBM established using patient-derived GSCs. Furthermore, we assessed the morphological and molecular phenotypes that may be associated with radioresistance in postirradiation relapse.

#### MATERIALS AND METHODS

#### Cell Culture

The GBM patient-derived GSC line 83NS was maintained in DMEM/F-12 (Corning, 10-090-cvr, Corning, NY, USA) supplemented with B27 (Invitrogen, 17504044, Carlsbad, CA, USA), epidermal growth factor (10 ng/mL; Prospec, cyt-217, Brunswick, NJ, USA), and basic fibroblast growth factor (5 ng/mL; Prospec, cyt-218). 83NS cells were provided by Dr. Ichiro Nakano (University of Alabama at Birmingham, Birmingham, AL, USA) (14). Prior to transplantation, cells were washed with PBS and dissociated. After centrifugation at 1,500 rpm for 5 min, the cell pellet was resuspended in PBS at a density of 1 × 10<sup>4</sup> cells/3 µL.

#### Establishment of the Orthotopic Mouse Model of Glioma and Radiotherapy

All animal experiments were conducted in accordance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) of the Korea Institute of Radiological and Medical Sciences (KIRAMS2018-0079). Athymic BALB/c nu/nu mice were purchased from Orient Bio Inc. (Seoul, Korea), housed under specific pathogen-free conditions, and supplied with standard rodent feed and tap water ad libitum. The animals' heads were fixed in a stereotactic frame using non-perforating bars, a midline incision was made on the scalp, and a burr hole was drilled 0.1 mm posterior and 2 mm left of the bregma. An 83NS cell suspension (1 × 10<sup>4</sup> cells) was injected into the left frontal cortex at the coordinate bregma using a stereotactic frame and microinjector. The average tumor size was determined to be 5 mm<sup>3</sup> by MRI. Mice were randomly divided into two subgroups: a control group and an irradiated group. Radiation (a single dose of 10 Gy) was administered to the entire head under anesthesia (30 mg/kg Zoletil and 10 mg/kg Rompun) using an X-Rad320 (Precision X-Ray, East Haven, CT, USA; filter: 2 mm AI; distance: 42 cm; 260 kVp, 10 mA, 10Gy/5 min).

#### MRI

All MRI was performed using a 9.4-T animal MR system and a specific mouse brain coil (Agilent Technologies, Palo Alto, CA, USA). The incidence and size of the orthotopic glioma were determined radiologically. Cranial irradiation was carried out 12 d after stereotactic 83NS cell transplantation when the tumor volume approached 5 mm<sup>3</sup> . To confirm the therapeutic effects of tumor irradiation, MR images were obtained 0, 3, 7, 14, and 21 d after irradiation and tumor size was compared. Before MRI, animals were anesthetized with 2.5% isoflurane in oxygen. A fast spin-echo MR sequence for T2-weighted imaging (T2-WI) was used with the following parameters: repetition time (TR), 2500 ms; effective echo time (TEeff), 30 ms; echo train length (ETL), 4; average number, 4; slice thickness, 0.8 mm; slice number, 6; matrix size, 192 × 192; field-of-view, 25 × 25 mm<sup>2</sup> ; and total imaging time, 605 s. Tumor volume was measured by summing all voxels within the tumor boundary of the anatomical T2-W images using ImageJ software (NIH Bethesda, MD, USA).

A fast spin-echo MR sequence for diffusion-weighted imaging was used with the following parameters: repetition time (TR), 2000 ms; echo time (TE), 23.5 ms; echo number, 1; average number, 1; slice thickness, 0.8 mm; matrix size, 192 × 192; field-of-view, 25 × 25 mm<sup>2</sup> ; b-value, 0 and 800 s/mm<sup>2</sup> ; and total imaging time, 512 s. Apparent diffusion coefficient (ADC) maps and values were processed using the intrinsic VnmrJ 4.0 workstation (Agilent Technologies, Inc.).

#### Histopathological Analysis

All mice were euthanized by CO<sup>2</sup> inhalation and brain tissue specimens were harvested in accordance with IACUC guidelines. The harvested brain tissue specimens were fixed in 4% paraformaldehyde, embedded in paraffin, and cut into 5µm sections using a microtome (Leica, Nussloch, Germany).

Immunofluorescence staining was performed as described previously (15). Briefly, specimens were blocked with blocking buffer (PBS with 1.5% normal horse serum and 0.1% Triton X-100), incubated with anti-Ki-67 (Acris, DRM004, Herford, Germany), anti-CD31 (R&D Systems, AF3628, Minneapolis, MN, USA), anti-CD34 (SantaCruz, sc-74499, Dallas, TX, USA), anti-Von Willebrand factor (VWF; Abcam, ab6994, Cambridge, MA, USA), anti-aquanporin 1 (AQP1; SantaCruz, sc-32737), and anti-aquaporin 4 (AQP4; Novus Biologicals, NBP1-87679, Littleton, CO, USA) antibodies overnight at 4 ◦C, and then incubated with Alexa Fluor 488-, Alexa Fluor 546-, and Alexa Fluor 647-conjugated secondary antibodies (Invitrogen). After washing the sections with PBST, nuclei were counterstained with DAPI and fluorescence visualized was using confocal microscopy (Carl Ziess, Oberkochen, Germany). Fluorescence intensity was measured using ImageJ software (NIH). Briefly, fluorescence positive areas were assessed and the ratio was calculated (16). A TUNEL assay was carried out to evaluate apoptotic glioma cells using an Apoptosis kit (Promega, g3250, Madison, WI, USA) in accordance with the manufacturer's instructions.

#### Quantitative Reverse Transcription Polymerase Chain Reaction (qRT-PCR) Analysis

Total RNA was isolated from cultured GSCs treated with or without radiation (10 Gy) using an RNeasy Mini kit (Qiagen, Valencia, CA, USA) and 1 µg was reverse-transcribed into cDNA using amfiRivert cDNA Synthesis Platinum Master Mix (GenDEPOT, Barker, TX, USA) in accordance with the manufacturer's protocol. qPCR was carried out in triplicate, using qPCR SYBR green 2× mastermix kit (M Biotech, 18303, Seoul, Korea) with a CXF-96 detection system (Bio-Rad Laboratories, Hercules, CA, USA). Relative gene expression was normalized to that of Gapdh using the comparative C<sup>T</sup> method with Bio-Rad CFX manager v2.1 (Bio-Rad Laboratories). The following primers were used: aqp1 (sense: 5′ -CGTGACCTTGGTGGCTCA G-3′ ; anti-sense: 5′ -GGACCGAGCAGGGTTAATCC-3′ ), aqp4 (sense: 5′ -AACGGACTGATGTCACTGGC-3′ ; anti-sense: 5′ -AA AGGATCGGGCGGGATTC-3′ ), and Gapdh (sense: 5′ - CATC GCTCAGACACCATG 3′ ; anti-sense: 5′ -TGTAGTTGAGGTCA ATGAAGGG-3′ ).

### Statistical Analysis

Data are presented as the mean ± standard deviation (SD). Differences between tumor stages were assessed by one-way ANOVA with statistical significance set at p < 0.05. All statistical analyses were carried out using GraphPad Prism Software version 7.0 (GraphPad Software, La Jolla, CA, USA).

### RESULTS

#### Effect of IR on the Survival of Orthotopic GBM Xenograft Model Mice

We assessed the survival of non-IR control and IR mice administered with 83NS cells. GBM xenograft model mice were orthotopically injected with 83NS cells and subjected to wholebrain RT (single 10 Gy dose) 12 d post-injection. As shown in **Figure 1A**, the IR group displayed significantly higher survival than the control group (p = 0.0035, Log-rank test) and the median survival of the IR group (32 ± 2.24) was 12.4 d longer than that of the control group (19.6 ± 0.89). We also monitored the body weight of each mouse. In the control group, body weight was maintained for 12 d after injection and then rapidly decreased when the tumors became detectable by MRI (**Figure 1B**). Weight loss in the IR group was temporarily delayed after IR but resumed 2 weeks later, achieving similar levels to the control group.

### IR Delayed Orthotopic GSC Glioma Progression

MRI has been widely used to characterize brain tumor growth, progression, and response to various treatments in clinical and

graph of mice implanted with 83NS cells and treated with or without radiation (*n* = 5, 1 × 10<sup>4</sup> cells injected per mouse). IR treatment was administered when tumors grew to their detectable size (12 d, 10 Gy). (B) Changes in body weight were monitored.

groups were determined using unpaired *t*-tests (*n* = 4–6). PT; primary tumors detectable by MRI; GT; growing tumors that progressed without IR; DT; delayed tumors

preclinical studies (17, 18). To evaluate the effect of IR on GSC glioma, we recorded whole tumor volume using anatomical MRI. As shown in **Figure 2**, the tumors in the control group progressed in 7 d from a detectable mass of ∼5 mm<sup>3</sup> (5.50 ± 0.94) to a lethal mass of 78 mm<sup>3</sup> (77.95 ± 4.47). In the IR group, tumor growth was temporarily attenuated for 14 d after RT; however, it eventually reverted to a lethal mass of ∼56 mm<sup>3</sup> (56.35 ± 5.12) after a further 7 d. Thereafter, we performed diffusion-weighted MRI (DW-MRI), which quantifies the movement of water within tumors by measuring the ADC, which is negatively correlated with tumor cell density. An increase in ADC values is a quantifiable indicator of antitumor efficacy and thus decreasing angiogenic activity and increasing apoptotic rate (17–19). **Figure 2D** shows that IR increased the ADC by up to 12% during the delay in GSC glioma growth (p < 0.05, one-way ANOVA, PT vs. DT), whereas the ADC remained largely unchanged during tumor progression in the non-IR group (+1% at 3 d and −3% at 7 d vs. 0 d). Based on GSC glioma relapse, the ADC was reduced by up to 13% 21 d after IR and was significantly lower than that of the non-irradiated group (p = 0.044, unpaired t-test, GT vs. RT), although the tumor volume of the relapsed tumor was significantly lower than that of the non-IR group (p = 0.029, unpaired t-test). Based on these results, increased ADC, which was induced by a decline in cell density and enlarged extracellular space after radiation, was decreased according to tumor relapse to an even greater extent than that in growing tumors (GTs), which suggests that RTs had a higher cell density than GTs. To evaluate the changes induced by IR and to allow additional analysis, the tumors were divided into four groups: primary tumors (PTs) that were detectable by MRI, growing tumors (GTs) that progressed without IR, delayed tumors after IR (DTs), and re-grown tumors after IR (RTs).

after IR; RT; re-grown tumors after IR.

xenografted mice were stained with hematoxylin and eosin to identify tumor regions. (B) Proliferative and apoptotic cells were visualized by immunofluorescence staining and TUNEL assays, respectively. Scale bar, 100µm. (C) The percentage of Ki-67+ and TUNEL+ cells of total tumor cells per unit area was determined. \**p* < 0.05 and \*\**p* < 0.01. Significant differences among the treated groups were determined by analysis of variance followed by Tukey's test for multiple comparisons (*n* = 4–6). PT; primary tumors detectable by MRI, GT; growing tumors that progressed without IR, DT; delayed tumors after IR, RT; re-grown tumors after IR.

### IR Temporally Inhibited GSC Proliferation and Attenuated GSC Proliferation During Tumor Progression

To investigate the effect of IR on GSC glioma growth, we enumerated the proliferative and apoptotic cells in the brain tumors. As shown in **Figure 3**, the number of Ki-67 positive proliferative cells decreased gradually in accordance with tumor growth in the GTs compared to the PTs, whilst the number of TUNEL-positive apoptotic cells increased. These results may have occurred due to spatial constraints during the progression of solid tumors which induce cell death and affect the delivery of oxygen and nutrients to the tumor (20). The number of Ki-67-positive cells decreased significantly 7 d after IR in the DTs compared to that in the PTs (p < 0.05), corresponding to the delay in tumor growth. These inhibitory effects on tumor growth are consistent with the low tumor cell density representing increased ADC in the DTs (**Figures 2B,D**). The number of Ki-67-positive cells was also significantly increased in RTs compared to that in DTs (p < 0.05) during tumor regrowth. Furthermore, no significant differences were observed in the apoptotic index during regrowth from DTs to RTs, unlike the tumor growth from PTs to GTs. Delayed tumor regrowth was accompanied by GSC re-proliferation and a significant decrease in ADC levels compared with GTs, indicating that 10 Gy of IR transiently inhibited GSC growth; however, GSCs regrew with increased aggressiveness.

### IR Suppressed Angiogenesis and Structurally Altered Microvasculature in the Late Phase of Tumor Progression

We investigated the effect of RT on GSC glioma angiogenesis by measuring the microvessel density (MVD) of tumors using CD31 as a pan-endothelial cell marker. As shown in **Figure 4**, MVD decreased significantly with tumor progression (p < 0.01, PT vs. GT) or re-growth (p < 0.05, DT vs. RT). During the early phase of tumor progression, IR slightly but non-significantly inhibited angiogenesis (p = 0.074, PT vs. DT), whereas in the late phase of tumor progression IR drastically inhibited MVD in RTs compared with PTs (p < 0.001) or DTs (p < 0.05). Vessel diameter increased sharply during delayed tumor regrowth compared to ordinary

tumor progression from PTs to GTs, consistent with the decrease in ADC (**Figure 2D**). Interestingly, RTs showed a significant increase in vessel diameter compared to that in GTs, which had similar tumor sizes. These results indicate that radiation inhibits angiogenesis and alters the vasculature of GSC gliomas.

### IR Altered Vascular Phenotypes in GSC Glioma

To verify the IR-induced morphological microvascular changes, we examined the molecular phenotypes of the tumor vessels by immunofluorescence staining with the additional endothelial cell markers CD34 and VWF during each stage of tumor progression. CD34 is a well-known marker of angiogenesis and endothelial progenitor cells (21, 22), whilst VWF is a multimeric plasma glycoprotein that mediates platelet adhesion to both the subendothelial matrix and endothelial surfaces (23) and is associated with tumor survival and angiogenesis (24, 25). As shown in **Figure 5**, in the non-irradiated group, CD34 and VWF expression were not markedly different in GTs compared to that in PTs. DTs showed only a mild increase in the number of CD34-positive cells and no changes in VWF expression in comparison with PTs, which displayed a similar tumor size. However, IR significantly increased the number of VWF- and CD34-positive cells in GTs compared to RTs. Due to regrowth, the tumors displayed decreased angiogenesis and enlarged vasculature; hence, increased CD34 and VWF may be involved in vascular abnormality rather than angiogenesis.

Since DW-MRI showed significant changes depending on the state of the tumor after IR, we also analyzed two major aquaporins (AQPs), AQP1 and AQP4, which are transmembrane water transporters that are primarily expressed in the brain tissue. In the non-irradiated group, AQP1 and AQP4 were expressed in GSCs at basal levels and no significant changes were observed in their expression during tumor progression. However, both AQPs were significantly upregulated following IR (p < 0.05; **Figure 6**). AQP1 is reportedly expressed in normal brain endothelial cells; however, AQP1 displayed a different expression pattern to CD31 (**Figure 6A**). Moreover, AQP4 is generally expressed in astrocytes; however, AQP4 displayed no colocalization with the astrocyte marker GFAP in the tumors (data not shown). We then examined aqp1 and aqp4 mRNA expression by qRT-PCR to determine whether IR upregulates AQPs in 83NS GSCs. As shown in **Figure 6C**, both AQPs were directly and significantly upregulated (p < 0.01) in 83NS 48 h after IR (10 Gy). These results show that IR directly alters GSCs and influences the microenvironment associated with tumor regrowth and aggressiveness.

by immunofluorescence staining using anti-CD34 and anti-VWF antibodies. Scale bar (white), 50µm. Scale bar (red), 10µm. (B) Proportions of CD34+ and VWF+ cells in tumor tissues were determined using ImageJ software. \**p* < 0.05, \*\**p* < 0.01, and \*\*\**p* < 0.005. Significant differences among the treated groups were determined by analysis of variance followed by Tukey's test for multiple comparisons (*n* = 6). PT; primary tumors detectable by MRI, GT; growing tumors that progressed without IR, DT; delayed tumors after IR, RT; re-grown tumors after IR.

(*n* = 3).

## DISCUSSION

GBM is one of the most intractable and angiogenic malignant tumors of the CNS. Despite recent advancements in the treatment of solid tumors, the treatment of these malignant gliomas remains essentially palliative since GBMs are extremely resistant to conventional radiation and chemotherapy. Moreover, despite significant technological improvements, radiotherapeutic effects are generally limited due to marked radioresistance in gliomas, particularly GSCs. Specific subpopulations of GSCs underlie this recurrence even when treated with aggressive multimodal therapies (26). In this study, we investigated the effects of therapeutic IR on GSCs in GBM by assessing histological and molecular alterations induced by IR in an orthotopic xenograft mouse model of GBM established using patient-derived GSCs. After IR, tumor progression was temporarily inhibited and the median survival increased for 12.4 days; however, tumors displayed rapid, and aggressive regrowth. Regrown GSC glioma, which displayed a smaller tumor volume than the non-irradiated group, was lethal to mice. Based on comparative histopathological analysis of different stages of tumor growth with or without IR, tumor regrowth after IR occurred alongside significant alterations in the vascular microenvironment. As shown in **Figure 5**, CD34 and VWF were significantly upregulated in regrown tumors with enlarged vessels. Unlike numerous studies reporting that CD34 and VWF are involved in angiogenesis, our study shows that CD34 and VWF expression coincide with reduced tumor vasculature. Consistent with our results, some studies have reported that primary GBM is characterized by increased angiogenesis, while recurrent GBM displays increased vasculogenesis and decreased angiogenic activity after RT (27). Hence, CD34 and VWF upregulation could be involved in abnormal vasculogenesis during tumor regrowth. Unlike ordinary vessel progression from PT to a GT which results from maturation and stabilization, vessel development from a DT to a RT after radiation may occur by vasculogenesis. Loss of angiogenic activity caused by radiation led to an influx of circulating cells which boost vasculogenesis, including endothelial progenitor cells and myeloid cells (3, 28). The increase in CD34+ and VWF+ cells in DTs and RTs caused by the infiltration of circulating cells might increase vessel diameter, which is crucial for tumor recurrence after radiation. Although CD34 is a marker of vascular endothelial progenitor cells (29, 30) and an optimal marker of microvascular density, a recent study showed that CD34 overexpression was associated with higher WHO glioma grades (III + IV) in 684 patients, suggesting that CD34 is a potential diagnostic and prognostic marker and therapeutic target for gliomas (31). In addition, VWF was only significantly upregulated in the late phase of GSC glioma regrowth; endothelial cell activation and the release of the procoagulatory protein VWF induce platelet aggregation, thus protecting cancer cells from immune cells, including NK cells, which is essential for malignancy (32). Therefore, these IR-induced alterations in the tumor microenvironment could contribute to resistance against further treatments, including RT.

This study demonstrated that IR upregulates AQP1 and AQP4 in GSC glioma tissue and 83NS cells in vitro. It has been reported that APQ expression is involved in water diffusion (33). To our knowledge, no studies have yet investigated IR-induced alterations in AQPs in GSCs or GBM. Since AQPs are involved in water transportation and edema, their upregulation in histopathological analysis was inconsistent with the reduced ADC in DW-MRI. In brain cancer, AQP1 expression is associated with brain capillary endothelial cells, which do not express AQP1 in normal brain tissue. The signals that induce AQP1 expression in the endothelium of brain tumors remain unclear but might include signals regulating the production and release of VEGF from cancer cells. As shown in **Figure 6A**, AQP1 was not expressed in tumor endothelial cells in GSC glioma. Hayashi et al. reported that AQP1 induction correlated with tumor cell metabolism and increased glycolysis and lactate dehydrogenase (LDH) activity, with patient GBM tissue exhibiting increased coincident AQP1, LDH, and cathepsin B expression levels which contributed to acidification of the extracellular milieu and glioma cell invasiveness (34). AQP4 is primarily an astroglial membrane protein localized in astrocytic endfeet that serves as a key functional component of the blood-brain barrier and is thought to be involved in brain edema pathogenesis (35). However, immunofluorescence staining and the ADC map of late phase regrown GSC glioma revealed differences compared to previous findings. In GSC glioma tissue, AQP4 was expressed on tumor cells, not tumor endothelial cells, and its expression levels decreased along with ADC levels, thus may not have contributed to edema. Interestingly, AQP4 expression is correlated with the incidence of epileptic seizures in GBM since patients with seizures have higher cell membrane AQP4 levels, suggesting that AQP4 expression is regulated post-transcriptionally (36). Furthermore, Lan et al. reported that AQP4 dissociates from

### REFERENCES


orthogonal particle arrays and is redistributed across the entire surface of glioma cells under tumor conditions; thus, AQP4 expression levels may correlate with tumor grade as AQP4 expression increases in higher glioma grades (37). These previous studies support our finding that increased AQP expression after IR may be associated with GSC glioma malignancy or aggressiveness; however, further studies are required to elucidate the role and underlying mechanism of AQPs involvement in GBM radioresistance.

There are numerous obstacles to improving the therapeutic efficacy of RT and further studies are required to investigate the basic molecular events in GBM. In particular, GSCs are responsible for post-treatment GBM relapse. To understand the molecular events that occur in radioresistant tumors, we used an orthotopic mouse model of glioma implanted with a patient-derived GSC xenograft to monitor tumor progression after RT and alterations in the tumor microenvironment over time. This study shows that CD34, VWF, and AQPs are associated with post-IR GSCs glioma relapse and provides essential insights for the development of treatment regimens for radioresistant tumors.

### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article.

### ETHICS STATEMENT

The animal study was reviewed and approved by the Institutional Animal Care and Use Committee at the Korea Institute of Radiological & Medical Sciences.

### AUTHOR CONTRIBUTIONS

Y-SS: data collection, data analysis, and manuscript writing. IK, HP, and YJ: data collection and data analysis. J-AP: data analysis. KK and M-JP: developed the study concept project development. H-JL: project development, developed the study concept, edited, manuscript writing, and revised the manuscript.

### FUNDING

This study was supported by a grants of the National Research Foundation of Korea (2015M2B2B1068627) and a grant from the Korea Institute of Radiological and Medical Sciences funded by the Ministry of Science and ICT (MSIT), Republic of Korea (50531-2019 and 50436-2019).


for malignant astrocytomas. Radiother Oncol. (1991) 20:99–110. doi: 10.1016/0167-8140(91)90143-5


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Seo, Ko, Park, Jeong, Park, Kim, Park and Lee. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Dual Cell Cycle Kinase Inhibitor JNJ-7706621 Reverses Resistance to CD37-Targeted Radioimmunotherapy in Activated B Cell Like Diffuse Large B Cell Lymphoma Cell Lines

Gro Elise Rødland<sup>1</sup> , Katrine Melhus <sup>2</sup> , Roman Generalov <sup>2</sup> , Sania Gilani <sup>1</sup> , Francesco Bertoni <sup>3</sup> , Jostein Dahle<sup>2</sup> , Randi G. Syljuåsen<sup>1</sup> and Sebastian Patzke1,2 \*

<sup>1</sup> Department of Radiation Biology, Institute for Cancer Research, Norwegian Radium Hospital, Oslo University Hospital, Oslo, Norway, <sup>2</sup> Research and Development, Nordic Nanovector ASA, Oslo, Norway, <sup>3</sup> Lymphoma and Genomics Research Program, Institute of Oncology Research, Università Della Svizzera Italiana, Lugano, Switzerland

#### Edited by:

Ira Ida Skvortsova, Innsbruck Medical University, Austria

#### Reviewed by:

Heng-Hong Li, Georgetown University, United States Richard Piekarz, National Institutes of Health (NIH), United States

\*Correspondence:

Sebastian Patzke spatzke@nordicnanovector.com; sebastip@rr-research.no

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 21 July 2019 Accepted: 11 November 2019 Published: 29 November 2019

#### Citation:

Rødland GE, Melhus K, Generalov R, Gilani S, Bertoni F, Dahle J, Syljuåsen RG and Patzke S (2019) The Dual Cell Cycle Kinase Inhibitor JNJ-7706621 Reverses Resistance to CD37-Targeted Radioimmunotherapy in Activated B Cell Like Diffuse Large B Cell Lymphoma Cell Lines. Front. Oncol. 9:1301. doi: 10.3389/fonc.2019.01301 The CD37 targeting radioimmunoconjugate <sup>177</sup>Lu-lilotomab satetraxetan (Betalutin) is currently being evaluated in a clinical phase 2b trial for patients with follicular lymphoma (FL) and in a phase 1 trial for patients with diffuse large B-cell lymphoma (DLBCL). Herein we have investigated the effect of <sup>177</sup>Lu-lilotomab satetraxetan in seven activated B-cell like (ABC) DLBCL cell lines. Although the radioimmunoconjugate showed anti-tumor activity, primary resistance was observed in a subset of cell lines. Thus, we set out to identify drugs able to overcome the resistance to <sup>177</sup>Lu-lilotomab satetraxetan in two resistant ABC-DLBCL cell lines. We performed a viability-based screen combining <sup>177</sup>Lu-lilotomab satetraxetan with the 384-compound Cambridge Cancer Compound Library. Drug combinations were scored using Bliss and Chou-Talalay algorithms. We identified and characterized the dual-specific CDK1/2 and AURA/B kinase inhibitor JNJ-7706621 as compound able to revert the resistance to RIT, alongside topoisomerase and histone deacetylases (HDAC) inhibitors.

Keywords: radioimmunotherapy (RIT), CD37, combination therapy, lymphoma, radiation resistance, aurora kinase, cyclin-dependent kinase, polo-like kinase

### INTRODUCTION

Diffuse large B-cell lymphoma (DLBCL) is the most common type of lymphoma, accounting for ∼35% of all newly diagnosed cases. Despite its phenotypical relatively homogenous appearance, DLBCL is a heterogeneous diseases (1). So far, the most commonly used sub-classification is based on the cell-of-origin (COO), identifying the germinal center B-cell (GCB)-type and the activated B-cell (ABC)-like DLBCL subtypes on the basis of gene or protein expression pattern reminiscent of normal germinal center or ABCs, respectively. ABC-DLBCL is associated with a worse outcome than GCB-DLBCL when patients are treated with the standard R-CHOP-like treatment (R-CHOP: a combination of the CD20-targeting antibody rituximab and the chemotherapeutics cyclophosphamide, doxorubicin, prednisolone, and vincristine).

Despite the major improvements in our understanding of the biology of DLBCL and the availability of a large number of novel compounds, no new regimen has shown superiority to R-CHOP (2–4). Recent studies have indeed uncovered a high degree of genetic heterogeneity even within the GCB and ABC-DLBCL subtypes (5, 6), thus indicating the need to target more specific subgroups of patients. Furthermore, over-expression of MYC and BCL2 proteins in the absence of chromosomal rearrangements identifies a subgroup of cases, the so-called double expressor lymphomas (DEL), with a particularly poor prognosis. DELs are more frequent in ABC than GCB-DLBCL (1, 7). In standard clinical practice firstline R-CHOP-like treatment of DEL includes etoposide in addition to prednisone, vincristine, cyclophosphamide and doxorubicin (R-ECHOP) [reviewed in (1, 8)]. Patients with refractory, relapsed or treatment resistant DLBCL are, if eligible, treated with intensive chemotherapy [in the case of chemoresistance: rituximab in combination with ifosfamide, carboplatin, and etoposide (R-ICE), or dexamethasone, high-dose cytarabine, and cisplatin (R-DHAP)] followed by autologous (or allogeneic) stem cell transplantation (ASCT). Still, the DEL patient group remain a particularly poor prognostic group with 5-year progression free survival (PFS) of <30% after relapse.

Radioimmunotherapy (RIT) is an alternative and targetspecific treatment for lymphomas. RIT is based on the conjugation of a short-ranged and short half-lived radioemitter to a lineage-specific monoclonal antibody (9). <sup>90</sup>Y-Ibriumomab and <sup>131</sup>I-tositumomab are examples of CD20 targeted FDAapproved, first-generation RITs for the treatment of relapsed or refractory follicular lymphoma (r/r-FL) and transformed FL [for a review (10)]. These showed promising results for treatment of FL and DLBCL, but, amongst other, logistic challenges resulted in underusage (11–14). <sup>131</sup>I-tositumomab was withdrawn from the market in 2014. CD37 is a transmembrane protein expressed almost exclusively on cells of the immune system, especially in mature B cells and B cell NHL (15), hence being an important alternative for CD20-targeting therapies, to which treatment-resistance can be developed (16). <sup>177</sup>Lu-lilotomab satetraxetan (Betalutin <sup>R</sup> ) is a CD37 specific murine monoclonal antibody (clone HH1) that is chelated via a p-SCN-benzyl-DOTA-linker (satetraxetan) to the β-emitting isotope <sup>177</sup>Lutetium (T1/<sup>2</sup> = 6.7 days) (17– 19). <sup>177</sup>Lu-lilotomab satetraxetan is currently being investigated as a single-injection mono-therapy for treatment of relapsed or refractory (r/r) FL (NCT01796171, Phase 2B) and r/r-DLBCL (NCT02658968, Phase 1), showing a promising overall response rate in r/r-FL of about 65–70% (20). <sup>177</sup>Lu-lilotomab satetraxetan may thus have potential in the treatment of highgrade DLBCL, though as in the case of r/r-FL treatment, resistance may occur. To explore and attempt to over-come the potential resistance we chose to investigate the sensitivity to <sup>177</sup>Lu-lilotomab in cell lines derived from ABC-DLBCL, the DLBCL subtype that has the lower sensitivity to standard regimens. Subsequently, we conducted a combinatorial drug screen for small molecular anti-cancer compounds preventing <sup>177</sup>Lu-lilotomab satetraxetan treatment resistance in the most resistant cell lines. Herein, we report the identification and pre-clinical characterization of a dual CDK1/2 and AURKA/B kinase inhibitor that was identified in the screen as a candidate compound to overcome <sup>177</sup>Lu-lilotomab satetraxetan therapy resistance.

#### RESULTS

#### Resistance of U-2932 and RIVA to CD37-Targeted <sup>177</sup>Lu-Radioimmunotherapy

We initially investigated the sensitivity of seven different ABC-DLBCL cell lines (HBL1, OciLy-3, Oci-Ly10, RIVA [RI-1], SU-DHL-2, TMD-8, U-2932) to treatment with <sup>177</sup>Lu-lilotomab satetraxetan. Cells were treated for 18 h with 11 different doses of <sup>177</sup>Lu-lilotomab satetraxetan ranging from concentrations of 0.01–20µg/ml (specific activity: 600 MBq/mg), washed and plated in 96-well plates. Mock treated cells were included as controls. The total DNA content in each well was assessed using the CyQuant reagent as a readout of cell proliferation. Comparative analysis of relative proliferation capacity compared to untreated control cells identified U-2932 and the RIVA cells as the most resistant cell lines, showing over 40% of signal intensity of untreated control cells even after treatment at 20µg/ml <sup>177</sup>Lu-lilotomab satetraxetan (**Figure 1A**). Conversely, Oci-Ly10 cells showed highest sensitivity to treatment with a 70% decreased proliferation capacity in response to treatment with 0.25µg/ml <sup>177</sup>Lu-lilotomab satetraxetan. Notably, CD37 mRNA and CD37 surface expression were not associated with the resistance to CD37-target RIT (**Table 1**). We confirmed the differential sensitivity of these three cell lines in a metabolic cell viability assay, utilizing MT RealTimeGlo, that allowed the monitoring of cell proliferation throughout a continuous period of 72 h (**Figures 1B,C**). Cells were treated as previously and the luminescent assay substrate added 72 h after plating into microwell titer plates. All cell lines and control treatment groups showed continuous proliferation throughout the observation period. Addition of cold, non-177Lu chelated lilotomab (HH1- DOTA) did not markedly inhibit proliferation in either cell line. Oci-Ly10 cells were sensitive to even the lowest tested dose of 0.05µg/ml <sup>177</sup>Lu-lilotomab satetraxetan and ceased proliferation at 0.25µg/ml. Confirming the observed resistance in the CyQuant assay, U-2932 and RIVA retained ∼60 and 40%, respectively, of the proliferation capacity of untreated cells at 5 days after treatment with 2µg/ml <sup>177</sup>Lu-lilotomab satetraxetan. Again, RIVA cells were more sensitive to <sup>177</sup>Lulilotomab satetraxetan than U-2932 and showed about 60% of the proliferation capacity of control cells at a dose of 0.5µg/ml, which is half of the dose required in U-2932 cells to reach a similar level of inhibition.

To conclude, U-2932 and RIVA have been shown to be <sup>177</sup>Lu-lilotomab satetraxetan treatment resistant ABC-DLBCL cell lines. Furthermore, the resistance of these cell lines to CD37 targeted RIT was not due to a reduced expression of and binding to CD37 on the cell surface.

### Combinatorial Drug Screen Identifies Cell Cycle Kinase Inhibitors as Candidate Drugs to Overcome Radioimmunotherapy Resistance

Only two out of seven cell lines of the ABC-DLBCL panel did not respond well to CD37-targeted RIT. These two cell lines, U-2932 and RIVA, may represent models of most challenging to


#### TABLE 1 | Characteristics of ABC-DLBCL cell lines.

WT, wild-type; T, translocated; OE, overexpressed, Amp, amplified; RPKM, Reads per kilo base per million (21–23).

treat DLBCL. To understand and overcome the lack of response to <sup>177</sup>Lu-lilotomab satetraxetan treatment, we set out to find combinatorial drug partners for <sup>177</sup>Lu-lilotomab satetraxetan capable of reversing CD37-targeted RIT resistance in both cell lines. U-2932 and RIVA cells were treated with 1 and 0.5µg/ml of <sup>177</sup>Lu-lilotomab satetraxetan for 18 h, respectively, washed and seeded onto micro-well plates pre-printed with a library of 384 anti-cancer compounds (Cambridge anti-Cancer Compound library; SelleckChem). Three days post-plating RealTimeGlo reagent was added and luminescence read on three consecutive days to assess the relative amount of metabolically active cells. The screen design is schematically presented in **Figure 2**.

Both cell lines continuously proliferated throughout the observation period, albeit U-2932 cells showed a higher growth rate and stronger resistance to <sup>177</sup>Lu-lilotomab satetraxetan monotreatment than RIVA cells in the primary screen (**Supplementary Figure 1**). Three different concentrations of drugs (U-2932: 10/1,000 nM; RIVA: 10/100 nM final concentration) were tested to account for compound-specific differences in potency. Inhibitory compounds were considered as a hit candidate if they: (1) in combination with <sup>177</sup>Lu-lilotomab satetraxetan inhibited cell proliferation over two consecutive days to a degree greater than the expected additive effect of the mono-treatments alone (Bliss theorem, see Materials and Methods for details), and (2) the inhibitory effect of the compound alone at the tested concentration was <90% relative to the untreated control. These criteria were met by 53 compounds in U-2932 (11 at 10 nM and 42 at 1µM) and 27 compounds in RIVA cell lines (8 at 10 nM and 19 at 100 nM) (**Tables 2**, **3**). **Figures 3A,B** summarize the screen results for each cell line and library concentration. Hit candidates are highlighted in dot-plots showing the relative proliferation of cells at day 5 treated with the compound alone vs. its combination with <sup>177</sup>Lu-lilotomab satetraxetan (datasets are included as excel spreadsheets in **Supplementary Material**). Topoisomerase inhibitors accounted for 13% of the hits in U-2932 and 23% in RIVA, and histone deacetylase (HDAC) inhibitors for 7% of the hits in U-2932 and 27% in RIVA cells (**Tables 2**, **3**). The enrichment defined topoisomerases and histone deacetylases (HDAC) as prime targets for co-inhibition in both cell lines. A third prominent group of hit candidates comprised inhibitors targeting mitotic cell cycle kinases, including AURKA, AURKB, CDK1, PLK1, and WEE1 (17% of total hits in U-2932; 12% in RIVA).

Since both topoisomerase inhibitors, such as doxorubicin or etoposide, and HDAC inhibitors are known to cause direct or indirect DNA damage, respectively, it is likely they might overcome resistance by potentiating the level of DNA damage in <sup>177</sup>Lu-RIT-targeted cells (24–26). Therefore, we focused on the third group of compounds, the mitotic kinase inhibitors, that affect kinases with critical functions for both mitotic entry and exit, and have a role in termination of the DNA damage-induced G2-checkpoint (27–29). In particular we further explored the combination of <sup>177</sup>Lu-lilotomab satetraxetan with the dual CDK1/2-Aurora A/B inhibitor JNJ-7706621, the Aurora A inhibitor alisertib (MLN8237), and the Plk1 inhibitor GSK461364, respectively. JNJ-7706621 and alisertib were hit candidates in both cell lines. GSK461364 did not score as a hit candidate in U-2932 cells only due to its high potency as a mono-therapy.

#### JNJ-7706621 Synergistically Reduces Viability of DLBCL When Combined With <sup>177</sup>Lu-Lilotomab Satetraxetan

To investigate the validity of the synergism observed following <sup>177</sup>Lu-Lilotomab satetraxetan with the effect of the cell cycle kinase inhibitors in resistant cell lines, U-2932 cells were treated or not with <sup>177</sup>Lu-lilotomab satetraxetan (0.5, 1, and 2µg/ml) for 18 h, washed and seeded onto 384-well plates preprinted with 11-step gradients of JNJ-7706621, alisertib, and GSK461364 ranging from 0 to 1,280 nM. Similar to the primary screen, viability was measured by RealTimeGlo. Dose-response profiles were recorded at day 5 and Combination Indexes (CI) calculated to test for a synergistic interaction of <sup>177</sup>Lu-Lilotomab satetraxetan-inhibitor combinations (Chou-Thalaly; CompuSyn software). CI values were calculated for combinations within the minimum and maximum effect range of mono-treatment of each inhibitor.

**Figure 4** shows dose-response curves for treatment with either drug alone or in combination with different doses of <sup>177</sup>Lu-lilotomab satetraxetan. The anti-proliferative effect of the dual CDK1/2-AURKA/B inhibitor JNJ-7706621 was moderate, even at high doses (**Figure 4A**). However, and confirming the

primary screen results, the combination with <sup>177</sup>Lu-lilotomab satetraxetan led to a greater reduction in the fraction of viable cells than each treatment alone. Synergism (CI < 1) was observed for all tested combinations (blue circles in Fa/CI plots). The AURKA inhibitor alisertib had a bi-phasic dose-response profile, with a decreasing anti-proliferative effect at doses above 160 nM (**Figure 4B**) and synergism was observed within a range of 10– 160 nM. U-2932 cells were highly sensitive to treatment with the PLK1 inhibitor GSK461364 with a near complete growth inhibition obtained at 20 nM (**Figure 4C**). At lower doses the sensitivity was greatly reduced, and when combined with <sup>177</sup>Lulilotomab satetraxetan growth inhibition was, however modestly, potentiated. Weak synergism (CI 0.75–0.95) was observed only at GSK461364 concentrations near the maximum effect (20 and 40 nM; Fa close to 1). Similar CI index profiles were obtained in RIVA cells (**Supplementary Figure 2**). The results of the validation screen identified the dual CDK1/2 and AURKA/B kinase inhibitor JNJ-7706621 as the best hit candidate. We hence controlled first for target-specificity/activity for inhibition of CDK1 and AURKB (**Supplementary Figure 3**). Synergism with <sup>177</sup>Lu-lilotomab satetraxetan was then confirmed in two additional combination experiments in U-2932 cells covering the range of 100–10,000 nM JNJ-7706621 (**Figure 4A**, yellow and red circles).

### The Dual CDK1/2 and AURKA/B Inhibitor JNJ-7706621 Synergizes With CD37-Targeted RIT by Potentiating Mitotic Slippage and Apoptotic Cell Death

The validation experiments confirmed a synergistic drug interaction of JNJ-7706621 and <sup>177</sup>Lu-lilotomab satetraxetan in the inhibition of proliferation of U-2932 and RIVA cells, as assessed by the decreased capacity in reducing the RealTimeGlo substrate. CD37-targeted RIT induced DNA damage and inhibition of CDK1/2 and AURKA/B kinases by JNJ-7706621 are independently of each other expected to result in cell cycle progression defects. For instance, in mitotic shake-off synchronized HeLa cells, inhibition of CDK1/2 and AURKA/B by JNJ-7706621 at a final concentration of 1–3µM was reported to delay exit from G1, to arrest cells at G2/M, and to induce endoreduplication (30). We thus investigated the effects of mono- and combination therapy on cell cycle progression in U-2932 and RIVA cells (**Figure 5**). Cells were treated with or without <sup>177</sup>Lu-lilotomab satetraxetan (18 h; U-2932: 1µg/ml, RIVA 0.5µg/ml), washed and treated with or without 500 nM JNJ-7706621, a concentration at which strong synergism was observed in both cell lines (**Figure 4** and **Supplementary Figure 2**).

Samples were taken immediately after treatment with <sup>177</sup>Lulilotomab satetraxetan and at 24, 72, and 144 h after inhibitor addition, and DNA content and cell death assessed by flow cytometry (**Figure 5A**). Medium was replenished (with or without inhibitor) at 72 h to allow for optimal growth conditions throughout the observation period. Treatment with JNJ-7706621 alone induced a prominent but transient accumulation of cells in G2/M-phase (4n DNA content) after 24 h of treatment in both cell lines (**Figure 5B**). Similarly, pre-treatment with <sup>177</sup>Lu lilotomab satetraxetan alone also induced a prominent but transient accumulation of cells in G2/M-phase in both cell lines. The addition of JNJ-7706621 to <sup>177</sup>Lu-lilotomab satetraxetan treated cells strongly reduced the fraction of cells in G<sup>1</sup> phase (2n) at the 24 h time time-point and at later times strongly increased appearance of single cells with <2n or >4n DNA content, indicative of cell death and endoreduplication/cytokinesis failure, respectively. We therefore quantitatively assessed the fraction

#### TABLE 2 | Hit candidate list from screen in U-2932 cells.

#### U-2932


(Continued)

#### TABLE 2 | Continued

#### U-2932


Color-shading indicates increasing Bliss score from yellow to dark green.

of endoreduplication (>4n DNA content), cell death (Pacific blue uptake), as well as cell size (median forward scatter) in both cell lines in two independent experiments (**Figure 6** and **Supplementary Figures 4, 5**). Combination treatment strongly induced the formation of endoreduplicating cells (20- and 12 fold increase in U-2932 and RIVA, respectively; Bliss CI U-2932 = 0.55, RIVA = 0.69; [CI= ((ERIT + EJNJ)–ERIT × EJNJ)/E[RIT+JNJ]]), which was accompanied by a 2-fold increase in relative cell size, compared to untreated and mono-treated cells. Visual inspection of cells by microscopy confirmed this result, revealing prominent occurrence of multinucleated cells and increased DNA content (**Supplementary Figure 6**). Most importantly, combination treatment led to a 5- and 13-fold increase in cell death compared to untreated cells at the 144 h time point in U-2932 (CI = 0.72) and RIVA (CI = 0.50), respectively. To further explore the mechanistic details behind the anti-tumor activity of JNJ-7706621 and <sup>177</sup>Lulitomab satetraxetan combination treatment, we used the same experimental setup as described in **Figure 5** but now monitored PARP cleavage, a late apoptotic event, and cell growth (**Figure 7**). In accordance with our RealTime-Glo data we found that cells exposed to the combination showed a significantly reduced growth rate between 24 and 72 h post-treatment as compared to control cells and cells receiving monotherapy (**Figure 7A**). The fraction of cells in late apoptosis (cleaved PARP positive cells), was at all tested time-points significantly increased in cells receiving combination therapy as compared to control cells, reaching about 50% after 144 h (**Figure 7B**). Induction of apoptosis was also significantly increased in cells receiving single




Color-shading indicates increasing Bliss score from yellow to dark green.

agent therapy at 72 and 144 h time points compared to control. At 144 h, 22% of JNJ-7706621 and 28% <sup>177</sup>Lu-litomab satetraxetan treated cells were determined as late apoptotic, indicating an additive effect of the combination treatment. Taken together, these results suggest that the synergistic anti-proliferative effect of the combination of JNJ-7706621 and <sup>177</sup>Lu-litomab satetraxetan is a consequence of JNJ-7706621 mediated mitotic infidelity of cells which have over-come the <sup>177</sup>Lu-litomab satetraxetan induced G2-arrest, leading to incompatibility with proliferation and to excessive cell death by apoptosis (**Figure 7C**).

#### DISCUSSION

Targeted radionuclide delivery for DNA damaging radiation by means of antibody-conjugates has shown promising efficacy in clinical studies in the treatment of hematological cancers. <sup>90</sup>Y-Ibriumomab and <sup>131</sup>I-tositumomab have demonstrated significant activity in indolent relapsed/refractory NHL. <sup>177</sup>Lulilotomab satetraxetan is emerging as a potential treatment option for patients with rituximab resistant relapsed/refractory FL as well as R-CHOP resistant (and ASCT in-eligible) DLBCL. Here, we identified two ABC-DLBCL cell lines, U-2932 and RIVA, with primary resistance to CD37-targeting <sup>177</sup>Lu-lilotomab satetraxetan treatment, derived from DE ABC-DLBCL with inactive TP53. Subsequently, we used these cell lines to screen for compounds able to prevent the resistance to RIT and we identified and characterized the dual-specific CDK1/2 and AURKA/B kinase inhibitor JNJ-7706621, alongside topoisomerase and HDAC inhibitors. Alike other RITs <sup>177</sup>Lulilotomab satetraxetan is likely to induce a DNA damage response mediated cell cycle G<sup>2</sup> arrest that resistant cells are required to overcome or adapt to. Our findings may thus be of particular importance as G<sup>1</sup> arrest abrogating subclonal TP53 mutations were recently found to be predictive of PFS in FL patients treated with CD20-targeting RIT-CHOP (131I Tositumomab), but not R-CHOP (31).

U-2932 and RIVA are notoriously treatment resistant cell line models of ABC-DLBCL, including radiation- and chemotherapy (32, 33). Importantly, loss of or decreased binding to CD37 was excludable as an underlying cause of resistance to CD37-targeting RIT. The unbiased screening for anti-cancer compounds, which in combination with <sup>177</sup>Lu-lilotomab satetraxetan synergistically impair proliferation of these cell lines identified three major classes of compounds that may be utilizable to overcome RIT-resistance: topoisomerase inhibitors, HDAC inhibitors, and inhibitors of mitotic cell cycle kinases. Since we tested only a limited set of concentrations in a sequential approach following exposure to RIT, our study might have missed additional compounds that might be beneficial in combination at different concentrations or when given before or concomitantly with RIT, although radiation is persistently delivered through binding of the radioisotope conjugated antibody to CD37.

Interestingly, all Topoisomerase and all HDAC inhibitors within the compound library scored in at least one cell line. Topoisomerase inhibitors, such as doxorubicin and etoposide, are essential constituents of standard and salvage chemotherapy regimens (R-CHOP and R-ICE/R-DHAP) for lymphoma treatment and may synergize with <sup>177</sup>Lu-lilotomab satetraxetan by exaggerating the cumulative DNA damage beyond the cell's capacity of repair. HDAC inhibitors have been shown to lead to excessive DNA damage in cancer cells (24), but other downstream aspects within the pleiotropic effects of epigenetic modification may contribute to or be essential for the observed synergistic interaction. They have shown anti-tumor activity especially in T-cell lymphomas and are currently under evaluation in different combination regimens for r/r-DLBCL (34). Hence, both classes of compounds might not only potentiate the amount of DNA damage in <sup>177</sup>Lu-RIT-targeted cancer cells but also confer off-targeted DNA damage in untransformed cells due to their systemic administration. We focused on the third enriched group composed of mitotic cell cycle kinases inhibitors, which represent a secondary main (independent) target different from induction of DNA damage. Their inhibition ultimately interferes with balanced segregation of chromosomes as well as

statistical criteria (See Figure 2). Enriched hits are specified with name of common target and are in separate colors [purple triangles: topoisomerase inhibitors, burgundy circles: aurora kinase inhibitors (including Alisertib) and green circles: histone deacetylase inhibitors]. In red is the pan-CDK1/2 and AuroraA/B inhibitor JNJ-7706621 and in brown the Plk1 inhibitor GSK461364. In light pink are all other hits.

separation of daughter cells, rather than the direct DNA damage induction of replicating cells (27–29). Mitotic kinase inhibitors, including the hit candidate MLN8237 (alisertib), are currently in clinical trials for B cell lymphomas, e.g., Friedberg et al. (35) and Kelly et al. (36) (NCT00807495; NCT00697346). The dualspecific inhibitor of CDK1/2 and AURA/B kinases, JNJ-7706621, outperformed the PLK1 inhibitor GSK461364 and the AURKA inhibitor alisertib in the validation screen on the basis of strong synergism within a broad concentration range, at which monotreatment with JNJ-7706621 showed little to no anti-proliferative efficacy. In contrast, synergism with PLK1 inhibition was only confirmed at concentrations near the maximum effect of the

FIGURE 4 | concentration gradient (0–1,280 nM). Viability was measured as in Figure 1B. In left panels the fraction of viable cells relative to untreated control at day 5 are plotted for cells exposed to single treatment or combinations for each drug dose (blue: drug alone, red: drug combined with 0.5µg/mL <sup>177</sup>Lu-lilotomab satetraxetan, green: drug combined with 1µg/mL <sup>177</sup>Lu-lilotomab satetraxetan and orange: drug combined with 2µg/mL <sup>177</sup>Lu-lilotomab satetraxetan) (A) JNJ-7706621, (B) Alisertib, and (C) GSK461364. Right panels show Fa/CIs plots obtained by the Chou-Talaly method using the CompuSyn software. The fraction affected (Fa) is the fraction of non-viable cells relative to untreated control. For each combination with a specific Fa value the CI indicates whether the combined treatment is synergistic (<1) or antagonistic (>1). In blue are data from the same experiment as that shown in left panels, whereas red and yellow circles represent two independent experiments performed in U-2932 with JNJ-7706621. In experiment 2, 100, 266, 707, 1,880, and 5,000 nM of JNJ-7706621 were combined with the same doses of <sup>177</sup>Lu-lilotomab satetraxetan as in experiment 1, and in experiment 3 the doses of JNJ-7706621 were; 200, 532, 1,410, 3,760, and 10,000 nM. Error bars: STDEV of triplicate samples.

highly potent GSK461364 inhibitor itself. The high efficacy of GSK461364 alone may suggest its applicability (or that of equivalent PLK1 inhibitors) in mono-therapy for treatment of CD37-targeting RIT-resistant DLBCL or aggressive ABC-DLBCL. Clearly, further pre-clinical evaluation will be required to support this hypothesis. GSK461364 showed dose-related anti-tumor activity in a phase 1 study, but concomitantly lead to a high incidence rate of venous thrombotic emboli (37). Furthermore, acquired resistance due to up-regulation of ATP-binding cassette drug transporters has been reported for GSK461364 and alternative PLK1 inhibitors, including BI2536 and the clinically most advanced BI6727/Volasertib, potentially favoring combination over monotherapy strategies (38–40). Synergism of <sup>177</sup>Lu-lilotomab satetraxetan with alisertib was less strong than with JNJ-7706621, and confined to a narrow concentration window, since alisertib demonstrated a loss of efficacy at concentrations above 360 nM. The latter may be a concentration-dependent compound de-activating artifact of formulation in DMSO or a consequence of an antagonistic interaction with a secondary "off-target" inhibited enzyme. AURKA (STK15) is a driver and essential gene in DLBCL (41) and the identification of several other AURKA inhibitors in our screening campaign (not validated here) are supportive of further investigations. Pre-clinical and clinical investigations of alisertib in DLBCL treatment strongly suggest the need for combination drug partners, since re-commitment of treatment induced senescent aneuploid cells to cell cycle progression and low mitotic index in primary tumors are apparent causes of treatment resistance (29, 42–44). Hence, a more elaborate analysis of <sup>177</sup>Lu-lilotomab satetraxetan with a panel of AURKA inhibitors may prove valuable. This is strongly supported by the robust synergistic interaction of CD37-targeting RIT with JNJ-7706621 reported here. Dual-specificity and activity of JNJ-7706621 against CDK1/2 and AURA/B kinases was shown in TP53 proficient and deficient cell lines (45) and observed in U-2932 and RIVA cells. This inhibitor has shown anti-tumor activity in mouse xenograft models of solid tumors, but not been tested in clinical trials (30, 46). Of note, pre-clinical studies of JNJ-7706621 activity investigated its effect in the µM range, where anti-proliferative activity in monotreatment is evident (30, 45). Here, we showed that even sub-µM concentrations of JNJ-7706621 were sufficiently strong enough to synergize with <sup>177</sup>Lu-lilotomab satetraxetan, reducing the potential toxicity-risk of this compound and concomitantly lowering its required active concentration. The in vivo synergistic interaction of JNJ-7706621 and CD37-targeting RIT remain, however, to be proven.

Kinetic studies of the effect of mono- and combination therapy of U-2932 and RIVA cells with JNJ-7706621 and <sup>177</sup>Lu-lilotomab satetraxetan are suggestive of a model (**Figure 7**) in which radiation damage induced G2-arrested lymphoma cells eventually enter mitosis (repair or escape) and mitotic entry, progression and exit are impaired by JNJ-7706621 mediated inhibition of CDK1/2 and AURKA/B. DNA damage in mitosis is a known driver of chromosomal instabilities (47, 48). The extended residence-time of cells in mitosis due to chromosome condensation and congression defects as well as spindle and mid-spindle assembly failure is pivotal for the increased sensitivity to persistent <sup>177</sup>Lu-lilotomab satetraxetan deposited DNA damage, ultimately promoting cytokinesis failure (multinucleation, aneuploidy, increased cell size) and cell death by apoptosis.

In conclusion, CD37-targeting <sup>177</sup>Lu-lilotomab satetraxetan RIT showed activity in several ABC-DLBCL lymphoma cell lines. CD37-independent RIT-resistance was identified in two cell lines representative of aggressive DE ABC-DLBCLs with inactive TP53, and reversed by subsequent inhibition of CDK1/2 and AURKA/B by JNJ-7706621. These findings are of specific relevance for ongoing clinical trials of <sup>177</sup>Lulilotomab satetraxetan in relapsed, ASCT-non-eligible DLBCL, and may also be more generally applicable to other <sup>177</sup>Lubased RITs and alternative radionuclide utilizing targeted therapies. Future pre-clinical investigations are required to elucidate the potential application of CDK1/2 and AURKA/B inhibitors as a strategy to revert RIT resistance in TP53 deficient cancers.

### MATERIALS AND METHODS

#### Cells and Reagents

ABC-DLBCL cell lines were maintained in RPMI 1640- GlutaMAX medium (Gibco 61870-044) supplemented with 15% fetal bovine serum (Biowest S181B-050) and 1% penicillinstreptomycin (Gibco 15140-122) at 37◦C in a humidified atmosphere containing 5% CO2. Cell lines were obtained as previously described (21) and their identity was authenticated by short tandem repeat DNA profiling (IDEXX BioResearch, Ludwigsburg, Germany). Real Time Glo was from Promega (G9713). The Selleck Cambridge Cancer Compound library was obtained from and printed onto 384-well plates [384 well, PS, F-bottom, µclear, white, lid, sterile, Greiner Bio-One 781098 (82050-076)] by the High-Throughput Chemical Biology Screening Platform at the Center for Molecular Medicine

FIGURE 7 | showing percentage of cells positive for cleaved PARP (n = 4; error bars represent standard error of mean (n = 4). (A,B) Statistical significance in differences between treatment groups were tested by One Way ANOVA: \*p < 0.05, \*\*p < 0.01, \*\*\*p < 0.001. (C) Model: treatment with <sup>177</sup>Lu-lilotomab satetraxetan leads to DNA-damage induced G<sup>2</sup> arrest and apoptotic cell death. Cells resistant to treatment adapt and recover from the arrest. Inhibition of CDK1 and AURKA/B interferes with bipolar- and mid-spindle assembly, causing chromosome congression and cytokinesis defects. Combined treatment with JNJ-7706621 and <sup>177</sup>Lu-lilotomab satetraxetan reverses resistance likely by potentiating the effect of persistent radiation due to extended residence time in and failure of mitosis, the cell cycle phase in which repair capacity is low.

Norway (NCMM). Antibodies used were: rabbit-anti-phospho-BRCA2(Ser3291) (AB9986 Millipore), rabbit-anti-Cleaved-PARP (Asp214) Alexa647-conjugate (#6987, Cell Signaling), rabbitanti-phospho-histone 3-Ser10 (06-570, Millipore), Alexa488 anti-rabbit (A21206, Life Technologies), and Alexa647 donkeyanti-mouse (Jackson ImmunoResearch, UK). The DNA stain FxCycleTM Far Red (200 nM FxCycle and 0.1mg/ml RNase A) (Thermo Fisher Scientific) was used together with Pacific Blue (Thermo Fisher Scientific, P10163) staining, and Hoechst 33258 (Sigma-Aldrich) (1.5µg/ml) in other experiments.

### Labeling of Antibody With <sup>177</sup>Lu

The chelator (p-SCN-Bn-DOTA, Macrocyclics, TX, USA) was dissolved in 0.005 M HCl, added to the HH1-DOTA (lilotomab) in a 6:1 ratio and pH-adjusted to ∼8.5 using carbonate buffer. After 45 min of incubation at 37◦C, the reaction was stopped by the addition of 50 µL per mg of Ab of 0.2 mol/L glycine solution. To remove free p-SCN-Bn-DOTA, the conjugated antibody was washed using Vivaspin 20 centrifuge tubes (Sartorius Stedim Biotech, Göttingen Germany) 4–5 times with NaCl 0.9%. Before labeling with <sup>177</sup>Lu, the pH was adjusted to 5.3 ± 0.3 using 0.25 mol/L ammonium acetate buffer. Between 120 and 220 MBq of <sup>177</sup>Lu (ITG, Garching, Germany) was added to 1 mg of satetraxetan-Ab and incubated for 15–30 min at 37◦C. The radiochemical purity (RCP) of the conjugate was evaluated using instant thin-layer chromatography. If RCP was below 95% the conjugate was purified by elution through a Sephadex G-25 PD-10 column (GE Healthcare Bio-Sciences AB, Uppsala, Sweden).

### Treatment With <sup>177</sup>Lu-Lilotomab Satetraxetan (Betalutin)

Cells were treated in 6-well plates (2.5 × 10<sup>6</sup> cells/mL) without shaking for 18 h with Betalutin (specific activity of ∼600 Mbq/mg) at a final concentration of 1µg/ml for U2932 and 0.5µg/ml for RIVA (or as otherwise specified). After treatment, PBS was added to the cells, and the cells pelleted. Cells were first resuspended in 1 ml PBS, then washed twice in PBS and finally diluted in growth medium to desired concentration for the assay applied.

For the initial screening of ABC-DLBCL cell lines cells were incubated in deep well plates (NuncTM 96-Well Polypropylene DeepWell Storage Plates from Thermo Scientific) with 12 different concentrations of <sup>177</sup>Lu-lilotomab satetraxetan, including a control with no treatment, for 18–20 h with shaking. Cells were washed three times with PBS using a plate washer (ELx405 Select Deep Well Washer from BioTek) and seeded in 96-well plates in 0.2 ml medium and incubated at 37◦C/5% CO2. Fifty microliter of fresh medium was added after 72 h. The cytotoxic effect was measured at 144 h using the CyQuant NF Cell Proliferation Assay kit (ThermoFischer) for Ascent FL multiplate reader (ThermoFischer. IC50 were calculated using Prism (GraphPad) and clustering analysis performed in J-Express Pro (49).

### Combinatorial Drug Screen

Cells treated with <sup>177</sup>Lu-lilotomab satetraxetan and untreated control cells were seeded onto 384-well plates pre-printed with the 384-compound Cambridge Cancer Compound library sourced from SelleckChem. The library was divided onto two plates and a total of 48 no-drug controls were included per plate. Cell seeding densities were 120 000 cells/mL, and a volume of 25 µL was seeded in each well-giving 3,000 cells/well. The drug library was screened at two concentrations for each cell line (10/100 nM for RIVA and 10/1,000 nM for U-2932). Three days after seeding, 25 µL of diluted NanoLuc <sup>R</sup> luciferase and MT Cell Viability Substrate (RealTime-Glo) was added to each well. Cells were incubated with the reaction mix for 1 h at 37◦C before measuring luminescence in a Tecan Spark multimode microplate reader, with integration time set to 1 s. Luminescence readings were repeated each day for three consecutive days. A microplate sample processor (Precision XS, BioTek) was used to facilitate the cell seeding and dispensing of RealTimeGlo reagent. Hit candidates were identified using the Bliss Independence test for synergy. The effect of each drug alone (Fa) at each concentration was calculated as the fraction of dead cells as compared to control cells

$$Fa = 1 - \left(\frac{RLU\_{drag}}{averageRLU\_{control}}\right)^2$$

a similar calculation was performed for the average effect of <sup>177</sup>Lu-lilotomab satetraxetan alone (Fb)

$$\overline{Fb} = 1 - \left(\frac{a\nuera \text{ReLU}\_{177Lu-likelihood}}{a\nuera \text{ReLU}\_{control}}\right).$$

Through the following equations we found the expected additive effect (E) and the measured effect (M) of the combination of drug + <sup>177</sup>Lu-lilotomab satetraxetan:

$$\begin{array}{l} E = Fa + \overline{Fb} - Fa \ast \overline{Fb} \\ M = 1 - (\frac{RLU\_{combination}}{averageRLU\_{control}}) \end{array}$$

The Bliss score was defined as:

$$Bliss\,score = \frac{M - E}{1 - Fa}$$

The normalization to the fraction of viable cells in samples treated with drug alone (1-Fa in equation above) was carried out to compensate for the reduced population of cells that could be affected by combined treatment with <sup>177</sup>Lu-lilotomab satetraxetan. The standard deviation for the effect of <sup>177</sup>Lulilotomab satetraxetan alone in the 48 control wells was calculated on each plate as follows:

$$s = \sqrt{\sum\_{i=1}^{n} \frac{\left(Fb\_i - \overline{Fb}\right)^2}{n-1}}$$

Drugs with a Bliss score two times higher than the standard deviation of <sup>177</sup>Lu-lilotomab satetraxetan treated controls were scored as potential hits. Finally, we excluded inhibitory drugs that alone reduced viability by >90% to that of untreated control. Each plate was treated individually. In validation experiments the same procedure was used with the following modifications; Cells were treated in 12-well plates (1.2 ml per well, with 2.5 × 10<sup>6</sup> cells/ml) without shaking for 18 h with <sup>177</sup>Lu-lilotomab satetraxetan at final concentrations of 0.5, 1, and 2µg/ml for U2932 and 0.25, 0.5, and 1µg/ml for RIVA. After treatment, cells were seeded on 384-well plates preprinted with JNJ-7706621, Alisertib and GSK461364 in an 11 step concentration gradient (0-1280 nM). A total of 60 no-drug controls were included on each plate. Synergy of drug/177Lu-lilotomab satetraxetan combinations was determined by calculating Combination Indexes (CI) using the Chou-Talaly theorem (CompuSyn software) (50), where CI < 1 represents synergy. In replicates of validation experiment cells were treated with <sup>177</sup>Lu-lilotomab satetraxetan in 96 deep-well plates (100 µL per well, with 2.5 × 10<sup>6</sup> cells/ml) for 18 h, diluted to 40 cells/µL and 25 µL transferred to 384-well plates. Immediately after seeding a Tecan D300 Digital dispenser was used to administer drug to wells at 100, 266, 707, 1,880, and 5,000 nM (f.c. experiment 2) or 200, 532, 1,410, 3,760, and 10,000µM (f.c experiment 3).

### Flow Cytometry

For live/dead discrimination, <sup>177</sup>Lu-lilotomab satetraxetan treated cells were diluted to 500 000 cells/mL, transferred to T-25 cell culture flasks and inhibitor (JNJ-7706621) added to a final concentration of 500 nM. <sup>177</sup>Lu-lilotomab satetraxetan treated and untreated control samples, with or without inhibitor, were harvested for flow cytometry analysis before and at 24, 72, and 144 h after addition of the inhibitor (**Figure 5A**). At 72 h, 6 mL of fresh medium w/wo inhibitor was added to allow continuous growth of the cells that were harvested at the latest timepoint (144 h). Before fixation, cell pellets were resuspended in 200 µL PBS with 18 ng/µL of Pacific Blue and incubated at 4◦C for 20 min for live/dead discrimination. One milliliter of PBS was added to the samples, and cells were pelleted and resuspended in 1 mL of ice-cold 70% ethanol for fixation. Samples were stained with FxCycle Far Red for visualization of DNA, and analyzed on an LSR II flow cytometer (BD Biosciences) using FlowJo software. The same experimental setup (omitting Pacific Blue) was followed for assessment of PARP cleavage. Cells were fixed in 1.5% neutral buffered formaldehyde for 5 min at r.t, washed in PBS, and stored in Methanol (−20◦C). For analysis, cells were rehydrated and stained with anti-cleaved-PARP-Asp214 and Hoechst. Cell counts in **Figure 7A** were determined in quadruplicates for each time-point using a Countess Automated Cell counter (Thermo Fisher). For testing efficacy of JNJ-7706621, cells were arrested in mitosis by culturing for 16 h in medium containing nocodazol (0.04µg/ml). Still in the presence of nocodazol, cells were treated with JNJ-7706621 for 1 or 6 h at a final concentration of 250, 500, or 1,000 nM before fixation in icecold 70% ethanol. Samples were stained with primary antibodies against CDK and Aurora B targets phospho-BRCA2-Ser3291 and phospho-histone H3-Ser10, respectively.

CD37-expression in U-2932, RIVA and Oci-Ly10 was measured by determination of the binding capacity of cold antibody (HH1-dota). For accurate comparison of different cell lines, we included barcoding with CellTracer stain and pooled the samples from the different cell lines together in a single tube before staining with the CD37 antibody. To this, 2 × 10e6 cells were pelleted and resuspended in 1 ml PBS and labeled with CellTrace stain (U-2932: CFSE 1µM f.c., RIVA: Violet 2µM f.c., Oci-Ly10: blank) in the dark for 20 min at 37◦C with gentle shaking every 3–5 min. Cells were diluted in 5 ml of pre-warmed medium. After 5 min, cells were pelleted and resuspended in 1 ml fresh medium, pooled and an additional 1 ml of medium added. Cells were kept on ice thereafter. The cell mixture was divided into four polystyrene tubes and pelleted. Primary antibody (HH1-dota) was diluted in medium and added to three of the four tubes (no antibody, 1, 2, and 20 µg). Cells were incubated on ice for 20 min, washed twice with cold PBS and resuspended in medium containing Alexa647 donkey-anti-mouse (Jackson ImmunoResearch, UK) at 1 µl/ml. After 20 min incubation on ice, cells were resuspended in 1 ml cold PBS w/1% Fetal Bovine Serum and analyzed on a LSR II flow cytometer.

#### Microscopy

Cells were imaged using a CellObserver microscope system (Carl Zeiss) equipped with a 20×/0.8 PlanApo Phase 2 lens, a Hamamatsu ORCA-Flash4.0 v3 camera, a temperature controlled XL-chamber, a temperature, humidity and CO<sup>2</sup> controlled stage incubator, a motorized coded X,Y-stage, a Definite Focus system and a HXP120 Metal-Halide illumination unit. Visual inspection of cells prepared for flow cytometry was conducted using 96-well flat-bottom plates (Greiner Bio-One, Kremsmünster, AUT).

### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

#### AUTHOR CONTRIBUTIONS

FB, GR, JD, KM, RS, and SP contributed to the conception and design of the study. GR, KM, RG, SG, and SP performed the experiments presented in the study. Data analysis was carried out by GR, JD, RS, and SP. GR and SP wrote the first draft of the manuscript. FB, GR, JD, RS, and SP contributed to the manuscript revision. All authors read and approved the submitted version.

#### ACKNOWLEDGMENTS

The authors wish to express their sincere gratitude to Helene Solhaug (Nordic Nanovector ASA) for technical assistance, and to Idun Dale-Rein and Dr. Trond Stokke at the core facility for flow cytometry (Institute for Cancer Research, Norwegian

#### REFERENCES


Radium Hospital, Oslo University Hospital) for their invaluable expert-advice in design and conduction of quantitative multicolor flow cytometry experiments.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.01301/full#supplementary-material


and mutation status is predictive of PFS when CHOP is combined with 131 iodine tositumomab but not with rituximab: an analysis of SWOG S0016. Blood. (2018) 132:919. doi: 10.1182/blood-2018-99-111399


apoptosis in double hit or double expressor diffuse large B-cell lymphomas. Mol Cancer Ther. (2017) 16:2083–93. doi: 10.1158/1535-7163.MCT-17-0089


**Conflict of Interest:** SP: Nordic Nanovector ASA: Employment, Patent. KM, RG, and JD: Nordic Nanovector ASA: Employment, Equity Ownership, Patents. GR: Patent. RS: Institutional research funds from Nordic Nanovector ASA, Patent. FB: Institutional research funds from Acerta, ADC Therapeutics, Bayer AG, Cellestia, CTI Life Sciences, EMD Serono, Helsinn, ImmunoGen, Menarini Ricerche, NEOMED Therapeutics 1, Nordic Nanovector ASA, Oncology Therapeutic Development, PIQUR Therapeutics AG; consultancy fee from Helsinn, Menarini; expert statements provided to HTG; travel grants from Amgen, Astra Zeneca, Jazz Pharmaceuticals, PIQUR Therapeutics AG.

The remaining author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Rødland, Melhus, Generalov, Gilani, Bertoni, Dahle, Syljuåsen and Patzke. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Extracellular, Cellular, and Nuclear Stiffness, a Trinity in the Cancer Resistome—A Review

Sara Sofia Deville1,2,3 and Nils Cordes 1,2,3,4,5 \*

<sup>1</sup> OncoRay – National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Helmholtz-Zentrum Dresden - Rossendorf, Technische Universität Dresden, Dresden, Germany, <sup>2</sup> Department of Radiation Oncology, University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, <sup>3</sup> Helmholtz-Zentrum Dresden - Rossendorf, Institute of Radiooncology - OncoRay, Dresden, Germany, <sup>4</sup> German Cancer Consortium (DKTK), Dresden, Germany, <sup>5</sup> Germany German Cancer Research Center (DKFZ), Heidelberg, Germany

Alterations in mechano-physiological properties of a tissue instigate cancer burdens in parallel to common genetic and epigenetic alterations. The chronological and mechanistic interrelation between the various extra- and intracellular aspects remains largely elusive. Mechano-physiologically, integrins and other cell adhesion molecules present the main mediators for transferring and distributing forces between cells and the extracellular matrix (ECM). These cues are channeled via focal adhesion proteins, termed the focal adhesomes, to cytoskeleton and nucleus and vice versa thereby affecting the pathophysiology of multicellular cancer tissues. In combination with simultaneous activation of diverse downstream signaling pathways, the phenotypes of cancer cells are created and driven characterized by deregulated transcriptional and biochemical cues that elicit the hallmarks of cancer. It, however, remains unclear how elastostatic modifications, i.e., stiffness, in the extracellular, intracellular, and nuclear compartment contribute and control the resistance of cancer cells to therapy. In this review, we discuss how stiffness of unique tumor components dictates therapy response and what is known about the underlying molecular mechanisms.

Keywords: stiffness, extracellular matrix, cancer resistome, radio(chemo)resistance, cell-extracellular matrix interaction, focal adhesions, solid stress

#### INTRODUCTION

Stiffness refers to the rigidity of a material or the extent to which the material can resist to deformation or deflection in response to an applied force (1). Typically, stiffness depends on properties of the material such as the composition and organization of the building elements. A stiff as compared to a flexible structure is less susceptible to deform under an external load and, consequently, apt to develop greater stress.

Generally, the composition of the extracellular matrix (ECM) determines the stiffness of a tissue (2, 3). Cells are surrounded by ECM providing structural and biochemical support. Eventually, these interactions present the fundamental organization unit of multicellular complex development into tissues. The ECM comprises two classes of macromolecules: polysaccharide chains and fibrillar proteins (4). The polysaccharide chains are covalently bound to transmembrane proteins and assemble into proteoglycans. The fibrillar proteins like collagens, fibronectin, elastin, and laminins have structural functions and serve as ligands for cell adhesion molecules. The proteoglycans

#### Edited by:

Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands

#### Reviewed by:

Jon Humphries, University of Manchester, United Kingdom Mirjam Zegers, Radboud University Medical Center, Netherlands

#### \*Correspondence: Nils Cordes nils.cordes@oncoray.de

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 05 September 2019 Accepted: 22 November 2019 Published: 06 December 2019

#### Citation:

Deville SS and Cordes N (2019) The Extracellular, Cellular, and Nuclear Stiffness, a Trinity in the Cancer Resistome—A Review. Front. Oncol. 9:1376. doi: 10.3389/fonc.2019.01376 form a gel-like structure in which the fibrillar proteins are embedded. Mesenchymal cells such as fibroblasts are responsible for the production and secretion of ECM proteins (5–8). The ECM is constantly reorganized and its dynamic is modulated by growth factors, cytokines, hormones, and extracorporal factors influencing significantly tissue physiology, morphology, homeostasis, and repair (9). A primary source of ECM restructuring is re-synthesis or proteolytic degradation by matrix metalloproteinases (MMPs) (10, 11). Several studies indicate that the spatio-temporal organization and dynamic re-modulation of the ECM has extensive biological implications for tumorigenesis promotion, progression, and metastasis.

In general, the outgrowth of a tumor produces an additional physical pressure, also defined as stress, on the host tissue and this is reciprocally balanced by the physical stress generated by the host tissue on the tumor. To overcome the stress enforced by the host tissue, tumor stiffening is essential for allowing host tissue displacement and growth in size (12). Tumors modulate their surrounding microenvironment including ECM, which results in alterations of tissue stiffness, porosity, and organization (13). A number of studies demonstrated specific changes in the mechanical properties of tumors over the time of their progression. When measured as single component, cancer cells and their nuclei become softer compared to normal cells (14, 15) suggesting a dis-regulation of cellular signaling pathways, cell proliferation, migration, survival, and treatment resistance (16, 17).

Fundamental for cell stiffness and mechanical forces are focal adhesions, serving as nexus between cytoskeleton and ECM (18–20). Cell adhesion elicits activation of different cytoplasmic signaling pathways for co-regulation of pro-survival mechanisms (5–8). Key mediators of this adhesion are integrins, an essential family among cell adhesion molecules. ECM reorganization drives significant changes in the integrin-mediated signaling pathways fundamental for tumor development and response to chemo- and radiotherapy (21–24). Various studies in normal (e.g., human fibroblasts and keratinocytes) and tumor cells (e.g., glioblastoma, pancreatic carcinomas, bronchial carcinomas, melanomas, breast cancers) documented adhesion to ECM to enhance resistance to ionizing radiation, chemotherapy, and molecular therapies (25, 26). These mechanisms are referred to as cell adhesion-mediated radioresistance (CAM-RR) and cell adhesion-mediated drug resistance (CAM-DR) (**Figure 1**) (25, 27).

FIGURE 1 | Extracellular matrix (ECM), cellular end nuclear stiffness are regulated by several factors. The ECM remodeling is highly dependent on cancer associated fibroblasts (CAFs). The cell stiffness instead is regulated by integrins and focal adhesion proteins (FAPs), which contribute to cancer radio- and drug-resistance by mediating cell adhesion to the extracellular matrix. Upon cell adhesion to ECM, integrins induce pro-survival signaling cascades mediating radiotherapy- and drug-resistance (CAM-RR and CAM-DR). Finally, the nuclear stiffness is regulated by the levels of lamin-A/C and chromatin condensation. Created with BioRender.

This review gives insights into recent findings of how tissue, cellular, and nuclear stiffness are associated with therapy resistance and discusses the underlying mechanisms.

#### REGULATION OF CANCER THERAPY RESISTANCE THROUGH ECM REMODELING AND STIFFNESS

During tumor progression, cellular and genomic alterations occur, which are accompanied by changes in mechanical properties in the intracellular- and extracellular environment. The ECM is a key component of the tumor microenvironment, which interacts with cancer cells and regulates signaling cascades through focal adhesion proteins (FAPs) (28–30).

The ECM of tumors, primarily composed of fibrous tissue, becomes stiffer due to an increase of fiber cross-linking (31, 32). This is in line with the development of desmoplasia during carcinogenesis. Desmoplasia is an intense fibrotic response characterized by the formation of dense ECM (31). Tumors with high desmoplasia are considered to be more aggressive and associated with worse prognosis (31). Reports showed cancerassociated fibroblast to (over-)secrete matrix and modulate tumor phenotype and therapy response (31). These dynamic ECM modifications alter the ECM mechanical properties: degradation, re-polymerization, and alignment, contributing to a re-arrangement of ECM fibers and strain-induced stretching (33–38). In order to remodel the matrix, cancer cells and CAFs release enzymes, such as MMP and lysyl oxidases (LOX), which degrade and crosslink the ECM, respectively (**Figure 2A**). A structural analysis of the fibrillary collagen revealed the presence of reorganized collagen in the tumor-stromal interphase (39). Moreover, it was demonstrated that an increased collagen alignment and fiber thickness is a negative prognostic marker for cancer, supporting the significance of ECM dynamic in cancer progression (40–43).

ECM stiffness is related to a high malignant tumor phenotype (16). This can be explained by: (a) limited distribution and penetration of drugs (44) and/or (b) alterations in integrin signaling, focal adhesions, Rho/Rho-associated protein kinase (ROCK) pathway activation, as well as actomyosinand cytoskeletal-dependent cell contractility and increased Ca2<sup>+</sup> influx through mechanosensitive channels (28, 45, 46). For instance, integrins respond to the force alteration by rearrangement and aggregation in clusters at the plasma membrane. The cluster is composed of multiple mechanosensors (e.g., talin, vinculin), signaling molecules [e.g., focal adhesion kinase (FAK), proto-oncogene tyrosine-protein kinase Src (SRC), Phosphoinositide 3-kinase (PI3K)], adapter proteins [paxillin, LIM, and senescent cell antigen-like-containing domain protein 1 (PINCH1)], and actin linker proteins (e.g., filamin, alphaactinin), which physically connect integrins to the cytoskeleton. On stiff substrates, the resistance to cellular tension leads to talin stabilization mediated by vinculin binding and also enhances FAK activation. These are some of the key mediators of the transmission of contractile forces to the cytoskeleton (17).

The higher aggressiveness also originates from matrix stiffness-induced epithelial to mesenchymal transition (EMT) being accompanied by cancer cell migration and invasion due to loss of intercellular adhesions (**Figure 2A**) (47, 48). EMT has been found to be related to treatment resistance (47, 48).

Another feature of the intra-tumoral microenvironment regulated by stiffness is the high interstitial hydrodynamic pressure induced by hypervascularization during tumor development. Such pressures have been found to promote tumor progression by impairing vessel function through constriction, thereby limiting tumor oxygen and nutriment supply, also known as hypoxia (44). Hypoxic tumors are known to be resistant to anticancer therapy, including radiation therapy, chemotherapy, and targeted therapy (44, 49).

Additional determinants of tumor stiffness are genetic mutations as they are not limited to initial driver mutations but encompass a wide genomic variation corresponding to the normal tissue where a tumor arises. High stiffness is correlated with dense collagen matrices resulting in small pore sizes for cells to transverse (50, 51). These events can drive genomic diversity through DNA damaged during migration. Metaanalyses showed that tumors originating from stiff tissues (e.g., lung, skin, bone) have substantially higher somatic mutations and chromosome copy numbers than malignancies originating from soft tissues (e.g., bone marrow, brain) (50). There are three hypotheses stating stiffness to drive genomic instability: (a) stiffness induces cell proliferations, increasing the probability to acquire spontaneous mutations; (b) stiffness increases the frequency of nuclear envelop rupture; (c) invasion of cancer cells through packed-tissue environments causes cell selection with a more aggressive phenotype (50).

Altogether, to gain a better understanding in the dynamics of cancer, it is necessary to uncover the effects cellular and extracellular mechanical properties elicit on tumor growth, metastatic spread and therapy resistance (**Table 1**). The basis of these events is cell behavior, which profoundly depends on mechanical properties and forces controlling signaling pathways involved in cell differentiation, proliferation, migration, and survival mechanisms (64–69).

The role of the ECM in treatment resistance was predominantly investigated for chemotherapeutics, with breast cancer being one of the most frequent models used (**Table 1**). As a matter of fact, one of the early detection methods for this tumor entity was the determination of abnormal stiffness by palpation or using medical devices. By means of an poly(ethylene glycol)-phosphorylcholine (PEG-PC) hydrogel system, Nguyen et al. examined the response of breast cancer cells to the Raf kinase inhibitor sorafenib in different stiffness substrates (55). The efficacy of sorafenib was reduced depending on stiffness and collagen concentration but independent of the commonly associated ROCK activity. Instead, triple negative breast cancer cells sustained an activation of JNK mediating the drug resistance (55). However, the combination of a JNK inhibitor with sorafenib eliminated the stiffness-mediated resistance. Strikingly, they found out that ERK (extracellular-signal-regulated kinase) and p38 (mitogen-activated protein kinases) were not involved in the drug resistance, it was rather regulated by β1 integrin (55).

FIGURE 2 | (A) The extracellular matrix (ECM) secretion depends mainly on cancer associated fibroblasts (CAFs). The dynamic reorganization of ECM is regulated by matrix metalloproteinase (MMP)-dependent matrix degradation and lysyl oxidase (LOX)-dependent ECM crosslinking. Changes in ECM and stiffening leads to: (a) epithelial-to-mesenchymal transition (EMT) enhancing cell migration and invasion, (b) limited drugs distribution, (c) genomic alterations resulting in clonogenicity and heterogeneity, and (d) the activation of key adhesion proteins, such as integrin. (B) Integrin-dependent outside-in signaling mechanisms regulated cell adhesion to ECM as part of their role in cancer radio- and drug-resistance. Many of these mechanisms involve the focal adhesion kinase (FAK). (C) The Linker of Nucleoskeleton and Cytoskeleton (LINC) complex is composed of two families: KASH located at the nuclear membrane exterior (NME) and SUN situated in the nuclear membrane interior (NMI). LINC regulates the physical transmission of forces generated by the ECM and cytoskeleton. Moreover, a low expression levels of lamin-A/C is correlated with a high cell migration and an increase of therapy resistance. Cells adjust to mechanical tensions by enhancing the expression level of lamin-A and phosphorylated emerin. LINC complex detaches from the nucleus and cytoskeleton to maintain DNA integrity when cells fail to manage the tension. Created with BioRender.

Moreover, Joyce et al. showed that extrinsic resistance is associated with matrix stiffness (56). As culture model, an innovative 3D alginate-based hydrogel system enabling dynamic ECM stiffening over time was used. The results displayed a stiffness-dependent response to the chemotherapeutic doxorubicin in triple negative breast cancer cells (MDA-MB-231), while a non-triple negative cell model (MCF7) failed to show the same stiffness-dependent resistance (56). This differential therapeutic response was correlated with a nuclear translocation of YAP, a marker of mesenchymal differentiation. In fact, a higher level of nuclear YAP was found in MDA-MB-231 relative to MCF7 cells (56).

Another example of a cancer entity with poor prognosis that seems to be dependent on ECM stiffness and EMT is pancreatic cancer (62). Pancreatic cancer is one of the stiffest human solid carcinomas characterized by a fibrotic reaction, leading to the


Research methods and treatment are included.

activation of EMT-related and prosurvival signaling pathways (62). Rice et al. reported that in vitro PDAC cell lines cultured on varying stiff polyacrylamide gels had different behavior than the corresponding tumors in vivo. Resistance to gemcitabine, a therapeutic drug that inhibits DNA synthesis and transcription, was shown to be unchanged with increased rigidity, although matrix rigidity still promoted EMT. In contrast, cells grown on stiff gels showed increased resistance to paclitaxel (a taxane that stabilizes microtubules preventing mitosis) compared with the softer conditions (62).

The second most studied tumor entity, in terms of matrix stiffness, is the hepatocellular carcinoma (HCC) since it often relates to liver fibrosis. Various studies demonstrated resistance to cisplatin, sorafenib, paclitaxel, 5-FU, and oxaliplatin to depend on ECM stiffness (52–55). It was also shown that a large number of cells were dormant and carrying stem cell-like characteristics in HCC when cultivated in low stiffness (52). Liu et al. cultured HCC cells in alginate gel beads with different degrees of stiffness (53). Cells cultured in the stiff matrices resisted to cisplatin, 5-FU, and paclitaxel, whereas cells in the soft environment were sensitive to these agents. Moreover, cells encapsulated in alginate beads highly express ABC transporters and endoplasmatic reticulumrelated proteins compared to 2D growth conditions. These proteins are supposed to contribute to drug resistance of solid tumors and treatment failure.

A recent study focused on the matrix stiffness-mediated effects in HCC stem cells (54). In this work, the authors showed that, when the substrate stiffness is increased, HCC cells exhibit an elevated number of CD133(+)/EpCAM(+) positive cells (stem cells markers). The increase in this cell population was accompanied by elevated expression levels of EpCAM, Nanog, and SOX2 (54). Moreover, the phosphorylation levels of Akt and mTOR were upregulated showing a greater self-renewing ability and oxaliplatin resistance. Interestingly, when these populations were subjected to integrin inhibition, all the previous described effects were attenuated, suggesting that integrin β1 may deliver higher stiffness signal inside HCC cells activating stemness associated signaling cascades (54).

Opposite to the results from You et al. (54), human laryngeal squamous cell carcinoma (Hep-2) cells cultured in a low stiffness environment showed an enhanced expression of stem cell markers (61). In addition, under the low stiffness environment, Hep-2 cells underwent less apoptosis to cisplatin and 5-FU. The authors suggested that the observed chemoresistance is related to increased Sox2 levels and an upregulation of the ABCG2 protein, a membrane xenobiotic transporter connected to multidrug resistance (61). These examples illustrate the diversity of resistance mechanisms in different tumor entities, suggesting that there is no "one-for-all" approach, and thus only tumor-specific studies shed light on the mechanisms.

Tokuda et al. studied the effect of stiffness on the treatment response of melanoma cells, showing a cell-line dependent effect (58). Cells were grown in different PEG hydrogels with variable tensile moduli and treated with a BRaf inhibitor— PLX4032. Cells derived from radial growth phase (WM35) presented stiffness-dependent chemoresistance in contrast to the metastatic melanoma cells (A375) (58). A recent study on therapeutic relapse to another BRaf inhibitor—vemurafenib used serial biopsies of genetically modified mice (59). Nextgeneration sequencing and single-cell transcriptomics enabled


TABLE 2 | Cell stiffness and related causes in different tumor entities, together with an overview of the methods for measuring cell stiffness.

tracking of the evolution of multiple cellular "compartments" within individual lesions during first-line treatment response, relapse, and second-line therapeutic interventions (59). It became clear that tumor relapse is genetically stable, while differentiation state and ECM contribute significantly to the resistant phenotype. The result from in vitro experiments presented a correlation between cell state changes and ECM remodeling, suggesting an increased tumor stiffness modulates tumor cell fate and reduces treatment responses (59).

For glioblastoma, the most common brain tumor in adults (70), no physiologically relevant model is currently available for exploring effects of cellular stiffness. The majority of investigations on stiffness applied 2D cultures system. Erickson et al. suggested a newly developed and characterized Chitosan-Hyaluronic Acid scaffold with varying stiffness for glioblastoma cell culture (63). They showed glioblastoma cells to form large spheroids in stiff scaffolds exhibiting a higher degree of drug resistance and a more invasive phenotype relative to 2D models (63).

Altogether, we conclude that an increase of ECM stiffness leads to enhanced therapy resistance, with some exceptions that might be tumor- or substrate/matrix-dependent. ECM stiffness, therefore, might be used as a physical marker for the prediction of tumor therapy resistance. Certain contradictory issues, especially in terms of stemness, need to be clarified. Cancer stem cells are a well-known factor of therapy resistance and more studies are necessary to understand how these subpopulations behave in different stiffness substrates.

#### REGULATION OF CANCER RESISTANCE THROUGH CELLULAR STIFFNESS

Regulation of cellular stiffness is typically dictated by a variety of factors such as cytoskeleton organization, number of focal adhesion clusters, and nuclear deformability. Generally, cancer cells tend to be softer than their normal counterpart (= tissue of origin) depending on the status of their malignant transformation (35, 71–77).

Using magnetic tweezers to probe cellular resistance to physical force, a study in ovarian cancer cells demonstrated that the migration and invasion potential are inversely proportional to cellular stiffness. Moreover, some treatments such as pharmacological myosin II inhibitors reduce cellular stiffness and, therefore, convert cancer cells into a more invasive phenotype (75, 78). Pathways regulating these mechanical cues may potentially serve as targets for molecular cancer therapy.

Cellular stiffness is also determined by particular membrane proteins found in focal adhesions. FAPs assemble into protein complexes and act as connecting and adaptor proteins between ECM and the cellular interior (18–20). The complexes transmit extracellular signaling and mediate a strong interaction with the actin cytoskeleton. In many cancers, these proteins are deregulated, resulting in abnormal cell-cell and cell-ECM adhesion. Integrins are commonly overexpressed in tumors and affect growth rate, cellular morphology, and invasiveness (28, 79, 80). Integrin activation triggers cytoskeletal re-arrangements through the regulation of signaling cascades like Src- and FAK and their downstream signaling pathways for therapy resistance (81).

The effects of cellular biophysical properties fundamental for therapy resistance remain to be clarified (**Table 2**). Liu et al. used a microfluidic platform to evaluate cancer cell transportability and invasiveness in heterogeneous breast cancer cells (90). Cell transportability is determined by cellular stiffness and cell surface frictional property, allowing the discrimination between more and less invasive phenotypes (90). The same principle was applied in another study. Leukemic cells treated with daunorubicin were sorted according to their cellular stiffness using a microfluidic device (88) uncovering cellular physics to serve as distinctive features between chemoresistant and -sensitive cells. Softer cells showed an alteration in multiple mechanisms related to drug resistance, including decreased sensitivity to apoptosis induction, enhanced metabolic activity, and regulation of key genes involved in extrusion of drugs such as CYP supergene family typically involved in drug resistance (88).

Using lab on chip technology, several studies investigated the influence of cell deformability on the chemotherapeutical response of ovarian, breast, and prostate cancer. It became clear that cisplatin-resistant ovarian cancer cells have more elastic deformation capability relative to cisplatin-sensitive cell populations (86). Although these results seem exciting, they are in contrast to the study of Sharma et al. showing that cisplatinresistant ovarian cancer cells are stiffer than their normal counterpart. This stiff phenotype is characterized by cytoskeletal long actin stress fibers mediated by Rho GTPases (85).

In line with this, paclitaxel-resistant prostate cancer cells were shown to be stiffer than the non-resistant counterpart. Kim et al. showed that paclitaxel-resistant cells gain mobility and invasiveness through increased EMT (84). Moreover, enhanced cell migration and invasion of paclitaxel-resistant cells was facilitated by increased cytoskeleton remodeling dynamics, stiffness, traction forces, and by a repression of keratin 8/18/19. In this work, the authors observed that resistant prostate cancer cells, despite being stiffer than the non-resistant cells, showed a more fluid-like behavior leading to a higher invasion capability (84).

In another study, matrices of different stiffness were used to understand the cellular behavior of different breast cancer cell lines (82). Interestingly, the most aggressive cells (MDA-MB-231) were softer when cultured on glass substrate, but when these cells were cultured on soft matrices they presented a stiffer phenotype compared to the other cell lines cultured in the same matrix. This is a good example of how the environment modulates cellular mechanical properties (82).

A similar work on breast cancer cells using matrices of different rigidity discovered a direct correlation between migration capacity and increase of matrix stiffness (83). Moreover, cells treated with cetuximab, an epidermal growth factor receptor (EGFR) inhibitor, had an increased elastic modulus followed by a decrease in migration ability. Here, the authors explained that cell mechanics are not only regulated by mechanical cues of the ECM but also by biochemical signals mediated through membrane receptors, such as EGFR (83).

Another study investigating environmental effects on liver cancer stem cells is from Sun et al. The authors investigated the effects of shear stress on cancer stem cell signaling regulating cellular migration, proliferation, and differentiation (87). It was found that certain shear stresses promote cell migration through activation of FAK and ERK1/2 signaling pathways. Moreover, shear stresses were responsible for lowering cellular stiffness in line with disrupted F-actin organization (87).

Environmental factors can also regulate epigenetic signatures such as methylation (89). Using cell lines with methylated tumor suppressor genes (e.g., hypermethylated in cancer 1—HIC1, Rasassociation domain family member 1A—RassF1A), a Taiwanese group investigated cell stiffness changes depending on the methylation status and found that the stiffness of the methylated cells was lost, followed by a decrease of tubulin expression and Factin disorganization (89). Further experiments involving cellular relaxation after cell compression showed that cancerous cells also have increased acto-myosin cortex contractility as compared to corresponding healthy cells (74, 91). Moreover, the higher the invasive level, the greater the cellular recovery behavior.

In contrast to ECM stiffness, cellular stiffness seems not to correlate with treatment resistance. Although there is a prevalence that a decrease cellular stiffness leads to an increase resistance, this assumption is often uncertain due to several factors: (1) measurement technique, (2) cell culture methodology, and (3) tumor entity/heterogeneity.

### INTEGRINS BRIDGING BETWEEN ECM AND CELLULAR STIFFNESS: EFFECTS ON THE RESISTOME

After years of research, it became obvious that cell adhesion is fundamental for cell survival (92). Furthermore, a number of studies showed that cell adhesion is associated with the refractory to cancer treatments (92, 93). The principles of treatment resistance of cancer modulated by cell adhesion were proposedly categorized into: (i) cell adhesion mediated radioresistance (CAM-RR) and (ii) cell adhesion mediated drug resistance (CAM-DR) (**Figure 1**). Diverse adhesion resistomes composed of integrins, adaptor proteins, kinases, and cytoskeleton mainly contribute to both resistance mechanisms (92, 93). Interestingly, the components of the adhesion resistomes are widely heterogeneous depending on tumor entities. These might be also related to the tissue of origin.

To form multicellular structures or tissues, cells need to attach to adjacent cells via cell-cell contacts and anchor to the ECM through the transmembrane adhesion receptors known as integrins. An integrin receptor is a non-covalent heterodimer consisting of an α and a β integrin subunit. To date, there are 18 α and 8 β integrin subunits allowing the formation of 24 different integrin receptors. These α and β combinations determine the binding specificity of the integrin (29). Essentially, integrins consist of a big extracellular ectodomain, a transmembrane domain and a short cytoplasmic tail (29).

In the last two decades, substantial studies on cell adhesion to ECM primarily focused on integrins. Integrins and their downstream FAPs are known as mechano-sensors and mechanotransducers that sense and transduce mechanical signals into chemical signals. Generally, normal tissues weakly express integrins and FAPs. In contrast, cells start to express them when cells are grown in an in vitro tissue culture surfaces, indicating that cell culture stiffness highly impacts on the expression of these proteins (94).

Most integrins are not constitutively active and are located at the cell surface in an inactive state. Integrins are bi-directional signal receptors stimulated in two ways: the inside-out and outside-in activation (95). Both activation pathways are based on a conformational change in the ectodomain of the integrin (**Figure 2B**). The ability of integrins to signal in an inside-out and outside-in manner may be exquisite in normal cells but it is deleterious in cancer cells (92). The outside-in signaling is better understood with regard to its role in the cell adhesion resistome to elicit CAM-DR and CAM-RR compared to the inside-out signaling, which is rarely investigated in cancer cells (96–107).

During the inside-out signaling, the cytoplasmic domain of the integrin binds and stimulates intracellular proteins such as kindlin or talin. By integrin conformational changes, there is an increased binding affinity for extracellular ligands. This activation mechanism controls, among other things, the migration of cells. With the help of outside-in activation, which is mainly dictated by ECM properties, integrins can introduce information into the cell. The extracellular binding of a ligand also leads to conformational changes of the integrin and activation of intracellular signaling pathways (**Figure 2B**). Often this signaling pathway recruits and activates kinases such as FAK and SCR, and also the RAS-MAPK (mitogen-activated protein kinase) and PI3K (phosphoinositide 3-kinase)—AKT (RAC-alpha serine/threonine Protein kinase) signaling nodes (42). Moreover, both signaling pathways, inside-out and outsidein, are powerful and can activate each other (108–114).

Binding of integrins to ECM proteins is mediated by short amino acid sequences. Motifs that can bind these sequences are (1) the RGD (arginine-glycine-aspartate) motif in fibronectin and laminin or (2) the DGEA (aspartate-glycine-glutamic acidalanine)—and the GFOGER (glycine-phenylalanine-glycineglutamic acid-arginine) motif in collagen (29, 115). Intracellular adapter proteins such as paxillin, parvin, or talin link integrins to the actin cytoskeleton, generating a bridge between ECM and cytoskeleton. Although integrins do not have intrinsic kinase activity, they recruit and activate a large spectrum of kinases to the cytoplasmic subunit. As a result, important cellular processes such as proliferation, apoptosis, differentiation, migration, or cell survival regulated (**Figures 1**, **2B**) (92, 116–119).

Beta-1 integrins are the largest subgroup of integrin adhesion receptors (29). Inhibition of β1 integrins, leads to an inactivation of a variety of integrin receptors such as for laminins, fibronectin, and collagens. This property, as well as the upregulated expression of β1 integrins in a variety of tumors, make β1 integrins a promising target molecule for cancer therapy (92, 120). The resistance of tumors to radiation and chemotherapy is dependent on the β1 integrin adhesion to ECM proteins. A collection of studies showed the importance of β1 integrin-mediated pathways for radiation resistance and survival, differentiation and proliferation, as well as for tumor progression and metastasis (34, 100, 101, 106, 107, 121–126).

In clinical trials, some inhibitors against β1 integrin receptors have been used. These include three inhibitors against the fibronectin receptor α5β1 integrin: ATN-161, Volociximab (M200) and JSM6427. ATN-161 is a peptide, which acts as an antagonist of the α5β1 integrin and blocks the receptor. Phase I studies showed that the use of ATN-161 had no risks or side effects (127). Volociximab, a humanized monoclonal antibody, has been reported as an angiogenesis inhibitor developed for solid tumors. Treatment with volociximab was in Phase I and no adverse reactions nor dose-related toxicity was observed (127). Further clinical studies in metastatic melanoma and renal cell carcinoma have shown promising effects upon volociximab treatment (128). The third α5β1 integrin inhibitor is the small molecule JSM6427 which inhibits angiogenesis and fibrosis and has so far only been tested in preclinical studies (127, 129).

Therefore, preclinical examinations have already described the importance of β1 integrin-mediated adhesion to ECM for survival of tumor cells after irradiation. Studies in several tumor entities were able to demonstrate that the inhibition of β1 integrin leads to radiation sensitization in glioblastoma cells (81, 130), lung carcinoma cells (131), colon carcinoma cells (132), breast carcinoma cells (133, 134), and HNSCC cells (121, 135). In vitro data from 3D cultured cells and data from xenograft tumors confirmed that inhibition of β1 integrins reduces significantly the radiation resistance of tumors (121, 134).

Depending on the integrin receptor and the tumor entity, integrins activate different survival-promoting signaling pathways. In breast cancer cells, the PI3K-AKT signaling pathway is mainly activated leading to adhesion-mediated radiation resistance (134). Integrins modulate the FADD (caspase-8/Fas-associated protein with death domain) signaling pathway, which is of importance for cell survival, resulting in the resistance to radiation induced cell death in leukemia cells (136). In HNSCC, FAK is the central signaling molecule for the β1 integrin-mediated signaling pathways and plays an essential role for cell survival after irradiation (116, 121). Data from our group showed that the inhibition of β1 integrin dephosphorylates FAK, causing the FAK/cortactin complex dissociation. This leads to the inactivation of JNK1 and the radio-sensitization of tumor cells (121). FAK consists of an N-terminal FERM (protein 4.1, ezrin, radixin, moesin homology) domain, a kinase domain, and a FAT (focal-adhesion targeting) domain (137, 138). The FERM domain mediates various interactions of FAK, e.g., with the EGFR. The FAT domain is responsible for the recruitment of FAK to the focal adhesion site. It binds integrin-associated adapter proteins such as talin or paxillin. In addition, FAK contains three proline-rich ones Regions (PRR1-3) that help proteins target a SH3 (SRC-homology 3) domain contained as e.g., p130Cas (139).

FAK can be phosphorylated on various tyrosine residues (139). The autophosphorylation of the tyrosine 397 site is triggered through the bond of β1 integrin to the ECM. In HNSCC cells, inhibition of β1 integrins leads to the dephosphorylation of FAK on tyrosine 397. This phosphorylation site is therefore used as a control of a functional β1 integrin inhibition (121). Interestingly, Lim and colleagues identified that FAK plays an important role in the nucleus. They showed that p53 binds to the FERM domain of FAK, thereby modulating cell survival and proliferation (140, 141). This observed function in the nucleus suggested that FAK has additional nuclear functions and, thus, might contribute to the rectification of radiation-induced DSB. In line, our group demonstrated that the non-homologous endjoining DNA double strand break repair pathway is partially co-controlled by β1 integrins via the FAK/JNK1 signaling axis (100). The significance of integrins in DNA repair processes was further emphasized by Christmann and colleagues. They observed that the αV/β3 integrin signaling axis coordinates the homologous recombination repair pathway in glioblastomas (142). Upon simultaneous temozolomide treatment and αV/β3 integrin knockdown, glioblastoma cells presented increased DNA double strand breaks and a depletion of Rad51 expression, indicating an impaired homologous recombination (142).

Furthermore, we have shown that β1 integrin targeting leads to an induction of the EGFR signaling cascade and the double targeting of β1 integrin and EGFR achieved a greater radiosensitization compared to the single targeting approaches in vitro and in vivo (101). This suggests a more efficient suppression of FAK/ERK (extracellular-signal-regulated kinase) prosurvival signaling upon the combination treatment of anti-β1 integrin/anti-EGFR treatment than the single therapy.

To date, only a few studies attempted to investigate the crosstalk between integrins and receptor tyrosine kinases (RTK) and its effect on cancer cell therapy resistance. Therefore, more studies are needed to identify the therapeutic potential of such combination therapy approaches. One of our recent study showed that HNSCC cells, which basically poorly respond to EGFR and β1 integrin blockage, were radiosensitized by the inhibition of targets identified from a whole exome sequencing (123). Briefly, we identified different gene mutation profiles in the non-responder HNSCC cell lines to EGFR and β1 integrin inhibition compared to the responder HNSCC cell lines. These profiles would allow the stratification of HNSCC patients and the identification of potential targets to address the treatment resistance. Kelch Like ECH Associated Protein 1 (KEAP1) and Mammalian target of rapamycin (mTOR) were identified as key targets. The pharmacological inhibition of KEAP1 or mTOR together with β1 integrin and EGFR effectively increased non-responder radiosensitization (123). The study suggested a therapeutic approach to identify a potential combination therapy and to promote identifications of novel targets.

In summary, we can assert that integrins and FAPs essentially contribute to therapy resistance and the possibility of targeting these proteins could be developed as a therapeutical option in combination with radiotherapy and chemotherapy.

### REGULATION OF CANCER RESISTANCE THROUGH NUCLEAR STIFFNESS

During tumorigenesis, in addition to altered ECM stiffness, contractility of the cytoskeleton, and cell adhesion, stiffness of the cell's nucleus actively or passively adjusts to the process of malignant transformation. A growing number of studies report a modified nuclear envelope structure and composition in cancer cells (143). The nuclear envelope, consisting mainly of lamins and nuclear pore complexes, was identified as the major structure that is modulated in cancer (143, 144). The nuclear envelope contributes to cellular mechanical properties and functions and determines nuclear deformability (145). It is also involved in mechano-transduction and transmission of forces to the nucleus. Cancer progression promotes modifications in the composition of the nuclear envelope generating softer and highly lobulated nuclei, which consequently allow cancer cells to invade dense tissues more easily (143, 146). The nuclear stiffness is predominantly modulated by mechano-signals communicating between ECM and nucleus. Physical interactions of nucleus and cytoskeleton are essential for cytoskeletal organization and cellular polarization, which influence cell migration for metastasis (147). Moreover, the interaction seems to induce rearrangements in chromatin structure and lamin expression via intranuclear signaling cascades (143, 144, 146).

A study using a microfluidic channel with a narrow constriction to investigate the stiffness of prostate cancer nuclei showed that the nuclear rigidity is reduced in more malignant phenotypes. Furthermore, prostate cancer cells expressed a more aggressive phenotype when a low expression level of lamin-A/C and a decreased chromatin condensation were present (148). This supports the hypothesis that cancer cells with softer nuclei metastasize more efficiently. The importance of nuclear stiffness in cellular migration was also shown in lung carcinoma and glioblastoma multiforme. Generally, lamin-Bs are more stably expressed than lamin-A, of which the expression level widely varies among normal and cancerous solid tissue cells. In this study, cells with low levels of lamin-A expression showed the most pronounced increase in 3D migration. Of key importance was the finding that the cellular migration was biphasic in lamin-A expressing cells as wildtype lamin-A protects cells against stress-induced cell death. In fact, knockout of lamin-A caused broad defects in stress resistance. Therefore, lamins impede 3D migration but also promote survival against migration-induced stress (149).

Remodeling of the nuclear structures is associated with mechanical stress transmitted via the ECM/FAPs/cytoskeleton/nuclear envelop protein axis. The mechanical stress transmission axis promotes the epigenetic changes and the modification of chromatin dynamics that influence on the nuclear behavior (150). FAPs, however, can become activated independent of ECM in certain cases e.g., in breast cancer cells (151, 152). During tumor progression, microenvironmental control of nuclear organization seems to be impaired but still dependent on β1 integrin signaling (152).

Of great interest is a finding showing that DNA repair proteins are mechanosensitive factors leading to a new field of mechano-genomics (153). The group of Discher focuses on the spatiotemporal changes of endogenous DNA damage and repair factors in cells migrating through rigid micropores and on the lasting perturbations to the genome. The study showed that multiple DNA repair proteins avoid mechanical stress upon pore migration, resulting in a cytoplasmic mislocalization sustained for many hours, which leads to delayed repair and consequently DNA sequence alteration (154, 155).

In the previous section, we discussed about signaling cascades activated from the integrin axis. These mechanical signals are then transduced to the nucleus though mechanosignaling, in other words biochemical mechanotransduction pathways. In addition to the mechanosignaling, there is a faster way to transmit physical signals directly to the nucleus possibly through the physical anchoring of the cytoskeleton with the nuclear lamina via the linker of nucleoskeleton and cytoskeleton (LINC) complex (17). This complex is composed of two family members, which are SUN (Sad1p, UNC-84) and KASH (Klarsicht/ANC-1/Syne Homology) located at the interior and the exterior of the nuclear membrane, respectively (**Figure 2C**). Typically, SUN is connected with the nuclear intermediate filament lamins, whereas, KASH interacts with cytoskeletal proteins, such as intermediate filaments, actin filaments and microtubules. SUN and KASH proteins interact within the perinuclear space forming a bridge between cytoskeleton and nucleoskeleton (156). Guilluy and colleagues studied the association of LINC complex with mechanical tension. They showed that an isolated nucleus adapts to mechanical tension induced by magnetic tweezers, which results in increased nuclear stiffness. The stiffening of the nucleus did not involve structural modification of chromatin or nuclear actin, but required an intact nuclear lamina and phosphorylation of emerin, a protein of the inner nuclear membrane (157).

In a recent study from the Swift group, the response of cells to cyclic tensile strain mimicking the dynamics of the microenvironment in vivo was investigated (158). A series of strains with different intensities was applied to cells. They observed that cells subjected to low levels of strain responded similar to cells exposed to an increased stiffness. In case cells were exposed to the high intensities, the composition of LINC complex was altered, specifically the SUN domain containing the SUN2 protein. This domain was significantly affected by protein levels and posttranslational modifications leading to a strain induced breakpoint in the linker complex. As a result, cells were able to detach the nucleus from the cytoskeleton in case of excess stress, conferring a protection to DNA (158).

Collectively, nuclear stiffness is associated with tumor aggressiveness, especially in migration and metastasis. However, more studies are required to understand the underlying mechanisms and to validate whether nuclear stiffness can be used as a predictive biomarker of therapy response.

#### CONCLUDING REMARKS AND PERSPECTIVE

We discussed recent studies showing how the tumor creates a microenvironment favorable for proliferation, invasion and treatment resistance. Cellular, nuclear, and ECM stiffness play essential and intertwined roles in the cancer response to therapy. Despite many investigations performed with regard to the impact of stiffness on chemotherapy response, it remains open if these results are similar and can be translated to the response to radiotherapy. We have shown that the presence of a 3D environment and matrix composition affects radiotherapy response upon the activation of FAPs (CAM-RR) (**Figure 1**) for pro-survival signaling. FAPs and extracellular matrices have been defined as important determinants of the hallmarks of cancer (30, 101, 159–161). In our previous studies,

#### REFERENCES


we demonstrated different growth conditions to modulate chromatin structure, DNA repair and cell survival upon radiation exposure (100, 162). Obviously, force transmission and mechanotransduction are mediated by FAPs to enable control over nuclear processes including therapy resistance. Together, the current body of literature strongly supports the concept of mechanical characteristics of the cellular environment to critically regulate the epigenetic and genetic landscape driving cancer cell radiochemoresistance.

Clearly, the matrix stiffness is a main element in cancer therapy resistance, especially in chemotherapy. Radiotherapy typically induces fibrotic reactions that, consequently, amplify tissue stiffening. This causes complications in normal tissues such as lung fibrosis. A combination of multiple factors like fibroblast activation, vascular damage, and leakage, etc., promotes ECM remodeling and excess matrix deposition (163–165). To date, it remains to be understood to what extent these therapyinduced changes contribute to tumor progression, resistance, and metastasis.

The role of stiffness in resistance and the potential of ECM, cellular, and nuclear stiffness as a biomarker for therapy response are still elusive. This ambiguity is also due to the heterogeneous set of data, which may sometimes be conflicting (**Table 2** provides some examples). Therefore, an optimized and standardized approach for the study of stiffness is necessary. Moreover, it would be of great benefit for the community to collaboratively standardize experimental setups and measurement techniques.

Despite the number of research groups studying cell behavior on different substrates with different stiffness, the impact of these matrices on cell function and therapy response has only recently been appreciated. Future efforts may focus on (1) how stiffness sensing occurs in different macro-micro-nanoscales (ECM/tissue, cell, nucleus) and (2) whether stiffness is a promising biomarker for therapy response or even a therapeutic target.

#### AUTHOR CONTRIBUTIONS

SD and NC formulated the topic of the review and drafted and approved the manuscript.

#### FUNDING

This project has received funding from the European Union's Horizon 2020 research and innovation program under the Marie Skłodowska-Curie Grant Agreement No. 642623 (to NC).


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Deville and Cordes. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Biological Determinants of Chemo-Radiotherapy Response in HPV-Negative Head and Neck Cancer: A Multicentric External Validation

Martijn van der Heijden1,2, Paul B. M. Essers 1,3, Monique C. de Jong<sup>3</sup> , Reinout H. de Roest <sup>4</sup> , Sebastian Sanduleanu<sup>5</sup> , Caroline V. M. Verhagen1,2 , Olga Hamming-Vrieze<sup>2</sup> , Frank Hoebers <sup>5</sup> , Philippe Lambin<sup>6</sup> , Harry Bartelink <sup>3</sup> , C. René Leemans <sup>4</sup> , Marcel Verheij 1,3,7, Ruud H. Brakenhoff <sup>4</sup> , Michiel W. M. van den Brekel 2,8 and Conchita Vens 1,3 \*

#### Edited by:

Ira Ida Skvortsova, Innsbruck Medical University, Austria

#### Reviewed by:

Panagiotis Balermpas, University Hospital Zürich, Switzerland Raghvendra Mohan Srivastava, Memorial Sloan Kettering Cancer Center, United States

#### \*Correspondence:

Conchita Vens c.vens@nki.nl

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 21 August 2019 Accepted: 09 December 2019 Published: 10 January 2020

#### Citation:

van der Heijden M, Essers PBM, de Jong MC, de Roest RH, Sanduleanu S, Verhagen CVM, Hamming-Vrieze O, Hoebers F, Lambin P, Bartelink H, Leemans CR, Verheij M, Brakenhoff RH, van den Brekel MWM and Vens C (2020) Biological Determinants of Chemo-Radiotherapy Response in HPV-Negative Head and Neck Cancer: A Multicentric External Validation. Front. Oncol. 9:1470. doi: 10.3389/fonc.2019.01470 <sup>1</sup> Division of Cell Biology, The Netherlands Cancer Institute, Amsterdam, Netherlands, <sup>2</sup> Department of Head and Neck Oncology and Surgery, The Netherlands Cancer Institute, Amsterdam, Netherlands, <sup>3</sup> Department of Radiation Oncology, The Netherlands Cancer Institute, Amsterdam, Netherlands, <sup>4</sup> Amsterdam UMC, Vrije Universiteit Amsterdam, Otolaryngology/Head and Neck Surgery, Cancer Center Amsterdam, Amsterdam, Netherlands, <sup>5</sup> Department of Radiation Oncology (MAASTRO), GROW—School for Oncology and Developmental Biology, Maastricht University Medical Centre, Maastricht, Netherlands, <sup>6</sup> The D-Lab and The M-Lab, Department of Precision Medicine, GROW—School for Oncology and Developmental Biology, Maastricht University, Maastricht, Netherlands, <sup>7</sup> Department of Radiation Oncology, Radboud University Medical Center, Nijmegen, Netherlands, <sup>8</sup> Department of Oral and Maxillofacial Surgery, Amsterdam UMC, Academic Medical Center, Amsterdam, Netherlands

Purpose: Tumor markers that are related to hypoxia, proliferation, DNA damage repair and stem cell-ness, have a prognostic value in advanced stage HNSCC patients when assessed individually. Here we aimed to evaluate and validate this in a multifactorial context and assess interrelation and the combined role of these biological factors in determining chemo-radiotherapy response in HPV-negative advanced HNSCC.

Methods: RNA sequencing data of pre-treatment biopsy material from 197 HPV-negative advanced stage HNSCC patients treated with definitive chemoradiotherapy was analyzed. Biological parameter scores were assigned to patient samples using previously generated and described gene expression signatures. Locoregional control rates were used to assess the role of these biological parameters in radiation response and compared to distant metastasis data. Biological factors were ranked according to their clinical impact using bootstrapping methods and multivariate Cox regression analyses that included clinical variables. Multivariate Cox regression analyses comprising all biological variables were used to define their relative role among all factors when combined.

Results: Only few biomarker scores correlate with each other, underscoring their independence. The different biological factors do not correlate or cluster, except for the two stem cell markers CD44 and SLC3A2 (r = 0.4, p < 0.001) and acute hypoxia prediction scores which correlated with T-cell infiltration score, CD8<sup>+</sup> T cell abundance and proliferation scores (r = 0.52, 0.56, and 0.6, respectively with p < 0.001). Locoregional control association analyses revealed that chronic (Hazard Ratio (HR) = 3.9) and acute hypoxia (HR = 1.9), followed by stem cell-ness (CD44/SLC3A2; HR = 2.2/2.3), were the strongest and most robust determinants of radiation response. Furthermore, multivariable analysis, considering other biological and clinical factors, reveal a significant role for EGFR expression (HR = 2.9, p < 0.05) and T-cell infiltration (CD8+T-cells: HR = 2.2, p < 0.05; CD8+T-cells/Treg: HR = 2.6, p < 0.01) signatures in locoregional control of chemoradiotherapy-treated HNSCC.

Conclusion: Tumor acute and chronic hypoxia, stem cell-ness, and CD8<sup>+</sup> Tcell parameters are relevant and largely independent biological factors that together contribute to locoregional control. The combined analyses illustrate the additive value of multifactorial analyses and support a role for EGFR expression analysis and immune cell markers in addition to previously validated biomarkers. This external validation underscores the relevance of biological factors in determining chemoradiotherapy outcome in HNSCC.

Keywords: HNSCC, chemoradiotherapy, radiation resistance, hypoxia, immune cell infiltration, expression profile analysis, head and neck cancer, radiation oncology

#### INTRODUCTION

In this study we set out to perform multifactorial analyses to gain understanding of the role and dependence of biological factors that have shown to influence tumor radiation response in preclinical studies and to be associated with radiotherapy response in clinical studies (1, 2). Chemo-radiotherapy is the primary treatment option for advanced head and neck squamous cell carcinoma (HNSCC). Cure and locoregional failure rates of around 50 and 25%, respectively, facilitate the evaluation of biological determinants of radiation response. Using biological characteristics of the tumors, outcome association studies revealed many potential determinants of prognosis and treatment response in HNSCC (3–7). This study evaluates complementarity and hierarchy of radiation response determining "HNSCC biology."

Head and neck squamous cell carcinoma (HNSCC) is the 6th most common cancer in the world, with smoking, alcohol and HPV infection as the main risk factors. Around two thirds of the patients present with advanced stage disease and have a poor prognosis with 5 year overall survival rates around 50% (8, 9). Allowing for organ preservation, curative treatment for advanced stage hypopharyngeal, laryngeal and HPV-negative oropharyngeal carcinomas shifted from extensive surgery to concomitant cisplatin-based chemo-radiotherapy in the last decades (10). Around two third of all HNSCC patients receive radiotherapy as part of their treatment. Among these, those with HPV-positive oropharyngeal tumors have a particularly good prognosis, reason to consider them as a new entity in the new TNM staging (11). As revealed by gene expression and mutational analyses, these tumors are also biologically very different (12, 13). HPV-negative HNSCC, in contrast, are characterized by poor prognosis. They exhibit frequent amplifications and mutations in proto-oncogenes (EGFR, MYC, HRAS) and in cell cycle genes that drive and support tumor proliferation (14–16). p53 is affected in almost all HPVnegative HNSCC.

Early radiobiology studies revealed determinants of tumor radiation response. Hypoxia, repopulation, driven by tumor cell proliferation, tumor stem cell density (i.e., clonogenic cell density) and cellular radiosensitivity (as for example determined by cellular DNA damage repair capacity) were shown to be among the most relevant biological factors that affect radiation or fractionated radiotherapy response in preclinical models of different cancers (1, 2). In recent years, increased interest emerged in immune response related markers and immune cells due to novel immunotherapeutic options (17–21). A series of preclinical and clinical studies highlight the potential relevance of immune-related markers in HNSCC [reviewed in (5, 6, 19, 22–24)].

HNSCC outcome association studies using many different biomarkers, demonstrated the clinical importance of some of these pre-clinically assessed tumor biology parameters (1, 5– 7). HPV and hypoxia are indeed the best studied biology related prognostic markers in HNSCC. Within the HPVnegative patients, tumor hypoxia marks patients with a poor prognosis (25–29). Confirming its role above marking poor prognosis patients, hypoxia biomarkers also predict response to hypoxia modification therapy (25, 30–33). Elaborating on a gene expression profile that captures the cellular changes caused by acute hypoxia, we recently showed the relevance of acute hypoxia in addition to chronic hypoxia (29). As predicted by the process they capture, these two classifications did not necessarily overlap in the samples and also reveal different outcome associations in HNSCC that result from a prominent role of acute hypoxia.

While the success of accelerated radiotherapy schedules (34) highlight the important role of tumor repopulation in HNSCC, there is a lack of biomarker data showing a link to cellular proliferation (35). Based on genetic mutation data, we find a small role for co-occurring CCND1 and CDKN2A mutations

in HPV-negative chemo-radiotherapy treated HNSCC that was however not visible in the locoregional control endpoints (36). Yet, the combination of radiotherapy with the epidermal growth factor receptor (EGFR) binding antibody cetuximab has shown efficacy and EGFR expression has been associated with poor survival, preferentially in non-accelerated schedules arguing for a role in tumor repopulation (37–40). The role of EGFR and cellular proliferation in radiotherapy response needs to be further elaborated (41–43). However proliferation, as determined by the proliferation marker by Starmans et al. has been linked to aggressive disease or disease progression in multiple cancer types; unfortunately this was not assessed in HNSCC (44).

Originated from CD44 expression data from de Jong et al. (45) in laryngeal cancer and confirmed in resected and chemoradiotherapy treated HNSCC for CD44 and SLC3A2 (27, 46) in subsequent studies, it also became clear that tumor "stem cell-ness" is important in radiotherapy outcomes since these stem cell related biomarkers were associated with poor prognosis (35, 47–49).

The consistent effect of CD8<sup>+</sup> T cell depletion on radiation induced tumor growth delays in preclinical studies expose the relevance of certain immune cell populations in radiation response and resistance (50, 51). Evidence in clinic of a possible interaction is less strong and current studies focus on strategies to optimize combinations with immune response modulators to improve radiotherapy outcomes (6, 18, 20, 21, 23, 27, 52– 55). Interestingly, Mandal et al. recently showed that markers for regulatory Tcells (Treg), NK cells and CD8<sup>+</sup> T cells are prognostic in head and neck cancer (56) in the TCGA dataset. Despite these interesting initial reports, the prognostic value of these gene expression based immune markers is still unknown for chemo-radiotherapy treated patients since all patients in the TCGA dataset have been treated with primary surgery. Immunohistochemically (IHC) determined high CD8<sup>+</sup> T-cell counts are associated with good prognosis in postoperative chemo-radiotherapy treated patients, further indicating its relevance for HNSCC (57). A good prognosis association with IHC CD8<sup>+</sup> TIL density was found in patients with oropharyngeal squamous cell carcinoma treated with surgery or (chemo) radiotherapy and in a similarly mixed treatment cohort of hypopharyngeal SCC patients (58–60).

Our previous studies emphasized the important role of functional and genetic DNA crosslink repair defects in HNSCC (61, 62) and provided the basis for machine learning generated models that predicted such DNA repair defects in clinical samples (63). The expression based DNA repair defect prediction models revealed an association with metastasis in HNSCC and linked DNA repair defects to migratory and invasive behavior in HNSCC cell lines (63). Given the relevance of Epithelial to Mesenchymal Transition (EMT) in many cancer types, we also developed a HNSCC-specific EMT model that classifies HNSCC according to epithelial or mesenchymal characteristics (64). The strong prognostic value of this HNSCC-EMT model also suggests an important role in radiation response.

Taken together, the individual roles of some of these biological factors important in radiation response have not been validated and the interrelation of these biological factors has not been investigated in the clinical setting. We therefore studied the role of the aforementioned biological factors in the context of head and neck cancer and chemo-radiotherapy. Previously published gene expression based signatures were used to detect these factors. In a set of nearly 200 patients with advanced stage HPV-negative HNSCC treated with chemo-radiotherapy, we used univariate and multivariate outcome analyses to examine these factors while also considering correlation and dependence to delineate their relative roles.

#### MATERIALS AND METHODS

#### Patient Data and Material

This retrospective study included material and data from patients that were diagnosed between 2001 and 2014 and treated with definitive cisplatin-based chemo-radiotherapy within three centers: the Netherlands Cancer Institute (Amsterdam, NL), the Amsterdam University Medical Center (Amsterdam, NL) or the MAASTRO clinic/MUMC+ (Maastricht, NL). Selection criteria for this gene expression study cohort were (i) concomitant radiotherapy and cisplatin treatment of unresected HNSCC, (ii) hypopharyngeal, laryngeal or HPV-negative oropharyngeal (iii) no prior treatment with chemotherapy or radiotherapy in the head and neck area. To minimize the number of variables, AJCC disease staging, summarizing TNM stage, was used to classify HNSCC patients after determining whether this classification also represented N-staging and its known association with survival well (**Supplementary Figure 1**). Received radiotherapy regimens were 70 Gy over 35 fractions (up to 77 Gy in ARTFORCE patients) in 7 or 6 weeks (DAHANCA scheme). All patients were treated with either of four different cisplatin regimens: daily [25 × 6 mg/m<sup>2</sup> Body Surface Area (BSA)], weekly (7 × 40 mg/m<sup>2</sup> BSA) or 3-weekly (3 × 100 mg/m<sup>2</sup> BSA) intravenous administration or weekly intra-arterial administration [4 × 150 mg/m<sup>2</sup> BSA, for 8 patients according to the RADPLAT trial protocol (65)]. Not all patients completed the full chemotherapy scheme. Therefore, cumulative cisplatin doses were calculated and patients were classified into < or ≥ or 200 mg/m<sup>2</sup> BSA cisplatin, according to literature (66, 67). Survival data was calculated from the start of treatment until the first event was detected. The primary outcome measure is loco-regional control (LRC) and implies absence of recurrences in the radiotherapy targeted regions of the head and neck area. Patient characteristics are provided in **Supplementary Table 1**. Institutional Review Boards at the Netherlands Cancer Institute, the Amsterdam University Medical Center and the MAASTRO clinic/MUMC+ approved biopsies and collection of fresh-frozen HNSCC tumor material and the use of genetic and clinical data from patients at their respective centers after anonymization. All patients granted written informed consent for biopsy, material use and data use. Pre-treatment tumor biopsy material available for the DESIGN study or collected from the NKI ARTFORCE (68) or RADPLAT trial patients were used for RNA preparation and sequencing. HPV-status of all oropharyngeal carcinomas was determined by immunohistochemical assessment of p16 by a dedicated head and neck pathologist (69) followed by a HPV

DNA test on the p16-immunopositive cases and/or confirmed using RNA-sequencing data.

#### Material Preparation and RNA-Sequencing

Fresh-frozen tumor samples were sectioned, collected for RNA preparation and in part subjected to tumor percentage evaluation by revision of HE stained coupes by senior head and neck pathologist Dr. S.M. Willems. Only samples with a tumor percentage of >40% proceeded to RNA-sequencing. RNA was isolated using the AllPrep DNA/RNA mini kit (Qiagen). Quality and quantity of total RNA was assessed by the 2100 Bioanalyzer using a Nano chip (Agilent, Santa Clara, CA). Only total RNA samples having RIN>7 were used for library preparation. Strandspecific libraries were generated using the TruSeq Stranded mRNA sample preparation kit (Illumina Inc., San Diego, RS-122- 2101/2) according to the manufacturer's instructions (Illumina, Part # 15031047 Rev. E). The libraries were analyzed on a 2100 Bioanalyzer using a 7500 chip (Agilent, Santa Clara, CA), diluted and pooled equimolar into a 10 nM multiplex sequencing pool and sequenced with 65 base single reads on a HiSeq2500 using V4 chemistry (Illumina Inc., San Diego). Reads were mapped against the GRCh38 human genome using TopHat2.1 (70), with options "fr-firststrand," "transcriptome-index," and "prefilter multi-hits." Read counts were determined using HTSeq-count (71) with options "stranded" and mode "union."

#### Expression and Patient Outcome Analyses

All analyses were performed in R 3.4.3 using Rstudio 1.1. Samples were classified and scored for the different analyzed biological characteristics using different gene expression profiles according to the protocols described in the original publication. If not possible due to the lack of original reference data, GSVA, a Bioconductor package for R, was used on raw read counts to calculate gene expression profile scores (72). Transcripts per million (TPM) was used for individual gene expression analyses. Patient outcome analyses were performed using Cox proportional hazard model. Time to event was defined as the time between the first day of treatment and the day the event was detected. Events in the locoregional control data (LRC) were defined by recurrences in the radiotherapy targeted region. Distant metastasis (DM) events were defined by tumors detected outside the head and neck area. A patient's death prior to a possible event led to censoring in the LRC and DM data and no event was recorded. Progression free survival (PFS) was defined as the time from the start of treatment to the day the patient died, had a locoregional recurrence or distant metastasis. Tests were considered significant when p < 0.05. A spearman correlation coefficient was computed between continuous variables.

In order to obtain a robust cut-off when transforming a continuous variable into a dichotomous variable we used the bootstrap procedure as described in Linge et al. (28). In brief, 197 sample values were randomly assigned into one bootstrap cohort (from the cohort of 197 patients) while data from the same patient could be chosen multiple times. This procedure was repeated to obtain 10.000 randomized cohorts. At each possible cut-off value of the marker of interest, the individual cohorts were split into a "low" and "high" group and Cox proportional hazards models were fit based on these splits. These models included, next to the newly grouped marker of interest, all clinical variables that were found to be significantly associated with the outcome of interest [Locoregional Control (LRC), Distant Metastasis (DM), Overall Survival (OS) or Progression Free Survival (PFS)]. The fraction of cohorts for which the marker of interest was significantly associated with survival (p < 0.05) was recorded for each cutoff. The values of nine adjacent cutoffs were averaged to smoothen the data. The cutoff with the highest fraction of significant associations was chosen for further analysis. Cutoffs that would result in patient subgroups with <10% of the patients were not considered to maintain statistical power. Note that, this analysis was repeated for each endpoint resulting in different cut-offs.

To reduce the number of possible variables included in multivariable analysis we used a backward selection procedure. The most frequent level of each variable was used as the reference level for this analysis. A Cox proportional hazard model was fit containing all biological markers and clinical variables. Then, each individual variable was removed from the model and improvements in model performance by this process were assessed using the Akaike Information Criterion (AIC) from the "stats" package in R. The best model (lowest AIC) was selected for further analysis in the multivariate Cox regression analysis. This process was repeated until removing variables from the model did no longer result in an improved model.

## RESULTS

### Role of Clinical Factors and Patient Characteristics in Chemo-Radiotherapy Outcome

In this retrospective multicenter study, 197 patients met all inclusion criteria and had sufficient tumor material available. All patients were treated with definitive cisplatin-based chemoradiotherapy for advanced stage HPV-negative oropharyngeal, hypopharyngeal, or laryngeal carcinoma. Patient characteristics are shown in **Table 1**. The median age in this patient cohort is 62 years and there is a male: female ratio of 3:1. Most patients reported ongoing or a history of alcohol and/or tobacco use. The largest subsite representation is oropharyngeal tumors with 85 patients, then hypopharyngeal with 78 and laryngeal carcinoma with 34 patients. Except for two patients, all patients had stage III/IV classified tumors. As expected, outcomes and survival curves differ according to stage (**Supplementary Figure 2**). Tumor volume data as determined by delineation on RT planning CT images were available for 166 patients with a median volume of 23.2 cm<sup>3</sup> . Not all patients finished chemotherapy, but 126 patients (63%) received a cumulative dose of and above 200 mg/m<sup>2</sup> body surface area. Locoregional recurrences occurred in 23.8% (N = 49) of cases and distant metastasis in 19.8% (N = 39).

Clinical factors were tested for their association with locoregional control and other survival outcomes (**Supplementary Figure 1**, **Supplementary Table 1**). Consistent with previous reports we find that locoregional control (LRC) is influenced by cumulative cisplatin dose levels (66, 67). The



cumulative cisplatin dose of < 200 mg/m<sup>2</sup> BSA was significantly associated with LRC failure (HR = 2.57, p = 0.0012). Female sex shows a trend toward better locoregional control (HR = 0.52, p = 0.072). This could however been confounded by the less prominent alcohol consumption characteristics or other differences in lifestyle in this particular patient group. More female patients reported to abstain from alcohol compared to male patients (21.8 vs. 7.4%, p = 0.019), which was however not the case for tobacco use (p = 0.66). Heavy past or ongoing alcohol consumption was associated with an increased risk for LRC failure (HR = 2.16, p = 0.041). Interestingly, age, tobacco, tumor subsite and AJCC stage is not significantly associated with LRC in our patient cohort. The other clinical outcomes (DM, PFS or OS) showed significant associations with sex, tumor volume, stage and cisplatin (**Supplementary Figure 1**, **Supplementary Table 1**).

#### Tumor Biology Assessment and (in)Dependence in HPV-Negative HNSCC

Preclinical radiobiology studies and clinical biomarker studies exposed many different determinants of radiation response. The number of variables that can be included in statistical analyses are however limited by the cohort size and number of events. Thus, in order to evaluate the relative role of different tumor biology parameters in clinic, we prioritized those with a reported clinical outcome association. The following 12 gene expression signatures were therefore selected to characterize the clinical samples using pretreatment HNSCC transcriptomic data: The Toustrup (1) and Seigneuric (2) expression signatures were used to assess the level of chronic (1) and acute (2) hypoxia, respectively (25, 73). Linked to tumor stem cell richness SLC3A2 (3) and CD44 (4) gene expression values (in TPM) were included since both have been reported to be associated with outcome in chemo-radiotherapy treated patients (74). Economopoulou et al. (5) EGFR expression (in TPM) and (6) the Starmans et al. "proliferation" expression signature (44) were selected as cellular proliferation markers which could influence tumor repopulation between radiotherapy fractions. To cover immunerelated factors we further included expression signatures from Senbabaoglu et al. (75) that originated from Bindea et al. (76) and assess (7) 'T-cell infiltration score' (TIS), (8) CD8<sup>+</sup> Tcells, (9) CD56dim natural killer (NK) cells abundance while considering the (10) CD8<sup>+</sup> vs. T regulatory (Treg) cell ratios. This immune status gene expression signature selection is based on the reported outcome association in resected HNSCC (56). Our own studies conducted in HPV-negative advanced HNSCC revealed an important role for EMT and DNA crosslink (CL) repair defects in treatment outcome and these prediction models for mesenchymal characteristics and tumor cell DNA crosslink repair defects (11). "HNSCC-EMT" and (12) "MMConly," were therefore included in this analysis and are referred to as "EMT" and "DNA CL repair" in this manuscript (63, 64). Most of these biological factors have been tested individually, predominantly in univariable analyses and in different settings in previous studies; however their mutual correlations and possible dependence between them are unknown.

The goal of this study is to pinpoint biological factors that are important for (chemo)-radiotherapy treatment failure and thus might validate their independent role in radioresistance. We therefore calculated scores for all aforementioned markers. **Supplementary Figure 3** shows the frequency of the scores and their distribution over the patient cohort. Next, we performed hierarchical clustering to investigate the presence of HNSCC subsets as defined by these characteristics. Tumor volume was included in this analysis as it promotes chronic hypoxia or may be associated with high proliferation scores. Despite the coexistence and correlation of some factors this does however not reveal any prominent clusters (**Figures 1A,B**). Surprisingly, we find that the acute hypoxia profile score correlates with the Starmans proliferation score (r = 0.58, p < 0.001) (**Figure 1C**) but also with the T-cell Infiltration Score (TIS) and the CD8<sup>+</sup> T cell scores (r = 0.51 and r = 0.54, p < 0.001).

Within their own category, stem cell related markers, CD44 and SLC3A2 (r = 0.57, p < 0.001), and the immune cell related markers correlate with each other. While the correlation of acute and chronic hypoxia is significant, it is fairly weak (R = 0.26, p < 0.001) and was in line with previous reports (29). EGFR expression and the proliferation score are correlated to some extent (r = 0.26, p < 0.01). The CD8<sup>+</sup> T cell to regulatory T cell ratio (CD8+/Treg) as determined by the expression signature scores is negatively associated with the abundance of CD56dim

natural killer (NK) cells and the TIS signature. While the link between tumor volume and chronic hypoxia (R = 0.23, p < 0.01) is expected, tumor volume is also associated with EMT (R = 0.23, p < 0.05). With a maximum variance inflation factor value of 3.3, correlations were not strong enough to exclude parameters from subsequent analyses. None of these markers show strong associations with any of the clinical factors. Among all, we find that the most independent tumor characteristics are the presence of DNA CL repair defects and tumor EMT status (and tumor volumes).

#### Role of Individual Biological Factors in Locoregional Control by Chemo-Radiotherapy in HNSCC

Since we aimed to evaluate tumor characteristics with respect to radiation resistance and response, we initially focused on locoregional control outcome values that are mainly determined by the success of the "local" radiotherapy treatment. Given the lack of strong correlations, all markers were individually tested for their association with locoregional failure. A 10.000 times bootstrapping method was employed to (a) determine a potential role for the biomarker across different cutoffs and (b) to identify a clinically robust cut-off for each so to compare the biomarkers among each other. In brief, each marker was tested for their association with the selected survival outcome for all possible cutoffs. This analysis was performed using a multivariable Cox proportional hazard model with all relevant clinical factors included, as determined above. Consequently, clinical variables were included according to outcome type: sex and cumulative cisplatin dose for LRC; sex, subsite and cumulative cisplatin dose for OS; stage, subsite and cisplatin dose for PFS; and sex and alcohol use for DM. Based on the results of these 10.000 bootstrap repeats (**Figure 2A**), we find that the hypoxia and stem cell related markers are most robustly associated with LRC across different score cut-offs. Proliferation, EGFR and immune cell signatures merely provide significant associations with LRC in a fraction of the randomly created cohorts and tested cut-offs.

Cut-offs with the most stable clinical association were selected for each biomarker for further analysis as depicted in **Figure 2A** and listed in **Supplementary Table 2**. These analyses confirm that both, chronic and acute hypoxia, are strongly associated with locoregional control. Using these calculated cutoffs in multivariable analyses with clinical factors, we find that among all chronic hypoxia is most strongly associated with a failure of locoregional control (HR = 3.95, p = 0.0038) followed by acute hypoxia (HR = 1.9, p = 0.03) and stem cell related, SLC3A2 (HR = 2.31, p = 0.026) and CD44 (HR = 2.03, p = 0.043; **Figures 2B**, **3**; **Supplementary Table 3**). Although not significantly, larger tumor volumes showed a trend toward worse locoregional control with a hazard ratio (HR = 1.63, p = 0.11) that is comparable to those previously reported by others (27). It should be however noted that most tumors in this cohort are relatively large and stage III/IV. This and the fact that the LRC measure also includes regional recurrences, may together affect the specific HR values. Trends toward a worse LRC prognosis were observed in patient groups with tumors with high proliferation and CD8+ T cell scores (HR = 1.89, p = 0.067 and HR = 2.35, p = 0.071, respectively) (**Supplementary Figure 4** and **Supplementary Table 3**).

To delineate this from general poor prognosis pattern and to investigate the radiation response link further, we repeated these analyses and compared the role of these biomarkers for overall survival, progression free survival and distant metastasis (**Figure 3**). Cut-off values were defined by the bootstrapping method described above for each biomarker; and multivariable Cox proportional hazard analyses with clinical factors were performed. Notably, gene expression signature scores or expression value cut-offs, as determined by their potential relevance in the 10,000x bootstrapping method, resulted to be different in some of the biomarkers such as "acute hypoxia," "chronic hypoxia," "EGFR," "TIS," "NK CD56dim." and "CD8+/Treg" (**Supplementary Figures 4**–**6** and **Supplementary Table 2**).

Most markers show an association with several of the outcome parameters. We find that distant metastasis is associated with EMT (HR = 3.14, p = 0.0086), acute hypoxia score (HR = 2.44, p = 0.0086), NK CD56dim score (HR = 2.47, p = 0.019) and EGFR expression (HR = 2.07, p = 0.032) (**Supplementary Figure 5**, **Supplementary Table 3**). Poor overall survival is associated with increased tumor volume (HR = 2.12, p = 0.00054), EMT score (HR = 2.15, p = 0.002), DNA CL repair defect (HR = 1.97, p = 0.0043), acute hypoxia (HR = 1.62, p = 0.023) and EGFR expression (HR = 1.68, p = 0.014) (**Supplementary Figure 6**, **Supplementary Table 3**). High EMT (HR = 2.19, p = 0.0014), EGFR (HR = 1.82, p = 0.0038), acute hypoxia (HR = 1.64, p = 0.017), DNA CL repair defect (HR = 1.8, p = 0.0085) and chronic hypoxia scores (HR = 1.7, p = 0.015) and tumor volumes (HR = 1.72, p = 0.036) are associated with a worse progression free survival (PFS) (**Supplementary Figure 7**; **Supplementary Table 3**). Interestingly, when comparing the distant metastasis and locoregional control failure data, locoregional control is increased in tumors with higher EGFR expression or containing few CD56dim NK cells while high values in both result in an increased risk of DM. It should be noted, however, that the bootstrapping defined cut-offs were different in both. Yet, as evident from the bootstrapping data chronic hypoxia was not linked to DM but LRC at many cut-offs. On the contrary, the similar shape of the results from the acute hypoxia bootstrapping supports its relevance in both, LRC and DM (**Figure 2** and **Supplementary Figure 5**).

Taken together, for biomarkers which have been previously reported to be prognostic in HNSCC these analyses validate their role in an independent data set. Most biological factors as determined by the selected biomarkers are significantly associated with PFS thereby confirming their relevance. Overall, we find a prominent role for acute and chronic hypoxia and CD44 and SCL3A2 in our cohorts. We show that, from all, chronic hypoxia appears to be the most specific to LRC. In contrast, HR values from EMT and proliferation based splits are greater when assessing DM. Furthermore, these data reveal a role for

marker, the cohort was split into a high and low group at the best cutoff determined in A. Hazard ratios for recurrences and corresponding p-values were obtained with a multivariate Cox proportional hazard analysis using the same variables as those used to determine the cutoff.

the immune cell and proliferation related biomarkers in HNSCC outcome after definitive chemo-radiotherapy.

#### The Relative Role of Biological Factors in Chemo-Radiotherapy Outcomes in HNSCC

The biological markers have been tested independently of each other and most are significantly associated with patient outcome thereby supporting their role in HNSCC and treatment response. However, tumor biology and determinants of radioresistance are multifactorial and may depend on the context and relation to each other. We therefore aimed to identify the most relevant markers in a multivariable analysis. To this end we used a backward selection method. This method creates a Cox proportional hazard model using all available factors. It then iteratively eliminates the least relevant factor until no further decrease in AIC, a measure of model performance, is possible. From these analyses (**Supplementary Table 4**), we conclude that chronic hypoxia, EGFR expression, CD8+/Treg, T-cell infiltration, and CD44 are the most relevant biological factors that are associated with locoregional control. Multivariable analyses (**Figure 4**) also demonstrate that they are independent from relevant clinical factors such as cumulative cisplatin dose or sex. Cisplatin dose, age and sex are the clinical factors most associated with locoregional control in this cohort (**Figure 4**

and **Supplementary Table 4**). Broadly consistent with the results from the multivariate analyses that were performed on the biomarkers on an individual basis, EGFR and immune cell related factors remain important in instituting an increased risk for distant metastasis, while chronic hypoxia and CD44 are less relevant. Instead, tumor EMT and proliferation affects progression free survival most profoundly and independent of other important factors such as tumor volume or cisplatin dose (**Supplementary Figure 8** and **Supplementary Table 4**). The consistent worse prognosis and distant metastasis association of patients with tumors that score high in the CD8<sup>+</sup> T cells and related gene expression signatures is remarkable. High CD8<sup>+</sup> T cell scores, as determined by these signatures or by immunohistochemistry, have been reported to be linked to good prognosis in other heterogeneous HNSCC cohorts (6, 56, 77) and prompted us to analyze this further (**Supplementary Figures 9**, **10**). These analyses suggest that the lack of a good prognosis association could be based on the absence of HPV-positive HNSCC which show overall higher CD8 expression and CD8<sup>+</sup> T-cell signature scores in the TCGA cohort (**Supplementary Figure 9A**). Within the HPV-positive group, high CD8 expression is strongly associated with good prognosis (**Supplementary Figure 9B**). Notably, CD8A/B expression and CD8<sup>+</sup> T-cell signature values do not correlate well (**Supplementary Figure 10**). Interestingly, the observed outcome associations in HPV-negative HNSCC appear to be dependent on cumulative cisplatin dose (**Supplementary Figure 10**).

The obvious divergence in the biomarker associations with local treatment outcomes compared to DM development risks prompted us to investigate this further. Unable to classify patients according to "true" biological parameter classes, we relied on bootstrapping methods to provide cut-offs for each outcome endpoints. As described above those resulted to be largely different in some cases such as for EGFR expression and pointed to a different influence in the respective biological mechanisms. We therefore compared the biological markers with respect to their influence in locoregional control or DM risk in a less cutoff-dependent manner by computing the AUC of the hazard ratio plots from multivariable regression analysis with clinical variables (**Supplementary Figure 11**). **Figure 5** shows an overview of the impact of the individual biological parameters on locoregional control or distant metastasis risk. This analysis highlights the difference in the collection of the most relevant survival determinants for each outcome endpoint. Notably, locoregional control is mainly determined by chronic hypoxia, but also acute hypoxia. CD44 expression and CD8<sup>+</sup> T-cell/Treg ratio are more relevant to LRC than DM, whereas distant

metastasis is predominantly influenced by EMT, acute hypoxia, proliferation and EGFR status.

was repeated with the best performing models until the removal of variables did no longer improve the models. Hazard ratios for locoregional recurrences (A) and distant metastasis (B) as determined by the final model are shown.

#### DISCUSSION

Here, we aimed to evaluate the relevance and interrelation of biological factors known to influence radiation response as determined in preclinical studies. Limited by the size of the study cohort, we restricted the study to markers for which discriminative power has been reported in clinical data in HNSCC and added clinical or biological factors that have shown an important association with radiotherapy outcomes. Using RNA-sequencing data from a large and relative uniform cohort of 197 HPV-negative advanced stage HNSCC patients that were all treated with cisplatin-based chemo-radiotherapy, we find an important role of immune cell (T cell) markers in locoregional control which suggests a role in radiation response. We also show that chronic and acute hypoxia are robustly associated with locoregional control. Similarly, we validated the equally important role of CD44 and SCL3A2, in part related to stem cells,

in our study cohort. When assessed in combination, hypoxia, immune cells, EGFR are the most discriminating independent factors in LRC. For DM, those also include EMT. Overall, considered in the context of clinical factors and each other, our study underscores the relevance of many of these biological factors in HNSCC chemo-radiotherapy outcomes.

To advance previous findings on determinants of chemoradiotherapy outcomes and to prioritize prognostic markers for multi-parametric prediction models, we focused on (i) the validation of expression-based prognostic markers in the chemo-radiotherapy setting and (ii) the evaluation of their complementarity and (iii) the assessment of any dependence to important clinical or other biologic factors. Many HNSCC studies highlight the role of biology for outcome (3–6, 78). A major drawback however for many of such studies is the heterogeneity of the HNSCC patient cohorts or a lack of contextual analyses (78). If focused on tumor site they often encompass many different treatments or if focused on treatment they combine different tumor sites, HPV-negative and positive. Large multicentric studies are therefore valuable contributions to the field (12, 13, 15, 16, 57) that provide insights to the biology of HNSCC and its link to patient outcome (5, 6, 27, 28, 79, 80). Clinical factors are important (81, 82) but are often not considered in multivariable analyses (78, 83). A lack of tumor volume data for example, even though clearly linked to LRC (81, 84–86), impedes the assessment of a role for or a bias from tumor volumes in such analyses. To minimize such treatment or tumor site related bias due to possible interactions; we deliberately excluded the biologically distinct oral cavity and HPV-positive oropharyngeal HNSCC (14, 16, 87–90) in our study. Our cohort also solely comprises definitive chemo-radiotherapytreated advanced HNSCC. In contrast to predetermined gene sets in nanostring technologies, the availability of full transcriptomic data by RNA-Seq allowed us to test selected gene expressions and signatures related to the biological processes that we queried. Together we were able to show that most of the selected markers or marker categories are not related, are independently linked to outcome and that outcome associations are not based on links with known important clinical factors. Overall, we observed little or no influence or interactions with clinical factors, with the notable exception of tumor volume and cumulative cisplatin dose, factors often not accounted for in other biomarker studies. While correlations within the immune markers were expected, here we reveal an association with acute hypoxia scores which in turn appears to be linked to proliferation. Such relations or complementarities can alter the prognostic value or impede a discrimination of the true source of the observed outcome relevance. It however highlights the importance to study such markers in the context of each other and within the same cohort. Overall, our study pinpoints expression markers that should be considered as valuable contributors of future multiparametric prediction models that combine clinical, radiologic, pathological and genetic variables for improved prognosis in advanced HPV-negative HNSCC (91, 92).It is difficult to discern factors that determine tumor radioresistance (83). A comparison of similar patient cohorts treated without or with different doses of radiotherapy would be required to strengthen such a link. Since cisplatin-based chemo-radiotherapy has become a standard treatment for HNSCC to improve quality of life by achieving organ preservation, surgically resected HNSCC patients with similar clinical tumor characteristics are rare, impeding such comparisons. However, in the absence of a comparable but non-radiotherapy treated study cohort, differences in LRC (mostly achieved by radiotherapy) as defined by the biomarker classification, can suggest a role in radiation response. In our study, we assured that important clinical factors that impact patient outcomes have been considered to limit bias or dependence. DM events may have occurred prior to LRC events and could have masked a greater impact in radiation response, such as in the case of acute hypoxia that also shows a strong association with DM. The comparison of LRC with DM further helped to discern a more radiation response specific role from a role in metastasis. Our data here and those reported by us and others do indeed confirm the role of hypoxia in determining radiation response as reflected by the LRC rates (25, 27, 29). In addition, hypoxia has been also implicated in tumor cell invasiveness, facilitating dissemination, and has been therefore associated with metastasis formation, a role that is also evident from our DM analyses. Similarly, we have recently shown that HNSCC cell lines with DNA crosslink repair defects are more migratory and invasive (63), a feature that may explain the association with DM prognosis but could also result in a greater regional spread and failure of locoregional control.

After the initial EGFR studies in clinic and the success of cetuximab combinations (40, 93, 94), cetuximab in HNSCC and the role of EGFR amplification and expression have been disputed since then (95). Most of these studies focused on the very high expressing or used a median cut-off to detect an association with the clinical endpoints analyzed. Here we see a clear role for EGFR in the outcome data when also considering hypoxia and other factors in multivariable analyses. Average to high EGFR expression, is linked to improved LRC when analyzed individually. The association of a low EGFR expressing group with poor LRC however becomes much clearer in the combined multivariable analyses that integrated all relevant biomarkers. It epitomizes the importance of combined analysis, as the prevalence of other, also clinical, factors in the different EGFR expression classified groups may have shifted or masked a possible influence in other studies if not accounted for, as revealed here. Given its role in promoting cell cycle progression, it is conceivable that increased EGFR levels mediate an increase in tumor repopulation between fractions; a radiotherapy response determining process that is counteracted by radiotherapy treatment acceleration or concurrent chemotherapy. This process is therefore limited in our patient population in contrast to some earlier studies that analyzed the influence of EGFR (96). Notably, the association with improved LRC is still discernible (HR = 0.57, p = 0.067) when reanalyzing the data using the higher EGFR expression cut-off that was used for the DM data. DM HR values however drop to 0.57 (p = 0.2) showing a DM link only in the top 25% EGFR expression group (HR = 3.19, p = 0.0056). This more aggressive nature of highly EGFR expressing tumors is consistent with other reports in HNSCC and other cancer types (97).

Our study is limited by statistical constraints due to the cohort size. This enforced us to limit the biological variables and apply selection processes such as the bootstrapping analyses. Yet, it becomes evident that the prognostic value of many of the factors could be validated in our cohort and withstood multivariable analyses with the important clinical variables. Among the clinical variables, we observe a trend toward poor outcomes in current smokers, however this does not reach significance in our cohort. Low numbers in the former smoker category but also the lack of more accurate smoking status values may have decreased the power to reveal the reported association with smoking (79, 98, 99). Since we focused on known determinants of radiation response, other biomarkers were not included despite their relevance or prognostic value in HNSCC (5, 6, 16, 100–106). Some, such as tumor mutational burden (TMB) are prevalent in laryngeal and HPV-negative pharyngeal HNSCC (14) but require DNA sequencing data. TMB was found to be associated with poor prognosis in HPV-negative chemo-radiotherapy treated patients in our previous study (36) and more strongly so in a cohort of patients that also included oral cavity cancers and HPV-positive oropharyngeal (107). Interestingly, low immune cell infiltration or CD8+ T cell values, as assessed by gene expression, have been assigned to HNSCC high in TMB or mutational signatures related to smoking (56, 107).

Other limitations result from technical challenges. Here we detect different biological processes and factors in clinical samples by using published and validated expression signatures—that are linked to these processes. These gene expression signatures may not be perfect identification tools for the specific biology in question (83); however they often reflect the abundance of certain biological elements well (108, 109). The DNA CL repair defect prediction model has for example been generated using functional endpoints and then validated in independent cell line panels or by genetic modification. On the other hand, markers such as CD44 are less clear defined. CD44 expression is associated with stem cell-ness in tumor cells (110), but it is also expressed under hypoxic conditions or in epithelial cells and is a marker for effector memory T cells (45). Therefore, it is particularly interesting to observe the correlation with SLC3A2 another stem cell related marker in our samples which confirms its link to tumor stem cell abundance. Notably, we find a correlation between acute hypoxia and TIS or CD8<sup>+</sup> T-cell scores, suggesting a higher T-cell content in acute hypoxic areas or tumors which could be proposed to be driven by hypoxia induced inflammatory cytokine release (111). This T-cell/acute hypoxia correlation may in part also be responsible for the consistent poor outcome association of the CD8<sup>+</sup> T-cell gene expression signatures. Reiterating the role of technical limitations, it should be noted that these gene expression signatures were based on transcriptional profiles of purified immune cell subsets. Through multiple adaptation steps, they evolved to markers that allowed further discrimination in the context of colorectal carcinoma and HNSCC (56, 75, 76). In terms of identification accuracy there are potential challenges with such technical approaches that can also explain discrepancies with immunohistochemistry determined factors. It is evident that the tumor context affects gene expression of the immune cells and, on the other hand, tumor gene expression features, if present in these signatures, can compound the identification. For instance, the CD8<sup>+</sup> T cell signature includes ZEB1 expression, a protein involved in EMT and a poor prognostic factor in HNSCC (56, 76, 112–115). We therefore assessed CD8A and B gene expression in our samples as a simple surrogate for CD8<sup>+</sup> T cells and show its limited complementarity with the CD8<sup>+</sup> T cell signature score and associations with outcome. The better LRC outcome of patients with CD8 positive tumors in the low cumulative cisplatin patient category is in line with previous report based on IHC (116) The lack of an association with outcome in patients that received high cisplatin doses however demonstrates treatment dependence and explains the discrepancy to other studies (6, 77) when considering this clinical variable in our cisplatin treated cohort. Despite a significant but weak correlation with CD8A expression, high CD8<sup>+</sup> T cell signature values are associated with poor outcomes, demonstrating the influence of the other features in this discriminating signature. Immune cell identification by gene expression may not be flawless. Yet, together, our data indicate a prognosis association that is linked to this particular patient treatment. One could speculate that hematologic toxicities associated with cisplatin administration could contribute to this pattern by abolishing the benefit from an immune cell rich tumor status in these individuals. On the other hand, recent studies suggest an enhancement of antitumor immunity by cisplatin that could also diminish the impact of the pre-treatment immune status (117, 118). While the primary emphasis for prognostic biomarkers lays in the discriminatory power to predict patient outcome, the focus of biomarkers for targeting opportunities is the achievement of an accurate representation of the marked biological process or elements. The signatures used here were selected based on their reported association with both immune cell infiltration and prognosis in HNSCC (56, 76). Yet the question remains whether they reflect CD8<sup>+</sup> T cell infiltration well.

Interestingly, we did find a seemingly independent and consistent role for CD8+, non-regulatory, T cells in our study cohort. Observed for resected HNSCC in overall survival outcome data before, here we show an association with both, LRC and DM, in chemo-radiotherapy treated HNSCC patients indicating those with a high abundance of such T cells to have a worse prognosis. To our knowledge this poor prognosis association with radiation response has not been reported previously (6, 77, 119). As detailed above, this discrepancy with other studies is only in part explained by the used technology (116, 120) (IHC CD8 expression vs. gene expression signatures) since the signatures showed a good prognosis association in the Mandal et al. (56). Careful inspection of the TCGA data revealed increased CD8<sup>+</sup> T-cell gene expression signature scores in the HPV-positive oropharyngeal that drive the good prognosis association. A pattern observed in other studies as well (56, 77, 121–124). Mandal et al. adjusted for HPVassociated outcome differences, which does not account for a possible interaction between the two variables (56). The CD8A and B expression HR plots in our analyses however suggest a stronger effect in the HPV-positive subgroup. Despite obvious evaluation challenges when using the different techniques and associated cut-offs, a similar argument applies to other studies based on immunohistochemistry determined CD8<sup>+</sup> T cell infiltration values. A significant HPV status association got lost in multivariable analyses that indicated a good prognosis association of CD8<sup>+</sup> T cells in oropharyngeal squamous cell carcinoma patients (59). Yet, some studies also show a good prognosis association with TIS or CD8<sup>+</sup> T cells in HPV-negative patients using other scorings, cut offs and expression signatures (125). Since the effect size can be small, patient treatment associations with survival are often not significant in small studies. Treatment could however alter prognosis in subsets of patients. For instance, patients with tumors with DNA crosslink repair defects benefit most from a high cumulative cisplatin dose (63). Similarly, possible immune cell infiltration links could depend on treatment. Despite worse PFS in cases that lack or show minimal CD8A or CD8B expression, we cannot observe the previously reported poor prognosis link in CD8<sup>+</sup> T cell signature low patients in our cohort. No associations between TIS or CD8<sup>+</sup> scores and clinical variables were found; and outcome association links derived from the correlation with acute hypoxia should have been accounted for by the multivariable analyses. Together, our data suggest a role for HNSCC treatment, in particular cisplatin, in immune cell infiltration determined outcomes. Early cancer immunotherapy trials in HNSCC with immune checkpoint inhibitors demonstrate a benefit and underscore the potential value of immune response and chemo-radiotherapy relevant biomarkers to identify patients that will benefit from such treatments (24, 126–134). Larger comparative studies are

therefore needed to disentangle the role of CD8<sup>+</sup> T cells in the individual genetic HNSCC context and the important clinical variables connected to its role in patient outcome (78, 83, 135).

Our patient cohort is fairly unique in that it consists of definitive chemoradiotherapy treated advanced HNSCC patients. Based exclusively on resected HNSCC, these cases are unfortunately not present in the TCGA data. Supported by the detailed clinical data and follow up, this allowed us to elaborate on the role of biological determinants of chemoradiotherapy response. A quarter of the patients suffered from loco regional recurrences after treatment; a treatment success rate that further stresses the relevance of the biological factors found to determine treatment failure. This study does not provide or test clinically applicable prognostic markers. It was designed to compare the individual factors in relation to each other to assess and understand their influence in HNSCC outcome. Optimal cutoffs identified by the bootstrapping method and illustrated in the hazard ratio plots require validation for further development into true prognostic markers. Based on our results, future studies should focus on the elaboration of prognostic models that incorporate these biological markers together with important clinical factors. The multivariable outcome association results and the lack of correlations suggest that these future models should include all biological factors. Discrepancy in the optimal cutoff values further points to the value of non-dichotomized variables in such efforts and also reveals a possible cause of incongruent outcome associations in previous studies. The value of the clinical factors is exemplified by the fact that some biological markers (i.e., DNA CL repair or CD8<sup>+</sup> T cells) lose their strength in patients groups with a high cumulative cisplatin dose.

While tumor stem cell targeting agents are still under development, some of the other biological factors are targetable. Next to high-dose alkylating agents, PARP inhibitors may help to exploit DNA CL repair defects (62) and different immunotherapy options are currently being tested in the HNSCC setting (136). The value of such biological markers in personalized treatments remains to be determined; however our study demonstrates that those patients are in need of improved therapy options.

In conclusion, this multicentric external validation study confirms the important and independent role of biological factors that embody hypoxia, stem cell-ness, tumor growth, EMT and DNA repair for locoregional control in chemoradiotherapy treated patients. The multifactorial analyses results highlight the need to consider these biomarkers in the context of each other and also revealed an important role for immune cell abundance in HNSCC treatment outcome.

### REFERENCES

1. Baumann M, Krause M, Overgaard J, Debus J, Bentzen SM, Daartz J, et al. Radiation oncology in the era of precision medicine. Nat Rev Cancer. (2016) 16:234–49. doi: 10.1038/nrc.2016.18

### DATA AVAILABILITY STATEMENT

The data in this study is available the European Genomephenome Archive (https://www.ebi.ac.uk/ega/) with the accession numbers: EGAD00001005721, EGAD00001005715, EGAD00001005716, EGAD00001005717, and study number EGAS00001004090.

### ETHICS STATEMENT

The studies involving human participants were reviewed and approved by Institutional Review Board The Netherlands Cancer Institute, Amsterdam, The Netherlands; Institutional Review Board VuMC, Amsterdam, The Netherlands; Institutional Review Board Maastro, Maastricht, The Netherlands. The patients/participants provided their written informed consent to participate in this study.

### AUTHOR CONTRIBUTIONS

CVen, MH, PE, and MJ contributed to the conception and design of the study. MH, RR, CVer, OH-V, FH, PL, HB, CL, MV, RB, and MB contributed to the acquisition of data. MH, RR, and FH established and/or curated databases. PE, MH, and SS performed data analysis. MH, CVen, PE, and MJ contributed to data interpretation. CVen, MH, and PE wrote the first draft of the manuscript. All authors contributed to manuscript revisions and approved the submitted version.

### FUNDING

Authors acknowledge financial support from the Dutch Cancer Society (KWF-A6C7072, acronym DESIGN), from the EU 7th framework program (no 257144, acronym ARTFORCE) and the ERC advanced grant (no 694812, acronym Hypoximmuno).

### ACKNOWLEDGMENTS

The authors are grateful for financial support from Brunel and Verwelius and would like to thank the NKI Genomics Core Facility for performing RNA sequencing, the NKI RHPC facility for providing computational resources and the Core Facility-Molecular Pathology and Biobank (CFMPB) for collecting and preparing tissue samples.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.01470/full#supplementary-material


benefit from accelerated radiation therapy in a randomized controlled trial. J Clin Oncol. (2005) 23:5560–7. doi: 10.1200/JCO.2005.06.411


after postoperative radiochemotherapy of locally advanced oropharyngeal carcinoma: results from a multicentre explorative study of the German Cancer Consortium Radiation Oncology Group. Radiother Oncol. (2014) 113:317–23. doi: 10.1016/j.radonc.2014.11.011


trials. Clin Transl Radiat Oncol. (2019) 15:62–9. doi: 10.1016/j.ctro.2019. 01.005


interference with outcome after concurrent chemoradiation in patients with oropharyngeal carcinomas. Oncoimmunology. (2019) 8:1614858. doi: 10.1080/2162402X.2019.1614858


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 van der Heijden, Essers, de Jong, de Roest, Sanduleanu, Verhagen, Hamming-Vrieze, Hoebers, Lambin, Bartelink, Leemans, Verheij, Brakenhoff, van den Brekel and Vens. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Ultra-High Dose Rate (FLASH) Radiotherapy: Silver Bullet or Fool's Gold?

Joseph D. Wilson1†, Ester M. Hammond1†, Geoff S. Higgins 1† and Kristoffer Petersson1,2 \* †

<sup>1</sup> Department of Oncology, The Oxford Institute for Radiation Oncology, University of Oxford, Oxford, United Kingdom, <sup>2</sup> Radiation Physics, Department of Haematology, Oncology and Radiation Physics, Skåne University Hospital, Lund, Sweden

#### Edited by:

Ira Ida Skvortsova, Innsbruck Medical University, Austria

#### Reviewed by:

Virginie Monceau, Collège de France, France Anne-Sophie Wozny, Université Claude Bernard Lyon 1, France Marie-Catherine Vozenin, Lausanne University Hospital (CHUV), Switzerland

#### \*Correspondence:

Kristoffer Petersson kristoffer.petersson@oncology.ox.ac.uk

#### †ORCID:

Joseph D. Wilson orcid.org/0000-0001-8878-5882 Ester M. Hammond orcid.org/0000-0002-2335-3146 Geoff S. Higgins orcid.org/0000-0003-3072-909X Kristoffer Petersson orcid.org/0000-0003-0300-5790

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 06 September 2019 Accepted: 24 December 2019 Published: 17 January 2020

#### Citation:

Wilson JD, Hammond EM, Higgins GS and Petersson K (2020) Ultra-High Dose Rate (FLASH) Radiotherapy: Silver Bullet or Fool's Gold? Front. Oncol. 9:1563. doi: 10.3389/fonc.2019.01563 Radiotherapy is a cornerstone of both curative and palliative cancer care. However, radiotherapy is severely limited by radiation-induced toxicities. If these toxicities could be reduced, a greater dose of radiation could be given therefore facilitating a better tumor response. Initial pre-clinical studies have shown that irradiation at dose rates far exceeding those currently used in clinical contexts reduce radiation-induced toxicities whilst maintaining an equivalent tumor response. This is known as the FLASH effect. To date, a single patient has been subjected to FLASH radiotherapy for the treatment of subcutaneous T-cell lymphoma resulting in complete response and minimal toxicities. The mechanism responsible for reduced tissue toxicity following FLASH radiotherapy is yet to be elucidated, but the most prominent hypothesis so far proposed is that acute oxygen depletion occurs within the irradiated tissue. This review examines the tissue response to FLASH radiotherapy, critically evaluates the evidence supporting hypotheses surrounding the biological basis of the FLASH effect, and considers the potential for FLASH radiotherapy to be translated into clinical contexts.

Keywords: FLASH, radiotherapy, hypoxia, normal tissue, immune

## INTRODUCTION

In the UK, almost 30% of diagnosed tumors are treated with radiotherapy (RT) (1). External beam RT is a non-invasive procedure whereby tumors are targeted with ionizing radiation causing lethal damage to cancer cells resulting in cell death. However, RT also inflicts acute and chronic toxicities to the normal tissue surrounding the tumor (2–6). These radiation-induced toxicities limit the dose of radiation that can be delivered and subsequently limits the extent to which RT can be curative. Furthermore, as the number of long-term cancer survivors increases, late onset toxicities resulting from RT are emerging that significantly impact the quality of life of those patients. Consequently, there is a need for novel RT strategies that maintain the anti-tumor effect whilst limiting the extent of toxicities induced in the surrounding healthy tissue. Limiting the induction of toxicities to normal tissue would subsequently increase the therapeutic index of RT regimes (7). A number of recent studies have demonstrated that irradiation at ultra-high dose rates (FLASH) diminishes the severity of toxicities in normal tissues compared to irradiation at the conventional dose rates (CONV) currently used in clinical practice (8–18). Notably, limited data also shows that FLASH-RT reduces normal tissue toxicities whilst maintaining the anti-tumor response of CONV-RT (8–10, 15, 17, 19). FLASH-RT delivery uses irradiators with a high radiation output that allows for the entire RT treatment, or large fraction doses, to be delivered in parts of a second, compared to several minutes for CONV-RT. The short treatment times used in FLASH-RT, often shorter than 0.1 s, have the added value of minimizing treatment delivery uncertainties caused by intrafraction motion. Carefully implemented, this would allow for smaller treatment margins and therefore smaller volumes of normal tissue being unnecessarily irradiated. Given both the radiobiological advantageous FLASH effect and its potential to "freeze" physiological motion (15, 20), FLASH-RT has the potential to be an important evolutionary step in cancer treatment. The biology underpinning the FLASH effect, however, remains unknown.

#### FLASH-RT LIMITS NORMAL TISSUE TOXICITY

Investigation of the dose rate at which RT is delivered harks back to the 1960s, when it was demonstrated that noncancerous mammalian cells irradiated at ultra-high dose rates had greater viability than those irradiated at conventional dose rates (21). More recently, this toxicity-limiting property of ultra-high dose rate was rediscovered and named FLASH by Favaudon et al. (10). In their study, they demonstrated that thoracic irradiation of mice with a single fraction of 17 Gy at conventional dose rates (0.03 Gy/s) induced "moderate" and "severe" regions of pulmonary fibrosis at 36 weeks postirradiation. In contrast, when mice received the same dose at ultra-high dose rates (40–60 Gy/s) the induction of pulmonary fibrosis was starkly reduced. A greater dose of 30 Gy delivered by FLASH-RT was required to induce comparable levels of pulmonary fibrosis as seen following CONV-RT (10). Whilst exploring this reduction in pulmonary fibrosis following FLASH-RT, the same group investigated any changes in the induction of the transforming growth factor beta (TGFβ) signaling cascade a well-documented molecular marker of radiation-induced pulmonary fibrosis (22). In accordance with their prior findings, CONV-RT of 17 Gy significantly induced TGFβ signaling; this signaling was reduced in mice that had been subjected to FLASH-RT. Once again, a greater dose of 30 Gy delivered by FLASH-RT was required to induce TGFβ signaling to the equivalent extent as seen following irradiation with CONV-RT (10). Limited TGFβ signaling following FLASH-RT has also been shown in vitro (23): this study demonstrated that even 24 h post-irradiation, CONV-RT induced 3-fold greater TGFβ signaling compared to FLASH-RT.

In addition to thoracic irradiation, it has been shown in several studies that whole brain irradiation using FLASH-RT confers neuroprotection compared to CONV-RT (13, 14, 24, 25). In one such study, mice were exposed to varying dose rates, ranging from 0.1 Gy/s to 10 Gy delivered in a single 1.8 µs pulse; at all dose rates mice were exposed to 10 Gy in a single fraction (14). Any radiation-induced neurotoxicity was measured by a novel object recognition test 2 months post-irradiation. Analysis of these data showed that mice irradiated at 0.1 Gy/s performed significantly worse on the novel object recognition test compared to the non-irradiated control. Notably, as dose rate increased, mice performed significantly better in the recognition test when irradiated at dose rates ≥ 30 Gy/s. Furthermore, there was no statistical difference in novel object recognition between mice irradiated at dose rates exceeding 100 Gy/s and non-irradiated mice (14).

In earlier studies, it was observed in rodent models that radiation-induced skin reactions could be significantly reduced at ultra-high dose rates (26, 27). Specifically, it was shown in a rat model that irradiation at 67 Gy/s induced less severe skin reactions, e.g., reddening, moist desquamation, and skin breakdown, in the short and long term compared to rats irradiated at either 1 or 0.03 Gy/s. This study also measured the deformity of the irradiated feet 6 months post-irradiation; consistent with the induction of skin reactions, the extent of deformation was less in the rats irradiated at 67 Gy/s compared to the two lower dose rates (26). Pre-clinical FLASH-RT studies have also been extended from rodent models to higher mammals such as mini-pigs and cats (16). As recently and succinctly reviewed (28), this study irradiated ten 26 mm in diameter circular patches of skin on the back of a single mini-pig to five different dose levels from 22 to 34 Gy (in 3 Gy increments), with either FLASH-RT at a dose rate of 300 Gy/s, or CONV-RT at 0.083 Gy/s. Examination 48 weeks post-irradiation showed that FLASH-RT had been well-tolerated, with only mild cutaneous depigmentation at the site of irradiation (16). In contrast, sites subjected to CONV-RT presented with clear fibronecrotic lesions. By way of extension, this study used FLASH-RT to treat six cats, all presenting with squamous cell carcinoma of the nasal planum, to a total dose ranging from 25 to 41 Gy. All six cats responded extremely well to treatment with complete remission of tumors with minimal toxicity; cats treated with the largest doses of radiation exhibited moist desquamation around the site of irradiation (16). An obvious limitation of this study is the lack of a parallel arm of cat subjects treated with CONV-RT.

Many pre-clinical studies have reported a successful FLASH normal tissue sparing effect, but it cannot be overlooked that there have also been several studies reporting no significant sparing of normal tissues following irradiation at ultra-high dose rates (29–33). For example, Smyth et al. delivered whole and partial body (abdominal or head) synchrotron irradiation to mice, at ultra-high dose rates of 37–41 Gy/s in the hope of characterizing the equivalent CONV-RT dose (32). However, comparing TD<sup>50</sup> values (dose predicted to cause toxicity, i.e., >15–20% weight loss, severe diarrhea, moribund behavior, in 50% of the animals), this study did not observe any differential sparing between broad beam irradiation of ultra-high and conventional dose rates. A similar study by Montay-Gruel et al. delivering whole brain synchrotron irradiation at a dose rate of 37 Gy/s to mice, did however show significant neurocognitive sparing compared to conventional X-ray irradiation (24). Synchrotron irradiation beams are very flat, several cm in width but with a height on the µm-mm scale, requiring the irradiated sample to be scanned through this beam slice. For studies investigating the FLASH effect with synchrotron irradiation, the dose rate within the beam slice is likely the most important parameter. So even though the average dose rate was similar in these two studies, and probably just high enough for a FLASH sparing effect (14), the height of the beam slice through which the mice were scanned was different by a factor 20 (50µm compared to 1 mm), corresponding to the same difference in dose rate in the slice (12 000 Gy/s compared to 600 Gy/s) (14, 32). This difference in beam slice dose rate, and of course the difference in the investigated end-points, could explain why one study found a FLASH sparing effect whilst the other study did not. A summary of in vivo studies investigating the tissue response to FLASH-RT compared to CONV-RT, across a range of tissue types, are shown in **Tables 1**, **2**, many of which have demonstrated a reduction in radiation-induced toxicities for FLASH-RT (10–16, 24–27, 34).

#### SIMILAR ANTI-TUMOR RESPONSE WITH FLASH-RT AS CONV-RT

In addition to limiting toxicities, there have also been reports of FLASH-RT maintaining the same tumor response as seen following CONV-RT (8, 10, 17, 19, 35). In one such study, breast cancer, and head and neck carcinoma xenografts were established in mice (10). Both tumor models were then exposed to either FLASH-RT or CONV-RT; tumor volume was controlled independent of dose rate in breast, and head and neck xenografts. In the same study, mouse lung carcinoma luciferase-positive (luc+) TC-1 cells were transpleurally injected to generate an orthotopic lung tumor model. Thoracic irradiation of the mice with either CONV-RT or FLASH-RT, and subsequent evaluation of tumor growth using bioluminescence, showed no difference in treatment efficacy (10). Similarly in another study, human glioblastoma (GBM) were engrafted to nude mice and locally irradiated with either FLASH-RT or CONV-RT, resulting in similar tumor growth retardation (19). In the study by Bourhis et al. H454-luc+ murine GBM cells were implanted orthotopically in the striatum of nude mice. This

TABLE 1 | Summary of irradiation parameters and outcomes for in vivo studies investigating the FLASH effect in normal tissues (organized in order of model species and targeted tissue, as well as color coded by radiation modality).


TABLE 2 | Summary of irradiation parameters and outcomes for in vivo studies investigating the FLASH effect in tumor tissues (organized in order of model species and targeted tissue, as well as color coded by radiation modality).


was subsequently followed by whole brain irradiation 3 days post-implantation with either single pulse (1.8 µs) FLASH-RT or CONV-RT (0.1 Gy/s) (8). The mice were irradiated with a 10 Gy single fraction, 3 times 8 Gy, or 5 times 5 Gy, with 24 h in-between fractions. Using bioluminescence to assess the tumor burden, no significant difference could be seen between FLASH-RT and CONV-RT for any of the fractionation schemes (8). In a study by Rama et al. Lewis Lung Carcinoma (LLC) cells were inoculated into the left lung of C57Bl/6J mice (36). Two weeks post-inoculation, the whole lungs of tumor-bearing mice were irradiated with a single fraction dose of 18 Gy, using a clinical pencil beam scanning proton system. One week post treatment, CT-scans were performed to measure tumor size. Tumor size was also measured with a caliper after the mice had been sacrificed 10 days post-treatment. Surprisingly, the tumors of the mice treated with proton FLASH-RT were smaller than the tumors of the mice treated with proton CONV-RT. Moreover, immuno?uorescent staining on harvested tumor sections showed an improved recruitment of T lymphocytes into the tumor microenvironment for tumors treated with FLASH-RT compared to CONV-RT (36). Evidentially in some cases, the anti-tumor response to FLASH-RT might even be better than that of CONV-RT.

### WHAT FACTORS INFLUENCE THE FLASH EFFECT?

An important caveat of the pre-clinical studies investigating FLASH-RT is the lack of consistency between variables that could potentially influence the induction of the FLASH effect such as: dose rate, total dose, pulse rate, fractionation, and modality of radiation (**Tables 1**, **2**). The study by Montay-Gruel et al. using a wide range of dose rates has helped to elucidate the extent to which dose rate modulates the FLASH effect (14). As previously described, a neuroprotective FLASH effect was apparent at dose rates ≥ 30 Gy/s with a maximal FLASH effect induced at dose rates ≥ 100 Gy/s. This relationship is important to consider when examining studies such as those by Favaudon et al. (10), and Vozenin et al. (15, 16), which used 40–60 and 300 Gy/s, respectively when administering FLASH-RT. In contrast to previously mentioned studies, a recent interesting study by Venkatesulu et al. showed a higher toxicity for FLASH-RT delivered at 35 Gy/s than for CONV-RT delivered at 0.1 Gy/s (33). This dose rate is probably on the low side for a sparing effect to occur but that does not explain the highly unexpected increased toxicity they found for FLASH-RT in all of their experiments, especially the increased toxicity of a factor 1.3–1.4 for their in vitro data. There could be many reasons for these results, e.g., the dose-rate needed for a FLASH sparing effect might not be universal but rather tissue-specific, model and/or assay specific, or there could be dosimetric differences between the two delivery modes/setups, all of which highlights the challenge in performing studies at these dose rates, finding, and exploring a beneficial FLASH effect (33). Furthermore, there is a large degree of variation in the total dose of radiation used in pre-clinical FLASH-RT studies. Compounding this, the majority of studies administer FLASH-RT in single fractions of 10 Gy or more; in many clinical situations, these are currently considered to be extremely large and unattainable fraction doses.

The source of radiation must also be considered when evaluating the FLASH effect. The FLASH effect has been predominantly observed following FLASH-RT using dedicated electron linear accelerators as the source of radiation (10, 14, 15, 18, 37). However, recent studies have expanded the FLASH field and include observations of a FLASH effect following proton (11, 23, 36) and X-ray (24) irradiation. Again, it must be noted that there have been a couple of studies that have been unable to induce a FLASH effect using proton and X-ray sources (**Table 1**). The reason for one X-ray study showing a FLASH effect and one study not showing an effect was discussed above. The proton study compared quasi-continuous proton beam delivery at a CONV-RT dose rate of 5 Gy/min to FLASH-RT of 100 Gy/s, without seeing any toxicity difference for zebrafish embryos (29). A reason for the absent FLASH effect might be the quasi-continuous proton beam delivery with several orders of magnitude lower dose rates within each micro-pulse (≈ 10<sup>3</sup> Gy/s) than the FLASH electron studies macro-pulses (≈ 10<sup>6</sup> Gy/s) (29). So, further to mean dose rate, total dose, and the source of radiation, the pulsatile nature of irradiation may also influence the FLASH effect. In order to induce a FLASH effect, it seems that the irradiation beam should ideally be pulsed at a frequency in the order of 100 Hz (**Figure 1**). Furthermore, within each pulse; irradiation should be delivered at sufficiently high dose-per-pulse, and dose rate within the pulse (≥ 1 Gy and ≥ 10<sup>6</sup> Gy/s, respectively). Together, resulting in a total treatment delivery time of maximum a few tenths of a second (**Table 1**). The range of variables and outcomes seen to date warrants further investigation to confirm that these are the key parameters for inducing the FLASH effect (**Figure 1**).

### HYPOTHESES TO EXPLAIN THE FLASH EFFECT

#### Oxygen Depletion Hypothesis

The biological mechanism responsible for the reduction in normal tissue toxicities following irradiation at FLASH dose rates is not currently understood, yet several non-mutually exclusive hypotheses have been proposed. Some researchers have suggested that the differential response between FLASH-RT and CONV-RT may be due to the radiochemical depletion of oxygen at ultra-high dose rates and subsequent radioresistance conferred to the irradiated tissue (32, 38, 39). It is widely accepted that hypoxic tissues are more radioresistant than well-oxygenated tissues. This is because in the presence of molecular oxygen there is fixation of indirect radiation-induced DNA damage. Indirect damage, the predominant mechanism by which low linear energy transfer (LET) radiation induces DNA damage, occurs when

radiation results in the radiolysis of water molecules and the subsequent generation of free radicals. Free radicals are then incorporated into DNA, causing damage—yet this can be easily resolved. However, if a free radical reacts with molecular oxygen, this yields a peroxyl radical. Peroxyl radicals have the potential to induce permanent damage, and are therefore a more efficacious DNA damaging agent. Hence, a lack of oxygen in the immediate environment of a cell limits the extent of radiation-induced DNA damage (40).

When considering the oxygen depletion theory, it is important to note the nature of physiologically relevant oxygen concentrations, or "physoxia" (41). Normal tissues in vivo are perfused at much lower oxygen concentrations than in vitro cell lines cultured in atmospheric oxygen concentrations. Depending on tissue type, physoxia generally lies between 3.4 and 6.8% oxygen (42). Especially relevant for current treatment with FLASH-RT limited to superficial tissues, physoxia in skin increases with depth from the surface of the skin to the dermis, from around 1.1–4.6% (43). Considering physoxia, and given the critical relationship between oxygen concentration and radiosensitivity radiochemical oxygen depletion has the potential to significantly dampen the radiobiological response.

A relationship between dose rate and oxygen consumption was proposed by Dewey and Boag in 1959 (44). They demonstrated that bacteria irradiated at ultra-high dose rates had greater survival compared to bacteria irradiated at what we now consider to be conventional dose rates. The survival curve generated following ultra-high dose rate irradiation was indicative of bacteria irradiated in a hypoxic environment. The authors hypothesized at the time that this response was a consequence of oxygen depletion following a large dose of radiation in such a short timeframe; the time for which the bacteria were irradiated for was shorter than the time required for oxygen to diffuse and restore the oxygen that had been depleted. Given that molecular oxygen is depleted as it reacts with free radicals generated from the radiolysis of water, irradiation at ultra-high dose rates is able to significantly deplete oxygen before it can replenish. This gives rise to a small window of radiobiological hypoxia.

The oxygen-depletion hypothesis has been strengthened by work demonstrating that as dose rate is increased, cellular survival mimics that of cells irradiated in an increasingly hypoxic environment (45, 46). Furthermore, it was subsequently shown in mammalian cells that the oxygen-dependent fixation of indirect DNA damage could be dampened at ultra-high dose rates (47). Importantly, the total dose at which these cells exhibited a hypoxic-like response was linear with respect to increasing the oxygen concentration in which the cells were cultured. The range of oxygen concentrations used in this study was relatively narrow (0.44–0.7% O2) and therefore the phenomenon could have been limited to cells already in hypoxic environments. However, the recent in vitro study by Adrian et al. used physiologically relevant oxygen concentrations (1.6–8.3% O2) and showed that the sparing effect of FLASH irradiation is dependent on oxygen concentration (48). An in vivo mouse model has also shown that irradiation of mouse tails at ultra-high dose rates induced radioresistance indicative of oxygen depletion (49).

Together, these data suggest that the irradiation of tissues with FLASH-RT results in radiochemical oxygen depletion, giving rise to an extremely acute period of hypoxia within the irradiated tissue and consequently a transient radioresistance (**Figure 2**). This phenomenon is not seen following irradiation with CONV-RT as radiation is delivered with much smaller pulses and over a longer timeframe. Hence during CONV-RT, oxygen depletion is limited, and there is sufficient time for oxygen to diffuse into the irradiated region to replace oxygen that has been lost. Therefore, oxygen concentration within the irradiated tissue is maintained.

There is growing interest surrounding other oxygen-based radicals as a potential mechanism bridging the local oxygen depletion observed following irradiation at ultra-high dose rates, and reduced toxicities to normal tissue. A recent study proposes that oxygen depletion at ultra-high dose rates promotes the protection of normal tissue by limiting the production of reactive oxygen species (ROS) (13). This study repeated previous work, demonstrating that whole brain irradiation of C57Bl6/J mice with FLASH-RT did not induce cognitive impairments at dose rates exceeding 100 Gy/s compared to non-irradiated controls. Moreover, in support of a critical role for oxygen in the FLASH effect, increasing the local oxygen concentration in mice brains through carbogen breathing reversed the cognitive protection conferred by FLASH-RT. Furthermore, zebrafish embryos were subjected to either FLASH-RT or CONV-RT in the presence or absence of two well-documented ROS scavengers: N-acetylcysteine (NAC), and amifostine (13). Giving weight to the involvement of ROS in the FLASH effect, zebrafish embryos exposed to FLASH-RT in combination with a ROS scavenger had no effect on zebrafish length 5 days post-irradiation. However, zebrafish embryos exposed to CONV-RT alone were significantly shorter than those exposed to CONV-RT in combination with a ROS scavenger (13). This provides crude but encouraging evidence suggesting that toxicities arising from CONV-RT are in part due to the generation of ROS, and that the generation of these species is reduced following FLASH-RT. The largest limitation of this study is that there are no direct measurements of ROS in a physiological context. Instead, water containing 4% aqueous oxygen was irradiated at either ultra-high or conventional dose rates; conventional dose rates generated significantly greater ROS than ultra-high dose rates (13). Despite this short fall, the interesting findings detailed upon irradiation in combination with antioxidants merits further exploration into the role of ROS for the FLASH effect.

The oxygen depletion hypothesis seems to explain the reduced toxicity of FLASH-RT to normal tissue. However, it does not easily explain how FLASH-RT can maintain tumor response relative to CONV-RT. Although tumors are more hypoxic compared to their normal tissue counterparts, most are not completely anoxic (42). Therefore, following FLASH-RT, there will also be radiochemical depletion of oxygen within the tumor, hence it would be expected that this would confer radioresistance to the tumor. In contrast to experimental data (8, 10, 19), one would subsequently expect to observe reduced tumor control following FLASH-RT relative to CONV-RT. Though, for highly hypoxic tumor models the reduced tumor control would be expected to be minimal (**Figure 2**). A possible explanation for the

maintained tumor control is proposed in a recent paper by Spitz et al. They hypothesized that higher levels of redox-active iron (labile iron) in tumor compared to normal tissue and differences in oxidative metabolism between normal and tumor tissues, with the more rapid removal and decay of the organic hydroperoxides and free radicals derived from peroxidation chain reactions in normal tissue, defines the beneficial therapeutic index of the FLASH effect (50). Interestingly, a recent computational model of oxygen depletion induced by FLASH-RT concluded that radiochemical oxygen depletion at an expected rate of 0.42 mmHg/Gy would be sufficient to confer radioresistance (51). However, this conclusion was predicated on the basis that radioresistance would only be conferred to already hypoxic tissues. To explore this, it would be interesting to compare the DNA repair proficiency of normal tissue relative to tumor tissue; perhaps radioresistance induced in tumor tissue by oxygen depletion is compensated for by a lower ability of DNA repair compared to normal tissue. Regions of hypoxia occur in the majority of solid tumors as opposed to the physoxia found in the surrounding normal tissue. This may well be relevant to the relative repair of DNA damage induced by FLASH-RT as exposure to hypoxia has also been described to lead to the repression of the DNA repair pathways including homologous recombination (HR), non-homologous end joining (NHEJ), and base excision repair (BER) (52, 53). To test this hypothesis, the rate of DNA repair, assayed for example by determining the appearance and resolution of 53BP1 foci, should be measured in both normal and tumor cells after exposure to FLASH-RT.

The vast majority of data pertaining to the oxygen depletion theory has been extrapolated from cell survival responses following irradiation at different dose rates (44–47, 49, 54). Therefore, there must be more direct measurements of any potential oxygen flux in tissues following irradiation at ultra-high dose rates. However, given the supposed brevity of any hypoxia induced by FLASH-RT, this is extremely difficult; it has been inferred that reoxygenation by diffusion of a tissue following FLASH-RT occurs after just 10−<sup>3</sup> s (54). Hypoxia for such a brief moment can certainly not be detected by measuring markers of a hypoxia-mediated transcriptional response, which would be observed following a longer period of hypoxia (41). However, it is unknown whether a chemical marker of hypoxia, such as pimonidazole (55) is sufficiently sensitive to detect such an acute period of hypoxia.

#### Immune Hypothesis

A modified immune response following FLASH-RT relative to CONV-RT has also been proposed as a potential mechanism for the FLASH effect (9, 38). The fractionated RT regimes commonly used in CONV-RT, result in the irradiation of a greater proportion of circulating lymphocytes compared to total dose delivered in a single fraction (56). Following a standard regime of thirty fractions of 2 Gy, 98.8% of the blood pool has been exposed to more than 0.5 Gy. Additionally, it has been reported that the induction of chromosomal aberrations in the circulating blood pool is dependent on the total volume of the blood pool irradiated (57). Therefore, in accordance with the short irradiation time, characteristic of FLASH-RT, it would follow that fewer lymphocytes would be irradiated and subsequently reduced induction of chromosomal aberrations (9, 38, 56). However, FLASH-RT would expose lymphocytes to a greater dose of radiation, albeit much fewer of them, in comparison to CONV-RT. If a modified immune response contributes to the FLASH effect, one would expect a fractionated FLASH-RT regime to, at least in part, reduce any protection conferred by the FLASH effect.

This hypothesis has been strengthened recently by a study that carried out genome-wide microarray analysis on mice following FLASH-RT and CONV-RT (11). This study reported that immune system wide activation and maturation was dampened in mice following FLASH-RT relative to CONV-RT. Also as mentioned above, the study by Rama et al. showed an improved recruitment of T lymphocytes into the tumor microenvironment for tumors treated with FLASH-RT compared to CONV-RT, which gives merit to this hypothesis (36). In several studies, immunocompromised animals were used to compare treatment efficacy of FLASH-RT and CONV-RT with no observed difference in tumor response (**Table 2**), which could be interpreted to further strengthen the hypothesis (7, 8, 10, 35). It is worth noting however, that any evidence linking an immune role to the FLASH effect is correlative rather than causative; it is unclear whether any differential immune response following irradiation at ultra-high dose rates contributes to the FLASH effect, or is a consequence of it. Additionally, since the FLASH effect has been observed in vitro in bacterial and cell culture models, which are devoid of a functioning immune system, any immunological component is likely to be responsible for only part of the underlying mechanism. More studies are needed to clarify if the immune response or other biological responses like DNA damage response or inflammation is different following FLASH-RT compared to CONV-RT, and if they are part of the underlying mechanism resulting in the FLASH effect.

#### CLINICAL APPLICATIONS OF FLASH-RT

The obvious endpoint of investigation into the FLASH effect is the translation of FLASH-RT to the clinic. FLASH-RT could be translated to the clinic to serve two general purposes. Firstly, the FLASH effect could be exploited to allow for escalation of total dose in the treatment of radioresistant tumors that are currently associated with poorer patient outcomes (8). In this case, it is hypothesized that a greater dose of radiation could be delivered to the tumor without inducing as severe toxicities to the normal surrounding tissue as would be expected following CONV-RT. Secondly, FLASH-RT could be used in situations in which RT confers good levels of tumor control but is associated with severe normal tissue toxicity—the same total dose would be administered, but hypothetically FLASH-RT would induce less severe toxicities compared to CONV-RT.

Despite these exciting potential applications of FLASH-RT, the extent to which it is clinically viable in practice is questionable. As reviewed above, there are some inconsistencies in the results from the pre-clinical studies. Furthermore, a proportion of these studies are designed with significant limitations, such as using a single subject and a lack of controls irradiated at conventional dose rates (15). Moreover, the results emerging from pre-clinical studies put into question the suitability of FLASH-RT in many clinical situations. Independent studies that have successfully observed a FLASH effect report a dose-modifying factor of about 20–40% in favor of FLASH-RT relative to CONV-RT (**Table 1**). However, these same studies only report a FLASH effect at total doses of 10 Gy or more. This point is particularly well-illustrated in the recent study by Vozenin et al. (16). In a zebrafish model, whereby zebrafish embryos were irradiated with FLASH-RT or CONV-RT at doses ranging from 5 to 12 Gy, increasing in 1 Gy increments, zebrafish length was recorded 5 days post-irradiation as a measure of radiation-induced toxicity. A significant difference in morphology between those irradiated with FLASH-RT or CONV-RT was only apparent at doses ≥ 10 Gy. Even when accounting for the dose modifying factor of FLASH-RT, an equivalent dose per fraction of 6–8 Gy given by CONV-RT may still be considered as too large a dose in various clinical scenarios (58–60), such as in the treatment of larger, locally advanced tumors. A previous phase I dose escalation study in locally advanced non-small cell lung cancer (NSCLC) utilized hypofractionated treatment with doses per fraction well-below those required for a FLASH effect (58). Six patients developed late onset, grade 4–5 toxicities that were attributed to damage to the proximal bronchial tree, ergo highlighting the need for caution when employing hypofractionated regimes. Hypofractionation is nevertheless getting more widely used in the clinic for a variety of treatments sites (59, 61–64), and could be proven even more useful together with FLASH-RT and its (potentially) lower level of normal tissue toxicity.

One of the most interesting advancements in the FLASH field is the first human patient treated with FLASH-RT (9). A 75-year-old male presenting with multiresistant CD30+ T cell cutaneous lymphoma was offered the opportunity to be first human subject of FLASH-RT. A 35 mm lesion was exposed to a dose rate exceeding 10<sup>6</sup> Gy/s in each of ten discreet 1 µs pulses to a total dose of 15 Gy. This equates to a mean dose rate of 167 Gy/s, and 1.5 Gy per pulse. Following treatment, shrinkage of the lesion was observed 10 days post-irradiation culminating in a complete tumor response 36 days post-irradiation which was maintained for the following 5 months. From the point at which the lesion initially began to shrink, the patient presented with redness and mild (grade 1) oedema and epithelitis around the site of irradiation. This was starkly different to the patient's other lesions treated with CONV-RT that resulted in high-grade acute reactions to the surrounding skin that took ∼3–4 months to heal (9). Despite the promising outcome for this patient, this should not be considered evidence confirming that FLASH-RT can be successfully translated to the clinic. This study was performed in a single patient that only allowed for limited comparison of the differential response between FLASH-RT and CONV-RT. An appropriately powered, randomized controlled trial with FLASH-RT and CONV-RT arms would be required to definitively show whether FLASH-RT is associated with superior clinical outcomes. At the very least, a positive phase II, single-arm study of FLASH-RT in a sample of participants truly representative of real-world patients is required before the routine adoption of FLASH-RT can be seriously entertained. If 4.5–20 MeV electron beams are to be used for the clinical trials, they would be limited to treating surficial tumors or treating tumors with intra-operative radiation therapy (IORT). Currently, FLASH-RT clinical trials on deep-seated tumors can only be performed with proton beams (**Table 3**). However, to treat tumors with a proton beam in a clinical trial, the beam needs to be scattered or scanned to cover the target volume which reduces the average dose rate (65). So before performing clinical trial, pre-clinical studies are needed to ensure that the FLASH effect is not lost due to either the increased LET in the Bragg peak or to the required scattering/scanning of the beam.

As previously mentioned, most studies showing a FLASH effect has dedicated electron linear accelerators as the source of radiation (9, 10, 14, 15, 18, 37). Recent studies have shown that clinical linear accelerators can be modified to deliver FLASH-RT with electrons, largely increasing the potential availability TABLE 3 | Some relevant advantages and disadvantages of current and prospective FLASH radiotherapy sources (color coded by radiation modality).


of FLASH-RT devices and facilitating the translation to clinical trials (66, 67). However, an obvious limitation is the depth penetration with 4.5–20 MeV electron beams, only reaching to a few cm depths in tissue (**Table 3**). Consequently, other treatment devices/techniques are needed for FLASH-RT to be clinically useful for more than superficial treatments with external beam RT or IORT. A solution to the limited depth penetration would be to use electron beams of higher energy, so called Very High Energy Electron (VHEE) beams, with beam energies of 100–250 MeV. Such beams have good depth penetration, sharp beam penumbra, and are less sensitive to tissue heterogeneity than conventional X-ray beams (68, 69). Also, using electromagnets, the beam can in theory be focused to the tumor volume, resulting in dose-to-target conformity with a single beam comparable to that of modern X-ray treatment techniques, e.g., intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT). A single beam delivery might prove essential for retaining the FLASH effect in clinical trials. Unfortunately, these beams are currently limited to research accelerators which are either rather large (linear accelerator) or suffers from a low pulse rate, a small beam size, and stability issues (laser-based accelerators) (68–71). A recent paper showed (using a 160 kV X-ray beam) that conventional X-ray tubes could potentially be used for FLASH-RT studies (72). This is interesting as such systems are small, relatively inexpensive and clinically available (**Table 3**). Similar however to the electron linear accelerators, the depth penetration is a limiting factor making it useful only down to a few mm depth in tissue, an additional limitation is the beam size of only a few cm. Synchrotron sources has similar beam energies as X-ray tubes but has the added advantage of the possibility of using spatially fractionated ultra-high dose rate microbeam radiation therapy (MRT). MRT is characterized by arrays of quasi-parallel microplanar beams with a width of 25–100µm, typically separated by 100–400µm (32). Since its invention in 1992, numerous preclinical studies have shown extraordinary tolerance of normal organs and blood vessels exposed to fractionated radiation doses in excess of 100 Gy in-beam (peak) doses, with dose rates exceeding several hundred Gy/s. The combined effect of spatially fractionated microbeams and FLASH dose rates have been shown in small animal models to achieve therapeutic ratios that clearly exceed those obtained by conventional X-ray with a homogeneous dose distribution and CONV-RT dose rates, in a range of malignancies, including gliomas, gliosarcomas, human squamous cell carcinomas, and glioblastomas (73). The disadvantage of this technique is the requirement of synchrotrons, which are very large, expensive, and therefore of limited availability. A platform that might solve both the size and stability issue of VHEE beams and also allow for the production of 6–10 MV FLASH X-ray beams, is PHASER (Pluridirectional High-energy Agile Scanning Electronic Radiotherapy). The PHASER concept has been presented by Maxim et al. and might be an ideal way for introducing FLASH into the clinic (20). Included in the concept is a novel and quick image-guided technique. New or highly adapted image-guidance techniques are needed for the clinical treatment of deep-seated tumors with FLASH-RT, regardless of radiation modality. The PHASER is reliant on technical advances and novel innovations in linear accelerator technology, radiofrequency science and medical physics, which in turn requires time and funding for research and development. Therefore, it is still under development (**Table 3**). Alternative concepts of producing 6–10 MV FLASH X-ray beams would be to use multiple synchronized linear accelerators or a powerful recirculating accelerator (74). Albeit large and expensive, a clinically available system for treating deep-seated tumors with FLASH-RT is with proton beams (75, 76). Clinical proton beams have good depth penetration, are often electromagnetically steered, and can produce conformal dose distributions with a single to a few beams (65). There have been studies (published and unpublished) with mixed reports on a FLASH effect with protons but significant resources have now been put into research on proton FLASH-RT by the principal vendors for proton RT devices, which should expedite the translation of proton FLASH-RT into clinical trials (77–79).

#### CONCLUSION

The FLASH effect is an extremely interesting radiobiological phenomenon that confers some degree of protection compared to CONV-RT. The FLASH effect has now been observed across a range of animal models, and more recently has been suggested in a human patient for the first time. Of equal importance, limited data would suggest that FLASH-RT maintains a similar tumor response to CONV-RT. Together, this raises the prospect that FLASH-RT will allow patients to receive a greater total dose of

#### REFERENCES


radiation prior to the induction of unacceptable toxicities that currently limit RT regimes.

There has been much speculation regarding the biological mechanism(s) underpinning the FLASH effect. It is wellestablished that irradiation results in the radiochemical depletion of oxygen; this is particularly prevalent at ultra-high dose rates. From the data currently available, we can safely conclude that oxygen depletion contributes, at least in part, to the FLASH effect. However, the extent of its contribution remains unknown and therefore warrants further investigation. Aside from oxygen depletion, an immune modulatory role has been broadly implicated in the FLASH effect, yet evidence to support this is currently sparse and preliminary. Likewise, any potential immune-mediated contribution to the FLASH effect requires much greater exploration.

Aside from mechanistic insights, the overarching question remains of the translational potential of FLASH-RT to clinical environments. Despite independent studies concluding that FLASH-RT confers a dose modifying factor of 20–40%, the repeated finding that the FLASH effect is only evident at total doses of 10 Gy or more means that FLASH-RT would not be suitable in many clinical cases. As a result of further investigation into the biological basis of the FLASH effect, it may eventually be possible to generate a FLASH effect at smaller doses, therefore further increasing the clinical potential of FLASH-RT. Another limiting factor in translating FLASH-RT to the clinic is the availability of radiation sources, capable of producing beams suitable for treatment of deep-seated as well as superficial tumors with ultra-high dose rates. In summary, with shorter treatment times and lower levels of toxicity, FLASH-RT may 1 day have the potential to be a paradigm shift in the field of RT. For this to be the case, however, there is a real need to identify the mechanism(s) behind the FLASH effect. The currently available data more than justifies this further investigation.

#### AUTHOR CONTRIBUTIONS

JW and KP wrote the article. EH and GH contributed ideas to, read, and edited the article.

#### FUNDING

JW was grateful for support from the Royal College of Radiologists, and Oriel College, Oxford.


dose-rates. Int J Radiat Biol Relat Stud Phys Chem Med. (1974) 26:17–29. doi: 10.1080/09553007414550901


radiation therapy for hepatocellular carcinoma patients with varying degrees of hepatic impairment. Int J Radiat Oncol Biol Phys. (2018) 100:577–85. doi: 10.1016/j.ijrobp.2017.11.030


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer M-CV declared a past supervisory role with one of the authors KP to the handling editor.

Copyright © 2020 Wilson, Hammond, Higgins and Petersson. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Linking NRP2 With EMT and Chemoradioresistance in Bladder Cancer

Alexander Schulz 1†, Ielizaveta Gorodetska1†, Rayk Behrendt <sup>2</sup> , Susanne Fuessel <sup>3</sup> , Kati Erdmann<sup>3</sup> , Sarah Foerster <sup>4</sup> , Kaustubh Datta<sup>5</sup> , Thomas Mayr <sup>4</sup> \*, Anna Dubrovska1,6,7,8 \* and Michael H. Muders <sup>4</sup> \*

<sup>1</sup> Faculty of Medicine and University Hospital Carl Gustav Carus, OncoRay-National Center for Radiation Research in Oncology, Technische Universität Dresden and Helmholtz-Zentrum Dresden-Rossendorf, Dresden, Germany, <sup>2</sup> Faculty of Medicine, Institute for Immunology, Technische Universität Dresden, Dresden, Germany, <sup>3</sup> Department of Urology, Technische Universität Dresden, Dresden, Germany, <sup>4</sup> Rudolf Becker Laboratory for Prostate Cancer Research, Center of Pathology, University of Bonn Medical Center, Bonn, Germany, <sup>5</sup> Department of Biochemistry and Molecular Biology, University of Nebraska Medical Center, Omaha, NE, United States, <sup>6</sup> Helmholtz-Zentrum Dresden - Rossendorf, Institute of Radiooncology – OncoRay, Dresden, Germany, <sup>7</sup> German Cancer Consortium (DKTK), Partner Site Dresden, Dresden, Germany, <sup>8</sup> German Cancer Research Center (DKFZ), Heidelberg, Germany

#### Edited by:

Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands

#### Reviewed by:

Hyuk-Jin Cha, Seoul National University, South Korea Michael Wayne Epperly, University of Pittsburgh, United States

#### \*Correspondence:

Thomas Mayr Thomas.Mayr@ukbonn.de Anna Dubrovska Anna.Dubrovska@OncoRay.de Michael H. Muders Michael.Muders@ukbonn.de

†These authors share first authorship

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 02 August 2019 Accepted: 05 December 2019 Published: 21 January 2020

#### Citation:

Schulz A, Gorodetska I, Behrendt R, Fuessel S, Erdmann K, Foerster S, Datta K, Mayr T, Dubrovska A and Muders MH (2020) Linking NRP2 With EMT and Chemoradioresistance in Bladder Cancer. Front. Oncol. 9:1461. doi: 10.3389/fonc.2019.01461 Neuropilin-2 (NRP2) is a prognostic indicator for reduced survival in bladder cancer (BCa) patients. Together with its major ligand, vascular endothelial growth factor (VEGF)-C, NRP2 expression is a predictive factor for treatment outcome in response to radiochemotherapy in BCa patients who underwent transurethral resection. Therefore, we investigated the benefit of combining cisplatin-based chemotherapy with irradiation treatment in the BCa cell line RT112 exhibiting or lacking endogenous NRP2 expression in order to evaluate NRP2 as potential therapeutic target. We have identified a high correlation of NRP2 and the glioma-associated oncogene family zinc finger 2 (GLI2) transcripts in the cancer genome atlas (TCGA) cohort of BCa patients and a panel of 15 human BCa cell lines. Furthermore, we used in vitro BCa models to show the transforming growth factor-beta 1 (TGFβ1)-dependent regulation of NRP2 and GLI2 expression levels. Since NRP2 was shown to bind TGFβ1, associate with TGFβ receptors, and enhance TGFβ1 signaling, we evaluated downstream signaling pathways using an epithelial-to-mesenchymal transition (EMT)-assay in combination with a PCR profiling array containing 84 genes related to EMT. Subsequent target validation in NRP2 knockout and knockdown models revealed secreted phosphoprotein 1 (SPP1/OPN/Osteopontin) as a downstream target positively regulated by NRP2.

Keywords: bladder cancer, Neuropilin-2 (NRP2), glioma-associated oncogene family zinc finger 2 (GLI2), secreted phosphoprotein 1 (SPP1), osteopontin (OPN), epithelial-to-mesenchymal transition (EMT), RT112, J82

#### INTRODUCTION

Bladder cancer (BCa) is the 9th most common malignancy in the world with the highest incidence in Europe and North America (1). There are three main stages of this disease, the non-muscle invasive bladder cancer (NMIBC), the muscle invasive bladder cancer (MIBC), and the metastatic BCa. At diagnosis, 70% of the patients present with NMIBC, 20% with MIBC, and 10% with metastatic disease (2). While NMIBC can be treated with good outcome by transurothelial resection

**192**

of the bladder tumor (TURBT) and adjuvant intravesical Bacillus Calmette-Guérin (BCG) or chemotherapy, the treatment options for the more aggressive MIBC consist of neoadjuvant and adjuvant cisplatin treatment and radical cystectomy. Despite the aggressive therapy regimen, MIBC has a 50% risk to progress to metastatic disease. The average survival time in these patients is 14–15 months (3), and no curative treatment option is available for these patients. Only recently, immune checkpoint inhibition has become available in patients with metastatic disease. The success rate for this treatment is still uncertain. Nevertheless, new therapy options are urgently needed for this disease stage. Radiochemotherapy has emerged as a promising new option for improving locoregional control and being able to preserve the bladder and hence quality of life (4–6).

Our group previously demonstrated that Neuropilin-2 (NRP2) is a prognostic indicator for reduced survival in BCa patients. Together with its major ligand, vascular endothelial growth factor (VEGF)-C, NRP2 expression is capable of predicting treatment outcome in response to radiochemotherapy in BCa patients who underwent transurethral resection (7). NRP2 is a co-receptor frequently overexpressed in cancers. Because NRP2 expression is significantly associated with poor prognosis in renal cell carcinomas, colorectal carcinomas, gastric carcinomas, osteosarcoma, breast, pancreatic, and bladder cancer (7–13), it has become an attractive target for cancer therapy. This is in part due to the fact that NRP2 is implicated in signaling pathways commonly hijacked by tumor cells.

Hedgehog (Hh) signaling is silenced in many adult tissues; however, during tumorigenesis, it is often reactivated (14). Canonical Hh signaling is induced by sonic, indian, or desert Hh ligands and functions via glioma-associated oncogene family zinc finger (GLI) proteins, the major transcriptional effectors of Hh signaling (14). GLI proteins contain activation (GLI1, GLI2, and GLI3) and repression domains (GLI2 and GLI3) (15), thus differentially affecting their downstream target genes. Hh signaling can also be induced by non-canonical pathways including transforming growth factor (TGF)β-induced signaling (14, 16, 17). Non-canonical Hh signaling by TGFβ (and Wnt) was shown to induce GLI2 expression and activation (14). In another non-canonical pathway, NRP2 also directly enhances Hh signaling in a ligand-independent manner (18).

Epithelial-to-mesenchymal transition (EMT), a complex molecular process, plays an important role in tumor progression, invasion, and metastasis and is induced, among others, by TGFβ (19). EMT signaling is associated with therapy resistance in various tumor entities, including breast cancer (20), pancreatic cancer (21), and BCa (22). Interestingly, TGFβ1-induced EMT highly increased NRP2 protein levels and NRP2 was subsequently identified as a receptor for both the latent and active form of TGFβ1 (23).

Taken together, NRP2 supports a vast number of tumorpromoting events but seems to be less crucial in most healthy tissues, and thus, it has become an attractive target for anticancer therapy. In this report, we aimed to elucidate NRP2's role in TGFβ-mediated EMT as well as in radio(chemo)therapy treatment of BCa models.

#### RESULTS

### The Relationship of NRP2 and GLI2 in BCa

Because NRP2 has previously shown to enhance TGFβ signaling, we first aimed to determine the correlation of NRP2 mRNA expression with the expression of other TGFβ regulated genes in bladder tumors. To achieve this aim, we employed data from 408 BCa patients of the provisional BCa cohort from The Cancer Genome Atlas (TCGA) data set. The complete list of genes can be found in **Supplementary Tables 1A,B**. One of the most interesting identified targets was the Hh transcription factor GLI2 (r = 0.709). It was more strongly associated with NRP2 expression than its related genes GLI1 (r = 0.396) or GLI3 (r = 0.310) (**Figure 1A**). Notably, this relation was confirmed in other TCGA data sets of breast and prostate cancer (**Supplementary Figures 1A,B**). Furthermore, we confirmed this strong correlation between NRP2 and GLI2 transcripts by qPCR in a panel of 15 human BCa cell lines (**Figure 1B** and **Supplementary Figure 1C**) and by analysis of NRP2 and GLI2 co-expression in the cell lines of urinary tract (n = 26) using RNA-sequencing (RNA-seq) data from the Broad Institute Cell Line Encyclopedia (**Supplementary Figure 1D**). In order to investigate the potential clinical impact of NRP2 and GLI2 expression levels, we compared overall survival of single gene signatures of either NRP2 or GLI2 to the combined NRP2/GLI2 signature in Kaplan–Meier plots with median separation. This analysis demonstrated that combining NRP2 and GLI2 gene expression results in a higher predictive value for overall survival (**Figures 1C–E**). Notably, the same trend was observed for disease-free survival (**Supplementary Figure 2**). This observation, together with the strong correlation of both transcripts, tempted us to investigate the relationship of NRP2 and GLI2 in more detail by selecting two BCa cell lines, namely, J82 and HS853T, showing robust mRNA levels of both NRP2 and GLI2 (**Supplementary Figure 1C**) for knockdown experiments. To further evaluate the role of NRP2 in TGFβ-induced EMT, we treated these cell lines with TGFβ1 in addition to the respective knockdown. siRNAmediated knockdown of NRP2 resulted in a reduction of GLI2 expression in both cell lines. On the other hand, induction of NRP2 expression by TGFβ1 is impaired following GLI2 knockdown (**Figure 2**). This suggests a co-dependency of both targets based on the ligand initiating the downstream pathways. Notably, we also checked the expression of isoforms NRP2a and NRP2b as well as GLI1, a direct target gene of GLI2 (**Supplementary Figures 3**, **4**). As expected, GLI1 expression was also induced in response to TGFβ1 but to a lesser extent than GLI2. Accordingly, GLI1 levels were reduced following GLI2 knockdown. Isoforms NRP2a and NRP2b were induced similarly in TGFβ1-treated samples and GLI2 knockdown led to a shift of these isoforms in favor of NRP2b. A complete list of all p values for all targets and samples is provided in **Supplementary Table 2**.

In addition to the NPR2 knockdown in these cell lines, we created two NRP2 knockout clones from the cell line RT112. Wild-type (WT) and knockout (KO) cells were subjected to

treatment with TGFβ1 and compared to untreated controls. KO cell lines may still produce NRP2 transcript but the resulting mRNA contains premature translational stop codons on all alleles. While WT cells significantly increased the level of NRP2 mRNA in response to TGFβ1, both KO clones failed to upregulate transcription, potentially hinting to a positive feedback loop of NRP2 enhancing its own transcription upon TGFβ-signaling (**Figure 3A**). Moreover, TGFβ1 highly induced GLI2 in both WT and KO cell lines, suggesting that NRP2 is not upstream of TGFβ1-mediated GLI2 regulation. The mRNA level of GLI2 in WT cells was comparable to both KOs in the untreated state. In the treated samples, TGFβ1 induced GLI2 transcription more prominently in KO cell lines. However, this difference was not significant. Hence, TGFβ1-induced GLI2 expression seems to be independent of NRP2 in this model or cell line.

The fact that NRP2 is induced by more than 5-fold in WT cells raised the question whether upregulation is a direct effect of TGFβ signaling or TGFβ1 leads to faster degradation of NRP2, which may prompt cells to upregulate its transcription for maintaining constant NRP2 protein levels. Therefore, we performed Western blot analysis of WT lysates, which indicated that NRP2 protein was not upregulated significantly in TGFβ1-treated samples compared to untreated samples (**Figures 3B,C** and **Supplementary Figure 5A**). Despite the minor increase on protein level, it is not comparable to the 5-fold upregulation of NRP2 transcript, suggesting that the effect on mRNA level may potentially be a compensatory mechanism.

### Knockout of NRP2 Alters Gene Expression of EMT Regulators

To investigate how NRP2 may enhance TGFβ-signaling, cDNA of WT and KO cell lines was analyzed by a PCR array covering 84 genes involved in EMT. By addition of TGFβ1, EMT was successfully and similarly induced in both KO clones and their wild-type parental cell line RT112 as visible by an increase in the EMT marker vimentin (**Supplementary Figure 5B**). With this approach, it was possible to identify four genes whose expression was altered in both KOs compared to WT without TGFβ1 treatment. When all cell lines received TGFβ1 treatment, one gene was found to be deregulated in KOs vs. WT cells (**Figure 4**). Validation of these targets by qPCR in four biological repeats demonstrated that upregulation of Caldesmon 1 (CALD1) and Cadherin 2 (CDH2, N-Cadherin) was not significant but Secreted Phosphoprotein 1 (SPP1) and Six Transmembrane Epithelial

FIGURE 2 | Validation of the relationship of NRP2 and GLI2 in BCa cells after NRP2 and GLI2 knockdown. Quantitative real-time PCR in two cell lines: J82 (A) and HS853T (B) with robust expression of NRP2 and GLI2 were subjected to knockdown of these gene products (siNRP2 or siGLI2) or scrambled control (siSCR) and treated with 5 ng/ml TGFβ1 or left untreated (±). All transcripts were induced by TGFβ1 treatment. GLI2 levels were reduced after NRP2 knockdown while NRP2 induction by TGFβ1 is inhibited following GLI2 knockdown. Normalized to housekeeping gene HPRT1 and plotted relative to untreated siSCR sample. Significance calculated by two-way ANOVA. Error bars indicate standard error of the mean. n = 3. Not all p values are shown. A plot of all targets including NRP2a, NRP2b, and GLI1 is provided in Supplementary Figure 3. The figure also contains additional graphs with normalization against two other housekeeping genes ACTB and GAPDH. All p values for all cell lines are provided in Supplementary Table 2. (C) Western blot analysis of NRP2 and GLI2 expression in HS853T cells in response to GLI2 or NRP2 knockdown. Cells were transfected with gene-specific siRNA (siNRP2 or siGLI2) or siSCR and treated with 5 ng/ml TGFβ1 or left untreated (±). Relative protein expression was normalized to GAPDH. Error bars indicate standard deviation. n = 2.

treatment of two independent NRP2 knockout clones (KO #1 and KO #2) and their parental wild-type BCa cell line RT112 (WT). Untreated samples were used as control. NRP2 transcripts are highly induced by TGFβ1 only when NRP2 protein is expressed. Normalized to housekeeping gene HPRT1. Significance calculated by two-way ANOVA. Error bars indicate standard error of the mean. n = 4. (B) Western blot of WT cell line for NRP2 and α-tubulin as loading control. (C) Calculation of optical densitometry. Significance determined by two-tailed, unpaired Student's t test. n = 4.

Antigen of Prostate Family Member 1 (STEAP1) mRNA levels were significantly downregulated in KO clones (**Figure 5A**). When cells were treated with TGFβ1, mRNA expression of Secreted Protein Acidic and Cysteine Rich (SPARC) was highly upregulated but remained non-significant (**Figure 5B**). As an example, SPP1 expression as determined by the human EMT


FIGURE 4 | Analysis of the expression levels of EMT-related genes in wild type and NRP2 knockout BCa cells using PCR gene expression array. (A) Clustergram for all samples and genes on the PCR array plate extracted from Qiagen data analysis center software. (B) Consistently deregulated genes in both KOs vs. WT samples and their correlation based on the TCGA data set. n = 2.

FIGURE 5 | Validation of the PCR array results. Target validation by qPCR in two NRP2 knockout clones (KO #1, KO #2) compared to their parental wild-type cell line RT112 (WT) (A) without TGFβ1 treatment or (B) including TGFβ1 treatment. Genes SPP1 and STEAP1 were significantly affected by NRP2 knockout in both knockout clones. For other targets, data were either inconsistent or not significant. Data were normalized to housekeeping gene HPRT1. Significance calculated by two-way ANOVA. Error bars indicate standard error of the mean. n = 4. (C) Western blot of samples that remained untreated for proteins NRP2, N-Cadherin, and E-Cadherin. α-Tubulin served as loading control. Untreated KO cells demonstrated increased N-Cadherin expression. (D) Western blot of TGFβ1-treated KO clones did not show a significantly upregulated N-Cadherin expression compared to parental cells but revealed downregulated E-Cadherin on protein level. Significance calculated by two-way ANOVA. Error bars indicate standard error of the mean. n = 3.

**196**

PCR array is lower in both KO clones compared to the wild-type parental cell line (**Supplementary Figure 6**).

CDH2 was the only target that demonstrated robust mRNA expression in both conditions. Therefore, validation on protein level seemed promising only for this target. Western blot analysis demonstrated that N-Cadherin (gene CDH2) appeared to be upregulated following knockout of NRP2 without TGFβ1 treatment (**Figure 5C**). Since E-Cadherin is known to be an opposing player of N-Cadherin in EMT, we used this target as control. Surprisingly, lysates from TGFβ1 treated samples showed significantly decreased levels of E-Cadherin in KOs compared to WT despite no change in gene expression was detected by the PCR array (**Figure 5D**). qPCR of CDH1 (E-Cadherin) confirmed that this change did not arise from altered transcript levels (**Supplementary Figure 7**). Because EMT-related signaling pathways are involved in the regulation of cancer stem cell (CSC) phenotype and properties in urothelial carcinoma including BCa, we analyzed if the absence of NRP2 has an impact on the CSC-related properties (24). Sphere forming assays revealed that the number of spheres formed by KO cells was significantly reduced (**Supplementary Figure 8A**).

To confirm the aberrant EMT signature in an additional cell line, we used conventional siRNA-mediated knockdown of NPR2 in the BCa cell line J82 and HS853T (**Figures 6A,B** and **Supplementary Figure 8B**). The results show that all except one gene (STEAP1) were significantly altered by NRP2 knockdown (**Figures 6A,B**). Analysis of BCa TCGA dataset also revealed that all validated genes positively correlate with expression of both NRP2 and GLI2 genes (**Supplementary Figure 8C**). However, only one gene transcript was regulated in a similar manner to NRP2 knockout in RT112 as well as NRP2 knockdown in J82 (**Figure 6C**). The SPP1 gene (Osteopontin, OPN) was previously reported to be induced by VEGF (25) and to be associated with decreased survival, disease stage, and grading in BCa (26, 27). Previous findings support the role of SPP1 as one of the key EMT regulators (28). We applied an EMT PCR array to analyze if SPP1 regulates gene expression of EMT regulators in our cell models. We found that SPP1 knockdown in J82 cells decreased expression of a number of key EMT genes including SNAI1, COL1A2, FGFBP1, and STAT3 (**Figure 6D**). In the TCGA BCa cohort, SPP1 expression positively correlated with both NRP2 and GLI2 (**Figure 6E**). Although our data did not confirm that SPP1 might be a regulator of BCa radiosensitivity on its own, analysis of TCGA BCa dataset showed that combined NRP2/SPP1 signature improved predictive value for diseasefree but not overall survival compared to single NRP2 gene expression (**Supplementary Figures 8D,E**, **9**, **10**).

#### The Relevance of NRP2 to Treatment With Radiochemotherapy

The standard curative treatment of BCa is surgery and chemotherapy. Only for progressed stages of disease

#### FIGURE 6 | Identification of SPP1 as one of the NRP2-regulated and EMT-associated genes. qPCR of all identified targets following NRP2 knockdown in cell line J82 (A) excluding or (B) including TGFβ1 treatment. Significance calculated by two-way ANOVA. Error bars indicate standard error of the mean. n = 3. (C) Comparison of deregulated genes by NRP2 knockout in RT112 or NRP2 knockdown in J82. SPP1 is the only gene reacting the same way to both depletions in two different cell lines. ns = not significant. (D) Analysis of J82 cells transfected with siSCR (three pooled biological repeats) and J82 cells transfected with siSPP1 (three pooled biological repeats) with EMT PCR array. (E) Correlation of NRP2 with SPP1 and GLI2 with SPP1 in a provisional bladder cancer cohort of The Cancer Genome Atlas (TCGA).

other therapy options like radiochemotherapy and immunotherapy gained importance. Based on the fact that NRP2 and its ligand VEGF-C predicted treatment response to radiochemotherapy in patients (7), we analyzed the response of our KO and WT cell lines to radiotherapy and combined radiochemotherapy with cisplatin (**Figures 7A,B** and **Supplementary Figure 11**). Significances for plating efficacy are shown in **Supplementary Figure 12**, and the alpha–beta ratio defined from interpolation of linear–quadratic cell survival curves are shown in **Supplementary Figures 13A,B**. Our results indicate that there was no immediately visible effect between both KOs and their parental WT cell line. All cell lines responded to additional cisplatin treatment with significantly reduced clonogenic survival and plating efficacy. To identify the potential benefit of chemotherapy in addition to radiation treatment, we calculated the radiobiological enhancement ratio (RER) for each subset. The RER is an indicator of the radiosensitizing effect of any potential agent used as it compares the ratio of the surviving fraction from the radiation only to the radiation in combination with any agent at a specific dose. Analysis of RER showed a higher benefit of radiochemotherapy for KO cells than for WT cells (**Figure 7C** and **Supplementary Figure 13C**).

#### DISCUSSION

### NRP2 and GLI2 Interplay Is Dependent on the Ligand

The close relationship of NRP2 to GLI2 was discovered by correlation of genes in a provisional TCGA BCa data set (**Supplementary Table 1A**). Interestingly, this correlation of GLI2 and NRP2 was even stronger (r = 0.709) than the correlation of GLI2 to either GLI1 (r = 0.555) or GLI3 (r = 0.252). This is surprising because GLI1 is known to be a direct target gene of GLI2 in both the canonical and non-canonical Hh pathway (14, 16, 29, 30) (**Supplementary Table 1B**). We analyzed GLI1 levels for additional functional verification of GLI2 knockdown and could confirm that GLI1 levels were reduced following knockdown with a GLI2-specific siRNA pool. Since GLI1 levels are directly dependent on the expression of GLI2, we cannot fully exclude the possibility that changes in NRP2 could also be mediated through GLI1 or other downstream targets of GLI2 despite the fact that correlation between NRP2 and GLI2 is by far highest compared to GLI1 or GLI3 in a patient cohort. Whether or not this interdependency of the two gene products is a direct or indirect effect was demonstrated in two human BCa cell lines that GLI2 levels are regulated in

FIGURE 7 | NRP2 regulates BCa radiochemosensitivity. (A) Surviving fraction of cell line RT112 either harboring (WT—black) or lacking (KO #1—light gray, KO #2—dark gray) endogenous NRP2 after radiotherapy treatment with doses of 0, 2, 4, 6, and 8 Gy (dots). All cell lines were additionally treated with the chemotherapeutic drug cisplatin (Cispl. IC5 for KO #2 = 1.52µM) for 24 h prior to radiation (squares). Plotted lines were fitted using the linear quadratic model for both the radiochemotherapy group (dashed lines) and the radiation only group (solid lines). Error bars indicate standard error of the mean. n = 3. (B) Table of significance of fitted curves. Significance was calculated by SPSS software. –, without cisplatin (Cispl.); +, with cisplatin. (C) Radiation enhancement ratios (RERs) were calculated at doses 2, 4, 6, and 8 Gy from interpolation of linear-quadratic cell survival curves using mean values of three independent experiments. Relative RER values for KO #1 and KO #2 were calculated as RERrel = Average ([RER KO, 2Gy]/[RER WT, 2Gy];[RER KO, 4Gy]/[RER WT, 4Gy];[RER KO, 6Gy]/[RER WT, 6Gy];[RER KO, 8Gy]/[RER WT, 8Gy]). Error bars indicate standard error of the mean. n = 3.

part through NRP2 and that NRP2 mRNA levels are partially regulated by GLI2 signaling in a TGFβ1-dependent manner. In contrast to cell lines J82 and HS853T, another cell line (5637) showed no changes in the expression of these genes possibly because TGFβ1 failed to induce their expression. Consequently, the relationship of NRP2 and GLI2 could not be observed in this cell line (**Supplementary Figure 14**). Because prediction of overall and disease-free survival could be improved by combining NRP2 and GLI2 gene expression in the TCGA patient cohort, these results suggest a functional interplay between NRP2 and GLI2 in regulating tumor growth, although the mechanisms of this interplay still remain elusive and merit further investigation. Notably, GLI2 knockdown also changed the ratio of isoforms NRP2a and NRP2b in favor of the latter in two BCa cell lines. It was recently described in a lung cancer model that while NRP2a is almost dispensable for tumor formation and metastasis, NRP2b severely impacted these traits (31). The authors demonstrated that TGFβ1 predominantly upregulated NRP2b and that TGFβ1-dependent stabilization was specifically dedicated to isoform NRP2b. In our model, both NRP2 isoforms were equally increased on mRNA level, but this induction might be the result of compensating for increased protein degradation following TGFβ1 treatment given that the protein level of total NRP2 was not significantly increased.

### NRP2 Positively Regulates Osteopontin Expression

The active ligand TGFβ1 is a potent inducer of EMT. For investigating how NRP2 might enhance EMT, we chose a qPCR array containing 84 genes involved in EMT signaling and checked cDNA from two NRP2 knockout cell lines and their parental wild-type cell line RT112. Using this approach, a number of genes were deregulated in both KO clones compared to the parental cell line irrespective of the housekeeping gene used. Validation of these genes in four independent biological repeats confirmed their altered expression, although the change for CALD1, CDH2, and SPARC1 was not significant. Because the expression of CDH2 mRNA was sufficient to detect this target by Western blot, we checked the expression of the corresponding protein N-Cadherin as well as its counterpart E-Cadherin as control in treated and untreated conditions. Although N-Cadherin expression doubled in KO clones without TGFβ1-treatment, only one KO clone showed a significant increase. However, when TGFβ1 treatment was applied, we detected a significant reduction in both KO clones, although the change was less dramatic than for N-Cadherin in the untreated condition. There have been numerous reports in the past stating the significance of cadherin switching in progression and malignancy of BCa pointing to the importance of elucidating the mechanism for targeted therapy (32–37). Moreover, it was already shown that NRP2 and E-Cadherin expression are connected in multiple cancer types (23, 38–41). However, unlike the positive correlation in our model system, all publications reported a negative correlation. It has previously been described for melanoma that Osteonectin (gene SPARC) can downregulate E-Cadherin (42). We could show a 4.5-fold (KO #2) to 9.8-fold (KO #1) upregulation of SPARC compared to WT when treated with TGFβ1. Although this change was not found to be significant, it was still the highest deregulated gene within the panel. Therefore, the reduced E-Cadherin (CDH1) expression might be a direct result of increased SPARC expression. Given that the authors of the cited paper only investigated E-Cadherin by Western blot but not by qPCR and the fact that the change of E-Cadherin in our model was only visible on protein level suggests that SPARC might not control E-Cadherin transcriptionally. Previous studies showed that SPARC induces β-catenin nuclear localization and binding to the transcriptional regulator lymphocyte-enhancer factor-1 (LEF-1) (43, 44), whereas E-Cadherin forms alternative complexes with β-catenin in the adherens junctions (AJ). These AJ complexes prevent β-catenin nuclear localization and transactivation as well as E-Cadherin internalization (45). We can hypothesize that SPARC can induce loss of β-Catenin in the AJ by triggering its nuclear translocation that might result in E-Cadherin endocytosis and degradation. Of note, protein levels of N-Cadherin (CDH2) after TGFβ1-treatment did not change dramatically anymore, potentially indicating that this change is independent of the ligand TGFβ1.

In order to investigate if the identified targets were specific to that cell line or if signaling pathways were deregulated for compensation of complete NRP2 loss in the knockout cells, we used siRNA-mediated depletion of NRP2 in another human cell line (J82). The results confirmed deregulation of all but one gene (STEAP1), but the direction of change was only consistent for one gene (SPP1/Osteopontin/OPN). Thus, changes in other genes are either cell line specific or long-term KO models adapt to missing NRP2 by deregulation of other EMT pathways that were initially downregulated in the short-term knockdown model (for example, CALD1, CDH2, and SPARC). But given that Osteopontin was the only target significantly downregulated in both NRP2 KO and knockdown models in different cell lines, we propose that this dependency might be a general mechanism. To our knowledge, this is the first report linking NRP2 and SPP1/OPN in any tissue or cancer entity by showing that NRP2 acts upstream of SPP1 in a TGFβ1-independent manner. This can be an explanation for the VEGF-induced OPN expression, which was demonstrated in a large number of cases (25). OPN was shown to be upregulated in multiple cancer types including breast and prostate cancer as well as glioblastoma and melanoma (46, 47). Regarding BCa, immunohistochemical staining of OPN demonstrated significant correlation with tumor stage (27). More recently, Wong and colleagues showed that OPN expression correlates with disease stage and grading and that higher OPN expression led to decreased survival in multiple patient cohorts (26). Of note, we could not see the same in our TCGA data set when applying median expression for overall and diseasefree survival (**Supplementary Figure 9**). However, combining NRP2 and OPN expression slightly improved prediction of disease-free but not overall survival (**Supplementary Figure 10**). Since OPN is a secreted soluble molecule, it may serve as an attractive non-invasive prognostic marker in serum or urine. One study investigated plasma OPN levels before and after tumor resection in 50 patients with BCa and found significantly higher preoperative OPN levels in patients with muscle invasive tumors despite the relatively low number of patients. OPN levels also increased significantly with T stage when patients had undergone radical cystectomy. The strong trend of correlation with tumor grade and predicting recurrence did not reach statistical significance, potentially indicating that patient number for this analysis was too low (48). Similar to OPN, the expression pattern of NRP2 was previously reported to be significantly associated with pathological stage and tumor grade in a BCa cohort, suggesting a prominent role of NRP2 in BCa progression (49).

Because OPN is also associated with bone matrix formation, it would be highly interesting to analyze if the connection between NRP2 and OPN is also true for cancer entities like breast and prostate cancer, where bone metastases remain a big challenge.

#### NRP2 as Target for Radiochemotherapy

Overall and disease-free survival data suggested NRP2 as prognostic indicator for a TCGA BCa cohort. Our group previously showed that expression of NRP2 as well as its ligand VEGF-C could predict treatment outcome of BCa patients following TURBT and radiochemotherapy. This clinical finding prompted us to use our NRP2 KO and WT cell lines for an in vitro assay to determine their clonogenicity after radio(chemo)therapeutic treatment. When looking at the radiotherapy group only, no significant differences between both KO and the WT cell lines were found, which confirmed the previous finding that NRP2 expression alone in the patient group receiving only local radiotherapy was not a prognostic factor (7). When the same analysis was applied to the group of patients, which received radiochemotherapy, NRP2 was highly prognostic for overall and cancer-specific survival. In accordance to this clinical observation, we revealed a radiation dose-dependent trend toward higher profit of additional chemotherapy in KO cells, suggesting that NRP2 downregulation results in BCa radiochemosensitization. The exact role of NRP2 for radiochemotherapy in BCa warrants further investigation using additional cell lines and animal models.

Taken together, our study demonstrated that mRNA expression of NRP2 and GLI2 highly correlate in BCa cell lines and the TCGA BCa cohort. They influence each other's expression depending on the presence or absence of TGFβ1, a potent inducer of EMT. Moreover, screening of 84 genes involved in EMT identified SPP1/Osteopontin as a downstream target of NRP2 in two different BCa cell lines using different model systems.

Future research is needed to evaluate the exact mechanism of how NRP2 and GLI2 communicate bidirectionally, how NRP2 modulates SPP1 transcription, and what implications this will have for development and progression of other cancer entities apart from bladder carcinoma.

### MATERIALS AND METHODS

#### Cell Lines and Cell Culture

The cell line RT112 (DSMZ) was maintained in MEM alpha medium with GlutaMAX supplemented with 10% FBS (fetal bovine serum, both Gibco, Life Technologies, Waltham, USA). The cell line and their knockout derivatives were validated to be RT112 by single nucleotide polymorphism (SNP) profiling (performed by Multiplexion GmbH, Friedrichshafen, Germany). Cell line J82 (ATCC) and HS853T (ATCC) were cultured in DMEM medium (4.5 g/L glucose) supplemented with 10% FBS (both Gibco, Life Technologies, Waltham, USA), 1% HEPES solution, and 1% MEM non-essential amino acids (both Sigma-Aldrich, St. Louis, USA). The cell line 5637 (ATCC) was cultured in RPMI supplemented with 10% FBS (both Gibco, Life Technologies, Waltham, USA). All cell lines were subject to regular testing for excluding mycoplasma contamination (last on 2nd April 2019). All cells were maintained at standard conditions in a humidified incubator with 5% CO<sup>2</sup> at 37◦C.

### CRISPR/Cas9-Mediated Deletion of NRP2 in RT112 Cells

Plasmid-based CRISPR/Cas9 technology was used to generate RT112 cell clones deficient in NRP2 expression according to the protocol established by Ran et al. (50). In brief, DNA double stranded oligonucleotides located in the exon 3 splice acceptor region (NRP2e3gu1-top: 5′ -CACCGGATAAAGTCAT ACCTGGGTG-3′ , NRP2e3gu1-bottom: 5′ -AAACCACCCAGG TATGACTTTATCC-3′ ) and exon 3 coding region (NRP2e3gu2 top sequence: 5′ -CCACCGGGTGAACTTGATGTAGAGCA-3′ ; NRP2e3gu1-bottom sequence: 5′ -AAACTGCTCTACATCAAG TTCACCC-3′ ) of the NRP2 locus were designed using the Benchling software (San Francisco, USA) and cloned into pSpCas9 BB2A-GFP for gu1 (PX458, Addgene, LGC Standards, UK) or pSpCas9 BB-2A-Puro for gu2 (PX459v2, Addgene). RT112 cells were transfected by calcium phosphate precipitation with a 1:1 ratio of PX458-NRP2e3gu1 and PX459v2-NRP2e3 gu2 for 8 h, kept under puromycin selection (0.5µg/ml) for 24 h and then seeded at low density. One hundred clones were picked after two weeks, grown and checked for the desired 110 bp deletion in NRP2 exon 3 by PCR on isolated genomic DNA using primers 5′ -AGTGCCCTTCGCTTATCCATC-3′ and 5′ -TC TAAGACGCCCATCTCCCG-3′ . Clones that carried the deletion in the NRP2 locus were further checked for mutations within the corresponding region of the NRP1 locus (primer sequences 5 ′ -GCTGGATGATGCTGGTGTCTA-3′ and 5′ -TTCTACCGTA AGCTGTTCACTC-3′ ) and for Cas9 (primer sequences 5′ -CG ACGACAGCCTGACCTTTA-3′ and 5′ -TTGATGCCCTCTTC GATCCG-3′ ) to exclude integration of transfected plasmids. Sanger sequencing of the PCR amplification products verified the deletions. TOPO cloning of DNRP2 PCR products and subsequent Sanger sequencing of single mutated alleles yielded the sequences of individual deletion alleles for NRP2 in two individual RT112 DNRP2 cell clones. Three independent mutant alleles were identified for both RT112 DNRP2 clones, with altered exon 3 splice acceptor sequences and/or introduction of frame shifts due to nucleotide insertions or deletions:

(clone: mutant alleles; mutation; location of mutation in NRP2 exon 3)

#J9 (KO#1): J9high-3 delAG; delCT delAG (−2;−1); delCT (107; 108)


#J9 (KO#2): #J32low-1 del110 deletion: −3 to +107 #J9 (KO#1): #J32-P2 del111 deletion: −4 to +107

Both RT112 DNRP2 cell clones were checked for the absence of NRP2 expression by Western blot immunostaining using an anti-NRP2 antibody (R&D Systems, Minneapolis, USA). For reasons of simplicity, the clones are referred to as knockout (KO) #1 (clone #J9) and KO #2 (clone #J32) in this publication.

### TGFβ1-Induced EMT

RT112 WT and derived KO cell lines were seeded at 1 × 10<sup>5</sup> cells per well in a 6-well culture plate containing 2 ml of serumreduced (5 % FBS) growth medium either including or lacking 5 ng/ml TGFβ1 (Miltenyi Biotec, Bergisch Gladbach, Germany). After 48 h in the incubator, the medium was renewed and cells were incubated for another 24 h before RNA or protein isolation. In total, four independent biological repeats were performed with cells at different passages. RNA of two repeats was used for "RT<sup>2</sup> Profiler PCR Array" for human EMT and all four repeats were used for validation of identified targets by qPCR. Protein was used for immunoblotting.

#### Transfection of Cell Lines With siRNA

Cell lines J82, HS853T, and 5637 were used for knockdown experiments. 2 × 10<sup>5</sup> (J82, HS853) or 5 × 10<sup>5</sup> (5,637) cells per well were seeded in a 6-well culture plate with 2 ml complete growth medium (10% FBS) and incubated 24 h to allow attachment. Next, medium was renewed and liposomal transfection was conducted according to the manufacturer's protocol using 12 µl of Lipofectamine RNAi MAX (Thermo Fisher Scientific, Waltham, USA) and 50 nM siRNA (SMARTpool by Dharmacon, Lafayette, USA). The catalog number of siRNA pools was D-001810-10-20 (siSCR, control), L-017721-00-0010 (siNRP2), L-0066468-00-0005 (siGLI2), and L-012558-00-0005 (SPP1). Cells were treated with TGFβ1 as described above.

#### RNA Isolation, cDNA Synthesis, and qPCR

RNA was isolated with the RNeasy Plus Mini Kit (Qiagen, Hilden, Germany) by adding 350 µl of lysis buffer RLT Plus supplemented with 1% β-mercaptoethanol (Sigma-Aldrich, St. Louis, USA) to a well of a 6-well plate that was previously rinsed with PBS. Using a scraper, lysed cells were collected from the plate and transferred to a DNA removal column. The following steps were carried out according to the manufacturer's protocol and RNA was eluted from the column by addition of 30 µl of RNase-free water (Qiagen, Hilden, Germany). cDNA from 1,000 ng of total RNA input was synthesized by employing the PrimeScriptTM RT Reagent Kit (Takara Bio Inc., Kusatsu, Japan) according to the manufacturer's instructions. cDNA was diluted 1:5 with RNase-free water before continuing with realtime quantitative polymerase chain reaction (qPCR). qPCR was conducted using the TB GreenTM Premix Ex TaqTM II (Takara Bio Inc., Kusatsu, Japan) according to the manufacturer's protocol for a total reaction volume of 20 µl. The qPCR cycling conditions were set on a StepOnePlus system (Applied Biosystems, Waltham, USA): 94◦C for 3 min, 40 cycles: 94◦C for 15 s, 58◦C 60 s, 72◦C 60 s followed by a melt curve to 95◦C in steps of 0.3◦C. All experiments were conducted using at least two (for housekeeping genes) or three technical replicates (other targets) and most experiments included three different housekeeping genes as control: ACTB, GAPDH, and HPRT1. All primer sequences are listed in **Supplementary Table 3**. cDNA of the following BCa cell lines was used for qPCR analyses of NRP2 and GLI2: 5637 (ATCC), 639V (DSMZ), Cal29 (DSMZ), EJ28 (University Frankfurt), HS853T (ATCC), HT1376 (ATCC), J82 (ATCC), UMUC-3 (ATCC), UMUC-14 (Sigma-Aldrich), UMUC-16 (Sigma-Aldrich), VMCUB (DSMZ), KU1919 (DSMZ), RT112 (DSMZ), T24 (DSMZ), and TCC-SUP (DSMZ). HPRT1 served as housekeeping gene. 1Ct values were used to calculate correlation in SUMO software (http:// angiogenesis.dkfz.de/oncoexpress/software/sumo/).

## RT<sup>2</sup> Profiler PCR Array for Human EMT

cDNA synthesis was performed following RNA isolation as described above (total volume: 10 µl). For each 96-well plate, a master mix was prepared composed of 1,050 µl of TB GreenTM Premix Ex TaqTM II, 42 µl of ROX dye, 1,000 µl of RNasefree water, and 10 µl of cDNA sample. From this master mix, 20 µl was added to each well. The PCR program was identical to the one described above. For the human EMT PCR array (Qiagen, Hilden, Germany), two out of four biological replicates were chosen that displayed values closest to the median. Later, all four biological replicates were evaluated for target validation. Data were extracted using the housekeeping gene HPRT1 from the EMT profiler plate.

#### Protein Isolation and Immunoblotting

Protein was isolated from 1-well of a 6-well plate by washing cells once with PBS before adding cold 200-µl RIPA buffer (Thermo Fisher Scientific, Waltham, USA) supplemented with Complete inhibitor (Roche, Basel, Switzerland), proteinase, and phosphatase inhibitor (both Thermo Fisher Scientific, Waltham, USA). Lysates were collected with cell scrapers, transferred to a 1.5-ml reaction tube and centrifuged at 10,000 rpm at 4◦C for 10 min in a 5415R cooling centrifuge (Eppendorf, Hamburg, Germany). Supernatant was transferred to a new reaction tube and protein was quantified by PierceTM BCA Protein Assay Kit (Thermo Fisher Scientific, Waltham, USA) according to the manufacturer's protocol. Thirty-five micrograms of total protein lysate was loaded into each pocket of a 12-well Bolt 4–12% Bis-Tris Plus Gel (Thermo Fisher Scientific, Waltham, USA) running in an XCell SureLockTM Electrophoresis Cell (Invitrogen, Carlsbad, USA) at 90 V for 2 h in MOPS buffer (Thermo Fisher Scientific, Waltham, USA). Wet transfer to a methanol-activated 0.2-µm AmershamTM HybondTM Low Fluorescence PVDF membrane (GE Healthcare, Chicago, USA) was achieved by applying 90 V for 4 h in a cooled Mini-PROTEAN <sup>R</sup> three transfer tank (Bio-Rad, Hercules, USA). Membranes were washed once with TBS-T before blocking the membrane for 1 h at room temperature with 2.5% ECL PrimeTM blocking agent (GE Healthcare, Chicago, USA) dissolved in TBS. Membranes were incubated at 4◦C overnight with primary antibodies (see **Supplementary Table 4** for the complete list of used antibodies) diluted in 2.5% ECL PrimeTM blocking agent solution before applying three washing steps with TBS-T. Then, membranes were incubated with the appropriate secondary antibodies diluted in 2.5% ECL PrimeTM blocking agent solution for 1 h at room temperature. Following another three washing steps with TBS-T, chemiluminescent detection was performed by first incubating the membrane with Pierce <sup>R</sup> ECL Western Blotting Substrate (Thermo Fisher Scientific, Waltham, USA) and subsequent detection of the signal in auto-rapid mode in a ChemiDocTM MP Imaging System (Bio-Rad, Hercules, USA). Colorimetric images were taken for determining the molecular weight of the signals. Calculation of optical densitometry was performed with ImageJ software.

### Colony Forming Assay Following Radiochemotherapy

For radiochemotherapy treatment, cells were seeded into a 6 well plate at a density of 4 × 10<sup>5</sup> cells per well and treated with cisplatin (TEVA GmbH, Ulm, Germany) for 24 h at concentrations of 1.52 × 10−<sup>6</sup> M that corresponded to the IC<sup>5</sup> value of KO #2 (**Supplementary Figure 11**). Twenty-four hours after start of the treatment, cells were trypsinized and used for radiobiological colony forming assay. Clonogenic survival was determined by seeding of 500 (ionizing radiation—IR only) or 750 (IR + cisplatin) cells in technical triplicates into 6-well plates containing 2 ml of complete growth medium. Cells were cultivated overnight and then irradiated with doses of 0, 2, 4, 6, and 8 Gy (Yxlon Y.TU 320; 200 kV X-rays, dose rate 1.3 Gy/min at 20 mA, filtered with 0.5 mm Cu). Irradiated cells were returned to the incubator for allowing recovery and growth for 6 days. Colonies were fixed by the addition of 600 µl of 37% formaldehyde solution (Merck, Darmstadt, Germany) directly to the culture medium and incubation at room temperature for 30 min. Following removal of this solution and a washing step with normal tap water, 1 ml of a 0.05% crystal violet solution (Sigma-Aldrich, St. Louis, USA) was added to each well for 30 min at room temperature for staining colonies. Crystal violet was removed, and wells were washed twice with normal tap water and dried overnight before manually counting colonies using a stereomicroscope (Zeiss, Oberkochen, Germany). Cell survival data were entered into the SPSS program for calculation of α and β values, curve fitting to the linear quadratic model, and determination of statistical significance.

### Analysis of the TCGA Patient Cohort Data

From the TCGA patient cohort data set, Pearson coefficient was determined using SUMO software and significance was calculated by two-tailed, unpaired Student's t test. For evaluation of combined NRP2/GLI2 signature expression, the data for each of these genes were normalized to median across the entire dataset and log2 transformed. Further, the mean of two genes was calculated for each patient and the subset with known survival data was extracted from complete cohort. Finally, the up- and down-regulated groups for Kaplan–Meier analysis were defined as the mean of NRP2/GLI2 expression was positive or negative value accordingly.

#### Statistical Analysis

The cell survival curves were analyzed using the Statistical Package for the Social Sciences (SPSS) v23 software as described previously (51) by linear–quadratic formula S(D)/S(0) = exp−(αD + βD2) using stratified linear regression. A p < 0.05 was considered statistically significant. Correlation of gene expression levels was evaluated by SUMO software using Pearson correlation coefficient. IC<sup>50</sup> and IC<sup>5</sup> values (50 and 5% inhibitory concentration) were determined by non-linear regression using GraphPad Prism software (San Diego, USA).

### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

### AUTHOR CONTRIBUTIONS

AS performed most experiments and wrote and edited the manuscript. IG performed revision experiments and edited the manuscript. RB helped to teach CRISPR/Cas9 technology and designed the guide RNAs. SFu and KE provided cell lines and cDNA of cell lines and edited the manuscript. SFo and KD edited the manuscript. TM generated and validated the knockout clones from cell line RT112. TM, AD and MM supervised and guided AS, edited the manuscript, and aided in scientific questions including experimental design. AD helped perform radiochemotherapy and colony forming assay.

### FUNDING

Work in AD lab was partially supported by grants from Deutsche Forschungsgemeinschaft (DFG) (273676790, 401326337, and 416001651), from Wilhelm Sander-Stiftung (2017.106.1), BMBF (Grant-No. 03Z1NN11), and DLR Project Management Agency (01DK17047). This work was in part supported by German Academic Exchange Service (DAAD) and the Federation of European Biochemical Societies (FEBS) for IG. Work in MM lab was partially supported by a grant from Deutsche Forschungsgemeinschaft (DFG) (273676790). The professorship of MM was also funded by the Rudolf Becker Foundation of Translational Prostate Cancer Research. MM was a member of the German BRIDGE consortium. This work was in part supported by the graduate academy of Technische Universität Dresden for AS.

### ACKNOWLEDGMENTS

The authors kindly want to thank Vasyl Lukiyanchuk for analysis of TCGA datasets and for help with analysis of PCR array data and SPSS statistical analysis. We thank Jakob Püschel for help with radiobiological clonogenic analysis. We also thank Dorothee Pfitzmann for technical advice and providing cisplatin. The authors thank all members of AD and MM lab for useful discussions.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2019.01461/full#Supplementary-material

#### REFERENCES


migration, invasion, and resistance to paclitaxel. Cancer Res. (2007) 67:1979– 87. doi: 10.1158/0008-5472.CAN-06-1479


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Schulz, Gorodetska, Behrendt, Fuessel, Erdmann, Foerster, Datta, Mayr, Dubrovska and Muders. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Role of Cancer Stem Cells in Radiation Resistance

Christoph Reinhold Arnold<sup>1</sup> \*, Julian Mangesius <sup>1</sup> , Ira-Ida Skvortsova1,2 and Ute Ganswindt <sup>1</sup>

*<sup>1</sup> Department of Therapeutic Radiology and Oncology, Medical University of Innsbruck, Innsbruck, Austria, <sup>2</sup> EXTRO-Lab, Tyrolean Cancer Research Institute, Innsbruck, Austria*

Cancer stem cells (CSC) are a distinct subpopulation within a tumor. They are able to self-renew and differentiate and possess a high capability to repair DNA damage, exhibit low levels of reactive oxygen species (ROS), and proliferate slowly. These features render CSC resistant to various therapies, including radiation therapy (RT). Eradication of all CSC is a requirement for an effective antineoplastic treatment and is therefore of utmost importance for the patient. This makes CSC the prime targets for any therapeutic approach. Albeit clinical data is still scarce, experimental data and first clinical trials give hope that CSC-targeted treatment has the potential to improve antineoplastic therapies, especially for tumors that are known to be treatment resistant, such as glioblastoma. In this review, we will discuss CSC in the context of RT, describe known mechanisms of resistance, examine the possibilities of CSC as biomarkers, and discuss possible new treatment approaches.

#### Edited by:

*Marco Durante, Helmholtz-Gemeinschaft Deutscher Forschungszentren (HZ), Germany*

#### Reviewed by:

*Emanuele Scifoni, National Institute of Nuclear Physics (INFN), Italy Claere Von Neubeck, University of Duisburg-Essen, Germany*

> \*Correspondence: *Christoph Reinhold Arnold christoph.arnold@i-med.ac.at*

#### Specialty section:

*This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology*

Received: *01 September 2019* Accepted: *30 January 2020* Published: *20 February 2020*

#### Citation:

*Arnold CR, Mangesius J, Skvortsova I-I and Ganswindt U (2020) The Role of Cancer Stem Cells in Radiation Resistance. Front. Oncol. 10:164. doi: 10.3389/fonc.2020.00164* Keywords: cancer stem cells, radiation resistance, radiation therapy, DNA damage repair, reactive oxygen species, stem cell niche, quiescence

## INTRODUCTION

Radiation therapy (RT) is one of the mainstays of cancer treatment. Roughly one half to two-third of all oncologic patients receive some form of RT in the course of their disease (1–5), either in a curative setting for primary treatment with or without other treatment modalities (i.e., surgery or chemotherapy, CHT) or in a palliative setting for the irradiation of symptomatic metastases. Importantly, the number of patients requiring RT is expected to increase in the foreseeable future (6). In short, RT exerts its effect by inducing DNA damage, either directly or indirectly via the production of water-derived radicals and reactive oxygen species (ROS) (7–9), which then interact with macromolecules including DNA, lipids and proteins. As a consequence, DNA damage response (DDR) is initiated and leads to the activation of the DNA damage repair machinery as well as the induction of checkpoint kinase pathways, which delay cell cycle progression in order to facilitate DNA repair (10–13). In case the DNA is damaged beyond repair, DDR signaling induces apoptosis, senescence or mitotic catastrophe, all of which imply the loss of reproductive capacity of a cell (14–16). Consequently, if successful, RT hinders cancer cells from further proliferation. In theory, every (cancerous) cell can be killed with RT given a high enough dose. However, the surrounding healthy tissue limits the applicable dose (17). RT is usually a balancing act between giving enough dose to achieve local tumor control and only as much dose as the surrounding tissue can tolerate. Despite a very high local tumor control rate, a non-negligible rate of therapy failure still constitutes one of the major limitations in radiation oncology (18, 19). Insufficient response to irradiation (i.e., radiation resistance) contributes to residual cancer mass, which is the key driver of locoregional or distant recurrence, both of which are negatively influencing the patient's prognosis as local recurrence often is associated with metastatic spread, which is almost always fatal.

In recent years evidence has accumulated showing that multiple genetically diverse clones co-exist within various kinds of tumors (20–24). Not all cells within a tumor are equally sensitive to RT. Understanding the diverse radiosensitivity of different tumor cell subpopulations is very important. It challenges the common practice of employing macroscopic bulk tumor responses (as measured with medical imaging) as the primary endpoint for determining the effectiveness of an antineoplastic treatment. While this is certainly a very practical approach, it is only then true if the bulk tumor response represents the response of all cells within the tumor, including the most resistant subpopulation within the tumor. This is most likely not the case for most tumors because not all subpopulations are equally affected by the treatment. The stem cell model of cancer development may explain genetic, functional, and phenotypical differences, such as increased therapy resistance, within a tumor, even within the same tumor clone. Cancer stem cells (CSC), albeit difficult to identify, are believed to contribute to resistance to various oncological therapies, including RT (25– 34), making them a primary target for anti-cancer therapy. Hence, a proper understanding of differential sensitivity of cancer cells, especially CSC, to irradiation is vital in order to develop new or improve existing anti-cancer therapies. Most research on CSC has been done in breast cancer and glioma (35, 36). However, as CSC differ between entities results cannot be transferred to other tumors, at least not without caution.

#### CANCER STEM CELLS

Today, there are basically two largely accepted models for the origin of cancer: the standard (hierarchical) CSC model and the clonal evolution model. In the latter, genetic mutations accumulate with time and theoretically any cell can have tumorigenic potential (37). The CSC model describes a hierarchical organization of tumors with tumorigenic CSC at the apex which divide asymmetrically to form new CSC as well as differentiated non-tumorigenic progenies (34). Adding to the complexity of the CSC theory, is the fact that differentiation may be bidirectional. In this way, differentiated non-tumorigenic tumor cells may, instructed by niche signals, re-differentiate into CSC to replace lost stem cells. Even though data supporting the CSC hypothesis with its hierarchical organization of tumors is more solid, it is feasible that both, the CSC hypothesis and the clonal evolution model are not necessarily mutually exclusive.

The generation of CSC is not conclusively clarified, and several hypotheses exist (38). CSC may originate from normal stem cells, where random mutations during DNA replication may lead to them becoming malignant (39). Additionally, aberrant stromal signaling and pro-inflammatory conditions can lead to the malignant degeneration of normal stem cell (40). Alternatively, as stated earlier, CSC can be derived from differentiated cells. This can occur via genomic instability of tumor cells, horizontal genetic transfer, or microenvironmental signals. Genetic instability describes the acquisition of additional genetic mutations that provide any given differentiated tumor cell with stem cell traits so that it becomes a CSC (41). It is, however, unclear, whether stem cell traits shift from one cell to another in a stochastic manner during tumor evolution. In horizontal gene transfer, a normal stem cell may phagocyte fragmented DNA from tumor cells leading to their reprogramming and CSC formation (42). Microenvironmental signals include proinflammatory cytokines such as interleukin-6 (IL-6), which has been shown to facilitate dedifferentiation of non-CSC into CSC, or nuclear factor-kB (NF-kB), which maintains CSC numbers (43).

The frequency of CSC within a tumor is difficult to estimate and largely depends on the type of malignancy. In solid tumors, reported CSC rates are in the range from below 1% of all tumor cells to more than 80% (44–48). Similarly, the frequency of CSC in hematologic malignancies also displays a broad spectrum and ranges from <1% in acute myeloid leukemia (AML) up to over 80% in acute lymphoblastic leukemia (ALL) (49, 50). It has been shown in glioblastoma that CSC seem to reside predominantly in niches that are hypoxic, low in nutrients and have a low pH (51).

CSC are a distinct subpopulation within the heterogenous tumor mass and share several properties with normal stem cells (SC), the most important being the ability to self-renew (i.e., the potential of unlimited cell division) and the ability to give rise to more differentiated, mature cancer cells (34). Like in healthy tissue, stem cells initiate, promote, and maintain tumor development and growth (and re-growth after treatment) (52–55). It has been shown in glioma and breast cancer that the number of CSC in a tumor at the time of treatment is inversely correlated with clinical outcome (56, 57). Furthermore, repopulation of CSC after fractionated RT is one of the most important factors that determine local tumor control (58, 59). Therefore, inactivation of all CSC within a tumor is the prerequisite for a curative cancer treatment (60).

One major challenge regarding CSC is their correct identification as there is not one specific CSC marker, not least because of the high intra- and inter-tumor heterogeneity as well as tumor plasticity and the associated changes in genotype and phenotype. However, there are some cell surface marker that seem robust enough to use them as indicators for CSC. Two of these biomarkers are CD44 (found on CSC in cancers of the colon, esophagus, stomach, pancreas, breast, brain, lung, ovaries, prostate, liver, and the head and neck region) and CD133 (found in cancers of the colon, esophagus, stomach, pancreas, brain, lung, ovaries, prostate, liver, skin, and the head and neck region) (38, 61). Naturally, there are more surface molecules and usually combinations of these markers are used to identify and isolate CSC depending on the type of tumor that is investigated. It is important to emphasize that CSC differ between tumor entities, both phenotypically and functionally, and results from one type of cancer should not be translated to other types.

#### RADIOCURABILITY AND RADIATION THERAPY RESISTANCE

Following RT-induced DNA damage the balance of pro- and antiapoptotic pathways skews toward cell death induction. However, in CSC pro-survival pathways seem to be more pronounced and protect these cells from cell death, rendering CSC radiation resistant (27). Radiation resistance of CSC may either be primary, i.e., due to the constitutive upregulation of certain molecules and pathways (see below). Alternatively, radiation resistance of CSC may be acquired. Following RT, as is the case with any other antineoplastic therapy, intratumoral heterogeneity can theoretically promote clonal evolution through Darwinian selection and lead to the development of adaptive responses with the result of more resistant, aggressive, and invasive tumors (62). CSC clones with genomic alterations that protect them against RT are selected for and continue to sustain the tumor (63). Indeed, it is known that RT preferentially kills non-CSC, thereby enriching the tumor for CSC (64). In addition, RT has been shown to induce the reprogramming of non-CSC in breast cancer as well as squamous cell carcinoma of the head and neck (HNSCC) leads to the acquisition of functional CSC traits in order to compensate for cell loss in the stem cell compartment in response to cellular injury as is the case after RT (65, 66). Finally, RT leads to the recruitment of CSC in breast cancer from a quiescent state into the cell cycle (67, 68). In this way, RT contributes to the acquired or adaptive radioresistance via selective repopulation from the surviving CSC.

The number of CSC within a tumor predicts the radiation dose needed to eradicate the tumor. Therefore, in tumors with a higher proportion of CSC a given dose of irradiation leads to a lower probability of local control as compared to tumors with fewer CSC (69–71). From a clinical point of view, this implies a dose-volume dependency, as radiocurability of tumors inversely correlates with tumor volume (72) and with intrinsic radiosensitivity in vitro (73). Furthermore, the probability of successful irradiation also correlates with the number, density, and intrinsic radiosensitivity of CSC (60, 71) and the absolute number of CSC increases with tumor volume (70–72, 74). Importantly, survival of one single CSC after RT can lead to tumor relapse. Hence, eradication of the entire CSC population is of utmost importance for the patient. Nonetheless, one must keep in mind that CSC differ between tumor types and there is no general radiation resistance of CSC, as many patients can be cured with current concepts of conventional RT.

### CELLULAR FACTORS FOR CSC RADIORESISTANCE

Several cellular features render CSC radioresistant. In the following, we will discuss the best-studied cellular factors, which include low levels of ROS, increased DNA damage repair capacity, or quiescence (**Figure 1**). These are common characteristics of healthy SC and CSC alike, presumably to protect their DNA from stress-induced damages.

ROS are involved in various physiological processes, such as proliferation, differentiation, metabolism, apoptosis, angiogenesis, wound healing, or motility (75, 76). Intracellular ROS levels are tightly and continuously regulated by ROS scavengers, which include superoxide dismutase, superoxide reductase, catalase, glutathione peroxidase, glutathione reductase, or apurinic/apyrimidinic endonuclease/redox

effector factor (Ape1/Ref-1, also known as APEX1) (77). There is a multitude of publications showing that ROS scavengers are upregulated and highly efficient in CSC of various tumors (78–83) leading to low levels of ROS and protecting CSC from RT-induced cell death, as ROS is essential for the effect of RT (84). This protective effect of upregulated ROS scavengers even outweighs the effect of oxygen, a known potent radiosensitizer (85). Along this line, it has been shown that CSC produce less ROS upon radiation compared to non-CSC (86).

Secondly, DNA damage repair capacity following RT, especially regarding DNA double strand breaks (DSB), has been shown to be higher in CSC as compared to their non-tumorigenic counterparts (25, 87–90). This has been shown in CSC of several tumors, including glioma, nasopharyngeal carcinoma, prostate, lung, and breast cancer, and is mainly attributed to the activation of checkpoint-pathways in response to RT (90– 98). The resulting delayed cell cycle progression allows for repair of the DNA damage. Interestingly, CSC have been shown to repair DNA damage preferably via homologous recombination (HR) instead of non-homologous end-joining (NHEJ), the latter being less accurate and more error-prone than HR (25, 99–101). A comprehensive review on DNA damage repair in CSC has recently been published (102).

Third, it has been shown in various studies that CSC proliferate more slowly than further differentiated cancer cells (78, 103, 104). This is of importance as RT is known to be more effective in killing rapidly proliferating tumor cells as compared to slowly dividing (i.e., quiescent or dormant) cells and quiescence is associated with relative radiation resistance (105, 106). This way, non-proliferating cells survive therapeutic irradiations and remain quiescent for a various amount of time, which can range from weeks (78, 104) to even decades (107, 108). Once they continue to proliferate, these cells can cause a recurrence.

#### TUMOR MICROENVIRONMENT

Similar to healthy stem cells, CSC reside in specific niches that provide microenvironmental factors such as autocrine signaling and signals coming from stromal fibroblasts (cancer associated fibroblasts, CAF), immune cells, endothelial cells, and extracellular matrix components (109, 110). The exact composition of the niche is not well-defined for most tumors, as are the exact supporting signals. It is known, however, that the niche supplies CSC with oxygen and nutrients, supports stem cell functions, protects from insults such as radiation, and regulates responsiveness to a therapy (111). For instance, in breast cancer it has been shown that deregulation of the stem cell niche by increased expression of bone morphogenetic protein 2 (BMP2) can initiated and promote malignant stem cell transformation (112). Furthermore, at least in glioblastoma tumor samples, there are different types of niches, such as hypoxic periarteriolar niches, peri-vascular niches, peri-hypoxic niches, periimmune niches, and extracellular niches (113). Whether other types of cancer have different types of niches, and what therapeutic implications this finding might have, still needs to be elucidated.

Like the tumor itself, the tumor microenvironment also responds to RT (114) (**Figure 1**). For instance, it has been shown that CAFs acquire a pro-malignant phenotype after RT of in colorectal cancer samples (115–117). Furthermore, CAFs induce autophagy following irradiation of lung cancer and melanoma cell lines leading to enhanced cancer cell recovery and tumor re-growth (118).

Additionally, RT induces pro-inflammatory cytokines in the tumor microenvironment (119, 120), including platelet-derived growth factor (PDGF), interleukin 1β (IL1β), tumor necrosis factor α (TNFα), transforming growth factor β (TGFβ), C-X-C motif chemokine 12 (CXCL12), and matrix metalloproteinases (MMP), and interleukin-6 (IL6). This leads to the upregulation of ROS scavengers in CSC (7) and activation of downstream STAT3 signaling, a cascade known to promote self-renewal in embryonic stem cells neu (121). Furthermore, this promotes survival of tumor cells, facilitates tumor regrowth and leads to the development of highly invasive CSC phenotypes (122, 123).

Another mechanism by which the niche protects CSC is hypoxia. Oxygen is a potent radiosensitizer and is needed for radiation-induced ROS production and in further consequence for cell death. Lack of oxygen is known to increase radiation resistance (124–126) and has been associated with early relapse after RT. Consequently, increasing tumor oxygenation improves the response to RT (127–129). In addition to the absolute lack of oxygen and the resulting low ROS levels, CSC in hypoxic niches upregulate ROS scanvengers (7, 130, 131), thereby further lowering ROS levels compared to normoxic CSC. This in turn leads to the activation of the hypoxia-inducible factor (HIF) signaling route (132–134). Interestingly, HIF1α and the respective regulated cytokines have also been shown to be increased following RT (135). HIF are important master regulators of transcription of hypoxia response elements, which activates pro-survival pathways such as the Notch, wingless and INT-1 (WNT) and Hedgehog pathway (136–138). These pathways have been shown to be important for CSC maintenance and can lead to radioresistance and accelerated repopulation of CSC during or after treatment, as has been shown in glioma, breast cancer, and prostate cancer (97, 139–142).

### CSC AS BIOMARKER

There is accumulating evidence that CSC could be used as biomarkers to predict treatment response and estimate the likelihood of tumor relapse in cancer patients. It has been shown in various tumors, including urothelial cancer (143– 145), gastric cancer (146–150), pancreatic cancer (151, 152), HNSCC (153–155), glioma (156–159), thyroid carcinoma (160, 161), hepatocellular carcinoma (162, 163), breast cancer (164), and lung cancer (165). However, it has been shown in ovarian cancer that the prognostic value of CD44 may depend on its isoform, with the transmembrane form indicating a better prognosis, while the presence of the soluble extracellular domain was associated with a worse prognosis (166). In a recent meta-analysis, overall CD44 expression in ovarian cancer was associated with a high TNM stage and a poor 5 year overall survival (167). Along this line, low expression of CD44 was shown to be an independent factor of poor prognosis in ovarian mucinous carcinoma (168). Interestingly, in breast cancer, it has been shown that CD44 was associated with longer diseasefree-survival (DFS) in estrogen-receptor (ER) positive women, while CD44 positive tumors were associated with poor outcome in ER-negative patients (169). In a recent meta-analysis, Han and colleagues tried to generalize the prognostic significance of CD44 and its variant isoforms in advanced cancer patients. In this analysis of 15 articles with more than 1,200 patients, CD44 was slightly linked to a worse overall survival, but there was no correlation between CD44 expression and DFS, recurrence-free survival (RFS), or progression-free survival (PFS) (170).

Two studies from the German Cancer Consortium Radiation Oncology Group (DKTK-ROG) have shown that CSC marker expression is a potential biomarker for favorable prognosis in patients with locally advanced HNSCC, both after primary chemoradiotherapy (171) as well as following post-operative chemoradiotherapy (172). In a recent validation study from the same group, the addition of CD44 could further improve the prognostic performance of models using tumor volume, p16 status, and N stage (173).

Activity of the 26S proteasome, a protease complex with regulatory functions in cell cycle, DNA repair, and cell survival, is another CSC marker (131, 174–177). In this regard, low 26S proteasome activity correlated with high self-renewal capacity and high tumorigenicity in HNSCC cell lines (178). A high 26S proteasome activity correlated with a longer survival and higher local control rates in patients who underwent chemoradiotherapy for HNSCC (95).

Another potential biomarker, especially for glioma, might be the stem cell marker CD133. In a recent meta-analysis including 21 articles with more than 1,550 patients, CD133 expression correlated with higher grade of gliomas and worse prognosis in glioma patients (179). Interestingly, in a recent in vitro analysis in glioma cell lines, CD133 expression could be downregulated by vincristine, a common chemotherapeutic drug (180). In a study by Wu and colleagues, CD133 promotor methylation was a significant prognostic factor for adverse PFS and overall survival, while there was no correlation between CD133 protein expression and survival (181). Additionally, there are other publications that showed no association of CD133 protein expression with survival (182, 183).

#### NOVEL TREATMENT APPROACHES

Increased understanding of CSC has led to new ideas for improving cancer therapy. It certainly seems feasible to combine conventional anti-neoplastic therapy (e.g., RT or CHT) to target the tumor bulk with CSC specific treatment in order to improve outcome as compared to monotherapies (184, 185). Inactivation of even only limited numbers of CSC might significantly improve local tumor control probability (70). However, the assumed plasticity of the CSC and non-CSC compartments, especially the possible shift of stem cell traits, increases the complexity of treatment responses of tumors. In an ideal situation where the CSC population is strictly static, drugs that specifically target CSC would lead to a massively improved treatment outcome, possibly even without the need of additional treatment modalities (e.g., RT or CHT). However, CSC plasticity render a CSC-targeting monotherapy virtually impossible, since after treatment, non-CSC may gain CSC traits and repopulate the tumor. Additionally, regarding the current CSC marker and their uncertainty in robustly identifying CSC, and to sufficiently distinguish them from healthy SC, a strictly CSC-based therapy seems to be still a long way off.

One conceivable possibility to eliminate CSC with RT more efficiently is to increase the applied dose. This is usually not feasible for the whole tumor due to dose constraints of the surrounding healthy tissue. Therefore, visualization of CSC could allow for larger doses of radiation in CSC rich regions while still respecting the dose constraints. Indeed, there are first studies in mice, where CD133+ glioblastoma cells could be detected non-invasively by PET and near-infrared fluorescence molecular tomography using antibody-labeled tracer (186). Subsequently, the same group showed that near-infrared photoimmunotherapy using phototoxic antibody conjugates was efficient not only in rendering CD133+ glioblastoma cells visible but also in inducing cell death (187).

Another means by which RT can be utilized to eliminate CSC more efficiently is the use of types of irradiation other than commonly used photon beams (188, 189). Particle beams of protons and carbon ions are being increasingly used due to their advantageous depth-dose curve and their higher cellkilling efficiency compared with photons (190). Preclinical studies deliver promising results when using proton irradiation. For instance, in CSC-like cells from two human NSCLC cell lines, irradiation with protons was more efficient than photon treatment in reducing cell viability. clonogenic survival, cell migration, and invasiveness, while increasing apoptosis and ROS levels (191). Furthermore, proton irradiation has been shown to be more cytotoxic, induce higher and longer cell cycle arrest, reduce cell adhesion and migration ability, and reduce the overall population of CSC in NSCLC cell lines compared to photon irradiation (192). Finally, in glioma stem cells from glioblastoma patients, similar results have been achieved (193). In this study, particle irradiation with protons and carbon ions has been shown to be more effective in cell killing compared to photons, likely because of the different quality of the induced DNA damage. Indeed, compared to photons, proton beam irradiation has been shown to increase ROS levels, induce more single and double strand DNA breaks, less DNA damage repair (as measured by H2AX phosphorylation), and decreased cell cycle recovery which led to increased apoptosis (194). Interestingly, primary human glioma stem cells that were resistant to photon treatment could be rendered sensitive with carbon ion irradiation via impaired capacity to repair carbon ion induced DNA double strand breaks (195). Importantly, this study also showed an individual heterogeneity in the amount and radiosensitivity of glioma stem cells from different patients, further complicating a one size fits all treatment. In the recent years, immune checkpoint inhibitors have greatly improved treatment outcomes for many cancers. In this regard, it has been shown that proton irradiation increased sensitivity of CSC from different cell lines, such as breast or prostate cancer, to cytotoxic T-cell killing (196). These findings offer a rationale for the combined use of proton irradiation with immunotherapy. Another important aspect of particle irradiation is the reduced dependence from tissue oxygenation. While photon irradiation is always strongly affected by the presence of oxygen in the induction and maintenance of DNA damage, high LET particle beams can be much less hindered by hypoxic conditions, which are often found in solid tumors. For example, Tinganelli et al. (197) showed the survival of mammalian cells exposed to different types of particle radiation in various oxygen concentrations, leading to a hypoxia-adapted irradiation plan. Hence, it seems feasible and promising to use particle irradiation in order to counteract tumor hypoxia. Taken together, these results suggest a potential advantage of particle beam irradiations in CSC eradication, eventually in combination with conventional photon irradiation: photon beam irradiation to the whole tumor and a boost of particle beams to hypoxic areas within the tumor. Alternatively, one can ideally use a properly optimized plan of carbon or oxygen beam, or, considering the potential of the most advanced particle centers, a multiple-ion irradiation (198). Another new strategy in combating glioblastoma is the interference with metabolic pathways. It has been shown that dichloroacetate (DCA), an orphan drug, has been shown to switch the metabolism in freshly isolated glioma stem cells from mitochondrial oxidative phosphorylation to cytoplasmic glycolysis, which in turn increased mitochondrial ROS and induced apoptosis (199). In the same study, glioblastoma patients treated with the standard of care (i.e., radiation therapy and temozolomide) received oral DCA for up to 15 months. The drug was well-tolerated, and some patients showed prolonged radiologic stabilization and decelerated tumor progression. Additionally, DCA, in combination with PENAO [4-(N-(Spenicillaminylacetyl)amino) phenylarsonous acid] has been shown to increase radiosensitivity of glioma cells by inducing a cell cycle arrest, elevating ROS production, depolarizing mitochondrial membrane potential, increasing DNA damage, and inducing apoptosis (200).

Early experiments with CSC-directed antibodies have shown promising results. Gurtner et al. (201) used antibodies directed against CD44 that were loaded with highly cytotoxic drugs. In this experimental setting, these antibodies, combined with irradiation, led to an improved local tumor control. Considering the high expression of ROS scavengers in CSC, it may be reasonable to target these ROS scavengers. Pharmacological depletion of ROS scavengers (e.g., by treatment with buthionine sulfoximine (BSO), has been shown to reduce radiation resistance as well as clonogenic properties of CSC in breast cancer and HNSCC (202, 203). Additionally, the use of antioxidants during RT, which reduce ROS levels, might prove nonsensical, as high ROS levels are needed, especially in CSC, in order to facilitate RT-induced cell death.

Checkpoint pathways can be inhibited pharmacologically to prevent delayed cell cycle progression and hamper DNA damage repair in CSC as has been shown in glioma and prostate cancer (98, 204). Additionally, direct inhibition of DNA damage repair signaling has been shown to reduce radiation resistance in breast cancer CSC (92). Chemical inhibition of Notch signaling, a developmental pathway that is known to be essential for tissue homeostasis (205) has been shown to increase radiosensitivity in glioma CSC (206). However, most clinical studies regarding these signaling pathways focus on chemotherapy (207). There is preclinical data in mice with glioblastoma multiforme that were treated with RT and temozolomide combined with a Notch inhibitor. In this study, Notch inhibition had an antiglioma stem cell effect which provided a survival benefit (208). Regarding clinical data, there is an ongoing trial testing a Notch inhibitor combined with whole-brain RT or stereotactic cranial RT [NCT01217411], but so far there are no results of this study available.

Overcoming tumor hypoxia is another method to improve cancer therapy. Nitroimidazole derivates can mimic the effect of oxygen and can produce reactive species in hypoxic cells. There have already been clinical trials with a clinically relevant

#### REFERENCES


sensitization and low toxicity (209, 210). Nitric oxide (NO) is also able to mimic oxygen and thereby to increase radiosensitivity in hypoxic tissue. In a first clinical trial, the NO donor glyceryltrinitrate (GTN) has been shown to reduce hypoxia-induced progression of prostate cancer (211). Taken together, targeting the hypoxic niche of CSC might eventually improve treatment outcome following RT. An important task in this regard will be the correct identification of CSC for a given tumor, since CSC can differ between tumors both functionally/metabolically as well as phenotypically. Additionally, CSC share many properties and surface molecules with normal stem cells, making a clear distinction difficult and increasing the risk of unwanted side effects. Finally, it seems conceivable that drugs that interfere with spontaneous as well as RT-induced reprogramming of non-CSC into CSC could also be of value for cancer treatment. However, it needs to be investigated if these mechanisms are the same in normal steam cells and CSC before such drugs can be developed.

Another interesting approach to amplify the effect of radiation therapy, particularly on CSC, is the use of nanostructures that, after being endocytosed by cancer cells and following irradiation, release secondary electrons and large amounts of ROS (212). This could be especially effective in CSC, where ROS levels are generally lower. In this study, a significant tumor growth suppression and overall improvement in survival rate has been demonstrated in an in vitro and in vivo model of triple negative breast cancer.

#### CONCLUSION

There is growing evidence of a radiation resistance tumor subpopulation with increased DNA damage repair, increased survival signaling, and decreased ROS, that is furthermore protected by its environment. With refined understanding of these cells and their role in development and progression of cancer come new possibilities to improve cancer therapy. Targeting CSC, based on phenotype or function, seems promising. Nonetheless, we are just at the beginning and clinical data is still scarce. Major issues concern their correct identification and reliable distinction from healthy cells and the plasticity of the CSC department. It will be exciting to see which position CSC-specific therapies will occupy within the row of current anti-neoplastic therapies.

#### AUTHOR CONTRIBUTIONS

CA and JM performed the literature research. CA wrote the manuscript. I-IS and UG supervised the project.

breast, and lung cancers: verification of model estimates of need with radiotherapy utilization data from British Columbia. Int J Radiat Oncol Biol Phys. (2011) 79:1507–15. doi: 10.1016/j.ijrobp.2009. 12.070

3. Zhang-Salomons J, Mackillop WJ. Estimating the lifetime utilization rate of radiotherapy in cancer patients: the multicohort current utilization table (MCUT) method. Comp Methods Programs Biomed. (2008) 92:99–108. doi: 10.1016/j.cmpb.2008.06.011


combined treatments. Adv Drug Deliv Rev. (2017) 109:63–73. doi: 10.1016/ j.addr.2016.02.002


with clinical outcomes of gastric cancer patients. Asian Pac J Cancer Prev. (2018) 19:1313–8. doi: 10.22034/APJCP.2018.19.5.1313


CD133+ tumor stem cells. Proc Natl Acad Sci USA. (2014) 111:E692–701. doi: 10.1073/pnas.1314189111


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Arnold, Mangesius, Skvortsova and Ganswindt. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Novel Therapeutic Strategies for Ovarian Cancer Stem Cells

Nastassja Terraneo<sup>1</sup> , Francis Jacob<sup>2</sup> , Anna Dubrovska3,4,5,6 and Jürgen Grünberg<sup>1</sup> \*

<sup>1</sup> Center for Radiopharmaceutical Sciences ETH-PSI-USZ, Paul Scherrer Institute, Villigen, Switzerland, <sup>2</sup> Ovarian Cancer Research, Department of Biomedicine, University Hospital Basel and University of Basel, Basel, Switzerland, <sup>3</sup> OncoRay-National Center for Radiation Research in Oncology, Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universität Dresden, Dresden, Germany, <sup>4</sup> German Cancer Consortium (DKTK), Partner Site Dresden, Dresden, Germany, <sup>5</sup> German Cancer Research Center (DKFZ), Heidelberg, Germany, <sup>6</sup> Helmholtz-Zentrum Dresden-Rossendorf, Institute of Radiooncology-OncoRay, Dresden, Germany

Ovarian cancer (OC) is one of the most lethal gynecologic malignancies. Due to the lack of specific symptoms and screening methods, this disease is usually diagnosed only at an advanced and metastatic stage. The gold-standard treatment for OC patients consists of debulking surgery followed by taxane combined with platinum-based chemotherapy. Most patients show complete clinical remission after first-line therapy, but the majority of them ultimately relapse, developing radio- and chemoresistant tumors. It is now proposed that the cause of recurrence and reduced therapy efficacy is the presence of small populations of cancer stem cells (CSCs). These cells are usually resistant against conventional cancer therapies and for this reason, effective targeted therapies for the complete eradication of CSCs are urgently needed. In this review article, we highlight the mechanisms of CSC therapy resistance, epithelial-to-mesenchymal transition, stemness, and novel therapeutic strategies for ovarian CSCs.

#### Edited by:

Paul N. Span, Radboud University Nijmegen Medical Centre, Netherlands

#### Reviewed by:

Yoshihiko Hirohashi, Sapporo Medical University, Japan Janneke Hoogstad-van Evert, Radboud University Nijmegen Medical Centre, Netherlands

\*Correspondence:

Jürgen Grünberg juergen.gruenberg@psi.ch

#### Specialty section:

This article was submitted to Radiation Oncology, a section of the journal Frontiers in Oncology

Received: 30 September 2019 Accepted: 21 February 2020 Published: 17 March 2020

#### Citation:

Terraneo N, Jacob F, Dubrovska A and Grünberg J (2020) Novel Therapeutic Strategies for Ovarian Cancer Stem Cells. Front. Oncol. 10:319. doi: 10.3389/fonc.2020.00319 Keywords: ovarian cancer, cancer stem cells, epithelial-to-mesenchymal transition, L1CAM, radioimmunotherapy, Auger electron and alpha particle emitters, therapeutic strategies

## INTRODUCTION

#### Ovarian Cancer

Ovarian cancer (OC), which is a generic term for several different malignant tumors, is one of the most frequent cancer types in females (1). OC is the leading cause of cancer death among gynecological malignancies, and according to estimates, in 2019 ∼22,530 new cases of OC will be diagnosed and 13,980 OC-related deaths will occur in the U.S. (2). About 75% of all ovarian tumors and 90–95% of ovarian malignancies are epithelial ovarian cancer (EOC) (3) of different origin (4). The 5-year survival rate of OC varies from 30 to 92%, depending on the dissemination of the disease at the time of diagnosis (5). This type of cancer usually shows non-specific symptoms during the early development (e.g., abdominal distension and pain, loss of appetite, or increased urinary frequency) and currently there are still no reliable early screening strategies available (6). The most common sign of advanced disease is abdominal swelling due to ascitic fluids accumulation (7). OC is consisting of different histological subtypes with distinctive molecular genetic features, clinical presentations, and prognostic outcomes. In 2014, the new WHO criteria recognized five principal epithelial OC histotypes: high-grade serous (HGSC), low-grade serous (LGSC), endometrioid, clear cell, and mucinous carcinoma (8). Among them, HGSC is the most common histologic subtype of OC, with a poor 5-year survival rate of 35–40% and accounting for 70–80% of OC deaths (5, 9). Because of the lack of diagnostic methods, ∼75% of patients show the metastatic spread in the

**217**

peritoneal cavity and adjacent organs at diagnosis, which corresponds to FIGO stages (International Federation of Gynecology and Obstetrics stages) III and IV (10). The FIGO OC staging system was first published in 1973 and revised in 2014 (11, 12). Stage I EOC (disease is confined to one or both ovaries) is usually associated with good survival rates and surgery alone is sufficient as a therapeutic approach; unfortunately, it is quite rare since most of the patients are diagnosed only at stages III and IV (13). At stage II, tumor includes either one or both ovaries with pelvic dissemination and spread into the uterus and/or fallopian tubes. This group makes up <10% and it is considered curable, usually with chemotherapy (12). Stage III EOC tumor implicates one or both ovaries with spread to the peritoneum, lymph nodes and/or other sites outside the pelvis. The majority of OC are HGSCs and patients are diagnosed at stage III (14). Stage IV is characterized by distant metastases affecting the liver, spleen and lymph nodes and/or to other organs or tissues outside the peritoneal cavity such as the lungs and bones.

### Ovarian Cancer Treatments

Current first-line treatment regimen for OC patients comprises complete debulking surgery. The reductive tumor procedure includes hysterectomy, omentectomy, and other affected tissues possible to remove. The goal of surgery is to reduce tumor burden and minimize residual disease, which is inversely proportional to survival (15). Indeed, residual lesions smaller than 2 cm have been associated with better survival than bigger ones (16). At the same time, debulking surgery allows to precisely establish the histologic subtype of the disease and, therefore, it is very important for diagnosis. Even though surgery is the basis for OC treatment, it is rarely curative alone for patients with advanced disease and it needs to be combined with chemotherapy.

In late 1990s, two phase III clinical trials combined cisplatin (CDDP) with paclitaxel (PTX) as adjuvant treatment for advanced stage OC (17). Ever since, the combination of taxane and platinum derivatives, like CDDP and carboplatin (CBT), has been used as a standard therapeutic approach for OC patients, leading to response rate, and complete clinical remission of 60–80% (18). Nevertheless, the majority of these patients will ultimately relapse with a median progression-free survival of 18 months (19). Usually, response rates to second-line chemotherapy are proportional to treatment-free interval (20). Different combinations of chemotherapeutics have been tested to overcome chemoresistance following first-line paclitaxelplatinum treatment, but clinical responses are short-lived and led to only minor survival improvements for patients with chemoresistant tumors (21). So far, radiation therapy (RT) has played a minor role in ovarian cancer. Abdominopelvic RT was associated with serious side effects and poor therapeutic efficacy for most of the patients (22, 23). Acute toxicity was most commonly due to cramps, diarrhea, nausea, vomiting and more severe myelosuppression, whereas long-term toxicity was associated with bowel obstruction (23, 24). The poor therapeutic effect was due to the limitation of dosage that causes toxicity to radiosensitive organs such as blood building system or kidney. Maybe new approaches in radiotherapy could lead to wider use of RT in OC [reviewed in Fields et al. (25) and Iorio et al. (26)]. Likewise, the use of radiolabeled antibodies for the management of advanced ovarian cancer after cytoreductive surgery and chemotherapy is limited and of no success so far. Antibodies directed to CA-125 (mAb OC-125) and MUC1 (mAb HMFG1) antigens labeled with iodine-131 or yttrium-90, respectively showed little or no therapeutic benefit in ovarian cancer patients (27, 28). Different reasons for the treatment failure were discussed. The dose of radiation may have been too low because of insufficient binding of the antibody or the lack of antigen expression in residual micrometastases. Furthermore, yttrium-90 is not the ideal isotope for irradiation of small tumor nodules. No pharmacokinetics was performed, and it is possible that there was limited systemic exposure to the intact radioimmunoconjugate. In addition, the anti-MUC1 antibody HMFG1 is not actively internalized into target tumor cells. In contrast to RT, radioimmunotherapy (RIT) is a targeted therapy were the antibody brings the radiation to the tumor site also to disseminated metastases. Non-specific irradiation of healthy tissue is normally low due to the clearing of the mAb from the blood. The dose-limiting organ is the bone marrow [reviewed in Larson et al. (29)]. New promising RIT approaches for the treatment of OC will be discussed during this report.

The better understanding of tumor biology and chemoresistance over the past years supported the development of molecular targeted therapies, improving survival and increasing the quality of life in OC patients. Many different inhibitors, such as tyrosine kinase inhibitors (30) and monoclonal antibodies (mAbs) targeting multiple crucial cancer pathways, including angiogenesis, cell survival, cell growth, metastasis formation and DNA repair, are currently tested in clinical trials (31). The most promising investigational agents include vascular endothelial growth factor (VEGF)-specific inhibitors and poly (ADP-ribose) polymerase inhibitors (PARPi). Bevacizumab (Avastin <sup>R</sup> , Genentech, Inc.), a recombinant humanized mAb against VEGF, blocks angiogenesis, enhancing the efficacy of standard therapy. In 2004, Bevacizumab has been clinically approved in the U.S. as the first angiogenesis inhibitor for colon cancer (32). In 2018, based on phase III GOG-0218 clinical study (NCT00262847), the FDA approved its use in combination with CBT and PTX, followed by single-agent bevacizumab for the treatment of patients with advanced (stage III or IV) ovarian epithelial, primary peritoneal or fallopian tube cancer after initial surgery (33). PARP enzymes are involved in different cellular functions, including DNA single-strand break (SSB) repair through base-excision repair by PARP1 (34). The first PARPi approved in the clinic was Olaparib (AZD2281, Ku-0059436 trade name Lynparza), an orally administered drug (35). In 2014, based on phase III SOLO-2 (NCT01874353) and phase II Study 19 (NCT00753545) clinical trials, Olaparib obtained an accelerated FDA approval as maintenance treatment for patients with a recurrent ovarian epithelial, fallopian tube or primary peritoneal cancer, who are in complete or partial response to platinum-based chemotherapy (35–38). In the same year, based on phase II Study 19 and phase II Study 42 (NCT01078662), the EMA authorized Olaparib as maintenance treatment for patients with platinum-sensitive, relapsed BRCA-mutated (germline or somatic) HGSC, who responded to the last platinum-based chemotherapy (38, 39). Other types of tumorigenic pathway inhibitors targeting PI3K/AKT, mTOR, Src, and FRα are still in the early phase of development (38). To date, no effective cure for OC has been found. Considering the heterogeneous nature of OC and the lack of a common deregulated pathway in most patients, individualized therapy seems to be essential to improve survival.

### TUMOR HETEROGENEITY

Two main models have been used to explain histological and molecular heterogeneity, a common feature of most solid tumors: the clonal evolution (CE) or stochastic model and the cancer stem cell (CSC) or hierarchical model. In recent times a third model, called the plasticity model, linking CE, and CSC models has been postulated (**Figure 1**).

#### The Clonal Evolution or Stochastic Model

In the 1970s, with the discovery that mutations in oncogenes and tumor suppressor genes trigger most human cancers, Nowell introduced the clonal evolution (CE) concept (40). The CE model assumes that every tumor cell is biologically equivalent and potentially able to drive tumor progression (41). The majority of cancer cells have only restricted proliferative potential and tumor progression is driven by the acquisition of gene mutations and epigenetic alterations in the original clone (42, 43). The progress from early to invasive carcinoma implicates the stepwise acquisition of random mutations in specific cancer genes, leading to uncontrolled proliferation and high tumor heterogeneity (44).

### The Cancer Stem Cell or Hierarchical Model

In the early 1990s, with the introduction of new technologies such as fluorescence-activated cell sorting (FACS) and mouse xenografts assays for hematopoietic stem cells, Dick and colleagues found that tumor engraftment in acute myelogenous leukemia (AML) could only be initiated by CD34+/CD38- cell population (45). In 2003, Clarke et al. applied for the first time the same experimental approach to solid breast cancer (46). Using xenograft assay, they showed that as few as 100 CD44+/CD24−/low breast cancer cells were sufficient to induce tumors, in contrast to thousands of cells expressing different markers (46). Moreover, this tumorigenic population has been passaged several times and these cells were always able to induce tumors recapitulating the original tumor composition, forming both CD44+/CD24−/low tumorigenic and non-tumorigenic cells. Afterwards, similar studies on other solid tumors, such as brain, breast, prostate and colon cancer, were published (47, 48). Most studies on CSCs have been performed in a similar way. Generally, small populations of cells defined by a specific marker or marker panel and expressed in a heterogeneous way in a particular type of tumor, are isolated from cell lines or primary tumors. When transplanted into immunodeficient mice, these cells can induce tumor growth over weeks or months and to reproduce the heterogeneity of the initial tumor (49). A frequency of CSCs present in a given tumor population can be analyzed in limiting dilution assay (LDA) by transplanting increasing dilutions of single tumor cell suspensions. LDA analysis in vivo is the gold standard to determine the CSC frequency in a given tumor cell population (50, 51). In addition to LDA, subsequent transplantation assays provide an important information about

FIGURE 1 | Models of tumor development and heterogeneity. (A) The cancer stem cell (CSC) model of tumor development. Genetic or epigenetic mutations activate stem-like programs in a single cell, generating a CSC. This CSC is able to indefinite self-renewal and/or differentiation and all derived tumor cells have a hierarchical inheritance pattern. (B) The clonal evolution (CE) model of tumor development. Due to the acquisition of epigenetic and genetic mutations through time, any cell might have tumorigenic potential. Tumor heterogeneity is due to the propagation of cells carrying genetic mutations. (C) The plasticity model highlights the plastic state of cancer stemness. Based on the model, differentiated non-tumorigenic cancer cells can potentially revert back to CSCs.

the long-term self-renewal and tumor regeneration capacity of the tentative CSC populations (52). An emerging gene engineering technologies including applications of CRISPR-Cas9 system for target genome editing substantially simplified generation of knockin or knockout cell and mice models. The genetically modified patient-derived organoid and mice models where a given cell population can be traced in vivo is an important tool to identify tumor cell of origin (53, 54). Nevertheless, because of technical issues, many theoretical and experimental details about the CSC model have remained unexplored and the frequency of CSCs in solid tumors is highly variable. Technical issues include inconstant purity of tumor cell isolation, the necessity of more solid and reliable markers and the challenges related to xenotransplant assays that offer a different environment than the original tumor niche (55).

The CSC model suggests that the origin and the progression of many cancers are driven by small subpopulations of cells with stem-like properties; however, this model does not address the question of whether tumors arise from normal stem cells. Instead, it suggests that, regardless of the cell-of-origin, many cancers are hierarchically organized in the same manner as normal tissues and CSCs share similar molecular properties to normal stem cells. In accord with this model, tumors have a hierarchical structure, with tumorigenic CSCs at the top that generate both intermediate progenitors (also called transitamplifying cells) and terminally differentiated cells. Considering that the same CSC populations can originate from different cancer subtypes, the frequency of CSCs can highly vary among tumor types and also within the same tumor, leading to tumor heterogeneity (56). CSCs, like non-neoplastic stem cells, have extensive proliferative potential and generate the differentiated progeny that form most of the tumor mass and it is highly sensitive to cancer therapies. Additionally, these cells can remain quiescent for prolonged periods of time, which renders them unresponsive toward radiation and chemical insults, including cytotoxic drugs designed to target fast-proliferating tumor cells (57). Interestingly, recent studies have highlighted some common features (58, 59) but also many differences in stem cell programs operating in CSCs and non-neoplastic stem cells (60).

### The Plasticity Model

It is now evident that one model does not exclude the other and both might contribute to cancer development, depending on tumor type and stage (61).

In recent years, an alternative model based on cellular plasticity, which links the CE and the CSC models, has emerged (61–63). The plasticity model proposes that cancer cells in different types of tumors including OC can switch between stem cell-like and differentiated states so that some differentiated non-tumorigenic cancer cells can de-differentiate to become CSCs (64). Therefore, CSC-like phenotype is flexible and dynamic, instead of being a fixed property of tumor cells. Signaling within the tumor microenvironment (tumor niche), including oxygenation, cell-to-cell contact and secreted factors, could induce differentiated tumor cells to re-acquire stem celllike properties (62). Additionally, radio- and chemotherapy treatments have been shown to enrich CSC subpopulations in residual tumors because of selective pressure on drug-resistant cells (65–67) and due to tumor cell plasticity (64). Even though the CSC state has high plasticity, it is of high clinical importance as a potential marker for clinical outcome and target for anticancer treatment (68, 69).

### OVARIAN CANCER STEM CELLS

Regardless of the high response rate to standard therapy, most OC patients develop recurrent chemoresistant disease (70). Recurrence is believed to be caused by the presence of residual tumor-propagating cells that cannot be completely eradicated by surgical and/or pharmacological regimens (9). Accumulating evidence suggests that among these residual cancer cells some have the key stem cell-like properties such as self-renewal and differentiation (71, 72). This small population of cells appears to form and to sustain the tumor bulk population, being responsible for disease recurrence after the first-line treatment (73). In some studies, these cells have been isolated by flow cytometry and were discovered to be enriched in a side population (SP) able to efflux the Hoechst33342 dye by cell transporters using the same mechanism with which normal cells efflux toxic drugs (74, 75). Further investigations revealed that these cells have several characteristics in common with normal tissue stem cells. In 2005, Bapat et al. were one of the first groups that characterized the presence of ovarian CSCs from patient ascites, showing tumorigenic properties of these cells (71). In addition to selfrenewal and the ability to give rise to more differentiated progeny, CSCs are highly tumorigenic and display increased resistance against conventional cancer therapies (76–78). As of today, there are still considerable controversies on the OC CSCs due to the heterogeneity of CSC phenotypes, plasticity of CSC states as well as limitations of the current research methodology for CSC characterization. Nevertheless, reliable markers of OC CSCs have been successfully validated by xenograft transplantations of the serial tumor cell dilutions (limiting dilution assay) along with serial tumor transplantation and lineage-tracing assays (79–82). These analyses are currently "gold standard" assays to measure key CSC properties such as self-renewal and multipotency (52). In support of these preclinical findings, clinical significance of CSCs was recently confirmed by a number of studies demonstrated the association of CSC markers with clinicopathological parameters and clinical outcomes of OC patients (69).

#### Ovarian Cancer Stem Cell Markers

The proportion of CSCs can vary depending on tumor type and in the context of OC, CSC frequency shows high interpatient variability (83). Considering CSCs resistance against conventional cancer treatments, the development of more efficient tumor therapies requires effective identification and functional characterization of these cell populations. The expression of several individuals or combined cell surface markers has been associated with CSCs (**Table 1**). The multiplicity of CSC markers, along with their plasticity, might pose a challenge to detect successful CSC-targeting therapeutic strategies. For this reason, it is important to select

#### TABLE 1 | List of some putative CSC markers.


reliable molecular markers that could be used to develop new therapies.

One of the best-characterized CSC surface markers is the transmembrane glycoprotein CD133 (also known as AC133 and Prominin-1), initially identified as hematopoietic stem cell marker supporting stem cell maintenance and expansion (85, 86). CD133 localizes to plasma membrane protrusions and microvilli, indicating its role in membrane organization (86). Several recent studies revealed the role of CD133 as a positive regulator of Wnt, PI3K and EGFR signaling pathways (87, 88). However, the exact physiological function of CD133 in normal and cancer cells is still elusive. CD133 was first characterized as CSC marker in glioblastoma (89) and later it was found to be widely expressed in tumor-initiating cells of different tumors (e.g., ovarian, liver, lung, pancreatic and prostate cancer) (90). Several groups have identified the expression of CD133 in OC cells, which is connected with tumor initiation, selfrenewal and chemoresistance (91, 92). Some published studies indicated the phenotypic heterogeneity and plasticity of CD133 expression (80). This finding might explain why some data did not support the link between CD133 and ovarian CSCs, showing inconsistent expression and no increased spherogenic or tumorigenic properties of CD133+ in comparison to the CD133− counterpart (93, 94). Inconsistent CD133 detection because of different immunoreactivity of primary anti-CD133 antibody clones might also account for discrepancies in the identification of CSCs (95).

The human genome encodes 19 aldehyde dehydrogenase (ALDH) enzymes implicated in detoxification of endogenous and exogenous aldehydes via NAD(P)+-dependent oxidation (96, 97). Upon chemotherapy and irradiation, these enzymes catalyze the oxidation of aldehydes (oxygen, carbon, and hydrogen) to carboxylic acids to prevent DNA damage. ALDH modulates the expression of drug transporters to efflux chemotherapeutic agents, contributing to cell-acquired drug resistance against conventional cancer therapies (98). Additionally, the ALDH protein family catalyzes the oxidation of retinoic acid, which regulates the differentiation of normal stem cells and CSCs (99). The ALDH1 subgroup is highly expressed in normal and CSCs. Among ALDH1 isozymes, ALDH1A1 is more widely expressed in CSCs of different cancer types than ALDH1A2 and ALDH1A3 (100). The study of the ALDH1 expression in 24 types of normal and six types of epithelial tumor tissues revealed that increased expression of ALDH1 was significantly associated with poor outcomes for 439 patients with serous OC (68). Furthermore, ALDH expression in combination with CD133 correlates with poor patient prognosis and characterizes an ovarian CSC population (76, 101). However, recent research suggested that ALDH is a better marker to identify ovarian CSCs than CD133, as ALDH correlates with tumorigenicity and spheroid formation (102). It is now well-known that only cancer progenitor cells have the ability to proliferate under nondifferentiating and non-adherent conditions, forming 3D tumor spheres (103). These spheres are enriched by cells displaying stem cell-like properties, including the upregulation of some stem cellspecific genes, high ALDH activity, self-renewal ability along with high proliferative and differentiation properties (102, 104, 105). Consequently, sphere-forming cells display aggressive growth, migration, invasion, clonogenic survival, anchorage-independent growth, and reduced drug responsiveness in vitro (105).

CD44 cell transmembrane glycoprotein has been associated with several signal transduction pathways, including NANOG and EGFR-Ras-ERK (106). The main CD44 ligand is hyaluronic acid, a component of the extracellular matrix, which is positively associated with OC migration and metastatic spread (107, 108). CD44 is expressed in both normal and ovarian CSCs and it is associated with sphere-forming ability, self-renewal, chemoresistance, tumorigenicity, proliferation and invasiveness (109). Additionally, high expression of CD44 correlates with epithelial-to-mesenchymal transition (110) and with stem celllike properties, contributing to tumor invasion, metastasis, disease recurrence and chemoresistance (111). The co-expression of CD44 with the c-Kit receptor CD117 was shown to define an ovarian subpopulation with tumor-initiating capacity (72). C-kit regulates survival, proliferation and chemoresistance of ovarian CSCs through PI3K/AKT and Wnt/β-catenin-ATPbinding cassette G2 signaling (112).

CD24 is a glycosylphosphatidylinositol-anchored membrane glycoprotein recognized as a positive or negative marker for CSCs in numerous cancer types. The low expression of CD24 combined with high CD44 (CD44+/CD24−/low) is used to identify breast CSCs (46). On the contrary, several data indicated CD24 as a positive marker for ovarian CSCs. A population of CD24+ cells isolated from OC samples and cell lines was shown to display high expression of stemness-related genes, high tumor initiation and fast tumor growth along with increased sphereforming ability (113, 114). Besides, Davidson and colleagues demonstrated that tumor cells collected from OC peritoneal fluids exhibited higher levels of CD24 than solid tumors, suggesting an enrichment of CSCs (115).

Epithelial cell adhesion molecule (EpCAM), also called CD326, is a transmembrane glycoprotein overexpressed in many carcinomas (116). Several studies described a controversial role of EpCAM in carcinogenesis. Being an adhesion molecule, EpCAM regulates homophilic adhesion interactions, which might inhibit metastatic invasion (116). However, depending on the microenvironment, EpCAM is also able to suppress E-cadherin adhesions, supporting metastasis formation. EpCAM has been recognized as an additional marker for CSCs in different tumor types (55). High expression of EpCAM in OC is associated with tumor recurrence, chemoresistance and poor patient prognosis (117). Cells co-expressing CD24, CD44, and EpCAM showed stem cell-like characteristics, high tumorigenicity in vivo as well as enhanced migration, invasion and colony-formation in vitro and this population could be enriched by chemotherapy (50, 118).

There are also several stem cell-associated genes that play a role in the maintenance and development of ovarian CSCs. LIN28, OCT4, SOX2, and NANOG transcription factors (TFs) regulate and maintain pluripotency in embryonic stem cells (119). The co-expression of LIN28 and OCT4 identifies a population of ovarian CSCs and correlates with advanced tumor grade (120). NANOG is closely associated with HGSC, high chemoresistance and poor overall patient survival (121). SOX2 is required to maintain ovarian CSCs and its expression correlates with chemoresistance and poor prognosis (122).

### Mechanisms of Cancer Stem Cell Therapy Resistance

Conventional therapies are often not sufficient to eliminate CSC populations because of intrinsic resistance mechanisms and epigenetic plasticity of the cells (123, 124). This means in many cases, tumor relapse is caused by the incomplete elimination of all CSCs, since a single CSC seems to be sufficient to regrow a tumor [**Figure 2**; (125)].

Several molecular mechanisms, such as enhancement of DNA repair and DNA damage response (DDR), increased drug efflux, efficient scavenging of reactive oxygen species (ROS) and signals from the microenvironment, support the resistance of CSCs toward conventional cancer therapies [**Figure 3**; (126)].

#### Increased DNA Pepaire Capacity

Radiation and many chemotherapeutic drugs (e.g., DNA crosslinkers, DNA synthesis, and topoisomerase inhibitors) induce DNA damage, which, if not repaired, can lead to cell death (127). Among different types of DNA damages caused by radiation and chemotherapy, DSBs are the most lethal lesions if not repaired (128). DSBs are usually repaired by error-free homologous recombination (HR) or error-prone non-homologous end joining (NHEJ) mechanisms (129). In response to radiation, DNA repair is initiated and controlled by the DNA damage response (DDR). For the purpose of stalling cell cycle progression and give time to the cells to repair DNA damage, DDR triggers the activation of checkpoint kinase signaling pathways such as ataxia telangiectasia mutated (ATM)-checkpoint kinase 2 (Chk2) and ATM-Rad3-related (ATR)-checkpoint kinase (Chk1) (130, 131). CSC populations of glioblastoma, prostate, lung, breast cancer and many others, have been shown to possess high DNA repair capacity, mainly due to increased activation of ATR-Chk1 and ATM-Chk2 pathways (130). Recent reports demonstrated increased HR-dependent DNA repair proficiency of CSCs in OC that makes them more resistant to PARP inhibition (132). Different populations of OC CSCs have a high resistance to the platinum agents due to the altered regulation of cell cycle checkpoint, upregulation of the Fanconi Anemia DNA repair proteins (FANCD2, FANCJ) (133), MLH1, and BRCA1 (134) and increased expression of DNA polymerase eta (135).

### ROS Scavenging

Radiation-induced damages include the formation of ROS, which are chemically reactive free radicals produced by oxygen metabolism. ROS are involved in many physiological signaling pathways regulating metabolism, cell proliferation, migration, angiogenesis and wound healing (136). However, in high amounts, ROS can produce oxidative DNA damage, alter proteins and cell membrane lipid bilayer, leading to cell arrest and cell death (137). The cells control ROS levels through ROS scavenging molecules (e.g., glutathione peroxidase, superoxide dismutase, and catalase), which balance the production and the elimination of these products (137). CSCs isolated from different tumors exhibit more efficient ROS scavenging systems and a lower level of ROS production compared to non-CSC

populations (138, 139). ALDH positive population in OC has an upregulation of NRF2 (Nuclear factor erythroid 2-like) signaling driving the cytoprotective response to the oxidative stress (140).

### Enhanced Drug Efflux by ABC Transporters

Another important mechanism for CSC drug resistance is the high expression of ATP-binding cassette (ABC) transporters. The human genome encodes 49 ABC genes and these proteins are essential to maintain homeostasis and to protect the cells against environmental insults in many normal tissues (e.g., liver, intestine, blood-brain barrier, placenta, kidney, and normal stem cells) (141). Three ABC transporter proteins, P-glycoprotein (Pgp, MDR1, ABCB1), multidrug resistance protein 1 (MRP1, ABCC1), and breast cancer resistance protein (BCRP, ABCG2), have been shown to play a role in multidrug resistance of various types of cancers like OC, breast, lung, colon and others (142). Due to broad substrate spectrum specificity, their expression provides tumor resistance toward the major classes of chemotherapeutic drugs (126). CSC populations of different tumors, including breast cancer, lung cancer, retinoblastoma, neuroblastoma, and glioblastoma, have been demonstrated to overexpress ABC transporters, which correlates with high drug resistance (143).

### Aldehyde Dehydrogenase Activity and Activation of Developmental Pathways

Supplementary intrinsic mechanisms underlying CSC therapy resistance are mediated by high ALDH activity and activation of developmental pathways essential for embryonic development and tissue homeostasis like the canonical Wnt/β-catenin, Notch and Hedgehog pathways (126). Notch signaling contributes to survival and platinum resistance of ovarian CSCs and it has been determined that Notch 3 expression is associated with poor prognosis for OC patients (144). Hedgehog signaling is aberrantly activated in OC and this pathway affects cell growth, motility, invasion, and tumorigenesis (145). Wnt/β-catenin is involved in stem cell proliferation and differentiation. CSCs of many tumors were found to overexpress β-catenin, which promotes stemness (146). In OC, genetic mutations in the Wnt pathway are rare, but recent data demonstrated that the activation of Wnt signaling could be regulated by the tumor microenvironment, contributing to ovarian tumorigenesis (147).

### Cancer Stem Cell Niche and Their Microenvironment

Growing bodies of evidence showed that microenvironmental stimuli coming from the specific niche in which CSCs reside could influence and regulate treatment responsiveness (**Figure 4**). Additionally, autocrine signaling and stimuli coming from stromal fibroblasts, immune cells and extracellular matrix (ECM) as well as oxygen, nutrient supply and tissue pH, might affect CSC properties and metastatic dissemination (136).

Matrix metalloproteinases (MMPs) are extracellular proteases able to degrade and cleave ECM molecules, remodeling CSC niche and releasing growth factors and cytokines, thus, being suggested to regulate signaling pathways that control proliferation, differentiation and tumor invasion (148).

Tumor and stroma-derived growth factors and cytokine regulating stemness and resistance comprise interleukin-6 (IL-6), interleukin-8 (IL-8), chemokine (C-X-C motif) ligand 12 (CXCL12), chemokine (C-X-C motif) receptor type 4 (CXCR4), chemokine (C-C motif) ligand 2 (CCL2), platelet-derived growth factor (PDGF), transforming growth factor-beta 1 (TGF-β1), tumor necrosis factor-alpha (TNFα), epidermal growth factor (EGF), VEGF and fibroblast growth factor (FGF) (62, 126). IL-6, which is secreted by tumor-associated stroma has been shown to play an important role in the enrichment of OC stem cells after treatment with platinum agents (149). Dual expression of CXCR4 and CD133 has been used to identify an OC population with stem cell-like properties that regulate tumor development and chemoresistance (92). The formation of new vessels, also known as angiogenesis, supports tumor growth and it is tightly regulated by angiogenic activators including VEGF, FGF, PDGF, and EGF (150). Bao et al. determined that glioblastoma CD133+ CSCs highly express VEGFA, which contributes to the angiogenic process by interacting with the microenvironment (151). A recent study revealed that VEGFA stimulates OC stem cells by activation of the epigenetic mechanisms inducing loss of miR128- 2 and upregulation of Bmi1 (152).

The association between inflammation and cancer development has been suggested for numerous types of tumors (153). The accumulation of proliferating tumor cells mimics a chronic inflammation and the activated stroma cells produce a number of cytokines, angiogenic factors and chemokines, which in turn recruit more immune cells including monocytes and macrophages. The most common inflammatory molecules released in the microenvironment by tumor-associated macrophages (TAMs) are the EMT-inducers TGF-β1 and TNFα (154). EMT confers stem cell-like phenotype to cancer cells, enhancing motility and drug resistance.

In addition, tissue oxygenation influence CSC properties, maintaining cells in a quiescent (hypoxia) or activated (normoxia) state. For example, hypoxia is a major stimulant of CXCL12, which is secreted by tumor stroma fibroblasts and is involved in the early stage of malignant transformation of OC (139). Hypoxia makes CSCs more resistant to various environmental insults and it is the major stimulant of angiogenesis. Hypoxia-inducible factor (HIF) target genes, such as GLUT-1, VEGF, OCT4 and Notch, are crucial regulating proliferation, tumorigenicity and self-renewal in different cancers (155). HIF induces stem cell properties in OC cells (156) and promotes OC stem cells adaptive stress response (157) and resistance to therapy (158). Hypoxia has been demonstrated to promote the acquisition of stem cell-like state through the increased expression of CSC markers like CD133 and CD44, along with CSC-related genes such as SOX2 (155). Moreover, hypoxia is also tightly associated with chemoand radioresistance. Low oxygen conditions maintain CSCs in a quiescent state with a low proliferation rate and since most conventional cytotoxic drugs target proliferating cells, CSCs cannot be destroyed. It is also recognized that the local concentration of oxygen improves radiotherapy efficacy since DNA lesions caused by ROS react with oxygen-generating stable DNA peroxides (159). Tumor oxygenation and oxygen therapeutics have been utilized as radiosensitizers to improve patients' response to radiotherapy (160).

### Epithelial-To-Mesenchymal Transition and Stemness

In recent years, high plasticity of CSCs and stemness have been increasingly linked to EMT. EMT is an essential cellular process usually involved in embryogenesis, wound healing and tumorigenesis. During tumor progression, neoplastic cells switch from an epithelial-like state to gain mesenchymal properties. Therefore, EMT allows cancer cells to become migratory and invasive, acquiring multiple properties associated with highgrade malignancies (161). EMT is suggested to be a reversible process where mesenchymal-like cells might transition back into an epithelial state, a process called mesenchymal-to-epithelial transition (MET). A number of TFs, referred as EMT-TFs, are associated with EMT and can be classified into three main protein families: Snail (Snail and Slug), ZEB (ZEB1 and ZEB2), and basic helix-loop-helix (TWIST1, TWIST2, and TCF3) families (162). These TFs coordinate changes in gene expression resulting in suppression of genes associated with the epithelial state, like E-cadherin, and in upregulation of genes correlated with the mesenchymal state, including N-cadherin and vimentin (162). The activation of some of these TFs, including Slug, ZEB1, ZEB2, TWIST1, and Snail1, has been linked to the expression of stem cell-related genes and self-renewal in various tumor types (163). The EMT process is controlled by a variety of cytokines and growth factors and among them TGF-β signaling plays a fundamental role. TGF-β signaling regulates ovarian CSCs through the modulation of tissue transglutaminase 2 (TGM2), which promotes EMT, metastasis and stem celllike phenotype (164, 165). Hypoxia, through the expression

of HIF-1α and HIF-2α, regulates CSC-associated genes and induces EMT via the TGF-β signaling pathway, which in turn contributes to stemness (166). Cancer cells with an intermediate EMT phenotype display stem cell-like properties accompanied by enhanced resistance to several anti-cancer drugs [**Figure 5**; (124)]. In 2011, Strauss and colleagues identified a minor population of OC cells in a transitory epithelial/mesenchymal hybrid stage (i.e., partial or intermediate EMT), which means that these cells expressed at the same time epithelial and mesenchymal markers, respectively linked to adhesion and migration (167). The observation that these cells drive tumor growth in vivo and have the capacity of self-renewal provided a link between stemness and phenotypic plasticity. Many other studies have highlighted a connection between partial EMT, drug resistance and tumor-initiating ability, showing that this phenotype enables the formation of highly metastatic circulating-tumor cells clusters (168). Therefore, EMT gives a hint about the plasticity model, showing that plasticity enhances invasion potential and tumorinitiating properties of cells that arrive at the metastatic site (169). Stemness is therefore not a fixed and inherent property but instead, CSCs and non-CSCs can bidirectionally interconvert between these two states (168).

### NOVEL THERAPEUTIC STRATEGIES FOR OVARIAN CANCER

High-throughput approaches have identified numerous potential therapeutic targets to treat cancer cells specifically. The current focus of anti-cancer drug discovery is the development of molecular targeted therapies rather than conventional cytotoxic drugs. A major challenge remains to find agents and strategies that select and eliminate the source of tumor recurrence after



Clinical trial information accessed via https://clinicaltrials.gov with National Clinical Trial Number (NCT Number).

therapy, along with bulk tumor cells. It became increasingly evident that CSCs are crucial for disease recurrence and chemoresistance, thus targeting these cells seems a promising way to reach complete tumor eradication.

#### Inhibition of Cancer Stem Cell-Associated Pathways

Novel therapies have been employed to target CSC-specific markers or pathways critical for regulation and maintenance of stem cell-like properties (**Table 2**).

Metformin hydrochloride is clinically used as an anti-diabetic drug for type 2 diabetes. Recent studies have reported its ability to enhance chemotherapy efficacy in different cancer types by targeting CSCs. Metformin showed a synergic effect with standard chemotherapeutic agents, reducing tumor relapse rate (170). Shank and colleagues elucidated a potential mechanism behind this effect, demonstrating that the treatment with metformin reduced ALDH<sup>+</sup> ovarian CSCs, proliferation and angiogenesis in vitro and in vivo. This effect was additive in combination with CDDP (171). The outcomes of combinational treatment with metformin and chemotherapy in patients with stage III and IV OC are currently investigated in phase II clinical trial (NCT02122185) (172). Treatment of OC cell lines with the FDA approved PARPi olaparib and rucaparib enriched cells with highly efficient DNA repair mechanisms. These cells expressed the CSC markers CD133 and CD117 (132). This might explain why patients who received PARPi treatment develop resistance in spite of a good initial therapy response. Several developmental pathways, including Notch, Wnt and Hedgehog, are crucial for self-renewal and regulation of CSCs (173). The inhibition of these pathways in combination with traditional chemotherapy has been proposed as a promising therapeutic strategy for recurrent diseases.

Sonidegib is an FDA approved Hedgehog inhibitor for basal cell carcinoma. Sonidegib was tested in phase I clinical trial in combination with PTX (NCT01954355) as a treatment for patients with advanced OC, demonstrating anti-tumoral activity. Based on this study, a recommended dose for phase II trial has been identified (174). The second Hedgehog inhibitor, Vismodegib, was used in phase II clinical trial as maintenance therapy for patients diagnosed with OC in second or third complete remission (NCT00739661) (175). However, no significant survival benefit was observed (5.8 months for placebo vs. 7.5 months for the treatment group), implying that the blockage of the Hedgehog pathway alone is probably not enough to destroy recurrent disease (175). The Wnt inhibitor Ipafricept reduces CSC frequency, promotes cell differentiation in patient-derived OC xenografts and in phase I clinical trial in association with carboplatin and PTX for recurrent platinumsensitive OC (NCT02092363), 82% of patients achieved a partial or complete response (176). Focal adhesion kinase (FAK) is a cytoplasmic protein overexpressed in CSCs of different tumors and it maintains interaction with stromal cells to activate intracellular signaling cascades (62). Defactinib, an oral inhibitor of FAK, exhibited modest activity in phase I clinical trial in combination with PTX for advanced OC (NCT01778803), suggesting that targeting the CSC niche might offer a suitable therapy strategy (177).

#### New Radioimmunotherapy Approaches

At the Center for Radiopharmaceutical Sciences of the Paul Scherrer Institute we developed the anti-L1 cell adhesion molecule (L1CAM) chimeric monoclonal antibody chCE7 for radioimmunotherapeutic (RIT) approaches to cancer. The L1CAM is a 200–220 kDa type I membrane glycoprotein of the immunoglobulin superfamily, containing six immunoglobulinlike domains and five fibronectin type III repeats followed by a transmembrane region and a highly conserved cytoplasmic tail (178–180). Firstly, L1CAM was identified as a protein of the nervous system and during brain development it was shown to play an important role in morphogenic events like neurite outgrowth, fasciculation, adhesion and neuronal cell migration (181). L1CAM is expressed in human peripheral nerve bundles and human kidneys, while it is absent in other human tissues comprising heart, lung, colon, liver, thymus or testis (181, 182). Low levels of L1CAM are also detectable in hematopoietic cells (181). L1CAM is overexpressed in various types of human cancers, including ovarian and endometrial carcinoma, colon cancer, melanoma and glioblastoma, and it usually correlates with advanced tumor stage and poor prognosis (179, 180). L1CAM expression in cancer supports motility and invasion, promoting aggressive tumor growth and metastasis formation (180). Additionally, L1CAM in combination with CD133 characterized a new ovarian CSC population (82). This protein exhibits static and motility-promoting functions, eliciting different signaling pathways depending on the binding partners; L1CAM homophilic interactions support static cell-cell binding while binding of L1CAM to integrin induces motility and invasiveness (180).

Indeed, the restricted expression of L1CAM allowed the use of anti-L1CAM mAbs for OC targeted therapies (183). Previous work revealed that chCE7 binds near the RGD sequence in the sixth Ig-like domain of human L1CAM, preventing its binding with integrin (184). RIT using copper-67 and the radiolanthanides lutetium-177 (177Lu) and terbium-161 (161Tb) was demonstrated to increase the efficiency of antibody-based L1CAM therapy in preclinical OC models (183, 185, 186). The antigen-antibody complex usually internalizes by endocytosis and the coupled metallic radionuclides are trapped in the cell (187). Therapeutic results for RIT in solid tumors (as adjuvant therapy after primary surgery and/or chemotherapy) have been modest during recent years and the main obstacles include low radiosensitivity, poor lymphatic drainage, limited diffusion of the antibody through tumor mass, poor vascularization and lack of homogeneous targeting (188). To overcome these limitations, several new strategies, including bioengineering development of different antibody formats to improve tumor penetration, pretargeting approaches, the use of alpha particle or Auger electron emitters, as well as dose fractionation, have been developed over the past years (188). Furthermore, RIT has been combined with other chemotherapeutics, including DNA-damaging agents (e.g., CDDP), microtubules-stabilizing agents (e.g., PTX) and protein kinase inhibitors (PKIs) providing cellular radiosensitization effects (189, 190). PTX enhances anti-L1CAM RIT effectiveness, blocking cancer cells in the radiosensitive G2/M cell cycle phase. Our group could recently demonstrate that in vivo combination therapy with <sup>177</sup>Lu-DOTA-chCE7 [177Lu was coupled to the mAb via the chelator 1,4,7,10-tetraazacyclododecane-N-N'-N"-N"' tetraacetic acid (DOTA)] and PTX significantly extended overall survival of ovarian tumor-bearing mice (55 days RIT+PTX vs. 18 days PTX and 29 days RIT) (189).

Different properties of alpha-particles and Auger electrons, including short path length and high density of released energy in tissue, make them appropriate for targeted therapies against small volume diseases like metastases (<1 cm diameter) and for eradicating radioresistant CSCs (191).

The low-linear energy transfer (LET) refers to the energy released over the track length of the radiation in biological tissues. At equal absorbed dose, high-LET particles result in relatively high deposition of energy in tumor cells, which renders these forms of radiation more cytotoxic than low-LET radiations (**Figure 6**).

Most beta-particles emit low-LET radiations of 0.1–1 keV/µm, while alpha-particles have high-LET values of 50– 230 keV/µm and Auger electrons intermediate-LET of 4 to 26 keV/µm (192). Low-LET of beta-particle emitters is due to intermediate energy (0.5–2.3 MeV) and long-range in biological tissues (0.5–12 mm), producing easily reparable sparse DNA lesions (193). Additionally, beta-particles kill tumor cells surrounding target cells independently from antigen expression, by the so-called "cross-fire" effect. Because of heterogeneous antigen expression among tumor cells, cross-fire irradiation is suitable when not all tumor cells can be specifically targeted by radiolabeled antibodies. However, the long path of these particles can also induce bone marrow toxicity. In contrast, alpha-particles have high energy (5–9 MeV) and intermediate path length (50– 100µm), which restricts their effect to 5–10 cell diameters (193). Auger electrons have low energy (0.001–1 KeV) and subcellular nanometer range (<1µm) (193). These characteristics minimize the non-specific irradiation of non-targeted surrounding healthy tissues. Most importantly, it makes it possible to locally deliver high-absorbed doses and lethally damage tumor cells, inducing densely localized DNA DSBs. Besides DNA damage,

these short-range electrons damage the cell membrane and elicit radiation-induced bystander effect (194). For this reason, alpha-particle and Auger electron-emitting nuclides are very attractive for RIT against therapy-resistant CSCs. Our group could recently demonstrate that L1CAM, combined with CD133, defines a new ovarian CSC population and L1CAM is responsible for the radioresistance of these cells (82). This cell population showed high spherogenic and clonogenic capabilities and highly expressed some CSC- and EMT-related genes. Additionally, these cells demonstrated high tumorigenicity, fast-tumor growth and self-renewal when transplanted in nude mice. CRISPR-Cas9 deletion of L1CAM demonstrated that L1CAM expression correlates with EMT phenotype. Since L1CAM is not expressed in normal stem cells and it is present in the bulk population of cancer cells, anti-L1CAM RIT using Auger electrons and alphaparticle emitters would be a promising new option for ovarian CSCs therapy. In this context, we could clearly demonstrate in a comparative anti-L1CAM RIT study using <sup>177</sup>Lu- and161Tblabeled mAb chCE7 that <sup>161</sup>Tb was better by 82.6% compared to <sup>177</sup>Lu under equitoxic conditions in a preclinical OC xenograft model (186). <sup>161</sup>Tb emits 16 times more Auger and conversion electrons per decay (3–50 keV) than <sup>177</sup>Lu. Both radiolanthanides have similar ß–energy of 134 keV (177Lu; mean) or 154 keV ( <sup>161</sup>Tb; mean). Data were taken from "National Nuclear Data Centre Brookhaven National Laboratory."

Some RIT approaches for OC using alpha-particle emitters showed their great potential to treat microresidual diseases in preclinical models (194, 195). In 2017, Kasten et al. used the mAb 376.96, which recognizes the B7-H3 epitope expressed on ovarian CSCs and on the bulk population, radiolabeled with lead-212 (212Pb) to treat tumor-bearing mice. Mice treated with <sup>212</sup>Pb-376.96, alone or combined with carboplatin, showed two to three times longer survival than control groups (195). Recent outcomes from a phase I clinical study indicated minimal toxicity of intraperitoneal (i.p) administration of <sup>212</sup>Pb-TCMCtrastuzumab in patients with advanced OC (196).

### CONCLUSIONS

The CSC heterogeneity and plasticity of CSC state is currently the largest obstacle to move CSC research toward clinical translation. Nevertheless, although the cancer stemness is now considered as a dynamic state rather than an entity, CSCrelated biomarkers might be a powerful tool for prediction of patients' clinical outcomes, and targeting of CSC OC population might prove beneficial for the treatment of this deadly disease. Common knowledge about ovarian CSCs has made remarkable progress over the last years; still, there are many limitations and challenges related to disease complexity and current experimental techniques. Future studies are expected to employ clinically relevant models such as patient-derived

#### REFERENCES


xenografts, orthotopic mice models, and organoid cultures. Genetically engineered mice models provide opportunity to trace and validate potential CSC populations in the context of a fully functional immune system and tumor stroma. The inherent heterogeneity of OC provides therapeutic challenges and it is further complicated by the acquired heterogeneity driven by microenvironmental and therapeutic pressure. Clinical trials should consider the high heterogeneity regarding CSC markers to select the proper patient cohort. CSC frequency and content should be monitored during clinical trials to assess therapy response. Considering the possibility of bulk tumor cell reprogramming, an optimal therapeutic regimen should combine therapeutic drugs showing wide cytotoxic effects on non-CSCs in combination and/or followed by a therapy targeting resistant CSCs (**Figure 7**) (197). To reach this goal, research needs to focus on the identification of new and reliable signaling pathways that influence CSC maintenance, differentiation, drug resistance, DNA damage repair, and their plasticity. In addition, it would be helpful to identify new CSC cell surface markers that are not expressed in normal stem cells but are also present in the bulk population of tumor cells, like L1CAM in ovarian CSCs. In recent years, there has been a veritable renaissance of radiotherapeutic approaches and new promising radioisotopes were introduced in pre- and clinical trials. Especially alpha- and Auger-emitters allow the sterilization of radioresistant cells such as cancer stem cells due to their high-LET.

### AUTHOR CONTRIBUTIONS

NT and JG designed the manuscript. NT wrote the first draft and illustrated the manuscript. NT, JG, FJ, and AD edited and reviewed the manuscript.

### FUNDING

This work was supported by the Swiss Cancer Research Foundation (KFS 3585-02-2015) to JG.

### ACKNOWLEDGMENTS

We would like to thank Prof. Dr. Roger Schibli, Prof. Dr. med. Viola Heinzelmann-Schwarz, and Dr. Martin Béhé for their support and useful scientific discussions.


signaling in hepatocellular carcinoma. Cancer Lett. (2017) 389:1–10. doi: 10.1016/j.canlet.2016.12.023


and induces Rac1 and Ras signaling during ovarian tumor cell migration and growth. J Biol Chem. (2001) 276:48679–92. doi: 10.1074/jbc.M106759200


generates residual cells with mesenchymal stem cell-like profile. J Cell Biochem. (2011) 112:2850–64. doi: 10.1002/jcb.23199


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Terraneo, Jacob, Dubrovska and Grünberg. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.