# PROCEEDINGS FROM ACCM19: CELL CYCLE, DNA DAMAGE RESPONSE AND TELOMERES

EDITED BY : Andrew Burgess and Liz Caldon PUBLISHED IN : Frontiers in Cell and Developmental Biology

#### 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 978-2-88966-115-2 DOI 10.3389/978-2-88966-115-2

#### 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

# PROCEEDINGS FROM ACCM19: CELL CYCLE, DNA DAMAGE RESPONSE AND TELOMERES

Topic Editors: Andrew Burgess, Anzac Research Institute, Australia Liz Caldon, Garvan Institute of Medical Research, Australia

Citation: Burgess, A., Caldon, L., eds. (2020). Proceedings From ACCM19: Cell Cycle, DNA Damage Response and Telomeres. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-115-2

# Table of Contents

*04 Editorial: Proceedings From ACCM19: Cell Cycle, DNA Damage Response and Telomeres*

Andrew Burgess and C. Elizabeth Caldon

*07 Why Be One Protein When You Can Affect Many? The Multiple Roles of YB-1 in Lung Cancer and Mesothelioma*

Thomas G. Johnson, Karin Schelch, Sunali Mehta, Andrew Burgess and Glen Reid


Winnie Tan, Sylvie van Twest, Vincent J. Murphy and Andrew J. Deans

*45 Two-Faced: Roles of JNK Signalling During Tumourigenesis in the*  Drosophila *Model*

John E. La Marca and Helena E. Richardson


Alvaro Gonzalez-Rajal, Jordan F. Hastings, D. Neil Watkins, David R. Croucher and Andrew Burgess


Nikita Panicker, Melody Coutman, Charley Lawlor-O'Neill, Richard G. S. Kahl, Séverine Roselli and Nicole M. Verrills


Jinbae Son, Katherine M. Hannan, Gretchen Poortinga, Nadine Hein, Donald P. Cameron, Austen R. D. Ganley, Karen E. Sheppard, Richard B. Pearson, Ross D. Hannan and Elaine Sanij

# Editorial: Proceedings From ACCM19: Cell Cycle, DNA Damage Response and Telomeres

Andrew Burgess 1,2 \* † and C. Elizabeth Caldon3,4 \* †

*<sup>1</sup> Cell Division Lab, ANZAC Research Institute, Concord Hospital, Concord, NSW, Australia, <sup>2</sup> Faculty of Medicine and Health, Concord Clinical School, University of Sydney, Sydney, NSW, Australia, <sup>3</sup> The Kinghorn Cancer Centre, Garvan Institute of Medical Research, Darlinghurst, NSW, Australia, <sup>4</sup> Faculty of Medicine, St. Vincent's Clinical School, UNSW Sydney, NSW, Australia*

Keywords: cell cycle, telomere, DNA damage, mitosis, JNK, radiation, chemotherapy, PP2A

**Editorial on the Research Topic**

**Proceedings From ACCM19: Cell Cycle, DNA Damage Response and Telomeres**

#### Edited and reviewed by:

*Philipp Kaldis, Lund University, Sweden*

#### \*Correspondence:

*Andrew Burgess andrew.burgess@sydney.edu.au C. Elizabeth Caldon l.caldon@garvan.org.au*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology*

Received: *27 July 2020* Accepted: *31 July 2020* Published: *03 September 2020*

#### Citation:

*Burgess A and Caldon CE (2020) Editorial: Proceedings From ACCM19: Cell Cycle, DNA Damage Response and Telomeres. Front. Cell Dev. Biol. 8:805. doi: 10.3389/fcell.2020.00805*

#### INTRODUCTION

The 19th ACCM meeting was held in Sydney, Australia, from the 17–19th of June 2019. The Australian Cell Cycle Meeting (ACCM) began in the late 1990s as a small Australian based workshop, to bring together cell cycle researchers from across the country once a year. A strong focus was placed on providing students and post-docs with the opportunity to give oral presentations of their work and establish new collaborations. It continued in this fashion for over a decade, establishing itself as a must-attend event for the local cell cycle community. In 2015, we expanded the meeting to incorporate the fields of DNA damage and telomere biology and switched to running every 2 years.

The once small local meeting has now become an internationally and respected meeting that attracts leading researchers from around the world. Our invited research leaders for 2019 were Agata Smogorzewska on genome maintenance (Rockefeller University, New York, USA), Agnel Sfeir on telomere biology (Skirball Institute of Biomolecular Medicine, New York University, Langone Medical Center, USA), Gerry Hanna on radiation oncology (Director of Radiation Oncology Peter MacCallum Institute Victoria, Australia), and Karlene Cimprich on genome stability and DNA replication (School of Medicine, Stanford University, Stanford, CA, USA). Importantly, the meeting maintained its early goal of providing a platform for students and junior scientists to present their research in a friendly and collaborative environment. The next meeting is scheduled for the end of 2021 and will return to Melbourne, where it commenced in 1999.

## CELL CYCLE, DNA REPAIR, AND TELOMERES: ELEMENTS THAT UNDERPIN CANCER BIOLOGY AND TREATMENT

The 2019 ACCM meeting brought together research covering DNA repair, telomere biology, RNA transcription, early developmental biology, cell-polarity and cell-signaling, and big-data. The exploration of these ideas in one forum provided an opportunity for in depth discussion of focused questions with crossfertilization of ideas from closely aligned research areas.

Two excellent reviews by Sia et al. and Martin and Martin delve into Radiation therapy (RT), which is used to treat more than half of all cancers. With recent technological advancements, RT is responsible for as much as 40% of all cancer cures. The review by Sia et al., examines the types and mechanisms of cell death induced by RT in cancer. RT induces a range of death processes, from classical apoptosis and mitotic cell death through to autophagy, which is mediated by numerous intrinsic pathways in combination with the local microenvironment and radiation specific factors. In their opinion piece, Martin and Martin, contextualize the use of radiation therapy treatment by discussing the notable side effects that many radiotherapy patients face, especially accelerated aging.

This work was complemented by two reviews focused on aspects of lung cancer. Work by Johnson et al., comprehensively examined the role of the oncogene Y-Box protein 1 (YB-1) in driving lung cancer. YB-1 is a multi-function protein capable of regulating transcription, translation, and DNA repair. It is commonly over-expressed in numerous cancers, and it is implicated in driving proliferation, metastasis, and resistance to chemotherapy. In parallel, Gonzalez-Rajal et al. have identified and reviewed the recent breakthrough studies that have shed light on the underlying mechanisms of innate platinum resistance in lung cancer. Notably, platinum chemotherapy is a cornerstone and front-line treatment for many lung cancers including mesothelioma. It is essentially curative in testicular cancer but is hampered by extensive innate resistance in lung cancer, with only 30% of patients responding. In this opinion article, Gonzalez-Rajal et al. highlight how TGFβ signaling, the cell cycle and DNA repair are key central players in regulating platinum resistance in lung cancer.

Three research manuscripts provided key findings on DNA repair and telomeres with respect to cancer biology mechanisms, improved cancer biomarkers, and methodological advancement. One essential pathway that regulates platinum resistance is the Fanconi Anaemia (FA) pathway, which is responsible for removing platinum induced interstrand crosslinks (ICLs) from DNA. Exciting new research by Tan et al. examines how two crucial components of the FA pathway, FANCD2 and FANCI, are regulated by direct phosphorylation by ATR kinase. This protects the FA complex from degradation, thereby ensuring cross-linked DNA is properly repaired. Notably, cancer cells often have defective DNA repair pathways making them susceptible to novel chemotherapies. One such treatment is the RNA polymerase I (PolI) transcription inhibitor CX-5461, which causes DNA damage and is currently in phase I clinical trials for solid tumors. Work from Son et al. demonstrates the number of active rDNA repeats positively correlates with sensitivity to CX-5461 in ovarian cancer cells, and hence may be a potential clinical biomarker for this exciting new chemotherapy. Notably, telomeres protect the chromosome ends from being recognized as DNA double-strand during normal replication. Disruption of telomeres is a hallmark of cancer, and hence the ability to accurately monitor telomere length is a critical assay not only for basic research but also clinical diagnosis. In exciting new research, Kahl et al. demonstrate the Telomere length Combing Assay (TCA), which can accurately measure telomere length in cell populations by pulling DNA fibers out onto glass coverslips using a constant stretching factor.

Telomeres and DNA repair also play essential roles in early developmental biology. Kafer and Cesare provide a comprehensive review of the replication stress and HR repair factors that are essential for early mammalian embryo development covering over 347 genes. Understanding embryo development and the use of developmental systems such as Drosophila (Fruit flies) has been essential for discovering and new drivers of cancer. Pre-eminent among these is the c-Jun N-terminal Kinase (JNK) signaling pathway, which plays a multitude of roles from regulating cell proliferation through to survival. Here, La Marca and Richardson provide a compelling review of how the JNK pathway acts as both promoter and inhibitor of tumorigenesis in Drosophila. Traditionally, early developmental biology, DNA repair, and cell cycle pathways have focused on the role of kinases, however, the essential role of counterbalancing phosphatases has recently gained prominence. This is highlighted in new research by Panicker et al., where they analyse the role of the major phosphatase PP2A. PP2A is a multi-complex phosphatase, with specificity for substrates controlled by the regulatory subunit (Rogers et al., 2016). Using CRISPR/Cas9, Panicker et al., demonstrate that knocking out the B55α regulatory subunit caused embryonic lethality in mice, due to failed epidermal stratification, highlighting the importance of phosphatases.

## SUMMARY

The ACCM conference and resulting special issue in Frontiers in Cell and Developmental Biology has highlighted the biological complexity that underpins cancer biology and treatment, and that major aspects of this biology are the cell cycle, DNA repair pathways, and telomere biology. A key aspect of our meeting is to bring together what are sometimes diverse fields and encourage the sharing of ideas that can accelerate knowledge gain and implementation. The studies published in this special issue cross over these fields and highlight the impact of cross-fertilization; examples include the importance of developmental biology on understanding of DNA damage pathways, and of how the basic understanding of the detection and repair of DNA lesions could alter cancer therapy.

## AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

## FUNDING

AB was supported by a NBCF Investigator Initiated Research Scheme (IIRS-18-103). CC was supported by a NBCF Career Development Fellowship (ECF17-002).

## REFERENCES

Rogers, S., McCloy, R., Watkins, D. N., and Burgess, A. (2016). Mechanisms regulating phosphatase specificity and the removal of individual phosphorylation sites during mitotic exit. Bioessays 38, S24–S32. doi: 10.1002/bies.201670905

**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 Burgess and Caldon. 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.

# Why Be One Protein When You Can Affect Many? The Multiple Roles of YB-1 in Lung Cancer and Mesothelioma

Thomas G. Johnson1,2,3,4, Karin Schelch<sup>5</sup> , Sunali Mehta6,7, Andrew Burgess2,3 and Glen Reid6,7 \*

<sup>1</sup> Asbestos Diseases Research Institute, Sydney, NSW, Australia, <sup>2</sup> Cell Division Laboratory, The ANZAC Research Institute, Sydney, NSW, Australia, <sup>3</sup> School of Medicine, The University of Sydney, Sydney, NSW, Australia, <sup>4</sup> Sydney Catalyst Translational Cancer Research Centre, The University of Sydney, Sydney, NSW, Australia, <sup>5</sup> Institute of Cancer Research, Medical University of Vienna, Vienna, Austria, <sup>6</sup> Department of Pathology, University of Otago, Dunedin, New Zealand, <sup>7</sup> Maurice Wilkins Centre, University of Otago, Dunedin, New Zealand

#### Edited by:

Dominic C. Voon, Kanazawa University, Japan

#### Reviewed by:

Joseph William Landry, Virginia Commonwealth University, United States Shang Li, Duke-NUS Medical School, Singapore

> \*Correspondence: Glen Reid glen.reid@otago.ac.nz

#### Specialty section:

This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology

Received: 26 July 2019 Accepted: 18 September 2019 Published: 01 October 2019

#### Citation:

Johnson TG, Schelch K, Mehta S, Burgess A and Reid G (2019) Why Be One Protein When You Can Affect Many? The Multiple Roles of YB-1 in Lung Cancer and Mesothelioma. Front. Cell Dev. Biol. 7:221. doi: 10.3389/fcell.2019.00221 Lung cancers and malignant pleural mesothelioma (MPM) have some of the worst 5 year survival rates of all cancer types, primarily due to a lack of effective treatment options for most patients. Targeted therapies have shown some promise in thoracic cancers, although efficacy is limited only to patients harboring specific mutations or target expression. Although a number of actionable mutations have now been identified, a large population of thoracic cancer patients have no therapeutic options outside of first-line chemotherapy. It is therefore crucial to identify alternative targets that might lead to the development of new ways of treating patients diagnosed with these diseases. The multifunctional oncoprotein Y-box binding protein-1 (YB-1) could serve as one such target. Recent studies also link this protein to many inherent behaviors of thoracic cancer cells such as proliferation, invasion, metastasis and involvement in cancer stem-like cells. Here, we review the regulation of YB-1 at the transcriptional, translational, posttranslational and sub-cellular levels in thoracic cancer and discuss its potential use as a biomarker and therapeutic target.

#### Keywords: lung cancer, mesothelioma, targeted therapy, biomarker, Y-box binding protein-1

## INTRODUCTION

Lung cancers are the leading cause of cancer death worldwide (Islami et al., 2015; Kris et al., 2017), and malignant pleural mesothelioma patients continue to experience some of the worst 5-year survival rates of all malignancies (Mutti et al., 2018). Therefore, advances in therapeutic options are urgently needed and require a more thorough understanding of the underlying biology of both.

While SCLC represents ∼15–20% of all lung cancers, NSCLC represent the majority of cases (∼80–85%). NSCLC are further subtyped into adenocarcinomas (ADC; ∼40–50% of NSCLC), squamous cell carcinomas (SCC; ∼20–40%) and large cell carcinomas (LGC; ∼20%). Whilst all of these carcinomas are significantly associated with tobacco consumption, this association is much stronger in SCLC and SCC than in ADC and LGC (Khuder, 2001).

Malignant pleural mesothelioma arises from the pleural linings of the lung and is strongly linked to asbestos exposure (Tossavainen, 2004). MPM is currently subtyped as epithelioid, sarcomatoid or biphasic, which are characterized by a mixture of epithelioid and sarcomatoid cells (Marshall et al., 2015). At times, this review refers to lung cancer and mesothelioma as "thoracic cancers," although we acknowledge that this term also encompasses tumors of the trachea, esophagus and thymus.

The current clinical practice guidelines for NSCLC, SCLC, and MPM all recommend the use of platinum-based chemotherapy in combination with other agents as the standard mode of care (Vogelzang et al., 2003; Rudin et al., 2016; Bradbury et al., 2017; Kris et al., 2017; Nagasaka and Gadgeel, 2018; Szolkowska et al., 2018). Diagnosis in the early stages of NSCLC affords better survival odds, however, the majority of patients are diagnosed with advanced disease (Kris et al., 2017; Visconti et al., 2017). Such individuals face a 5-year survival rate of only 23% and treatment options are often limited to chemotherapy (Kris et al., 2017). SCLC patients face similarly poor survival odds. Patients usually respond initially to platinum-based chemotherapy but inevitably develop chemoresistant tumors (Rudin et al., 2016). Overall survival rates of SCLC patients currently sit at 10–12 months post diagnosis (Rudin et al., 2016). In MPM, the standard of care consists of a combination of cisplatin with pemetrexed, providing an overall survival rate of only 12.1 months (Vogelzang et al., 2003; Mutti et al., 2018). Epithelioid mesotheliomas present with the best prognosis, with the median overall survival being between 12 and 27 months (Yap et al., 2017). Patients with biphasic mesothelioma have median overall survival rates of 7– 18 months, while sarcomatoid patients are afforded the worst prognosis of 4–12 months (Yap et al., 2017). Recent trials of immunotherapy strategies, such as the anti-PD-1 checkpoint inhibitors pembrolizumab and nivolumab, have shown promise as first-line and second-line therapies in some thoracic cancers (Visconti et al., 2017; Forde et al., 2019). However, response to immunotherapy is unpredictable due to a lack of robust biomarkers, so predicating which patients will respond is not yet possible (Ventola, 2017). Acquired resistance to these drugs also remains a significant problem (Ventola, 2017). Improved treatment options for patients suffering malignancies of the lung and mesothelial linings are therefore still desperately needed.

#### Toward Personalized Therapy for Thoracic Cancer Patients

The development of next-generation sequencing has fostered a deeper understanding of the molecular drivers and mutational landscape of thoracic cancers. Multi-region whole-exome sequencing of 100 early stage NSCLC patients demonstrated that clonal alterations of oncogenes such as the growth receptor EGFR and the kinases MET, and BRAF were commonly found in ADC (Jamal-Hanjani et al., 2017). These were accompanied by sub-clonal modifications of the oncogene PIK3CA and the tumor suppressor neurofibromin 1 (Jamal-Hanjani et al., 2017). Alterations of PIK3CA, the transmembrane receptor NOTCH1, growth factor receptor FGFR1 and transcription factor SOX2 were also observed in early SCC (Jamal-Hanjani et al., 2017). TP53 or p53 mutations were frequent clonal events in both subtypes, while oncogenic ALK translocations were not observed in any tumors (Jamal-Hanjani et al., 2017).

As for MPM, next-generation sequencing of 216 MPM patients showed that the tumor suppressors BAP1, NF2, and SETD2 were significantly mutated through gene fusions and splicing alterations (Bueno et al., 2015). CDKN2A, which encodes the tumor suppressor p16INK4a, is also frequently deleted in up to 75–90% of MPM cases (Ladanyi, 2005; Sementino et al., 2018). Data from TCGA reflects the above findings, apart from ALK alterations in ADC, which were present in 7% of cases (**Figures 1A–C**). An important distinction must between lung cancer and MPM is that lung cancers are generally characterized by an increase in oncogenic drivers, while MPM appears to be more commonly defined by loss of tumor suppressors (Ladanyi, 2005; Bueno et al., 2015; Jamal-Hanjani et al., 2017; **Figure 1C**). This makes identifying new therapeutic targets in MPM more challenging. Apart from bevacizumab, which targets vascular endothelial growth factor A, no targeted therapies are currently available to MPM patients (Brosseau et al., 2017).

**Abbreviations:** 1NP63α, Isoform p63α – protein isoform of p63; ADC, lung adenocarcinoma; AKT, lrotein kinase B; ALK, anaplastic lymphoma kinase; APE1, apurinic/apyrimidinic endonuclease 1; BAP1, BRCA associated protein-1; Bcl-2, B-cell lymphoma 2; BER, base-excision repair; BMP7, bone morphogenetic protein 7; BRAF, proto-oncogene B-Raf; C1QBP, complement component 1 Q subcomponent-binding protein; CAR10, CAR intergenic 10; CCND1, cyclin D1; CDC25A, M-phase inducer phosphatase 1; CDKN2A, cyclin-dependent kinase inhibitor 2A; COPD, chronic obstructive pulmonary disease; CRS, cytoplasmic retention signal; CSC, cancer stem-like cell; CSD, cold shock domain; CTD, C-terminal domain; DANCR, differentiation antagonizing non-protein coding RNA; E/M, hybrid epithelial/mesenchymal state; EGF, epidermal growth factor; EGFR, epidermal growth factor receptor; EMT, epithelial-mesenchymal transition; EPO, human erythropoietin; EZH2, enhancer of zeste homolog 2; FGFR1, fibroblast growth factor receptor 1; FOXO3a, forkhead box O3; G3BP1, Ras GTPase-activating protein-binding protein-1; GAS5, growth arrest specific 5; HACE1, HECT domain and ankyrin repeat containing E3 ubiquitin protein ligase 1; HER2, human epidermal growth factor receptor 2; HIF1α, Hypoxiainducible factor 1-α; HMGB1, high mobility group box 1 protein; HULC, highly up-regulated in liver cancer; LGC, large cell carcinoma; LINC00312, long intergenic non-protein coding RNA 312; LMO3, LIM domain only protein 3; lncRNA, long non-coding RNA; LRP/MVP, lung resistance protein/major vault protein; MAPK/ERK, Mitogen-activated protein kinase/extracellular-signalregulated kinase; MDCK, Madin-Darby canine kidney cell; MDR1, multi-drug resistance 1; MET, tyrosine-protein kinase Met; MIR22HG, MIR22 host gene; miRNA, microRNA; MPM, malignant pleural mesothelioma; MRP1, multidrug resistance-associated protein 1; mTOR, mammalian target of rapamycin; NANOG, homeobox protein NANOG; ncRNA, non-coding RNA; NEIL1, nei like DNA glycosylase 1; NER, nuclear excision repair; NF2, neurofibromatosis type 2; NLS, nuclear localization signal; Notch1/NOTCH1, notch homolog 1; Notch3, notch receptor 3; NSCLC, non-small cell lung cancer; nt, nucleotide; Oct4, octamerbinding transcription factor 4; OGT, O-linked N-acetylglucosamine transferase; PABP, Poly(A)-binding protein; PARP1, Poly(ADP-ribose) polymerase 1; PARP2, Poly(ADP-ribose) polymerase 2; PCNA, proliferating cell nuclear antigen; PIK3CA, phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit α; RBBP6, retinoblastoma binding protein 6; RSK p90, ribosomal S6 kinase; SAHA, suberoylanilide hydroxamic acid; SCC, lung squamous carcinoma; SCLC,small cell lung cancer; SETD2, SET domain containing 2; SILAC, stable isotope labeling with amino acids in cell culture; siRNA, small interfering RNA; Snail, Zinc finger protein SNAI1; SOX2, sex determining region Y-box 2; SPHK1, sphingosine kinase 1; SRC3, steroid receptor co-activator 3; ssDNA, single-stranded DNA; TCGA, The Cancer Genome Atlas; TGFβ1, transforming growth factor β1; TP53/p53, tumor protein 53; TP53TG1, TP53 target 1; Twist1, Twist-related protein 1; XPC, xeroderma pigmentosum; YB-1, Y-box binding protein-1; YB-1/p18, 18 kDa fragment of YB-1.

and RPPA protein Expression for (A) Lung Adenocarcinoma (ADC; n = 584), (B) Lung Squamous Cell Carcinoma (SCC; n = 511) and (C) Mesothelioma (MPM; n = 87). Panels (A–C) were adapted from the open-source platform cBioPortal for Cancer Genomics (cBioPortal.org). (D) High YBX1 expression correlates with poor prognosis in NSCLC patients (p = 1.5 × 10−10). Kaplan-Meir plot of 1,926 NSCLC patients generated using Lung Cancer KM plotter. Univariate analysis with probe set 20862\_s\_at (YBX1) using auto-selected cutoff and excluded biased arrays. (E) High YBX1 expression correlates with poor prognosis in MPM patients (p = 8.6 × 10−<sup>3</sup> ). Kaplan-Meir plot was generated using PROGgene V2 with the TCGA mesothelioma dataset (n = 83) using "DEATH" as the survival measure and median as the cutoff.

The story for SCLC patients is similar with no breakthrough changes in treatment in over 25 years despite decades of research. The only exception to this is the approval of topotecan as a second-line therapy (Hirsch et al., 2017), and immunotherapy, which has shown some promise in Phase I/II trials in PD-L1 positive relapsed SCLC patients (Ott et al., 2015). Unfortunately, immunotherapy success has been limited by rapid disease progression, which can result in patient death before an effective anti-tumor response has time to occur (3–6 months), and severe immuno-related toxicities (encephalitis or myasthenia gravis) that are already highly associated with SCLC (Oronsky et al., 2017). Other drugs such as PARP inhibitors and

transcription inhibitors have shown some preclinical promise, but have yet to translate into clinical benefits for SCLC patients (Oronsky et al., 2017).

For NSCLC, targeted therapies have provided promising, albeit limited, results. The best known of these are the EGFR tyrosine kinase inhibitors such as erlotinib and osimertinib, which have proved effective for EGFR mutant ADC tumors (Hirsch et al., 2017; Winther-Larsen et al., 2019). In the ADC TCGA dataset, 21% of patients had EGFR alterations (**Figure 1A**), although the occurrence of EGFR mutations can vary between populations in ADC and NSCLC as a whole. For example, while EGFR mutation can occur in up to 40% of all NSCLC patients of Asian descent, the frequency of mutation in non-Asian NSCLC populations drops to 10–20% (Hirsch et al., 2017). Another problem is that response to EGFR inhibitors is almost always followed by the emergence of resistance (Hirsch et al., 2017). ALK inhibitors are similarly effective in patients harboring ALK translocations (Hirsch et al., 2017), present in 2% of all NSCLC patients (Hirsch et al., 2017). Alterations of ALK in ADC tumors specifically is found in up to 7% of cases, according to TCGA data (**Figure 1A**). Inhibitors targeting BRAF mutant tumors (3–5% of lung cancers), MET overexpressing tumors (3– 4% of ADC cases) and tumors harboring RET fusion proteins (1–2% of NSCLC) are also currently undergoing preclinical and clinical studies (Hirsch et al., 2017). The remaining majority of patients with ADC have no known actionable targets.

Patients with SCC have even fewer options with only ∼13% of SCC tumors reported to harbor at least one currently actionable alteration (Lindquist et al., 2017). There is no subset of patients known to benefit from targeted drugs at the moment, although there is some benefit from immunotherapy (Hirsch et al., 2017; Friedlaender et al., 2019). TP53 mutations are a common alteration in SCC patients (33%; **Figure 1B**), however, existing targeted TP53 treatments have proven ineffective (Friedlaender et al., 2019). PIK3CA is also frequently altered in thoracic cancer, particularly in SCC (Friedlaender et al., 2019) (71%; **Figure 1B**), indicating that it may have significance as a therapeutic target. However, despite promising preclinical studies of PIK3CA inhibitors, the benefit of these drugs appears to be negligible in trials with NSCLC patients (Friedlaender et al., 2019). This has also been the case in other cancers where, generally, patients show limited response and many experience prohibitive toxicity (Janku et al., 2018). This pathway mediates a multitude of downstream effects, which may attest to the observed relative ineffectiveness of PIK3CA inhibitors in lung cancer. FGFR1 amplification occurs in 20–25% of SCC cases (Friedlaender et al., 2019; **Figure 1B**), but again, targeting it in the clinic has provided limited efficacy and its potential as a viable target remains under contention (Hirsch et al., 2017; Friedlaender et al., 2019). There are a few targets that have been the focus of preclinical studies showing promising results, such as the transcription factor SOX2. SOX2 is involved in cell lineage-survival (Friedlaender et al., 2019) and is often upregulated in SCC (Karachaliou et al., 2013; Friedlaender et al., 2019) (60%; **Figure 1B**), as well as SCLC (Rudin et al., 2012; Karachaliou et al., 2013) and to a lesser degree ADC (Karachaliou et al., 2013). Finding such targets and translating them to the clinic is essential to improve outcomes for patients with SCC.

The heterogeneity of thoracic cancer biology makes finding clinically relevant therapeutic targets inherently difficult. Identifying other penetrant driver events in thoracic cancers may uncover alternative targets, which could yield more therapeutic options for patients down the line. One such potential target is YB-1. YB-1 is downstream of the commonly dysregulated PI3K-AKT-mTOR pathway, so targeting it may refine the effects of inhibiting this signal cascade. Thus, anti-YB-1 agents may provide more tumor-specific results than their upstreamtargeting counterparts, such as PIK3CA inhibitors (Janku et al., 2018). Adding to this, YB-1 upregulates PIK3CA at the transcriptional level in breast cancer (Astanehe et al., 2009). This implies that YB-1 may be involved in a feed-forward loop with the PI3K-AKT pathway and that targeting it could be an effective strategy in tumors with heightened PIK3CA, such as SCC. YB-1 is also upstream of SOX2 (Jung et al., 2014) and a host of other oncogenic drivers (Lasham et al., 2013), so the downstream effects of YB-1 inhibition may still be broad enough to make it an interesting candidate. This review therefore outlines recent literature focusing on YB-1 in cancer and makes the case for its possible use as a biomarker and future therapeutic target in thoracic malignancies.

### Y-Box Binding Protein-1 in Thoracic Cancers: An Overlooked Target?

Y-box binding protein-1, encoded by the YBX1 gene, is a multifunctional oncoprotein involved in many hallmarks of cancer development including driving proliferation, invasion and metastasis, CSC biology, resistance to chemotherapy, hypoxic response, DNA repair and exosomal sorting. Despite these links YB-1 has received limited attention as a therapeutic target or biomarker in oncology (Lasham et al., 2013). Although mutations of YBX1 are rare [∼1% in all cancers types (Cerami et al., 2012; Gao et al., 2013)], overexpression of YB-1 is found in a wide range of cancers and is often associated with poor prognosis (Lasham et al., 2013), including NSCLC and MPM. Analysis of TCGA data shows that elevated YBX1 expression was highly prognostic in a cohort of 1,926 NSCLC patients (Gyorffy et al., 2013 ˝ ; **Figure 1D**) and in 83 mesothelioma patients (Goswami and Nakshatri, 2014; **Figure 1E**). This supports the results of a recent meta-analysis of data from 692 NSCLC patients which found that high YB-1 protein expression significantly correlated with poorer overall survival and clinicopathological features (Jiang et al., 2017). YB-1 is overexpressed in mesothelioma compared to non-malignant mesothelial cells in vitro (Johnson et al., 2018) and a small study of 33 MPM patients showed a trend toward higher YB-1 expression in sarcomatoid MPM tumors, which are associated with shorter survival (Iwanami et al., 2014). Unfortunately, TCGA data is currently not available for SCLC, likely because surgically resected tissue specimens are relatively rare (Byers and Rudin, 2015) and, to our knowledge, a prognostic study on YB-1 expression in SCLC is yet to be conducted.

In the above datasets, alterations were seen in 12, 11, and 8% in ADC, SCC, and MPM, respectively, and mRNA upregulation was predominant (**Figures 1A–C**). While only PIK3CA and SOX2 were significantly co-expressed in the SCC dataset (q < 0.001;

**Figure 1B**), notably, this analysis did not show YBX1 alteration to be significantly associated with the current targetable oncogenes ALK, BRAF or EGFR in ADC (**Figure 1A**), despite there being a small proportion of tumors that had elevated levels of both YBX1 and EGFR. This suggests that YB-1 deregulation may represent a unique subpopulation of patients that may not have a targetable mutation. This combined with the prognostic significance of YB-1 in NSCLC and MPM, suggests that YB-1 may be a clinically relevant target worthy of further investigation.

### YB-1: A MALIGNANT JACK OF ALL TRADES

#### A Driver of Malignant Phenotypes

Y-box binding protein-1 was first discovered as a negative transcriptional factor of major histocompatibility complex Class II where it binds to the Y-box (5<sup>0</sup> -CTGATTGG-3<sup>0</sup> ) (Didier et al., 1988). Further investigation found that YB-1 stimulated the transcription of a wide variety of genes, including important oncogenes such as EGFR and HER2 (Lasham et al., 2013). YB-1 is a part of the cold-shock protein superfamily and contains a conserved nucleic acid binding region termed the CSD (Wolffe et al., 1992; **Figure 3**). Along with the CSD, YB-1 is comprised of two other highly disordered domains, the alanine/proline rich variable N–terminal domain and the C–terminal domain (CTD), each facilitating different biological interactions (Lyabin et al., 2014; Suresh et al., 2018). This versatility affords YB-1 a range of functions including transcriptional regulation, DNA repair and pre-mRNA splicing (Lyabin et al., 2014). YB-1 is also a major component of messenger ribonucleoprotein complexes and is integrally involved in mRNA stabilization and the translational activation or repression of many genes (Suresh et al., 2018). This assortment of functions manifest themselves in an equally broad spectrum of biological roles in cancer (Lasham et al., 2013; Lyabin et al., 2014). The general cancer-related activities of YB-1 have been previously reviewed (Lasham et al., 2013; Kosnopfel et al., 2014; Lyabin et al., 2014) and therefore we will primarily focus on recent publications on the role of YB-1 specifically in lung cancer and MPM here. The evidence supporting each phenotype driven by YB-1 and the relevant interaction partners for the following sections is summarized in **Table 1**.

### A Promoter of Cell Proliferation and Cell Cycle Progression

The proliferative role of YB-1 in cancer has been demonstrated in many malignancies, driven by its regulation of highly penetrant downstream oncogenic growth promoting genes (Lasham et al., 2013). A prime example is the transcriptional activation of EGFR by YB-1. A study of 105 NSCLC samples showed that YB-1 and EGFR were significantly co-expressed and knockdown of YB-1 in two NSCLC cell lines resulted in reduction of EGFR (Hyogotani et al., 2012). Similar results have also been observed in basallike breast cancer and spinal chordoma (Stratford et al., 2007; Liang et al., 2019). Notably, overexpression of EGFR in lung cancer and mesothelioma promotes cell growth, invasion and angiogenesis (Ciardiello et al., 2004; Destro et al., 2006). Several cell cycle regulators are also under YB-1 control, including the E2F family. YB-1 specifically binds to the promoter of cell cycle activators transcription factor E2F1 and transcription factor E2F2 and YB-1 knockdown reduced cell proliferation of a NSCLC cell line in vitro and in vivo (Lasham et al., 2011). In NSCLC cells, YB-1 transcriptionally activates CCND1 a protein critical for progression through the G1 phase (Harada et al., 2014). YB-1 also binds to and activates the promoter of the dual specific phosphatase CDC25A, driving G1/S cell cycle progression (Zhao et al., 2016). These studies demonstrate the important role of YB-1 by showing that its knockdown with siRNA induces G0/G1 cell cycle arrest in vitro and in vivo (Harada et al., 2014; Zhao et al., 2016). Similarly, we have also shown that targeting YB-1 with siRNA can inhibit the growth of MPM cells in vitro (Johnson et al., 2018).

Y-box binding protein-1-driven proliferation may require a region within its N–terminal domain. Breast cancer cells overexpressing a YB-1 CTD fragment (from amino acid 125 onward) exhibited proliferation inhibition in vitro and in vivo (Shi et al., 2016). It is possible that the removal of Ser102, a site commonly phosphorylated and associated with growth (discussed further in section "Post-Translational Modification in the Control of YB-1 Activity and Localization"), could explain the lack of growth promotion here. However, as growth was actively inhibited in response to the upregulation of the YB-1 CTD, this could also suggest that YB-1, or certain regions of it, may inhibit proliferation under specific gene dosages or biological contexts. For example, YB-1 overexpression in Ras-MAPK activated breast cancer cells led to YB-1-mediated translational repression of growth-promoting genes, lowering proliferation rates. This was accompanied by the induction of EMT-like changes which promoted migration, invasion and allowed cells to survive in anchorage-independent conditions (Evdokimova et al., 2009b). This suggests that YB-1 expression levels determine its function, driving either a proliferative or invasive phenotype.

## YB-1 Is a Central Player in EMT, Invasion and Metastasis

Invasion and metastasis are key behaviors of lung cancer and mesothelioma cells that contribute to patient death and the poor prognosis observed with these tumors. YB-1 is known to play a role in the migration of thoracic cancer cells. Stable overexpression of YB-1 in lung ADC cells induced E-cadherin downregulation, N-cadherin upregulation, accelerated TGFβ1 induced EMT and cell migration (Ha et al., 2015). In support, silencing YB-1 inhibited the invasion and metastasis of lung cancer cells in vitro and in vivo (Guo et al., 2017). YB-1 overexpression also significantly increased the invasive capacity of these cells in vitro (Guo et al., 2017). Similarly, knockdown of YB-1 inhibited lung cancer migration (Zhao et al., 2016) and MPM migration and invasion (Johnson et al., 2018) in vitro. YB-1 has also been implicated in the migration and invasion of breast cancer (Lim et al., 2017), melanoma (Jia et al., 2017), nasopharyngeal cancer (Zhou et al., 2017b), skin squamous

TABLE 1 | Roles and interaction partners of YB-1 related to thoracic cancer biology.


cell carcinoma (Wang W. et al., 2004) and spinal chordoma (Liang et al., 2019).

Epithelial-mesenchymal transition is thought to be a primary mechanism facilitating cancer cell invasion and metastasis through inducing phenotypic plasticity (Brabletz, 2012). Current evidence suggests that EMT is a progressive, transient and reversible process and that cells in a hybrid E/M state - partial EMT – exhibit significantly higher tumorigenic potential compared to exclusively epithelial or mesenchymal cells (Pastushenko et al., 2018; Kröger et al., 2019).

Hybrid epithelial/mesenchymal state hybrids can be promoted by Zinc finger protein SNAI1 (Snail, gene SNAI1) transcription factor activity, the expression of which is specific to E/M populations of basal breast cancer cells (Kröger et al., 2019). Snail protein was found to be 5-fold higher in such cells compared to mesenchymal populations, while epithelial cells displayed undetectable levels (Kröger et al., 2019). However, this was only accompanied by a 1.5-fold increase in SNAI1 transcript expression, implying that translational activation is more important in Snail overexpression than transcriptional regulation (Kröger et al., 2019). YB-1 translationally upregulates Snail expression (Evdokimova et al., 2009a,b), suggesting that YB-1 could also be a key promoter the E/M state. In support, stable YB-1 overexpressing epithelial Madin-Darby canine kidney (MDCKYB−<sup>1</sup> ) cells exhibited a partial EMT-like phenotype and establish viable tumor xenografts in mice, while parental MDCK

cells did not (Gopal et al., 2015). This increased tumorigenicity was also accompanied by elevated secretion of angiogenic factors (Gopal et al., 2015). Treatment of endothelial cells with concentrated conditioned medium from MDCKYB−<sup>1</sup> cells also stimulated cell migration (Gopal et al., 2015).

Wnt signaling is also a primary driver of partial and complete EMT. β-catenin-dependent canonical Wnt signaling is thought to be preferentially active in E/M populations (Reya and Clevers, 2005; Kröger et al., 2019), while β-catenin-independent noncanonical signaling is more associated with a mesenchymal state, migration and invasion (Weeraratna et al., 2002; Gujral et al., 2014). Knockdown of YB-1 in hepatocellular carcinoma cells disrupted stemness and suppressed β-catenin protein expression and nuclear translocation, which was rescued by overexpression of the active form of β-catenin (Chao et al., 2017). This regulation of β-catenin-dependent Wnt signaling further supports a potential role for YB-1 in driving a partial EMT state. Interestingly, populations in the partial EMT state are also enriched with CSCs (Kröger et al., 2019), suggesting that YB-1 may also play are role in regulating these important cancer progenitors.

#### Involvement in Cancer Stem-Like Cells

Cancer stem-like cells are becoming recognized as important drivers of disease progression and are thought to be a major contributing factor toward metastasis, the development of drug resistance and recurrence of most cancers, including those of the thorax (Leon et al., 2016; MacDonagh et al., 2016; Makena et al., 2018). CSCs are a heterogeneous, slow growing population of cells within a tumor. They have selfrenewal ability but one subpopulation, termed metastatic CSCs, can disseminate through blood vessels and initiate metastasis (Dalerba and Clarke, 2007). This was clearly demonstrated in pancreatic cancer, where eradicating the metastatic CSC population dramatically reduced metastatic but not tumorigenic potential, implying that a subgroup of CSCs are responsible for metastasis (Hermann et al., 2007).

One recent study has shown that YB-1 enforces lung cancer metastatic CSC-like properties in vitro and in vivo through transcriptional upregulation of NANOG, a marker of CSCs required for the invasion and sphere formation of ADC cells in vitro (Guo et al., 2017). Supporting this, knockdown of YB-1 in hepatocellular carcinoma cells reduced NANOG and Oct4, as well as α-fetoprotein transcript expression (Chao et al., 2017). This follows findings showing NANOG and Oct4 are upregulated in ADC, which induce sphere formation, drug resistance and EMT (Chiou et al., 2010). YB-1 also regulates SOX2 in breast CSCs, maintaining stemlike properties and tumorigenic potential (Jung et al., 2014). Given the probable importance and frequent upregulation of SOX2 in lung cancer (Rudin et al., 2012; Karachaliou et al., 2013; Friedlaender et al., 2019; **Figure 1B**), a study investigating the relationship between YB-1 and SOX2 in thoracic cancer may further implicate YB-1 in the biology of these diseases.

Y-box binding protein-1 has been shown to be important in other cancer CSCs as well. Brain CSCs were shown to have high expression of YB-1 which was utilized in a YB-1-based virotherapy in vitro (Mantwill et al., 2013). The reexpression of the microRNA miR-382 in osteosarcoma cells significantly decreased the CSC population resulting in reduced relapse after doxorubicin treatment, EMT and metastasis both in vitro and in vivo (Xu et al., 2015). The authors attributed these tumor suppressive functions of miR-382 to targeting and downregulating YB-1 (Xu et al., 2015). This microRNA is downregulated in NSCLC and exogenous miR-382 expression inhibits NSCLC growth, migration and invasion via the suppression of SETD2 (Chen T. et al., 2017) and LMO3 (Chen et al., 2019). In breast cancer, inhibition of p90 RSK, a major kinase involved in YB-1 phosphorylation; see section "Post-Translational Modification in the Control of YB-1 Activity and Localization") using the small molecule LJI308 eradicated the population of breast CSCs and induced apoptosis in breast cancer cells (Davies et al., 2015). RSK is thought to have potential as a therapeutic target as it is involved in the proliferation of lung cancer (Poomakkoth et al., 2016). Furthermore, knockdown of WAVE3, a protein required for nuclear translocation of YB-1, prevented YB-1 mediated transcriptional activation NANOG, SOX2 and Oct4 in breast CSCs (Bledzka et al., 2017). WAVE3 expression was also correlated with that of YB-1 and more aggressive phenotypes of breast cancer (Bledzka et al., 2017).

#### YB-1 Is Involved in Hypoxic Response

The maintenance of CSCs is intertwined with the effects of hypoxia (Li and Rich, 2010). Supporting its role in thoracic CSC biology, hypoxia promotes an aggressive phenotype in MPM and upregulates Oct4, a marker of CSCs (Kim et al., 2018). Oct4 is also important in gefitinib-resistant lung CSCs and cisplatin-induced stemness in NSCLC has been linked to hypoxia-inducible factors (Kobayashi et al., 2016). Hypoxia occurs in most solid tumors and has been linked to CSC maintenance and behavior (Li and Rich, 2010; Bao et al., 2012), as well as disorganized tumor vascularization, EMT and metastasis (Muz et al., 2015). Factors such as HIF1α drive hypoxia-mediated transcription, influencing cell immortalization, metastasis and vascularization (Semenza, 2014). YB-1 translationally regulates HIF1α (El-Naggar et al., 2015) and acts as a transcriptional repressor for the HIF1α inhibitor FOXO3a via competition for p300 during vascular development (Emerling et al., 2008; Chou et al., 2015). Under hypoxic conditions YB-1 translocates to the nucleus where it binds to hypoxia response elements within the 3<sup>0</sup> enhancer of the EPO gene and blocks its expression (Rauen et al., 2016). Hypoxia plays an important role in driving malignant cellular behavior, including resistance to chemotherapy (Rohwer and Cramer, 2011). While YB-1-driven response to hypoxia may contribute toward chemoresistance, its activity as a transcription factor may also play a role in drug inefficacy.

### A Possible Role for YB-1 in Resistance to Platinum-Based Chemotherapy

Although the role of YB-1 has not yet been studied in lung cancer or MPM, it has been shown to be involved in the

chemoresistance of many cancers including that of platinumbased chemotherapies (To et al., 2010; Kang et al., 2013; Lasham et al., 2013; Shiota et al., 2014; Yamashita et al., 2017). Silencing YB-1 induces cisplatin sensitization in neuroblastoma (Wang H. et al., 2017), esophageal SCC (Xu and Hu, 2016), bladder cancer (Shiota et al., 2011) and melanoma (Schittek et al., 2007). Treatment with cisplatin also stimulates YB-1 production in bladder cancer (Shiota et al., 2010), while ovarian cancer cells with acquired cisplatin resistance show an increase in nuclear YB-1 expression (Yahata et al., 2002), suggesting that cancer cells may increase YB-1 production as a protective measure. The reasons why YB-1 may provide protection are still unclear. However, YB-1 does transcriptionally upregulate LRP, aka MVP (Stein et al., 2005), the principal component of vaults in human cells. Vaults are highly conserved ribonucleoproteins which have been suggested to play a role in the resistance of cancer cells to cisplatin, among other chemotherapies, by sequestering drugs away from their intended targets (Wang W. et al., 2004; Lara et al., 2011). YB-1 knockdown in lung cancer cell lines resulted in LRP downregulation and nuclear staining of YB-1 correlated with LRP expression in 105 NSCLC samples, conferring significantly lower overall survival (Hyogotani et al., 2012). However, this study did not investigate the effect of this knockdown on the chemoresistance of any drug.

Y-box binding protein-1 has also been linked to MRP1 gene activation (Stein et al., 2001; Mantwill et al., 2006), an efflux ATP-binding cassette transporter which is thought to contribute toward multidrug resistance (Stefan and Wiese, 2019). High levels of LRP and MRP1 correlated with lower response to cisplatin chemotherapy, poorer progression free survival and overall survival in advanced NSCLC patients receiving cisplatin-based chemotherapy (Li J. et al., 2009; Li X.Q. et al., 2009). Treatment with cisplatin also induces heightened LRP expression in ADC and SCC cell lines (Xu et al., 2017a) and LRP gene expression was significantly increased compared to control pleura samples in a study of MPM patients (Singhal et al., 2003). MDR1 gene (encoding P-glycoprotein 1), which is dependent on the nuclease and base excision repair enzyme APE1 expression, has also been implicated in YB-1-driven cisplatin resistance (Ohga et al., 1998; Chattopadhyay et al., 2008). However, the evidence supporting a clear role for P-glycoprotein 1 as an integral player in the chemoresistance of lung cancer and mesothelioma remains contentious, implying that other targets may be more important (Soini et al., 2001; Wangari-Talbot and Hopper-Borge, 2013).

#### An Agent of DNA Repair in Response to Cisplatin and Oxidative Stress

Y-box binding protein-1 may drive chemoresistance through the upregulation of the above targets and through driving a hypoxic response. However, some of its other functions may also contribute, such as its role as part of the DNA repair machinery. Oxidative stress and resulting chronic inflammation has long been implicated as a primary driver of cigarette smoking-related diseases, including lung cancer (Park et al., 2009; Sears, 2019). Altered DNA repair pathways have been implicated in the carcinogenesis of lung cancer in response to cigarette smoke-related DNA damage, particularly the NER and BER pathways (Sears, 2019). There is also a body of evidence supporting the suggestion that COPD leads to the development of lung cancer, or at least that the two are correlated (Sears, 2019). Chronic inflammation caused by asbestos-related oxidative stress is a major driver of MPM carcinogenesis (Benedetti et al., 2015; Chew and Toyokuni, 2015), implying that aberrations in DNA repair machinery in response to oxidization play a role in the progression of many thoracic cancers.

Y-box binding protein-1 has been suggested to be part of the DNA repair machinery as it binds to enzymes involved in BER, mismatch repair and DNA double-stranded break repair, previously reviewed (Alemasova and Lavrik, 2017). YB-1 binds preferentially to cisplatin-damaged DNA complexed with PCNA, where it works to separate cisplatin-damaged DNA strands, recruit DNA repair proteins and displays weak endonucleolytic and exonucleolytic function (Ise et al., 1999; Gaudreault et al., 2004). PARP1 has also been shown to catalyze the poly(ADP-ribosyl)ation of YB-1 in the presence of DNA damage, further supporting a role for YB-1 in DNA repair (Alemasova et al., 2015).

Y-box binding protein-1 is also involved in NER and BER in response to oxidative stress. DNA damage-related stress stimulates YB-1 nuclear translocation (Cohen et al., 2010) (discussed further in section "Control of YB-1 Subcellular Localization") where it can bind to oxidized DNA lesions, structurally altering DNA to allow access to the damaged site while recruiting and scaffolding proteins involved in BER including PARP1, PARP2, NEIL1, and PCNA, among others (Dutta et al., 2015; Alemasova et al., 2016). In ssDNA, YB-1 suppresses NEIL1-mediated apurinic/apyrimidinic site cleavage, and it has been suggested that the role of YB-1 in DNA repair can prevent ssDNA breaks and induce oxidative nucleotide repair in double-stranded DNA (Dutta et al., 2015). YB-1 has also been linked to NER. Cross-talk between YB-1 and XPC (an important player in NER which has significance in lung cancer carcinogenesis and is affected by germline mutation in MPM), results in their assembly at DNA damage sites (Jin et al., 2014; Fomina et al., 2015; Betti et al., 2017; Sears, 2019).

Y-box binding protein-1 was found to be involved in mitigating cigarette smoke-induced guanine oxidization in lung fibroblasts and mice chronically exposed to cigarette smoke, and that lung samples of late-stage COPD patients exhibited significantly lower YB-1 levels compared to early mid stage patients or patients without COPD (Deslee et al., 2010). The role YB-1 plays in DNA repair (particularly from oxidization) and the fact that it is secreted under oxidizing conditions (see section "YB-1 is Secreted Into the Extracellular Space Under Cellular Stress") implies that YB-1 may promote the oxidation-related carcinogenesis of lung cancer and MPM. Cigarette-induced oxidative stress has additionally been suggested to induce the release of exosomes (Ryu et al., 2018), the sorting of which are also mediated in part by YB-1.

### YB-1 and Exosomal RNA Sorting

fcell-07-00221 November 21, 2019 Time: 15:7 # 9

Extracellular vesicles such as exosomes are used by cells for intercellular communication to both their immediate and distant surroundings (Mashouri et al., 2019). Exosomes carry factors such as proteins, mRNA and miRNA to mediate processes including embryonic development, injury response and homeostasis (Mashouri et al., 2019). Exosomes also play versatile and key roles in cancer cell behavior and remodeling of the tumor microenvironment (Mashouri et al., 2019). A malignant role for exosomes in lung cancer is well documented, where exosomes can induce proliferation, angiogenesis, EMT changes and metastasis (Vanni et al., 2017; Zhou et al., 2017a; Ryu et al., 2018). Exposure to cigarette smoke is also thought to induce the release of extracellular vesicles, such as exosomes, which has been linked to the development of COPD and possibly the development of lung cancers (Ryu et al., 2018). Asbestos exposure also alters the exosomal cargo of lung epithelial cells in vitro and exposing non-malignant mesothelial cells to these exosomes induces gene expression changes related to EMT and other cancer related pathways (Munson et al., 2018). This indicates that exosomes may play an integral role in the carcinogenesis of mesothelioma. MPM cell lines also secrete higher levels of exosome-associated proteins linked to stress response and proliferation compared to their nonmalignant counterparts (Creaney et al., 2017). Supporting this, exosomes from MPM cells have a distinct oncogenic signature and stimulate the migration of fibroblasts and endothelial cells (Greening et al., 2016).

Y-box binding protein-1 is known to be involved in exosomal RNA-sorting, reviewed previously (Suresh et al., 2018), which may indicate it is involved in altering malignant exosomal expression profiles. Briefly, the presence of YB-1 in exosomes has been shown in both malignant and non-malignant cells alike where it helps to define the levels of several RNA species, including miRNA and tRNA (Shurtleff et al., 2017; Suresh et al., 2018). However, to our knowledge no study has investigated YB-1 in lung cancer and mesothelioma exosomal sorting. Future studies following this line may shed further light into the underlying mechanisms of exosomes and their role in thoracic cancer biology.

## A Role in Immune Evasion?

Evidence in other tumor types suggests that the upregulation of YB-1 could drive immune evasion. For example, in doxorubicin-resistant hepatocellular carcinoma cells, YB-1 is overexpressed, which in turn transcriptionally upregulates the expression of PD-L1 and decreases the secretion of the chemokines IL1β, IL10, and TGFβ in vitro (Tao et al., 2019). High YB-1 was also associated with resistance to cisplatin, gemcitabine, docetaxel, dasatinib and gefitinib in this study (Tao et al., 2019). This suggests that resistance to these drugs may also result in heightened PD-L1 and subsequent immunosuppression via YB-1 upregulation, at least in hepatocellular carcinoma. In light of these results, investigating the potential of a similar role in thoracic cancers would be of great interest.

## YB-1 Regulation: A Complex Network of Transcriptional, Translational and Post-translational Control

The wide-ranging roles of YB-1 in cell biology imply that its expression, localization and function must be tightly regulated in normal physiology. As YB-1 is frequently overexpressed in cancer, dysregulation of these controlling systems may play a role in malignant transformation. The expression and localization of YB-1 is controlled by a complex network of transcriptional, translational and autoregulatory signals discussed below.

### Transcriptional Control

Several transcription factors have been found to promote YB-1 transcription by binding to motifs in the YBX1 promoter. For example, YBX1 transcription has been shown to be promoted by GATA transcription factors, although recent evidence suggests the GATA family is less important for promoting YBX1 expression in ADC (Yokoyama et al., 2003; Murugesan et al., 2018). Possibly more important are the six E-boxes located in the promoter of YBX1 (Makino et al., 1996). The first is located at 48–53 nucleotide residues away from the promoter, the second at 353–358, the third at 458– 463, the fourth at 531–536, the fifth at 1147–1152, and the sixth at 1201–1206 (Makino et al., 1996). The E-box binding transcription factor Twist1 also stimulates YBX1 transcription, driving cell growth and EMT (Shiota et al., 2008; He et al., 2015; **Figure 2**). A recent meta-analysis of 572 NSCLC patients showed that high Twist1 expression significantly correlated with poorer patient prognosis, recurrence-free survival and lymph node or other metastasis (Li et al., 2018). A small retrospective study of mesothelioma samples also showed that Twist1 expression was significantly higher in sarcomatoid tumors (expressed in 7/7 of samples) compared to biphasic (6/10) and epithelioid tumors (7/17) (Iwanami et al., 2014). Although the percentage of samples positive for YB-1 was almost identical to that of Twist1 in this study (6/7 in sarcomatoid, 6/10 in biphasic and 7/17 in epithelioid), whether YB-1 and Twist1 were co-expressed in the same samples was not determined (Iwanami et al., 2014).

An E-box within the YB-1 promoter is also trans-activated by Myc and p73 to drive the transcription of YBX1 (Uramoto et al., 2002; **Figure 2**). The ability of Myc to transcriptionally activate YBX1 is interesting, not only as Myc drives malignant behavior and is often associated with poor prognosis in thoracic cancers (Jiang et al., 1992; Volm and Koomagi, 2000; Riquelme et al., 2014), but because YB-1 can itself initiate Myc translation by acting as a specific internal ribosome entry segment-trans-activating factor (Cobbold et al., 2010). YB-1 was also shown to regulate Myc at the transcriptional level in bladder cancer, with implications on aerobic glycolysis (Warburg effect) (Xu et al., 2017b). This feed forward loop was first described in multiple myeloma (Bommert et al., 2013), however, it is quite possible that a similar feed forward loop accounts for both YB-1 and Myc overexpression in thoracic cancers, driving malignant progression and aggressiveness.

## Translational Regulation of YB-1

Y-box binding protein-1 expression is also regulated at the translational level, most notably via signaling through the mTOR pathway (**Figure 2**), which regulates cell growth, motility, survival, transcription and protein synthesis via the integration of signals from hormone and growth factor stimulation, availability of nutrients, and stress (Zarogoulidis et al., 2014). mTOR signaling promotes the translation of YBX1 and increases the phosphorylation of RSK, a serine/threonine kinase which phosphorylates and thereby activates YB-1 (Mendoza et al., 2011; Lyabin et al., 2012). RSK has been implicated in lung cancer proliferation and has itself been suggested as a target with therapeutic significance (Poomakkoth et al., 2016).

The division rate of eukaryotic cells affects YBX1 translation via mTOR regulation. Slow dividing and serum-starved cell populations exhibit attenuated mTOR signaling, which in turn inhibits YBX1 translation (Lyabin et al., 2012). This pathway is frequently activated in lung cancer and antagonizing mTOR in such cells has proven to be a potential therapeutic avenue (Zarogoulidis et al., 2014). The PI3K/mTOR pathway is also highly activated in mesothelioma, but not in non-malignant mesothelial cells (Zhou et al., 2014) or adjacent tissue (Hoda et al., 2011), and phospho-mTOR was significantly associated with poorer overall survival in a cohort of 107 mesothelioma patients (Bitanihirwe et al., 2014). Dactolisib (BEZ235) treatment inhibited mesothelioma cell growth by targeting mTOR (Zhou et al., 2014) and similarly, treatment with the mTOR inhibitor temsirolimus stopped MPM cell proliferation and was synergistic with cisplatin treatment in vitro and in vivo (Hoda et al., 2011). It stands to reason that YB-1 overexpression is likely to be, at least in part, linked to the prominent role mTOR signaling plays in thoracic cancers.

## Autoregulation of YB-1 – An Unsolved Piece of the Puzzle

Y-box binding protein-1 is controlled by an autoregulatory feedback loop in which YB-1 binds its own mRNA at two 8 nucleotide motifs at (nt) 1133–1145 and nt 1165–1172, inhibiting translation prior to 40S ribosomal subunit binding (Skabkina et al., 2005; **Figure 2**). PABP competes with YB-1 at one of these overlapping sites (nt 1149–1196), and stimulates YBX1 translation (Skabkina et al., 2003, 2005). Considering YB-1 overexpression is frequently observed in cancer, this feedback system may be dysregulated before or during malignant progression. It is possible that PABP upregulation could cause a bias for PABP translational activation of YBX1, although PABP itself is controlled by a similar autoregulatory loop (Ma et al., 2006). Nonetheless, recent expression and interactome analysis

FIGURE 3 | Post-translational modification of YB-1. The presence of various sites before or after proteasomal cleavage of YB-1 modulates its function and localization, which has implications on antibody use. YB-1 is comprised of a CSD shown in dark gray, an N-terminal domain in white and a CTD, also in white. Within the CTD there are three nuclear localization signals (NLS-1 from amino acid (aa) 149–156, NLS-2 from aa 185 to 194 and NLS-3 from aa 276 to 292), shown in light gray, and one cytoplasmic retention signal (CRS from aa 247 to 267), shown in black. YB-1 is proteolytically cleaved at Glu216 and Glu219 (shown in red and highlighted with a scissors icon), which is thought to stimulate YB-1 translocation. Three commonly used antibodies targeting YB-1 are also shown, two of which have been validated using mass spectrometry (in yellow) and one which is known to cross react with hnRNP1A (in red). If the proteolytic theory of YB-1 translocation is correct, this would also have implications on the use of antibodies. Various post-translational modifications also effect the downstream function and nuclear localization of YB-1. Green dots indicate acetylation, yellow glycosylation, blue phosphorylation, red sumoylation and orange ubiquitination. Solid black arrows indicate a post-translational modification that is produced by a known upstream regulator, or a known function of YB-1. Dotted black arrows indicate a post-translational modification or function that is yet to be fully proven. Blue and dotted blue arrows indicate the movement or supposed movement of YB-1 throughout cellular compartments, respectively. Created with BioRender.com.

of YB-1 showed that PABP significantly correlated with YB-1 expression in ADC, implying it plays a central role in YB-1 upregulation and highlighting the need for further investigation into the PABP/YB-1 loop (Murugesan et al., 2018).

## Non-coding RNA Modulate YB-1 Expression

Various families of ncRNAs also play a role in regulating YB-1 levels (**Figure 2**). One such family are miRNAs – a conserved class of short, ncRNAs that regulate gene expression by binding to and initiating RNA-induced silencing complex-mediated degradation of target mRNA (Jonas and Izaurralde, 2015). The inhibition of YBX1 by several miRNAs has been shown in other cancers, although to our knowledge such interactions have not been investigated in thoracic cancers. We recently demonstrated regulation of YBX1 by miR-137 in MPM cells, inhibiting growth, migration and invasion (Johnson et al., 2018). This miRNA is also known to act as a tumor suppressor in lung cancer by targeting SRC3 (Chen R. et al., 2017) and BMP7 (Yang et al., 2015). Another miRNA known to target YBX1 is miR-216a, which suppresses YB-1-mediated metastasis in pancreatic cancer (Lu et al., 2017). MiR-216a acts as an tumor suppressor in SCLC by targeting and downregulating the anti-apoptotic protein B-cell lymphoma 2 (Bcl-2) (Wang et al., 2018), although it is likely that these effects are also, in part, due to YBX1 downregulation.

The lncRNA DANCR has been implicated in reducing the levels of this miRNA through its complimentary miR-216a binding site, sequestering it away from miR-216a targets (Zhen et al., 2018). DANCR is associated with advanced tumor grade and poor prognosis in lung cancer and promotes ADC cell growth in vitro and in vivo (Zhen et al., 2018). Dysregulation of DANCR and subsequent lowering of miR-216a could represent one mechanism of YB-1 overexpression in thoracic cancer, representing an area which requires further investigation.

In addition to DANCR, other lncRNAs as well as transfer RNA-derived fragments can also play a role in regulating YB-1 expression, reviewed previously (Suresh et al., 2018). One example is the lncRNA GAS5, which interacts with YB-1 protein and activates YBX1 translation, upregulating p21 and initiating G1 cell cycle arrest in stomach cancer (Liu et al., 2015). Interestingly, GAS5 knockdown did not affect YBX1 mRNA expression, something the authors attribute to possible interactions with other proteins (Liu et al., 2015). GAS5 is a known competing endogenous RNA for miR-137, which targets YBX1 in thoracic cancers (see above in this section) (Chen et al., 2018), so it is possible that this may contribute toward YBX1 translational upregulation. However, GAS5 knockdown does not affect YBX1 mRNA expression (Liu et al., 2015), as would be expected by an increase in miR-137 availability, so this does not fully explain this relationship. Further inquiry into the GAS5/YB-1 and possibly miR-137 relationship is required. GAS5 acts as a tumor suppressor and is lost in lung cancer and mesothelioma (Renganathan et al., 2014; Shi et al., 2015), which is consistent with findings in other cancer types (Gutschner et al., 2018). The apparent discrepancy between the tumor suppressive function of GAS5 and GAS5-mediated translational

upregulation of the oncogene YBX1 remains unanswered and also warrants further study.

More recently, the lncRNA MIR22HG was shown to prevent proteasomal degradation of YB-1 in lung cancer cells, which might contribute to YB-1 overexpression (Su et al., 2018). LINC00312 also interacts with YB-1 driving invasion, migration and vascular mimicry of ADC cells, and LINC00312 is associated with metastasis in ADC patients (Peng et al., 2018). HULC is another lncRNA that binds to YB-1 in hepatocellular carcinoma cells, promoting Ser102 phosphorylation, the significance of which is further described in section "Post-Translational Modification in the Control of YB-1 Activity and Localization" (Li et al., 2017). HULC is overexpressed in NSCLC and can promote proliferation via SPHK1 upregulation, which is upstream of the PI3K/AKT pathway (Liu L. et al., 2018). This implies that HULC may also be involved in PI3Kmediated YB-1 activation. TP53TG1, yet another lncRNA, can also bind to YB-1 and inhibit its nuclear translocation, stopping it from transcriptionally activating its oncogenic targets (Diaz-Lagares et al., 2016). TP53TG1 is downregulated in NSCLC and its upregulation sensitized cisplatin resistant NSCLC cells to cisplatin (Xiao et al., 2018). This was attributed to the downregulation of miR-18 (Xiao et al., 2018), however, considering the likely role of YB-1 transcriptional regulation in cisplatin resistance (see section "A Possible Role for YB-1 in Resistance to Platinum-Based Chemotherapy"), it is possible that cytoplasmic retention of YB-1 also played a part in the cisplatin sensitivity seen here. Finally, CAR10 binds to and stabilizes YB-1, leading to the upregulation of EGFR in lung cancer and promoting proliferation (Wei et al., 2016). ncRNA therefore play an integral role in the expression and activity of YB-1, and dysregulation of these families is likely to contribute to YB-1 overexpression in cancer.

### Post-translational Modification in the Control of YB-1 Activity and Localization

The activity of YB-1 is modulated through various posttranslational modifications (**Figure 3**), with phosphorylation being the best studied. Ser102 (located in the CSD of YB-1) is currently the most comprehensively studied phosphorylation site. This site is a target of AKT and RSK, making it downstream of both the MAPK/ERK and PI3K/AKT pathways (Sutherland et al., 2005; Stratford et al., 2008; Mendoza et al., 2011). Several additional phosphorylation sites on YB-1 have been identified including Tyr281, which is located within a NLS toward the C-terminal of YB-1 and correlates with the nuclear localization of either a YB-1 C-terminal fragment or full length YB-1 (van Roeyen et al., 2013) (refer to next section for more detail). Tyr162 on YB-1 is also reportedly phosphorylated by FGFR1 (Kasyapa et al., 2009), an important oncogenic driver in lung cancer (Jamal-Hanjani et al., 2017; Friedlaender et al., 2019) and mesothelioma (Schelch et al., 2014; Quispel-Janssen et al., 2018), however, to our knowledge the significance of this modification has not yet been established. Ser165 and Ser176 on YB-1 are also phosphorylated, each promoting distinct groups of nuclear factor-κB target gene expression. This pathway is commonly

dysregulated in thoracic cancers and drives cell survival, chemoand radiotherapy resistance (Chen et al., 2011; Nishikawa et al., 2014; Prabhu et al., 2015; Martin et al., 2017).

In the case of Ser102, it seems that phosphorylation is linked to the hexosamine biosynthetic pathway, in which OGT and O-linked N-acetylglucosamine add or remove N-acetylglucosamine groups to serine or threonine residues, respectively. OGT-mediated O-linked glycosylation of YB-1 at Thr216 aids in the phosphorylation of Ser102 and subsequent transcriptional activity of YB-1 in hepatocellular carcinoma (Liu et al., 2016; **Figure 3**).

Sumoylation, acetylation and ubiquitination are also prominent post-translational modifications that can contribute toward regulating YB-1 activity and localization. In addition, the nuclear localization of YB-1 has been linked to three NLS, mapped to amino acid residues 149–156 (NLS-1), residues 185–194 (NLS-2) and residues 276–292 (NLS-3) (van Roeyen et al., 2013).

Y-box binding protein-1 is sumoylated at three distinct sites in response to circadian rhythm in zebra fish cells, which has implications on its nuclear shuttling (Pagano et al., 2017). One of these sites is a canonical inverted sumoylation site (at amino acids 287–290 within NLS-3), while the other two are non-canonical sites (at 60–63 which is within the CSD and at 151–154, within NLS-1; **Figure 3**) (Pagano et al., 2017). Circadian disruption has been correlated with an increased risk of cancer development (Hansen, 2017; Liu W. et al., 2018) and many processes integral to tumorigenesis follow circadian rhythms (cell cycle regulation and DNA repair, for example). Although one study failed to find a link between night shift work and lung cancer among a cohort of female textile workers in Shanghai, China (Kwon et al., 2015), preclinical data indicates that disturbance of the circadian clock can promote lung tumor growth in vivo (Papagiannakopoulos et al., 2016). Modulation of YB-1 localization in response to light may represent one contributing factor in the observed correlation between circadian rhythm and cancer and warrants further investigation.

Acetylation of YB-1 has been reported to occur in lung cancer cells, however, the significance of this remains unclear. YB-1 was one of 542 proteins acetylated by the histone deacetylase inhibitor SAHA in SILAC experiments in a NSCLC cell line (Wu et al., 2015). Here, YB-1 was acetylated at Lys64 (**Figure 3**). Lys301/304 of YB-1 can also be acetylated and the amount of acetylated YB-1 is significantly increased in monocytes of hemodialysis patients (Ewert et al., 2018).

Ubiquitination may also play an important role in YB-1 expression and subcellular localization. RBBP6 initiates proteasomal degradation of YB-1 by binding to and ubiquitinating it within a 62-residue fragment of the YB-1 CTD (Chibi et al., 2008). The protein isoform of p63 1Np63α counteracts this by preventing proteolysis of full-length YB-1 and stimulating accumulation of poly-ubiquitinated YB-1 in the nucleus (di Martino et al., 2016), possibly supporting the role of proteolytic cleavage-dependent YB-1 nuclear shuttling (discussed further in section "Control of YB-1 Subcellular Localization"; **Figures 3**, **4**). Further supporting this theory, UV irradiated DNA damage stimulates YB-1 ubiquitination at Lys64 (Boeing et al., 2016) (the same lysine residue that is acetylated, above in this section; **Figure 3**). Considering the DNA repair function of YB-1 and the aforementioned ubiquitination-driven proteasomal cleavage of YB-1, this possibly induces a similar nuclear translocation of YB-1. This is further supported by results showing that YB-1 is shuttled to the nucleus upon UV irradiation (Koike et al., 1997).

Ubiquitination is also important in the secretion of YB-1 via the multi-vesicular body pathway. The E3 ligase activity of HACE1 polyubiquitinates YB-1 at K27, facilitating tumor susceptibility gene 101 binding, which initiates YB-1 secretion (Palicharla and Maddika, 2015). In summary, post-translational modification influences the levels, activity and localization of YB-1, which in turn impacts the downstream effects of YB-1.

#### Control of YB-1 Subcellular Localization

In non-malignant cells, YB-1 is primarily located in the cytoplasm and functions as a major component of free messenger ribonucleoprotein complexes, where it can inhibit or stimulate cap-dependent translation depending on the ratio of YB-1 to mRNA (Suresh et al., 2018). Under certain stresses such as cisplatin treatment (Yahata et al., 2002), hypoxia (Rauen et al., 2016), UV radiation (Koike et al., 1997), and hyperthermia (Stein et al., 2001), YB-1 translocates to the nucleus, however, the underlying mechanism of this remains unclear. As above, YB-1 has three NLS sites which have been mapped to amino acid residues 149–156, residues 185–194 and residues 276–292 (van Roeyen et al., 2013), which are recognized by transportin-1 (Mordovkina et al., 2016) and WAVE3 (Bledzka et al., 2017). In addition YB-1 also contains a CRS at residues 247–267 (Woolley et al., 2011; **Figure 3**). The locations of these sites are postulated to regulate YB-1 nuclear-cytoplasmic translocation.

One line of evidence suggests that nuclear translocation is preceded by a specific proteolytic cleavage by the 20S proteasome of YB-1 at Glu216 and Glu219 under cellular stress (Sorokin et al., 2005; Kim et al., 2013; **Figures 3**, **4**). This results in loss of a 105 amino acid sequence from the C-terminus, including the CRS, and accumulation of the remaining N–terminal fragment in the nucleus (Sorokin et al., 2005; Kim et al., 2013). The presence of an NLS in the CTD suggests that a C–terminal fragment may also be shuttled to the nucleus, presumably if the nearby CRS has been cleaved off (van Roeyen et al., 2013; **Figures 3**, **4**). Supporting this, breast cancer cells preincubated with the proteasome inhibitor MG-132 before treatment with doxorubicin displayed reduced nuclear and enhanced cytoplasmic levels of YB-1 (visualized with a C-terminal-targeting antibody; **Figure 3**), compared to cells treated with doxorubicin alone (van Roeyen et al., 2013). However, this does not rule out whether full-length YB-1 translocation occurs by some other mechanism.

Countering the proteasomal theory is one study that suggests the YB-1 N-terminal fragment has been misidentified as another protein, hnRNP1A, and that only full-length YB-1 is found in the nucleus (Cohen et al., 2010). Full-length YB-1 nuclear translocation could be facilitated by its phosphorylation. For example, there is evidence showing that YB-1 is phosphorylated at Ser102 by the serine/threonine kinase AKT before being shuttled to the nucleus (Sutherland et al., 2005; **Figures 3**, **5**).

This may cause a conformational change which could block the CRS of YB-1, stimulating its nuclear shuttling. However, a recent study found that while ionizing radiation, EGF stimulation and overexpression of the KRAS G12V mutant induced Ser102 phosphorylation of YB-1 in both the nucleus and the cytoplasm, there was no increase in YB-1 expression in nuclear fractions (Tiwari et al., 2018). The authors attribute this to nuclear translocation of RSK, phosphorylating pre-existing nuclear YB-1 – not the shuttling of YB-1 itself (**Figure 4**). It may be that the translocation of either YB-1, RSK or both is dependent on the type of cellular stress applied. As mentioned in section "Post-Translational Modification in the Control of YB-1 Activity and Localization," phosphorylation of Thr281 within the NLS 276– 292 of YB-1 also correlates with its nuclear translocation (van Roeyen et al., 2013), however, it is not yet clear whether this modification is actively involved in YB-1 shuttling.

The localization of YB-1 also appears to be dependent on its ability to bind RNA and other proteins in the cytoplasm as YB-1 nuclear localization is hampered by higher cytoplasmic mRNA levels (Tanaka et al., 2018). This group also found that p53 (along with 4 other nucleocytoplasmic-shuttling proteins) binds to a YB-1 NLS and co-accumulates with YB-1 in the nucleus in response to actinomycin D treatment (Tanaka et al., 2016). This implies that YB-1 nuclear localization is a p53-mediated response to DNA stress. Another factor, C1QBP inhibits nuclear localization by binding to and blocking an NLS region (Matsumoto et al., 2018). C1QBP binding also moderately attenuated YB-1-mediated mRNA stabilization (Matsumoto et al., 2018). It is likely that the balance of this cytoplasmic interactome determines where YB-1 is localized under different conditions and that a disruption of this balance may lead to malignant progression.

### YB-1 Is Secreted Into the Extracellular Space Under Cellular Stress

Stress-related secretion of factors found in the nucleus and cytoplasm have been found to be biologically relevant in thoracic cancer biology and may serve as potential non-invasive biomarkers. Secretion of the nuclear protein HMGB1 in response

to asbestos-related necrosis in MPM cells, which acts as an alarmin to stimulate inflammation is one example (Yang et al., 2010). Serum HMGB1 has shown prognostic relevance as a possible biomarker in MPM (Tabata et al., 2013).

Y-box binding protein-1 is related on an evolutionary level to HMGB1 and is also secreted under certain cellular stresses. This was first evident in monocytes stimulated with bacterial lipopolysaccharide through an active, non-classical pathway and appears to require the same two lysine residues (Lys301/304) that are the site of acetylation in hemodialysis patients (Frye et al., 2009; Ewert et al., 2018; **Figures 3**–**5**). Secreted YB-1 stimulated DNA synthesis, cell proliferation and migration of kidney cells (Frye et al., 2009). More pertinent to thoracic cancer, YB-1 is also secreted under oxidative stress. YB-1 translationally upregulates G3BP1 under oxidative stress and localizes to cytoplasmic stress granules where it is involved in pro-survival mRNA reprogramming (Somasekharan et al., 2015). G3BP1 also promotes the invasion and metastasis of sarcoma cells in vivo (Somasekharan et al., 2015). In support, YB-1 enrichment in stress granules is also linked to its secretion to the extracellular space under oxidizing conditions (Guarino et al., 2018; **Figures 4**, **5**). Secretion of YB-1 resulted in depletion of cytoplasmic YB-1, leaving nuclear expression intact (presumably to allow for YB-1 mediated DNA repair), while secreted YB-1 inhibited the growth of neighboring keratinocytes (Guarino et al., 2018).

Extracellular YB-1 acts as a ligand for Notch3, binding to epidermal growth factor-like repeats 17–24 on Notch3 and subsequently promoting YBX1 translation in a feed-forward, autoregulatory fashion (Rauen et al., 2009; Raffetseder et al., 2011; Gera and Dighe, 2018; **Figures 4**, **5**). Notch3 is frequently overexpressed in NSCLC where high levels correlate with poor prognosis, making it a candidate target for therapeutic intervention (Zong et al., 2016). Considering the prevalence of oxidative stress and Notch3 in thoracic cancers, the secretion of YB-1 may be biologically important, although more studies are required to fully understand this process.

#### YB-1 IN THORACIC CANCERS: CLINICAL RELEVANCE

#### YB-1 as a Prognostic Biomarker

There is evidence supporting the use of YB-1 as a prognostic biomarker in thoracic cancers, and subcellular localization is

important in this regard. Analysis of TCGA data demonstrate that high levels of YBX1 mRNA correlate significantly with poor prognosis in NSCLC and MPM patients (**Figures 1D,E**). YB-1 protein has been correlated with poor prognosis previously (Shibahara et al., 2001; Gessner et al., 2004), supported by a recent meta-analysis of six studies covering data on 692 NSCLC patients, where YB-1 was associated with worse overall survival, tumor stage and depth of invasion (Jiang et al., 2017). A study in MPM patients also supports the TCGA data (albeit tentatively due to the low number of patients in the cohort) (Iwanami et al., 2014). Here, YB-1 levels were shown to be higher in sarcomatoid MPM tumors, which confer the worst prognosis (Iwanami et al., 2014).

There has been some contention surrounding the use of particular YB-1 antibodies in prognostic studies across cancer types. One N-terminal targeting YB-1 antibody that binds to residues 23–52 has been used in prognostic studies in the past (**Figure 3**). However, this antibody has been shown to cross reacts with the ubiquitously expressed hnRNPA1 protein via mass spectrometry making it unsuitable for such application (Woolley et al., 2011). Antibodies targeting the extreme N-'terminus of YB-1 (residues 1–12) or residues 299–313 in the CTD (Cterminal) have been shown to be specific for YB-1, again by mass spectrometry (Woolley et al., 2011; **Figure 3**). However, the N–terminal antibody has been suggested as more suitable for prognostic applications as this region does not interact with other proteins, so this epitope may be more accessible (Woolley et al., 2011). Notably, all prognostic studies cited in this review utilize the C-terminal targeting antibody. Regardless, a universal standardization of one reliable antibody would significantly enhance the prognostic potential of YB-1 for diagnosis using traditional pathological tissue staining.

Secreted YB-1 may also have prognostic significance in cancer. One study of 44 breast cancer patients with bone metastases found that serum YB-1 was present in 50% of patients and associated with extra-bone metastases and faster bone disease progression (Ferreira et al., 2017). There was a trend toward poorer overall survival in high-YB-1 patients, although a bigger cohort is needed to provide a more definitive answer (Ferreira et al., 2017). Another group found an YB-1/p18 in the plasma of patients with various diseases (including 32/38 lung cancers) but not in healthy controls via Western Blot using a monoclonal YB-1 antibody (Tacke et al., 2014). This study found no prognostic significance of YB-1/p18 in any of the cancers tested, but they assert that YB-1/p18 may have diagnostic significance (Tacke et al., 2014). The small sample number in this study should be noted before the prognostic applicability of secreted YB-1 is ruled out. Investigating the prognostic significance of secreted fulllength or other fragments of YB-1, not just YB-1/p18, may also be of interest. The potential of YB-1 as a circulating biomarker is intriguing as a non-invasive method of prognosis and diagnosis, although more studies with larger cohorts are required.

#### Targeting YB-1: An Achievable Feat?

In the past YB-1 has been overlooked as a therapeutic target because of its role as a transcription and translation factor, which have been traditionally hard to target with small molecule inhibitors. However, recent advancements in the delivery of RNA-based drugs has opened up new potential avenues of targeting oncoproteins such as YB-1 (Seton-Rogers, 2012; Afonin et al., 2014). We and others have shown that miRNA or siRNA can be used to target YBX1 in thoracic cancer cells in preclinical studies (Xu et al., 2015; Johnson et al., 2018). The delivery of miRNA mimics in the clinic is now thought to be a viable anti-cancer strategy. For example, MRX34 (a liposomal miR-34a mimic) showed evidence of efficacy and safety in a phase I trial in patients with various solid tumors including 1 NSCLC patient (Beg et al., 2017). More pertinently, a phase 1 clinical trial delivering miR-16-based mimics using bacterial minicells (EnGeneIC Dream Vectors) in mesothelioma and advanced NSCLC patients demonstrated the safety and efficacy of miRNA-based therapy (van Zandwijk et al., 2017), evidencing the potential for miRNA replacement therapy in patients with thoracic cancer.

There are a number of systems which pose as attractive options to deliver RNA-based drug payloads in thoracic cancer such as lipid, RNA, inorganic and polymer-based nanoparticles, all with their respective advantages and drawbacks (Shu et al., 2014). The delivery of siRNA or miRNA using nanoparticles in lung cancers, and to a lesser extent MPM, has been achieved in vitro and in vivo, evidencing the potential of these delivery systems (Lee et al., 2016). The in vivo transport of siRNA to large cell lung carcinoma tumors using lipoprotein nanoparticles (Tagalakis et al., 2018) and ADC tumors using polyethylene glycol nanoparticles (Wen et al., 2017) has demonstrated the applicability of nanoparticle delivery systems for targeted therapy. However, these studies treated subcutaneously grown tumors, which do not reflect the orthotopic context of thoracic cancer and the problems with delivery that come with it. Recently though, an siRNA targeting anti-EZH2 was successfully delivered to orthotopically grown NSCLC tumors in mice using modified polyethyleneimine nanoparticles (Yuan et al., 2017), and delivery and retention of amiloride-sensitive epithelial sodium channelspecific siRNA into the lungs of normal mice was achieved (Tagalakis et al., 2018). The successful delivery of miR-215-5p miRNA mimics complexed in an atelocollagen vehicle was also recently achieved in an orthotopic MPM mouse model, which significantly suppressed tumor growth (Singh et al., 2019). The advances in RNA-based drug delivery in preclinical and clinical studies mean that siRNA or miRNA delivery is an appealing YB-1 targeting strategy in thoracic cancers. However, other potential strategies may also be of interest, although these are yet to be investigated in humans.

Inhibiting YB-1 activation may be one such viable targeting strategy. Fisetin (3,7,3<sup>0</sup> ,40 -tetrahydroxy flavone) is a flavanol that binds to the CSD of YB-1, inhibiting its phosphorylation at Ser102 and blocking EMT in prostate cancer cells in vitro (Khan et al., 2014). Targeting YB-1 using fisetin also attenuated the growth of melanoma cells in vitro and in vivo (Sechi et al., 2018). Fisetin was also found to inhibit mTOR and PI3K/Akt signaling in NSCLC cells, both of which are important in both thoracic cancer biology and YB-1 regulation (see section "Translational Regulation of YB-1") (Khan et al., 2012). Another possible method for targeting YB-1 was demonstrated by using an interference cell permeable peptide that prevented YB-1 Ser102

phosphorylation. This led to an inhibition of EGFR expression and reduced growth of prostate and breast cancer cells, but not of non-malignant mammary epithelial cells (Law et al., 2010). Upstream inhibitors such as those targeting mTOR may also be an option (Hoda et al., 2011; Zhou et al., 2014), but would not be specific. A recent study showed that 2,4-dihydroxy-5-pyrimidinyl imidothiocarbomate antagonizes YB-1, inhibits YB-1 nuclear translocation and increases doxorubicin accumulation in breast cancer cells (Gunasekaran et al., 2018).

The use of oncolytic viruses that require YB-1 for replication is another potential therapeutic approach. XVir-N-31-mediated lysis of brain CSCs and virus production was significantly reduced in non-malignant astrocyte cells that expressed significantly less YB-1 compared to CSC cells (Mantwill et al., 2013). XVir-N-31 also repressed the growth of bladder cancer cells with strong YB-1 expression in vitro and intratumor delivery significantly repressed tumor volume in vivo (Lichtenegger et al., 2018). Consequently, virotherapy may prove to be an interesting avenue for targeting YB-1 overexpressing lung cancer and MPM.

Preclinical evidence in other tumors suggests that targeting YB-1 could also benefit immunotherapy in some cases. YB-1 knockdown increased the efficacy of IFN-α in renal cell carcinoma cells in vitro and in vivo (Takeuchi et al., 2013). IFN-α in combination with cisplatin provided a partial response in five out of ten patients in an open-label non-comparative phase II study of NSCLC patients (Chao et al., 1995). A phase II randomized study in SCLC patients with limited disease also showed a survival benefit of IFN-α in combination with a chemotherapy regime of carboplatin, ifosfamide and etoposide (Zarogoulidis et al., 2013). Based on these results, further investigating whether targeting YB-1 could increase the modest efficacy of IFN-α in thoracic cancer is warranted.

The use of YB-1 as a tumor-associated antigen in therapeutic vaccination has also shown promise in other cancers. YB-1 was identified as a tumor-associated antigen in neuroblastoma by serological expression of cDNA expression libraries (Zheng et al., 2009). In the context of regulatory T-cell depletion, YB-1 immunization enhanced CD8<sup>+</sup> response against neuroblastoma cells and conferred significantly higher mouse survival compared to control groups (Zheng et al., 2012). Adoptive T-cell therapy from immunized mice into neuroblastoma tumor-bearing mice also conferred a significant survival benefit and reduced tumor growth (Zheng et al., 2012). Again, further study in the context of thoracic cancer is warranted.

It must be noted that as with all current targeted therapies, it is likely that a YB-1-based approach to thoracic cancer management would benefit only a sub-population of patients. YB-1 overexpression, rather than mutation, would probably be the best predictive marker as mutations of YB-1 are very rare

FIGURE 6 | Further study required to understand the role of YB-1 and use it in the treatment and management of thoracic cancer patients. Various upstream and downstream regulatory loops and the role of YB-1 in platinum drug resistance, exosomal sorting and proliferation need further study to fully understand the biology of YB-1 in lung cancer and MPM. The mechanism of YB-1 nuclear localization is also under contention and the occurrence and significance of secreted YB-1 is yet to be determined. Standardization of a suitable YB-1 antibody for prognostic application would also be a step forward. Finally, evaluating the current strategies of YB-1 inhibition in vivo further would build a stronger case for translation into humans. Created with BioRender.com.

(Cerami et al., 2012; Gao et al., 2013). TCGA data (**Figure 1**) suggests that ∼10% of thoracic cancer patients would benefit, making it comparable to ALK inhibitors in ADC according to these datasets.

### FUTURE DIRECTIONS AND AREAS REQUIRING FURTHER STUDY

Throughout this review we have highlighted some avenues for potential future research that currently require further consideration, summarized briefly below. The YB-1/SOX2 axis needs to be further investigated in lung cancer, particularly in SCC and SCLC where the development of new therapeutic strategies is most urgent. The feed-forward loop of YB-1 and Myc also requires further investigation in the context of thoracic cancer. The roles of certain ncRNA in the dysregulation of YB-1 are also still unclear, namely the relationship between GAS5, miR-137 and YB-1 and the potential DANCR/miR-216a/YB-1 loop. Also, the apparent tumor suppressor function of GAS5 does not fit with its role in promoting YB-1 translation, which is another area requiring further attention.

While there is strong evidence supporting YB-1-driven resistance to platinum chemotherapy in other cancers, a study looking at the effect of YB-1 knockdown on cisplatin or other platinum drug sensitivity in lung cancer or MPM cells is still required. Also, while YB-1 has been shown to upregulate LRP and MRP1, the effect of these interactions on cisplatin resistance are yet to be determined. Determining the involvement of YB-1 in thoracic cancer exosomes would also be of interest. And while the mechanism underlying YB-1-driven growth in lung cancer has been studied well, similar studies in MPM cells are yet to be conducted.

Perhaps the most contentious area warranting further study relates to the regulation of YB-1 localization. Determining whether the proteolytic theory, phosphorylation theory or both is correct remains an important determination to be made. While these theories represent the most studied lines of evidence covering YB-1 nuclear localization, other posttranslational modifications could also play a role and deserve further attention, including the phosphorylation of Tyr281. However, what upstream regulator phosphorylates YB-1 here and whether this post-translational modification actually plays an important role is not yet known. Determining whether sumoylation and circadian-related YB-1 translocation occurs and is important in lung cancer and MPM patients would also be of interest.

The secretion of YB-1 into the extracellular space in response to oxidative stress has been reported in other cell types but is yet to be studied in thoracic cancers. If secretion does occur in these contexts, it would be interesting to determine whether acetylation of Lys301/304 is required, as in immune cells. Evaluating the potential interaction between secreted YB-1 and Notch3 here would also be interesting. It is also possible that secreted YB-1 could be used as a biomarker down the line, however, studies with larger patient numbers are required to determine this. Regardless, the evidence supports utilizing YB-1 as a prognostic tissue biomarker, however, universal standardization of an appropriate YB-1 antibody is would be favorable.

Finally, YB-1 remains an interesting target in thoracic cancer, but further in vivo studies delivering YB-1-targeting drugs need to be done before translation into humans can occur. This section is summarized in **Figure 6**.

## CONCLUSION

In summary, this review covers recent advances in the understanding of YB-1 in cancer biology with a focus on thoracic cancers. YB-1 plays an important role in the malignant behaviors of lung cancer and MPM including proliferation, invasion and metastasis. It also has been shown to be involved in the maintenance and behavioral regulation of CSCs. The demonstrated prognostic significance of YB-1 and developments in the delivery of RNAbased drugs mean that utilizing this multifunctional oncoprotein in the management of thoracic cancer may soon become a reality.

## AUTHOR CONTRIBUTIONS

This review was drafted by TJ and critically revised by KS, SM, AB, and GR.

## FUNDING

TJ was supported by a Ph.D. Scholarship from the Asbestos Diseases Foundation of Australia and a Top-Up Ph.D. Scholarship from the Sydney Catalyst Translational Cancer Research Centre, Australia (scholarship ID 470119385). KD was supported by a Firnberg Program Grant from the FWF (Austrian Science Fund, project number T 1062 Firnberg Program). AB was supported by a NBCF Investigator Initiated Research Scheme (IIRS-18-103). We would also like to thank the Asbestos Diseases Research Institute and Sydney Catalyst for funding the publication costs associated with this review.

## ACKNOWLEDGMENTS

We would like to acknowledge the contributions of and thank Dr. Yuen Yee Cheng and Prof. Ken Takahashi from the Asbestos Diseases Research Institute, Prof. Fiona Blyth from the University of Sydney and Prof. Monica Robotin from the University of Notre Dame for their helpful suggestions and additions to this work.

## REFERENCES


fibroblasts. Am. J. Respir. Cell Mol. Biol. 43, 576–584. doi: 10.1165/rcmb.2009- 0221OC






lung cancer cells. Cancer Chemother. Pharmacol. 80, 235–242. doi: 10.1007/ s00280-017-3355-0


**Conflict of Interest:** GR has an issued patent, US 9,006,200, covering use of microRNAs for cancer therapy.

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 Johnson, Schelch, Mehta, Burgess and Reid. 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: Why Be One Protein When You Can Affect Many? The Multiple Roles of YB-1 in Lung Cancer and Mesothelioma

#### Thomas G. Johnson1,2,3,4, Karin Schelch<sup>5</sup> , Sunali Mehta6,7, Andrew Burgess 2,3 and Glen Reid6,7 \*

*<sup>1</sup> Asbestos Diseases Research Institute, Sydney, NSW, Australia, <sup>2</sup> Cell Division Laboratory, The ANZAC Research Institute, Sydney, NSW, Australia, <sup>3</sup> School of Medicine, The University of Sydney, Sydney, NSW, Australia, <sup>4</sup> Sydney Catalyst Translational Cancer Research Centre, The University of Sydney, Sydney, NSW, Australia, <sup>5</sup> Institute of Cancer Research, Medical University of Vienna, Vienna, Austria, <sup>6</sup> Department of Pathology, University of Otago, Dunedin, New Zealand, <sup>7</sup> Maurice Wilkins Centre, University of Otago, Dunedin, New Zealand*

Keywords: lung cancer, mesothelioma, targeted therapy, biomarker, Y-box binding protein-1

#### **A Corrigendum on**

Edited and reviewed by: *Dominic C. Voon, Kanazawa University, Japan*

#### \*Correspondence: *Glen Reid glen.reid@otago.ac.nz*

#### Specialty section:

*This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology*

Received: *22 October 2019* Accepted: *06 November 2019* Published: *22 November 2019*

#### Citation:

*Johnson TG, Schelch K, Mehta S, Burgess A and Reid G (2019) Corrigendum: Why Be One Protein When You Can Affect Many? The Multiple Roles of YB-1 in Lung Cancer and Mesothelioma. Front. Cell Dev. Biol. 7:293. doi: 10.3389/fcell.2019.00293*

#### **Why Be One Protein When You Can Affect Many? The Multiple Roles of YB-1 in Lung Cancer and Mesothelioma**

by Johnson, T. G., Schelch, K., Mehta, S., Burgess, A., and Reid, G. (2019). Front. Cell Dev. Biol. 7:221. doi: 10.3389/fcell.2019.00221

In the original article, there was an error. The authors wrote "adenylation" instead of "acetylation".

A correction has been made to the **YB-1: a malignant jack of all trades**section, subsection **YB-1 is secreted into the extracellular space under cellular stress**, paragraph two:

"Y-box binding protein-1 is related on an evolutionary level to HMGB1 and is also secreted under certain cellular stresses. This was first evident in monocytes stimulated with bacterial lipopolysaccharide through an active, non-classical pathway and appears to require the same two lysine residues (Lys301/304) that are the site of acetylation in hemodialysis patients (Frye et al., 2009; Ewert et al., 2018; Figures 3–5). Secreted YB-1 stimulated DNA synthesis, cell proliferation and migration of kidney cells (Frye et al., 2009). More pertinent to thoracic cancer, YB-1 is also secreted under oxidative stress. YB-1 translationally upregulates G3BP1 under oxidative stress and localizes to cytoplasmic stress granules where it is involved in pro-survival mRNA reprogramming (Somasekharan et al., 2015). G3BP1 also promotes the invasion and metastasis of sarcoma cells in vivo (Somasekharan et al., 2015). In support, YB-1 enrichment in stress granules is also linked to its secretion to the extracellular space under oxidizing conditions (Guarino et al., 2018; Figures 4, 5). Secretion of YB-1 resulted in depletion of cytoplasmic YB-1, leaving nuclear expression intact (presumably to allow for YB-1-mediated DNA repair), while secreted YB-1 inhibited the growth of neighboring keratinocytes (Guarino et al., 2018)."

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

## REFERENCES


Copyright © 2019 Johnson, Schelch, Mehta, Burgess and Reid. 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.

# ATR-Mediated FANCI Phosphorylation Regulates Both Ubiquitination and Deubiquitination of FANCD2

#### Winnie Tan1,2, Sylvie van Twest<sup>1</sup> , Vincent J. Murphy<sup>1</sup> and Andrew J. Deans1,2 \*

<sup>1</sup> Genome Stability Unit, St Vincent's Institute of Medical Research, Fitzroy, VIC, Australia, <sup>2</sup> Department of Medicine (St Vincent's Hospital), The University of Melbourne, Melbourne, VIC, Australia

DNA interstrand crosslinks (ICLs) are a physical barrier to replication and therefore toxic to cell viability. An important mechanism for the removal of ICLs is the Fanconi Anemia DNA repair pathway, which is initiated by mono-ubiquitination of FANCD2 and its partner protein FANCI. Here, we show that maintenance of FANCD2 and FANCI proteins in a monoubiquitinated form is regulated by the ATR-kinase. Using recombinant proteins in biochemical reconstitution experiments we show that ATR directly phosphorylates FANCI on serine 556, 559, and 565 to stabilize its association with DNA and FANCD2. This increased association with DNA stimulates the conjugation of ubiquitin to both FANCI and FANCD2, but also inhibits ubiquitin deconjugation. Using phosphomimetic and phosphodead mutants of FANCI we show that S559 and S565 are particularly important for protecting the complex from the activity of the deubiquitinating enzyme USP1:UAF1. Our results reveal a major mechanism by which ATR kinase maintains the activation of the FA pathway, by promoting the accumulation of FANCD2 in the ubiquitinated form active in DNA repair.

#### Keywords: FANCI, phosphorylation, FANCD2, ubiquitination, deubiquitination

## INTRODUCTION

Many chemotherapeutic drugs kill cancer cells by inducing toxic DNA interstrand crosslinks (ICLs). ICLs prevent DNA strand separation and therefore stall DNA transcription and replication complexes (Deans and West, 2011). A critical step in the repair of replication forks stalled by ICLs is the biochemical modification of FANCD2 protein by mono-ubiquitination. Genetic deficiency in this pathway leads to Fanconi anemia (FA), characterized by hypersensitivity to DNA crosslinking agents, bone marrow failure, infertility, and cancer predisposition (Garaycoechea and Patel, 2014; Tsui and Crismani, 2019).

FANCD2 monoubiquitination is temporally and spatially controlled at stalled forks by the FA core complex (a RING E3 ligase), USP1:UAF1 (a deubiquitinating enzyme) and ATR kinase (Ishiai et al., 2017). FANCI is the heterodimeric partner of FANCD2 and plays an essential role in this regulation. In particular, FANCI itself is a target for monoubiquitination by the FA core complex (Smogorzewska et al., 2007), is a substrate of ATR kinase (Chen et al., 2015) and contains a USP1:UAF1 binding site necessary for deubiquitination of FANCD2 (Cohn et al., 2009). FANCD2 associates with FANCI as a heterodimer during ICL repair to signal DNA repair proteins that contain ubiquitin binding motifs, to promote DNA repair via homologous recombination or translesion synthesis (Smogorzewska et al., 2010; Klein Douwel et al., 2014).

#### Edited by:

Andrew Burgess, Anzac Research Institute, Australia

#### Reviewed by:

Indrajit Chaudhury, University of Minnesota, Morris, United States Angelos Constantinou, Centre National de la Recherche Scientifique (CNRS), France Michael G. Kemp, Wright State University, United States

#### \*Correspondence:

Andrew J. Deans adeans@svi.edu.au

#### Specialty section:

This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology

Received: 22 November 2019 Accepted: 03 January 2020 Published: 04 February 2020

#### Citation:

Tan W, van Twest S, Murphy VJ and Deans AJ (2020) ATR-Mediated FANCI Phosphorylation Regulates Both Ubiquitination and Deubiquitination of FANCD2. Front. Cell Dev. Biol. 8:2. doi: 10.3389/fcell.2020.00002

A crystal structure of FANCI:FANCD2 complex revealed that the lysine residues targeted for monoubiquitination in FANCI (K522 in mouse corresponding to K523 in human) and FANCD2 (K559 in mouse corresponding to K561 in human) are embedded in the dimer interface (Joo et al., 2011). This raised questions regarding the accessibility of ubiquitin and E3 enzyme into the dimer during the conjugation reaction. In contrast, several validated ATR kinase sites of FANCI (S556, 559, and 565) within an ST/Q cluster (Chen et al., 2015; Cheung et al., 2017) were found to be exposed on the FANCI surface adjacent to the heterodimer interface (Joo et al., 2011). As ATR kinase activity is required for optimal FANCD2 monoubiquitination (Shigechi et al., 2012), it was proposed that ATR phosphorylation of FANCI occurs prior to FANCI:FANCD2 ubiquitination to promote a partial opening of the FANCI:FANCD2 complex, and access to the ubiquitin ligase complex. Support for this comes from the observation that FANCI phosphorylation at serine sites 559 and 565 occurs predominantly on the monoubiquitinated form (Cheung et al., 2017). The UAF1-binding SIM domain of FANCI is next to serines 559 and 565, so an alternative explanation is that ATR acts on these sites post-monoubiquitination, to prevent USP1:UAF1 binding. In this manner, phosphorylation would prevent deubiquitination by USP1. Indeed, phosphomimetic FANCI-S559D/S565D promotes partial cellular resistance to the USP1 inhibitor ML323 (Cheung et al., 2017), but the complication in living cells is that inhibition of deubiquitination actually prevents the correct localization of FANCD2 to new DNA breaks (Castella et al., 2015). The complex intersection of phosphorylation, monoubiquitination, and deubiquitination in regulation of the FA pathway requires investigation using a defined biochemical system.

We previously reconstituted the in vitro monoubiquitination and de-ubiquitination of FANCI:FANCD2 using purified proteins. Maximal monoubiquitination required the FANCB-FANCL-FAAP100 (BL100 enzyme module) and FANCC-FANCE-FANCF (CEF substrate adaptor module) components of the FA core complex (Swuec et al., 2017; van Twest et al., 2017). Deubiquitination was more nuanced – USP1:UAF1 could efficiently remove ubiquitin from FANCD2-Ub-FANCI but not FANCD2-Ub-FANCI-Ub. However if FANCD2-Ub-FANCI-Ub is dissociated from DNA, it then becomes a USP1:UAF1 substrate (van Twest et al., 2017). In this way, USP1:UAF1 drives FANCI:FANCD2 complex toward a uniformly di-ubiquitinated state, that can only be de-ubiquitinated post-repair. We have now used this robust reconstituted system to determine if FANCI phosphorylation regulates monoubiquitination and/or deubiquitination of the complex.

## MATERIALS AND METHODS

#### Protein Purification

**Table 1** outlines the plasmids and bacmids used in this study and their derivation. Plasmids were propagated using NEB-10-beta competent cells and purified using Monarch miniprep kits (NEB). Bacmids were generated using the Multibac system (Berger et al., 2004) and purified using alkaline lysis method followed by isopropanol precipitation and resuspension in TE.

Human FANCI:FANCD2 complex and Avi-ubiquitin was purified as described in Tan et al. (2020). Xenopus laevis (frog) FANCI:FANCD2, human FANCB:FANCL:FAAP100, FANCC:FANCE:FANCF and UBE2T were expressed and purified as described in van Twest et al. (2017). USP1:UAF1, HA-ubiquitin and UBE1 were purchased from Boston Biochem. ATR-ATRIP was purchased from Eurofins DiscoverX. Lambda phosphatase was purchased from New England Biolabs.

#### FANCI Phosphomutants

Xenopus laevis StrepII-FANCD2, Flag-FANCI and human Flag-FANCI were cloned into pFastBac1 plasmid (Thermo Fisher). Expression plasmids for StrepII-FANCD2, Flag-FANCI, FANCI phosphomimic mutant (S6D) and phosphodead mutant (S6A) were previously described (Knipscheer et al., 2009; Sareen et al., 2012; van Twest et al., 2017). Xenopus FANCI with six codons encoding for serine (S) residues S557, S560, S566, S597, S618, and S630 (corresponding to serine residues S556, S559, S565, S595, S617, and S629 in human FANCI) mutated to encode either for aspartic acid (D) residues (FANCI6S:D) or alanine residues (FANCI6:A) were kindly provided by Alexandra Sobeck lab (Sareen et al., 2012). Different permutations of FANCI phosphomimic (:D) or phosphodead (:A) in the S3 clusters were generated as indicated in **Figure 2A**. Recombinant baculoviruses were generated by standard protocols (Berger et al., 2004). Trichoplusia ni (Hi5) insect cells were co-infected with Xenopus FANCI and FANCD2 viruses or infected only with human or Xenopus FANCI (Flag-tagged) or FANCD2 (StrepIItagged) (MOI = 2) and harvested after 72 h. Cell pellets were washed in 1X PBS and resuspended in 9 mL Flag Lysis buffer (50 mM Tris-HCI pH 8.0, 100 mM NaCl, 1 mM EDTA, 1X mammalian protease inhibitor (Sigma-Aldrich), 10% glycerol) or Strep Lysis buffer (50 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 1X mammalian protease inhibitor (Sigma-Aldrich), 10% glycerol, 20 µg/mL avidin, 1 mM DTT). Lysates were briefly sonicated and cleared by centrifugation for 45 min at 16,000 × g, 4◦C. 1 mL M2 Flag resin (Sigma-Aldrich) or 1 mL StrepTactin Sepharose resin (VWR International) was washed with 5 CVs each of water, followed by 0.1M glycine pH 3.5 or 0.5M NaOH, and equilibrated with 10 CVs buffer A (20 mM Tris-HCl pH 8.0, 100 mM NaCl, 10% glycerol). Lysate was added to the Flag or Streptactin resin, and incubated with gentle mixing at 4◦C for 2 h. Resin was washed with 10 CVs of buffer A and eluted in 1 CV buffer A with 5 µg/mL Flag peptide (Assay Matrix) or 100 µg/mL D-desthiobiotin (Sigma-Aldrich). Flag fractions containing FANCI:FANCD2 complex were pooled, and loaded onto a MiniQ anion purification column (GE) and equilibrated with buffer A. Using a gradient between buffer A and buffer B (20 mM Tris-HCl pH 8.0, 1M NaCl, 10% glycerol), FANCI:FANCD2 complex were eluted between 250 and 350 mM NaCl. The peak fractions were pooled, and assessed by SDS-PAGE. Protein concentration was determined by Nanodrop (Absorbance at 280 nm and calculated extinction coefficients).


TABLE 1 | Plasmids and Bacmids used in this study.

fcell-08-00002 January 31, 2020 Time: 17:12 # 3

#### In vitro Kinase and Phosphatase Assays

Ten µg of recombinant FANCI:FANCD2, FANCI, or FANCD2 were incubated in 60 µL 20 mM Tris-HCl pH 7.4, 10 mM MgAc, 0.5 mM DTT, 0.05% Tween-20, 100 mM KCl and 0.2 mM ATP in the presence of 0.1 µg of ATR:ATRIP for 30 min at 30◦C. For phosphatase experiments, 600 units of lambda phosphatase was added to reactions together with 1 mM MnCl<sup>2</sup> and incubated for 30 min at 30◦C prior to establishment of ubiquitination reaction.

### In vitro Ubiquitination and Deubiquitination Assays

Standard ubiquitination reactions contained 10 µM recombinant human AviTag-biotin-ubiquitin, 50 nM human recombinant UBE1, 100 nM UBE2T, 100 nM pUC19 plasmid or dsDNA oligonucleotide substrate, 2 mM ATP, 100 nM FANCI:FANCD2 complex wild type (WT) or ubiquitination-deficient (KR), in reaction buffer (50 mM Tris-HCl pH 7.4, 2.5 mM MgCl2, 150 mM NaCl, 0.01% Triton X-100). The dsDNA substrates were generated using oligonucleotide 1 (5<sup>0</sup> -ACGC TGCCGAATTCTACCAGTGCCTTGCTAGGACATCTTTGCC CACCTGCAGGTTCACCC-3<sup>0</sup> ) and oligonucleotide 2 (5<sup>0</sup> -GG GTGAACCTGCAGGTGGGCAAAGATGTCCTAGCAAGGCA CTGGTAGAATTCGGCAGCGT-3<sup>0</sup> ). 20 µL reactions were set up on ice and incubated at 25◦C for 90 min. To perform deubiquitination assays, FANCI:FANCD2 monoubiquitination was arrested using apyrase (NEB) and 100 nM recombinant USP1:UAF1 (Boston Biochem) were added for 30 min at room temperature. Reactions were stopped by adding 10 µL NuPage LDS sample buffer and heated at 80◦C for 5 min. Reactions were loaded onto 4–12% SDS PAGE and run using NuPAGE <sup>R</sup> MOPS buffer and assessed by western blot analysis using Flag (Jomar Life Research) or StrepII (Abcam) antibody.

## In gel Proteolytic Digestion and Mass Spectrometry Analysis of FANCI Proteins

Proteins were separated by SDS-PAGE, and the band excised for manual digestion to maximize sensitivity and efficiency. Protein bands were destained and dehydrated with 500 µL acetonitrile (ACN). Subsequently, proteins were reduced with 500 µL 10 mM dithiothreitol (DTT) in 25 mM ammonium bicarbonate (NH4HCO3) at 55◦C for 1 h and alkylated with 50 µL 55 mM iodoacetamide in 25 mM NH4HCO<sup>3</sup> at room temperature for 45 min in the dark. Samples were incubated overnight with 20 µL trypsin (125 ng, 37◦C). The resulted proteolytic peptides were subjected to sonication with 50 µL of 50% ACN, 5% formic acid and analyzed by mass spectrometry after concentration under vacuum to a 10–15 µL final volume. Mass spectra of digested protein gel bands were obtained on ESI-quadropole-timeof-flight mass spectrometer coupled to reverse-phase HPLC-MS/MS. The analysis program MASCOT was used to identify phosphorylation sites on FANCI.

## Biolayer Inferometry (BLItz) Kinetic Analysis

Kinetic titration series were performed in buffer F (20 mM Tris-HCl pH 8.0, 150 mM NaCl). 100 µg/ml FANCD2 or FANCI was diluted in buffer F and further diluted three times with a dilution factor of two. To measure the interaction between FANCD2 and FANCI, the association and dissociation times were 180 and 300 s, respectively, for every analyte concentration. In total, four Streptavidin sensors (ForteBio) were used to measure four different analyte concentrations in parallel, while one sensor was used to measure the buffer reference. All steps were performed at 25◦C with an agitation speed of 1000 rpm. Sensorgrams were measured on a ForteBio BLItz instrument and referenced against the buffer reference signal using the Data Analysis software 7.1.0.36 (ForteBio). The sensorgrams obtained with the concentrations: 334.7, 167.3, 83.7, and 41.8 nM were fitted with the BiaEvaluation software 4.1 from Biacore using a 1:1 binding model.

## Electromobility Shift Assay (EMSA)

Sixty bp fluorescently labeled dsDNA substrates were prepared as described in van Twest et al. (2017) by annealing oligonucleotide XOm1 5<sup>0</sup> -labeled with IRDye-700 (IDTDNA) and oligonucleotide XOm1.com. 25 nM dsDNA were incubated with the indicated amounts of protein for 30 min at room temperature in a 15 µL reaction containing 6 mM Tris pH 7.5, 0.1 mM EDTA, 1 mM DTT, 6% glycerol. The reaction was resolved by electrophoresis through a 6% non-denaturing polyacrylamide gel in 1X TBE (100 mM Tris, 90 mM boric acid, 1 mM EDTA) buffer and visualized by Licor Odyssey system.

#### Pull-Down Assays

Two µg of purified Flag-FANCI proteins were incubated with 2 µg of His-USP1:UAF1 (Boston Biochem) for 30 min at 25◦C in 40 µL of pull-down buffer (20 mM Tris-HCl pH 8.0, 100 mM

NaCl and 5% glycerol). 10 µL of TALON metal affinity resin (Takara Bio) were added to the reaction mixtures and were gently mixed for 30 min at 25◦C. The resin were then washed three times with 1 mL pull-down buffer. The proteins bound to the resin were separated by SDS-PAGE and stained with Coomassie blue. The band intensities of FANCI were quantitated and visualized by Licor Odyssey system.

#### RESULTS

#### Dephosphorylation of Recombinant FANCI:FANCD2 Complex Inhibits Its in vitro Monoubiquitination

We determined that a percentage of recombinant FANCI purified from baculovirus infected insect cells is phosphorylated within a previously described S/TQ cluster domain (Ishiai et al., 2008 and **Supplementary Figure 1**). Tandem mass spectrometry (MS/MS) analysis of the purified protein revealed FANCI phosphorylation of multiple serine residues (**Supplementary Figures 1B–D**), including residues previously reported S557, S560, and S566 (Cheung et al., 2017). These correspond to exposed surface residues that are conserved during evolution (**Figure 1A**), but differ in their structure between the unbound and FANCD2 bound FANCI (**Figure 1B**). ATR is the kinase predicted to phosphorylate S557, S560, and S566 in FANCI.

To determine the contribution of the identified phosphorylation sites to FANCD2:FANCI monoubiquitination, we performed an extended phosphatase treatment and then performed in vitro ubiquitination reactions using recombinant FA core complex. In both human (**Figure 1C**) and Xenopus FANCI:FANCD2 (**Figure 1D**), treatment with λ-phosphatase almost completely eliminated the monoubiquitination of FANCI, and significantly reduced the monoubiquitination rate of FANCD2. We show that addition of recombinant ATR-ATRIP kinase can restore the mono-ubiquitination levels in these preparations. As such, we observed the reappearance of positive bands when using phosphospecific antibodies raised against these three residues, only in ATR-ATRIP treated human FANCI:FANCD2 samples (**Figure 1E**). Together using phospho-specific antibodies and mass spectrometry analysis (**Supplementary Figure 1**), we show that three FANCI serine residues are the substrates of ATR kinase, and are required for optimal monoubiquitination of FANCI and FANCD2.

Free hFANCI, and to some extent hFANCD2, can also be substrates for monoubiquitination by the FA core complex (Longerich et al., 2009; Hodson et al., 2014). We therefore tested the requirement for ATR-phosphorylation in mediating the monoubiquitination of each subunit in isolation. Surprisingly, we found that human FANCI is monoubiquitinated faster in a free state, than when it is in complex with FANCD2, while free FANCD2 remains unubiquitinated (**Figure 1F**). Furthermore, free FANCI is ubiquitinated at the same fast rate as ATR:ATRIP treated FANCI, but is very slow when dephosphorylated (**Figure 1G**). This result suggests phosphorylation of FANCI specifically mediates its effects on FANCI even when it is not bound to FANCD2.

## FANCI S6D Phosphomimic Mutant Binds to FANCD2 With Reduced Affinity

Three S/TQ residues, in additional to those described above, were also predicted to be conserved ATR sites in human (Ishiai et al., 2008) and Xenopus FANCI (Sareen et al., 2012). However, these predictions were made prior to the derivation of the FANCI:FANCD2 crystal structure. The residues, S597, S618, and S630 turned out to be completely buried in the solenoid structure of FANCI and unlikely to be accessible to any kinase without complete unfolding of the protein structure. None-the-less, the xFANCI-S6D (6 serines changed to aspartate "phosphomimic") and xFANCI-S6A (6 serines changed to alanine "phosphodead") mutants were subsequently used to hypothesize that FANCI phosphorylation leads to the dissociation of FANCD2 and FANCI (Sareen et al., 2012). In our in vitro studies, ATR phosphorylation of FANCI did not cause dissociation of the FANCD2:FANCI heterodimer. This led us to suspect that phosphomutant S6A and phosphomimic S6D amino acid substitutions at these residues could be affecting the integrity of the FANCI structure and/or heterodimer formation. We examined xFANCI-S6D (where all 6 serines in the SQ cluster are changed to phosphomimic aspartate residues, **Figure 2A**) and confirmed that it has reduced copurification with xFANCD2 (**Figure 2B** and **Supplementary Figure 2**). This was a result of an order of magnitude decrease in the affinity of xFANCI-S6D compared to WT, as measured by biolayer inferometry (**Figures 2C,D,F**). xFANCI-S6A, showed normal co-purification with xFANCD2 and a similar affinity to FANCD2 as WT (**Figures 2B,E**).

We never observed S597, S618, and S630 to be phosphorylated in our mass spectrometry studies of recombinant ATR-ATRIP phosphorylated xFANCI or hFANCI protein, and a recent publication failed to identify phosphorylation of these sites in endogenous human FANCI (Cheung et al., 2017). We therefore conclude that the S557/S560/S566 (the 3SQ cluster) is likely to be the functional ATR-phosphorylation cassette in FANCI. So, we repeated the co-purification experiments using phospho-mimic/dead mutants of the 3SQ cluster in various permutations to determine if affinity changes were also observed when these sites only are altered. Unlike for xFANCI-S6D, each of the xFANCI 3SQ mutants bound to xFANCD2 in a 1:1 ratio (**Figure 2B** and **Supplementary Figure 2**). We conclude that ATR-regulation of FANCI on the surface 3SQ cluster does not lead to dissociation of the heterodimer, and likely plays some other role in regulation of FANCD2 and/or FANCI monoubiquitination.

### Phosphorylation of FANCI in the S3Q Cluster Promotes FANCI:D2 Monoubiquitination by Increasing Its Retention on DNA

To test if phospho-mimic or phospho-dead mutants within the xFANCI 3SQ cluster affect FANCI:FANCD2 ubiquitination, we performed in vitro ubiquitination assays of FANCI:FANCD2

WT and each of the FANCI-3SQ-mutants using recombinant FA core complex proteins. Four of the xFANCI phosphomimic mutants (ADD, ADA, DAD, and DDD) showed faster monoubiquitination rates than for WT xFANCI:FANCD2 complex. In particular, the FANCI-ADD and -DDD variants stimulated the highest rate of xFANCD2 monoubiquitination. Conversely, two of the FANCI-3SQ mutants (AAD and AAA) and both of the FANCI-6SQ mutants (S6D and S6A) showed

ubiquitination time course of ATR:ATRIP treated isolated human FANCI, FANCD2, or FANCI:FANCD2 complex. (G) Coomassie stained SDS-PAGE gel showing the

ubiquitination time course of untreated, phosphatase or ATR:ATRIP treated human FANCI. All data are representative of three experiments.

FIGURE 2 | FANCI phosphomimic mutant at six serine sites but not three serine sites dissociates ID2 complex. (A) Schematic of Xenopus FANCI, indicating the six phosphorylation sites within the SQ cluster region. The phosphorylation sites serine 557, 560, 566, 597, 618, and 630 were replaced either by aspartate (DQ) or alanine (AQ). The FANCI mono-ubiquitination site lysine 524 is also shown. (B) Coomassie stained SDS-PAGE of Flag-ID2 complex with StrepII-FANCD2 co-expressed with various flag-FANCI phosphomutants. (C–E) Biolayer inferometry (BLItz) sensorgrams obtained using StrepII-FANCD2-loaded biosensors in 20 ng/mL solution, with red dotted lines indicating the start of binding (left) and dissociation (right) phases. Biosensors loaded with StrepII-FANCD2 were incubated with different concentrations of FANCI wild type (WT), S6A or S6d mutants, as indicated to generate a series of sensorgrams. (F) Summary of dissociation constant (KD), association (Ka), dissociation (Kd) rate constants obtained from the sensorgrams. Data are representative of four experiments.

lower xFANCD2 monoubiquitination than WT (**Figure 3A**). These results demonstrate that phosphomimic mutations in S560 together with S565 are sufficient to maximally stimulate monoubiquitination of FANCI:FANCD2 complex by the FA core complex (**Figure 3B**).

Previous studies have shown that optimal FANCI:FANCD2 monoubiquitination requires the association of the complex with DNA (Sato et al., 2012; Longerich et al., 2014; van Twest et al., 2017) leading us to speculate that phosphorylation may increase the affinity of FANCI:FANC2 complex to DNA. To examine this hypothesis, we performed electromobility shift assays (EMSAs). Remarkably, the DNA binding measured by EMSA shift was greatly reduced in lambda-phosphatase treated FANCI:FANCD2 complex or FANCI AAA mutant compared to control (**Figure 3C**). DNA binding was not only restored after phosphorylation by ATR and in the FANCI DDD mutant, but occurred at significantly lower FANCI:FANCD2 concentration (i.e., higher affinity). We further tested the association of isolated FANCI with DNA and found that phosphatase treatment also caused a reduction in DNA binding that was restored by ATR-kinase treatment (**Figure 3D**). These results are consistent with ATR-phosphorylation creating a higher affinity of FANCI (and associated FANCD2) for DNA binding, leading to a greater stimulation of monoubiquitination of FANCD2 by the FA core complex.

#### FANCI Phosphorylation or Phosphomimics Also Protect FANCD2 From Deubiquitination

Published evidence from cell-based studies suggests phosphorylation of FANCI by ATR could also regulate FANCD2 deubiquitination (Cheung et al., 2017). To examine whether FANCI deubiquitination activity by USP1:UAF1 is regulated by the 3SQ cluster, we performed in vitro deubiquitination reactions where xFANCIUb:xFANCD2Ub was treated with lambda phosphatase after ubiquitination. Most of the xFANCIUb was resistant to deubiquitination in the native state, but dephosphorylation accelerated the deubiquitination. xFANCD2Ub was also significantly faster. Re-phosphorylation of the xFANCIUb:xFANCD2Ub restored the slower rate of USP1:UAF1-mediated deubiquitination (**Figure 4A**). This was due to a direct effect of ATR phosphorylation on FANCI because xFANCD2 and FANCI-(AAA)Ub was also deubiquitinated significantly faster than xFANCIUb or xFANCI-(ADD)Ub . Overall, the rate of FANCD2 deubiquitination when bound to FANCI-ADD or WT-FANCI is fourfold slower than for FANCI-AAA (**Figures 4B,C**).

Taken together, our results suggest that ATR-mediated FANCI phosphorylation both promotes monoubiquitination and inhibits deubiquitination of FANCD2.

#### DISCUSSION

In humans, ATR kinase strongly influences the activation of the Fanconi Anemia DNA repair pathway (Andreassen et al., 2004). In particular, FANCI phosphorylation by ATR kinase was proposed to be an "on switch" for FANCD2 monoubiquitination (Ishiai et al., 2008; Cheung et al., 2017). The present work uncovers additional evidence that it is the direct phosphorylation of FANCI by ATR that influences the biochemical rate of FANCD2 ubiquitination and deubiquitination to reveal fundamental insights into how FANCD2-monoubiquitination is maintained at sites of DNA damage.

We found that three sites on the surface of FANCI, serines 557, 560 and 565, are the main targets of recombinant ATR kinase in vitro. Phosphomimic mutations at these sites, create a charge effect similar to true phosphorylation, and stimulated the in vitro monoubiquitination of FANCD2. However, we did not see this effect when we used a mutant called S6D. We propose that the additional three serines (S597, S618, and S630) mutated in this variant may cause structural malformation of FANCI, given their location within α-helices of the solenoid structure of the protein. As such, FANCI-S6D only binds weakly to FANCD2 and does not stimulate its mono-ubiquitination. This finding supports the in vivo cell-based studies of Cheung et al. (2017) where only these three phospho-serines were identified in massspectrometry based analysis, but does not support the concept of FANCI dissociation from FANCD2 after phosphorylation or monoubiquitination (Sareen et al., 2012). The actions of ATR kinase on the FANCI:FANCD2 complex are therefore likely to be restricted to these sites associated with DNA damage response.

We previously showed that DNA-PKcs, a kinase related to ATR, could phosphorylate FANCI but could not stimulate FANCD2 monoubiquitination (van Twest et al., 2017). This indicates that DNA-PKcs is a poor substitute for ATR, or that it cannot catalyze the correct combination of FANCI phosphorylation events necessary for activation. However, it's possible that other kinases could further modulate the temporal and/or spatial localization of FANCD2 and FANCI monoubiquitination in vitro or in cells. For example, Casein Kinase 2 phosphorylation of FANCD2 at a cluster of serines between residues 882–898 inhibits its DNA association and subsequent monoubiquitination (Lopez-Martinez et al., 2019), while ATM phosphorylates many residues in FANCD2 independent of monoubiquitination, but only during S-phase and after ionizing radiation (Taniguchi et al., 2002b; Ho et al., 2006). Other kinases with an important role in DNA replication and DNA damage response have yet to be explored, although several such as Chk1 and CDK also phosphorylate subunits in the FA core complex (Deans et al., 2006; Wang et al., 2007). Kinases may also regulate enigmatic dimerization and monoubiquitination-independent functions of FANCI and FANCD2, that appear to be necessary only after certain types DNA damage (Sareen et al., 2012; Chaudhury et al., 2013; Chen et al., 2016). A thorough biochemical investigation of the complex interplay of known and unknown kinases in the FA pathway is likely to be fruitful.

Deubiquitination of FANCD2 and FANCI appears to be as important as ubiquitination, in the regulation of the Fanconi Anemia pathway. There are at least two reasons for this: (1) it prevents the retention of FANCD2 at spurious, non-repair sites in the nucleus and (2) it allows completion of DNA repair (Cohn et al., 2009; Tan and Deans, 2017). A major function

of FANCIUb appears to be the prevention of FANCD2Ub deubiquitination, but only while it is DNA associated. Our data show that a phosphomimic mutant of FANCI is further protective to FANCD2Ub deubiquitination. Supporting our results, Cheung et al. (2017) showed that FANCI phosphomimic mutant at serine 560 and 566 prevents deubiquitination of FANCD2 in vivo (Cheung et al., 2017). Together, these results suggest that FANCI dephosphorylation might be required for efficient FANCI:FANCD2 deubiquitination. Additional studies will be required to understand the dephosphorylation of FANCI

and whether it alters FANCD2 ubiquitination in cells. CTDP1 and PTEN are phosphatases known to act in the FA pathway, although neither have been demonstrated to act directly on FANCI (Vuono et al., 2016; Hu et al., 2019). Our in vitro reconstitution system will be useful for answering mechanistic questions regarding FANCI dephosphorylation and for identifying potential phosphatases that regulate the FA pathway.

FANCD2 monoubiquitination by FA core complex and deubiquitination by USP1:UAF occurs at steady state levels associated with ongoing replication (Taniguchi et al., 2002a; Liang et al., 2016). We propose that upon replication fork stalling, ATR kinase is activated, with locally high concentrations of active ATR promoted by the activity of FANCM:FAAP24:hCLK2 complex (Collis et al., 2008). ATR phosphorylates the FANCI-3SQ cluster. From structural models, phosphorylation of these residues are predicted to increase the surface area of interaction between FANCD2 and FANCI by at least twenty-five percent (Joo et al., 2011), which likely explains the increased affinity of the complex on DNA. This would stabilize the complex at damage sites, leading to its increased monoubiquitination by the FA core complex, which is also localized by FANCM (Deans and West, 2009). We also predict that a conformational change induced by phosphorylation of the S3Q cluster alters the association of the adjacent SLIM domain of FANCI with USP1:UAF1, as a mechanism for the observed reduction in USP1:UAF1 binding in S3Q phosphomimic proteins. Therefore, deubiquitination is prevented until repair is completed and the ATR kinase is deactivated (**Figure 4D**).

Given the critical importance of FANCI as an ATR target, the phosphorylation state of the FANCI-3SQ cluster may be an appropriate biomarker of the effectiveness of several classes of ATR kinase inhibitor currently in clinical trial for the treatment of cancer (Lecona and Fernandez-Capetillo, 2018). Our work also suggests that direct modulation of FANCI phosphorylation plays a twofold role in stabilizing FANCD2 monoubiquitination, with relevance to understanding and treating both Fanconi anemia and cancer.

#### DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

#### AUTHOR CONTRIBUTIONS

fcell-08-00002 January 31, 2020 Time: 17:12 # 10

WT performed protein purification, biolayer inferometry, ubiquitination, deubiquitination, and DNA binding assays. ST performed deubiquitination assays. VM generated FANCI phosphomutants and assisted with protein purifications. WT and AD wrote the manuscript and designed experiments for this article.

#### FUNDING

This work was supported by grants from the Fanconi Anemia Research Fund, the National Health and Medical Research

#### REFERENCES


Council (GNT1123100 and GNT1181110 to AD), and the Victoria government IOS program. WT was supported by an Australian Government Research Training Scheme postgraduate scholarship. AD was a Victorian Cancer Agency mid-career fellow.

#### ACKNOWLEDGMENTS

We would like to thank Dr. Alexandra Sobeck (University of Minnesota) for providing xFANCI WT, monoubiquitinationdead (KR), phosphomimic (S6A) and phosphodead (S6D) mutant constructs. We thank Robert Cheung for antibodies against FANCI serine 556, 559, and 565. We thank Genome Stability lab members for thoughtful contributions.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2020.00002/ full#supplementary-material


**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 Tan, van Twest, Murphy and Deans. 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.

fcell-08-00002 January 31, 2020 Time: 17:12 # 11

# Two-Faced: Roles of JNK Signalling During Tumourigenesis in the Drosophila Model

John E. La Marca\* and Helena E. Richardson

Richardson Laboratory, Department of Biochemistry and Genetics, La Trobe Institute for Molecular Science, La Trobe University, Melbourne, VIC, Australia

The highly conserved c-Jun N-terminal Kinase (JNK) signalling pathway has many functions, regulating a diversity of processes: from cell movement during embryogenesis to the stress response of cells after environmental insults. Studies modelling cancer using the vinegar fly, Drosophila melanogaster, have identified both pro- and antitumourigenic roles for JNK signalling, depending on context. As a tumour suppressor, JNK signalling commonly is activated by conserved Tumour Necrosis Factor (TNF) signalling, which promotes the caspase-mediated death of tumourigenic cells. JNK pathway activation can also occur via actin cytoskeleton alterations, and after cellular damage inflicted by reactive oxygen species (ROS). Additionally, JNK signalling frequently acts in concert with Salvador-Warts-Hippo (SWH) signalling – either upstream of or parallel to this potent growth-suppressing pathway. As a tumour promoter, JNK signalling is co-opted by cells expressing activated Ras-MAPK signalling (among other pathways), and used to drive cell morphological changes, induce invasive behaviours, block differentiation, and enable persistent cell proliferation. Furthermore, JNK is capable of non-autonomous influences within tumour microenvironments by effecting the transcription of various cell growth- and proliferation-promoting molecules. In this review, we discuss these aspects of JNK signalling in Drosophila tumourigenesis models, and highlight recent publications that have expanded our knowledge of this important and versatile pathway.

Keywords: JNK, Drosophila, tumourigenesis, scrib, Ras, apoptosis

## INTRODUCTION

Jun N-terminal kinase signalling is a conserved Mitogen-Activated Protein Kinase (MAPK, N.B. for a glossary of abbreviated terms refer to **Supplementary Table 1**) signalling pathway which, through a conserved kinase cascade, acts to influence gene transcription, and hence the cellular response to various stimuli. In Drosophila, the sole JNK is Basket (Bsk; orthologue of human JNK1, JNK2, and JNK3, also known as MAPK8, MAPK9, and MAPK10, respectively), which acts to phosphorylate and activate a number of transcription factors (TFs). The best known JNK-activated TFs are Jun-related antigen (Jra) and Kayak (Kay; whose closest human orthologues are JUN (Jun proto-oncogene, AP-1 TF subunit) and FOS (Fos proto-oncogene, AP-1 TF subunit), respectively) (**Figure 1**), which together make up the heterodimeric Activator Protein-1 (AP-1) complex. While other Bsk targets also exist, upstream of Bsk the signalling network is much more complex. In order to activate JNK/Bsk there are at least two JNK kinases (JNKKs) – Hemipterous (Hep) and MAP kinase kinase 4 (Mkk4; orthologues of human MAP2K7 and MAP2K4, respectively) – and at least four JNKK kinases (JNKKKs) – Slipper (Slpr, human orthologues MAP3K9, MAP3K10,

#### Edited by:

Liz Caldon, Garvan Institute of Medical Research, Australia

#### Reviewed by:

Jormay Lim, National Taiwan University, Taiwan Rajprasad Loganathan, Johns Hopkins University, United States

#### \*Correspondence:

John E. La Marca e.lamarca@latrobe.edu.au

#### Specialty section:

This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 12 November 2019 Accepted: 17 January 2020 Published: 05 February 2020

#### Citation:

La Marca JE and Richardson HE (2020) Two-Faced: Roles of JNK Signalling During Tumourigenesis in the Drosophila Model. Front. Cell Dev. Biol. 8:42. doi: 10.3389/fcell.2020.00042

MAP3K11, and MAP3K21), Wallenda (Wnd, human orthologues MAP3K13 and MAP3K12), TGFβ-associated kinase 1 (Tak1, human orthologue MAP3K7), and Apoptotic signal-regulating kinase 1 (Ask1, human orthologues MAP3K15 and MAP3K5) (**Figure 1**). These kinases follow an ever-growing multitude of signalling pathways, molecules, and stimuli that feed into the activation of JNK. Despite this complexity, the kinase core of Hep-Bsk is generally considered to be the canonical and main effector of JNK signalling.

Originally identified in the Heidelberg genetic screens as a mutant that had improper dorsal closure during embryogenesis (Nüsslein-Volhard et al., 1984), it was not until a decade later that bsk was determined to be the orthologue of the mammalian JNK genes (Riesgo-Escovar et al., 1996; Sluss et al., 1996), a discovery that followed closely on the heels of the identification of hep as a JNKK (Glise et al., 1995). Since then, astonishingly large bodies of work have identified JNK signalling as being critical in a multitude of biological processes, such as regulating cell morphology and migration behaviours (via inducing the expression of genes like the actin cross-linker cheerio (cher) (Pastor-Pareja et al., 2004; Külshammer and Uhlirova, 2013), or by upregulating targets like the integrin-associated scaffolding protein Paxillin (Huang et al., 2003; Llense and Martín-Blanco, 2008; Leong et al., 2009)), regulating organ size (Willsey et al., 2016), and promoting cell death by upregulating genes like head involution defective (hid) and reaper (rpr) (Moreno et al., 2002; Luo et al., 2007).

With such diverse functionality, it is perhaps no surprise that JNK signalling has also emerged as a key player in tumourigenesis in Drosophila, something it shares with its mammalian orthologue (reviewed in Wagner and Nebreda, 2009; Wu et al., 2019). Almost all human orthologues of the core JNK signalling hierarchy have been implicated in multiple cancers, though their roles are often not well understood, and are often context dependent. We have summarised some of the recent literature concerning the links between JNK signalling and human cancer in our **Supplementary Material** (**Supplementary Table 2**). The role of JNK in tumourigenesis in flies is relatively better understood, but still exceedingly complex, with the pathway fulfilling different, seemingly opposing roles depending on the context. Simply put, JNK signalling is capable of both eliminating pre-tumourigenic cells via apoptosis, but also can cooperate with various genetic insults to promote tumourigenesis. In this review, we will examine these pro- and anti-tumourigenic roles of JNK signalling, non-autonomous roles of the pathway during tumourigenesis, and the various activation modes of the pathway in these contexts.

## ANTI-TUMOURIGENIC JNK SIGNALLING

The anti-tumourigenic effect of JNK signalling ultimately induces cell death due to the upregulation of apoptosisinducing genes like hid and rpr, and the activation of caspases (Moreno et al., 2002; Luo et al., 2007; Shlevkov and Morata, 2012; Li et al., 2019). One scenario where this role is well documented is upon the clonal disruption of cell polarity.

FIGURE 1 | Conservation of JNK signalling core. The kinase core of JNK signalling is well-conserved between flies and mammals. In Drosophila, there are at least four JNKKKs: Slpr (Slipper), Wnd (Wallenda), Tak1 (TGFβ-associated kinase 1), and Ask1 (Apoptotic signal-regulating kinase 1). All have multiple human orthologues, which are Mitogen-activated protein kinase kinase kinases (MAP3Ks), with the most highly conserved being shown here. Slpr is most closely related to MAP3K9, MAP3K10, MAP3K11, and MAP3K21, Wnd to MAP3K13 and MAP3K12, Tak1 to MAP3K7, and Ask1 to MAP3K15 and MAP3K5. The Drosophila JNKKs are Hep (Hemipterous) and Mkk4 (MAP kinase kinase 4), which have human orthologues amongst the Mitogen-activated protein kinase kinases (MAP2Ks). Hep is most closely related to MAP2K4, and Mkk4 to MAP2K7. While Drosophila has only one JNK, Basket (Bsk), there are three conserved Mitogen-activated protein kinase (MAPK) orthologues to Bsk in humans: MAPK10, MAPK8, and MAPK9. JNKs upregulate the activity of various TFs, the best known of which are those that form the heterodimeric AP-1 complex. In Drosophila, those TFs are Kay (Kayak) and Jra (Jun-related antigen). In humans, the orthologues of Kay are FOS (Fos proto-oncogene, AP-1 transcription factor subunit), FOSL1 (FOS like 1, AP-1 transcription factor subunit), and FOSL2 (FOS like 2, AP-1 transcription factor subunit), while the orthologues of Jra are JUN (Jun proto-oncogene, AP-1 transcription factor subunit) and JUND (JunD proto-oncogene, AP-1 transcription factor subunit).

Cell polarity is the asymmetric distribution of proteins within a cell, and the disruption of polarity is considered one of the hallmarks of cancer (Hanahan and Weinberg, 2011).

In Drosophila, apico-basal polarity is critical to the proper formation of larval epithelial tissues, such as the wing and eye-antennal imaginal discs, and is controlled predominantly by the mutually antagonistic behaviours of three polarity protein modules: Scribble/Discs large 1/Lethal (2) giant larvae (Scrib/Dlg1/L(2)gl), Crumbs/Stardust/Patj (Crb/Sdt/Patj), and Bazooka/Par-6/atypical protein kinase C (Baz/Par-6/aPKC) (reviewed in Tepass, 2012). The largest body of work has examined Scrib/Dlg1/L(2)gl, where animals wholly mutant for any of these components produce neoplastic tumours, in which tissues overproliferate and show aberrant differentiation alongside a disorganised morphology – scrib, dlg1, and l(2)gl are therefore referred to as neoplastic tumour suppressor genes (nTSGs) (Bilder, 2004). However, while these wholly mutant tissues overgrow, clonal patches of epithelial tissue mutant for these genes are eliminated via a process termed cell competition. Cell competition is a surveillance mechanism that leads to the active elimination of cells that are "less fit" by their "more fit" neighbouring cells (reviewed in Fahey-Lozano et al., 2019; Ohsawa, 2019). Clones mutant for scrib (scrib−/−) are eliminated by apoptosis in Drosophila imaginal tissues, and this process is dependent on JNK signalling activity, as blocking JNK enables the cells to survive (**Figure 2**; Brumby and Richardson, 2003). These polarity mutant clones are therefore thought of as pretumourigenic, since if they are not removed tumours will develop. Furthermore, while scrib−/<sup>−</sup> cells have enhanced proliferative capacity via JNK-independent upregulation of the cell cycle regulator, Cyclin E (CycE) (Brumby and Richardson, 2003; Leong et al., 2009), JNK signalling promotes their apoptosis, and the balance between these opposing phenotypes can be pushed in either direction by enhancing or disrupting JNK (Uhlirova and Bohmann, 2006).

Jun N-terminal kinase signalling was found to be primarily upregulated in cells at the borders of scrib−/<sup>−</sup> clones and wildtype tissue, suggesting that its upregulation was not a direct consequence of scrib mutation (Leong et al., 2009). What, then, was the source? It was determined that JNK signalling, and the elimination of scrib or dlg1 mutant clones, was dependent on activation of the pathway by TNF signalling – the Drosophila TNF, Eiger (Egr), binds to the TNF Receptors (TNFRs) Wengen (Wgn) and/or Grindelwald (Grnd), and eventually triggers activation of the kinase core of the JNK signalling pathway (**Figure 2**; Igaki et al., 2009; Andersen et al., 2015). Mislocalisation of Egr to endosomes within the scrib−/<sup>−</sup> cells, rather than its upregulation, was determined to be the cause of the ectopic JNK signalling, with endocytosis increased in the clones – though, notably, endocytosis was only increased when wildtype tissue was adjacent to the scrib−/<sup>−</sup> cells (Igaki et al., 2009). Although Egr was detectable in all the epithelial cells in the scrib−/<sup>−</sup> mosaic tissue, genetic analyses showed that it acts in an autocrine manner within the scrib−/<sup>−</sup> cells (Igaki et al., 2009), but this is unlikely to be the whole story – Egr was later shown to also be produced by haemocytes, circulating macrophage-like cells within the Drosophila haemolymph, and that its presence in these cells was sufficient for the activation of JNK in scrib−/<sup>−</sup> cells (**Figure 2**; Vidal, 2010). While not investigated in scrib−/<sup>−</sup> cells specifically, haemocyte attraction was shown to depend on JNK-mediated secretion of a cleaved form of the protein Tyrosyl-tRNA synthetase (Casas-Tintó et al., 2015). Additionally, egr is necessary for the elimination of dlg1-knockdown cells in wing imaginal discs, and for apoptosis within wholly scrib (or dlg1) mutant animals (Cordero et al., 2010). Regardless of the source of Egr, JNK signalling has a key role in eliminating scrib−/<sup>−</sup> cells during cell competition by promoting apoptosis – however, blocking apoptosis in these scrib−/<sup>−</sup> clones is not, in fact, as effective at preventing their elimination (and thus promoting tumourigenesis) as simply blocking JNK signalling, suggesting other removal mechanisms are at play (Brumby and Richardson, 2003). To wit, JNK signalling has been demonstrated to upregulate the genes slit, roundabout 2 (robo2), and enabled (ena), which together act to promote scrib−/<sup>−</sup> cell extrusion from the tissue – Sli-Robo2-Ena signalling disrupts Shotgun (shg, a.k.a. E-cadherin), and also forms a positive feedback loop with JNK signalling by promoting F-actin accumulation (**Figure 2**; Vaughen and Igaki, 2016).

Jun N-terminal kinase signalling is not exclusively active within polarity-impaired cells during cell competition. The pathway has also been shown to be active within their wildtype neighbours – Egr-dependent JNK activation in the wildtype cells promotes signalling via PDGF- and VEGF-receptor related (Pvr), which in turn activates Ced-12 and Myoblast city (Mbc) to promote engulfment and removal of the mutant cells by their healthy neighbours (**Figure 2**; Ohsawa et al., 2011). Furthermore, mechanisms have been identified that are involved in the recognition of polarity-impaired cells. Protein tyrosine phosphatase 10D (Ptp10D) is expressed on the surface of scrib−/<sup>−</sup> cells, and is bound and activated by the ligand Stranded at second (Sas) expressed on the surface of their wildtype neighbours (Yamamoto et al., 2017). Activated Ptp10D suppresses epidermal growth factor receptor (Egfr) activity, allowing JNK signalling to act in its anti-tumourigenic capacity (Yamamoto et al., 2017). If Egfr activity were permitted due to sas or Ptp10D downregulation, activated Ras-MAPK signalling would occur alongside JNK signalling, the consequences of which we will discuss in a later section ("Pro-tumourigenic JNK signalling").

Interestingly, l(2)gl mutant (l(2)gl−/−) clones and tissues behave somewhat differently to scrib−/<sup>−</sup> cells, though they also upregulate JNK signalling, and are eliminated by JNK-dependent apoptosis (Froldi et al., 2010; Grzeschik et al., 2010; Menéndez et al., 2010; Tamori et al., 2010). Autocrine Egr is dispensable in l(2)gl−/<sup>−</sup> clones, as they still upregulate JNK signalling even when egr is knocked down in these cells; however, it is thought that l(2)gl−/<sup>−</sup> tissue growth and survival is more dependent on levels of the oncogenic TF Myc than on JNK signalling (Froldi et al., 2010).

As mentioned, scrib−/<sup>−</sup> clones exhibit ectopic proliferation, but their potential to overgrow is modulated by JNK signallinginduced apoptosis. Inhibiting JNK allows these clones to overgrow, but where does this capability come from? One important growth regulating pathway is the SWH signalling pathway, which is a conserved inhibitor of tissue growth, which functions by phosphorylating and thus cytoplasmically sequestering the TF coactivator, Yorkie (Yki). This prevents Yki from interacting with and activating the TF Scalloped

by upregulating targets such as CycE and Diap1. Gene and protein name abbreviations used in the diagram are as follows: Eiger (Egr), Grindelwald (Grnd), Wengen (Wgn), PDGF- and VEGF-receptor related (Pvr), Myoblast city (Mbc), Yorkie (Yki), Cyclin E (CycE), Death-associated inhibitor of apoptosis 1 (Diap1), Wallenda (Wnd), reaper (rpr), head involution defective (hid), unpaired 1 (upd1), unpaired 2 (upd2), unpaired 3 (upd3), roundabout 2 (robo2), enabled (ena), Shotgun (Shg).

(Sd, a TEAD family TF) and its target genes, which include the cell cycle regulator CycE and the apoptosis inhibitor Deathassociated inhibitor of apoptosis 1 (Diap1) (reviewed in Misra and Irvine, 2018). Many reports have indicated that both the JNK and SWH signalling pathways are interwoven but elucidating exactly how they interact in polarity-impaired tumours has been difficult. Initial experiments in eye-antennal imaginal disc scrib−/<sup>−</sup> clones indicated that reporters for certain Yki target genes were highly expressed (though variable), suggesting some level of SWH inhibition may be in effect and, indeed, scrib−/<sup>−</sup> cell overproliferation was found to depend on Yki and Sd activity – however, blocking JNK signalling resulted in SWH inhibition, as revealed by the upregulation of Yki targets (Doggett et al., 2011). More light was shed on these results in a later study, which more closely examined SWH signalling in scrib−/<sup>−</sup> clones in both eye-antennal and wing imaginal discs and found that Yki

activity was specifically dependent on whether cell competition was occurring (Chen et al., 2012). Specifically, scrib−/<sup>−</sup> clones not facing cell competition (by artificially lowering the fitness of their neighbours, or in wholly scrib−/<sup>−</sup> tissue) showed elevated Yki activity and the cells overgrew, but in scrib−/<sup>−</sup> cells undergoing competition Yki activity was downregulated, and this downregulation was mediated by JNK signalling (though other mechanisms are likely to contribute) (**Figure 2**; Chen et al., 2012). Notably, Yki activation is also required in the wildtype neighbours for their compensatory proliferation, where it is thought to act parallel to Janus kinase-Signal Transduction and Activator of Transcription (Jak-STAT) signalling to promote the elimination of the scrib−/<sup>−</sup> cells (**Figure 2**; Chen et al., 2012; Schroeder et al., 2013).

Somewhat incongruous with these results is data from wing imaginal disc regions (not clones), where the induction of apoptosis in regions of disc tissue induced Yki activation in adjacent cells to promote compensatory proliferation (Sun and Irvine, 2011). Here, Yki activation was dependent on JNK signalling and, indeed, JNK signalling initiation was sufficient to induce Yki activity (Sun and Irvine, 2011). Interestingly, initiating neoplastic growth via the knockdown of dlg1/l(2)gl in large tissue regions also upregulated Yki activity and, in these instances, Yki upregulation was again dependent on JNK activity – possibly this is similar to the aforementioned scrib−/<sup>−</sup> clones when dodging cell competition (Sun and Irvine, 2011). JNK signalling in these contexts may mediate Yki activation via downregulation of SWH signalling at the level of the Warts (Wts) protein kinase – JNK activity phosphorylates Ajuba LIM protein (Jub), which in turn binds and inactivates Wts (Sun and Irvine, 2013).

Interactions between JNK signalling and SWH/Yki in the context of JNK acting in an anti-tumourigenic role are not limited to the context of disrupted cell polarity. Inducing cytokinesis failure in wing imaginal disc cells promotes aneuploidy, which can lead to tumourigenesis, but JNK signalling is also upregulated in these cells and acts to downregulate Diap1 and a cell cycle regulator, String (Stg, orthologue of the human CDC25 proteins), thus promoting cell death and suppressing cell proliferation (Gerlach et al., 2018). However, SWH inhibition or Yki activation can bypass this JNK-mediated tumourigenesis prevention by facilitating Diap1 and Stg upregulation (Gerlach et al., 2018).

Overall, JNK signalling, primarily in its capacity as a proapoptotic regulator, plays a fundamental role as an antitumourigenic signal. This is particularly apparent (and best studied) in the context of polarity-deficient pre-tumourigenic cells, where it acts both autonomously and non-autonomously in facilitating their elimination. Furthermore, JNK signalling has a complex relationship with the SWH signalling pathway, where its interactions vary depending on context – in polarity-impaired, pre-tumourigenic cells, activated JNK signalling suppresses Yki activity (Doggett et al., 2011; Chen et al., 2012), whereas in regenerating wing imaginal disc tissue JNK signalling suppresses SWH signalling, and hence promotes Yki activity (Sun and Irvine, 2011). This role in promoting Yki activity is a hint at the two-faced nature of JNK signalling – depending on context, it can also be pro-tumourigenic.

## PRO-TUMOURIGENIC JNK SIGNALLING

Pro-tumourigenic JNK signalling in Drosophila was discovered during the study of cooperative tumourigenesis. Cancer is a multi-step process, and cooperative tumourigenesis is the phenomenon by which different genetic lesions in a cell, or in different cells, can cooperate to drive the initiation and progression of cancer. In Drosophila, cooperative tumourigenesis was discovered by one group of researchers looking at the consequences of introducing oncogenic mutations into scrib−/<sup>−</sup> clones (Brumby and Richardson, 2003), and simultaneously by another group of researchers screening for mutations that cooperate with oncogenic mutations to produce metastatic tumours (Pagliarini and Xu, 2003). Coming from opposite directions, both groups identified that expressing an activated form of Ras oncogene at 85D (Ras85D – the most commonly used activated form is often referred to as RasV12, but hereafter is referred to as Ras85DV12) cooperated with mutations in cell polarity regulator genes such as scrib to produce overgrown and invasive tumours in eye-antennal imaginal discs (**Figure 3**; Brumby and Richardson, 2003; Pagliarini and Xu, 2003). Ras85D is a GTPase, and canonically acts via the "Ras-MAPK" signalling pathway to effect gene transcription.

These initial studies did not determine a role for JNK signalling in Ras85DV12/scrib−/<sup>−</sup> tumours, but it was clear that JNK-mediated apoptosis must be blocked in some way (Brumby and Richardson, 2005). Surprisingly, the JNK signalling pathway was in fact strongly upregulated in Ras85DV12/scrib−/<sup>−</sup> tumours (and was not upregulated in the benign tumours formed after expression of Ras85DV<sup>12</sup> in isolation) (Igaki et al., 2006; Uhlirova and Bohmann, 2006). Indeed, JNK signalling was necessary (and sufficient when induced via activated Hep, but insufficient when induced via Egr overexpression) for Ras85DV12/scrib−/<sup>−</sup> tumour invasiveness, which was further demonstrated to be due to the JNK-induced transcription of Matrix metalloproteinase 1 (Mmp1) (**Figure 3**; Igaki et al., 2006; Uhlirova and Bohmann, 2006). Mmp1 is from a family of genes strongly linked to cell motility, and is necessary for basement membrane degradation and invasive behaviours by Ras85DV12/scrib−/<sup>−</sup> tumours (Srivastava et al., 2007). Other JNK signalling targets that are thought to contribute to invasive behaviours include the actin cross-linker cher (Pastor-Pareja et al., 2004; Külshammer and Uhlirova, 2013) and the integrinassociated scaffolding protein Paxillin (Huang et al., 2003; Llense and Martín-Blanco, 2008; Leong et al., 2009).

It was therefore thought that JNK signalling was switched from pro-apoptotic to pro-growth/proliferation in the face of Ras85DV12, a role that was known from experiments in undead cells – when apoptosis is triggered, but caspase activity is prevented via expression of the effector caspase inhibitor p35, cells are referred to as "undead" and undergo behaviours associated with cell death, but remain in the tissue and can induce non-autonomous effects (see also section "Non-autonomous effects of JNK signalling") (reviewed in Martín et al., 2009). In Drosophila wing imaginal disc cells, undead cells induce JNKdependent overproliferation in their wildtype neighbours, which reflects this observed role reversal (Pérez-Garijo et al., 2004;

Ryoo et al., 2004). Similarly, JNK signalling resulting from non-apoptotic levels of caspase activation in undead cells also autonomously promotes invasiveness via Mmp1 upregulation (Rudrapatna et al., 2013).

The aforementioned studies demonstrate that the malignancy of Ras85DV12/scrib−/<sup>−</sup> tumours depends on JNK signalling, and the downstream effectors of that JNK signal have recently been identified. Three key TFs act downstream of JNK signalling in Ras85DV12/scrib−/<sup>−</sup> tumours – Kay (a.k.a. Fos), Ftz TF 1 (Ftz-f1), and Ets at 21C (Ets21C) (Külshammer et al., 2015). Similarly, another study demonstrated that a majority of the phenotypes seen in Ras85DV12/scrib−/<sup>−</sup> tumours can be traced back to a network of around 10 interconnected TFs that act downstream of JNK, SWH, and Jak-STAT signalling (Atkins et al., 2016), with Jak-STAT signalling being a key contributor to Ras85DV12/scrib−/<sup>−</sup> tumour overgrowth (**Figure 3**; Wu et al., 2010; Atkins et al., 2016). Regardless, it was shown that Kay was solely responsible for JNK-related differentiation defects and Mmp1 upregulation, but both Kay and Ftz-f1 are necessary for tumour invasiveness, and Ets21C overexpression can cooperate with Ras85DV<sup>12</sup> to produce invasive (but non-overgrowing) clones (Külshammer et al., 2015).

Jun N-terminal kinase signalling inhibits differentiation of Ras85DV12/scrib−/<sup>−</sup> tumour cells via an unclear mechanism (**Figure 3**), but it can be observed in the eye imaginal disc tissue due to decreased expression of embryonic lethal abnormal

vision (Elav), a marker of photoreceptor cell differentiation, which is restored upon JNK signalling inhibition (Leong et al., 2009). When cooperative tumourigenesis between Ras85DV<sup>12</sup> and scrib−/<sup>−</sup> was first identified, it was also demonstrated that an activated form of Notch (N – the activated form is commonly and hereafter referred to as NACT) also cooperated with scrib−/<sup>−</sup> to induce tumourigenesis (Brumby and Richardson, 2003). While invasiveness of N ACT/scrib−/<sup>−</sup> tumours is driven by JNK signalling, differentiation suppression is not, as blocking JNK did not rescue differentiation as indicated by Elav expression (Leong et al., 2009). However, blocking aPKC and JNK simultaneously was able to completely rescue the differentiation defects in, and the overgrowth and invasion phenotypes of, N ACT/scrib−/<sup>−</sup> tumours (Leong et al., 2009). Interestingly, it was recently found that N ACT can also cooperate with l(2)gl mutation to produce tumours, which also have their invasiveness driven by JNK-induced Mmp1 activity (Paul et al., 2018). Similarly, ectopic JNK signalling also contributes to tumourigenesis driven by Ras85DV<sup>12</sup> expression and l(2)gl mutation – Ras85DV12/l(2)gl−/<sup>−</sup> tumours upregulate JNK signalling, which is thought to proceed via Src oncogene at 42A (Src42A), Ubiquitin-conjugating enzyme variant 1A (Uev1A), and the E2 ubiquitin ligase, Bendless (Ben) (Ma et al., 2013a,b). Uev1A and Ben play a highly conserved role in regulating the DNA damage response in cells via their role in K63-linked polyubiquitination, but whether such activity is required in JNK signalling remains to be determined (Bai et al., 2018). Activation of the Wingless (Wg, orthologue of the human WNT family) signalling pathway by JNK signalling is also thought to drive invasiveness via upregulation of Mmp1 expression and activity in Ras85DV12/l(2)gl−/<sup>−</sup> tumours (Zhang et al., 2019). Indeed, direct activation of JNK signalling together with Ras85DV<sup>12</sup> expression is sufficient for neoplastic tumourigenesis (Brumby et al., 2011).

While there are clearly differences between Ras85DV12/scrib−/<sup>−</sup> and N ACT/scrib−/<sup>−</sup> tumours, research has also uncovered many genetic similarities between them. Microarray data from Ras85DV12/scrib−/<sup>−</sup> and N ACT/scrib−/<sup>−</sup> tumours has identified just over 500 genes that were similarly misregulated between the two tumours, as well as 103 genes that were specifically responsive to JNK signalling shared between them (Doggett et al., 2015). Four of those genes were BTB-zinc finger TFs, and one of those was chronologically inappropriate morphogenesis (chinmo), which was shown to be capable of cooperating with both Ras85DV<sup>12</sup> and N ACT to drive tumourigenesis, even if JNK signalling was blocked (Doggett et al., 2015). A similar role was identified for the BTB-zinc finger TF fruitless, while another BTB-zinc finger TF, abrupt, was able to compensate for chinmo removal in driving tumourigenesis of scrib−/<sup>−</sup> /Ras85DV<sup>12</sup> clones (Doggett et al., 2015). This indicates that these BTB-zinc finger TFs are important transcriptional targets of JNK signalling in cooperative tumourigenesis.

As with JNK in its anti-tumourigenic role, questions exist regarding the source of pro-tumourigenic JNK signalling. Egr is suspected, and haemocytes appear to be attracted to tumourous tissue just as they are attracted to pre-tumourigenic tissue (**Figure 3**), though not necessarily by the same mechanism – Ras85DV12/scrib−/<sup>−</sup> tumours, in their undead-like state, have been shown to co-opt the activity of caspases to generate reactive oxygen species (ROS), which can attract haemocytes, which then activate JNK signalling and caspases in the tumourigenic cells, forming a feedback loop (Pérez et al., 2017). It is also thought that the increased JNK signalling in Ras85DV12/scrib−/<sup>−</sup> tumours may increase haemocyte proliferation via JNK-dependent upregulation of the unpaired (upd1, upd2, and upd3) family of genes (behaviourally similar to mammalian IL-6), which act as ligands for the proliferation-promoting Jak-STAT signalling pathway (Pastor-Pareja et al., 2008; Cordero et al., 2010; Wu et al., 2010; Bunker et al., 2015). Regardless, haemocytes produce Egr when associated with pre-tumourigenic and tumourigenic tissue, but while Egr activates apoptosis promoting TNF-JNK signalling in pre-tumourigenic tissue, it is thought to promote tumour growth in Ras85DV12/scrib−/<sup>−</sup> tumourigenic tissue, as well as invasive capacity via TNF-JNK-mediated Mmp1 upregulation (Cordero et al., 2010). However, there is debate regarding the importance of Egr in JNK activation in polarityimpaired tumourigenesis. It has recently been shown that, in two different JNK-driven tumourigenesis models in the eye-antennal imaginal discs – tumourigenesis induced via polarity-impairment (Ras85DV12/scrib−/−) or via chromosomal instability (CIN) – JNK signalling initiation primarily derives from the tumourous epithelia itself, rather than recruited haemocytes or mesenchymal myoblasts (Muzzopappa et al., 2017). Furthermore, it was found that egr and grnd were dispensable in the process – instead, JNK signalling derived from signalling through the JNKKKs Wnd and Ask1 in polarity-impairment- and CIN-induced tumourigenesis, respectively (**Figure 3**; Muzzopappa et al., 2017). The authors reasoned this was due to CIN-induced tumourigenesis producing ROS, to which Ask1 is sensitive (Sekine et al., 2012), while Wnd mediates JNK signalling in response to polarity-impairment (see also section "Upstream regulation of JNK signalling") (Ma et al., 2016; Muzzopappa et al., 2017).

We have highlighted how anti-tumourigenic JNK signalling is thought to be partly responsible for blocking Yki activation in pre-tumourigenic scrib−/<sup>−</sup> clones (Doggett et al., 2011; Chen et al., 2012). However, it is currently unclear whether co-opting JNK signalling into being pro-tumourigenic in Ras85DV12/scrib−/<sup>−</sup> tumours alters its effect on SWH signalling. In Ras85DV12/scrib−/<sup>−</sup> tumourigenesis, SWH signalling is impaired, and Yki is active and contributes to the observed neoplastic overgrowth, but not invasion (**Figure 3**; Doggett et al., 2011). Conversely, it has been demonstrated that Yki-driven overgrowth in wing and eye-antennal imaginal discs is suppressed via JNK-mediated Wts activity (Enomoto et al., 2015). However, when Ras85DV<sup>12</sup> is coupled with active JNK signalling (via egr overexpression) tumourigenesis occurs similar to that driven by Ras85DV12/scrib−/−, and the combination of Ras-MAPK and JNK signalling leads to Yki activation via the accumulation of F-actin, dependent on the actin regulators Jub, Diaphanous (Dia), and Rac1, as well as inactivation of Wts (Enomoto et al., 2015). These data suggest SWH signalling modulation and Yki upregulation contribute to Ras85DV12/scrib−/<sup>−</sup> tumourigenesis, but more research is needed to understand how JNK signalling interacts with SWH signalling and Yki during the process.

As briefly discussed, tumourigenesis can be modelled in Drosophila by more mechanisms than just oncogene activation coupled with polarity-impairment (as per Ras85DV12/scrib−/−), and many of these alternative mechanisms are also reliant on JNK signalling. However, while activated Ras85D is known to block apoptosis (Bergmann et al., 1998; Kurada and White, 1998) and thus co-opt JNK signalling into promoting tissue growth (**Figure 3**), in other cases of cooperative tumourigenesis it is not clear how the apoptosis promoting role of JNK is halted. Examples of both such JNK-driven tumour types have recently been described elsewhere (reviewed in Richardson and Portela, 2018). We list some newly identified examples of JNK-driven tumourigenesis in various Drosophila tissues below.


signalling (Woodfield et al., 2013). Similar phenomena are observed upon overexpression of Vacuolar H+ ATPase 44kD subunit (Vha44), the C-subunit of V-ATPase, which is involved in the acidification of endosomes, and also leads to JNK-dependent tumourigenesis (Petzoldt et al., 2013). More recently, it has been shown that endocytic nTSG clones generated in eye-antennal imaginal discs also have some degree of polyploidy due to JNK and Yki coactivation, with JNK signalling downregulating the G2-M phase cell cycle regulator Cyclin B (CycB) and Yki upregulating Diap1 to promote polyploidy-inducing endoreplication (Cong et al., 2018). Interestingly, it was shown that polyploid cells also form in Ras85DV12/scrib−/<sup>−</sup> tumours due to CycB downregulation, and blocking their formation inhibits the invasive behaviours of these tumours (Cong et al., 2018).


In summary, the second face of JNK signalling is as a powerful driver of tumourigenesis. Pro-proliferation and survival functionalities are co-opted by apoptosis suppression signals such as Ras-MAPK signalling, and JNKs regulation of cell movement and migration is converted into promoting invasion and metastasis. Modulation of SWH signalling is thought to be involved with the pro-tumourigenic roles of JNK, but more research is needed to fully clarify this signalling cross-talk. JNK signalling in Drosophila is therefore a powerful pro-tumourigenic force, but context is key.

## NON-AUTONOMOUS EFFECTS OF JNK SIGNALLING

Most of our previous discussions regarding the effects of JNK signalling dealt with autonomous induction and action of the pathway, however, non-autonomous effects of JNK signalling have been identified. One of the earliest explorations of nonautonomous JNK signalling effects came from examinations

of cooperative tumours generated via scrib mutation and Raf oncogene (Raf) activation (using a Raf gain-of-function allele (RafGOF)) (Uhlirova et al., 2005). These RafGOF/scrib−/<sup>−</sup> tumours are indistinguishable from Ras85DV12/scrib−/<sup>−</sup> tumours. Upregulating JNK signalling in RafGOF/scrib−/<sup>−</sup> tumours via hepACT expression led to a reduction in the size of eye-antennal imaginal discs, but adult eyes increased in size, and in both cases the GFP-positive cells (where the different transgenes were clonally expressed) were eliminated, suggesting a nonautonomous effect on the growth of the surrounding wildtype tissue (Uhlirova et al., 2005). The researchers suggested that the addition of hepACT overcomes the apoptosis inhibition of RafGOF , and prompts compensatory proliferation from the wildtype cells, but the secretion of cytokines by the RafGOF/scrib−/−/hepACT cells leads to the malformation observed (Uhlirova et al., 2005).

Compensatory proliferation is one of the key mechanisms through which non-autonomous JNK signalling is realised, and it is usually effected via secretable JNK targets that induce proliferation, such as Wg or the Upd family ligands (Uhlirova et al., 2005; Pastor-Pareja et al., 2008; Sun and Irvine, 2011). While a healthy level of compensatory proliferation maintains tissue homeostasis in the face of wounding or cell competition, the process can be corrupted if apoptosis is induced (e.g., by Diap1 mutation or X-ray exposure) but the pathway is blocked by p35 expression – these undead cells upregulate JNK signalling, which promotes the expression and secretion of Wg and Decapentaplegic (Dpp), ligands that then promote the nonautonomous overproliferation and overgrowth of neighbouring cells (**Figure 4A**; Pérez-Garijo et al., 2004; Ryoo et al., 2004). Interestingly, TNF-JNK signalling in undead cells can also, conversely, trigger non-autonomous apoptosis, with cell death being induced in different wing imaginal disc compartments, a process the researchers termed "apoptosis-induced apoptosis" (Pérez-Garijo et al., 2013).

Perhaps unsurprisingly, non-autonomous effects of JNK signalling can be tumourigenic, such as the phenomenon of interclonal cooperation, where scrib−/<sup>−</sup> (or l(2)gl−/−) clones

adjacent to Ras85DV12-expressing clones (Ras85DV12//scrib−/−) cooperate to induce tumourigenesis of the Ras85DV<sup>12</sup> cells – these tumours appear functionally identical to those where scrib is mutated and Ras85DV<sup>12</sup> is expressed in the same cells (Ras85DV12/scrib−/−) (**Figure 4B**; Wu et al., 2010). Researchers found that the interclonal cooperation was due to the secretion of the Upd family ligands, which activate Jak-STAT signalling and are targets of JNK-mediated transcription in scrib−/<sup>−</sup> cells (**Figure 4B**; Wu et al., 2010; Bunker et al., 2015). Indeed, coexpression of upd1, upd2, or upd3 and Ras85DV<sup>12</sup> replicated the interclonal cooperation phenotype, and inhibiting JNK signalling rescued Ras85DV12//scrib−/<sup>−</sup> tumourigenesis, but not Ras85DV12/upd1/2/3 tumourigenesis (**Figure 4B**; Wu et al., 2010). In this example of interclonal cooperation, the scrib−/<sup>−</sup> cells upregulate JNK signalling and, while they are eventually eliminated, the JNK signal non-autonomously allows for the tumourigenic overgrowth of the Ras85DV12-expressing cells (Wu et al., 2010).

Another example of non-autonomous JNK signalling activity is the case of Ras85DV<sup>12</sup> cooperation with mutated genes from the mitochondrial respiratory system, including NADH dehydrogenase (ubiquinone) PDSW subunit, mitochondrial ribosomal protein L4, and Cytochrome c oxidase subunit 5A – their cooperation induces overgrowth in wildtype neighbours, but not in the mutated cells (Ohsawa et al., 2012). ROS production in the Ras85DV12/mitochondrial gene mutant clones (Ras85DV12/mito−/−) promotes ectopic JNK signalling, which contributes to SWH downregulation, Yki upregulation, and the transcription of wg and upd1/upd2/upd3 (Ohsawa et al., 2012). The expression and secretion of Wg and Upd1/Upd2/Upd3 acts on the surrounding wildtype cells to promote their proliferation (**Figure 4C**), but if the surrounding cells overexpress Ras85DV<sup>12</sup> , they develop into neoplastic invasive tumours (Ohsawa et al., 2012). Whether tumourigenic or not, the Upd gene family being a transcriptional target of JNK signalling is a common theme in how non-autonomous JNK signalling is effected – this is also seen in clones mutant for the early endosomal regulatory gene Rab5, where the concomitant disruptions to endocytic processes lead to upregulated TNF-JNK and Ras-MAPK signalling, Yki activation, and Upd ligand expression and secretion to drive overgrowth of surrounding tissue (Takino et al., 2014). Research using the same model system (Ras85DV12/mito−/<sup>−</sup> clones) further dissected how JNK signalling was regulated and acted. It was shown that Ras85DV12/mito−/<sup>−</sup> clones displayed phenotypes associated with cellular senescence – their cell cycle was arrested in G1, they upregulated various senescenceassociated markers, the individual cells were overgrown, and they displayed a senescence-associated secretory phenotype (SASP) (**Figure 4C**; Nakamura et al., 2014). In these cells, ROS production and p53 upregulation contribute to the activation of JNK signalling, which then induces Upd1/Upd2/Upd3 expression and secretion, leading to non-autonomous tissue growth effects (**Figure 4C**; Nakamura et al., 2014). A somewhat related role for JNK signalling was recently observed for cells upon wounding in wing imaginal discs – wounding induced JNK signalling, and cells became transiently "stalled" in the cell cycle at G2 phase (or near-permanently "arrested" in G2 phase after JNK induction via egr overexpression in the rotund expression domain) (Cosolo et al., 2019). Researchers found that JNK signalling induces G2 phase stalling/arrest via downregulating the activity of Stg (an inducer of mitosis) and, furthermore, that cells with this G2 profile were protected from JNK-mediated apoptosis (Cosolo et al., 2019). It was also shown that the G2-biased profile of clones mutant for wts or dlg1 was due to JNK signalling (Cosolo et al., 2019). Interestingly, the wildtype tissue adjacent to these mutant clones overgrows, and was suggested by researchers to be another example of JNK-induced non-autonomous tissue growth (Cosolo et al., 2019). Similar cell cycle arrest has also been observed in wholly scrib−/<sup>−</sup> wing imaginal disc tumours – strong JNK signalling in periphery cells early in tumourigenesis induces G2/M phase arrest, but JNK signalling decreases over time in these cells and, together with a concomitant increase in Ras-MAPK signalling, the cell cycle arrest ceases and the tumours overgrow (Ji et al., 2019).

One final example of JNK signalling-induced nonautonomous effects concerns imaginal disc clones overexpressing Src64B that are eliminated via cell competition, but cause the overgrowth of their wildtype neighbours (Enomoto and Igaki, 2013). Clonal Src64B-overexpression activates JNK signalling, in part via the induction of F-actin accumulation (Enomoto and Igaki, 2013) – further links have been drawn between the actin cytoskeleton and JNK signalling activation, as Src genes have also been shown to promote JNK signalling via Rho1-induced actin remodelling (Rudrapatna et al., 2014). Regardless, the F-actin accumulation also promotes Yki activation in the Src64B-overexpressing clones – though their overgrowth is opposed by the upregulated JNK signal, the Yki signal propagates to the wildtype neighbours in a JNK-dependent manner, and there upregulates Yki and promotes non-autonomous overgrowth (Enomoto and Igaki, 2013). Interestingly, however, if JNK signalling is inhibited in the Src64B-overexpressing clones, Yki activity promotes the tumourigenic overgrowth of those cells instead (Enomoto and Igaki, 2013).

Non-autonomous JNK signalling is necessary for the maintenance of tissue homeostasis, regulating as it does the process of compensatory proliferation after cell competition or wounding events. However, the above examples also show that JNK signalling is two-faced, and can be co-opted to effect non-autonomous tumourigenesis, inducing or enhancing the overgrowth and invasion of otherwise benign cells.

#### UPSTREAM REGULATION OF JNK SIGNALLING

The activation of JNK signalling is complex. While the kinase core remains largely the same, upstream activation contexts can vary wildly. Most of the cases discussed so far (where it has been examined) are thought to have utilised the TNF-JNK signalling pathway in various ways, but clearly this is not the only way JNK can be activated. In this section, we will discuss some of the more unique ways in which JNK

signalling can be activated when acting in a pro- or antitumourigenic fashion.

### Regulating JNK Signalling in Development and Tissue Homeostasis

Besides acting as a pro-apoptotic signal, JNK signalling is arguably best understood as a regulator of cell morphology and migration in a variety of developmental contexts (reviewed in Harden, 2002; Ríos-Barrera and Riesgo-Escovar, 2013). During Drosophila embryogenesis, JNK signalling plays a critical role during the epithelial sheet migration process of dorsal closure, and similarly in thoracic closure during pupariation (**Figure 5**). Activation of JNK signalling during thoracic closure is mediated by Pvr signalling via Crk oncogene (Crk), Ced-12, Mbc, and Rac1 (Ishimaru et al., 2004). As an aside, constitutively activated Pvr signalling is oncogenic, and in wing imaginal discs activates JNK signalling (alongside Ras-MAPK and PI3K signalling) to effect metabolic reprogramming of the tumour cells (Wang et al., 2016). Another process in which both Pvr and JNK signalling are involved is border cell migration (BCM), a process during oogenesis involving the movement of a cluster of "border cells" from the apical end of the egg chamber to the surface of the oocyte itself, and which is an established model of cell migration and invasion in Drosophila (reviewed in Montell et al., 2012). During BCM, JNK signalling regulates clustering and migratory behaviours of the border cells, and is thought to be activated by the GTPases Rho1 and Cdc42 (Mathieu et al., 2007; Llense and Martín-Blanco, 2008; Melani et al., 2008). It is further thought that JNK signalling contributes to the Pvr signalling-mediated guidance of the border cells, but this interaction is not fully understood (Llense and Martín-Blanco, 2008).

Returning to dorsal and thoracic closure, in both processes Rac1 is thought to activate JNK signalling via the JNKKK Slpr (Garlena et al., 2010), whereas in thoracic closure the JNKKK Wnd has been shown to be dispensable (Ma et al., 2015b). JNK signalling via Slpr during dorsal closure is believed to be initiated by the activity of the Src family proteins (and/or Btk family kinase at 29A (Btk29A)) and their downstream targets, SH2 ankyrin repeat kinase (Shark) and Downstream of kinase (Dok), and that Rac1 may act downstream of these molecules (reviewed in Ríos-Barrera and Riesgo-Escovar, 2013) – however, the upstream regulation of JNK signalling in these developmental contexts is not fully understood (**Figure 5**). The role of JNK signalling during dorsal closure has also been shown to rely on additional signalling pathways. Transforming growth factor-β (TGF-β) signalling, activated by Dpp, has been demonstrated as acting to suppress the pro-apoptotic activity of the JNK signalling pathway – while the JNK-induced AP-1 TF complex promotes rpr expression and apoptosis, the TGF-β-induced TF Schnurri (Shn) suppresses rpr expression (Beira et al., 2014). This is an elegant example of how JNK signals can be co-opted during developmental events, like in tumourigenesis, via the mechanism of apoptosis suppression.

It is believed that JNK signalling-mediated cell death during development generally proceeds via input from the TNF signalling pathway. However, research in this vein has largely examined how TNF-JNK signalling proceeds after induction, rather than how endogenous TNF-JNK signalling is regulated or initiated. In the Drosophila eye, JNK-mediated cell death induced by Egr expression acts predominantly via Tak1, though Wnd plays a small role, whereas JNK-mediated cell death induced by Rac1 expression acts predominantly via Wnd, but Tak1 is dispensable (Ma et al., 2015b). Furthermore, cell death (and invasion) within the Drosophila wing imaginal disc can be induced via scrib knockdown in the ptc expression domain – this was also suppressed by wnd knockdown, which appears to act via both Hep and Mkk4 in this context (Ma et al., 2015b). This study demonstrates the complex nature of JNK signalling regulation. Research suggests Rho1, like Rac1, can activate JNK signalling, but does not do so identically. As mentioned, it has been shown that Rho1 is likely responsible for JNK signalling upregulation in response to polarityimpairment in Ras85DV12/scrib−/<sup>−</sup> tumours (Muzzopappa et al., 2017). Furthermore, it has been shown that knockdown of Rho1 (and wnd) suppressed scrib knockdown-induced cell death and invasion in the ptc expression domain in wing imaginal discs (**Figure 5**; Ma et al., 2016). Overexpression of wnd or Rho1 in the ptc expression domain were found to promote epithelial to mesenchymal transition (EMT)-like invasive phenotypes and Mmp1 upregulation (**Figure 5**; Ma et al., 2016). Knocking down wnd while overexpressing grnd (and vice versa) led to a rescue of the invasion phenotypes (though actin remodelling was not dependent on JNK signalling), suggesting some kind of feedback loop may occur (Ma et al., 2016). Furthermore, while Rac1 overexpression in the ptc expression domain also led to JNK-dependent invasive behaviours, it was shown that wnd was dispensable (in contrast to the aforementioned necessity of wnd in Rac1-induced JNK-mediated cell death in the eye (Ma et al., 2015b)) and, curiously, that both Rho1 and Rac1 overexpression-induced phenotypes were rescued upon simultaneous disruption of Tak1 and Slpr (Ma et al., 2016).

Overexpression of wnd has also been shown to cooperate with Ras85DV<sup>12</sup> to generate tumours where, similar to what is seen in Ras85DV12/scrib−/<sup>−</sup> tumours, it induces JNK signalling to promote upregulation of the Wg signalling pathway, promoting cell proliferation (Ma et al., 2016). Various cell morphology regulators have also been identified as interacting with Ras85DV<sup>12</sup> to promote both tumourigenesis and JNK signalling. These include the aforementioned Rho-family GTPases Rho1 and Rac1, as well as their partner Rho guanine nucleotide exchange factor 2 (RhoGEF2), which cooperate with Ras85DV<sup>12</sup> to enhance adult eye overgrowth phenotypes when overexpressed, due to their role as positive regulators of JNK signalling (Brumby et al., 2011). Moreover, clonally, Rac1, RhoGEF2, or activated Rho1 expression in isolation led to clones that were eliminated, but cooperated when expressed alongside Ras85DV<sup>12</sup> to induce the formation of invasive tumours via activation of JNK signalling (Brumby et al., 2011). Specifically, RhoGEF2 and Rho1 act upstream of Rho kinase (Rok) and Spaghetti squash (Sqh, a.k.a Myosin II Regulatory Light Chain) to activate JNK signalling and cooperate with Ras85DV<sup>12</sup> in tumourigenesis, but they also promote actin/myosin contractility

FIGURE 5 | Upstream regulation of JNK signalling. JNK signalling can be activated by a variety of different upstream mechanisms. JNK signalling mediates cell morphology changes, such as those during embryonic dorsal closure, where it is thought to act via Src42A and Src64B (and/or Btk29A) to activate the kinase Shark and its adapter, Dok. This signalling may in turn activate Rac1 and promote JNK signalling via Slpr. In cells with chromosomal instability, accumulated ROS promote Ask1 signalling, stimulating the JNK signalling pathway to promote apoptosis. Arguably, the best understood pathway is the TNF-JNK signalling pathway, which is generally considered to be activated as a response to cellular stresses. Egr, the Drosophila TNF, binds to Grnd or Wgn (TNFRs), which activate Msn (JNKKKK) and Tak1 (JNKKK), via the adapter proteins Traf4, Tab2, and Traf6. JNK signalling is activated in polarity-impaired cells, which is thought to occur via the stimulation of Wnd by the actin cytoskeleton regulator Rho1, although TNF signalling contributes to amplify JNK activity. Lastly, in Drosophila wing imaginal discs, differential JNK signalling regulation by the SWH signalling pathway has been observed. While non-active SWH signalling allows Yki to promote JNK activation via Rho1, activated SWH signalling suppresses Yki activity, preventing bantam transcription, a miRNA that suppresses Rox8, which acts as a positive regulator of JNK signalling, possibly by downregulating dlg1. Msn is thought to be capable of activating Tak1, Wnd, and Slpr, but has not yet been shown to activate Ask1. Tak1, Wnd, and Ask1 are then thought to be capable of activating both the JNKKs, Hep and Mkk4, while Slpr has only yet been shown to act via Hep. Both Hep and Mkk4 can activate the sole Drosophila JNK, Bsk, which positively regulates a number of TFs, including the well known Jra and Kay. These TFs promote transcription of a number of important genes, including the apoptosis promoters hid and rpr, the Jak-STAT ligands upd1, upd2, and upd3, the invasion promoter Mmp1, and the negative JNK regulator puc. Dotted lines represent uncertain interactions. Wavy lines represent known physical interactions between core pathway members and (Continued)

#### FIGURE 5 | Continued

their adapters. Gene and protein name abbreviations used in the diagram are as follows: Src oncogene at 64B (Src64B), Src oncogene at 42A (Src42A), Btk family kinase at 29A (Btk29A), Downstream of kinase (Dok), SH2 ankyrin repeat kinase (Shark), Eiger (Egr), Grindelwald (Grnd), Wengen (Wgn), Misshapen (Msn), TNF-receptor-associated factor 4 (Traf4), TAK1-associated binding protein 2 (Tab2), TNF-receptor-associated factor 6 (Traf6), Slipper (Slpr), Wallenda (Wnd), TGF-β activated kinase 1 (Tak1), Apoptotic signal-regulating kinase 1 (Ask1), Hemipterous (Hep), MAP kinase kinase 4 (Mkk4), Basket (Bsk), Jun-related antigen (Jra), Kayak (Kay), Ets at 21C (Ets21C), Activating transcription factor 3 (Atf3), PAR-domain protein 1 (Pdp1), Inverted repeat binding protein 18 kDa (Irbp18), Ftz transcription factor 1 (Ftz-f1), puckered (puc), Matrix metalloproteinase 1 (Mmp1), wingless (wg), decapentaplegic (dpp), cheerio (cher), reaper (rpr), head involution defective (hid), unpaired 1 (upd1), unpaired 2 (upd2), unpaired 3 (upd3), Yorkie (Yki), discs large 1 (dlg1).

and cell shape changes independently of JNK signalling (Khoo et al., 2013).

### TNF-JNK Signalling

TNF-JNK signalling has already been discussed in earlier sections, but a more detailed exploration is pertinent, as it is one of the best understood activation contexts of the JNK pathway. The TNF ligand of the pathway, Egr, was identified in 2002 as an orthologue of multiple members of the mammalian TNF gene family (Igaki et al., 2002; Moreno et al., 2002). Identification of the first Drosophila TNFR orthologue, Wgn, quickly followed (Kanda et al., 2002). It is thought that TNF-JNK signalling occurs primarily via interactions between activated TNFRs, the JNKKK kinase (JNKKKK) Misshapen (Msn), and the JNKKK Tak1, which are mediated by the adapter proteins TNFreceptor-associated factor 4 (Traf4), TNF-receptor-associated factor 6 (Traf6), and TAK1-associated binding protein 2 (Tab2) (**Figure 5**; reviewed in Igaki and Miura, 2014). The pathway then drives transcription via the canonical kinase core of Hep and Bsk, and TFs including the AP-1 complex members Kay and Jra – these effectors are, as we have discussed, primarily involved in mediating apoptosis as a response to cellular stresses, such as polarity-impairment, but can be co-opted into a role in tumourigenesis.

Recently, another Drosophila TNFR-encoding gene, grnd, was identified, and found to also play a role in the apoptosis of polarity-impaired cells, as well as in cooperative tumourigenesis (**Figure 5**; Andersen et al., 2015). It was found in a screen for genes necessary for neoplastic growth induced by knockdown of the endocytic gene Syx7, where most genes found to rescue tumourigenesis when disrupted were JNK pathway components (Andersen et al., 2015). Overexpression of egr in the adult eye results in cell death, which was rescued upon knockdown of grnd, but not of wgn (Andersen et al., 2015). It is thought that Grnd binds Egr, prevents its diffusion, and hence controls the autonomy of cell death – it was shown that wing imaginal disc clones overexpressing egr were eliminated via apoptosis, and co-expressing RNAi against Tak1 prevented that autonomous cell death, but co-expressing RNAi against grnd prevented autonomous cell death while also promoting non-autonomous cell death (Andersen et al., 2015). Interestingly, proper identification of a classic Drosophila tumour suppressor gene, lethal(2)tumorous imaginal discs, as ALG3, alpha-1,3- mannosyltransferase (Alg3), has shed light on how Grnd might be regulated – Alg3 mutants fail to glycosylate (and thus inactivate) Grnd, enabling persistent TNF-JNK signalling activation via Egr secreted by the fat body, which promotes JNK-mediated tissue overgrowth via SWH signalling inhibition and Yki activation (de Vreede et al., 2018). Furthermore, Grnd is likely to be involved in both polarity-impairment-induced cell competition and cooperative tumourigenesis – clones with scrib knockdown die, while those with knockdown of both scrib and grnd survive, and the invasiveness of Ras85DV12/scrib−/<sup>−</sup> tumours is blocked via grnd knockdown, as is their overexpression of the JNK target, Mmp1 (Andersen et al., 2015).

As we have discussed, there is some debate over how JNK signalling is activated in polarity-impaired cells – some studies implicate Egr as the key effector of the pathway (Igaki et al., 2009; Vidal, 2010), but others have suggested JNK signalling is initiated by direct activation of JNKKKs by molecules involved in cytoskeletal regulation, such as Rho1 (Ma et al., 2016; Muzzopappa et al., 2017). In this vein, one recent study has demonstrated a new method by which Egr can regulate polarity-impaired cell elimination. Knocking down scrib via RNAi in wing imaginal discs, within the spalt major expression domain, leads to large areas of polarity-impaired cells upregulating JNK signalling (Poernbacher and Vincent, 2018). Disrupting egr in those same cells was shown to only partially rescue the ectopic JNK signalling, while disrupting grnd or adenosine receptor (AdoR) completely rescued the ectopic JNK signalling (Poernbacher and Vincent, 2018). AdoR positively regulates JNK signalling when scrib is knocked down in various expression domains, and also in wholly scrib−/<sup>−</sup> imaginal discs (Poernbacher and Vincent, 2018). Researchers found that the activation of JNK signalling was likely due to an increase in extracellular adenosine secreted by the polarity-impaired cells stimulating AdoR activity, which was then also necessary for egr transcript upregulation (Poernbacher and Vincent, 2018).

### Feedback Loops and ROS

An assortment of feedback loops involving JNK signalling have been identified in various contexts, allowing for persistent activation of the pathway. One of the key positive feedback loops identified occurs during apoptosis in response to stress, where death regulator Nedd2-like caspase (Dronc), an initiator caspase, is activated by and activates JNK signalling (Shlevkov and Morata, 2012; Rudrapatna et al., 2013). Other examples of positive feedback loops include when JNK signalling is activated due to signalling from the Src family members (Rudrapatna et al., 2014), and during Ras85DV12/scrib−/<sup>−</sup> tumourigenesis when haemocyte recruitment and ROS production promote JNK signalling (Pérez et al., 2017). Another feedback loop was recently identified that stresses the importance of JNK signalling

activation in tumourigenesis progression, and also illustrates the stepwise process by which cooperative tumourigenesis can occur. Transient initiation of JNK signalling (via irradiation, p53 expression, or activated hep (hepACT) expression) in cells where apoptosis was functional led to activity of the pathway gradually ceasing (Pinal et al., 2018). Conversely, where apoptosis was blocked (via Ras85DV<sup>12</sup> expression or apoptotic pathway component mutation) JNK signalling persisted, leading to overgrowth when Ras85DV<sup>12</sup> was expressed and upregulation of growth-promoting JNK targets such as the Wg and Jak-STAT signalling pathway ligands (Pinal et al., 2018). Researchers identified that the transient JNK signalling-induced ROS production, and the sustained JNK signalling in apoptosis-deficient cells was dependent on ROS production and the ROS-producing gene moladietz, itself a JNK signalling target, and hence a feedback loop was formed (Khan et al., 2017; Pinal et al., 2018). As we have mentioned, ROS production can also occur during CIN-induced tumourigenesis, where it is thought to activate JNK signalling via Ask1 activity (**Figure 5**; Muzzopappa et al., 2017).

#### SWH Signalling

We have discussed how JNK signalling can differentially regulate the SWH signalling pathway – upregulating Yki during compensatory proliferation and tumourigenesis (Doggett et al., 2011; Ohsawa et al., 2011; Sun and Irvine, 2011, 2013; Enomoto et al., 2015), and downregulating Yki during cell competition and tissue growth regulation (Doggett et al., 2011; Chen et al., 2012). However, it has also been recently shown that the opposite is possible – SWH signalling is capable of regulating JNK signalling. Yki/Sd activity was shown to promote tissue overgrowth, as well as Rho1 transcription which, as we have discussed, is capable of then promoting JNK signalling – specifically, researchers found that Rho1 activated JNK signalling via Tak1 and Hep (**Figure 5**; Ma et al., 2015a). Interestingly, this interaction was only observed in the wing, and not eye-antennal, imaginal discs, suggesting it is context dependent (Ma et al., 2015a). Furthermore, Ykimediated overgrowth was rescued by blocking JNK signalling, and was phenocopied via coupling Rho1 and p35 expression (promoting JNK signalling and blocking apoptosis) (Ma et al., 2015a). Conversely, it has also been demonstrated that activating SWH signalling, and hence blocking Yki, can lead to invasive behaviour, which is primarily governed by JNK signalling – indeed, JNK signalling was upregulated and responsible for the invasiveness (Ma et al., 2017). Mechanistically, researchers showed that the Yki target, bantam (a microRNA), suppresses Rox8, which acts as a positive regulator of JNK signalling, possibly via its role in downregulating dlg1 (**Figure 5**; Ma et al., 2017). Therefore, an interesting regulatory situation occurs in the wing epithelium: SWH signalling inactivation promotes tissue growth via Yki-mediated activation of JNK signalling (Ma et al., 2015a), while SWH signalling activation promotes JNK-mediated invasiveness (Ma et al., 2017). Another interesting example of interaction between Yki and JNK signalling occurs during wound healing, a process JNK is well

known to be involved in via its capacity as a regulator of cell movement and morphology (Lesch et al., 2010). Researchers found that Yki was required (alongside Sd) for wound closure in larval epidermal tissue not via its canonical roles in proliferation, but rather by regulating actin polymerisation (Tsai et al., 2017). Furthermore, they found that Yki activity during wound closure occurred independently of Pvr signalling, but likely acts parallel to or downstream of JNK signalling (Tsai et al., 2017).

#### Other Regulators

Lastly, while these previous examples highlight how JNK signalling can be activated due to certain signalling pathways, or as a result of different biological phenotypes, the pathway can also be induced or repressed by simple genetic lesions. As such, these mutations can potentially contribute significantly to tumourigenesis, and so we feel it is important to call attention to some of the more recently identified examples. JNK signalling can be activated by mutations in genes such as jumeau (jumu, a Fork head family TF) (Wang et al., 2019) or pontin (pont, member of the ATPases Associated with various cellular Activities (AAA+) family) (Wang et al., 2018), or by overexpression of genes such as growth arrest and DNA damage-inducible 45 (Gadd45, orthologue of human GADD45G) (Camilleri-Robles et al., 2019). Other negative regulators of JNK signalling include the Striatin interacting phosphatase and kinase (STRIPAK) complex members, Connector of kinase to AP-1 (Cka) and Striatin interacting protein (Strip) – when JNK signalling is activated via Immune Deficiency (IMD) signalling, part of the Drosophila innate immune system (reviewed in Hoffmann, 2003; Kaneko and Silverman, 2005), these molecules act to suppress JNK pathway activity (Bond and Foley, 2009; Ashton-Beaucage et al., 2014). However, it has also been shown that mutating or knocking down Cka or Strip during Drosophila larval spermatogenesis induces JNK signalling via Egr, independently of IMD signalling, suggesting Cka and Strip may act more universally (La Marca et al., 2019). Conversely, JNK signalling can be inhibited by mutations in genes such as deltex (dx, a positive regulator of Notch signalling) (Dutta et al., 2018), Glycine N-acyltransferase (Glyat, orthologue of human GLYATL3) (Ren et al., 2017), and various members of the Toll (a.k.a. NFκB) signalling pathway (Wu et al., 2015). Interestingly, while Toll signalling is necessary for JNK-mediated cell death (Wu et al., 2015), it is repressed in scrib−/<sup>−</sup> cell competition where, if activated, Toll signalling in scrib−/<sup>−</sup> cells then occurs alongside simultaneous activation of JNK signalling and accumulation of F-actin, which promotes Yki activity and tumourigenesis (Katsukawa et al., 2018). Some more unusual examples are genes that promote JNK signalling when overexpressed, but inhibit it when mutated, such as licorne (lic, orthologue of human MAP2K6) (Sun et al., 2019) or tankyrase (Tnks, encodes NAD(+) ADP-ribosyltransferase) (Feng et al., 2018).

In summary, a wide variety of activation contexts exist for JNK signalling, each of which seem to drive markedly different cellular behaviours and outcomes. Direct regulation of the

JNKKKs appears to be a key method for inducing different roles for the pathway, and occurs through mechanisms as diverse as actin cytoskeleton regulation and ROS production. The Rho-family GTPases are, in particular, granted a key role in these non-canonical activations, and the pathway is also capable of activating itself via a number of different feedback loops. Many more JNK-activating signals are doubtless waiting to be discovered.

#### CONCLUSION AND PERSPECTIVES

Jun N-terminal kinase signalling is a complex process. An intricate array of upstream signalling molecules feed into the activation of the single titular JNK in Drosophila, Bsk, which then activates an equally vast and detailed collection of downstream TFs and target genes. However, this highly conserved pathway is significantly more complex in mammals and, hence, the relative simplicity of JNK signalling in Drosophila, coupled with the powerful genetic techniques available to researchers using this model organism, means Drosophila has been (and will doubtless continue to be) an indispensable tool in uncovering the molecular basis for the twofaced nature of JNK: being both pro- and anti-tumourigenic. There has been a massive increase in our understanding of these processes over the last decade, which we have attempted to capture in this review. This final section will look at some of the future directions the field may take, as informed by a number of recent and unique explorations of JNK signalling.

While the upstream complexity of JNK signalling contributes to the power of the pathway to drive such varied outcomes, it is also an obstacle to obtaining a comprehensive understanding of its biological contributions. Further complications arise due to the possibility that components of the pathway are effecting roles unrelated to the central JNK signalling cascade. One such example has been shown for Egr – in tumourigenic tissue wholly mutant for dlg1, Egr secreted by attracted haemocytes sensitises the cells of the tumour to activity of the antimicrobial peptide Defensin, which promotes tumour cell death (Parvy et al., 2019). Though a role for JNK signalling in the process was not explicitly ruled out, the possibility of more JNK-unrelated roles for JNK pathway members lends itself to potential new avenues of research.

Alongside the various upstream actuators of JNK signalling, there is considerable potential to explore how the intensity of the JNK signal affects its pro- and anti-tumourigenic properties. As discussed, one recent study found that Cno overexpression led to JNK-mediated tissue overgrowth, but slightly modulating JNK signalling then led to massive overgrowth (Ma et al., 2019). The researchers determined that Cno overexpression upregulated both JNK and Ras-MAPK signalling, each of which downregulated SWH signalling to promote Yki-dependent tissue growth, and suggested that subsequent modulation of JNK signalling inhibited its antitumourigenic role in promoting apoptosis, while leaving its pro-tumourigenic role as a SWH pathway inhibitor intact

(Ma et al., 2019). In a similar vein another recent study examined JNK signalling levels in response to tissue damage. It was shown that high levels of ROS produced in damaged tissue phosphorylated and activated Ask1 and, therefore, strong JNK signalling and apoptosis (Santabárbara-Ruiz et al., 2019). However, it was found that the ROS signal propagated to undamaged neighbouring tissue, and Ask1 was then phosphorylated by both the ROS signal and Akt1, a downstream protein kinase of the PI3K signalling pathway (Santabárbara-Ruiz et al., 2019). This altered activation context for Ask1 led to a lower level of JNK signal, which promoted cell proliferation and survival rather than apoptosis (Santabárbara-Ruiz et al., 2019). Both these studies are examples of the potential for altering JNK signalling levels to profoundly alter the outcome of the pathway, and are exemplars of an exciting new avenue of research regarding pro- and anti-tumourigenic JNK signalling.

The importance of subcellular localisation of JNK signalling components is also an area that is ripe for exploration. A critical role for endocytosis in JNK signalling has been demonstrated, with increased endocytosis thought to be key in upregulating the TNF-JNK signalling observed in scrib−/<sup>−</sup> clones (Igaki et al., 2009). Furthermore, another study has shown Rho1 specifically localised to the cell membrane can activate apoptosis-inducing JNK signalling that acts primarily via Slpr and Tak1 and, indeed, regulates the subcellular localisation of Slpr via physical interaction (Neisch et al., 2010).

The relationship between JNK and SWH signalling is, as yet, unresolved. There is clearly complex interplay between the pathways, highly dependent on cellular context, but its elucidation is critical for obtaining a more complete understanding of pro- and anti-tumourigenic JNK signalling. While recent research discussed in this review has undoubtedly advanced our knowledge greatly (Doggett et al., 2011; Sun and Irvine, 2011; Chen et al., 2012; Enomoto et al., 2015; Ma et al., 2015a, 2017; Gerlach et al., 2018), there is still much to be discovered, particularly with regard to how the pathways interact during cooperative tumourigenesis. This relationship between JNK and SWH signalling is undoubtedly one that will be examined closely in coming years.

There is a preponderance of different model systems used in the study of JNK signalling within Drosophila (e.g., different tissues and tissue regions, different mutant clones, different models of cooperative tumourigenesis). While this is certainly a great strength of the organism as a model, it is also a complicating factor, making it difficult to generate a cogent, unified model for JNK signalling (if this indeed exists). While the exploration of JNK signalling in a wide variety of contexts is undoubtedly beneficial, and should be encouraged and highlighted, we believe that a discussion regarding standardised systems in which observations regarding JNK signalling might be examined and replicated is worthwhile.

Finally, it is worth reiterating the relevance of Drosophila studies to explorations of mammalian genetics and disease. We have touched upon how JNK signalling has two faces in humans, as it does in flies – it can be both

pro- and anti-tumourigenic (or pro- and anti-survival), depending on the context (reviewed in Wagner and Nebreda, 2009; Wu et al., 2019). The inherently increased complexity of human JNK signalling makes the pathway difficult to work with, however, and this leaves Drosophila research at an advantage, able to use parallels between the highly conserved systems to position itself as a foundational body of knowledge for the field – conclusions drawn in flies can be adapted and extended into mammalian research. For example, we have discussed how Rho-family GTPases have emerged as key activators of JNK signalling in Drosophila (Garlena et al., 2010; Brumby et al., 2011; Khoo et al., 2013; Rudrapatna et al., 2014; Ma et al., 2016), and similar conclusions have been reached in mammals – RAC1 gain-offunction mutations are common in BRAF and NRAS-driven melanomas, and said mutations can drive RAC1 binding to and activation of targets like PAK1 (p21 (RAC1) activated kinase 1) and MAP3K11 (mitogen-activated protein kinase kinase kinase 11, a.k.a. MLK3) (Krauthammer et al., 2012). Both PAK1 (Qing et al., 2012) and MAP3K11 (reviewed in Gallo and Johnson, 2002) have been shown to be activators of JNK signalling in mammals. Interestingly, while the Drosophila PAK1 orthologue, p21-activated kinase (Pak), is not known to act via JNK signalling (Conder et al., 2004), the orthologue of MAP3K11 is Slpr, thought to be the key JNKKK in cell morphological changes and embryogenesis processes like dorsal closure (reviewed in Ríos-Barrera and Riesgo-Escovar, 2013). Another study examining BRAF-driven melanoma found that ectopic JNK signalling via JUN was likely partly responsible for the invasiveness of the tumour cells (Ramsdale et al., 2015). As discussed, JUN is one half of the AP-1 TF complex, alongside FOS, which is conserved in Drosophila. Furthermore, pharmacologically inhibiting both BRAF and JNK signalling proved an effective way of suppressing invasiveness and increasing tumour cell death (Ramsdale et al., 2015). This example is clearly reminiscent of the invasive tumourigenesis seen when overactive Ras-MAPK and JNK signalling are coupled in Drosophila, as discussed throughout this review.

While a direct connection has not yet been made between JNK signalling research in Drosophila and the development of human cancer therapeutics, the JNK signalling pathway is clearly two-faced during tumourigenesis in both flies and humans (**Supplementary Table 2**), and therefore the extensive research generated in the fly model may serve to inform which types of human cancer types are likely to be JNKdependent, in order to triage these patients for treatment with JNK pathway inhibitors. For example, genetic analyses of the Drosophila Ret oncogene (Ret) protein tyrosine kinase pathway revealed that the JNK pathway was anti-tumourigenic, which was paralleled in cases of the human Ret-driven cancer, multiple endocrine neoplasia type 2 (Read et al., 2005), thereby implying that targeting the JNK pathway in these cancers would not be beneficial. There has been considerable interest in developing drugs that target JNK signalling for use in the treatment of cancers (and other diseases), and some success has been obtained, but the pathway is so context dependent, with such a diverse array of targets and effectors, that determining how best to exploit the pathway pharmacologically can be difficult (reviewed in Messoussi et al., 2014; Cicenas, 2015; Cicenas et al., 2017). Nonetheless, research into Drosophila JNK signalling has, does, and will continue to support our understanding of JNK signalling in humans but, as it relates to cancer progression, more research is clearly needed. With the advent of transgenic RNAi (reviewed in Perrimon et al., 2010) and, more recently, CRISPR/Cas9 technology in Drosophila (reviewed in Housden and Perrimon, 2016; Bier et al., 2018), the generation of more sophisticated models of human cancer in flies has been possible, which can be used for chemical screening for anti-cancer compounds (Richardson et al., 2015; Sonoshita and Cagan, 2017; Cagan et al., 2019), a process already achieving promising results (Bangi et al., 2016, 2019). Similar methodologies using Drosophila as avatars for specific human cancers could be applied specifically to investigate whether the JNK pathway is pro- or anti-tumourigenic in particular contexts, and how it might be targeted to be beneficial therapeutically. Moreover, Drosophila JNK-dependent cancer models can be used in the development of JNK pathway inhibitors that are more bioavailable, and which have greater efficacy and lower toxicity, as has been achieved using a Drosophila model of Ret-driven cancer (Dar et al., 2012; Sonoshita et al., 2018; Ung et al., 2019). With the adoption of these new technologies, the Drosophila model still has much to offer toward the understanding and targeting of the JNK pathway in human cancers.

## AUTHOR CONTRIBUTIONS

JL wrote the manuscript and prepared the figures. HR provided intellectual direction and editorial advice.

## FUNDING

JL is supported by Australian Research Council (DP170102549) awarded to HR, who is supported by funding from the School of Molecular Sciences, La Trobe University.

## ACKNOWLEDGMENTS

Apologies to all authors whose work we were unable to include in this review. Thanks to Sarah T. Diepstraten for assistance with figure preparation, as well as manuscript proofreading and suggestions. Thanks also to Natasha Fahey-Lozano and Marta Portela for manuscript proofreading and suggestions.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2020.00042/ full#supplementary-material

## REFERENCES




tumor invasion. Proc. Natl. Acad. Sci. U.S.A. 104, 2721–2726. doi: 10.1073/pnas. 0611666104


**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 La Marca and Richardson. 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.

# Molecular Mechanisms of Radiation-Induced Cancer Cell Death: A Primer

#### Joseph Sia1,2,3, Radoslaw Szmyd4,6, Eric Hau5,6 and Harriet E. Gee5,6 \*

<sup>1</sup> Department of Radiation Oncology, Peter MacCallum Cancer Centre, Melbourne, VIC, Australia, <sup>2</sup> Division of Cancer Research, Peter MacCallum Cancer Centre, Melbourne, VIC, Australia, <sup>3</sup> Sir Peter MacCallum Department of Oncology, The University of Melbourne, Melbourne, VIC, Australia, <sup>4</sup> Children's Medical Research Institute, Sydney, NSW, Australia, <sup>5</sup> The University of Sydney, Sydney, NSW, Australia, <sup>6</sup> Sydney West Radiation Oncology Network, Sydney, NSW, Australia

Radiation therapy (RT) is responsible for at least 40% of cancer cures, however treatment resistance remains a clinical problem. There have been recent advances in understanding the molecular mechanisms of radiation-induced cell death. The type of cell death after radiation depends on a number of factors including cell type, radiation dose and quality, oxygen tension, TP53 status, DNA repair capacity, cell cycle phase at time of radiation exposure, and the microenvironment. Mitotic catastrophe (a pathway preceding cell death that happens in mitosis or as a consequence of aberrant mitotic progression) is the primary context of radiation-induced cell death in solid cancers, although in a small subset of cancers such as haematopoietic malignancies, radiation results in immediate interphase apoptosis, occurring within hours after exposure. There is intense therapeutic interest in using stereotactic ablative body radiotherapy (SABR), a precise, high-dose form of RT given in a small number of fractions, to prime the immune system for cancer cell killing, but the optimal radiation dose and fractionation remain unclear. Additionally, promising novel radiosensitisers targeting the cell cycle and DNA repair pathways are being trialled. In the context of the increasing use of SABR and such novel agents in the clinic, we provide an updated primer on the major types of radiation-induced cell death, focussing on their molecular mechanisms, factors affecting their initiation, and their implications on immunogenicity.

Keywords: radiotherapy, radiation therapy, stereotactic ablative radiotherapy, cell death, immunogenic cell death, abscopal effect

## INTRODUCTION

Radiation therapy (RT) is a major cancer treatment modality and is responsible for at least 40% of cancer cures (Ringborg et al., 2003), yet treatment resistance remains a clinical problem. A primary reason for this is the capacity for cancer cells to evade radiation-induced cell death. Treatment paradigms have traditionally viewed cancer as a cell-autonomous problem of dysregulated proliferation while side-lining host-tumour interactions, but this dogma has undergone a remarkable revolution in the last few decades, with increasing appreciation of the tumour stroma and immune milieu in shaping tumour evolution. Trials of novel radiosensitisers targeting not only key cell death pathways but also the stromal and immune microenvironment, particularly together with stereotactic ablative body radiotherapy (SABR), have gained intense interest.

Here, we provide an updated primer on the major types of radiation-induced cell death, focussing on their molecular mechanisms, factors affecting their initiation, and their implications

#### Edited by:

Andrew Burgess, Anzac Research Institute, Australia

#### Reviewed by:

Brian Gabrielli, The University of Queensland, Australia Suneil Jain, Queen's University Belfast, United Kingdom

#### \*Correspondence:

Harriet E. Gee harriet.gee@sydney.edu.au; Harriet.Gee@health.nsw.gov.au

#### Specialty section:

This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 14 November 2019 Accepted: 17 January 2020 Published: 13 February 2020

#### Citation:

Sia J, Szmyd R, Hau E and Gee HE (2020) Molecular Mechanisms of Radiation-Induced Cancer Cell Death: A Primer. Front. Cell Dev. Biol. 8:41. doi: 10.3389/fcell.2020.00041

on immunogenicity. We conclude with discussing these aspects in the context of the increasing use of SABR and novel agents in the clinic. We point readers interested in further detail to the excellent referenced reviews.

## THE CELL CYCLE AND TYPES OF RADIATION-INDUCED CELL DEATH

The cell cycle is a highly regulated process occurring in two major phases: interphase (consisting of the G1, S, and G2 phases) and mitosis (cell division). During interphase, the cell grows its organelle counts (G1 phase), copies its DNA (S phase), and reorganises contents in preparation for division (G2 phase). Radiation-induced DNA damage is sensed by the kinases ataxia-telangiectasia mutated (ATM) and ataxia-telangiectasia and Rad3-related protein (ATR), which activate downstream proteins to initiate the DNA damage response (reviewed in Maier et al., 2016). This places the cell into cell cycle arrest, during which DNA double-strand breaks (DSBs) are repaired mainly by two pathways: non-homologous end joining (NHEJ) and homologous recombination (HR). NHEJ is error-prone but active throughout the cell cycle, while HR is error-free but requires an undamaged sister chromatid as the repair template, and therefore is only available in the late S and G2 phases. If damage is too significant for repair, cell death will occur via one of the types described below, at varying time points after the initial irradiation event (interphase or mitotic). In classical radiobiology, cell death is defined by loss of replicative capacity (i.e., replicative or reproductive death) and is determined by clonogenic assays. While this has undoubtedly served the field tremendously, it ignores the types and effects of cell death, discussed in this review.

#### Mitotic Catastrophe and Mitotic Death

Mitotic catastrophe is a mechanism for the control of cells unable to complete mitosis, by the triggering of mitotic arrest and ultimately regulated cell death (RCD) or senescence. Mitotic death refers to RCD (usually intrinsic apoptosis) that is driven by mitotic catastrophe (Galluzzi et al., 2018). Because cells in mitotic catastrophe are almost always incapable of further replication, mitotic catastrophe is often confused as a bona fide cell death type. However, these cells eventually trigger one of the 'other' cell death pathways, or less commonly escape from this state (Gascoigne and Taylor, 2008; Vakifahmetoglu et al., 2008). Dysfunctional cell cycle checkpoints, a common hallmark of cancer cells, allow radiation-damaged cells to enter mitosis prematurely with misrepaired DNA, leading to mitotic catastrophe. The exact mechanisms for the initiation of mitotic catastrophe and the deciding of final cell fate are unclear. Several attempted divisions can occur before sufficient genetic damage is accumulated to trigger mitotic death, underlining why solid tumours often demonstrate delayed responses to RT.

#### Apoptosis

Apoptosis is a highly regulated form of cell death with characteristic morphological and molecular features. The intrinsic (mitochondrial) apoptotic pathway is activated when the DNA damage repair machinery, with p53 as a central player, disrupts the balance between pro- and anti-apoptotic factors, resulting in the release of cytochrome c from the mitochondria into the cytoplasm to activate the intrinsic pathway-specific caspase 9 (Eriksson and Stigbrand, 2010). The extrinsic pathway, in contrast, is initiated by external signalling via binding of the tumour necrosis factor (TNF) family of ligands to plasma membrane death receptors, eventually causing downstream activation of the extrinsic pathway-specific caspase 8. Irradiated cells can upregulate death receptors, making them susceptible to death through this pathway (Wu et al., 1997; Chakraborty et al., 2003; Sheard et al., 2003). The third pathway, the ceramide pathway, is triggered by radiation-induced activation of acid sphingomyelinase in the plasma membrane, producing ceramide via hydrolysis of sphingomyelin. Radiation-induced DNA damage can also activate mitochondrial ceramide synthase for de novo synthesis of ceramide. Ceramide acts as a second messenger in initiating the apoptotic programme, but its signalling targets are complex (Pettus et al., 2002; Kolesnick and Fuks, 2003). The intrinsic, extrinsic and ceramide pathways converge in the activation of caspase 3 and 7, which sets off a cascade of controlled degradation of cellular components.

#### Necrosis

Necrosis is an unregulated, chaotic form of cell death, triggered by unfavourable conditions such as extreme changes in pH, energy loss, and ion imbalance within the cell and its microenvironment after irradiation (Kroemer et al., 2009). Traditionally identified in histological specimens, it is a diagnosis of exclusion and based on morphology, characterised by a gain in cellular volume, plasma membrane rupture and spill of intracellular contents.

#### Senescence

Senescence refers to permanent cell cycle arrest. Radiationinduced senescence is triggered by DNA damage and the induction of the p53 and pRb pathways causing cell cycle block, but other factors including oxidative stress may also be relevant triggers in the irradiation context (Sabin and Anderson, 2011; Li et al., 2018). Importantly, senescent cells while "dead" in terms of clonogenicity continue to be viable and metabolically active, over time developing a specific expression pattern of immunomodulatory factors. This senescence-associated secretory phenotype is a result of the extrusion of cytoplasmic chromatin fragments in senescent cells, which act on the DNA sensing and effector cGAS-STING pathway to drive expression of interferon and NF-kB elements (Dou et al., 2017; Gluck et al., 2017).

#### Autophagy

Autophagy describes the process of sequestrating damaged or old cytoplasmic organelles within vesicles for lysosomal degradation in response to cellular stress. It is usually a cytoprotective process, but autophagic cell death occurs when the autophagic response is excessive (Kroemer and Levine, 2008). The autophagic machinery is a highly regulated pathway involving the ATG genes and is known to be triggered by irradiation, likely via the endoplasmic stress module and mTOR pathway, although the exact mechanisms are unclear (Tam et al., 2017). Autophagy observed after irradiation may be a mechanism of treatment resistance or of cell death, being likely context-dependent (Levy et al., 2017; Dikic and Elazar, 2018).

## Other Types of Cell Death

fcell-08-00041 February 11, 2020 Time: 18:13 # 3

More recently, forms of RCD have been identified in response to irradiation, including necroptosis and ferroptosis, although the extent to which these occur in vivo are uncertain (Gong et al., 2019; Lang et al., 2019; Lei et al., 2019). Necroptosis has similar morphological appearances to necrosis, but is regulated. It can be activated by death receptors, but signalling is transduced via caspase-independent pathways converging on RIPK3 and MLKL. Ferroptosis is a type of cell death triggered by the accumulation of lipid peroxides as the lethal event due to decreased degradation by glutathione peroxidase (GPX4). With irradiation, this is shown to happen via ATM, which represses the cystene-glutamate antiporter system xc-, resulting in decreased glutathione, a co-factor for GPX4 (Lang et al., 2019).

## FACTORS AFFECTING TYPE OF CELL DEATH

#### Cell-Intrinsic Factors

The type of cell death after irradiation depends on a number of factors, with many remaining uncertainties. Cell type is a key determinant. In a subset of tumours such as haematopoietic malignancies (Radford et al., 1994; Jonathan et al., 1999), radiation results in immediate interphase apoptosis. In the majority of solid tumours however, mitotic catastrophe is the most frequent context of radiation-induced cell death (Eriksson and Stigbrand, 2010), whereas normal tissues commonly undergo senescence after irradiation (Nguyen et al., 2018).

The importance of cell type could be related to the cellular status and function of p53 and ATM. p53 is a key regulator of apoptosis and senescence. Intact p53 is required for interphase apoptosis (Fridman and Lowe, 2003), while disruptive TP53 mutations are associated with increased radioresistance via the inhibition of senescence (Skinner et al., 2012). As most tumour cells have lost normal p53 function or inactivated its downstream pathways, the G1/S checkpoint is impaired and DNA damage repair is consequently dependent on ATR/Chk1-mediated intra-S and G2/M arrest (Soussi and Beroud, 2001; Brady et al., 2011; Dillon et al., 2014). Thus, irradiated cells with inactivated p53 will enter mitosis with unrepaired DNA damage, resulting in mitotic catastrophe rather than immediate apoptosis or senescence (Ianzini et al., 2006). While manipulation of p21 or p53 levels may dramatically alter the kinetics of cell death after irradiation, it may not always correlate with the overall loss of replicative capacity (Wouters et al., 1997; Brown and Wouters, 1999). Similarly, ATM impacts cell death by regulating apoptosis, autophagy, and possibly necroptosis (Liang et al., 2013; Tripathi et al., 2013; Chen et al., 2015) (reviewed in Wu et al., 2017), but it is not clear yet how ATM may affect the decision between these types of cell death.

The phase of the cell cycle in which irradiation occurs also influences timing and type of cell death. Studies have demonstrated cells irradiated in G1 phase divide more times and survive longer before undergoing apoptosis compared to cells irradiated in G2 phase, while cells irradiated in mid-to-late S phase died without undergoing mitosis. Post-mitotic apoptosis however appeared to be stochastic and variable between cell lines (Forrester et al., 1999; Endlich et al., 2000).

### Radiation Factors

It may be tempting to speculate that ablative doses incline cells to a different type of cell death compared to conventional doses per fraction, but whether this occurs as a direct effect of irradiation is uncertain. The complexities of establishing this link surround dissecting the effect of dose per fraction from biological effective dose (BED), and accounting for temporal confounding by repopulation and cell death dynamics. Worth noting here is that the induction of sufficient damage to non-nuclear cellular components to evoke upfront necrosis may require clinicallyirrelevant high doses (Warters et al., 1978). Nonetheless, the rate of apoptotic events after irradiation can be proportional to the number of fractions given, in line with the reassortment principle (cells cycling inter-fraction into states predisposed for interphase apoptosis) (Verheij and Bartelink, 2000).

Radiation quality may also affect mode of cell death. High linear energy transfer (LET) radiation such as alpha particles has been shown to result in enhanced chromosome rearrangements and reproductive death (Franken et al., 2011), due to both the complexity and absolute number of DNA damage clusters (Pinto et al., 2005; Obe et al., 2010).

#### Microenvironment Factors

The above factors are strongly influenced by the cellular microenvironment. In brief, oxygen tension modulates cell death after irradiation, with a reduction in chromosomal aberrations and reproductive capacity of cells irradiated under hypoxic conditions. This is presumed due to kinetic competition between the oxygen 'fixation' of DNA damage and chemical repair processes (reviewed in Stewart et al., 2011). Under the hypoxic, nutrient-deprived conditions in vivo, cell death will be influenced both by the chemical properties of oxygen fixation and indirect effects of hypoxia on cellular processes including cell metabolism, DNA repair, and hypoxia inducible factor (HIF)-related survival mechanisms (Harris, 2002). An acidic microenvironment and serum deprivation also suppress the progression of cells to apoptosis and the formation of micronuclei after irradiation (Paglin et al., 1997; Park et al., 2000).

## RADIATION-INDUCED IMMUNOGENIC CELL DEATH

The observation that an intact host immune system improves the tumour control probability of RT, first made in 1979 (Stone et al., 1979), suggests that the anti-tumour effects of RT must extend beyond its direct effects on cancer cells. Over the last decade, the concept of immunogenic cell death (ICD) has emerged to redefine cancer cell death from an immune-functional aspect, whereby the dying cell through peri-mortem processes is able to evoke anti-tumour adaptive immune responses (Ma et al., 2010; Kepp et al., 2014).

## Defining ICD

fcell-08-00041 February 11, 2020 Time: 18:13 # 4

Historically, necrosis has been traditionally thought to be immunogenic (Scaffidi et al., 2002; Sancho et al., 2009), while apoptosis has been viewed as immunologically silent or even tolerogenic (Voll et al., 1997; Krysko and Vandenabeele, 2010). This relationship is an oversimplification. Apoptosis (Casares et al., 2005; Obeid et al., 2007b), necrosis (Scaffidi et al., 2002; Sancho et al., 2009), and autophagy (Michaud et al., 2011; Ko et al., 2014), have all been demonstrated to be capable of inducing immunogenic responses, while necrosis can in fact contribute to a chronic inflammatory microenvironment that is protumourigenic and immunosuppressive (Vakkila and Lotze, 2004).

The current framework for understanding ICD disengages from the traditional forms of cell death and requires two components: the exposure of danger signals, termed damageassociated molecular patterns (DAMPs), from the dying cell to alert antigen presenting cells (APCs), which include dendritic cells (DCs); and an antigenic epitope to be cross-presented by APCs for training of T cells (Galluzzi et al., 2017).

## Observations That RT Can Induce ICD

Bona fide ICD induced by RT has been directly demonstrated with in vivo vaccination assays in mouse models (Apetoh et al., 2007a; Obeid et al., 2007a; Gorin et al., 2014). A strong body of work now corroborate these landmark findings across different murine and human cell lines using surrogate measures (Deng et al., 2014; Gameiro et al., 2014; Golden et al., 2014; Ko et al., 2014; Lim et al., 2014). Radiation-induced ICD can also be implied from a growing number of pre-clinical and clinical studies combining RT with immunotherapy showing improved systemic tumour responses (reviewed in Deloch et al., 2016; Grassberger et al., 2019). Although the link is indirect, the release of DAMPs and antigenic determinants to evoke cancer-specific immunity in these studies are thought to be at least partially mediated through radiation-induced ICD. Other important trials testing RT for this role have yielded negative results for yetuncertain reasons (Kwon et al., 2014; Voorwerk et al., 2019), suggesting that while the phenomenon of radiation-induced ICD is real, its desired clinical impact is a highly selective event.

## Mechanisms of Radiation-Induced ICD

Our understanding of the molecular mechanisms of radiationinduced ICD and the determinants for their efficient triggering are in its nascency. Calreticulin, HMGB1, and ATP, the classically described DAMPs, act on CD91, TLR4, and purinergic receptors on DCs respectively, to attract, activate, and promote phagocytosis of cellular corpses by DCs (Ma et al., 2010). Calreticulin, an endoplasmic reticulum chaperone, is translocated to the cell surface as part of the unfolded protein response in stressed cells as a pre-apoptotic event (Obeid et al., 2007b). HMGB1 is a nuclear chromatin-binding protein that is released on necrosis (Scaffidi et al., 2002; Apetoh et al., 2007b). Unlike HMGB1 where passive release can occur, optimal secretion of ATP in radiation-induced ICD is active and dependent on intact pre-mortem autophagic machinery for the accumulation of ATP within autolysosomes (Michaud et al., 2011; Ko et al., 2014). A host of other DAMPs have since been identified, including cytosolic double stranded DNA (dsDNA), the heat shock proteins 70 kDa (HSP70) and 90 kDa (HSP90), F-actin, and interleukin-33 (reviewed in Krysko et al., 2012; Galluzzi et al., 2017).

Cytosolic dsDNA and the cGAS-STING axis have recently emerged as subjects of intense therapeutic interest. dsDNA is released from either the mitochondria and/or nucleus of irradiated cells into the cytoplasm and binds to its sensor cGAS, producing the cyclic dinucleotide cGAMP. Micronuclei that accumulate when irradiated cells progress through mitosis with radiation-induced DSBs also constitute a source of dsDNA (Harding et al., 2017). cGAMP ligates with the adaptor protein STING to ultimately upregulate the expression of type I interferon, an important immunomodulatory cytokine (Burdette and Vance, 2013; Paludan and Bowie, 2013). Importantly, dsDNA does not need to activate STING in a cell-intrinsic fashion, but dsDNA or cGAMP can be released upon ICD and taken up by myeloid cells, including DCs, to activate STING within those cells (Deng et al., 2014; Diamond et al., 2018; Schadt et al., 2019). This axis can be shut down by ablative radiation doses due to the simultaneous induction of the exonuclease TREX1, which degrades cytosolic dsDNA (Vanpouille-Box et al., 2017). While this mechanism explains the negative impact of ablative radiation doses on abscopal responses in a murine mammary carcinoma model (Dewan et al., 2009), the observation is not consistently held up in non-abscopal readouts of tumour immunity, suggesting that other factors may be involved in this regulation (Burnette et al., 2011; Deng et al., 2014; Schadt et al., 2019). On the other hand, generation of DAMPs such as ATP and HMGB1 appears to be proportional to radiation doses of up to 100 Gy in vitro (Gameiro et al., 2014), but whether this translates to increased immunogenicity in vivo is unknown.

On the antigenicity side, RT increases the processing and MHC-I-restricted presentation of surface antigens in a cellautonomous manner, at least partially via the mTOR pathway (Santin et al., 1997; Verbrugge et al., 2014). More provocatively, as a result of altered transcriptional activity and antigen processing, irradiated tumour cells in vitro demonstrate a diversification of epitope repertoire presented on MHC-I (Reits et al., 2006). A recent clinical trial combining RT with immune checkpoint blockade reported a patient case with expansion of T cell clones against a radiation-induced neoantigen, serving as the first in-human proof of this concept (Formenti et al., 2018).

## DISCUSSION AND CLINICAL RELEVANCE

Increasingly, our appreciation of the anti-tumour effects of RT extends beyond a DNA and cancer cell-centric perspective

FIGURE 1 | The anti-tumour activity of radiation therapy (RT) is multi-faceted. Tumour irradiation induces both direct and indirect effects in controlling the tumour. Direct effects are the result of significant radiation-induced DNA damage in cancer cells leading to their death, which may occur at various time points after the initial irradiation event. Radiation therapy (RT) also exerts anti-tumour activity via indirect effects, which include, but are not limited to, injury of tumour vasculature and priming of host anti-tumour immune responses. A range of factors influence the triggering and magnitude of these effects, broadly falling into cell-intrinsic, radiation, and microenvironment factors. A deeper knowledge of the underpinning mechanisms and their interplay will reveal opportunities for enhancing the overall anti-tumour activity of RT. ATM, ataxia-telangiectasia mutated. DAMP, danger associated-molecular pattern.

(**Figure 1**). The need to better understand the molecular pathways of radiation-induced cell death is even more important in light of recent technological advances in the safe delivery of RT and the increasing use of novel agents in clinic. SABR is progressively adopted into routine clinical practice with a growing number of studies backing its remarkable efficacy, greater than predicted by a simple extrapolation from lower doses per fraction. For example, it has been shown to double median survival for oligometastatic disease (Palma et al., 2019) and more than double the response rate to immunotherapy (Formenti et al., 2018; Theelen et al., 2019), leading to the hypothesis that the biology of tumour response to irradiation is different when given high dose per fraction. Theoretically, SABR would induce DNA damage that was more difficult to repair and decrease tumour cell repopulation due to reduced overall treatment time, albeit with a cost from decreased inter-fraction cellular reassortment and reoxygenation. There is currently insufficient data to confidently propose a model by which SABR achieves such high efficacy (Moding et al., 2016; Song et al., 2019). Indeed, some authors argue that the improved local control seen with SABR purely results from a higher dose (Brown et al., 2013, 2014).

Others posit that SABR has additional indirect effects on cancer cells. Significant tumour vascular injury has been shown to result from SABR, leading to hypoxic cancer cell death

several days after irradiation (Song et al., 2019). Interestingly, the ceramide apoptotic pathway in the tumour vascular endothelium, activated with ablative radiation doses, is thought to be especially important (Garcia-Barros et al., 2003; Rao et al., 2014). These results, however, have been directly challenged in other models (Moding et al., 2015), while others contend that hypoxia induced by SABR in fact inhibits its efficacy (Carlson et al., 2011).

The other reason measurements of radiation-induced cancer cell death in vitro fail to represent the extent of clinically observed responses is due to host immune responses (Golden et al., 2012; Haikerwal et al., 2015). For example, inhibition of the pro-survival autophagy machinery radiosensitises cell kill in vitro, but in fact impairs radiation tumour control in immunocompetent mice due to loss of autophagydependent ATP release as an immune adjuvant (Ko et al., 2014). As such, a pressing question is the impact of radiation dose-fractionation on cell death pathways, especially those related to ICD. Studying fractionated regimens in the preclinical setting is complicated because of the physiologically artificial context and logistical difficulties of irradiating animals in vivo over an extended number of fractions. Pre-clinical studies therefore typically employ moderately hypofractionated or ablative regimens (Demaria and Formenti, 2012; Deloch et al., 2016), potentially biassing the commonlyheld hypothesis that such regimens are better inducers of ICD.

Finally, greater understanding of the mechanisms of cell death will enable the application of novel radiosensitisers (Moding et al., 2016). For example, the dependence of many cancer cells specifically on the G2/M checkpoint has led to the development of agents that target this checkpoint, particularly Chk1, Wee1, ATM, and ATR (Dillon et al., 2014). Hypoxia-modifying therapy might prevent HIF-mediated revascularisation, recurrence, and metastasis (Moding et al., 2016). Additionally, targeting previously under-appreciated cell death types such as ferroptosis may support the combination of RT

fcell-08-00041 February 11, 2020 Time: 18:13 # 5

and immunotherapy (Lang et al., 2019). Alongside these developments, biomarker driven studies must not be neglected to help guide patient selection.

#### CONCLUSION

fcell-08-00041 February 11, 2020 Time: 18:13 # 6

Cancer cells undergo a range of interphase and mitotic death after irradiation via direct and indirect effects of RT. An increased understanding of the molecular mechanisms of

### REFERENCES


radiation-induced cell death will reveal novel opportunities for improving the overall anti-tumour efficacy of RT.

#### AUTHOR CONTRIBUTIONS

JS, RS, EH, and HG conceived and designed the study, revised the manuscript, and read and approved the submitted version. JS wrote the first draft of the manuscript. HG wrote sections of the manuscript.


is shown by all lines examined. Int. J. Radiat. Biol. 65, 217–227. doi: 10.1080/ 09553009414550251


**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 Sia, Szmyd, Hau and Gee. 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.

fcell-08-00041 February 11, 2020 Time: 18:13 # 8

# Cancer Radiotherapy: Understanding the Price of Tumor Eradication

Olga A. Martin<sup>1</sup> \* and Roger F. Martin<sup>2</sup>

*<sup>1</sup> Sir Peter MacCallum Department of Oncology, The University of Melbourne, Melbourne, VIC, Australia, <sup>2</sup> School of Chemistry, The University of Melbourne, Melbourne, VIC, Australia*

#### Keywords: cancer radiotherapy, quality of life, normal tissue toxicity, aging, systemic effects, DNA damage, DNA repair, radioprotectors

Cancer radiotherapy (RT) is involved in the treatment of more than a half of all cancer patients, because it is highly effective; 40% of cancer cures can be attributed to RT (Baskar et al., 2012). Moreover, the efficacy of RT is steadily improving, largely due to the striking progress in technology, aimed at maximizing the radiation dose to the tumor and minimizing the dose to normal tissues. This continual improvement contributes to the increasing numbers of cancer survivors. In Australia, the 5-years relative survival from all cancers (excluding skin cancer) increased from 48% in 1984–1988 to 68% in 2009–2013<sup>1</sup> . In 2012, 410,530 ex-cancer patients were alive 5 years after treatment; 1.8% of the population. In the USA, there are now 14M cancer survivors; ∼4% of the population (Travis, 2006; Travis et al., 2013). Accordingly, increasing attention is now directed to the quality of life (QoL) of cancer survivors, particularly to treatment-related toxicities (Stone et al., 2003; De Ruysscher et al., 2019), as highlighted in a recent report from the National Cancer Research Institute in the UK<sup>2</sup> .

#### Edited by:

*Liz Caldon, Garvan Institute of Medical Research, Australia*

#### Reviewed by:

*Alexandros G. Georgakilas, National Technical University of Athens, Greece Michael Jackson, University of New South Wales, Australia*

> \*Correspondence: *Olga A. Martin olga.martin@petermac.org*

#### Specialty section:

*This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology*

> Received: *13 December 2019* Accepted: *27 March 2020* Published: *24 April 2020*

#### Citation:

*Martin OA and Martin RF (2020) Cancer Radiotherapy: Understanding the Price of Tumor Eradication. Front. Cell Dev. Biol. 8:261. doi: 10.3389/fcell.2020.00261*

Normal tissue toxicity from RT can be attributed to three different etiologies. The most obvious of these can be defined as "targeted", due to relatively high radiation doses to normal tissues in the vicinity of a tumor. Ironically, the technological improvements in dose delivery that have diminished this problem, have contributed to the second category of normal tissue toxicity. Modern RT techniques (e.g., Intensity-Modulated RT, IMRT) use multiple moving beams that sculpt a volume of high dose encompassing the tumor, so quite large volumes of normal tissues are 'bathed' in low doses, within and between beams (Kry et al., 2005; Harrison, 2017). This category also includes scattered radiation that spreads out in different directions from each radiation beam. The third category can be considered as "systemic," reflecting the radiation-induced abscopal ("out-of-field") effect (RIAE). This is attributed to the localized stress in the irradiated volume, that triggers a systemic biological response that is propagated to sites distant from the irradiated volume, and is largely mediated by the immune system (Reynders et al., 2015; Siva et al., 2015). In a sense, the RIAE can be considered as the systemic counterpart of the cellular radiation-induced bystander effect (RIBE), although the historical understanding of the phenomena was quite different. The recognition of the RIBE (Nagasawa and Little, 1992; Prise and O'Sullivan, 2009) is much more recent, compared to early observations the RIAE by radiation oncologists, that manifest both as out-of-field tumor responses and out-of-field RT-associated toxicities (Mole, 1953; Siva et al., 2015).

The best-known RT-induced normal tissue toxicities are targeted effects (tissue responses in the higher dose volume), the subject of many classical and contemporary radiobiological studies (Stewart and Dorr, 2009). They can be acute (appear within weeks of irradiation), late (months to years after RT), or both. For targeted effects, there is a wide spectrum of individual radiosensitivity (RS) manifested as normal tissue toxicity (Barnett et al., 2009). Low dose- and RIAE-generated "silent" toxicities, e.g., chronic inflammatory responses and mutagenesis in radiosensitive tissues,

<sup>1</sup>Cancer in Australia Australian Association of Cancer Registries, 2017. <sup>2</sup>https://www.ncri.org.uk/lwbc/

can also lead to long-term tissue dysfunction, even for future generations (Dubrova, 2003). Just as it is well-established for targeted effects, one can expect that there will be a spectrum of individual RS for low dose and systemic effects.

Epidemiological findings in long-term cancer survivors treated with RT indicate the increased incidence of degenerative pathological conditions normally associated with aging, or age-related diseases (e.g., cardio- and cerebrovascular disorders, neurodegeneration including dementia, hormonal disturbances, cataracts, bone marrow insufficiency, immune system dysfunction, second cancers, and overall life shortening) (Cupit-Link et al., 2017). Evidence is accumulating for similar consequences of low dose IR exposure (Majer et al., 2014; Harrison, 2017). Therefore, aging may be the common link between the diverse late morbidities and RT. By amplifying the mechanisms that are responsible for cellular aging (Sabatier et al., 1995; Dubrova et al., 2002; Dubrova, 2003; Miller et al., 2008; Paulino et al., 2010; Azzam et al., 2012; Sabatino et al., 2012; Merrifield and Kovalchuk, 2013; Ungvari et al., 2013; Sprung et al., 2015; Shimura et al., 2016; Yin et al., 2018), RT may induce a premature aging manifested as an accelerated onset of chronic degenerative disorders in some patients. **Figure 1** highlights the similarities between the response to IR and aging, but there are some differences, e.g., differences in the spectra and severity of DNA lesions (Nikitaki et al., 2015). Premature cellular senescence (Nakamura et al., 2008) is also an important common feature of the two pathologies. It is important to note that the idea of RT-induced accelerated aging is not a new one (Richardson, 2009), but given the growing aging population, it has increasingly important consequences for the cost of community health care. Moreover, the availability of improved biomarkers provides a means of monitoring both the need for intervention, and the efficacy of proposed interventions.

Therefore, to fully understand the role of RT in accelerating the aging process, research aimed at the following objectives is required: (1) Development of a "signature" profile of systemic markers to identify RT cancer patients susceptible to development of premature aging; (2) Improving mechanistic understanding of systemic propagation of genotoxic events and the associated aging phenotype following local exposure to IR; (3) Development of strategies for prevention, protection and mitigation of RT-related systemic genotoxic events.

#### IDENTIFICATION OF A SIGNATURE OF SYSTEMIC MARKERS FOR PREMATURE AGING IN RT CANCER PATIENTS

The kinetics of aging biomarkers could be monitored in blood of RT-treated cancer patients and compared with the pre-treatment values. Suitable patients, scheduled for treatment with RT, would be <50–60 years of age, and without evidence of non-cancer morbidities at the time of work-up. Examples of a suitable cohort would be breast or head & neck cancer patients, with an anticipated 5-years survival >90%.

Studies indicate that accumulated unrepaired systemic DNA damage underlies RT-induced pathologies (De Ruysscher et al.,

2019). The DNA damage response (DDR) varies in young, mature and old mice, slow down with age, making old mice the most vulnerable to radiation effects (Kovalchuk et al., 2014). DDR declines in senescent cells and during normal and premature human aging (Sedelnikova et al., 2008), and individual RS continuously rises with age (Schuster et al., 2018). Novel functional assays of radiation-induced DNA damage recognition and repair efficiency in ex-vivo irradiated primary human fibroblasts and peripheral blood mononuclear cells (PBMC) have been recently developed. The tests are based on post-irradiation formation of nuclear repair foci at the sites of DNA double-strand breaks (DSBs) for two DNA damage markers, phosphorylated ATM kinase ("ATM nucleo-shuttling") (Bodgi and Foray, 2016; Pereira et al., 2018) and histone H2AX (γ-H2AX) (Martin et al., 2013; Lobachevsky et al., 2015a, 2016). Both assays efficiently separated radiosensitive individuals with impaired DDR from those with normal RS. In our retrospective study, ex-RT patients who had documented to have severe RT-induced toxicity (and matched controls who responded normally) were recalled for blood sampling. A novel statistical algorithm was developed by Lobachevsky et al. (2016), based on non-linear regression analysis of the kinetics of repair of γ-H2AX foci, following exvivo irradiation of the PBMC. Subsequently the same dataset was analyzed by Bayesian modeling (Herschtal et al., 2018). Both methods of analysis distinguished the radiosensitive patients from controls, but the Bayesian statistics also outlined the importance of assessment of both the initial radiation-induced DNA damage and DNA damage repair. In a later study, the exvivo γ-H2AX response was assayed in PBMC collected before and during RT, and this showed that RT itself can affect individual RS, as reflected by changes in DSB repair efficiency in PBMC (Yin et al., 2018), adding a further dimension to the challenge of implementation.

Also, the number of endogenous γ-H2AX foci per cell in PBMC of normal individuals increases with age in a linear fashion (Sedelnikova et al., 2008; Schurman et al., 2012). The outliers identified in the linear regression analysis (Schurman et al., 2012) included elevated γ-H2AX foci/cell in patients with clinical morbidities. Interestingly, DDR has been linked with the immune response for normal and tumor tissues, as evidenced by cumulative bioinformatics studies (Georgakilas et al., 2015).

Therefore, the numbers of γ-H2AX foci/cell and efficiency of DDR in PBMC could provide a basis for identification of RT patients susceptible to RT-induced premature aging. However, it is more likely that a combination of markers will be required to constitute an effective "signature" to identify patients requiring added attention. Candidates for such auxiliary biomarkers include those reflecting immune and epigenetic alterations, increased immune cell senescence, oxidative stress, and mitochondrial dysfunction.

### IMPROVING THE MECHANISTIC UNDERSTANDING OF SYSTEMIC PROPAGATION OF GENOTOXIC EVENTS AND THE ASSOCIATED AGING PHENOTYPE FOLLOWING LOCAL EXPOSURE TO IR

Conventional RT triggers systemic biological effects in animal models (Koturbash et al., 2006, 2008; Mancuso et al., 2008), but due to significant scatter, RIAEs are difficult to interpret. The scatter problem associated with conventional radiation sources is much reduced with Synchrotron radiation, providing a useful tool to study RIAE. The defined geometry and coherence of the synchrotron beam delivers IR to small volumes with lower scatter, and the high dose rate (up to >1,000 Gy/sec) minimizes motion artifacts, but also introduces the "FLASH" effect (Durante et al., 2018). Ventura et al. (2017) reported that that various synchrotron settings (IR dose, volume, beam modality) trigger similar systemic effects in normal mouse tissues of wild-type C57BL/6 mice. Depending on the level of scatter radiation (Lobachevsky et al., 2015b), these effects were attributed to either true abscopal signaling, or to direct low-dose scatter radiation. RIAE was abrogated in mice with immune deficiencies, e.g., in mice with non-functional macrophages (Lobachevsky et al., 2019). Possible extensions of these studies using synchrotron irradiation include comparison of targeted and out-of-field effects of IR in young and old mice of wild-type and immunedeficient strains, as well as verification of salient features using a model with conventional radiation beams. The objective would be to understand the pathways by which IR modulates the aging processes in various organs crucial for the development of IR-related late pathologies (e.g., spleen, bone marrow, heart, vasculature, gonads, brain). These experimental models could also be used to evaluate potential therapeutic targets that emerge from the clinical studies described in the previous section.

#### DEVELOPMENT OF STRATEGIES FOR PREVENTION, PROTECTION AND MITIGATION OF RT-RELATED SYSTEMIC GENOTOXIC EVENTS

Targeted radiation effects and RIAE are both associated with elevated DNA damage and genome instability mediated by reactive oxygen species (ROS), so it is critical that any strategy aimed at reducing systemic genotoxic events does not compromise the cancer therapy, mediated by targeted radiation effects. Similarly, whilst normal tissue toxicities associated RIAE are mediated by the immune response, the tumor response to RT also involves the immune response (Haikerwal et al., 2015; Xing et al., 2019). However, kinetic studies on the impact with immunomodulators on the response of targeted tumors, out-of-field metastases and RIAE in normal tissues may reveal differences in response kinetics enabling selective suppression of RIAE in normal tissues. Such kinetic differences enabled scatter effects to be distinguished from the systemic RIAE (Ventura et al., 2017).

Our mouse studies revealed that molecules that block cytokines/cytokine receptors and macrophages can be expected to mitigate abscopal genotoxic events in normal tissues (Ventura et al., 2017; Lobachevsky et al., 2019). Our extensive review of potential strategies for prevention of RT-induced second cancers (Martin et al., 2016) illustrates the range of approaches that can be considered for all toxicities mediated by RIAE. A review on radiation-induced cardiotoxicity (Stewart et al., 2013), which noted the role of systemic effects, also discussed strategies for prevention. Another report extensively reviewed strategies for amelioration of radiation effects on the eye (Kleiman et al., 2017), some of which could be applicable to RT. In this context, a relatively new family of radioprotectors developed by one of the authors (RFM) is of interest. The first example, methylproamine (Martin et al., 2004), is a potent radioprotector in vitro; a dose modification factor of 2.0 at a concentration of 10 uM (Lobachevsky et al., 2011) and improved analogs, including in vivo activity, have been reported in the patent literature (Martin et al., 2011). Such radioprotectors have the potential to take advantage of the slow kinetics observed for the RIAE, illustrated by the report of the delayed appearance of DNA damage in eyebrow hair follicles after RT of lung cancer patients; 24 h after the first fraction (Siva et al., 2015). A delayed DDR is well-established for the cellular RIBE (Sedelnikova et al., 2007). Interestingly, methylproamine protects bystander cells in the in vitro RIBE setting, e.g., if present with recipient cells at the time of transfer of media irradiated cells (Burdak-Rothkamm et al., 2015). By contrast, in the context of targeted radiation effects, methylproamine must be present before and during irradiation to endow radioprotection of cultured cells (Lobachevsky et al., 2011), consistent with the mechanism (DNAbinding antioxidant) of radioprotection (Martin and Anderson, 1998). Thus, one can envisage an RT scenario in which such radioprotectors could be administered after irradiation, and thus not compromise response of the tumor, but nevertheless mitigate the subsequent RIAE mediated toxicity in normal tissues.

Whilst scheduling that avoids the possibility of compromising the tumor response might be challenging in the setting of conventional fractionation, this would be less problematical for hypofractionation modalities.

#### CONCLUSIONS

RT has an established role in cancer therapy and is unlikely to be superseded in the foreseeable future. In addition to pursuing better treatments, it is time now for more focus on the QoL of cancer RT survivors. The priorities include the need to understand the biological basis of treatment side-effects and their management, and, in particular, the mechanisms responsible for RT-induced aging phenotype and associated pathologies. This new knowledge is expected to enable development of systemic markers to identify patients most susceptible to accelerated aging, and the early stages of that process, as well as novel interventions

#### REFERENCES


Haikerwal, S. J., Hagekyriakou, J., MacManus, M., Martin, O. A., and Haynes, N. M. (2015). Building immunity to cancer with radiation therapy. Cancer Lett. 368, 198–208. doi: 10.1016/j.canlet.2015.01.009


for prevention and mitigation. Thus, the overall objective is early diagnosis, monitoring and management of RT-related morbidities, and identification of those cancer patients at most risk of these morbidities so their treatments can be modified accordingly.

#### AUTHOR CONTRIBUTIONS

OM and RM contributed ideas and wrote the manuscript.

#### ACKNOWLEDGMENTS

The authors acknowledge continuous support and advice of clinical colleagues from the Department of Radiation Oncology, Peter MacCallum Cancer Center. OM's membership in the Multidisciplinary European Low-Dose Initiative (MELODI) contributed to her awareness of ionizing radiation-induced premature aging.


mouse cerebellum. Proc. Natl. Acad. Sci. U.S.A. 105, 12445–12450. doi: 10.1073/pnas.0804186105


**Conflict of Interest:** RM has a commercial interest associated with his intellectual property on DNA-binding radioprotectors.

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 © 2020 Martin and Martin. 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.

# Breathing New Life into the Mechanisms of Platinum Resistance in Lung Adenocarcinoma

Alvaro Gonzalez-Rajal <sup>1</sup> , Jordan F. Hastings <sup>2</sup> , D. Neil Watkins 3,4, David R. Croucher 2,5 and Andrew Burgess 1,6 \*

*<sup>1</sup> ANZAC Research Institute, Concord, NSW, Australia, <sup>2</sup> The Kinghorn Cancer Centre, Garvan Institute of Medical Research, Sydney, NSW, Australia, <sup>3</sup> Research Institute in Oncology and Hematology, Cancer Care Manitoba, Winnipeg, MB, Canada, <sup>4</sup> Department of Internal Medicine, Rady Faculty of Health Sciences, University of Manitoba, Winnipeg, MB, Canada, <sup>5</sup> St Vincent's Hospital Clinical School, University of New South Wales, Sydney, NSW, Australia, <sup>6</sup> The University of Sydney Concord Clinical School, Faculty of Medicine and Health, Sydney, NSW, Australia*

Keywords: cisplatin, TGF-β, DNA damage, DNA repair, follistatin, cell cycle, p21, p53

## INTRODUCTION

Lung cancer accounts for approximately 11% of all cancer cases, however the 5-year survival rate is often below 20%. Consequently, lung cancer is the leading cause of cancer related mortality worldwide (Bray et al., 2018). There are two major types of lung cancer; small cell lung cancer (SCLC), which accounts for ∼15% of cases and non-small cell lung cancer (NSCLC), which accounts for ∼85% (Herbst et al., 2018). NSCLC is further separated into lung adenocarcinoma (LUAD, ∼50%), squamous cell carcinoma (∼30%) and multiple smaller subtypes (∼20%). Notably, up to 75% of NSCLC patients are diagnosed with advanced stage III/IV lung cancer (Walters et al., 2013), limiting surgical intervention.

#### Edited by:

*Weimin Li, Washington State University, United States*

#### Reviewed by:

*Joseph William Landry, Virginia Commonwealth University, United States*

\*Correspondence:

*Andrew Burgess andrew.burgess@sydney.edu.au*

#### Specialty section:

*This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology*

> Received: *17 December 2019* Accepted: *07 April 2020* Published: *08 May 2020*

#### Citation:

*Gonzalez-Rajal A, Hastings JF, Watkins DN, Croucher DR and Burgess A (2020) Breathing New Life into the Mechanisms of Platinum Resistance in Lung Adenocarcinoma. Front. Cell Dev. Biol. 8:305. doi: 10.3389/fcell.2020.00305*

While smoking is strongly associated with all lung cancer types, at least 20% of LUAD cases are from non–or never smokers (Herbst et al., 2018). Furthermore, while LUAD is characterized by a high somatic mutation rate, with deletion or mutation of TP53 occurring in up to 46% of cases, <20% of patients carry targetable mutations such as those within EGFR, ALK, or BRAF or NTRK (Arbour and Riely, 2019). Consequently, the overwhelming majority of LUAD patients receive platinum-based chemotherapy as standard of care.

Unfortunately, response rates to platinum in LUAD are below 30%, due to innate/acquired resistance and rate-limiting side-effects such as nephrotoxicity (Marini et al., 2018). Importantly, potential synergy between platinum chemotherapy and immunotherapy has emerged as a therapeutic opportunity in LUAD (Mathew et al., 2018). Therefore, improving platinum efficacy and identifying mechanism of resistance could significantly improve patient outcomes. In this opinion article, we cover several of the latest landmark publications that shed new light on the mechanisms of platinum resistance in LUAD.

## OVERVIEW OF PLATINUM CHEMOTHERAPY

The anti-tumor abilities of cisplatin were identified over 50 years ago (Rosenberg et al., 1969). Since then platinum has become one of the most successful chemotherapeutics developed. It is essentially curative in testicular cancer, with survival rates >90% (Koster et al., 2013). It is also used with varying degrees of success to treat ovarian, head and neck, bladder and cervical cancer. Second and third generation cisplatin analogs have now been developed with the aim of lessening nephrotoxicity, neurotoxicity, ototoxicity, or providing better bioavailability and overcoming tumor resistance. Of these, carboplatin and oxiplatin are the most well-known, however nedaplatin, heptaplatin, lobaplatin and satraplatin are also used clinically (Wang and Lippard, 2005).

Cisplatin and its derivates rely on their platinum group to exert killing. Platinum compounds can bind to many biological targets including DNA, RNA, and proteins (Stordal and Davey, 2007). The binding of cisplatin to DNA forms platinum-DNA adducts (**Figure 1**), which must be repaired by the cell. Approximately 90% of cisplatin-induced adducts are intra-strand crosslinks that are rapidly repaired mostly by the base-excision and nucleotide excision repair (BER, NER) pathways during G1 phase (Slyskova et al., 2018). In contrast, inter-strand crosslinks (ICL) represent <5% of cisplatin-induced adducts but are far more difficult for cells to remove as they are "hidden" within the DNA helix. ICLs prevent the unzipping of the double helix, creating a physical barrier to efficient DNA replication. The removal, largely by the Fanconi anemia (FA) pathway (Michl et al., 2016; Niraj et al., 2019; Smogorzewska, 2019), results in the formation of single and double strand breaks (SSBs and DSBs). The damaged DNA is then repaired by either the high-fidelity homologous recombination (HR) pathway during S/G2-phase (Karanam et al., 2012) or by the error-prone non-homologous end joining (NHEJ) pathway in G1 phase (Enoiu et al., 2012). The extent of, or failure to repair the DNA damage caused by cisplatin can result in cell death, accounting for the cytotoxic mode of action for most platinum agents. The exception is oxiplatin, which kill cells through increasing ribosome biogenesis stress (Bruno et al., 2017). For simplicity, here we will only focus on the mechanisms of cisplatin resistance in LUAD.

#### Screening for Platinum Sensitisers

To date over 147 mechanisms of platinum resistance have been proposed (Stewart, 2007), yet there remains a lack of viable clinical options to improve response rates. To overcome this, several recent publications (Cheng et al., 2016; Jhuraney et al., 2016; Jin et al., 2018; Marini et al., 2018; Ding et al., 2019; Hsu C.- H et al., 2019), have looked at potential mechanisms of resistance in LUAD using unbiased screens, and detailed preclinical models. Based on these new data, it is clear that the main points of resistance arise from alterations to DNA repair, TGF-β signaling, cell cycle and apoptosis (**Figure 1**). Put simply, the ability of cisplatin to kill cells requires actively cycling cells that generate sufficient DNA damage and a functional apoptotic pathway to induce death. Consequently, disruption at any point along these pathways can prevent cell death, thereby reducing sensitivity to platinum mediated killing. Conversely, synergising therapies in general either block inhibitory cell death pathways, thereby lowering the threshold required to trigger death or increase the amount of damage induced by platinum.

#### DNA Repair and Resistance to Platinum

The DNA Damage response (DDR), has been extensively reviewed (Jackson and Bartek, 2009; O'Connor, 2015; Pearl et al., 2015; Matt and Hofmann, 2016), as has its role in LUAD (O'Grady et al., 2014). Briefly, the DDR comprises of several functional layers including sensors (e.g., MRN complex, RPA, ATRIP), signaling kinases (e.g., ATM, ATR), damage mediators (e.g., 53BP1, BRCA1/2, H2AX), downstream kinases (e.g., CHK1/2), and cell cycle checkpoint effectors (e.g., p53, p21, WEE1). Unsurprisingly, defects at each level have been reported to regulate sensitivity to cisplatin in a variety of cancers including LUAD. Perhaps the best example of this is the well-reported link between BRCA1/2 mutations and sensitivity to cisplatin in breast cancer (Tutt et al., 2018). Similarly, the BRACness phenotype, which is defined as any defect that impacts HR repair and phenocopies the mutation or loss BRCA1/2 (Byrum et al., 2019b), is also strongly linked with sensitivity to platinum and PARP inhibitors (Ding et al., 2019), especially in ovarian (Pillay et al., 2019) and breast cancer (Tung and Garber, 2018). The links with BRCA1/2 mutations, BRACness and cisplatin sensitivity are less clear in LUAD. Although recent reports indicate that DNA methyltransferase inhibitors can induce a BRACness phenotype in NSCLC cells, sensitizing them to PARP inhibitors (Abbotts et al., 2019), and hence may extend to other DNA damage chemotherapies such as cisplatin (**Figure 1**). Early preclinical studies showed significant promise for directly inhibiting ATR kinase activity (Hall et al., 2014; Vendetti et al., 2015) to enhance cisplatin killing of LUAD cells. Interestingly, inhibition of ATM does not appear to synergise with cisplatin (Schmitt et al., 2017), although it may reduce the metastatic potential of cisplatin resistant LUAD cells (Shen et al., 2019). Furthermore, co-depletion of ATM and MCL-1 can re-sensitize cells to cisplatin (Zhang et al., 2017). While phase 1/2 trials of the CHK1 inhibitor LY2603618 in combination with cisplatin showed promising anti-tumor activity, but also caused significant thromboembolic side-effects (Wehler et al., 2017), indicating that despite promising results in SCLC (Sen et al., 2017; Hsu W.- H. et al., 2019; Nagel et al., 2019), Chk1 inhibitors may not translate to LUAD. Indirectly targeting the DDR has also shown some promise, with inhibition of the JMJD2 histone demethylase family re-sensitizing resistant LUAD to cisplatin by preventing ATR association to sites of DNA damage, thereby weakening the DDR (Duan et al., 2019). Similarly, targeting specific forms of the PP2A phosphatase complex (PPP2R2A2), which are responsible for dephosphorylating and inactivating ATM and ATR, enhanced sensitivity to PARP inhibition in LUAD by maintaining the DDR response (Kalev et al., 2012). What is becoming clear is that there are a number of non-canonical DDR pathways, many of which become upregulated during oncogenesis and can increase replication fork stability and counterbalance BRACness and BRCA mutations (Chaudhuri et al., 2016). A surprising recent example is the discovery that the mitotic kinase Aurora A and its targeting factor TPX2 can regulate 53BP1 and HR repair in a pathway parallel to BRCA1 (Byrum et al., 2019a), possibly explaining why shRNA knockdown of Aurora A sensitized LUAD cells to cisplatin (Cheng et al., 2016). These results highlight the need for additional research that maps all of the pathways regulating the DDR in LUAD.

## TGF-β Signaling, EMT and Resistance

The sensing and repair of cisplatin adducts does not happen in isolation from the rest of the cell or its local environment. The DDR signaling pathway is intimately integrated into multiple signaling networks, with a prime example being the transforming growth factor β (TGF-β) pathway. TGF-β regulates a multitude of cellular pathways including the DDR, cellular proliferation and the epithelial-mesenchymal transition (EMT). It plays

both positive and negative roles in cancer development and progression. In established tumors, high TGF-β expression can drive metastasis, tumor heterogeneity and chemoresistance (Li J. et al., 2019). We recently demonstrated that members of the TGF-β pathway, including ACVR1B, TGFBR1, TAK1 and GDF11, mediated innate cisplatin resistance in LUAD (**Figure 1**), a possible consequence of epithelial airway cell lineage (Kretser et al., 2011). Critically, inhibition of activin receptor signaling reversed the resistance, as did blockage of activin A and GDF11 by the endogenous protein Follistatin (Marini et al., 2018). The mechanisms for TGF-β resistance are multifaceted, likely acting to suppress cell proliferation, apoptosis, and the DDR. In support, the antiapoptotic protein MCL-1 decreased upon chemical inhibition of the TGF-β pathway in cisplatin treated cells (Marini et al., 2018). TAK1 has also recently been shown to phosphorylate p38 MAPK and IKKα after DNA damage (Colomer et al., 2019), promoting ATM phosphorylation and increasing DNA repair, leading to chemoresistance. In turn, ATM can feedback into the TGF-β pathway, phosphorylating c-Cbl, stabilizing TβRII receptor and activating TFG-β signaling (Li Y. et al., 2019),

creating a positive feedback loop (**Figure 1**). TGF-β can also drive EMT (Hao et al., 2019) and chemoresistance (Fischer et al., 2015), in part due increased YAP1 mediated transcription of TGF-β target genes (Pefani et al., 2016). Consequently, TGFBR1 and YAP1 inhibitors have been shown to be synergistic in GATA4 deficient (Hao et al., 2019) and EGFR-mutant (Cheng et al., 2016) lung cancers, respectively (Gao et al., 2019), offering another potential therapeutic approach to enhancing cisplatin selectivity.

## Cell Cycle and Apoptosis

In general, non-cycling cells are more resistant to cytotoxic chemotherapies such as cisplatin, however, proliferating cells that increase repair or reduce death signaling are more resistant, and often more deadly. Once a proliferating cell encounters DNA damage it must halt cell cycle progression so that repair can occur. If the damage is deemed too great, then apoptosis will be initiated, thereby preventing the damage being passed on to subsequent generations. The key central regulator of this decision pathway is p21waf1/kip, which inhibits G1 and G2 cell cycle progression (Burgess et al., 2019) and blocks caspase 3 dependent apoptosis (Suzuki et al., 2000). Interestingly, intermediate "goldilocks" levels of p21 strongly correlate with continued cell proliferation post cisplatin exposure, while low or high levels result in damaged cells undergoing senescence (Hsu C.-H et al., 2019). Similarly, over-riding the protective cell cycle checkpoints in S and G2 phase through WEE1 inhibition has also shown promise, especially in p53 null and mutant cell lines (Jhuraney et al., 2016; Richer et al., 2017). Interestingly, some resistant cycling cells become highly dependent on glutamine for a multitude of metabolic reactions. Consequently, removal of glutamine makes resistant cells highly sensitive to cisplatin, and lowers the threshold required to trigger apoptosis (Guidi and Longo, 2018). Similarly, metformin, which blocks glucose uptake and ATP production, has also been linked with increasing sensitivity to cisplatin (Liu et al., 2017; Riaz et al., 2019). While, inhibition of NRF2, which protects against hypoxia and reactive oxygen species (ROS), synergises with cisplatin by enhancing DNA damage (Singh et al., 2016; Shi et al., 2019). Notably, NRF2 is commonly upregulated in LUAD by KRAS (Tao et al., 2014) and mutant p53 (Tung et al., 2015). Disrupting apoptosis is another common mechanism, with upregulation of MAST1 in LUAD cells resulting in a rewiring of downstream MEK signaling and a reduction in pro-apoptotic protein Bim (**Figure 1**), thereby increasing the threshold required to trigger apoptosis (Jin et al., 2018). Likewise, mutations in SET containing 2 (SETD2), a histone methyltransferase, confers cisplatin resistance in LUAD by altering ERK signaling and inhibiting apoptosis (Kim et al., 2019). While, as mention, TAK1-p38 signaling results in an increase in anti-apoptotic MCL1 levels, raising the threshold required to trigger apoptosis (Marini et al., 2018).

#### DISCUSSION AND CONCLUSION

A more complete understanding of the signaling, repair and apoptotic networks that are re-wired in LUAD will be key

#### REFERENCES


to improving platinum efficacy in LUAD. In addition, better temporal information on the dynamic nature of the signaling responses will greatly aid in the identification and prediction of resistance mechanisms. Any models will need to take into account cell cycle status, repair pathway and apoptotic thresholds in order to identify suitable synergising treatments. Finally, better preclinical models that more accurately model the dosing of platinum will be essential. Currently, the majority of studies rely on prolonged exposure, often >10-fold higher than what is achievable in patients (Urien and Lokiec, 2004; Jacobs et al., 2005). Screening of synergistic treatments using this extreme exposure may have increased the rate of false positives and failure of some preclinical studies to translate clinically. This is further cofounded by the disparate effect that platinum has on various organs (Yimit et al., 2019), especially the kidneys. Consequently, treatments such as Follistatin (Marini et al., 2018), which not only protect these vital organs but also enhance tumor selective killing, may have significant clinical potential. In summary, the advent of large-scale screens combined with detailed preclinical studies has given a greater understanding of the mechanisms of cisplatin resistance in LUAD, breathing new life into this stalwart of chemotherapy.

### AUTHOR CONTRIBUTIONS

AG-R and JH co-wrote the initial draft. DW and DC cowrote and edited the manuscript. AB conceived and wrote the article.

#### FUNDING

AB was supported by a NBCF Investigator Initiated Research Scheme (IIRS-18-103). DC is also supported by NBCF Investigator Initiated Research Schemes (IIRS-18-121 and IIRS-20-032).

forks during replication stress by counteracting 53BP1 function. J. Cell Biol. 218, 422–432. doi: 10.1083/jcb.201803003


resistance in KEAP1-deficient NSCLC tumors. Acs Chem. Biol. 11, 3214–3225. doi: 10.1021/acschembio.6b00651


**Conflict of Interest:** DW is a coinventor on a patent application relating to components of this work (U.S. 20180125936-A1).

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 © 2020 Gonzalez-Rajal, Hastings, Watkins, Croucher and Burgess. 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.

# A Survey of Essential Genome Stability Genes Reveals That Replication Stress Mitigation Is Critical for Peri-Implantation Embryogenesis

#### Georgia R. Kafer and Anthony J. Cesare\*

Genome Integrity Unit, Children's Medical Research Institute, The University of Sydney, Westmead, NSW, Australia

#### Edited by:

Liz Caldon, Garvan Institute of Medical Research, Australia

#### Reviewed by:

Fumiko Esashi, University of Oxford, United Kingdom Yibo Luo, The University of Toledo, United States

#### \*Correspondence:

Anthony J. Cesare tcesare@cmri.org.au

#### Specialty section:

This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 10 February 2020 Accepted: 05 May 2020 Published: 29 May 2020

#### Citation:

Kafer GR and Cesare AJ (2020) A Survey of Essential Genome Stability Genes Reveals That Replication Stress Mitigation Is Critical for Peri-Implantation Embryogenesis. Front. Cell Dev. Biol. 8:416. doi: 10.3389/fcell.2020.00416 Murine development demands that pluripotent epiblast stem cells in the periimplantation embryo increase from approximately 120 to 14,000 cells between embryonic days (E) 4.5 and E7.5. This is possible because epiblast stem cells can complete cell cycles in under 3 h in vivo. To ensure conceptus fitness, epiblast cells must undertake this proliferative feat while maintaining genome integrity. How epiblast cells maintain genome health under such an immense proliferation demand remains unclear. To illuminate the contribution of genome stability pathways to early mammalian development we systematically reviewed knockout mouse data from 347 DDR and repair associated genes. Cumulatively, the data indicate that while many DNA repair functions are dispensable in embryogenesis, genes encoding replication stress response and homology directed repair factors are essential specifically during the periimplantation stage of early development. We discuss the significance of these findings in the context of the unique proliferative demands placed on pluripotent epiblast stem cells.

Keywords: early development, embryology, pluripotency, DNA damage response, DNA repair, DNA replication, replication stress response

## INTRODUCTION

#### Overview

Pluripotent cells in early mammalian embryos proliferate at a phenomenal rate. This is necessary to maintain embryo growth and reach critical developmental milestones within defined temporal windows. Because all somatic tissues are derived from these early pluripotent precursors, it is critical that genome integrity is maintained during early development. Embryonic pluripotent stem cells are thus subjected to unique challenges to maintain their DNA health. To elucidate which genome stability pathways are essential for early development we probed the Mouse Genome Informatics Gene Ontology Project (MGI-GO) database (Bult et al., 2019). Within MGI-GO we identified 347 genes grouped within the ontologies of major DNA repair pathways (MGI-GO designations: DNA damage checkpoint, nucleotide excision repair, mismatch repair, base excision repair, homologous recombination, and non-homologous end joining). Of these genes, we identified 297 with a validated mouse knockout. From these 297 murine models, only 108 gene knockouts were lethal during embryonic development (**Supplementary Table S1**). Within the grouping of 108 embryonic

lethal genes, 10 knockouts were lethal during preimplantation development prior to E4.5 (**Table 1**), and 36 knockouts were lethal during somite stages from E8.5 (**Supplementary Table S1**). Notably, most of the targeted genes that conferred embryonic lethality, 62 genes, did so specifically during the period of rapid cell proliferation occurring with peri-implantation development (E4.5 to E8.5) (**Table 2**). Below we briefly review pre- and peri-implantation murine development before considering the function of essential genome stability factors across the early stages of embryonic development. Finally, we discuss why the unique cells of the peri-implantation embryo appear to specifically require replication stress response factors for cell viability.

#### Early Murine Development

Embryonic development consists of a series of events occurring in chronological progression. Murine development takes 19 to 20 days depending on mouse strain (**Figure 1**; Murray et al., 2010). Preimplantation development occurs between fertilization (E0) and the initiation of embryo implantation in the uterine wall (around E4.0). Following fertilization, embryonic cells are uniformly totipotent and identical until formation of the morula at E2.5 (Condic, 2014) (see **Box 1** for a detailed explanation of cell potency). Within the morula cells undergo polarization and compaction (Humi˛ecka et al., 2017). By E3.0 the embryo forms a blastocyst structure containing two distinct cell populations; an outer layer of multipotent trophectoderm cells that eventually derive the placental tissues, and an inner group of pluripotent inner cell mass (ICM) cells which primarily serve as precursors for the embryo proper (Morris et al., 2010).

Peri-implantation is the developmental period from implantation to organogenesis (E4.0 to E8.5). Implantation begins at E4.0 when the free-floating embryo contains 64 cells (Behringer et al., 2014). At this point, the embryo loses its glycoprotein zona pellucida shell and the outer trophoblast attaches to the uterine wall. Single cell transcriptomics demonstrate that at E4.0 the ICM has differentiated into two pluripotent cell types: the primitive endoderm (PrE, sometimes referred to as the hypoblast) and the epiblast (Mohammed et al., 2017). Peri-implantation development is associated with exceptionally rapid cell proliferation during "gastrulation," which begins at E5.5 as the embryo elongates and a luminal pro-amniotic cavity opens in the center of the epiblast cell mass (Snow, 1976). At E6.5 the primitive streak emerges from the epiblast, and from the epiblast all three dermal lineages of the conceptus will arise (Tam et al., 1993). These are the ectoderm

BOX 1 | Potency states. Totipotent cells are present in the embryo between fertilization and morula formation. They can give rise to any cell type from any stage of the animal's life, including germ and placental cells. Pluripotent cells exist in the inner cell mass (ICM) and epiblast region of the developing embryo from the blastocyst stage to immediately prior to organogenesis. Pluripotent cells can give rise to any cell of dermal lineage (mesoderm, endoderm or ectoderm) but not placental or germ cells. Multipotent cells exist in newly formed tissues or organs and can develop into a limited number of cell types within their original dermal lineage.

 stability factors essential for preimplantation development.


#### TABLE 2 |Genome stabilityfactors essential for peri-implantation development.


(Continued)

DNA Repair in Early Development

#### TABLE 2 | Continued


(Continued) DNA Repair in Early Development

DNA Repair in Early Development

fcell-08-00416 May 29, 2020 Time: 17:2 # 5

#### TABLE 2 | Continued


Key: (E) = Embryonic day; DDR = DNA damage response/damage sensing; CC = cell cycle control; Rep = DNA replication and/or replication stress; HDR = Homologous directed repair; EJ = End joining pathways; BER = Base excision repair; NER = Nucleotide excision repair; MMR = Mismatch repair; Telomere = telomere protein; TLS = Translesion synthesis. Gene functions are cross-referenced with the literature and may diverge from the MGI-GO categorization. Factors labeled with \* indicate when the gene encodes a protein with additional functions outside genome stability that likely contribute to embryonic lethality.

exocoelom becomes the yolk sack. Between E8.5 and E9.5 the embryo turns, resulting in complete envelopment of the embryo within the supporting

(nervous system and skin precursor), endoderm (gut precursor), and mesoderm (precursor for all other tissues) (Watson and Tam, 2001). Because continued gastrulation requires a critical mass of 1000 cells, delays in epiblast proliferation at E6.5 may result in the embryo forming a small primitive streak but developing no further (Snow and Tam, 1979; Tam, 1989; Power and Tam, 1993).

extraembryonic tissues. Embryo depictions are not to scale.

Between E6.5 and E7.5 the rate of epiblast proliferation is exceptionally pronounced, with some cell cycles completed in 2.2 h (Snow, 1977). Elongation of the primitive streak between E6.5 and E7.0 is associated with the first evidence of epithelial to mesenchymal transition (EMT) (Carver et al., 2001). EMT mobilizes epiblast cells to form the endoderm and mesoderm lineages, while cells that do not undergo EMT become the ectoderm (Acloque et al., 2009). At E7.5 the mouse embryo contains over 14,000 cells derived from epiblast progenitors. Proliferation slows at this point resulting in an increased average cell cycle duration of 8.1 h (Snow, 1977). There is evidence to suggest a that a specialized "proliferative zone" with sub 4 h cell cycles is maintained within the primitive streak after E7.5, however, definitive evidence remains to be substantiated (Tam et al., 2013).

### Genome Instability Differentially Impacts Cell Viability in Pre- and Peri-Implantation Embryos

Embryonic lethality implies that one or more cell populations are compromised during development at the time of embryo failure. This may arise through DNA damage or genome instability, with multiple lines of evidence indicating that preimplantation embryos are more resistant to genome perturbations than periimplantation embryos. For example, gamma irradiation-induced DNA damage at E6.5 or E7.5 confers apoptosis specifically within the epiblast (Heyer et al., 2000). Conversely, apoptosis levels are low in preimplantation embryos irradiated between E3.5 and E5.5, or during organogenesis from E8.5. It is therefore the epiblast that is particularly sensitive to these types of DNA lesions. Similarly, despite tetraploidy driving additional

chromosomal instability in preimplantation cells (Paim and FitzHarris, 2019) chimeric mouse embryos with a mixture of diploid and tetraploid cells will develop through preimplantation before dying during peri-implantation (Horii et al., 2015). Notably, tetraploid sensitivity in vivo appears to be specific to the epiblast as embryos with tetraploid trophoblast cells and diploid epiblast cells can generate live pups (Wen et al., 2017). Mouse embryos containing a mixture of diploid and aneuploid cells will also develop to peri-implantation before the aneuploid cells are specifically depleted in the epiblast through apoptosis (Bolton et al., 2016).

As with somatic tissues, the tumor suppressor TP53 (p53) plays a central role regulating stem cell outcomes following genomic insult. p53 orchestrates growth arrest or apoptosis following activation of the DNA damage response (Mello and Attardi, 2018). Concordantly, inhibiting p53 dependant signaling pathways enables chimeric embryos made from tetraploid preimplantation murine embryonic stem cells (mESCs) to survive until birth (Horii et al., 2015). Deleting TP53 also reduced apoptosis levels in irradiated E6.5 embryos (Heyer et al., 2000) and extended the survival of embryos codeleted for essential DNA repair factors (Jones et al., 1995; Haupt et al., 1997; Ludwig et al., 1997; Kim et al., 2002; McCarthy et al., 2003; Cang et al., 2006; Reinhardt and Schumacher, 2012). Not surprisingly, TP53 was identified as a critical mediator of apoptosis in the gastrulating epiblast (Laurent and Blasi, 2015). However, when activated in pluripotent stem cells, p53 also influences the expression of pluripotency factors to regulate differentiation (Lin et al., 2005; Li et al., 2012; Akdemir et al., 2014; Jain et al., 2016). p53 therefore functions through canonical and unique pathways in early development to regulate cellular outcomes. This highlights that our classic understanding of genome stability pathways may not strictly apply to early development or certain pluripotent cell types (Zaveri and Dhawan, 2018).

#### DNA DAMAGE RESPONSE AND REPAIR PATHWAYS

#### Replication Stress Response

Somatic mammalian cells prepare for DNA replication in G1 phase by licensing replication origins and loading inactive Cdc45-MCM-GINS replicative helicase complexes (Bleichert, 2019; Miller et al., 2019). Cyclin dependent kinase activity promotes E2F transactivation to initiate replication

at the G1/S transition (Kent and Leone, 2019). Replication then proceeds throughout the S-phase with origins firing in temporal coordination and DNA synthesis occurring across the entirety of the genome (Burgers and Kunkel, 2017; Limas and Cook, 2019). Intrinsic and extrinsic factors may disrupt replication fork processivity: a phenomenon known as "replication stress" (Zeman and Cimprich, 2014). Replication stress is sensed through the accumulation of RPA binding to its single strand DNA (ssDNA) substrate (Bhat and Cortez, 2018). When replication stress stalls DNA synthesis the replicative helicase continues to unwind its substrate exposing ssDNA for RPA coating (Byun et al., 2005). ATR kinase is the master regulator of the replication stress response (Saldivar et al., 2017). RPA coated ssDNA recruits ATR and its associated protein ATRIP (Cortez et al., 2001) to stalled replication forks through parallel pathways mediated by TopBP1 and ETAA1 (Kumagai et al., 2006; Bass et al., 2016; Haahr et al., 2016). Once localized to the stalled fork, ATR is activated and propagates a signaling cascade resulting in engagement of the replication stress response. This includes activation of the downstream effector CHK1 kinase to arrest S phase until replication stress is resolved (Zhang and Hunter, 2014). During the replication stress response, stalled replication forks are often remodeled into a four-way structure and protected before engaging one of many diverse repair mechanisms dependent upon the underlying stress the fork encountered (Quinet et al., 2017; Cortez, 2019).

If replicative stress is unresolved, arrested replication forks may collapse into one-ended double strand breaks (DSBs) (Ait Saada et al., 2018). Additionally, persistent replication stress can result in under-replicated DNA persisting through S-phase, the second growth (G2) phase, and into the mitotic (M) phase of the cell cycle (Mankouri et al., 2013). Specialized repair mechanisms address replication defects carried into mitosis (Minocherhomji et al., 2015), during which time the canonical DSB repair pathways are inhibited (Orthwein et al., 2014). Replication defects passed into mitosis can confer chromosome segregation errors resulting in aneuploidy (Burrell et al., 2013; Wilhelm et al., 2019), or if severe mitotic death (Masamsetti et al., 2019). If a replication stressed cell escapes mitosis this is often evident in the daughter cells where the under-replicated DNA is present as a scar in the first growth phase (G1) that is repaired in the subsequent S phase (Lukas et al., 2011; Spies et al., 2019).

#### Double Strand Break Repair

DSBs are a major threat to genome stability because a failure in repair may result in loss of an entire chromosome arm (Scully et al., 2019). Additionally, chromosome segregation errors resulting from inadequate DSB repair are rife with implications for genome instability, including chromothripsis and kataegis (Maciejowski et al., 2015; Ly and Cleveland, 2017). DSBs are sensed by Ku70/80 (Gottlieb and Jackson, 1993; Ochi et al., 2015), PARP1 (Ali et al., 2012; Liu C. et al., 2017), and the MRN (MRE11, RAD50 and NBS1) complex (Lee and Paull, 2005; Stracker and Petrini, 2011). MRN facilitates recruitment of the ATM kinase to DSBs (Uziel et al., 2003) while Ku70/80 recruits a related kinase DNA-PKcs (Hnizda and Blundell, 2019). Once recruited to the break, ATM is activated and engages downstream DDR pathways (Dupre et al., 2006; Maréchal and Zou, 2013). This includes phosphorylation of the histone variant H2AX (termed γ-H2AX when phosphorylated) within the break adjacent chromatin and assembly of factors at the break locus to facilitate repair (Rogakou et al., 1998). ATM activation also results in growth arrest, including activation of the p53 pathway (Banin et al., 1998). Subsequent repair of DNA breaks is then orchestrated by homology directed repair (HDR) or end-joining (EJ) pathways.

Homology directed repair utilizes the homologous sister chromatid as a template for repair and is thus limited to the S and G2 cell cycle phases (Hustedt and Durocher, 2016). HDR initiates with resection at the broken DNA end to provide a 3<sup>0</sup> ssDNA overhang for insertion into homologous regions of the sister chromatid (Symington, 2016). Once resection occurs at a canonical two-ended DSB, a series of enzymatic steps facilitate strand-invasion of the broken end into the sister chromatid, formation of a displacement-loop, template copying from the invaded DNA end, potential formation of a "Holliday Junction," and resolution of strand invasion (Scully et al., 2019). HDR factors also function within the replication stress response though diverse mechanisms including the remodeling and stabilization of stressed replication forks into four-way structures prior to repair (Neelsen and Lopes, 2015). Additionally, HDR factors participate in resolving one-ended DSBs that arise from collapsed replication forks to restart replication (Ait Saada et al., 2018). HDR is thus intrinsically linked to the replication stress response.

Conversely, EJ involves the covalent ligation of broken DNA ends to repair DSBs and thus does not require an additional chromatid template. End-joining mechanisms function in S and G2, but notably are the only DSB repair pathway available in G1 (Chang et al., 2017). Classical non-homologous end joining (c-NHEJ) directly ligates DNA ends, and if unregulated can also drive chromosome translocations (Ghezraoui et al., 2014) or telomere fusions (Celli et al., 2006; Van Ly et al., 2018). Cells can also engage alternative non-homologous end joining (alt-NHEJ), where DNA resection creates 3<sup>0</sup> overhangs to align the DNA ends for repair through microhomology (Sfeir and Symington, 2015). This often requires a fill-in reaction by the error prone polymerase Pol θ which can introduce errors into the repaired sequence (Mateos-Gomez et al., 2015; Mateos-Gomez et al., 2017).

## Nucleotide and Base Damage

In addition to DNA breaks, cells must also contend with damage to nucleotide bases, base mismatches, and bulky DNA lesions that distort the DNA double helix. Base excision repair (BER) mends small non-distorting DNA lesions such as oxidized DNA bases (Wallace, 2014), while nucleotide excision repair (NER) corrects bulky lesions such UV-induced pyrimidine dimers (Schärer, 2013). BER is mediated by DNA

glycosylases that cleave and remove the damaged bases at the lesion. A correct base is then inserted by specialized DNA polymerases (Wallace, 2014). In NER, a 25–30 nucleotide patch of ssDNA containing the bulky lesion is excised from the double strand helix before specialized DNA polymerases fill in the ssDNA gap (Schärer, 2013). Mismatch repair (MMR) proofreads the newly replicated DNA to identify misincorporated bases (Liu D. et al., 2017). When a mismatch is identified, the newly synthesized strand is nicked and resected creating a patch of exposed ssDNA for fill in by a highfidelity DNA polymerase.

#### Essentiality of DNA Repair During Embryogenesis

Not all DNA repair activities are essential for early development (**Supplementary Table S1**). The most striking example is the dichotomy between the major DSB repair pathways. While many HDR factors are essential, the core c-NHEJ factors are either dispensable for development [Ku70 (Ouyang et al., 1997), Ku80 (Nussenzweig et al., 1996; Gu et al., 1997), Prkdc (DNA-PKcs) (Kurimasa et al., 1999), Nhej1 (XLF) (Li et al., 2008)] or only required for organogenesis [Lig4 (Frank et al., 1998), Xrcc4 (Dickinson et al., 2016)]. Some alt-NHEJ components are required for early development. Alt-NHEJ is promoted by the polymerase Pol θ which primes the DNA ends for repair by removing RPA from exposed ssDNA regions (Kent et al., 2015; Mateos-Gomez et al., 2017). Alt-NHEJ is also dependent on PARP-1, XRCC1 and LIG3 (Audebert et al., 2004), with PARP-1 having a critical role in synapsis, and XRCC1 and LIG3 mediating DNA ligation (Audebert et al., 2004). While Lig3 and Xrcc1 knockout embryos die during gastrulation (Tebbs et al., 1999; Puebla-Osorio et al., 2006), both genes encode factors that have functions in other DNA metabolic activities outside EJ (Wallace, 2014). Conversely Pol θ knockout mice are viable (Shima et al., 2004; Masuda et al., 2005), as are Parp1 knockouts (Wang et al., 1995), suggesting that alt-NHEJ is not essential for embryogenesis.

Preference for HDR over NHEJ is further evident in mechanistic studies of DDR engagement within embryonic stem cells (Tichy et al., 2010). Following DSB induction, ATMdependent γ-H2AX phosphorylation at the break locus recruits the scaffolding protein MDC1 (Stewart et al., 2003). MDC1 subsequently recruits the ubiquitin ligases RNF8 and RNF168 that signal downstream recruitment of 53BP1 (Jackson and Durocher, 2013). 53BP1 then establishes a physical domain at the repair locus in coordination with RIF1 to inhibit end resection and HDR in favor of c-NHEJ (Chapman et al., 2013; Zimmermann et al., 2013; Lou et al., 2019; Ochs et al., 2019). Notably, many factors that promote c-NHEJ are dispensable for embryonic development, including Atm (Barlow et al., 1996; Xu et al., 1996), H2ax (Celeste et al., 2002), Mdc1 (Lou et al., 2006), Rnf8 (Valnegri et al., 2017), Rfn168 (Zong et al., 2019), and Trp53bp1 (53BP1) (Ward et al., 2003). Additionally, multiple reports demonstrate that 53BP1 does not localize to DSB foci in the ICM of preimplantation embryos or mESCs grown in culture (Ziegler-Birling et al., 2009; Kafer et al., 2016). In vivo, 53BP1 localizes to damage in E5.5 embryos, but only in the epiblast cells and not the endoderm or trophoblast cells (Laurent and Blasi, 2015). While Trp53bp1 expression is unchanged in early development, Rnf168 expression is limited prior to the epiblast stage which may explain differential engagement of DDR pathways in pre- and peri-implantation cells (Laurent and Blasi, 2015).

#### GENES ESSENTIAL FOR PREIMPLANTATION DEVELOPMENT

Consistent with a greater tolerance of genome instability in preimplantation cells, only 10 knockouts of genome stability factors that we reviewed were lethal during preimplantation (**Table 1** and **Supplementary Table S1**). Factors encoded by many of these preimplantation essential genes also possessed critical functions outside DDR or repair. Additionally, preimplantation embryos commonly survive to the morula stage even if a targeted gene is essential for viability. This is because embryonic genomes are not immediately active following fertilization while the inherited maternal mRNA guides protein translation (Jukam et al., 2017).

### Multifunctional Factors Essential for Preimplantation Development

Of the 10 genome stability factors required for preimplantation development, only Ercc2 deletion did not permit morula generation. Ercc2−/<sup>−</sup> embryos fail to develop past the 2 cell stage in vitro and there is no evidence of embryo implantation in vivo (de Boer et al., 1998). Ercc2 encodes the XPD helicase that functions both in NER and as a core component of the general transcription factor IIH (TFIIH) complex. TFIIH opens DNA and anchors the cdkactivating kinase (CAK) complex to facilitate transcription (Egly, 2001). Given that Ercc2 knockouts are compromised during embryonic genome activation (EGA), a reasonable conclusion is that preimplantation lethality is associated with a loss of XPD transcriptional activity. Likewise, Xab2 is a TFIIH and NER component and Xab2 deletion is lethal prior to the blastocyst stage (Yonemasu et al., 2005). However, Xab2 null embryos survive past the 2-cell stage suggesting Xab2 is not required for EGA.

Pcna and Rpa1 both function in multiple genome maintenance pathways and deleting either gene confers preimplantation lethality (Wang et al., 2005; Roa et al., 2008). PCNA (proliferating cell nuclear antigen) forms a homotrimeric sliding clamp that encircles DNA. The protein clamp travels with the replication fork to promote replicative polymerase processivity and mediate diverse functions in DNA replication and the replication stress response (Mailand et al., 2013). PCNA also functions in NER and MMR (Strzalka and Ziemienowicz, 2011). Rpa1 (Replication Protein A 1) codes

for the 70 kDa subunit of the heterotrimeric RPA ssDNA binding complex that participates in replication, but which also functions in ATR activation, replication fork repair, HDR and NER (Bhat and Cortez, 2018). Analysis of Pcna mRNA showed a significant reduction of maternal transcripts from the zygote to the 2-cell stage consistent with the observed early embryonic lethality (Hamatani et al., 2004; Roa et al., 2008; Bult et al., 2019). Rpa1 null embryos form blastocysts at E3.5 that are smaller than their heterozygous littermates and which fail in blastocyst outgrowth assays, indicating Rpa1 is required for trophectoderm and ICM growth (Wang et al., 2005). While both Rpa1 and Pcna function in diverse aspects of genome maintenance, both genes are also required for DNA replication. Preimplantation failure of Pcna or Rpa1 null embryos likely stems from an inability to polymerize nascent DNA coupled with simultaneous attenuation of multiple DNA repair pathways.

Two additional genes classified as DNA repair factors in the MGI-GO database are also essential during preimplantation. Nop53 encodes a nucleolar protein with suggested roles regulating p53, DNA repair, and the cellular response to mitochondrial stress (Lee et al., 2012; Yoon et al., 2014). Nop53 null embryos fail to develop past the blastocyst stage, while Nop53−/<sup>−</sup> mESCs are viable, suggesting that Nop53 functions in morula to blastocyst maturation (Sasaki et al., 2011). In somatic cells, Pot1a encodes a telomere specific ssDNA binding protein that suppresses ATR and p53 activation by the chromosome end (Hockemeyer et al., 2005, 2006; Denchi and de Lange, 2007). Loss of Pot1a likely confers robust ATR activation which explains the failure of Pot1a−/<sup>−</sup> embryos to form an ICM (Hockemeyer et al., 2006).

#### Cell Cycle and Checkpoint Factors Essential for Preimplantation Viability

In the event of replication stress or DSBs, checkpoints slow the cell cycle to provide time for DNA repair. Deletion of several genes that encode checkpoint regulatory factors confer preimplantation embryo failure at the morula stage. These include Wee1 (Tominaga et al., 2006), Cdk1 (Diril et al., 2012), and Plk1 (Lu et al., 2008; Wachowicz et al., 2016). CDK1 promotes the G2/M transition through interaction with cyclin B, and CDK1-cyclin B activity is regulated by WEE1 (Harvey et al., 2005; Gavet and Pines, 2010). ATR and CHK1 activate the G2/M checkpoint by stimulating WEE1 to inhibit CDK1 (O'Connell et al., 1997; Nigg, 2001). Polo-like kinase 1 (PLK1) controls numerous mitotic activities, participates in the G2/M transition, may function in DNA replication, and is implicated in HDR through phosphorylation of RAD51 (Watanabe et al., 2004; Mandal and Strebhardt, 2013; Yata et al., 2014; Pintard and Archambault, 2018). Dtl is also required for preimplantation development (Liu et al., 2007). Dtl encodes a component of the CUL4A-DDB1 E3-ubiquitin ligase complex that degrades the replication licensing factor CDT1 to stimulate replication and S phase progression (Nishitani et al., 2001; Li and Blow, 2005). Because these essential cell cycle genes function in DDR-regulated checkpoints and normal cell cycle transitions, the mechanism of embryonic lethality following their deletion likely involves deregulation of both the DDR and normal cell cycles.

#### GENES ESSENTIAL FOR PERI-IMPLANTATION DEVELOPMENT

Peri-implantation is the period of embryonic development most associated with lethality following deletion of genome stability factors. Within the MGI-GO derived gene list, it stands out that 54 of the 62 genes essential for periimplantation development have direct or indirect functions in DNA replication, the replication stress response, and/or HDR (**Table 2** and **Supplementary Table S1**). Below we focus specifically on these pathways.

### Genes Regulating Replication Initiation, DNA Polymerization, and the Replication Stress Response Are Essential for Peri-Implantation Development

Because of the rate of cell proliferation in peri-implantation embryos, it is perhaps unsurprising that factors which regulate the G1/S transition and initiate DNA replication are essential for peri-implantation development. This includes Cdc7 (Kim et al., 2002), Cdc45 (Yoshida et al., 2001), Cdk7 (Ganuza et al., 2012), Mnat1 (Rossi et al., 2001), Gins4 (Mohri et al., 2013), Cul4A (Li et al., 2002), and Ddb1 (Cang et al., 2006). Additionally, the Pold3 subunit of the replicative polymerase Polδ is required for peri-implantation development (Uchimura et al., 2009; Zhou et al., 2018), as is Fen1, which plays a central role in processing Okazaki fragments on the lagging strand of DNA synthesis (Larsen et al., 2003). Lethality following deletion of most of these genes occurs on or near E4.5 and is associated with proliferative failure within the ICM after blastocyst formation but prior to gastrulation. A likely explanation is that peri-implantation development is rapidly compromised when fundamental functions necessary for cellular proliferation and/or DNA replication are dysregulated.

Even in the absence of exogenous threats to replication, the exceptionally fast cell cycles within the developing epiblast are expected to confer intrinsic replication stress. For example, blastocyst derived mESCs display replication stress in unperturbed cultures in vitro (Ahuja et al., 2016). mESCs appear to manage this replication stress by effectively coupling replication and repair activities to facilitate near continuous DNA synthesis punctuated by brief G1 and M phases (Ahuja et al., 2016). Epiblast cell cycles in vivo are more rapid than cell cycles in cultured mESCs (Snow, 1977; Mohammed et al., 2017) suggesting that a similar reliance on efficient coupling of DNA replication and the repair activity is required to manage replication stress in peri-implantation embryos.

The replication stress response begins when ATR/ATRIP localizes to RPA-coated ssDNA, and ATR is activated through

TOPBP1 and the RAD9-RAD1-HUS1 (9-1-1) protein complex (Kumagai et al., 2006; Navadgi-Patil and Burgers, 2009; Choi et al., 2010). 9-1-1 is a sliding clamp loaded onto DNA (Bermudez et al., 2003). TOPBP1 bridges ATR/ATRIP and 9-1-1, thereby stabilizing ATR at the fork. Once stabilized at the site of replication stress, ATR regulates downstream replication stress repair and cell cycle arrest via CHK1 kinase and CDC25A (Liu et al., 2000; Xiao et al., 2003; Delacroix et al., 2007). TIMELESS, a component of the replication fork protection complex (Kemp et al., 2010; Couch et al., 2013; Buisson et al., 2017), and the ubiquitin ligase PRP19 (Prpf19) (Maréchal et al., 2014), also function in ATR and CHK1 activation. Embryos null for Atr (Brown and Baltimore, 2000; de Klein et al., 2000; Murga et al., 2009), Topbp1 (Jeon et al., 2011), Hus1 (Weiss et al., 2000), Chek1 (CHK1) (Takai et al., 2000; Zaugg et al., 2007), Cdc25A (Ray et al., 2007), Timeless (Gotter et al., 2000), and Prpf19 (Fortschegger et al., 2007) are all compromised during peri-implantation.

In somatic tissues, ATR or CHK1 inhibition coupled with replication stress induces replication catastrophe and S-phase apoptosis (Myers et al., 2009; Toledo et al., 2013; van Harten et al., 2019). Similarly, Atr null embryos form blastocysts that in vitro display widespread apoptosis within the ICM but not in the trophoblast, and which are eventually compromised by day three of culture (Brown and Baltimore, 2000; de Klein et al., 2000; Murga et al., 2009). ATR activity is therefore essential in the pluripotent and rapidly proliferating epiblast cells that will become the embryo proper, but not in the multipotent placental precursors. Chek1 null embryos also form blastocysts, but unlike Atr null embryos, are unable to attach to the culture vessels in outgrowth experiments (Takai et al., 2000). The more severe embryonic response to a loss of CHK1 activity compared to ATR inhibition is consistent with observations in somatic cells (Buisson et al., 2015).

## HDR Genes Are Essential for Peri-Implantation Development

Common HDR factors manipulate DNA substrates at twoended DNA breaks and stressed replication forks (Ait Saada et al., 2018). During replication stress, this entails regressing and protecting stalled forks in 4-way DNA structures to facilitate repair (Quinet et al., 2017). HDR factors essential for peri-implantation that function in canonical DSB repair and replication fork protection include: Brca1 (Hakem et al., 1996; Ludwig et al., 1997; Taglialatela et al., 2017), Brca2 (Sharan et al., 1997; Lemaçon et al., 2017; Mijic et al., 2017), Rad51 (Lim and Hasty, 1996; Tsuzuki et al., 1996; Zellweger et al., 2015), Bard1 (McCarthy et al., 2003; Shakya et al., 2008; Daza-Martin et al., 2019), Palb2 (Bowman-Colin et al., 2013), and Rad51C (Kuznetsov et al., 2009; Somyajit et al., 2015). Similarly, common factors promote resection during HDR repair of canonical two-ended DSBs and at stalled replication forks (Ait Saada et al., 2018). This includes the essential peri-implantation genes: Ctip (Chen et al., 2005; Przetocka et al., 2018); Ptip (Cho et al., 2003; Ray Chaudhuri et al., 2016); Dna2 (Lin et al., 2013; Thangavel et al., 2015), and the MRN complex [Mre11, Rad50, and Nbs1 (Xiao and Weaver, 1997; Luo et al., 1999; Zhu et al., 2001; Theunissen et al., 2003; Buis et al., 2008; Schlacher et al., 2011).

Homology directed repair activity in replication fork remodeling and protection is an emerging topic in the DNA repair field, and it remains unclear which HDR functions are essential for cell viability. However, lethality in Brca2 null mESCs is rescued when fork protection is facilitated by inhibiting PARP activity to protect replication forks from MRE11 nuclease (Ding et al., 2016). For BRCA2, fork protection may be the essential function. In vivo, deleting factors that participate in both canonical HDR-dependent DSB repair and fork remodeling or resection, typically confers reduced cell proliferation and embryo failure during gastrulation. For example, Brca1−/<sup>−</sup> embryos are severely malformed with an underdeveloped pro-amniotic cavity and mesoderm (Hakem et al., 1996; Ludwig et al., 1997). Concordantly, Brca1 null embryos display a significant reduction of DNA synthesis by E6.5, increased p21 [a factor induced by p53 transactivation to mediate cell cycle arrest (Georgakilas et al., 2017)], and a high incidence of G1 arrested cells (Hakem et al., 1996). Similarly, Rad51 null embryos suffer from reduced proliferation and begin to degenerate during gastrulation by E7.5, but development can be extended to E9.5 in TP53/Rad51 double knockouts (Lim and Hasty, 1996). Loss of HDR factors therefore appears to primarily confer proliferative failure through p53-induced growth arrest and subsequently through accumulated genome instability if p53 function is compromised.

## Additional Peri-Implantation Genes Linked to HDR or Replication

Peri-implantation development additionally requires several genes that are functionally linked to HDR or replication stress mitigation. Such genes include Apex1 (Xanthoudakis et al., 1996), Rnaseh2B (Hiller et al., 2012), Rif1 (Chapman et al., 2013), Blm (Chester et al., 1998), Rev3L (Bemark et al., 2000), Usp7 (Kon et al., 2010), and Ino80 (Min et al., 2013). Apex1 does not directly function in replication, but is critical for the repair of abasic DNA sites that if left unresolved confer replication stress (Boiteux and Guillet, 2004). Similarly, Rnaseh2B resolves RNA within RNA-DNA hybrids, including mis-incorporated ribonucleotides, which if not resolved can impede replication (Pizzi et al., 2015). In addition to functioning in DSB repair pathway choice, Rif1 also coordinates replication timing (Yamazaki et al., 2012; Foti et al., 2016). Blm encodes a helicase that functions in homologous recombination, the replication stress response (Davies et al., 2007; Machwe et al., 2011), promotes replication at telomeres (Barefield and Karlseder, 2012), and participates in a specialized form of break induced replication at chromosome ends (Sobinoff et al., 2017). Rev3L encodes the catalytic subunit of POL ζ which facilitates translesion synthesis to enable replication forks to negotiate lesions within DNA without stalling (Bhat et al., 2013).

Usp7 is a deubiquitylating enzyme that promotes efficient DNA replication (Lecona et al., 2016) and Ino80 encodes a chromatin remodeling factor that functions in both replication and HDR (Poli et al., 2017).

## MAINTAINING GENOME STABILITY DURING PERI-IMPLANTATION DEVELOPMENT

While HDR and replication associated genes are not the only genome stability factors required for peri-implantation development, their predominance on the list of genes essential for this developmental window is compelling. Conditions enabling in vitro culture of peri-implantation murine epiblast stem cells are now established (Brons et al., 2007; Tesar et al., 2007). However, there are no published studies focusing on genome maintenance in isolated epiblast cells. It is thus necessary to extrapolate from studies of other rapidly dividing cells why HDR and replication factors may be essential during periimplantation development.

### Replication Stress in Peri-Implantation Cells

The fastest cell cycles in the lifetime of a mouse are likely during the brief window of epiblast development. Outside of the embryo, rapid proliferation contributes to replication stress in hematopoietic and cancerous tissues by robbing cells of the time to properly execute DNA polymerization (Pilzecker et al., 2017). For example, during a somatic G1-phase replication is facilitated by increasing the abundance of MCM proteins and loading these factors at replication origins in preparation for S-phase (Boos et al., 2012). Late G1 cells also upregulate dNTP production for the oncoming burst of DNA synthesis (Hirschi et al., 2010; Dick and Rubin, 2013). When origins are not established properly, or dNTP production is dysregulated, cancer and aging hematopoietic cells are susceptible to replication stress (Flach et al., 2014; Aird and Zhang, 2015; Alvarez et al., 2015; Garzon et al., 2017).

Pluripotent cells in the early embryo display almost nonexistent G1 phases and appear to use unique mechanisms to ensure S-phase progression is not impeded. For example, human ESCs counter G1 brevity by loading MCM proteins faster than cultured non-pluripotent cells (Matson et al., 2017). Retinoblastoma protein is also constitutively hyperphosphorylated in mouse and human ESCs (Conklin et al., 2012; Ter Huurne et al., 2017), which maintains high E2F transactivation to promote unperturbed dNTP synthesis (Hirschi et al., 2010; Dick and Rubin, 2013). To ensure rapid S-phase progression, mouse ESCs also license more dormant origins than non-embryonic progenitor cells (Ge et al., 2015). This increases the ability of ESCs to complete replication during a short S phase (Courtot et al., 2018). ESCs also demonstrate an increased ability to restart stalled replication forks (Ahuja et al., 2016; Zhao et al., 2018). Human induced pluripotent stem cells (iPSC) also license replication origins rapidly in their abbreviated G1 phase (Matson et al., 2017), and dormant origins are critical to maintain rapid proliferation in hematopoietic stem cell pools (Alvarez et al., 2015). Although not yet directly tested, we anticipate that epiblast cells employ similar countermeasures to expedite DNA synthesis.

Despite these efforts, human and mouse ESCs display endogenous replication stress in unperturbed cultures (Ahuja et al., 2016; Lamm et al., 2016). It is not clear why replication is stressed in pluripotent cells, however in proliferating cancer cells transcription commonly drives replication stress through collisions between RNA and DNA polymerases, or persistent RNA/DNA hybrid molecules termed R-loops (Crossley et al., 2019). Primary somatic cells mitigate transcription and R-loop interference by temporally coordinating transcription and replication during S-phase (Meryet-Figuiere et al., 2014). It is probable that peri-implantation epiblast cells facilitate rapid genome duplication by simultaneously engaging large numbers of origins. Such activity could increase the probability of replisome and transcriptome collisions and/or R-loop induced replication stress.

Under-replication due to endogenous stress is described in mESCs and human iPSCs (Ahuja et al., 2016; Vallabhaneni et al., 2018). Lengthening the G1 phase reduces replication defects in mESCs consistent with the notion that replication stress is driven by rushed cell cycles (Ahuja et al., 2016). Pluripotent cells are reported to have enhanced DNA repair capacities and it is possible this translates to highly effectively management of inefficient replication in embryos (Maynard et al., 2008; Luo et al., 2012). Alternatively, pluripotent blastocyst cells persist for a limited number of cell divisions in vivo, and it is feasible that pre-implantation embryos simply tolerate under-replication for this brief period. Tolerance of replication defects is consistent with observations that preimplantation embryonic cells are more resistant to genomic insult than periimplantation cells. Compromising the replication stress response should confer similar DDR activation, chromosomal segregation errors, and aneuploidy in both pre- and peri-implantation cells. However, cells with supernumerary chromosomes persist in the blastocyst before elimination during peri-implantation (Bolton et al., 2016). This may explain why many genome stability factors become essential specifically during peri-implantation development. In vivo, activating p53 through deletion of its negative regulator Mdm2 leads to lethality during periimplantation (Jones et al., 1995). This suggests that p53 signaling during peri-implantation is a conduit to remove genomically unstable cells in vivo.

Additionally, it is important to recognize that genome stability factors may possesses cryptic essential functions during early development. For example, PARP1 is DDR factor involved in multiple aspects of genome integrity (Ray Chaudhuri and Nussenzweig, 2017). However, in mESCs, Parp1 has roles preventing the trans-differentiation of extraembryonic trophoblast cells (Nozaki et al., 1999; Hemberger et al., 2003). This pluripotency-specific function is linked to the DNA binding ability of PARP1 which confers an epigenetic-like regulation of pluripotency (Roper et al., 2014). Additionally, pluripotency-specific repair factors may provide essential

functions in early development. The Filia-Floped protein complex is abundantly expressed in pluripotent but not somatic cells, and is suggested to increase the abundance of essential repair factors including BLM and promote ATR activation to encourage fork restart (Zhao et al., 2015; Zhao et al., 2018; Zheng, 2020).

### Essential Roles for HDR Factors During Peri-Implantation

It is possible the canonical G1-phase c-NHEJ DSB repair pathway is dispensable during peri-implantation because cells rapidly transition through their brief G1 phase and engage alt-NHEJ or HDR in S-phase. Conversely, c- and alt-NHEJ remain active in HDR-deficient embryos, and in principle EJ functions could provide redundant DSB repair activity. An explanation why c- or alt-NHEJ cannot rescue HDR during peri-implantation may stem from the diversity of substrates repaired by HDR. While twoended DSBs are readily repaired by EJ, collapsed replication forks present as one-ended DNA breaks which c- or alt-NHEJ cannot process (Feng and Jasin, 2017; Scully et al., 2019). One-ended breaks, however, can be repaired through a specialized HDR mechanism termed break-induced replication (BIR). During BIR, the exposed DNA end is resected, and following strand-invasion to form a displacement loop on the sister chromatid, DNA is polymerized from the invaded strand in a conservative manner (Kramara et al., 2018). While BIR is commonly studied in yeast, the analogous mechanism in vertebrates remains poorly defined. Notwithstanding, Blm promotes BIR at chromosome ends in human cancer cells that maintain telomere length through the alternative lengthening of telomeres mechanism suggesting that Blm may promote BIR elsewhere in the genome following replication stress (Sobinoff et al., 2017). Additionally, Pold3 is also implicated in vertebrate BIR (Costantino et al., 2014). Conversely, the non-essential gene Rad52 also functions in vertebrate BIR casting doubt on the requirement for BIR in early development and indicating the need for future studies (Sotiriou et al., 2016).

Multiple HDR factors essential for peri-implantation also play central roles in replication stress management. This includes HDR factors that regress, stabilize, or resect stalled replication forks including Brca1/2, Bard1, Rad51 and Rad51C, Ctip, Ptip, Dna2, and genes encoding the MRN complex. Interestingly, the helicases and translocases that promote fork regression, including Smarcal1, are not essential for periimplantation (Baradaran-Heravi et al., 2012). This may reflect redundancy in the mechanisms that drive formation of 4 strand structures at stalled replication forks (Quinet et al., 2017). Given the essentiality of the replication stress response for epiblast development, it stands to reason that replication fork remodeling and protective functions encoded by HDR factors play a role in the viability of peri-implantation embryos. Failure to facilitate these activities would confer genome instability, chromosome segregation errors, and molecular outcomes consistent with unrepaired replication stress. As described above, this would activate p53 surveillance systems during peri-implantation. In the future it will be interesting to determine which specific HDR functions are essential for embryogenesis.

### CONCLUSION AND FUTURE DIRECTIONS

While most DDR and repair factors are non-essential, this does not exclude non-essential genome stability pathways from playing an important role in healthy development. Deletion of non-essential DDR and repair genes commonly results in the birth of live pups exhibiting a wide array of deleterious phenotypes (**Supplementary Table S1**). Additionally, cancer predisposition and a reduced lifespan are common in mice lacking non-essential genome stability factors (**Supplementary Table S1**). The difference between essential and non-essential genome stability genes is likely derived from the necessity of the targeted repair pathway to resolve lethal genome instability that arises during a specific window of development.

The critical genome stability pathways required for cell survival are determined by the type of genomic lesions the cell encounters and when in the cell cycle those lesions arise. Early embryonic development is largely protected from exogenous influence. Preimplantation embryos are wholly contained, and maternal blood supply begins at E9.5 (Behringer et al., 2014). The critical early stages of development likely benefit from sequestration from external threats to genome stability. We suggest the imminent threat to genome stability in early development stems from the need to mitigate endogenous replication stress within the epiblast to sustain rapid cell cycles. If left unresolved, replication stress drives genome instability, chromosome segregation errors, growth arrest and/or cell death. Dwindling cell numbers of epiblast cells that fail to effectively replicate their genome will progressively lead to embryo compromise. This premise is supported by the timing of embryo failure associated with deletion of essential DDR and repair genes. Consistent with systemic proliferative failure, deletion of factors required for basal DNA replication, or up-stream signaling in the replication stress response, typically induce an early embryonic demise (E5.0 or before). Whereas deleting HDR factors induces death at a subsequent time (E5.5 or later), potentially as the additive outcome of progressive genome instability.

Another caveat of reports reviewed here is that many were pioneering studies that utilized the newly developed technology of gene targeting in mice. In the intervening decades, our understanding of early development and experimental capability has blossomed. Development of new stem cell culture technologies now enables mechanistic study of peri-implantation development within three-dimensional gastrula structures in vitro. Co-culture of ICM derived ESCs, trophectoderm derived trophoblast stem cells, and extraembryonic endoderm stem cells results in the spontaneous self-assembly of a structure remarkably like the gastrulating embryo. These structures, termed "ETX embryos" will enable directed gene deletion within specific embryonic cell types (Sozen et al., 2018). Coupled with precise mechanistic investigations, ETX embryo models will reveal new insights on the mechanisms of cell compromise

during embryogenesis. It will be exciting to learn in the coming years why a limited number of DDR and repair pathways are essential for development, and the underlying reasons for their importance within the peri-implantation embryo.

#### DATA AVAILABILITY STATEMENT

MGI-GO designations discussed in this review can be found at the following. DNA damage checkpoint: http://www.informatics. jax.org/go/term/GO:0000077; nucleotide excision repair: http:// www.informatics.jax.org/go/term/GO:0006289; mismatch repair: http://www.informatics.jax.org/go/term/GO:0006298; base excision repair: http://www.informatics.jax.org/go/term/GO: 0006284; DSB repair via homologous recombination: http:// www.informatics.jax.org/go/term/GO:0000724; DSB repair via non-homologous end joining: http://www.informatics.jax.org/ go/term/GO:0006303.

### AUTHOR CONTRIBUTIONS

GK and AC conceived of the review manuscript. GK compiled the information presented in **Tables 1**, **2** and **Supplementary Table**

#### REFERENCES


**S1** from the MGI-GO database as indicated. GK made the figures. GK and AC composed the manuscript.

#### FUNDING

The Cesare laboratory is supported by grants from the Australian National Health and Medical Research Council (1162886 and 1185870) and The Neil and Norma Hill Foundation.

#### ACKNOWLEDGMENTS

The authors thank members of the Cesare and Tam laboratories at the Children's Medical Research Institute for their helpful discussion.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2020.00416/ full#supplementary-material


in vitro. Proc. Natl. Acad. Sci. U.S.A. 100, 1633–1638. doi: 10.1073/pnas. 0437927100








**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 Kafer and Cesare. 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.

# Ppp2r2a Knockout Mice Reveal That Protein Phosphatase 2A Regulatory Subunit, PP2A-B55α, Is an Essential Regulator of Neuronal and Epidermal Embryonic Development

Nikita Panicker1,2, Melody Coutman1,2, Charley Lawlor-O'Neill1,2, Richard G. S. Kahl1,2 , Séverine Roselli1,2† and Nicole M. Verrills1,2 \* †

<sup>1</sup> School of Biomedical Sciences and Pharmacy, Faculty of Health and Medicine, Priority Research Centre for Cancer Research, Innovation and Translation, University of Newcastle, Callaghan, NSW, Australia, <sup>2</sup> Hunter Cancer Research Alliance, Hunter Medical Research Institute, New Lambton, NSW, Australia

#### Edited by:

Andrew Burgess, ANZAC Research Institute, Australia

#### Reviewed by:

Joshua Oaks, Lynx Biosciences, Inc., United States Veerle Janssens, KU Leuven, Belgium

#### \*Correspondence:

Nicole M. Verrills nikki.verrills@newcastle.edu.au

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 17 February 2020 Accepted: 22 April 2020 Published: 05 June 2020

#### Citation:

Panicker N, Coutman M, Lawlor-O'Neill C, Kahl RGS, Roselli S and Verrills NM (2020) Ppp2r2a Knockout Mice Reveal That Protein Phosphatase 2A Regulatory Subunit, PP2A-B55α, Is an Essential Regulator of Neuronal and Epidermal Embryonic Development. Front. Cell Dev. Biol. 8:358. doi: 10.3389/fcell.2020.00358 The serine/threonine protein phosphatase 2A (PP2A) is a master regulator of the complex cellular signaling that occurs during all stages of mammalian development. PP2A is composed of a catalytic, a structural, and regulatory subunit, for which there are multiple isoforms. The association of specific regulatory subunits determines substrate specificity and localization of phosphatase activity, however, the precise role of each regulatory subunit in development is not known. Here we report the generation of the first knockout mouse for the Ppp2r2a gene, encoding the PP2A-B55α regulatory subunit, using CRISPR/Cas9. Heterozygous animals developed and grew as normal, however, homozygous knockout mice were not viable. Analysis of embryos at different developmental stages found a normal Mendelian ratio of Ppp2r2a−/<sup>−</sup> embryos at embryonic day (E) 10.5 (25%), but reduced Ppp2r2a−/<sup>−</sup> embryos at E14.5 (18%), and further reduced at E18.5 (10%). No live Ppp2r2a−/<sup>−</sup> pups were observed at birth. Ppp2r2a−/<sup>−</sup> embryos were significantly smaller than wild-type or heterozygous littermates and displayed a variety of neural defects such as exencephaly, spina bifida, and cranial vault collapse, as well as syndactyly and severe epidermal defects; all processes driven by growth and differentiation of the ectoderm. Ppp2r2a−/<sup>−</sup> embryos had incomplete epidermal barrier acquisition, associated with thin, poorly differentiated stratified epithelium with weak attachment to the underlying dermis. The basal keratinocytes in Ppp2r2a−/<sup>−</sup> embryos were highly disorganized, with reduced immunolabeling of integrins and basement membrane proteins, suggesting impaired focal adhesion and hemidesmosome assembly. The spinous and granular layers were thinner in the Ppp2r2a−/<sup>−</sup> embryos, with aberrant expression of adherens and tight junction associated proteins. The overlying stratum corneum was either absent or incomplete. Thus PP2A-B55α is an essential regulator of epidermal stratification, and is essential for ectodermal development during embryogenesis.

Keywords: PP2A, knockout mouse, embryo, lethality, skin, epidermal barrier

## INTRODUCTION

fcell-08-00358 June 4, 2020 Time: 15:47 # 2

Mammalian development is a highly complex process requiring exquisite regulation of cellular growth, differentiation and morphogenesis. The signaling pathways controlling these processes are mediated, in large part, by phosphorylation cascades. Protein phosphatases control the rate and duration of these signals, yet our understanding of the functional roles of protein phosphatases in mammalian development is poor. Protein phosphatase 2A (PP2A) is a ubiquitously expressed serine-threonine phosphatase that has key roles in growth, differentiation and morphogenesis (Olsen et al., 2006; Moorhead et al., 2007). Serine and threonine residues dominate the human phosphoproteome, together representing 98% of phosphorylation sites identified. Together with protein phosphatase 1 (PP1), PP2A is responsible for over 90% of Ser/Thr de-phosphorylation in most cells (Eichhorn et al., 2009). Dysregulated PP2A activity is associated with numerous diseases, including, but not limited to, Alzheimer's disease, various cancers, diabetes, cardiac disease, asthma, inflammation and auto-immune conditions (Calin et al., 2000; Takagi et al., 2000; Suzuki and Takahashi, 2003; Neviani et al., 2005; Kowluru and Matti, 2012; Collison et al., 2013; Crispin et al., 2013; Sontag and Sontag, 2014; Lei et al., 2015; Goldsworthy et al., 2016; Ross et al., 2017; Refaey et al., 2019).

Protein phosphatase 2A is a complex family of enzymes, composed of structural (A), catalytic (C) and regulatory (B) subunits. The PP2A core dimer consists of a structural and catalytic subunit, of which there are two isoforms (α and β) expressed in mammals (Hemmings et al., 1990). The tissue and substrate specificity of PP2A activity is mediated by binding of one of the four families of regulatory B-subunits: B55, B56, B <sup>00</sup>, and B<sup>000</sup> (each with multiple isoforms) to the core dimers. This results in over 80 different potential PP2A configurations (Smith A.S. et al., 2011), which can target a multitude of largely mutually exclusive substrate proteins (e.g., AKT, ERK, cJun, etc.), in diverse signaling pathways, including DNA damage repair, cell cycle progression and mitosis, proliferation, apoptosis and metabolism. Most regulatory B-subunits are ubiquitously expressed, but some demonstrate tissue specificity. For example, members of the B56 family are highly expressed in almost all major organs and tissues (Reynhout and Janssens, 2019). In contrast, the B<sup>00</sup> family generally shows low expression in most tissues, with the exception of high B00α in the heart and high B <sup>00</sup>γ in embryonic brain. The B55 family is highly expressed in the brain, with B55γ almost exclusively expressed in embryonic and adult brain, and other family members to a lower extent. The B55α subunit shows the widest expression pattern within the B55 family (Reynhout and Janssens, 2019). The specific substrates and the functional roles of each regulatory subunit in health and disease, however, have not been well studied.

In vivo studies, using knockout mouse models for genes encoding the catalytic and structural subunits, have revealed a vital role for PP2A in embryonic development [recently reviewed in Reynhout and Janssens (2019)]. Knockout of PP2A-Cα (Ppp2ca) causes lethality at embryonic day (E) 6.5 (Götz et al., 1998), and knockout of PP2A-Aα (Ppp2r1a) results in lethality before E10.5, implying no functional redundancy with the β isoforms of either the A or C subunits in embryonic development, despite high sequence similarity (Ruediger et al., 2011). Of the regulatory B-subunits, knockout of three members have been reported. In the case of B56δ, there are opposing reports, with one study finding homozygous Ppp2r5d−/<sup>−</sup> animals were viable (Louis et al., 2011), and another that mice heterozygous for a strongly hypomorphic Ppp2r5d gene-trap allele were viable, but no homozygotes were recovered (Kapfhamer et al., 2010). Constitutive homozygous deletion of Ppp2r5c (B56γ) resulted in neonatal death at post-natal (P) day 1–2 due to heart defects (Varadkar et al., 2014); and gene-trap mediated constitutive Ppp2r5a (B56α) knockout mice were viable, however, displayed heart and nerve defects (Little et al., 2015), skin lesions, hyperproliferation of the epidermis and hair follicles, and increased hematopoiesis (Janghorban et al., 2017). Thus, specific PP2A B56 subunits play diverse roles in mammalian development. To date, however, the role of B55 subunits in mammalian development is not known.

The PP2A-B55α subunit (encoded by the PPP2R2A gene) is expressed in all tissues, with highest levels observed in the embryonic central nervous system and limbs, as well as adult brain, bladder, adrenal glands, ovaries and placenta (Reynhout and Janssens, 2019). Inactivation or genetic loss of B55α has been implicated in human diseases including cancer and Alzheimer's disease (Sontag et al., 2004; Ruvolo et al., 2011; Kalev et al., 2012; Beca et al., 2015; Watt et al., 2017). Therefore understanding the functional role of B55α in normal physiology is of particular interest. In cellular models, PP2A-B55α complexes have been shown to exert positive regulation of the ERK/MAPK pathway, but negative regulation of the PI3K/AKT pathway (Smith A.M. et al., 2011). PP2A-AB55αC complexes have also been implicated in DNA damage repair pathways by dephosphorylating ATM; cell cycle regulation by dephosphorylation and activation of the retinoblastoma-related protein p107; control of mitotic exit by deactivation of Cyclin-dependent kinase B-Cdk1 and many of its target proteins (Burgess et al., 2010; Gharbi-Ayachi et al., 2010; Manchado et al., 2010; Mochida et al., 2010; Schmitz et al., 2010; Cundell et al., 2013); and cell adhesion and migration by dephosphorylation of Rac1 and AP-1 (Gilan et al., 2015; Bousquet et al., 2016). PP2A-B55α has emerged as a tumor-suppressor in many epithelial and blood cancers. The PPP2R2A gene is commonly deleted in human breast (Curtis et al., 2012) and prostate tumors (Cheng et al., 2011), and PPP2R2A knockdown in breast cancer cell lines increases tumorigenicity (Watt et al., 2017). PPP2R2A is also commonly down-regulated in non-small cell lung carcinomas (Kalev et al., 2012). A recent ENU-induced mutagenesis study reported a splice-site mutation in Ppp2r2a resulting in reduced PP2A-B55α expression. Generation of double heterozygous mice for this Ppp2r2a mutation and a null allele of the gene encoding the insulin receptor, resulted in a diabetic phenotype characterized by hyperglycemia, hyperinsulinemia, impaired glucose tolerance, and glycosuria (Goldsworthy et al., 2016), suggesting PP2A-B55α may play a role in metabolism and insulin signaling.

PP2A-B55α dephosphorylates β-catenin during Wnt signaling, suggesting it may play a role in development (Zhang et al., 2009). In support of this, ex vivo knockdown of PP2A-B55α suggested an essential role for this subunit in mouse oocyte maturation (Liang et al., 2017) and in epidermal barrier formation (O'Shaughnessy et al., 2009). To investigate the function of this subunit in vivo, we have generated the first constitutive Ppp2r2a knockout mouse using CRISPR/Cas9. Homozygous Ppp2r2a deletion resulted in embryonic lethality post E10.5. Ppp2r2a−/<sup>−</sup> embryos displayed neural tube defects, limb defects and impaired epidermal barrier formation. The latter was associated with aberrant polarization of basal keratinocytes and aberrant keratinocyte differentiation. Therefore PP2A-B55α has essential roles in embryonic development, in particular in epidermal barrier formation.

### MATERIALS AND METHODS

#### Generation of Ppp2r2a Knockout Mice

Ppp2r2a knockout mice were generated at the Australian Phenomics Network (Monash University, Melbourne, Australia), using the CRISPR/Cas9 technology. The CRISPR Design site http://crispr.mit.edu/ was used to identify guide RNA target sites flanking exon ENSMUSE00000482200 (exon 4) of the Ppp2r2a gene. The following guide RNAs were used: – 5<sup>0</sup> TACGATAAAGCAGCCTAGTT 3<sup>0</sup> for the 5<sup>0</sup> end of exon 4, and – 5<sup>0</sup> TTTGCTTTCAGGTACTACAT 3<sup>0</sup> for the 3<sup>0</sup> end of exon 4. Complementary oligonucleotides corresponding to the RNA guide target sites were annealed and cloned into BbsI (NEB) digested plasmid pX330-U6-Chimeric\_BB-CBh-hSpCas9 (Addgene plasmid #42230). Single guide RNAs (sgRNA) were generated using the HiScribeTM T7 Quick High Yield RNA Synthesis Kit (NEB) according to the manufacturer's instructions. The sgRNAs were purified using the RNeasy Mini Kit (Qiagen). Cas9 mRNA (30 ng/mL, Sigma) and the sgRNAs (15 ng/mL) were microinjected into the cytoplasm of C57BL/6J zygotes at the pronuclei stage. Injected zygotes were transferred into the uterus of pseudopregnant F1 (C57BL/6 × CBA) females. Forward (5<sup>0</sup> GTGTTCCAGCCAGCTGTTTCT 3<sup>0</sup> ) and reverse (5<sup>0</sup> GACACTGCTGCCTATGTCTGCT 3<sup>0</sup> ) genotyping PCR primers flanking the targeted region and amplifying a product of 819 bp from the wild-type DNA were used to characterize gene editing events in the resulting mice.

## Animals and Genotyping

Animals were used in accordance with the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes and all protocols were approved by the University of Newcastle Animal Care and Ethics Committee. Mice were housed in individually ventilated cages on a 12 h light/dark cycle and fed standard chow ad libitum. Timed matings between heterozygous breeding pairs were set up with midday the next day designated as 0.5 days post coitus (dpc), and pregnant mice identified via the presence of a vaginal plug. Mice were euthanased using CO<sup>2</sup> asphyxiation, and embryos harvested at 10.5 days of development (E10.5), E14.5 and E18.5. Pregnant females were monitored and the day of birth of the pups was designated P0. Euthanasia of E18.5 embryos and neonates was performed by decapitation. Genotyping was performed by PCR on DNA extracted from tail tips, using the REDExtract-N-AmpTM Tissue PCR Kit (Sigma-Aldrich #XNAT) as per the manufacturer's recommendations. The same PCR primers flanking exon 4 and used to characterize the gene editing events (described above) were used.

#### Protein Extraction and Immunoblotting

Protein extraction was conducted by grinding whole embryos to a fine powder in liquid nitrogen, using a mortar and pestle, and proteins solubilized using RIPA lysis buffer (0.05M Hepes pH 7.4, 1% Triton X-100, 0.1% SDS, 50 mM Sodium Fluoride, 0.05M EDTA, 1 mM Sodium Orthovanadate, 2.5% Protease Inhibitor Cocktail (Sigma #P8340), 5% Sodium Deoxycholate). Protein was quantitated using a BCA assay, separated using 4–12% Bis-Tris SDS-PAGE and transferred to nitrocellulose. Immunoblotting was performed using a commercial anti-PP2A-B55α rabbit polyclonal (Cell Signaling Technology #4953); an inhouse rabbit polyclonal antibody raised against the B55α peptide FSQVKGAVDDDVAE (residues 14–27) (Strack et al., 1998); a polyclonal rabbit anti-PP2A-A (Merck Millipore #07-250); a mouse monoclonal anti-PP2A-C (Merck Millipore #05-421); and anti-rabbit/mouse HRP-conjugated secondary antibodies. A HRP-conjugated anti-actin antibody (Sigma #A3854) was used as a loading control. Images were captured on a Biorad ChemiDocTM Imaging system using ECL.

### RNA Isolation, cDNA Synthesis and qPCR

Total RNA was prepared from embryonic tissue using the Isolate II RNA Mini Kit (Bioline #BIO-52072) and 1 µg was used to generate cDNA using the SensiFASTTM cDNA Synthesis Kit (Bioline #BIO-65053) according to manufacturer's instructions. Gene expression analysis was conducted using the SensiFASTTM SYBR <sup>R</sup> Hi-ROX Kit (Bioline #BIO-92020) and detected using an Applied Biosystems 7900HT Fast Real-Time PCR system. The cycling conditions used were 95◦C for 2 min, followed by 40 cycles of 95◦C for 15 s and 60◦C for 1 min, and a final melt curve analysis at 95◦C for 15 s, 60◦C for 15 s and 95◦C for 15 s. Primers were used at a final concentration of 0.3 µM. The primers used were Ppp2r1a (forward 5<sup>0</sup> ACTCTTCTGCATCAATGTGTT 3<sup>0</sup> , reverse 5<sup>0</sup> ATACGAAGAACTGTGGGCA 3<sup>0</sup> ), Ppp2r1b (forward 5 <sup>0</sup> GCTTCAGATGAACAGGACTCT 3<sup>0</sup> , reverse 5<sup>0</sup> AGACAGT AACTGGGCAATGC 3<sup>0</sup> ), Ppp2ca (forward 5<sup>0</sup> CTTGTAGCTCT TAAGGTTCG 3<sup>0</sup> , reverse 5<sup>0</sup> TCTGCTCTCATGATTCCCTC 3<sup>0</sup> ), Ppp2cb (forward 5<sup>0</sup> TTCTTGTAGCATTAAAGGTGCG 3<sup>0</sup> , reverse 5<sup>0</sup> TCCATACTTCCGTAGGCAC 3<sup>0</sup> ), Ppp2r2a (forward 5 <sup>0</sup> CCGTGGAGACATACCAGGTA 3<sup>0</sup> , reverse 5<sup>0</sup> AACACTGT CAGACCCATTCC 3<sup>0</sup> ), Ppp2r2b (forward 5<sup>0</sup> GGACCTCAAC ATGGAAAATC 3<sup>0</sup> , reverse 5<sup>0</sup> CGCTGTCTGACCCATTCCAT 3 0 ), Ppp2r2c (forward 5<sup>0</sup> AGCGGGAACCAGAGAGTAAG 3<sup>0</sup> , reverse 5<sup>0</sup> GTAGTCAAACTCCGGCTCG 3<sup>0</sup> ), Ppp2r2d (forward 5 <sup>0</sup> TTACGGCACTACGGGTTCCA 3<sup>0</sup> , reverse 5<sup>0</sup> TTCGTCGT GGACTTGCTTCT 3<sup>0</sup> ), Rpl19 (forward 5<sup>0</sup> CTCAGGAGATACC GGGAATCCAAGAAGA 3<sup>0</sup> , reverse 5<sup>0</sup> CACATTCCCTTTGAC

CTTCAG 3<sup>0</sup> ). Genes of interest were normalized to the included housekeeping gene Rpl19, and relative gene expression was quantified using the comparative cycle (Ct) method (2−11Ct) relative to wildtype.

### Histology and Immunohistochemistry

Embryos fixed in neutral buffered formalin (NBF) at various stages (E10.5, E14.5, and E18.5) were embedded in paraffin and sectioned at 5 µm before subsequent hematoxylin and eosin (H&E) staining or immunohistochemistry (IHC). Occasionally for H&E only, embryos were fixed in Bouins (**Figure 4A** E14.5 embryos). The H&E stained sections were analyzed by a veterinary anatomical pathologist at APN for stage-specific developmental features. IHC was performed as previously described (Pundavela et al., 2015). Briefly, following deparaffinization and rehydration steps using standard procedures, heat induced epitope retrieval (HIER) was carried out in a low pH, citrate-based antigen unmasking solution (Vector Laboratories, Burlingame, CA, United States, #H-3300) using a decloaking chamber (Biocare, West Midlands, United Kingdom) at 105◦C for 5 min. After quenching of endogenous peroxidases in 0.3% H2O2, primary antibodies were prepared in 1% BSA, 0.1% Tween-20 in phosphate-buffered saline (PBS), and applied to the sections. Primary antibodies were rabbit polyclonal anti-cleaved caspase 3 (1:200; Cell Signaling Technology #9661), rabbit monoclonal anti-Ki67 [SP6] (1:100; Abcam, ab16667) and rabbit monoclonal p-cJun (1:100; Cell Signaling Technology #2361). The ImmPRESS (Peroxidase) Polymer secondary antibody kit and ImmPACT DAB (Vector Laboratories) were used for detection of the primary antibodies as per the manufacturer's recommendations and the sections were counterstained with hematoxylin, dehydrated in a series of ethanol washes, and cleared in xylene before mounting with Ultramount No. 4 (Fronine #FNNII065C). Slides were scanned using a Leica Biosystems Aperio AT2 Scanner at 20× magnification.

#### Immunofluorescence

Embryos were fixed overnight in 4% paraformaldehyde and cryopreserved through a sucrose gradient (10 and 30%) before freezing in O.C.T. Compound (Tissue-Tek <sup>R</sup> ; Proscitech IA018) using isopentane cooled down in dry ice. Eight µm thick frozen sections were prepared using a Leica cryostat and immunofluorescence experiments were performed at room temperature, using a slightly modified procedure to previously described (Naudin et al., 2017). Briefly, sections were thawed, air dried, rehydrated in PBS and free aldehyde groups were blocked using a solution of 0.1M glycine in PBS for 10 min. Sections were then permeabilized in 0.3% Triton X-100 in PBS, and blocked with a solution of 1% BSA, 0.1% Tween-20 and 5% donkey serum in PBS for 30 min. Incubation with the primary antibodies was carried out at room temperature for 1 h before washing with PBS and applying the secondary antibodies for 1 h. The primary antibodies used were rabbit antikeratin1 at 1:400 (Biolegend #Poly19052), rabbit anti-keratin14 at 1:400 (Biolegend #Poly19053), rabbit anti-loricrin at 1:500 (Biolegend #Poly19051), rabbit anti-B55α at 1:100 (Cell Signaling Technology #4953), an in-house rabbit polyclonal antibody against a PP2A-C peptide [PHVTRRTPDYFL (Sim et al., 1998)] at 1:500, a rabbit anti-Collagen IV at 1:100 (Millipore #AB756P), rat anti-Laminin γ1 at 1:1000 (Chemicon #1914), rat anti-Integrin β1 at 1:500 (BD Pharmingen #553715), rat anti-Integrin β4 at 1:200 (BD Biosciences #346-11A), rabbit anti-β-Catenin at 1:250 (Abcam # ab32572), and rabbit anti-ZO-1 at 1:100 (Thermo Fisher Scientific #61-7300). Alexa fluor-488 and - 594 secondary antibodies (Abcam) were used at 1:500. All antibody dilutions were made in 1% BSA, 0.1% Tween-20 in PBS. After the secondary antibody incubation, slides were washed in PBS and mounted with Prolong gold antifade mountant with DAPI (Thermo Fisher Scientific #P36941), cured for 24 h in the dark at room temperature, coverslipped and stored at 4 ◦C. Imaging was performed on a Fluoview FV1000 confocal microscope (Olympus).

## Keratinocyte Cultures

Primary keratinocyte cultures were derived as previously described by Li et al. (2017). Briefly, the skin from E18.5 embryos was collected, washed in PBS then incubated overnight at 4◦C in 4 mg/mL ice cold dispase II (Merck #D4693) made up in Keratinocyte Serum Free Medium (ThermoFisher, #17005042) with added 1% Penicillin-Streptomycin antibiotic (Sigma #P4333). The next morning epidermis was isolated and incubated with the basal layer down on a 500 µL drop of 0.25% trypsin-EDTA in PBS with gentle rocking at room temperature for 20 min. Keratinocytes were mechanically detached by vigorously rubbing each epidermal sheet three times in 2 mL media. The detached keratinocyte suspensions were centrifuged at 180 × g for 5 min and seeded onto collagen I (Trevigen #3442- 100-01) coated plates at a density of 1.67 × 10<sup>5</sup> cells/mL. After growing to confluency, keratinocyte differentiation was induced by increasing the CaCl<sup>2</sup> concentration in the media from 0.06 to 0.2 mM. Cells were imaged using a Zeiss Axiovert 200 inverted light microscope and Zeiss AxioVision software.

## Epidermal Barrier Staining

Immediately upon retrieval, E18.5 embryos were washed in PBS and incubated in X-gal reaction mix (100 mM NaPO4, 1.3 mM MgCl2, 3 mM K3Fe(CN)6, 3 mM K4Fe(CN)<sup>6</sup> and 1 mg/mL X-Gal (Promega #V3941) at pH 4.5, with gentle rocking at 37◦C for 3– 4 h, followed by overnight incubation at room temperature. The following day, embryos were imaged using a digital camera.

## RESULTS

## Constitutive Deletion of Ppp2r2a Is Lethal

We generated Ppp2r2a knockout mice by targeting exon 4 using CRISPR/Cas9. Genotyping PCR primers flanking the targeted region and amplifying a product of 819 bp from the wildtype allele were used to characterize gene editing events in the resulting mice (**Figure 1**). Sequence analysis of the PCR products identified two mice with the expected deletion of

FIGURE 1 | Generation of Ppp2r2a null mice lacking B55α protein expression. (A) Schematic of Ppp2r2a wildtype (+/+) allele and Ppp2r2a knockout alleles (–/–) from the A-line and B-line, showing the extent of deletion of exon 4 and partial flanking introns. The arrows show primer binding regions. gRNA 1 and 2 show binding site of CRISPR/Cas9 guide RNAs. (B) DNA genotyping gel of E10.5 embryos showing Ppp2r2a wild-type, heterozygous and knockout progeny. (C) Representative immunoblot of two E10.5 embryos of each genotype, revealing complete loss of PP2A-B55α expression in Ppp2r2a knockout (–/–) embryos using a primary polyclonal antibody to PP2A-B55α (CST #4953) and an in-house primary polyclonal antibody to B55α as described in the methods. Representative immuno-blots of PP2A-A and –C subunits. Actin was used as a loading control. (D) The mRNA expression of the B55 subunit genes (Ppp2r2a, Ppp2r2b, Ppp2r2c, Ppp2r2d), the PP2A catalytic subunit genes (Ppp2ca and Ppp2cb) and the PP2A structural subunit genes (Ppp2r1a and Ppp2r1b) were assessed by quantitative real time PCR. Data was normalized to the housekeeping gene, Rpl19, which was constant between genotypes and embryonic stages, and is represented as a fold change relative to the respective wildtype embryos for each stage. n = 2, \*p < 0.05, \*\*p < 0.01, two-tailed t-test compared to wildtype.

exon 4 plus part of the flanking introns (**Figure 1A** and **Supplementary Figure S1**). These mice were used as founders to establish two independent Ppp2r2a knockout mouse lines, named Ppp2r2a-line A and B, respectively. The genomic deletion covers 298 bp in line A and 552 bp in line B (**Supplementary Figure S1**), removing Ppp2r2a exon 4 and resulting in a frameshift that prevents the expression of the full-length protein.

The founders were crossed twice to C57BL/6J before setting up heterozygous breeding pairs. In line A, out of 170 pups born from heterozygous breeding pairs, 100 (36%) were Ppp2r2a+/+, and 174 (64%) were Ppp2r2a+/<sup>−</sup> (**Table 1**). Similarly, in line B, out of 104 pups born from heterozygous breeding pairs, 43 (41%) were Ppp2r2a+/+, and 61 (59%) were Ppp2r2a+/<sup>−</sup> (**Table 1**). No homozygous knockout pups were observed at the time of genotyping (post-natal day 14) in either line, suggesting that constitutive deletion of Ppp2r2a results in early lethality with complete penetrance.

Genotyping of embryos at E10.5 confirmed the presence of homozygous Ppp2r2a−/<sup>−</sup> embryos (**Figure 1B**) and immunoblot analysis using two different anti-B55α antibodies of whole E10.5 embryos revealed absence of B55α protein in Ppp2r2a−/<sup>−</sup> embryos, and reduced B55α protein in Ppp2r2a+/<sup>−</sup> embryos (**Figure 1C**). This was further corroborated by immunoblot analysis of E14.5 embryos (**Supplementary Figure S2**). There was no change in protein expression of the PP2A-C catalytic subunit, or the PP2A-A structural subunit.

Similarly, at the mRNA level, Ppp2r2a expression was decreased in Ppp2r2a+/<sup>−</sup> embryos and further reduced in Ppp2r2a−/<sup>−</sup> embryos compared to Ppp2r2a+/<sup>+</sup> at both E10.5 and E18.5 (**Figure 1D**). This suggests that the altered mRNA synthesized from the Ppp2r2a knockout allele is unstable and mostly degraded, with only a small residual amount detected in Ppp2r2a−/<sup>−</sup> animals. We next assessed the expression of genes encoding the other B55 isoforms, and the catalytic and structural PP2A subunits. A significant increase in Ppp2r2b (B55β) expression was observed with Ppp2r2a knockout at E18.5. The expression of the Ppp2ca (encoding the PP2A-Cα subunit) was also increased with Ppp2r2a knockout at E18.5. The increased Ppp2r2b and Ppp2ca may be an attempt to compensate for the lack of B55α.

Heterozygous Ppp2r2a+/<sup>−</sup> mice in both the A and B lines grew normally, appeared healthy, were fertile and females nursed their pups. There were no major differences in size between Ppp2r2a+/<sup>+</sup> and Ppp2r2a+/<sup>−</sup> mice monitored up to 24 months of age, and no obvious gross anatomical or behavioral problems were observed in Ppp2r2a+/<sup>−</sup> mice (**Supplementary Figures S3A,B**), despite a reduction in B55α protein expression observed in adult tissues (**Supplementary Figure S3C**). Further investigation into the Ppp2r2a−/<sup>−</sup> animals was focused on the A line.

### Ppp2r2a Is Required for Late Embryonic Development

The absence of homozygous Ppp2r2a−/<sup>−</sup> animals at 14 days of age suggested neonatal or in utero lethality. At E10.5, 25% of embryos harvested were Ppp2r2a−/−, consistent with


TABLE 1 | Number of mice per genotype at various embryonic and post-natal stages.

a-8 litters; b-13 litters; c-17 litters; d-5 litters; e-70 litters; f-24 litters; g, this neonate was found dead the day of birth.

the expected Mendelian ratio from a heterozygous cross. The percentage of Ppp2r2a−/<sup>−</sup> embryos at E14.5, however, dropped to 18% and it decreased further at E18.5, to only 10% (**Table 1**). At birth, this percentage dropped to 3%, corresponding to only one knockout offspring found amongst a total of 35 pups from 5 litters. It must be noted, however, that this neonate was found dead while litter mates were still being born, suggesting that it was still-born or died very soon after birth, indicating perinatal lethality.

### Late-Stage Ppp2r2a−/<sup>−</sup> Embryos Are Small With Neural, Cranial and Limb Abnormalities

To investigate the cause of the embryonic lethality, we first examined embryo size and gross morphology. At midgestation (E10.5), embryos of all three genotypes appeared macroscopically normal, and there was no difference in size, as determined by measuring the crown to rump length (**Figure 2A**). In contrast, Ppp2r2a−/<sup>−</sup> embryos at E14.5 and E18.5 were significantly smaller than Ppp2r2a+/<sup>+</sup> and Ppp2r2a+/<sup>−</sup> embryos, and displayed abnormal morphology (**Figures 2A,B** and **Supplementary Figure S4**). Notably, a lack of divergent digits (syndactyly) was a common feature in all E14.5 (n = 19) and E18.5 (n = 10) Ppp2r2a−/<sup>−</sup> embryos, and was confirmed by histopathology analysis (**Figure 2C**).

While no macroscopic abnormalities were observed in E10.5 Ppp2r2a−/<sup>−</sup> embryos, histopathology analysis revealed differences in neural development. One litter analyzed displayed inconsistent neuro-epithelium in all three Ppp2r2a−/<sup>−</sup> embryos, compared to the one wild-type litter mate. Lamination of the cerebral cortex involves organization of the neurons into six layers in mammals (Chen et al., 2017). At E10.5 the Ppp2r2a+/<sup>+</sup> embryo displays this emerging lamination (**Figure 3A**). This was not observed in the Ppp2r2a−/<sup>−</sup> embryos, but rather the neuro-epithelium displayed cellular degeneration and debris (**Figure 3A**). In contrast, in a second litter analyzed containing two Ppp2r2a−/<sup>−</sup> embryos, no histological differences were observed between the Ppp2r2a+/<sup>+</sup> and Ppp2r2a−/<sup>−</sup> embryos (data not shown), suggesting incomplete penetrance of this phenotype at E10.5.

Out of 13 Ppp2r2a−/<sup>−</sup> embryos analyzed at E14.5, gross changes in cranial and/or embryo shape were observed in five embryos. Three embryos displayed a bulbous cranium, almost translucent compared to the rest of the body (**Supplementary Figure S4A**-representative). Another had a slight bulge at the cranial apex and small raised bulge along the upper dorsal region (**Supplementary Figure S4B**). The last also had a slight bulge at the cranial apex and a raised ridge-like structure along the dorsal region (**Supplementary Figure S4C**). These features suggested possible internal defects in the neural tube, such as exencephaly and spina bifida, and were confirmed by histopathology, with three Ppp2r2a−/<sup>−</sup> embryos displaying exencephaly (**Figure 3B**). One of these embryos also had cranial bulging due to edema at the cranial apex, and a protruding and poorly contained spinal cord in the cervical dorsal region (not shown). A further embryo was found to have spina bifida (**Figure 3B**).

Out of 10 Ppp2r2a−/<sup>−</sup> pups collected at E18.5, all showed movement or reacted to touch at the time of dissection. However, the skin of all E18.5 Ppp2r2a−/<sup>−</sup> embryos was shiny and displayed significant erythroderma, with visible superficial hemorrhaging particularly at the tips of the tail and limbs (**Figure 2B**). The morphology and histopathology of E18.5 Ppp2r2a−/<sup>−</sup> placentas (n = 3) appeared normal (data not shown). The only abnormal feature noted among the wildtype or heterozygous littermates was syndactyly in 1/30 (3.33%) Ppp2r2a+/<sup>+</sup> embryos, however, this embryo was partially degenerated at the time of dissection and thus was likely a runt already undergoing the process of degradation. Histopathological analysis of four E18.5 Ppp2r2a−/<sup>−</sup> embryos found that one had cranial vault collapse (**Figure 3C**).

The one dead Ppp2r2a−/<sup>−</sup> pup observed at P0 (**Table 1**) was much smaller than Ppp2r2a+/<sup>+</sup> and Ppp2r2a+/<sup>−</sup> littermates, and was partially degraded (**Supplementary Figure S5**). While Ppp2r2a+/<sup>+</sup> and Ppp2r2a+/<sup>−</sup> pups were pale pink all over, the Ppp2r2a−/<sup>−</sup> pup was deep red in color, with the exception of a white tail, and exhibited an incomplete skin layer with large patches missing from the abdomen and a smaller patch on the lower spine, suggesting a potential epidermal barrier defect. As observed in the E18.5 embryos, the digits were also poorly defined.

We next investigated whether the cranial defects observed were associated with changes in aberrant cell proliferation or cell death. IHC staining for the proliferation marker Ki67, showed similar levels of Ki67<sup>+</sup> cells in comparable brain regions of Ppp2r2a+/<sup>+</sup> and Ppp2r2a−/<sup>−</sup> embryos at E14.5 (**Supplementary Figure S6**). There were very occasional cleaved caspase 3 (CC3) positive cells in the Ppp2r2a+/<sup>+</sup> cerebellum and forebrain and none in the thalamus. In contrast, CC3<sup>+</sup> cells were found in all three brain regions in the Ppp2r2a−/<sup>−</sup> brain (**Figure 3D**). This suggests the neuronal defects in Ppp2r2a knockout embryos are associated with increased apoptosis.

## Ppp2r2a−/<sup>−</sup> Embryos Display a Thin Stratified Epidermis With a Defective Stratum Corneum

At E10.5, the epidermal committed basal layer in Ppp2r2a−/<sup>−</sup> embryos was much thinner and disrupted compared to the wildtype. In some instances, this layer was also detached from the underlying dermis (**Figure 4A**). As noted above, the skin of E18.5 Ppp2r2a−/<sup>−</sup> embryos was deep red with multiple sites of hemorrhaging. This was supported by histopathology analysis, with Ppp2r2a−/<sup>−</sup> embryos demonstrating superficial epidermal hemorrhage and edema at both E14.5 and E18.5 (**Figure 4A**). At E14.5, Ppp2r2a−/<sup>−</sup> embryos had a thinner basal epidermal layer than Ppp2r2a+/<sup>+</sup> mice, with disorganized and irregular

cell orientation, and a complete lack of an epidermal layer in some segments (**Figure 4A**, black arrowhead). Furthermore, they showed no emerging hair follicles – a hallmark feature of normal epidermis at E14.5 (Mann, 1962; Hardy, 1992; Cheng et al., 2014).

At E18.5, the epidermis of Ppp2r2a−/<sup>−</sup> embryos was also much thinner than Ppp2r2a+/<sup>+</sup> embryos at the dorsal and caudal regions. The basal layer was often disorganized, the spinous and granular layers were thin, and the upper stratum corneum was less distinct and often missing altogether. These key phenotypic differences observed in the E14.5 and E18.5 epidermis were most prominent in the dorsal region. E18.5 Ppp2r2a−/<sup>−</sup> embryos also lacked dermal appendages, such as sweat glands, sebaceous glands and hair follicles, which are a normal feature of developing skin at this stage (**Figure 4A**; Hardy, 1992; Chu and Loomis, 2004; Schmidt-Ullrich and Paus, 2005).

Immunofluorescence analysis revealed PP2A-B55α expression in the basal and suprabasal layer of the epidermis, as well as dermal cells, in wild-type embryos at E14.5. At E18.5 PP2A-B55α was observed in all layers, with strongest expression in the granular layer (**Figure 4B**). The faint staining in the Ppp2r2a−/<sup>−</sup> embryos may be due to cross-reactivity of this antibody with the closely related B55δ subunit, however, lack of a specific B55δ antibody for immunofluorescence precludes us from confirming this. Immunofluorescence for PP2A-C showed a similar localization to PP2A-B55α at both embryonic stages (**Figure 4B**).

### Ppp2r2a−/<sup>−</sup> Keratinocyte Cultures and Epidermis Display Impaired Differentiation

To further examine the epidermal phenotype of the Ppp2r2a−/<sup>−</sup> embryos, keratinocytes were isolated from the epidermis of E18.5 Ppp2r2a+/<sup>+</sup> and Ppp2r2a−/<sup>−</sup> embryos.

Interestingly, the epidermal layer separated more rapidly from the dermis of Ppp2r2a−/<sup>−</sup> than wild-type embryos, suggesting a weaker connection between the two layers. Less keratinocytes were consistently isolated from Ppp2r2a−/<sup>−</sup> skin, and these cells proliferated less than the wild-type cells (**Figure 4C**). Furthermore, the addition of high Ca2<sup>+</sup> induced differentiation in the wild-type keratinocytes but not in knockouts. After 48 h in high Ca2<sup>+</sup> the cytoplasm of Ppp2r2a+/<sup>+</sup> keratinocytes expanded and the cells arranged in a distinctive cobble-stone pattern with clear cell-cell adhesions (**Figure 4C**). In contrast, the shape and pattern of the Ppp2r2a−/<sup>−</sup> keratinocytes showed very little change to untreated cells (**Figure 4C**), suggesting both proliferation and differentiation of isolated keratinocytes is impaired with PP2A-B55α loss.

We next examined the epidermal proteins Keratin 14 (K14), a marker of proliferating basal cells; Keratin1 (K1), a marker of the spinous layer; and loricrin, a marker of the granular layer, by immunofluorescence labeling. At E14.5, K14 was present in both Ppp2r2a−/<sup>−</sup> and Ppp2r2a+/<sup>+</sup> epidermis (**Figure 5A**). The Ppp2r2a+/<sup>+</sup> epidermis was 1–2 cells thick, with all basal cells staining for K14, and the most prominent expression observed at the baso-lateral side of cells. In contrast, only one layer of cells was evident in the Ppp2r2a−/<sup>−</sup> epidermis, which was strongly positive for K14, with expression at both the apical and basal side, as well as the lateral junctions between cells (**Figure 5A**). Many cells in this single layer appeared flatter and irregularly positioned compared to the wellorganized, regular cuboidal-type basal layer of the wild-type epidermis, indicating a potential defect in cell polarity in the Ppp2r2a−/<sup>−</sup> basal layer. In the Ppp2r2a+/<sup>+</sup> epidermis, there was also a thinner layer of K14-negative cells (the intermediate layer) emerging above the K14-positive basal layers (**Figure 5A** indicated by a white arrow head), which is absent in the knockout epidermis.

At E18.5 in wild-type embryos all four layers of the epidermis have developed: basal, spinous, granular and stratum corneum (**Figure 5B**). Both the histological and immunofluorescence analysis showed that each layer of Ppp2r2a−/<sup>−</sup> epidermis was thinner than in the Ppp2r2a+/<sup>+</sup> mice. K14 marked basal and spinous layer cells in wild-type E18.5 epidermis, however, similar to that observed at E14.5, only a single basal layer of K14<sup>+</sup> cells was observed throughout most of the E18.5 Ppp2r2a−/<sup>−</sup> epidermis, and the cells were highly disorganized, further supporting a polarity defect (**Figure 5B**). K1<sup>+</sup> cells were observed in spinous and granular layers of E18.5 Ppp2r2a+/<sup>+</sup> epidermis. In contrast, strong K1 immunolabelling was observed in many cells within the basal layer of Ppp2r2a−/<sup>−</sup> epidermis. The overlying spinous layer was much thinner than the wild-type, and contained very few K1<sup>+</sup> cells, and the very thin granular layer was negative for K1 (**Figure 5B**). Immunostaining for loricrin confirmed a thinner granular layer in the Ppp2r2a−/<sup>−</sup> epidermis (**Figure 5B**). Given cells in the stratum corneum are anuclear, the few unstained nuclei above the granular layer in the Ppp2r2a−/<sup>−</sup> epidermis (**Figure 5B**, indicated by white arrowheads) indicate a defective and persisting periderm.

Immunofluorescent labeling for epidermal markers Keratin 14 (K14), Keratin 1 (K1) and Loricrin (green) in E14.5 (A) and E18.5 (B) embryos, with nuclei stained by DAPI (blue). White arrowheads show the unstained superficial epidermal cell layer. Dashed white line indicates the dermal-epidermal junction. All scale bars = 25 µm.

## Reduced Basement Membrane and Integrins in Ppp2r2a−/<sup>−</sup> Epidermis

Proper epithelial formation and integrity relies on formation of the basement membrane (BM), a specialized extracellular matrix (ECM) that is produced, secreted and assembled by the basal keratinocytes (Fuchs and Raghavan, 2002). The lack of mature stratified epithelium in the Ppp2r2a knockouts suggested there might be a defect in the BM. Immunolabelling for collagen IV and laminin γ1, two major components of the BM, was similar in both genotypes at E14.5 (**Figure 6A**), however, was markedly reduced in the Ppp2r2a−/<sup>−</sup> BM at E18.5 (**Figure 6B**).

The major cell surface receptors that recognize and assemble the BM are integrins; heterodimeric transmembrane proteins made up of an α and β subunit. Laminin binding to α3β1 integrin results in the assembly of focal adhesions, while binding to α6β4 integrin results in formation of hemidesmosomes, specialized adhesions that link the BM to the intermediate filament network (Tsuruta et al., 2008; Wickstrom et al., 2011). These integrin-mediated adhesions play critical roles in keratinocyte polarization and migration (Stepp, 1999; Rippa et al., 2013). At E14.5 β4 integrin was expressed on the basal side of the

IV (green), Laminin γ1 (red), Integrin β4 (red) and Integrin β1 (red) in Ppp2r2a+/<sup>+</sup> and Ppp2r2a−/<sup>−</sup> epidermis at both E14.5 (A) and E18.5 (B). DAPI is shown in blue. Inserts show a region from each image without the DAPI to highlight the protein labeling. All scale bars = 25 µm.

Ppp2r2a+/<sup>+</sup> basal and suprabasal cells, whereas in the single layer of Ppp2r2a−/<sup>−</sup> cells, β4 integrin was expressed but did not show a preferential basal orientation. At E18.5, β4 integrin immunolabelling is strongest along the basement membrane of wildtype embryos. In contrast, in the Ppp2r2a−/<sup>−</sup> epidermis staining is weak and discontinuous (**Figures 6A,B**). Similarly, β1 integrin immunolabelling along the BM is weaker in the Ppp2r2a−/<sup>−</sup> dorsal skin, most notably at E18.5 (**Figures 6A,B**).

(green) in E18.5 Ppp2r2a+/<sup>+</sup> and Ppp2r2a−/<sup>−</sup> epidermis. DAPI is shown in blue. Dashed white line indicates the dermal-epidermal junction. All scale bars = 25 µm.

The reduced BM and integrins would likely compromise the adhesion between the epidermis and dermis, and thus contribute to the blistering and ready separation of these layers described above.

## Ppp2r2a−/<sup>−</sup> Embryos Have Aberrant Junctions in the Epidermis

In addition to the hemidesmosomes and focal adhesions attaching the epidermis to the BM, adherens and tight junctions play a key role in cell-cell attachment and formation of the protective epidermal barrier. β-catenin links membrane bound E-cadherin to the actin cytoskeleton at adherens junctions, and is also a key effector of the canonical Wnt pathway which is important in epidermal stratification (Capaldo and Macara, 2007; Hartsock and Nelson, 2008; Zhu et al., 2014). β-catenin labeling of the E14.5 Ppp2r2a+/<sup>+</sup> epidermis is strongest along the cellcell junctions between keratinocyte layers, and also between keratinocytes within a layer (**Figure 7A**). In contrast, β-catenin labeling in the Ppp2r2a−/<sup>−</sup> epidermis is inconsistent and patchy (**Figure 7A**). At E18.5, β-catenin is still highly expressed on the apico-lateral junctions between keratinocytes of the basal layer, and to a lesser extent in the spinous layers (**Figure 7B**). In comparison, weaker β-catenin is observed in the apico-lateral regions of the basal layer, and is absent from the thin suprabasal layers, in the Ppp2r2a−/<sup>−</sup> epidermis.

The tight junction protein, ZO-1 (Capaldo and Macara, 2007; Hartsock and Nelson, 2008) in E18.5 Ppp2r2a+/<sup>+</sup> epidermis was localized at the cell-cell junctions of keratinocytes of the upperspinous and granular layers (**Figure 7B**). Interestingly, ZO-1 junctional labeling was much stronger in the granular/spinous

layers of Ppp2r2a−/<sup>−</sup> embryos compared to wildtypes but did not appear as well structured as seen on the wildtype (**Figure 7B**). Thus, the loss of PP2A-B55α affects the composition and/or organization of cell-ECM and cell-cell junctions in the developing skin.

## Late Stage Ppp2r2a−/<sup>−</sup> Embryos Have Incomplete Epidermal Barrier Acquisition

The epidermis is essential for life as it provides a protective barrier, preventing water loss and protecting from external environmental factors (O'Shaughnessy et al., 2009). To test whether the epidermal abnormalities affected epidermal barrier function, a skin permeability assay was conducted on E18.5 embryos. In embryos with an incompletely formed epidermis the X-gal substrate will penetrate the skin, be cleaved by the endogenous β-galactosidase enzyme in the skin and form a blue precipitate in low pH conditions (Hardman et al., 1998; Guttormsen et al., 2008). In all Ppp2r2a+/<sup>+</sup> and Ppp2r2a+/<sup>−</sup> embryos, only the edges of skin where the tail and umbilical cord were cut, turned slightly blue due to experimentally interrupted barrier. In contrast Ppp2r2a−/<sup>−</sup> embryos developed patchy blue staining all over the epidermis, confirming the presence of a disrupted epidermis, resulting in defective barrier function (**Figure 8A**).

## Altered Proliferation in Ppp2r2a−/<sup>−</sup> Epidermis

The primary cultures suggested that proliferation may be impaired in the epidermis of Ppp2r2a−/<sup>−</sup> embryos. Therefore we stained for the proliferation marker, Ki67, via IHC. At E14.5, as noted above, the dorsal Ppp2r2a−/<sup>−</sup> epidermis was generally 1-cell thick, with only an occasional suprabasal layer, whereas the Ppp2r2a+/<sup>+</sup> epidermis was 1–3 cells thick. However, in both genotypes the majority of epidermal cells were Ki67<sup>+</sup> (**Figures 8B,C**). In E18.5 Ppp2r2a+/<sup>+</sup> embryos, Ki67<sup>+</sup> cells were present in the basal layer of the epidermis and surrounding the dermal appendages, with only a few positive cells in the spinous and subsequent outer stratified layers (**Figures 8B,C**). In comparison, the Ppp2r2a−/<sup>−</sup> epidermis contained fewer Ki67<sup>+</sup> in the basal layer. In the suprabasal layer, where the cells are more compact in the knockout compared to the wild-type, there were reduced total numbers of cells, however, more cells were Ki67+. Therefore, homozygous Ppp2r2a deletion results in slightly reduced epidermal basal cell proliferation, but increased proliferation in the more apical suprabasal cells, which at this stage should be ceasing proliferation and starting to terminally differentiate.

## The Epidermal Defect in Ppp2r2a−/<sup>−</sup> Mice Is Associated With Increased Phosphorylation of the PP2A Target cJun

cJun is a subunit of the transcription factor AP-1, and functions in epidermal development downstream of PP2A and AKT (Karin, 1995; Kallunki et al., 1996). We found increased phosphorylation of cJun at S63, a site which enhances cJun transcriptional activity (Pulverer et al., 1991; Smeal et al., 1991;

FIGURE 8 | Defective epidermal barrier in Ppp2r2a knockout embryos is associated with changes in proliferation, and p-cJun signaling. (A) X-gal substrate penetration (permeability) assay in E18.5 embryos. Areas stained blue on the Ppp2r2a−/<sup>−</sup> embryo indicates regions with an impaired epidermal barrier. White arrowheads mark regions (umbilical cord and tail) that stained blue as they were cut during dissection. Representative of n = 3. Scale bar = 1 cm. (B) Ppp2r2a+/<sup>+</sup> and Ppp2r2a−/<sup>−</sup> epidermal regions at E14.5 and E18.5 showing Ki67 (proliferation marker) staining. Dashed black line indicates the dermal-epidermal junction. (C) Quantitation of Ki67 positive and negative cells in E14.5 and E18.5 epidermis. Ki67<sup>+</sup> cells were significantly increased in the Ppp2r2a−/<sup>−</sup> suprabasal epidermis compared to the wild-type. \*p < 0.05, unpaired two-tailed t-test. (D) IHC for p-cJun (S63) in Ppp2r2a+/<sup>+</sup> and Ppp2r2a−/<sup>−</sup> epidermal regions. Dashed black line indicates the dermal-epidermal junction. All scale bars = 50 µm.

Deng and Karin, 1994), in the basal, spinous and granular layers of dorsal and ventral Ppp2r2a−/<sup>−</sup> epidermis, compared to wildtype epidermis (**Figure 8D**), suggesting loss of PP2A-B55α leads to hyperphosphorylation and activation of cJun in the developing skin.

### DISCUSSION

We have generated and characterized the first constitutive Ppp2r2a knockout mouse. Homozygous deletion of Ppp2r2a resulted in embryonic lethality between E10.5 and birth, while heterozygous mice were viable and grew normally. Late stage Ppp2r2a knockout embryos were significantly smaller in size than littermate controls, and displayed various abnormalities such as syndactyly, neural defects and epidermal defects. The epidermal defect was characterized by thinner basal, spinous and granular layers, and a discontinuous stratum corneum. The structural defects of the epidermis were accompanied by a lack of dermal appendages and increased epidermal hemorrhage and edema. Functionally, the morphological skin defects translated to impaired epidermal barrier integrity, which is the likely cause of Ppp2r2a−/<sup>−</sup> neonatal death.

While this is the first reported in vivo analysis of Ppp2r2a deletion in mice, a study in zebrafish embryos revealed the importance of PP2A-B55α in cytoskeletal regulation to promote stable angiogenesis (Martin et al., 2013). Similarly, a prior ex vivo mouse embryo study had suggested that PP2A-B55α was essential in early embryogenesis (Liang et al., 2017). Double-stranded RNA-mediated Ppp2r2a knockdown in fertilized mouse zygotes arrested in vitro embryonic development, in association with increased DNA damage and apoptosis (Liang et al., 2017). In contrast, we observed close to Mendelian ratios of all three genotypes at E10.5, showing that PP2A-B55α is not essential for early embryonic development in vivo. However, the percentage of Ppp2r2a−/<sup>−</sup> embryos recovered at E14.5 and E18.5 steadily declined, suggesting that some knockout mice die between E10.5 and E14.5, others between E14.5 and E18.5, and a final few between E18.5 and P0. Therefore, there is significant heterogeneity in the penetrance of the phenotypes caused by Ppp2r2a loss.

The lethality observed in our Ppp2r2a knockout mice occurred later in embryogenesis than that reported in knockout mice for the PP2A structural and catalytic subunits (Götz et al., 1998; Ruediger et al., 2011). PP2A-Aα (Ppp2r1a) deletion resulted in the absence of homozygous knockout embryos by E10.5, but the cause of lethality was not determined (Ruediger et al., 2011). Given our E10.5 Ppp2r2a knockout embryos appeared grossly normal, we can infer that PP2A-B55α is not essential for implantation or gastrulation. However, the few occurrences of neural-related defects observed between E10.5 and E18.5 suggest that neurulation (∼E7.5) may not be fully functional. Homozygous PP2A-Cα knockout (Ppp2ca−/−) embryos had a developmental block at E6.5, before completing gastrulation (Götz et al., 1998). Moreover, Ppp2ca−/<sup>−</sup> embryos examined at E7.5 were found to be amorphous, smaller in size than controls and have a degenerated embryonic ectoderm (Götz et al., 1998). The early developmental block makes it difficult to compare the phenotype exactly to that of our Ppp2r2a knockout, however, some similarities are apparent; Ppp2r2a knockout embryos were also significantly smaller than wild-type littermates, and showed defects in neural and epidermal tissue, both of which derive from the ectodermal germ layer (Loebel et al., 2003). Thus, PP2A complexes containing B55α play an essential role in the development and differentiation of the ectoderm, but other PP2A regulatory subunit(s) are necessary, or are able to compensate for the loss of B55α, in the earliest stages of development.

Lamination refers to the differentiation of neurons into morphologically distinct layers in the neuroepithelium and begins around E10.0-10.5 (Chen et al., 2017). The emergence of these layers was seen in the E10.5 Ppp2r2a+/<sup>+</sup> neuroepithelium (**Figure 3A**), but was not apparent in the Ppp2r2a knockout embryos. At E14.5 there is a continuation of neuroepithelium stratification and differentiation in wild-type mice, while at this stage Ppp2r2a−/<sup>−</sup> embryos exhibited neural tube defects. To our knowledge, PP2A has not previously been implicated in these developmental disorders. Exencephaly and spina bifida occur due to neurulation defects, such as inadequate neural tube folding, apposition or fusion, resulting in failure of neural fold closure (Harris and Juriloff, 2007; Chen et al., 2017). These defects begin early, with neurulation starting at ∼E7.5 and neural tube folding and fusion by ∼E10.5. The physical separation of the neuroectoderm from the surface ectoderm (future skin) also begins at ∼E10.5 (Chen et al., 2017). Thus, PP2A-B55α is important for normal development of the neuroepithelium.

There is a wide variety of genetic causes of neural tube defects [reviewed in Harris and Juriloff (2007, 2010)]. Mutation or reduced expression of numerous members of Wnt, FGF, or BMP pathways, for example FGFr1 and Axin, can lead to exencephaly and spina bifida (Juriloff and Harris, 2000). Neural cells develop from the ectoderm in the absence of canonical Wnt signaling when FGF signaling predominates and inhibits BMP. Conversely, an epidermal fate is adopted when active Wnt and BMP signaling predominates and promotes β-catenin activity (Finley et al., 1999; Stern, 2005; Fuchs, 2007). PP2A is known to regulate various proteins within these pathways and thus can have direct and indirect effects on neurulation. For example, it can positively regulate Raf-1, negatively regulate Erk downstream of FGF (Abraham et al., 2000; Jaumot and Hancock, 2001; Kubicek et al., 2002; Strack, 2002; Adams et al., 2005), and can exert both positive and negative regulation on Wnt/β-catenin signaling (Götz et al., 2000; Ivaska et al., 2002; Su et al., 2008; Clevers and Nusse, 2012; Mitra et al., 2012; Stamos and Weis, 2013; Thompson and Williams, 2018). PP2A-B55α complexes in particular can dephosphorylate β-catenin, leading to its stabilization and promotion of Wnt signaling (Zhang et al., 2009). Thus PP2A-B55α loss would be predicted to reduce Wnt signaling, and hence there may be no problem in committing to a neural fate, but rather in subsequent neurulation steps.

Further neural tube folding is promoted by the Sonic hedgehog (SHH) protein, alone and interacting with Wnts/BMPs (Colas and Schoenwolf, 2001; Ybot-Gonzalez et al., 2002). Loss of function of GLI3, a key SHH effector causes embryonic exencephaly, as does ectopic SHH expression (Echelard et al.,

1993; Hui and Joyner, 1993). PP2A is also involved in SHH signaling via negative regulation of GLI3 (Krauß et al., 2008). AP-2 is a retinoic acid inducible transcription factor that is expressed by non-neuronal ectodermal cells but is also vital for neural tube closure, and AP-2 null mice develop exencephaly (Zhang et al., 1996). B55α containing PP2A complexes can dephosphorylate and decrease AP-2 activity (Ricotta et al., 2008). Similar pathways and proteins are also involved in syndactyly development, including the HOX genes (SHH and Indian hedgehog), FGFs, BMPs and Wnt, implying that PP2A-B55α regulates these pathways in multiple regions of the developing embryo (Manouvrier-Hanu et al., 1999).

The cellular degeneration and debris observed at E10.5 indicates that cell death is also occurring in Ppp2r2a knockouts. Whether this is due to PP2A specifically regulating cell death pathways, or is a consequence of the defective differentiation and the beginning of embryo degradation and resorption, is not clear. The neural defects observed at E14.5, were associated with increased apoptosis. Apoptosis is a normal and important part of shaping neuroepithelial development; however, it occurs in a very precise spatiotemporal manner. At later stages of gestation (E12–18), apoptosis occurs in regions that become the ventricular zone, intermediate zone, and the developing cortical plate and dorsal root ganglion of the cerebral cortex (Haydar et al., 2000; Yeo and Gautier, 2004; Chen et al., 2017). In the E14.5 Ppp2r2a−/<sup>−</sup> neuroepithelium, however, apoptosis was found in the thalamus, cerebellum and forebrain regions, which at this stage should be undergoing enlargement and differentiation (Chen et al., 2017).

Ppp2r2a deletion led to severe defects in epidermal stratification, including separation of the dermo-epidermal junction, subcutaneous hemorrhage and edema, and a thin and disorganized epidermis. Defective Ppp2r2a−/<sup>−</sup> keratinocyte differentiation was also evidenced in culture. These epithelial defects further indicate a failure of normal ectoderm differentiation with Ppp2r2a deletion. After neural tube fusion, the overlying epidermal ectoderm is separated from the neural tube and becomes what will be the skin at the back of the embryo (Colas and Schoenwolf, 2001). Interestingly, the most severe epidermal defects in the Ppp2r2a knockout occurred in the dorsal skin. Wnt/FGF/BMP signaling is important in induction of an epidermal fate and is also continually required through epidermal development and stratification (Zhu et al., 2014), providing further evidence for PP2A-B55α acting on these pathways in different regions of the developing embryo. In support of this hypothesis, B55α has been implicated in the regulation of BMP/TGF-β signaling by enhancing the TGF-β/Activin/Nodal signaling pathway (Batut et al., 2008). This pathway is essential for cell fate determination during development and for regulation of epidermal differentiation. Indeed, as mentioned previously, Wnt and BMP signaling are required for the ectoderm to develop into the epidermis, and this is further promoted by combined gradients of BMP and Nodal signaling during embryonic development which block neuroectoderm formation (Camus et al., 2006) and regulate patterning of the embryonic axes and neural and gut tubes (Arnold and Robertson, 2009; Pauklin and Vallier, 2015).

PP2A has previously been implicated in epidermal barrier formation. Mice with K14-Cre driven conditional knockout of PP2A-Cα had significant hair loss and disruption in the hair follicle regeneration cycle, as well as stunted size, melanin deposition and hyper-proliferation at the base of their claws (Fang et al., 2016). Furthermore, ex vivo studies have implicated the B55α subunit in epidermal development (O'Shaughnessy et al., 2009; Gerner et al., 2013; Youssef et al., 2013). Of the Ppp2r2 family, the Ppp2r2a isoform is the most highly expressed during epidermal barrier acquisition in mouse embryos (O'Shaughnessy et al., 2009). Indeed, we found B55α protein to be expressed in all epidermal layers of E14.5 and E18.5 wild-type embryos. The epidermal defects observed with Ppp2r2a deletion now confirm an essential functional role for PP2A-B55α in epidermal development.

The formation of the epidermal barrier is a key element of late embryonic development that facilitates survival ex-utero, and should be completely formed by E18.5 in C57BL/6 embryos (O'Shaughnessy et al., 2009). However, the epidermal barrier in E18.5 Ppp2r2a−/<sup>−</sup> embryos was patchy and penetrable, which would inevitably result in the rapid death of pups at birth, as evidenced by the one Ppp2r2a−/<sup>−</sup> pup found dead at P0 that had visible skin defects. In the Ppp2r2a−/<sup>−</sup> epidermis the basal, spinous and granular layers were thin and poorly defined, and the stratum corneum was inconsistent or absent. The stratum corneum plays a major role in preventing water loss and protecting against the external environment (Sandilands et al., 2009). The presence of a few cells with intact nuclei in the uppermost layer of the Ppp2r2a−/<sup>−</sup> epidermis (above the loricrin stained granular layer **Figure 5B**) may indicate a persisting periderm, the transitory layer that normally disappears by ∼E17.5 (Sengel, 1976; M'Boneko and Merker, 1988; Zhang et al., 2005), and adds to the evidence of defective final cornification.

In the Ppp2r2a−/<sup>−</sup> epidermis there was also a lack of dermal appendages, which develop into hair follicles, sebaceous glands and sweat glands (Liu et al., 2013). Dermal appendages arise from interactive signaling between the epidermis and underlying dermis (mesoderm), and is mediated by similar signaling proteins and pathways as neural and epidermal development (Wnt, BMP, FGF, SHH) (Duverger and Morasso, 2009; Sennett and Rendl, 2012; Biggs and Mikkola, 2014; Ahn, 2015). Epidermal Wnt production modulates a BMP-FGF signaling cascade in the dermis and is essential for proper stratification and formation of the spinous layer, and for hair follicle initiation (Fu and Hsu, 2013; Zhu et al., 2014). This Wnt signaling cascade also promotes normal p63 expression in proliferating basal keratinocytes, and is required for the normal stratification process. When epidermal Wnt signaling is disrupted, the subsequent loss of p63 expression ablates the proliferative capacity of basal keratinocytes to properly stratify, resulting in a hypoplastic spinous layer, similar to the thin spinous layer of Ppp2r2a knockouts, thus further implicating aberrant Wnt signaling in lethality of these knockouts (Zhu et al., 2014).

Epidermal stratification and hair follicle formation also rely on β1-integrin mediated remodeling of the basement membrane. Epidermal specific (K14-Cre) β1-integrin knockout mice exhibit severe skin blistering and hair defects, accompanied by

massive failure of BM assembly/organization, hemidesmosome instability, and a failure of hair follicle keratinocytes to remodel BM and invaginate into the dermis (Raghavan et al., 2000). Thus, the reduced β1-integrin in our Ppp2r2a knockouts likely contributes to the failure of developing hair follicles to invaginate into the underlying dermis.

The interaction and signaling between the epidermis and dermis via the BM, is also essential for keratinocyte polarization, proliferation and differentiation. α6β4 integrin heterodimers form hemidesmosomes by attaching to intracellular keratin filaments and anchoring basal keratinocytes to the BM (Jones et al., 1998). The BM layer was reduced and inconsistent in the Ppp2r2a−/<sup>−</sup> skin, and the expression of β4-integrin was also markedly reduced. β4-integrin null mice have hemidesmosome loss and show defective epidermal stratification, weak attachment to the basal lamina and basal and spinous keratinocyte disorganization leading to gross skin denuding (Dowling et al., 1996). Indeed, the separation of the dermis and epidermis, or blistering, and associated hemorrhage observed in the Ppp2r2a−/<sup>−</sup> skin (**Figures 2B**, **4A**), are reminiscent of these β4 integrin knockouts. Furthermore, the stratified layers of these mice displayed mitotic basal-like keratinocytes, similar to the proliferative (Ki67+) cells observed in the suprabasal layers of the Ppp2r2a−/<sup>−</sup> embryos. Correct timing of hemidesmosome internalization and detachment from the BM is vital for formation of the spinous layer (Poumay et al., 1994). The ability of B55α null and β4-integrin null keratinocytes to maintain expression of basal cell markers and undergo mitosis indicates that loss of attachment to the BM and impaired polarization leads to premature release of basal cells into the stratified layers, without appropriate signals to terminally differentiate (Stepp, 1999; Rippa et al., 2013).

Keratinocyte polarity is not only vital for the appropriate cell-BM attachment, but also for correct orientation of the mitotic spindle, which under normal conditions is oriented parallel to the BM during basal layer formation, and changes to perpendicular to the BM to form suprabasal keratinocyte layers, with distinct differentiation markers (Smart, 1970; Lechler and Fuchs, 2005; Muroyama and Lechler, 2012). PP2A-B55α is known to play a role in cytoskeletal regulation and spindle dynamics during mitosis (Martin et al., 2013; Williams et al., 2014; Wang et al., 2015, 2018; Liang et al., 2017) thus, loss of PP2A-B55α-mediated regulation of cell polarity could cause defective "asymmetric" division, leading to accumulation of immature and mitotic suprabasal keratinocytes.

In addition to the reduction of β1- and β4-integrins, the Ppp2r2a−/<sup>−</sup> basal and spinous layers displayed reduced and inconsistent membrane labeling of the adherens junction associated protein, β-catenin. In contrast, the tight-junction associated ZO-1 labeling in the thin granular layers of Ppp2r2a−/<sup>−</sup> basal keratinocytes was stronger than the wildtype. Whether this is a compensatory mechanism to adjust for reduced hemidesmosomes, focal adhesions and adherens junction attachments remains to be determined.

The expression of K1 in Ppp2r2a−/<sup>−</sup> basal keratinocytes, which is normally restricted to keratinocytes committed to terminal differentiation (Sur et al., 2006), further indicates aberrant differentiation, and is reminiscent of 1Np63 null mice (Romano et al., 2012). 1Np63, a transcription factor with high homology to the tumor suppressor p53, and downstream mediator of Wnt signaling, is a master regulator of epithelial development and differentiation, and its deletion results in severe developmental abnormalities including truncated forelimbs, the absence of hind limbs, and a poorly developed stratified epidermis comprising isolated clusters of disorganized epithelial cells, with premature expression of markers associated with terminal differentiation, such as K1 (Romano et al., 2012). Like the Ppp2r2a knockouts, 1Np63 null mice also had dramatic reduction in collagen and laminin BM proteins. PP2A was recently reported in 1Np63 immunoprecipitates from squamous cell carcinoma cell lines (Katoh et al., 2016), but whether B55α directly or indirectly regulates 1Np63 activity remains to be determined.

Taken together, our data suggests that PP2A-B55α is involved in epidermal development via regulation of WNT and BMP signaling pathways, together with cell polarity and adhesion regulation. Which specific substrates are involved in mediating these functions is not fully clear. cJun is a known target of PP2A-B55α (Gilan et al., 2015), and O'Shaughnessy et al. (2009) showed that there is a pulse of p-AKT leading to transient phospho-cJun (p-cJun) dephosphorylation during barrier acquisition (∼E17.5) and proposed that this occurs via PP2A-B55α. They further found that Ppp2r2a knockdown in skin explant cultures resulted in increased cJun phosphorylation and epidermal barrier incompetence (O'Shaughnessy et al., 2009). The increased p-cJun in the E18.5 Ppp2r2a knockout epidermis supports PP2A-B55α being the effector of AKT-mediated cJun dephosphorylation, and provides the first in vivo evidence for Ppp2r2a loss leading to defective epidermal barrier acquisition through increased aberrant p-cJun signaling, thus disrupting this spatiotemporally sensitive sequence of events that must occur at this time.

During the very late stages of embryonic and epidermal development, PP2A can also regulate the conversion of profilaggrin to free filaggrin monomers, which contribute to the protein scaffold upon which the uppermost impenetrable stratum corneum will attach (Kam et al., 1993; Sandilands et al., 2009). The lack of stratum corneum and dysfunctional epidermal barrier in the Ppp2r2a knockouts suggests that the PP2A-B55α regulatory subunit may regulate filaggrin. However, it is unlikely that PP2A-B55α regulation of cJun or filaggrin at these late stages of epidermal stratification are the only factors leading to the epidermal defects observed, particularly given defects were observed as early as E14.5. PP2A regulation of components of the Wnt, β-catenin, BMP, FGF, and SHH signaling pathways, are more likely to be driving neural development early on, and limb and epidermal development in later stages. Integrins, keratins, catenins, and tight junction proteins such as ZO-1, are all regulated by phosphorylation, therefore hyperphosphorylation of one or more of these proteins in the absence of B55α may also contribute.

In summary, our study provides the first in vivo evidence for the requirement of Ppp2r2a in embryonic development, with Ppp2r2a knockout causing limb, neural and epidermal

defects. These aspects of embryonic development are for the majority controlled by key signaling pathways driving ectoderm development, such as Wnt/β-catenin, BMP, FGF and SHH. This highlights a novel role for PP2A-B55α in the control of ectodermal development. Future identification of the specific PP2A-B55α substrates mediating these effects will shed further light on the functional role of this essential protein phosphatase in normal development and disease.

#### DATA AVAILABILITY STATEMENT

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

### ETHICS STATEMENT

The animal study was reviewed and approved by the University of Newcastle Animal Care and Ethics Committee.

## AUTHOR CONTRIBUTIONS

All authors contributed to this manuscript. NP, MC, CL-O, RK, and SR performed the experiments. NP, SR, and NV wrote the manuscript. NV and SR designed the study.

#### REFERENCES


#### FUNDING

This work was supported by grants from the Cancer Council NSW (APP1086334) and the Australian Research Council (FT170100077). NP was supported by a Research Training Program Scholarship, University of Newcastle Vice Chancellors' Scholarship, and NV by an Australian Research Council Future Fellowship. The APN was supported by the Australian Government through the National Collaborative Research Infrastructure Strategy (NCRIS) Program.

#### ACKNOWLEDGMENTS

The authors thank Prof. Allison Cowin, University of South Australia, and Prof. Christopher Dayas, University of Newcastle, for critical review of the manuscript. The Ppp2r2a mouse line were produced by the Monash University node of the Australian Phenomics Network (APN). This study also utilized the Australian Phenomics Network Histopathology and Organ Pathology Service, the University of Melbourne.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2020.00358/ full#supplementary-material

to deregulation of the cyclin B-Cdc2/PP2A balance. Proc. Natl. Acad. Sci. U.S.A. 107, 12564–12569. doi: 10.1073/pnas.0914191107



Fuchs, E., and Raghavan, S. (2002). Getting under the skin of epidermal morphogenesis. Nat. Rev. Genet. 3, 199–209. doi: 10.1038/nrg758


skin development in AP-2 gamma mutant mice. Dev. Biol. 317, 187–195. doi: 10.1016/j.ydbio.2008.02.017


transcription factor GLI3. Cancer Res. 68, 4658–4665. doi: 10.1158/0008-5472. CAN-07-6174


doi: 10.1016/j.semcdb.2012.08.011


**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 Panicker, Coutman, Lawlor-O'Neill, Kahl, Roselli and Verrills. 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.

# Telomere Length Measurement by Molecular Combing

Vivian F. S. Kahl<sup>1</sup>† , Joshua A. M. Allen<sup>1</sup>† , Christopher B. Nelson<sup>1</sup> , Alexander P. Sobinoff<sup>1</sup> , Michael Lee<sup>1</sup> , Tatjana Kilo<sup>2</sup> , Raja S. Vasireddy<sup>2</sup> and Hilda A. Pickett<sup>1</sup> \*

<sup>1</sup> Telomere Length Regulation Unit, Children's Medical Research Institute, Faculty of Medicine and Health, The University of Sydney, Sydney, NSW, Australia, <sup>2</sup> Department of Hematology, Children's Hospital at Westmead, Sydney Children's Hospitals Network, Sydney, NSW, Australia

Telomeres are repetitive regions of DNA bound by specialized proteins at the termini of linear chromosomes that prevent the natural chromosome ends from being recognized as DNA double strand breaks. Telomeric DNA is gradually eroded with each round of cell division, resulting in the accumulation of critically short or dysfunctional telomeres that eventually trigger cellular senescence. Consequently, telomere length is indicative of the proliferative capacity of a cell. Multiple methods exist to measure telomere length and telomere content, but a simple and reliable technique to accurately measure individual telomere lengths is currently lacking. We have developed the Telomere length Combing Assay (TCA) to measure telomere length on stretched DNA fibers. We used TCA to measure telomere erosion in primary human fibroblasts, and to detect telomere lengthening in response to activation of telomere maintenance pathways. TCA was also used to accurately measure telomere length in healthy individuals, and to identify critically short telomeres in patients with telomere biology disorders. TCA is performed on isolated DNA, negating the need for cycling cells. TCA is amenable to semiautomated image analysis, and can be fully automated using the Genomic Vision molecular combing platform. This not only precludes sampling bias, but also provides the potential for high-throughput applications and clinical development. TCA is a simple and versatile technique to measure the distribution of individual telomere lengths in a cell population, offering improved accuracy, and more detailed biological insight for telomere length measurement applications.

#### Edited by:

Liz Caldon, Garvan Institute of Medical Research, Australia

#### Reviewed by:

Shang Li, Duke-NUS Medical School, Singapore Predrag Slijepcevic, Brunel University London, United Kingdom

#### \*Correspondence:

Hilda A. Pickett hpickett@cmri.org.au †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 14 February 2020 Accepted: 25 May 2020 Published: 16 June 2020

#### Citation:

Kahl VFS, Allen JAM, Nelson CB, Sobinoff AP, Lee M, Kilo T, Vasireddy RS and Pickett HA (2020) Telomere Length Measurement by Molecular Combing. Front. Cell Dev. Biol. 8:493. doi: 10.3389/fcell.2020.00493 Keywords: telomere, telomere length, senescence, Telomere length Combing Assay, telomerase

## INTRODUCTION

Telomeres are specialized nucleoprotein structures at the ends of linear chromosomes that function to protect the chromosome ends, thereby maintaining the stability of the genome. Telomeric DNA comprises repetitive sequences of the hexanucleotide TTAGGG<sup>n</sup> repeat unit, bound in a sequence-specific manner to the protein complex shelterin, and assembled into macromolecular structures called telomere-loops (t-loops; Griffith et al., 1999; Blackburn, 2000; Van Ly et al., 2018). In normal human somatic cells, telomeres range from 5–15 kb in length (Pickett et al., 2011), and telomere length variability exists between individual telomeres and between different cell types. Inter-individual variability is also observed across the human population, superimposed on the well-established age-associated decline in telomere length (Aubert and Lansdorp, 2008).

The negative correlation between telomere length and chronological age is attributed to terminal replication limitations, oxidative damage, and nucleolytic degradation (Harley et al., 1990; Baird, 2008a).

The telomere attrition that accompanies cellular proliferation eventually leads to an accumulation of critically short or unprotected telomeres, which signals the onset of replicative senescence (Hayflick and Moorhead, 1961; Kaul et al., 2012). Senescence provides a barrier to unlimited cellular proliferation, thereby fulfilling a potent tumor suppressive role (Reddel, 2010). In some rare cases, cells are able to bypass replicative senescence by inactivating tumor suppressor pathways, allowing cells to proceed into crisis, which is characterized by catastrophic telomere shortening and widespread genome instability (Reddel, 2010). This process provokes the emergence of cancer cells with tumorigenic advantage. However, oncogenic progression necessitates stabilization of the genome to overcome crisis, which is dependent upon activation of a telomere maintenance mechanism (TMM; Maciejewski and de Lange, 2017). Telomere maintenance is achieved by one of two defined mechanisms. First, activation of the ribonucleoprotein enzyme telomerase, which uses a template sequence embedded within the RNA component of the enzyme to reverse transcribe telomeric sequences directly onto the chromosome termini (Bryan et al., 1998; Reddel, 2014). Second, the Alternative Lengthening of Telomeres (ALT) pathway, which co-opts homology-directed repair mechanisms to drive template-mediated telomere extension (Cesare and Reddel, 2010; Reddel, 2014; Pickett and Reddel, 2015).

Telomerase is also active in the germline, during embryogenesis, and in hematopoietic, stem and rapidly renewing cells, but is suppressed to undetectable levels upon differentiation in human somatic cells. In this capacity, telomerase supports the proliferative requirements of the organism by maintaining telomere length. When telomere length cannot be adequately maintained, highly proliferative tissues become impacted. Telomere biology disorders (TBDs) are a group of diseases caused by germline mutations in genes involved in telomere maintenance and function (Bertuch, 2016). The clinical manifestations of these disorders include bone marrow failure, aplastic anemia, pulmonary fibrosis and acute myeloid leukemia, which are attributed to the premature loss of stem cell populations. TBDs are typically characterized by telomere lengths in the bottom percentile of the normal population, and telomere length measurement is used for the clinical diagnosis of patients, bone marrow donor screening, and to direct effective treatment regimens for bone marrow transplant. Milder deficiencies in telomere length have also been implicated as risk factors in several diseases, including cardiovascular disease, diabetes mellitus, obesity, liver cirrhosis, and cancer (Epel et al., 2004; Sampson and Hughes, 2006; Willeit et al., 2010; Haycock et al., 2014; Opresko and Shay, 2016).

A variety of telomere length measurement methods exist, each with its own strengths and weaknesses. Accuracy and precision are critical for the utility of telomere length measurement in clinical, epidemiological and research studies. Many current measurement techniques provide an average or relative measurement of telomere length or telomere content; however, it is well established that it is the shortest telomeres, rather than average telomere length, that trigger cellular senescence (Hemann et al., 2001; Herbig et al., 2004; Zou et al., 2004; Kaul et al., 2012). Measurement techniques that provide the distribution of telomere lengths in a cell population offer improved and more informative data regarding telomere length dynamics.

Here, we have developed the Telomere length Combing Assay (TCA) as an accurate and robust technique to measure the distribution of telomere lengths in a cell population. TCA involves stretching DNA fibers onto coated glass coverslips using a constant stretching factor (Schurra and Bensimon, 2009). Telomeric DNA is then visualized with telomere-specific PNA probes, and individual telomere lengths are measured manually or using automated software and converted to absolute telomere length measurements according to the stretching factor. We demonstrate that telomere length measurements obtained by TCA are comparable to other widely utilized methods, and that TCA can detect dynamic changes in telomere length. TCA can be used to measure telomere lengths in healthy individuals and can accurately identify TBD patients with critically short telomeres. Finally, we demonstrate that TCA is amenable to semi-automated image analysis through two open-source software platforms and fully automated image analysis using the Genomic Vision molecular combing platform, and can therefore be adapted for high-throughput applications.

#### MATERIALS AND METHODS

#### Cell Culture

The cell lines U-2 OS, HeLa, HT1080, HT1080 hTR, and IIICF/c were cultured in Dulbecco's Modified Eagle's Medium (DMEM; Gibco, Paisley, United Kingdom) with 10% fetal bovine serum (FBS) at 37◦C in 10% CO2. MRC-5 cells were cultured similarly, but in 5% CO2. The CCRF-CEM cell line was cultured in RPMI (Thermo Scientific, Norwood, Australia) with 10% FBS. All cell lines were authenticated by 16-locus short-tandem-repeat profiling and confirmed to be free of mycoplasma contamination by CellBank Australia (Children's Medical Research Institute, Westmead, Australia).

#### Subjects

Peripheral blood was collected from healthy individuals and clinically referred individuals suspected of having a TBD through the Department of Hematology Telomere Length Testing Facility, Sydney Children's Hospitals Network. Informed consent was obtained from all participating individuals, and the studies were approved by the Human Research Ethics Committee of the Sydney Children's Hospitals Network.

#### Genomic DNA Extraction

Cells were harvested with 0.05% Trypsin-EDTA (Gibco, Grand Island, United States), washed in PBS and lysed in DNA extraction buffer [100 mM Tris–HCl pH 7.6, 100 mM NaCl, 10 mM EDTA, and 1% (w/v) N-lauroylsarcosine]. Lysates were digested with 50 µg/mL RNase A for 20 min at room

temperature and then with 100 µg/mL proteinase K at 55◦C for 12–16 h. DNA was extracted using three rounds of phenol/chloroform/isoamyl alcohol (25:24:1) solution (Sigma-Aldrich, Castle Hill, Australia) in MaXtract High Density tubes (Qiagen, Maryland, United States). DNA from the aqueous phase was then precipitated with 0.1 volume of 3 M sodium acetate pH 5.2 and 2.5 volumes of cold 100% ethanol. Finally, genomic DNA was washed with 70% ethanol, air dried and dissolved in 10 mM Tris–HCl pH 8.0, 1 mM EDTA. All subsequent assays (TCA, TRF, and qPCR) were carried out on the same sample of genomic DNA extracted from U-2 OS, HT1080, HT1080 hTR, HeLa, IIICF/c, and MRC-5 cells to minimize experimental variability.

Whole blood from healthy individuals and TBD patients was collected through venipuncture. For flow-fluorescence in situ hybridization (FISH), peripheral mononuclear blood cells (PBMCs) were isolated from whole blood using Histopaque-1077 (Sigma-Aldrich, Castle Hills, Australia) after the blood was homogenized with Hanks Balanced Salt Solution (Sigma-Aldrich, Castle Hills, Australia) to maintain pH and osmotic balance. For qPCR procedures, DNA from the whole blood of healthy individuals and TBD patients was extracted using QIAGEN QIAamp DNA Mini Kit (Qiagen, Maryland, United States) according to the manufacturer's protocol.

### Telomere Length Combing Assay (TCA)

A detailed working protocol is provided in the **Supplementary Material** and an overview of the method is shown in **Figure 1**. Briefly, cells were isolated by trypsinization, embedded in agarose plugs (Sigma-Aldrich, Castle Hills, Australia), and subjected to proteinase K (0.5 M EDTA pH 8.0, 10% (v/v) sarcosyl/0.5 M EDTA, and 20 mg/mL proteinase K) digestion at 50◦C for 12–16 h. Plugs were dissolved with agarase (Thermo Scientific, Norwood, Australia) for 12–16 h. The DNA solution was then transferred to a reservoir (Genomic Vision, Paris, France). Molecular combing was performed using the FiberComb <sup>R</sup> Molecular Combing System (Genomic Vision, Paris, France) with a constant stretching factor of 2 kb/µm using vinylsilane coverslips (20 × 20 mm; Genomic Vision, Paris, France), according to the manufacturer's instructions. Quality and integrity of stretched DNA fibers were checked using the YOYO-1TM Iodide counterstain (Thermo Scientific, Norwood, Australia). Combed coverslips were incubated at 60◦C for 4 h to minimize photobreaking, followed by serial ethanol dehydration (70–100%). Coverslips were hybridized with telomeric C- or G-rich PNA probe [TAMRA-OO-(CCCTAA)<sup>3</sup> or TAMRA-OO-KK(TTAGGG)3] (Panagene, Daejeon, South Korea) in PNA hybridization buffer [70% (v/v) deionized formamide, 0.25% (v/v) NEN blocking reagent (PerkinElmer), 10 mM Tris–HCl (pH 7.5), 4 mM Na2HPO4, 0.5 mM citric acid, and 1.25 mM MgCl2] at room temperature for 12–16 h. Coverslips were washed and counterstained with YOYO-1TM Iodide. Telomere fibers were detected on a Zeiss Axio Imager microscope and analyzed using ZEN v2.3 Pro software (Carl Zeiss, North Ryde, Australia).

## Terminal Restriction Fragment (TRF)

Terminal restriction fragments were obtained from genomic DNA by complete digestion with the restriction enzymes HinfI and RsaI. TRFs were separated by pulsed-field gel electrophoresis. Gels were dried, denatured and subjected to in-gel hybridization with a γ-[32P]-ATP-labeled (CCCTAA)<sup>4</sup> oligonucleotide probe. Gels were washed and the telomeric signal visualized by PhosphorImage analysis (Conomos et al., 2012). TRFs were either visually compared or processed by MultiGauge image analysis software (Fuji Pharma) to quantitate mean telomere length.

### Quantitative PCR (qPCR)

Telomere qPCR was performed relative to the single copy gene 36B4 or HBG (Cawthon, 2002; Lee et al., 2014). Results were expressed relative to U-2 OS and presented as relative telomere content (arbitrary units). Experiments were carried out in a Rotor-Gene Q platform (Qiagen, Maryland, United States) and analyzed using Rotor-Gene 6000 series software (Qiagen, Maryland, United States).

## Quantitative-Fluorescence in situ Hybridization (Q-FISH)

Cells were treated with 10 µg/mL colcemid (Gibco, Grand Island, United States) for 3 h to arrest cells in metaphase, harvested by trypsinization, and resuspended in 75 mM KCl at 37◦C for 20 min. Cells were fixed in methanol:acetic acid (3:1), and chromosome spreads were obtained according to standard cytogenetic methods. Microscope slides were subjected to serial ethanol dehydration (70–100%), followed by hybridization with 0.3 µg/mL Alexa-488-OO-(CCCTAA)<sup>3</sup> telomeric PNA probe (Panagene, Daejeon, South Korea) at 80◦C for 5 min, then at room temperature for 8 h. Slides were washed at 43◦C for 5 min with each of the following solutions: 50% formamide in 2xSSC; 2xSSC; and 0.1% Tween-20/2xSSC. Slides were mounted using ProLongTM Gold antifade with DAPI counterstaining (Invitrogen, Norwood, Australia). One hundred and fifty metaphases from each cell line were scored using MetaSystems software (MetaSystems, North Ryde, Australia). Results were shown as mean telomere intensity (arbitrary units × 1000).

## Flow-Fluorescence in situ Hybridization (Flow-FISH)

Equal numbers (3 × 10<sup>6</sup> ) of control CCRF-CEM cells were mixed with experimental cells prior to denaturation and overnight hybridization at room temperature with 0.3 nM FITC-OO-(CCCTAA)<sup>3</sup> telomeric PNA probe (Panagene, Daejeon, South Korea). Unstained duplicate tubes were run in parallel to determine autofluorescence. Cells were washed twice (70% formamide, 0.1% Tween-20), once (PBS, 0.1% Tween-20), followed by incubation at 4◦C for 3 h with 500 µL of PBS containing 0.1% BSA, RNase A, and DNA binding dye LDS-751 (Santa Cruz Biotech, CA, United States). Data were acquired using FACS-Canto (BD Biosciences, NJ, United States) by acquisition of 1 × 10<sup>4</sup> cells (with and without PNA probe incubation) to calculate mean fluorescence intensity (FI). Data were analyzed using FACSDivaTM software (BD Biosciences, NJ, United States). Cells were gated to include the tetraploid CCRF-CEM cells. Relative telomere length (RTL) was calculated using the following formula: RTL = (mean FI of experimental cells

with PNA probe) – (mean FI of experimental cells without PNA probe)/(mean FI of CCRF-CEM cells with PNA probe) – (mean FI of CEM cells without PNA probe). Results were shown as RTL.

### Semi-Automated Image Analysis for TCA

Two user-friendly semi-automated pipelines were developed using free, open-source platforms for quantification of TCA: CellProfiler v2.2.0 (Jones et al., 2008) and ImageJ v1.8.0 for Windows (NIH, Bethesda, MD, United States). Overall, both pipelines use a similar general approach for semi-automatic detection and size annotation of telomere fiber signals. However, some program-specific adaptations were made to improve the accuracy of telomere fiber identification.

#### CellProfiler Pipeline

Image manipulation (tubeness) was applied to smooth telomere fibers and reduce background. Telomere fibers were subjected to a maximum correlation threshold (MCT) algorithm and selected based on the following parameters: (1) ratio between major and minor axis > 2; (2) eccentricity ≥ 0.75; (3) width > 5 and < 15 pixels; (4) orientation of the major axis of the object is ± 15◦ of vertical relative to X-axis of the image. Telomere fibers that met the selection criteria were refined further by closing small gaps along the length of the fiber. Finally, telomere fiber identification was manually checked for accuracy and corrected, if necessary, using the manually edit objects function.

#### ImageJ Pipeline

Prior to telomere identification a separate pipeline was used to create a binary mask of the image for background detection and refinement. This involved applying a series of open shape selector indicators as disk, horizontal lines and the 45◦ and 145◦ lines, using the MorphoLibJ plugin (v1.4.1). Next, a tubeness processing filter was used to smooth telomere fibers and reduce background. Telomere fibers were then subjected to a user-defined threshold (chosen with a slider bar) and a binary mask was created. Finally, the background mask generated in the first pipeline was subtracted from the newly generated telomere mask image and telomere fibers were further selected by the following parameters: (1) size > 0.250 µm<sup>2</sup> ; (2) circularity from 0 to 0.5. Telomere fibers that met the selection criteria were refined further by closing small gaps along the length of the fiber (MorphoLibJ plugin, v1.4.1).

#### Automated Image Analysis for TCA Using the Molecular Combing Platform

The FiberVision <sup>R</sup> scanner (Genomic Vision, Paris, France) was used to acquire multi-color images of the whole surface of combed coverslips. FiberStudio <sup>R</sup> software (Genomic Vision, Paris, France) was used to detect and measure telomere fiber signals using a custom-designed algorithm. Samples were prepared as described in the Telomere length Combing Assay (TCA) section, with minor modifications. First, coverslips were hybridized with telomeric C-rich PNA probe (Alexa647-OO- (CCCTAA)3; 0.3 µg/mL; Panagene, Daejeon, South Korea). Second, for imaging and analysis purposes, coverslips were mounted in barcoded cartridges (Genomic Vision, Paris, France), and loaded onto the FiberVision <sup>R</sup> scanner. Third, images were automatically acquired at 40×, imported into FiberStudio <sup>R</sup> software, and automatically scored. FiberStudio <sup>R</sup> software allows the scoring to be reviewed, and generates a report containing the telomere length distribution for all regions of interest (ROI).

#### Statistics

Graphs and statistical analysis were generated using GraphPad Prism v5.04. The two-sided Student's t-test was performed on normally distributed data, while the two-sided Mann-Whitney test was performed on data assumed to be non-normally distributed. Linear regression was used to test the correlation between datasets. Further details of statistical analyses are provided in the figures and figure legends.

## RESULTS

## Molecular Combing Can Be Used to Measure Telomere Length

Molecular combing, or DNA fiber analysis, is a powerful technique used to visualize genomic loci and repetitive sequences in the genome (Parra and Windle, 1993). DNA fiber analysis involves extracting DNA from cells such that the chromatin is deproteinized leaving naked DNA, which is then stretched onto a glass microscope slide. This can be combined with FISH for the detection and measurement of specific genomic sequences (Ersfeld, 1994). More recently, DNA fiber analysis has been used to study replication dynamics by tracking the sequential incorporation of halogenated nucleotides into replicating DNA (Norio and Schildkraut, 2001). This technique has been further adapted to analyze the progression of replication forks through telomeric DNA (Sfeir et al., 2009), as well as for the measurement of telomere extension events (Sobinoff et al., 2017; Lu et al., 2019).

An underexplored application of molecular combing is the direct measurement of telomere length (Wang et al., 2013). More recently, a CRISPR-Cas9 nickase system has been used to label telomere tracts, followed by nanochannel array analysis to measure discrete telomere lengths (McCaffrey et al., 2017). We have developed the TCA to measure the distribution of individual telomere lengths in a cell population. A schematic representation of TCA is presented (**Figure 1A**), and a detailed protocol is provided (**Supplementary Material**). To preserve the integrity of the telomeric DNA and to minimize shearing, cells were embedded in agarose plugs, prior to protein digestion to release the DNA from the chromatin scaffold. The DNA was then solubilized. Vinylsilane treated glass coverslips were submerged in the solution and withdrawn mechanically, allowing the DNA molecules to be stretched at a constant stretching factor of 2 kb/µm. The utilization of a constant stretching factor that is both irrespective of sequence content and DNA fiber length, and uniform across the glass surface (Schurra and Bensimon, 2009), obviates the requirement to normalize the fiber length internally, providing that the stretching dynamics are regularly calibrated. DNA fibers were then hybridized to a telomerespecific PNA probe and subjected to DNA counterstaining. The YOYO-1 counterstain was used to provide additional

information pertaining to the terminal location of the telomere, thus minimizing the inclusion of interstitial repeat arrays or extra-chromosomal telomeric repeats (ECTRs) in the telomere length measurements. Finally, telomere fibers were detected by fluorescence microscopy (**Figure 1B**), and the length of the telomere repeat tracts measured using the associated microscope software, or by automated image analysis (**Figure 1C**). TCA was able to accurately measure telomere lengths in human cells or extracted DNA within a range of <1 kb and >80 kb. TCA is a simple and robust technique to measure the telomere lengths of individual DNA molecules, providing the distribution of telomere lengths in a cell population.

#### Comparison of TCA With Established Telomere Length Measurement Techniques

Established telomere length measurement techniques have various strengths and limitations (Aubert et al., 2012; Montpetit et al., 2014; Lai et al., 2018), with most techniques providing a relative or average measurement of length or telomere content. TCA overcomes this limitation by measuring individual telomeres. We measured the distribution of telomere lengths by TCA in four different cell lines (**Figure 2A**). These cell lines were selected based on their well-established differences in telomere length and TMM, with HeLa, and HT1080 having short telomeres and utilizing telomerase as the TMM, and IIICF/c and U-2 OS having longer and heterogeneous telomere lengths and utilizing the ALT pathway of telomere maintenance (**Supplementary Table S1**; Lee et al., 2014). TCA demonstrated different telomere length distribution profiles in the four different cell lines. Mean telomere length measurements were 4.55 kb for HeLa (min: 0.96 kb; max: 14.29 kb), 7.60 kb for IIICF/c (min: 0.87 kb; max: 62.09 kb), 36.11 kb for U-2 OS (min: 1.41 kb; max: 145.10 kb), and 6.95 kb for HT1080 (min: 1.82 kb; max: 16.46 kb; **Figure 2A**). Telomere length analysis by TCA further identified the presence of very long (>80 kb) as well as very short (<1 kb) telomere lengths in U-2 OS and IIICF/c cells, consistent with these cell lines utilizing the ALT mechanism of telomere maintenance (**Figure 2A**). This contrasted with the short homogeneous telomere lengths observed in the two telomerase-positive cell lines (HeLa and HT1080).

We then compared TCA to the most commonly utilized telomere length measurement techniques in the same four cell lines (**Figures 2A–E**). TRF analysis is the gold standard telomere length measurement technique employed by molecular biology labs (Kimura et al., 2010; Mender and Shay, 2015). TRF analysis was used to visualize the intensity and size distribution of TRFs (**Figure 2B**), and the mean telomere length from the contributing cell population was estimated from the size of the heterogeneous telomere smear (**Figure 2B**). Quantitative-FISH (Q-FISH) can be applied to interphase or metaphase cells to provide high resolution measurements of individual

FIGURE 1 | Overview of the Telomere length Combing Assay (TCA). (A) Cells are embedded in agarose plugs, prior to protein digestion. High molecular weight DNA is stretched onto vinylsilane coverslips and hybridized with a telomeric PNA probe. Telomeric fibers are then detected and measured. (B) Representative images of DNA counterstained by YOYO-1 (left panel) and telomere fibers (right panel). Telomere fibers are indicated by white arrows. Scale bars are 5 µm. (C) Examples of telomere fibers. Scale bars are 2 µm.

FIGURE 2 | Comparison of telomere length measurement methods. (A) Distribution of individual telomere lengths measured by TCA in HeLa (telomerase-positive), IIICF/c (ALT), U-2 OS (ALT) and HT1080 (telomerase-positive) cell lines. Tukey boxplots of median from >450 telomeric fibers from n = 3 technical replicates. (B) TRF analysis of telomere length (left panel) and mean telomere length quantitated using MultiGauge image analysis software (right panel). Error bars are mean ± SEM from n = 3 independent experiments. (C) Telomere length measurements by Q-FISH. Results are presented as arbitrary units (×1000). Error bars are mean ± SEM from ≥3,000 telomere signals. (D) Telomere length measurements by flow-FISH. Results are presented as relative telomere length (RTL). Error bars are mean ± SEM from n = 3 independent experiments with 3 technical replicates each. (E) Telomere content analyzed by qPCR relative to the single copy gene 36B4 and normalized to the U-2 OS reference sample. Error bars are mean ± SEM from n = 3 independent experiments with 3 technical replicates each.

telomere lengths (Lansdorp et al., 1996; Lai et al., 2018), while flow-FISH has been extensively validated for clinical diagnostic purposes and employs flow cytometry to detect telomere-bound fluorescently labeled probes in individual viable cells (Rufer et al., 1998; **Figures 2C,D**). Quantitative PCR (qPCR) has been broadly adopted for clinical and epidemiological studies, and benefits from its technical simplicity and the requirement for small amounts of DNA, but produces variable results (Cawthon, 2002; Lai et al., 2018). qPCR of amplified telomeric products was compared to a reference single copy gene, to provide a measure of relative telomere content in the four cell lines (**Figure 2E**).

Overall, we found that telomere lengths were comparable across all five techniques (**Figures 2A–E**), with the U-2 OS cell line consistently displaying the longest and most heterogeneous telomere lengths. The HeLa and HT1080 cell lines consistently displayed short telomere lengths. Some variability regarding the cell line with the shortest telomeres was observed, with flow-FISH indicating that HT1080s had the shortest telomere lengths, in contrast to all other methods, which identified HeLa as having the shortest telomeres. Variability in the RTL of the four cell lines was also observed between methods. TCA and Q-FISH were the only techniques employed here that measure individual telomere lengths. The distribution of telomere lengths measured by TCA and Q-FISH were further compared using telomere length frequency histograms, with TCA demonstrating increased resolution of very long telomeres in the two ALT cell lines compared to Q-FISH (**Supplementary Figure S1**). By directly comparing telomere lengths measured by TCA to measurements made using the other methods, the most tightly correlated techniques were TCA and qPCR (R <sup>2</sup> = 0.93), followed by TCA and Q-FISH (R <sup>2</sup> = 0.91), TCA and TRF (R <sup>2</sup> = 0.80), and TCA and flow-FISH (R <sup>2</sup> = 0.78).

We then aimed to determine the accuracy and reproducibility of TCA, and its versatility to different sample preparations. TCA is typically performed on freshly harvested cells embedded in agarose plugs and in-gel protein digestion to minimize DNA shearing, prior to stretching of DNA fibers. We compared embedded cells versus phenol-chloroform extracted DNA from U-2 OS and HeLa cell lines and from the MRC-5 mortal cell strain (**Supplementary Figure S2**). Telomere length measurements were comparable for HeLa and MRC-5 cells, which have short telomere lengths, whilst significant variability in telomere length was observed between the extracted DNA samples compared to the cell preparation samples for U-2 OS cells (**Supplementary Figure S2**). Specifically, telomere length distributions revealed an under-representation of the longest telomeres in U-2 OS cells, indicative of some level of DNA shearing. This suggests that, whilst TCA is a reproducible and versatile technique, the requirement for embedded cells is more critical when analyzing cells with very long telomeres, in contrast to shorter telomeres, which are less sensitive to DNA processing. This is unlikely to be a concern when analyzing human samples, as only ALT cells with extremely long telomeres were significantly impacted by the skewed representation. Overall, these data support the utility of TCA as an accurate and robust technique to measure the distribution of individual telomere lengths in a cell population.

## TCA Can Be Used to Measure Dynamic Changes in Telomere Length

Telomere attrition and TMMs contribute to telomere length regulation, and telomere length ultimately dictates the proliferative capacity of a cell. To address whether TCA provides sufficient sensitivity to detect dynamic changes in telomere length, we used three separate cell systems. First, we used primary human fibroblasts that undergo telomere attrition at a rate of 50–150 bp per cell division (Harley et al., 1990), until telomere lengths become critically short and the cell reaches senescence. Cellular senescence is characterized by an absence of mitotic cell division and permanent disengagement from the cell cycle (Reddel, 2000). TCA was used to measure telomere lengths in the MRC-5 mortal human lung fibroblast cell strain at early, mid and late population doublings (PDs; **Figure 3A**). Telomere shortening was observed with increasing PDs. Specifically, early and mid PDs demonstrated mean telomere length measurements of 10.55 kb and 9.32 kb, respectively. The late timepoint (PD 60) corresponded with the approximate onset of cellular senescence (Kaul et al., 2012), and had a mean telomere length of 4.52 kb. This demonstrates the sensitivity of TCA to accurately determine telomere length changes of ≥1 kb. Absolute telomere length measurements were used to calculate a rate of attrition of approximately 70 bp/cell division, consistent with normal rates of telomere attrition in cells lacking a TMM. Importantly, TCA is not dependent on mitotic cells for the measurement of individual telomere lengths, allowing senescent cells to be measured by TCA.

Telomere length measurements by TCA were then compared to telomere lengths measured by TRF analysis and by qPCR in MRC-5 cells at early, mid and late PDs (**Figures 3B,C**). All three techniques demonstrated telomere shortening with increasing PD (**Figures 3A–C**); however, the difference in telomere length by TRF was only significant between early and late PD, indicative of the lower sensitivity of TRF analysis to detect dynamic changes in telomere length (**Figure 3B**). Interestingly, absolute telomere length measurements at early and mid-timepoints were similar for TCA and TRF analysis. These two techniques provide a measure of telomere length in kb, although TRFs include variable amounts of subtelomeric regions, dictated by the position of the terminal restriction enzyme site. At the later timepoints telomere lengths appeared shorter when measured by TCA compared to TRF analysis, supporting TCA being the more sensitive technique (**Figures 3A,B**).

Second, we used the previously described HT1080 hTR supertelomerase cell line to measure telomerase-mediated telomere lengthening (Pickett et al., 2009). HT1080 hTR cells that exogenously express hTR display elevated levels of telomerase activity and progressive telomere lengthening over 150 PDs, until telomere lengths plateau (Pickett et al., 2009). TCA identified progressive telomere lengthening in the HT1080 hTR cell line at early (mean telomere length: 11.12 kb) and late (26.59 kb) PDs compared to the parental HT1080 wild-type (5.25 kb) cell line, with the late timepoint corresponding to approximately 90 PDs post stable overexpression of hTR (**Figure 3D**). Telomere lengthening was similarly observed by TRF analysis and by

FIGURE 3 | TCA can accurately measure telomere attrition and telomerase-mediated telomere extension. (A) Telomere length measurements in MRC-5 fibroblasts at early (20), mid (40), and late (60) population doublings (PDs). Scatterplots showing the distribution of ≥150 individual telomere fiber lengths measured by TCA. Error bars represent the mean ± SEM from n = 3 technical replicates, \*\*p < 0.01 and \*\*\*\*p < 0.0001. Student's t-test. (B) TRF analysis (left panel) and quantitation using MultiGauge image analysis software (right panel). Error bars represent the mean ± SEM from n = 2 technical replicates, \*p < 0.05 and n.s. p > 0.05. Student's t-test. (C) Telomere content analyzed by qPCR relative to the single copy gene 36B4 and normalized to the U-2 OS reference sample. Error bars represent the mean ± SEM from n = 3 technical replicates, \*p < 0.05 and \*\*p < 0.01. Student's t-test. (D) Telomere length measurements in HT1080 wild type (wt) and HT1080 hTR at early (10) and late (90) PDs. Scatterplots showing the distribution of ≥150 individual telomeric fiber lengths measured by TCA. Error bars represent the mean ± SEM from n = 3 technical replicates, \*\*\*\*p < 0.0001. Student's t-test. (E) TRF analysis (left panel) and quantitation using MultiGauge image analysis software (right panel). Error bars represent the mean ± SEM from n = 2 technical replicates, \*\*p < 0.01 and \*\*\*p < 0.001. Student's t-test. (F) Telomere content analyzed by qPCR relative to the single copy gene 36B4 and normalized to the U-2 OS reference sample. Error bars represent the mean ± SEM from n = 3 technical replicates, \*\*\*p < 0.001, \*\*\*\*p < 0.0001, and n.s. p > 0.05. Student's t-test.

qPCR (**Figures 3E,F**). TCA was able to provide a more detailed distribution profile of telomere lengths than the telomere length measurements obtained by TRF and qPCR analysis. This was most striking at the later timepoints when telomere lengths were longer and more heterogeneous (**Figure 3D**).

Third, we used TCA to measure the telomere length changes that occur in ALT cells in response to manipulation of SLX4 and BLM levels (Sobinoff et al., 2017). Specifically, we used SLX4 overexpression to promote resolution of telomere recombination intermediates, thereby causing telomere shortening, and BLM overexpression to promote ALT-mediated telomere extension (Sobinoff et al., 2017). Telomere length distribution profiles obtained by TCA demonstrated telomere shortening in response to SLX4 overexpression, and telomere lengthening in response to BLM overexpression in IIICF/c (mean telomere lengths: empty vector: 14.24 kb; BLM + : 20.77; and SLX4 + : 10.92), and U-2 OS (empty vector: 22.73 kb; BLM + : 25.91; and SLX4 + : 12.30) ALT cell lines, but not in HT1080 hTR telomerasepositive cells (empty vector: 10.05 kb; BLM + : 10.92; and SLX4 + : 10.79; **Supplementary Figure S3**), consistent with previous reports (Sobinoff et al., 2017). Notably, TCA was able to provide more precise measurements of the distribution of individual telomere lengths, identifying a striking decrease in long telomeres in the SLX4 overexpressing cells, and a noticeable increase in very long telomeres in the BLM overexpressing cells (**Supplementary Figure S3**). Overall, these data demonstrate that TCA can accurately measure dynamic changes in telomere length, including normal telomere attrition in primary human cells, and both telomerase- and ALT-mediated telomere extension in cancer cells.

### TCA Can Be Used to Measure Telomere Length in the Human Population

Telomere length decreases with advancing age and is considered a biomarker of chronological aging. Nevertheless, inter-individual telomere lengths at any given age are highly variable in the human population (Harley et al., 1990; Aubert and Lansdorp, 2008). To determine whether TCA can detect age-associated decline in telomere length, we accessed DNA extracted from peripheral blood mononuclear cells (PBMCs) from twelve healthy individuals spanning ages 4 to 75 years of age, collected as part of a cohort of 240 healthy individuals used to obtain normal telomere length percentiles for different ages through the Department of Hematology Telomere Length Testing Facility, Sydney Children's Hospitals Network, Australia. TCA provided a distribution of telomere lengths for each individual, demonstrating an overall decrease in telomere length with chronological age (**Figure 4A**).

Telomere biology disorders are premature aging diseases caused by abnormally short telomeres, with lengths typically falling below the bottom percentile of the normal healthy population. The clinical characteristics of TBDs are diverse, but patients frequently develop bone marrow failure and idiopathic pulmonary fibrosis (Mangaonkar and Patnaik, 2018). Telomere length analysis is used to identify individuals suspected of having TBDs, but can also provide prognostic information regarding disease onset and clinical manifestations, as well as directing treatment regimens (Alder et al., 2018). We used TCA to measure the telomere lengths of five individuals (ages 1, 1.5, 14, 45, and 48 years old) that had previously been diagnosed with TBDs, using DNA extracted from PBMCs. TCA analysis showed that these individuals all had strikingly short telomere length distribution profiles, and that their mean telomere lengths were substantially shorter than those of healthy individuals (**Figure 4A**).

We then compared telomere length measurements by TCA to telomere lengths measured by flow-FISH and qPCR for the healthy individuals and the TBD patients. Normal telomere length percentiles for the cohort of 240 healthy individuals were included for the flow-FISH and qPCR datasets (**Figures 4B,C**). Telomere lengths for the healthy individuals were somewhat variable across the three different techniques, but an overall decline in telomere length with chronological age was always observed (**Figures 4A–C**). TCA correlated well with qPCR (R <sup>2</sup> = 0.77), but not as well with flow-FISH (R <sup>2</sup> = 0.52), for this specific dataset (**Figure 4D**). This is consistent with previous reports that indicate variable correlations between different telomere length measurement methods (Khincha et al., 2017). All three techniques identified the TBD patients as having abnormally short telomeres (**Figures 4A–C**), falling below the tenth percentile of the healthy cohort. These data demonstrate that TCA can be used to measure telomere lengths in PBMCs from healthy individuals. Importantly, TCA can reliably identify patients with critically short telomere lengths, supporting a potential practical application for TCA in the diagnosis and treatment of TBDs.

## Software Automation for TCA Quantification

Detection and measurement of telomere fibers was performed manually using image manipulation tools. Each technical replicate generated an average of 300 individual telomere fibers to be scored. In order to adapt TCA for high-throughput applications, we developed pipelines for the semi-automated detection and size annotation of telomere lengths using the open-source image processing platforms CellProfiler and ImageJ, and we evaluated fully automated telomere length measurement using FiberStudio <sup>R</sup> automated analysis and reporting software, in collaboration with Genomic Vision (France).

The detection and analysis pipelines for CellProfiler and ImageJ are presented (**Supplementary Figures S4A,B**). Tubeness processing allows both CellProfiler and ImageJ algorithms to close small gaps along the identified telomeric fiber. Both CellProfiler and ImageJ provide an optional step for the user to check and modify identified fibers. This was employed here to validate the pipelines. Cell Profiler and ImageJ pipelines require manual imaging of the telomere fibers through microscope tiling image manipulation tools and are, therefore, semi-automated pipelines. In contrast, the FiberVision <sup>R</sup> scanner automatically images the slide, and custom-designed FiberStudio <sup>R</sup> software detects, quantifies, and annotates the telomere fibers, providing a fully automated platform.

to 75 years and five TBD patients. Scatterplots showing the distribution of ≥150 individual telomeric fiber lengths measured by TCA. Error bars are mean ± SEM from n = 2 technical replicates (left panel). Mean telomere length determined by TCA plotted against age for healthy individuals and TBD patients. Annotated numbers indicate mean telomere length (kb; right panel). (B) RTL measured by flow-FISH plotted against age for healthy individuals and TBD patients. Lines indicate the 1st, 10th, 50th, 90th, and 99th telomere length percentiles from a cohort of 240 normal controls. (C) Telomere content analyzed by qPCR. Ratio between telomere content and single copy gene HBG plotted against age for healthy individuals and TBD patients. Lines indicate the 1st, 10th, 50th, 90th, and 99th telomere length percentiles from a cohort of 240 normal controls. (D) Linear regression of TCA against flow-FISH (R <sup>2</sup> = 0.52) and qPCR (R <sup>2</sup> = 0.77) for healthy individuals and TBD patients. Spearman correlation test.

While differences in image manipulation and telomere signal detection exist between the two semi-automated programs, both CellProfiler and ImageJ performed similarly, and produced comparable measurements of mean telomere fiber length to the measurements obtained from manual analysis in the HeLa, IIICF/c, U-2 OS, and HT1080 cell lines (**Figure 5**). FiberStudio <sup>R</sup> also demonstrated accuracy in measuring telomere lengths from HeLa, IIICF/c, U-2 OS, and HT1080 cell lines when compared to manual scoring (**Figure 5**).

To demonstrate the accuracy of TCA, we measured the same fifty fibers from HeLa cells three separate times. Measurements were made manually, and each replicate

was scored independently. Telomere length measurements demonstrated a very high level of accuracy between replicates (**Supplementary Figure S5A**).

Finally, to determine the potential utility of automated TCA analysis, we directly compared the lengths of fifty telomere fibers from HeLa cells measured manually to the lengths of the same fifty telomere fibers measured using CellProfiler, ImageJ, and FiberStudio <sup>R</sup> . The individual telomere length measurements obtained by manual assessment correlated similarly with those obtained using CellProfiler (R <sup>2</sup> = 0.80) and ImageJ (R <sup>2</sup> = 0.81), and correlated slightly better with measurements obtained using FiberStudio <sup>R</sup> (R <sup>2</sup> = 0.84; **Supplementary Figure S5B**). These results demonstrate that TCA is suitable for both semi-automated and fully automated image analysis. The successful application of CellProfiler, ImageJ, and FiberStudio <sup>R</sup> to the detection and accurate measurement of telomere fibers, with high correlation to manual measurements, demonstrates the utility of these programs in providing an unbiased approach for the detection and measurement of individual telomere fibers, with potential high-throughput capabilities.

#### DISCUSSION

We have developed TCA as a simple and accessible method to accurately measure the distribution of individual telomere lengths in a cell population. TCA involves stretching DNA fibers onto coated glass coverslips using a constant stretching factor. This achieves unbiased and uniform stretching of millions of DNA molecules, which can then be probed with telomere-specific PNA probes for the detection and measurement of individual telomere lengths. The major strengths of TCA are its simplicity, adaptability to image automation for high-throughput purposes,

lack of requirement for cycling cells, and ability to accurately detect all telomere lengths in a cell population, including very short (<1 kb), and very long (>80 kb) telomeres. The main limitations of TCA are that it is relatively time-consuming and expensive (approximately twice the cost of flow-FISH), and relies on specialized molecular combing equipment and consumables.

By directly comparing TCA to other widely utilized telomere length measurement techniques (TRF, Q-FISH, flow-FISH, and qPCR), we demonstrated that TCA is sensitive and accurate for telomere detection and length measurement. TCA provides important and relevant information regarding the distribution of individual telomere lengths in a cell population. This is not achievable using current methods, which include TRF for molecular research, flow-FISH for clinical telomere length measurement applications and the diagnosis of TBDs, and qPCR for large-scale clinical and epidemiological applications. Other techniques, including Single TElomere Length Analysis (STELA), universal STELA (U-STELA; Bendix et al., 2010), and Telomere Shortest Length Assay (TeSLA; Lai et al., 2017), are able to detect individual telomere lengths, but include a PCR step that precludes the amplification and detection of very long telomeres, and are somewhat limited by their technical complexity (Lai et al., 2017, 2018).

TCA is versatile and can be applied reliably to both freshly harvested cells and extracted DNA, although the use of agaroseembedded cells is preferable for the measurement of very long telomeres to prevent shearing through mechanical manipulation. The requirement for isolated DNA as the starting material means that telomere length measurements by TCA are not dependent on cycling cells, in contrast to flow-FISH and Q-FISH methods. This is particularly important for the analysis of cells with a low mitotic index or cells approaching or at senescence (Kaul et al., 2012). In addition, the removal of nucleosomes by protease digestion leaves the DNA naked and amenable for stretching. This means that telomere length measurements are not affected by aberrant chromatin compaction that has the propensity to influence the binding of telomere FISH probes (Bandaria et al., 2016; Timashev et al., 2017; Vancevska et al., 2017).

TCA provides absolute measurements of individual telomere lengths, rather than an estimate of RTL or content. The telomere length distribution profile is representative of the range of telomere lengths in the cell population, and absolute measurements provide precise information regarding telomere length. In this study, we were able to measure telomere lengths as short as 0.87 kb and as long as 145.1 kb. The proportion of very short or very long telomeres in a cell population is biologically relevant (Aubert and Lansdorp, 2008; Baird, 2008a,b; Muntoni et al., 2009; Stanley and Armanios, 2015), but is typically not considered in clinical, epidemiological and research-based investigations. TCA provides the means to interrogate telomere length associations in far more detail than currently achievable.

TCA is sufficiently sensitive to detect dynamic changes in telomere length. Specifically, we measured telomere attrition in primary human fibroblasts, telomere lengthening in response to increased telomere maintenance by telomerase or ALT, and telomere shortening following the inhibition of telomere maintenance by ALT. TCA provided additional biological insight

into these processes by uncovering the changes that occurred at the telomere length extremes, as well as identifying the overall shift in individual telomere length distribution profiles. The sensitivity of TCA to detect these intricacies will provide a more complete understanding of how telomere length drives replicative senescence and tumorigenic escape, as well as how inhibition of TMM pathways impacts telomere length and cell proliferation.

Telomere length measurements are increasingly being used as an indicator of lifetime health, to assess risk of many age-associated diseases, and to diagnose TBDs (Epel et al., 2004; Sampson and Hughes, 2006; Willeit et al., 2010; Haycock et al., 2014; Bertuch, 2016; Opresko and Shay, 2016; Khincha et al., 2017; Maciejewski and de Lange, 2017; Alder et al., 2018; Mangaonkar and Patnaik, 2018). We demonstrated that TCA can be used to measure age-associated telomere length in PBMCs isolated from healthy individuals spanning over 70 years and can accurately detect short telomeres in patients with TBDs. Interestingly, the telomere length distribution profiles obtained from patients with TBDs were very distinctive, showing very short and homogeneous telomere lengths. This supports the utility of TCA as an improved telomere length measurement technique for clinical testing and diagnosis.

We also demonstrated that TCA is readily amenable to automated image analysis. Specifically, we implemented two open-source software pipelines using CellProfiler and ImageJ, to semi-automatically measure the manually imaged TCA samples from HeLa, IIICF/c, U-2 OS, and HT1080 cell lines. Both pipelines were comparable to manual scoring. We then compared fully automated TCA using the FiberVision <sup>R</sup> automated scanner and accompanying custom-designed FiberStudio <sup>R</sup> software, and demonstrated an even higher correlation to manual scoring. Importantly, automation precludes bias and subjectivity, and enables rapid and accurate analysis of extensive datasets.

In summary, TCA is a simple and precise method for the measurement of telomere length. TCA has the potential to supersede currently available methods by fulfilling several advantages. First, TCA measurements provide additional relevant biological insight by measuring the distribution of individual telomere lengths, including both telomere length extremes. Second, TCA is versatile in terms of the sample preparation and starting material required. Third, TCA is adaptable to automated image analysis, which precludes bias and will facilitate highthroughput applications. TCA may be further improved by the inclusion of subtelomeric probes to enable the measurement of chromosome-specific telomeres, and the combined use of variant telomere repeat probes as an indicator of telomere directionality and integrity. Automated TCA represents an exciting and

#### REFERENCES

Alder, J. K., Hanumanthu, V. S., Strong, M. A., DeZern, A. E., Stanley, S. E., Takemoto, C. M., et al. (2018). Diagnostic utility of telomere length testing in a hospital-based setting. Proc. Natl. Acad. Sci. U.S.A. 115, E2358–E2365. doi: 10.1073/pnas.1720427115

promising telomere length measurement technique for clinical and research applications.

#### DATA AVAILABILITY STATEMENT

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

#### ETHICS STATEMENT

The studies involving human participants were reviewed and approved by Human Research Ethics Committee of the Sydney Children's Hospitals Network. Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.

#### AUTHOR CONTRIBUTIONS

HP, JA, and VK conceived the idea and designed the experiments. VK, JA, CN, AS, ML, TK, and RV performed the experiments and analyzed the data. HP and VK wrote the manuscript. All authors contributed to the manuscript and approved the submitted version.

#### FUNDING

This work was supported by the National Health and Medical Research Council Australia (1162886 to HP), Cancer Australia (1160171 to VK and 1099299 to AS), and Cancer Institute of New South Wales (ECF171269 to AS).

#### ACKNOWLEDGMENTS

The authors would like to acknowledge their collaboration with Genomic Vision, Bagneux, France. Telomere qPCR and microscopy were performed in the ACRF Telomere Analysis Centre, supported by the Australian Cancer Research Foundation and the Ian Potter Foundation.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2020.00493/ full#supplementary-material




**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 Kahl, Allen, Nelson, Sobinoff, Lee, Kilo, Vasireddy and Pickett. 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.

# rDNA Chromatin Activity Status as a Biomarker of Sensitivity to the RNA Polymerase I Transcription Inhibitor CX-5461

Jinbae Son1,2, Katherine M. Hannan3,4, Gretchen Poortinga1,2,5, Nadine Hein<sup>3</sup> , Donald P. Cameron<sup>3</sup> , Austen R. D. Ganley<sup>6</sup> , Karen E. Sheppard1,2,4 , Richard B. Pearson1,2,4,7, Ross D. Hannan1,2,3,4,7,8 \* † and Elaine Sanij1,2,9 \* †

Edited by:

Andrew Burgess, Anzac Research Institute, Australia

#### Reviewed by:

Tom Moss, Laval University, Canada Wenbing Xie, Johns Hopkins Medicine, United States Jaclyn Elizabeth Quin, Stockholm University, Sweden

#### \*Correspondence:

Ross D. Hannan ross.hannan@anu.edu.au Elaine Sanij elaine.sanij@petermac.org †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Cell Growth and Division, a section of the journal Frontiers in Cell and Developmental Biology

Received: 16 December 2019 Accepted: 15 June 2020 Published: 03 July 2020

#### Citation:

Son J, Hannan KM, Poortinga G, Hein N, Cameron DP, Ganley ARD, Sheppard KE, Pearson RB, Hannan RD and Sanij E (2020) rDNA Chromatin Activity Status as a Biomarker of Sensitivity to the RNA Polymerase I Transcription Inhibitor CX-5461. Front. Cell Dev. Biol. 8:568. doi: 10.3389/fcell.2020.00568 <sup>1</sup> Peter MacCallum Cancer Centre, Melbourne, VIC, Australia, <sup>2</sup> Sir Peter MacCallum Department of Oncology, The University of Melbourne, Parkville, VIC, Australia, <sup>3</sup> ACRF Department of Cancer Biology and Therapeutics, The John Curtin School of Medical Research, Australian National University, Canberra, ACT, Australia, <sup>4</sup> Department of Biochemistry and Molecular Biology, The University of Melbourne, Parkville, VIC, Australia, <sup>5</sup> Department of Medicine, St. Vincent's Hospital, The University of Melbourne, Parkville, VIC, Australia, <sup>6</sup> School of Biological Sciences, The University of Auckland, Auckland, New Zealand, <sup>7</sup> Department of Biochemistry and Molecular Biology, Monash University, Clayton, VIC, Australia, <sup>8</sup> School of Biomedical Sciences, The University of Queensland, Brisbane, QLD, Australia, <sup>9</sup> Department of Clinical Pathology, The University of Melbourne, Parkville, VIC, Australia

Hyperactivation of RNA polymerase I (Pol I) transcription of ribosomal RNA (rRNA) genes (rDNA) is a key determinant of growth and proliferation and a consistent feature of cancer cells. We have demonstrated that inhibition of rDNA transcription by the Pol I transcription inhibitor CX-5461 selectively kills tumor cells in vivo. Moreover, the firstin human trial of CX-5461 has demonstrated CX-5461 is well-tolerated in patients and has single-agent anti-tumor activity in hematologic malignancies. However, the mechanisms underlying tumor cell sensitivity to CX-5461 remain unclear. Understanding these mechanisms is crucial for the development of predictive biomarkers of response that can be utilized for stratifying patients who may benefit from CX-5461. The rDNA repeats exist in four different and dynamic chromatin states: inactive rDNA can be either methylated silent or unmethylated pseudo-silent; while active rDNA repeats are described as either transcriptionally competent but non-transcribed or actively transcribed, depending on the level of rDNA promoter methylation, loading of the essential rDNA chromatin remodeler UBF and histone marks status. In addition, the number of rDNA repeats per human cell can reach hundreds of copies. Here, we tested the hypothesis that the number and/or chromatin status of the rDNA repeats, is a critical determinant of tumor cell sensitivity to Pol I therapy. We systematically examined a panel of ovarian cancer (OVCA) cell lines to identify rDNA chromatin associated biomarkers that might predict sensitivity to CX-5461. We demonstrated that an increased proportion of active to inactive rDNA repeats, independent of rDNA copy number, determines OVCA cell line sensitivity to CX-5461. Further, using zinc finger nuclease genome editing we identified that reducing rDNA copy number leads to an increase in the proportion of active rDNA repeats and confers sensitivity to CX-5461 but also induces genome-wide instability and sensitivity to DNA damage. We propose that the proportion of active to inactive rDNA repeats may serve as a biomarker to identify cancer patients who will benefit from CX-5461 therapy in future clinical trials. The data also reinforces the notion that rDNA instability is a threat to genomic integrity and cellular homeostasis.

Keywords: RNA polymerase I, CX-5461, ovarian cancer, DNA damage response, rDNA copy number

#### INTRODUCTION

fcell-08-00568 July 2, 2020 Time: 16:19 # 2

Transcription of the rDNA repeats by Pol I within the nucleoli is a critical step in ribosome biogenesis and accounts for over 60% of all cellular transcription (Warner, 1999; Moss and Stefanovsky, 2002; Moss et al., 2007). The rDNA encodes the 47S pre-rRNA precursor of the 18S, 5.8S, and 28S rRNAs, which together with 5S rRNA constitute the RNA component of ribosomes. The rDNA is organized into large blocks of tandem repeats, with 400–600 repeats divided among the five pairs of acrocentric chromosomes in the human genome (Stults et al., 2009). Modulation of transcriptional rates can be achieved by regulating Pol I transcription initiation, elongation, and/or processing of the 47S rRNA precursor (Stefanovsky et al., 2001; Schneider, 2012; Goodfellow and Zomerdijk, 2013; Hung et al., 2017). Remarkably, only a fraction of the rDNA repeats are actively transcribed at any one time, providing an additional layer of regulation with transcription output being determined at two levels: the active copy number in combination with the rate of transcription per rDNA repeat (Sanij et al., 2008; Zentner et al., 2011; Hamdane et al., 2014). However, the majority of short-term regulation affects rDNA transcription rate through changing the rate of transcription from active genes, reviewed in McStay and Grummt (2008), Sanij and Hannan (2009), Goodfellow and Zomerdijk (2013). In mammalian cells, the rDNA chromatin can exist in active or inactive states (**Figure 1A**) [reviewed in Sanij and Hannan (2009), Hamperl et al. (2013), Poortinga et al. (2014), Moss et al. (2019), Potapova and Gerton (2019), Warmerdam and Wolthuis (2019)]. Active rDNA repeats are defined as open/accessible chromatin structures, bound by the upstream binding factor (UBF), which is essential for decondensing rDNA chromatin and determining the active rDNA state (Sanij et al., 2008; Hamdane et al., 2014). Active repeats can be either transcriptionally active or transcriptionally competent but nontranscribed, depending on cell cycle phase, cellular signaling, nutrient availability and/or stress stimuli (Schneider, 2012; Xie et al., 2012; Goodfellow and Zomerdijk, 2013; Zhao et al., 2016; Hung et al., 2017). Inhibition of Pol I transcription by loss of the initiation factor RRN3 or upon treatment with the selective inhibitor CX-5461 has no effect on UBF binding nor the proportion of active to inactive rDNA ratio (Quin et al., 2016; Herdman et al., 2017). Thus, UBF binding, not transcription, establishes the active rDNA fraction consistent with (Moss et al., 2019). Inactive rDNA repeats, which lack UBF binding, can be CpG methylated at the rDNA promoter and stably silenced, or non-methylated and hence be in a "pseudo-silent" state (Sanij et al., 2008; Sanij and Hannan, 2009; Hamperl et al., 2013). UBF binding/release is critical for the conversion between active/inactive rDNA repeats, termed rDNA class switching (Sanij et al., 2008; Hamdane et al., 2014; Diesch et al., 2019).

Dysregulated rDNA transcription is linked to a diverse range of human disorders including cancer (Hannan K.M. et al., 2013; Valori et al., 2020). This link is underscored by the modulation of rDNA transcription and rDNA silencing during differentiation and tumorigenesis (Conconi et al., 1989; Sanij et al., 2008; Grummt and Langst, 2013; Hamperl et al., 2013; Hayashi et al., 2014; Savic et al., 2014; Woolnough et al., 2016; Diesch et al., 2019; Prakash et al., 2019). Increased rDNA transcription is a well-known hallmark of cancer. We have shown that a high rDNA transcription rate is required for the oncogene MYC to drive transformation (Chan et al., 2011; Bywater et al., 2012). Thus, rather than being a housekeeping process, rDNA transcription is a key driver of oncogenesis (Hein et al., 2013). Interestingly, variation in the number of rDNA repeats has also been associated with cancer (Stults et al., 2009). This variation in copy number is due to the tandem arrangement of the rDNA rendering it susceptible to recombination, which can result in rDNA instability, such as the loss and gain of rDNA copies. Intriguingly, spontaneous alterations in rDNA organization were over 100-fold elevated in cells lacking Bloom Syndrome protein, a RECQ helicase involved in homologous recombination (HR) DNA repair, and 10-fold elevated in cells lacking ATM (ataxia-telangiectasia, mutated) compared with wild-type controls (Killen et al., 2009). These rDNA alteration phenotypes seem to correlate with the increased cancer predisposition reported in Bloom syndrome and ataxia-telangiectasia patients (Killen et al., 2009). These results suggest that rDNA instability may mediate the predisposition for cancer progression. It has been proposed that reduction of rDNA silencing and rDNA instability underpin global genomic instability, and that this can drive the etiology and progression of cancer (Diesch et al., 2014; Wang and Lemos, 2017; Xu B. et al., 2017; Udugama et al., 2018; Valori et al., 2019). Nevertheless, it appears that increased rRNA production can be achieved even if the rDNA copy number is reduced (Wang and Lemos, 2017), which can occur by regulating the rate of Pol I transcription per rDNA repeat and/or the number of active rDNA repeats.

Pol I transcription is a therapeutic target for small anticancer drugs (Bywater et al., 2012; Hein et al., 2013; Peltonen et al., 2014). The first-in-class Pol I transcription inhibitor, CX-5461 is a promising cancer therapy as a single agent and in combination therapy in pre-clinical models of lymphoma, acute myeloid leukemia, prostate and ovarian cancer (Bywater et al., 2012; Devlin et al., 2016; Rebello et al., 2016; Hein et al., 2017; Yan et al., 2019; Sanij et al., 2020). Recently, the sensitivity profile of CX-5461 was shown to closely resemble a topoisomerase II (TOP2) poison (Olivieri et al., 2019; Bruno et al., 2020). TOP2a is an essential component of the Pol I pre-initiation complex (Ray et al., 2013) and while CX-5461 demonstrates highly selective

primers (Supplementary Table S3) specific to the external transcribed spacer 5'ETS. Expression levels in each cell line were normalized to Vimentin mRNA and expressed as fold change relative to TOV112D cells. Each dot represents the mean value of n = 3 biologically independent experiments per cell line (Individual data points are provided in Sanij et al., 2020). Error bars represent mean ± SD. Statistical analysis was performed using a two-tailed unpaired t test. (D) OVCA cell lines display a range of rDNA content. Top panel, a schematic of the transcribed region of the rDNA repeat detailing the position of the probe used for Southern blotting and a representative Southern blot of gDNA from 15 OVCA cell lines, ordered by increasing GI50 values of CX-5461 as reported in Sanij et al. (2020). gDNA was isolated from 10<sup>6</sup> cells and Southern blotting for rDNA performed and quantitated using ImageQuant pTL software (GE Healthcare). Graph represents rDNA dosage expressed as fold over that for TOV112D; n = 3; mean ± SEM. CX-5461-sensitive cell lines are indicated in red and resistant cell lines are in blue. Statistical analysis was performed using two-sided one-way ANOVA Dunnett's multiple comparisons test. (E) Correlation analysis of rDNA dosage and sensitivity to CX-5461 (GI50) in 15 OVCA cell lines using GraphPad Prism. Error bars represent mean ± SD. Pearson's r is -0.16, NS (not significant) denotes p-value > 0.05; Spearman's rho is −0.29, NS (not significant) denotes p-value > 0.05.

inhibition of Pol I transcription initiation, it is possible that it does so, in part, by trapping TOP2a at the rDNA repeats. Importantly, our first-in-human trial of CX-5461 in patients with advanced hematological cancers established on-target efficacy in targeting Pol I transcription and demonstrated single-agent anti-tumor activity in p53 wildtype and p53-mutant hematologic malignancies (Khot et al., 2019). CX-5461 is also in Phase I clinical trial in solid tumors and has shown preliminary activity in patients with HR deficiency tumors (Xu H. et al., 2017; Hilton et al., 2020). In order to maximize the clinical impact and success of CX-5461, it is crucial to identify biomarkers of response to enable patient stratification.

CX-5461 induces the p53-dependent "nucleolar stress response" (Bywater et al., 2012; Devlin et al., 2016; Pelletier et al., 2018) and a p53-independent checkpoint induced by targeted activation of the DNA-damage response (DDR) upon the induction of chromatin defects and replication stress at the rRNA genes (Quin et al., 2016; Sanij et al., 2020). Activation of each checkpoint results in different cellular phenotypes depending on cell type and cellular context (Yan et al., 2017; Sanij et al., 2020). The nucleolar stress response results in p53-mediated G1/S arrest and apoptosis, with apoptosis being the major response in MYC-driven lymphoma. The p53-independent checkpoint results in S and G2/M phase cell cycle arrest and is the predominant phenotype of solid tumor cells (Quin et al., 2016; Sanij et al., 2020). CX-5461 has been shown to activate ATM/ATR (ataxia telangiectasia and RAD3 related) signaling and a G2/M cell cycle checkpoint in ovarian cancer (OVCA) cells with differential sensitivities observed across a panel of 32 OVCA cell lines (Sanij et al., 2020). OVCA cell lines with higher rates of Pol I transcription are more sensitive to CX-5461 (Sanij et al., 2020). Thus, rDNA copy number and rDNA chromatin status may function as biomarkers of response to inhibition of Pol I transcription.

Here we investigated whether alterations in rDNA copy number and changes in the proportion of active to inactive rDNA repeats correlate with sensitivity to CX-5461 across a panel of OVCA cell lines. We found that an increase in the proportion of active rDNA repeats correlates with increased OVCA cell sensitivity to CX-5461. Further, deleting rDNA copies led to an increase in the proportion of active rDNA repeats, which also correlated with increased sensitivity to CX-5461 and genomewide instability. Therefore, we propose that an increased fraction of active rDNA repeats is a potential biomarker of response to CX-5461 therapy. Our data also demonstrates that deleting rDNA copies is associated with increased sensitivity to DNA damage highlighting the strong interplay between rDNA and genome-wide instability.

#### MATERIALS AND METHODS

#### Cell Culture

Individuality and the identity of OVCA cell lines (listed in **Supplementary Tables S1, S2**) were confirmed by a PCR-based short tandem repeat (STR) analysis using six STR loci. Cell lines were maintained in culture (**Supplementary Table S1**) for a maximum of 8–10 weeks. CX-5461 was purchased from Synkinase and 10 mM stocks were prepared in 50 mM NaH2PO4. Proliferation time course and growth curves for the OVCA cell lines were obtained by assessing cell confluency using the Incucyte ZOOM (Essen Instruments) imaging system. Doubling time for each cell line was calculated using non-linear fit of exponential growth using GraphPad prism software.

#### 47S rRNA Expression

For 47S rRNA expression analysis, cells were lysed, RNA extracted, and first-strand cDNA synthesized using random hexamer primers and Superscript III (Invitrogen). Quantitative reverse transcription PCR (qRT-PCR) was performed in triplicate using FAST SYBR Green on the StepOnePlus real-time PCR system (Applied Biosystems, United States). Primer sequences are listed in **Supplementary Table S3**. Measurement of baseline (basal) rDNA transcription rates of exponentially growing OVCA cell lines was reported in Sanij et al. (2020), determined by qRT-PCR analysis using primers specific to the external transcribed spacer 5'ETS (ETS2). Expression levels in each cell line were normalized to Vimentin mRNA and expressed as fold change relative to TOV112D cells. The mean and standard error of the mean (SEM) values of n = 3 biologically independent experiments per cell line were utilized in this study.

#### ZFN Gene Editing

Zinc Finger Nucleases (ZFNs) induce double strand DNA breaks (DSBs) at a specific target region, recognized by a zinc finger DNA-binding domain fused with DNA-cleavage domain Gaj et al. (2013). TOV112D cells were transduced with Lentiviruses expressing empty vector or co-transduced with two ZFN targeting rDNA sequences (**Supplementary Figure S1**) followed by selection with puromycin for 5 days, then FACS to generate clonal cell lines.

#### Measurement of rDNA Copy Number

qPCR analysis of 100 ng of genomic DNA (gDNA) was performed in triplicate using FAST SYBR Green on the StepOnePlus realtime PCR system (Applied Biosystems, United States). Primer sequences are listed in **Supplementary Table S3**. Changes in abundance were to normalized to corresponding Vimentin levels as a single copy locus control and expressed as fold change relative to TOV112D by 2(−11CT).

For quantification using Southern blotting, gDNA was isolated from 10<sup>6</sup> cells, digested with SalI, and separated on a 0.9% agarose gel, and alkaline southern blotting was performed. rDNA was visualized using a <sup>32</sup>P (Amersham) labeled probe (+1601-2089 base pair relative to transcription start site) within the 5'ETS (external transcribed spacer) region of rDNA and binding detected using a Phosphorimager (GE Healthcare). Signal quantitation was performed using ImageQuant (TLv2005.04; GE Healthcare).

#### Psoralen Cross-Linking Assay

Cells were lysed in 10 mM Tris–HCl, pH 7.4, 10 mM NaCl, 3 mM MgCl2, and 0.5% NP-40, and nuclei were pelleted, resuspended

Son et al. Active rDNA Chromatin and CX-5461

in 50 mM Tris–HCl, pH 8.3, 40% glycerol, 5 mM MgCl2, and 0.1 mM EDTA, and irradiated in the presence of 4,5,8<sup>0</sup> trimethylpsoralen (Sigma-Aldrich) with a 366 nm UV light box at a distance of 6 cm (Conconi et al., 1989). 200 µg/ml psoralen was added at 1:20 dilution every 4 min for a total irradiation time of 20 min. gDNA was isolated, digested with SalI, and separated on a 0.9% agarose gel, and alkaline Southern blotting was performed. To reverse psoralen cross-linking, filters were treated with 254 nm UV at 1,875 × 100 µJ/cm<sup>2</sup> using a UV cross-linker (Stratalinker 2400; Agilent Technologies). The membrane was then hybridized to a purified <sup>32</sup>P (Amersham) labeled rDNA probe (+ 1601-2089), visualized by scanning on a PhosphoImager (GE Healthcare), and quantitated using ImageQuant (TLv2005.04; GE Healthcare).

## Chromatin-Immunoprecipitation (ChIP)

Chromatin-immunoprecipitation was performed as described previously (Poortinga et al., 2004; Sanij et al., 2008). Cross-linking was achieved with 0.6% formaldehyde and assays performed using 10<sup>6</sup> cells per IP. For all ChIPs, 8 µl of sera was used per IP. Samples were analyzed in triplicate using the FAST SYBR green dye on the StepOnePlus real- time PCR system (Applied Biosystems). To calculate the percentage of total DNA bound, unprecipitated input samples from each condition were used as a reference for all qPCR reactions. Primer sequences are listed in **Supplementary Table S3**.

### Immunofluorescence (IF)

Cells were fixed in 4% paraformaldehyde (10 min at room temperature), permeabilized with 0.3% Triton X-100 in PBS for 10 min on ice, washed with PBS, and blocked with 5% goat serum and 0.3% Triton X-100 in PBS for 30 min. Cells were sequentially incubated with the primary and secondary antibody (**Supplementary Table S4**), then washed with PBS. Nuclei were counterstained with DAPI in VECTASHIELD mounting media (Vector Labs). Images were acquired on an Olympus BX-61 microscope equipped with a Spot RT camera (model 25.4), using the UPlanAPO 60X, NA 1.2 water immersion objective and Spot Advanced software (v.4.6.4.3). The gamma adjust and background subtract settings for adjusting the image after acquisition were identical for all images.

#### rDNA Fluorescent in situ Hybridization (FISH)

Following performing IF, slides were fixed in methanol:acetic acid (3:1) for 5 min at room temperature then dehydrated through a 70%–80% ethanol series. Slides were denatured in 70% formamide/2XSSC (saline-sodium citrate) or 10 min at 83◦C, then dehydrated through the ethanol series and airdried. Probes derived from the intergenic spacer of the human ribosomal gene repeat provided by Prof. Brain McStay, NUI Galway. 100 ng of denatured biotin-labeled probe were combined with 30 µg salmon sperm DNA and 18 µg Cot1 carrier DNA (Invitrogen) in 2XSSC with 50% formamide and 20% dextran sulfate and added per slide then hybridized at 37◦C for 24 h in a humidified chamber. Slides were washed in 50% formamide/2XSSC at 42◦C for 15 min and 0.1XSSC at 60◦C for 15 min. Streptavidin-Alexa fluor 488 was added for 1 hr at 37◦C and the slides then washed in 0.05% Tween-20/4XSSC for 15 min. Slides were mounted in DAPI. Images were acquired on an Olympus BX-61 microscope as described above.

#### COMET Assay

Cells were collected and processed as described in the manufacturer's protocol (Trevigen, Comet Assay 4250-050-K). Images were acquired on an Olympus BX-61 microscope using the Olympus UPlanAPO 203, NA 1.2 water immersion objective as described above.

#### Statistical Analysis

Pearson correlation coefficient, Spearman's rank correlation coefficient, one-way ANOVA multiple tests and Student's t-test were employed as indicated in figure legends.

## RESULTS

#### A Higher Proportion of Active rDNA Repeats Correlates With OVCA Cell Sensitivity to CX-5461

Our aim was to characterize rDNA features (**Figure 1B**) that correlate with sensitivity to CX-5461 in order to identify possible predictive biomarkers of response to CX-5461. To do this, we employed a panel of established human OVCA cell lines from a range of histological subtypes. We previously reported that the concentration of drug that induces a 50% reduction in cell proliferation (GI50) varied profoundly between individual OVCA cell lines, and these cell lines were defined as resistant or sensitive to CX-5461 if the GI50 was above or below the geometric GI50 median of 360 nM, respectively (Sanij et al., 2020). We also reported that CX-5461 sensitivity correlates with basal rDNA transcription rate (i.e., rDNA transcription rate of exponentially growing cells) (Sanij et al., 2020) (shown here in **Figure 1C**). Specifically, we demonstrated that sensitive OVCA cell lines exhibited higher rDNA transcription rates than resistant lines. In this study we extend on the earlier work of Sanij et al. (2020) to assess whether differences in basal rDNA transcription rate were associated with rDNA copy number, and therefore whether copy number may explain OVCA cell line sensitivity to CX-5461. To test this, we measured rDNA copy number (dosage) in 15 OVCA cell lines by Southern blotting, and expressed rDNA dosage as fold change relative to the rDNA copy number of TOV112D cells. A wide range of rDNA dosage, from 0.07-fold in OVCAR8 to 3-fold in OVCAR3, were observed (**Figure 1D**). We also measured rDNA dosage using FISH in 4 OVCA cell lines and showed that the range of rDNA dosage concorded between the two assays confirming the differences in rDNA dosage between the OVCA cell lines (**Supplementary Figure S2A**). However, despite this wide range in rDNA dosage, we did not observe a correlation between rDNA dosage and CX-5461 GI50 values

(**Figure 1E**). Thus, these results do not support the hypothesis that rDNA copy number directly determines OVCA cell line sensitivity to CX-5461.

We have previously shown that CX-5461 activates nucleolar DDR by inducing chromatin defects and replication stress at the rDNA (Quin et al., 2016; Yan et al., 2019; Sanij et al., 2020). Therefore, we examined whether the number of active rDNA repeats and/or the proportion of active to inactive rDNA repeats are determining factors for CX-5461 sensitivity. To do this, we performed psoralen cross-linking followed by Southern blotting of gDNA from the 15 OVCA cell lines. Actively transcribed and open chromatin rDNA states are accessible to psoralen crosslinking, thus active repeats migrate slower than inactive repeats during agarose gel electrophoresis, allowing the proportion of active/inactive repeats to be quantified (Conconi et al., 1989; Sanij et al., 2008). The relative proportion of active to inactive rDNA repeats varied substantially across the OVCA cell lines (**Figure 2A** and **Supplementary Figure S2B**). We found that CX-5461-sensitive cell lines exhibited higher relative proportions of active rDNA repeats compared to resistant cell lines (**Figure 2B**). We considered whether this sensitivity simply reflects the absolute number of active rDNA repeats. To test this, we calculated active rDNA dosage by normalizing the proportion of active repeats to the relative rDNA dosage in **Figure 1D**. Interestingly, OVCA cell line sensitivity to CX-5461 showed a trend in correlation with the active rDNA dosage (**Figures 2C,E**) but not the inactive rDNA dosage (**Figures 2D,F**). Since sensitivity to CX-5461 correlated with higher basal rate of rDNA transcription (**Figure 1C**; Sanij et al., 2020), we examined whether this also correlated with parameters of rDNA chromatin activity (**Supplementary Figures S3A–C**). Basal rate of rDNA transcription did not correlate with rDNA dosage nor the proportion of active to inactive repeats (**Supplementary Figures S3A,B**) but did show a trend in correlation with active rDNA dosage (**Supplementary Figure S3C**). Therefore, we tested whether OVCA cell line sensitivity to CX-5461 correlates with rDNA transcription rate normalized to active rDNA dosage, but found no correlation with sensitivity to CX-5461 (**Figure 2G**). Thus, of all the rDNA activity parameters examined, only the ratio of active to inactive genes and the basal rate of rDNA transcription (Sanij et al., 2020) correlated with CX-5461 sensitivity in OVCA cell lines. However, we cannot dismiss that the technically challenging nature of the assays examined lead to variabilities that may limit our ability to detect correlations.

Recent studies proposed that variation in rDNA copy number is an adaptive response to DNA replication stress, specifically allowing cells with reduced rDNA copy number to rapidly complete replication and cell cycle progression (reviewed in Salim and Gerton, 2019). Therefore, we investigated whether rDNA dosage or the proportion of active to inactive rDNA repeats correlated with OVCA cell line doubling time (**Supplementary Table S2**). While we observed no correlation between rDNA dosage and OVCA cell line doubling time (**Figure 2H**), a trend in correlation between doubling time and the proportion of active rDNA repeats (**Figure 2I**) and sensitivity to CX-5461 (**Figure 2J**) was observed. Our data therefore suggests that OVCA cell lines with a higher proportion of active rDNA repeats proliferate faster (**Figure 2I**), and are more sensitive to CX-5461 (**Figure 2J**). As replication of active rDNA chromatin occurs in early S phase whereas the silent repeats are replicated from mid to late S-phase (Li et al., 2005), it is plausible that cells with elevated proportions of inactive rDNA require a longer S phase in order to complete DNA replication and thus exhibit a longer doubling time.

UBF has an essential role in establishing and maintaining active rDNA chromatin (Sanij et al., 2008, 2015). To test whether UBF loading on rDNA correlates with sensitivity to CX-5461, we examined UBF and Pol I loading at the rDNA in TOV112D, OVCAR4 and EFO21 cell lines. These cell lines were chosen as they exhibit differential sensitivity to CX-5461 and also vary with respect to rDNA dosage and the ratio of active to inactive rDNA repeats (**Figures 1D**, **2A**). The TOV112D cell line, which has higher rDNA dosage and proportion and of active rDNA than the other two cell lines, also showed a higher occupancy of UBF and Pol I across the transcribed region of the rDNA, while UBF and Pol I rDNA occupancy did not differ between EFO21 and OVCAR4 (**Figure 3A**). Since UBF and Pol I bind to the active, psoralen-accessible rDNA repeats, their occupancy was normalized to the proportions of active rDNA (**Figure 3B**). EFO21, which was the least sensitive to CX-5461 (Sanij et al., 2020) and had the lowest active rDNA dosage (**Figure 2C**) of the three OVCA cell lines, had the highest loading of UBF and Pol I normalized to the proportions of active rDNA (**Figure 3B**). Thus, while UBF binding defines active repeats, the amount of UBF binding per repeat can vary. Consistent with the elevated UBF and Pol I loading, this cell line also has a higher rate of Pol I transcription normalized to active rDNA dosage compared to the other cell lines tested (**Figure 3C**). Therefore, the data suggests that the level of Pol I loading and transcription rate normalized to the proportions of active rDNA do not influence sensitivity to CX-5461, but rather sensitivity to CX-5461 is determined by the proportion of active to inactive rDNA repeats and the basal rate of rDNA transcription (i.e., total rDNA transcription output).

### Reducing rDNA Copy Number Increases the Proportion of Active rDNA Repeats and Is Associated With Elevated Genomic Instability

To obtain independent evidence supporting a role for the proportion of active to inactive rDNA repeats in mediating sensitivity to CX-5461, we reduced the rDNA copy number, which has been shown in yeast to mediate an increase in the activity of the remaining repeats (French et al., 2003; Ide et al., 2010). We did this by utilizing zinc-finger nuclease (ZFN) genome-editing. Two pairs of ZFNs were designed to specifically induce double strand DNA breaks (DSBs) targeting the intergenic spacer (IGS) region upstream and downstream of the transcribed unit of the rDNA repeat thus mediating a loss in rDNA copy number (**Supplementary Figure S1**). Copy number was measured by qPCR in eight empty vector

FIGURE 2 | The proportion of active to inactive rDNA chromatin correlates with sensitivity to GI50 by CX-5461 in 15 OVCA cell lines. (A) A representative psoralen cross-linking Southern blot analysis of 15 OVCA cell lines. The proportion of active versus inactive rDNA was quantified as a % total rDNA; n = 3–4; mean ± SEM. Statistical test of change relative to TOV112D was performed using unpaired t-test, p-values are indicated. (B) Correlation analysis of OVCA cell lines proportion of active rDNA and sensitivity to CX-5461 (GI50). The sensitive cell lines are marked as red dots while the resistant cell lines are blue. Error bars represent mean ± SD. (C) The active rDNA dosage and (D) inactive rDNA dosage were calculated by multiplying the mean rDNA dosage (Figure 1D) with the mean proportion of active or inactive rDNA, respectively and expressed as fold over TOV112D; n = 3, mean ± SEM. Statistical analysis was performed using two-sided one-way ANOVA Dunnett's multiple comparisons test. \*p-value < 0.05, \*\*p-value < 0.01, \*\*\*p-value < 0.001, compared to TOV112D. Correlation analysis of OVCA cells sensitivity to CX-5461 (GI50) and: (E) active rDNA dosage; (F) Inactive rDNA dosage. (G) A correlation analysis of OVCA cells sensitivity to CX-5461 (GI50) with rDNA transcription rate normalized to active rDNA dosage [calculated by dividing the basal rate of rDNA transcription in Figure 1C (Sanij et al., 2020) by active rDNA dosage from (C)]. (H) Correlation analysis of OVCA cell doubling time (Supplementary Table S2) with rDNA dosage, (I) the proportions of active rDNA repeats and (J) OVCA cells sensitivity to CX-5461 (GI50). Error bars on all correlation graphs represent mean ± SD. Significant p-values p < 0.05 are highlighted by the rectangles.

two-sided unpaired t-test,\*\*\*p-value < 0.001 compared to TOV112D.

(EV) transduced TOV112D cell lines and ten ZFN clonal TOV112D cell lines, then expressed as fold change relative to the first EV cell line (EV1). A robust decrease in rDNA copy number (∼20–50%) was observed in the ZFN expressing cell lines compared to EV controls (**Figure 4A**). We measured the effect of reducing the rDNA copy number on cell proliferation using the IncuCyte (**Figures 4B,C**). Overall, the ZFN cell lines exhibited reduced proliferation and longer doubling times (25.6 to 196.8 h) compared to the EV clones (21.9 to 34.9 h), suggesting that the ZFN-mediated reduction in rDNA copy number leads to defects in proliferation. However, it is also plausible that the ZFNs induced multi-DSBs within the rDNA loci leading to activation of a DNA damage response and growth inhibition.

We next evaluated a ZFN clone (Z38) that exhibited a ∼68% reduction in rDNA copy number compared to the EV cell line (**Figure 4D**). Quantitation of rDNA copy number performed using Southern blotting (**Figure 4D**) and rDNA-FISH combined with IF for UBF (**Figure 4E**) confirmed the reduced rDNA dosage in Z38 cells compared to EV cells. Psoralen cross-linking assays demonstrated a higher proportion of active rDNA in Z38 cells compared to EV cells (76%

FIGURE 4 | Reducing rDNA copy number is associated with an increased proportion of active to inactive rDNA chromatin and rate of rDNA transcription in the remaining rDNA pool. (A) TOV112D cells were infected with Lentivirus expressing empty vector (EV) or 2 ZFNs targeting rDNA sequencing and clonal cell lines generated by puromycin selection. gDNA was extracted from 8 EV and 10 ZFNs exponentially growing clones and rDNA dosage measured by qPCR using the 5'ETS (ETS2) primers, then normalized to Vimentin as a single copy locus control (Supplementary Table S3). Data is represented as fold change over EV1; \*\*\* indicates p < 0.001 according to unpaired t-test. (B) The proliferation rate of EV and ZFN clonal cell lines was monitored and the % of cell confluency determined using the IncuCyte; n = 3 of technical replicates, mean ± SD. (C) Doubling time of cell lines was determined using the IncuCyte measurements and analyzed using GraphPad prism. Correlation analysis of rDNA dosage (A) and doubling time for the EV and ZFN clones was performed; Pearson's r is -0.61, \*\* indicates p < 0.01; Spearman's rho is -0.75, \*\*\*p < 0.001. (D) A representative rDNA Southern blot from EV and Z38 cells (upper panel) with quantitation expressed as fold over control (EV); mean ± SEM of n = 3 (lower panel). Paired t test analysis was performed. (E) IF-FISH analysis of rDNA (green: white arrows) and UBF (pink) and DAPI (blue) stained EV and Z38 cells. The intensity of rDNA FISH signal was quantitated using Definiens Tissue Software (Definiens) and graphed as mean ± SD of n = 150 cells analyzed over 3 biological replicates, \*\*\* indicates p < 0.001 according to two-sided Mann-Whitney t-test. (F) A representative of psoralen cross-linking (x-linking) analysis of EV and Z38 cells (upper panel) and quantitation of n = 3; mean ± SEM (lower panel). Paired t test analysis was performed. (G) qChIP analysis of UBF and Pol I (POLR1A subunit) loading on the rDNA. UBF and Pol I enrichment was calculated as described in Figure 3A and normalized to the mean proportion of active rDNA as determined by psoralen cross-linking in (F), mean ± SEM of n = 3. (H) The abundance of the 47S pre-rRNA was measured by qRT-PCR and expressed as fold change over control (EV); mean ± SEM of n = 3. Paired t test analysis was performed. (I) The basal rate of rDNA transcription normalized to active rDNA dosage in EV and Z38 cells was calculated by multiplying the rate of rDNA transcription in (H) with the mean active rDNA dosage from (D,F) and expressed as fold over control (EV); mean ± SEM of n = 3. Paired t test analysis was performed.

compared to 60%; **Figure 4F**), which was associated with an increase in UBF and Pol I occupancy normalized to the proportion of active rDNA (**Figure 4G**). However, the EV and Z38 cells displayed similar rates of basal rDNA transcription (**Figure 4H**). Together, the data suggest that as a consequence of reducing rDNA copy number, the proportion of active rDNA repeats increases, concomitant with an increase in Pol I transcription rate normalized to active rDNA dosage (**Figure 4I**), to maintain total rDNA transcriptional output (**Figure 4H**).

We next determined the sensitivity of Z38 cells to CX-5461. Exponentially growing cells were treated with increasing concentrations of CX-5461 for one hour followed by determination of 47S rRNA abundance. Interestingly, CX-5461 IC50 values for Pol I transcription inhibition by CX-5461 decreased by 50% in the Z38 clone (123 nM) compared to EV (214 nM) (**Figure 5A**). Thus, Z38 cells are more sensitive to Pol I transcription inhibition by CX-5461 than the control cells (EV). We also determined the rate of proliferation 48 h after treatment. Similarly, GI50 for CX-5461 in Z38 cells (29 nM) was reduced by 60% compared to EV cells (70 nM) (**Figure 5A**). Thus, the increase in the proportion of active rDNA repeats in Z38 cells is associated with increased sensitivity to CX-5461. This is consistent with the correlation between the ratio of active to inactive rDNA repeats and sensitivity to CX-5461 (**Figures 2A,B**).

Several studies suggest a strong correlation between rDNA chromatin activity status and genome integrity (reviewed in Diesch et al., 2014). In mouse cells, loss of rDNA silencing results in destabilization of the perinucleolar heterochromatin, which is crucial for ensuring genome stability (Guetg et al., 2010). In addition, yeast with low-copy strains have impaired DNA damage repair during S-phase and consequently higher sensitivity to DNA damaging agents such as ultraviolet radiation and methyl methanesulfonate (Ide et al., 2010). Hence, variation in rDNA chromatin states may predispose cells to genomic

instability and influence cellular responses to DNA damage. To investigate this possibility, we assessed micronuclei formation, a marker of genome instability, and used the alkaline comet assay to measure single strand DNA breaks, DSBs and other DNA lesions at a single cell level in Z38 cells. The Z38 cells exhibited a higher abundance of micronuclei formation (**Figure 5B**) and significantly brighter and longer comet tails (**Figure 5C**) compared to the EV cells, suggesting a higher degree of basal DNA damage although at a lower level compared to EV cells treated with doxorubicin. Furthermore, while the basal level of phosphorylated histone variant (γH2AX), a marker of DNA damage, was similar in the Z38 and EV cells, low dose doxorubicin (Doxo, 10 nM) treatment for 3 h mediated a significant increase in γH2AX foci staining specifically in the Z38 cells (**Figure 5D**). Thus, Z38 cells with a higher proportion of active rDNA repeats exhibit greater sensitivity to DNA damage mediated by doxorubicin. Altogether, these results suggest that a reduction in rDNA copy number by ZFNs mediates an increase in the ratio of active to inactive rDNA repeats and enhances the cells sensitivity to CX-5461 and DNA damage.

## DISCUSSION

In this report, we utilized a panel of human OVCA cell lines to identify potential predictive biomarker(s) of therapeutic response to CX-5461. Our data revealed that sensitivity to CX-5461 significantly correlates with the basal rDNA transcription rate (total rDNA transcriptional output), the proportion of active to inactive rDNA repeats and doubling time. Our analyses also showed a correlation trend for sensitivity to CX-5461 with active rDNA dosage, but there was no correlation between CX-5461 sensitivity and rDNA dosage (copy number), inactive rDNA dosage or rDNA transcription rate normalized to active rDNA dosage. However, we cannot exclude that the high variation in rDNA copy number and proportion of active rDNA repeats measurements may limit our ability to detected significant correlations. Such correlations may be revealed in time should methods that more precisely measure these parameters be developed.

The strong association of higher proportions of active rDNA with sensitivity to growth inhibition by CX-5461 is consistent with CX-5461's mode of action in triggering defects associated with open chromatin and replication stress at the rDNA (Quin et al., 2016; Yan et al., 2019; Sanij et al., 2020), including potentially acting as a TOP 2 poison (Bruno et al., 2020) selectively at the rDNA and/or across the genome. We have demonstrated that CX-5461 activates nucleolar ATM and ATR leading to activation of cell cycle checkpoints and global replication-mediated DNA damage (Quin et al., 2016; Yan et al., 2019; Sanij et al., 2020). Our data therefore suggests that cells with a higher ratio of active rDNA are more sensitive to CX-5461-mediated nucleolar DDR and activation of cell cycle checkpoints, with faster proliferating cells being more responsive to cell cycle arrest.

We found that the proportion of active rDNA repeats does correlate with OVCA cell doubling time. This finding is important in the context of recent bioinformatic analyses of whole genome sequencing data demonstrating that rDNA repeats tend to be lost in cancers (Wang and Lemos, 2017; Xu H. et al., 2017). These results are consistent if we assume that the lower rDNA copy number reported for these cancers has resulted in an increased proportion of active rDNA repeats. However, we also found that rDNA copy number did not correlate with doubling time (**Figure 2H**), suggesting that the relationship between copy number and proliferation time is more complex, which might reflect the multiple ways cells can achieve a certain level of rDNA transcription. Consistent with this, we also showed that the baseline rate of rDNA transcription does not correlate with the proportion of active rDNA repeats or rDNA dosage, although there is a trend in correlation with active rDNA dosage (**Supplementary Figure S3**), possibly due to the rDNA transcription rate being dependent on both the number of active rDNA repeats and the density of Pol I loading/transcription elongation at these active repeats (Conconi et al., 1989; French et al., 2003; Schneider, 2012; Hung et al., 2017). In this case, the increased rate of rDNA transcription observed in cancer cells (Drygin et al., 2010; Hannan R.D. et al., 2013) could be mediated independently of rDNA copy number. We therefore propose that an increase in the proportion of active rDNA repeats is likely to be a more consistent phenotype of proliferative cancers than a reduction in rDNA copy number.

We demonstrated that rDNA copy number can be reduced using dual ZNF targeting. Whether this occurred via precise deletion of whole rDNA repeats, thus by homologous recombination-mediated repair of DSBs, or by non-homologous end joining of the break sites remains unclear. Distinguishing between these two possibilities requires sequencing of the ZNF clones, but the interpretation is likely to be complicated by reports suggesting there are pre-existing incomplete units in the human rDNA (Caburet et al., 2005). The reduced rDNA copy number in the ZFN-targeted TOV112D cells is associated with an increased proportion of active rDNA repeats, in agreement with previous reports in yeast showing altering rDNA copy number modulated rDNA chromatin states (Ide et al., 2010). The increased proportion of active rDNA repeats may result from rDNA class switching, specifically the reduced pool of rDNA repeats promotes UBF binding to any inactive repeats thus converting them to active repeats (Diesch et al., 2019). Alternatively, ZNF deletion may selectively target inactive rDNA repeats. However, CRISPR Cas9-mediated DSBs preferentially occurs in euchromatic regions in the genome, suggesting gene editing is more efficient in euchromatin than in heterochromatin (Jensen et al., 2017). Furthermore, dynamic switching between active and inactive repeats was reported upon ATRX depletion, which resulted in reduced rDNA silencing, specific loss of the inactive rDNA repeats and increased proportions of active rDNA repeats (Udugama et al., 2018) thus making the distinction in genome editing bias toward active or inactive rDNA repeats largely void. Importantly, our data demonstrates that the reduction in rDNA copy number observed in Z38 cells induced genome-wide

instability, as illustrated by increased micronuclei formation and other markers of DNA damage (**Figure 5**). Furthermore, Z38 was more sensitive to doxorubicin-induced DNA damage, consistent with results in yeast showing reduced rDNA copy number increases sensitivity to DNA damaging agents (Ide et al., 2010). Whether this increased genomic instability is mediated by the reduced rDNA copy number, the increased proportion of active rDNA repeats, or another feature due to ZNF treatment remains to be determined.

Taken together, this study demonstrates a significant correlation between OVCA sensitivity to CX-5461 and the proportion of active to inactive rDNA repeats. These data suggest that rDNA chromatin states may be a useful biomarker for sensitivity to targeted Pol I transcription therapies. Validation of this parameter as a predictive biomarker of response to CX-5461 in patient samples in future clinical trials will be important to translate these findings to the clinic. A potential barrier to the effectiveness of rDNA chromatin status as a biomarker is the lack of precision with which the proportion of active rDNA repeats can currently be determined, however our results suggest there is value in developing improved methods for measuring rDNA activity state in situ.

### REFERENCES


### AUTHOR CONTRIBUTIONS

ES, RH, and AG conceptualized and designed the study. ES, KH, GP, NH, DC, and KS developed the methodology. JS and ES acquired the data. JS, ES, KH, RH, AG, and RP analyzed and interpreted the data. ES, RH, RP, and AG supervised the study. All authors wrote, reviewed, and/or revised the manuscript.

#### FUNDING

This work was supported by the National Health and Medical Research Council (NHMRC) of Australia project grants (#1100654 and #1162052). RH and RP were supported by NHMRC fellowships. AG was supported by the New Zealand Marsden Fund (14-MAU-053).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2020.00568/ full#supplementary-material



**Conflict of Interest:** RH is a Chief Scientific Advisor to Pimera Inc.

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 © 2020 Son, Hannan, Poortinga, Hein, Cameron, Ganley, Sheppard, Pearson, Hannan and Sanij. 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.