# REDOX AND METABOLIC CIRCUITS IN CANCER

EDITED BY : Salvatore Rizza, Andrea Rasola, Danyelle M. Townsend and Giuseppe Filomeni PUBLISHED IN : Frontiers in Oncology

#### Frontiers Copyright Statement

© Copyright 2007-2018 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.

The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.

Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.

Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.

As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.

All copyright, and all rights therein, are protected by national and international copyright laws.

The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use. ISSN 1664-8714 ISBN 978-2-88945-635-2 DOI 10.3389/978-2-88945-635-2

#### 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

# REDOX AND METABOLIC CIRCUITS IN CANCER

Topic Editors:

Salvatore Rizza, Danish Cancer Society Research Center, Denmark Andrea Rasola, Università degli Studi di Padova, Italy Danyelle M. Townsend, Medical University of South Carolina, United States Giuseppe Filomeni, Danish Cancer Society Research Center, Denmark; University of Rome Tor Vergata, Italy

"Dynamic vision of cancer cell mitochondria" Image: Salvatore Rizza

Living cells require a constant supply of energy for the orchestration of a variety of biological processes in fluctuating environmental conditions. In heterotrophic organisms, energy mainly derives from the oxidation of carbohydrates and lipids, whose chemical bonds breakdown allows electrons to generate ATP and to provide reducing equivalents needed to restore the antioxidant systems and prevent from damage induced by reactive oxygen and nitric oxide (NO)-derived species (ROS and RNS).

Studies of the last two decades have highlighted that cancer cells reprogram the metabolic circuitries in order to sustain their high growth rate, invade other tissues, and escape death. Therefore, this broad metabolic reorganization is mandatory for neoplastic growth, allowing the generation of adequate amounts of ATP and metabolites, as well as the optimization of redox homeostasis in the changeable environmental conditions of the tumor mass. Among these, ROS, as well as NO and RNS, which are produced at high extent in the tumor microenvironment or intracellularly, have been demonstrated acting as positive modulators of cell growth and frequently associated with malignant phenotype. Metabolic changes are also emerging as primary drivers of neoplastic onset and growth, and alterations of mitochondrial metabolism and homeostasis are emerging as pivotal in driving tumorigenesis.

Targeting the metabolic rewiring, as well as affecting the balance between production and scavenging of ROS and NO-derived species, which underpin cancer growth, opens the possibility of finding selective and effective anti-neoplastic approaches, and new compounds affecting metabolic and/or redox adaptation of cancer cells are emerging as promising chemotherapeutic tools.

In this Research Topic we have elaborated on all these aspects and provided our contribution to this increasingly growing field of research with new results, opinions and general overviews about the extraordinary plasticity of cancer cells to change metabolism and redox homeostasis in order to overcome the adverse conditions and sustain their "individualistic" behavior under a teleonomic viewpoint.

Citation: Rizza, S., Rasola, A., Townsend, D. M., Filomeni, G., eds (2018). Redox and Metabolic Circuits in Cancer. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-635-2

# Table of Contents

*06 Editorial: Redox and Metabolic Circuits in Cancer* Salvatore Rizza, Andrea Rasola, Danyelle M. Townsend and Giuseppe Filomeni

#### 1. MITOCHONDRIAL METABOLISM IN CANCER


#### 2. MITOCHONDRIA AND REDOX SIGNALING CROSSTALK IN CANCER


Nikos Koundouros and George Poulogiannis

#### 3. S-NITROSYLATION, AN EMERGING PLAYER OF CANCER REDOX SIGNALING


#### 4. THE ROLE OF HYPOXIA IN CANCER


#### 5. REDOX AND METABOLIC ADAPTATION OF CANCER CELLS


Venturina Stagni, Claudia Cirotti and Daniela Barilà

*146 Involvement of NADPH Oxidase 1 in Liver Kinase B1-Mediated Effects on Tumor Angiogenesis and Growth*

Elisabetta Zulato, Francesco Ciccarese, Giorgia Nardo, Marica Pinazza, Valentina Agnusdei, Micol Silic-Benussi, Vincenzo Ciminale and Stefano Indraccolo

*156 microRNA-494 Favors HO-1 Expression in Neuroblastoma Cells Exposed to Oxidative Stress in a Bach1-Independent Way* Sabrina Piras, Anna L. Furfaro, Rocco Caggiano, Lorenzo Brondolo, Silvano Garibaldi, Caterina Ivaldo, Umberto M. Marinari, Maria A. Pronzato, Raffaella Faraonio and Mariapaola Nitti

#### 6. BENCH TO BEDSIDE TRANSLATION

*163 Novel Mitochondria-Targeted Furocoumarin Derivatives as Possible Anti-Cancer Agents*

Andrea Mattarei, Matteo Romio, Antonella Managò, Mario Zoratti, Cristina Paradisi, Ildikò Szabò, Luigi Leanza and Lucia Biasutto

*171 Pushing the Limits of Cancer Therapy: The Nutrient Game* Daniele Lettieri-Barbato and Katia Aquilano

# Editorial: Redox and Metabolic Circuits in Cancer

#### Salvatore Rizza<sup>1</sup> , Andrea Rasola<sup>2</sup> , Danyelle M. Townsend<sup>3</sup> and Giuseppe Filomeni 1,4 \*

*<sup>1</sup> Redox Signaling and Oxidative Stress Research Group, Cell Stress and Survival Unit, Center for Autophagy, Recycling and Disease, Danish Cancer Society Research Center, Copenhagen, Denmark, <sup>2</sup> Department of Biomedical Sciences, University of Padova, Padova, Italy, <sup>3</sup> Department of Drug Discovery and Pharmaceutical Sciences, Medical University of South Carolina, Charleston, SC, United States, <sup>4</sup> Department of Biology, University of Rome Tor Vergata, Rome, Italy*

Keywords: cancer metabolism, redox signaling, tumorigenesis, cancer progression, reactive oxygen species

#### **Editorial on the Research Topic**

#### **Redox and Metabolic Circuits in Cancer**

Biological processes in living cells require a constant supply of energy that primarily derives from the oxidation of biomolecules such as carbohydrates, proteins and lipids. The catabolism of biomolecules relocates electrons to the redox couples NAD+/NADH, NADP+/NADPH and FAD/FADH2, which represent the principal cofactors of dehydrogenases and reductases used by cells to sustain all endergonic process. While the couple NADP+/NADPH is crucial for the antioxidant response and anabolic metabolism, NAD+/NADH and FAD/FADH2 convey electrons to the mitochondrial transport chain resulting in the oxygen-dependent production of ATP, the energetic molecule sustaining the activity of the majority of cellular processes. As predictable, there are tight connections between metabolic fluxes, redox balance, oxygen availability, mitochondria function, and turnover.

A peculiar feature of living cells is their extraordinary adaptability to fluctuations in nutrient availability and environmental conditions due to the high plasticity of their biochemical machinery. The back side of the coin emerges, however, in pathologic settings. For instance, cancer cells reprogram the metabolic circuitries in order to sustain their high proliferation rate, invade other tissues, and evade death. An extensive reorganization of cell metabolism is, indeed, a prerequisite for neoplastic transformation and facilitates tumor progression and metastasis. Cancer cells need to increase the levels of the molecular building blocks for membranes, nucleic acids, and proteins biosynthesis and, at the same time, need to produce elevate levels of ATP to sustain cell proliferation. This metabolic rearrangement, as well as exposure of cancer cells to diverse extracellular environments, inexorably results in the increase of reactive oxygen and nitrogen species (ROS and RNS, respectively), which act as positive modulators of cell growth and are frequently associated with malignant phenotype. The antioxidant capacity of cancer cells, as a consequence, readapts in order to tolerate the increased nitro-oxidative stress, this aspect having profound effects on chemoresistance to drugs. Metabolic rewiring, thus, generates cells that are able to face the adverse conditions they encounter in the process of tumor growth, such as nutrient paucity and nitroxidative stress or anticancer therapies.

The study of the intimate connection between redox and metabolic circuities is becoming a hot field in cancer biology, as it has the potency to provide selective targets for innovative chemotherapeutic tools that interfere with metabolic and/or redox adaptations of cancer cells. In this Research Topic we have assembled a collection of review articles that, we hope, will help the readers obtain a broad overview on different aspects of cancer metabolism and redox signaling. Moreover, we have included a substantial number of original research papers offering new insights on redox/metabolic pathways of cancer cells.

Edited and reviewed by: *Paolo Pinton, University of Ferrara, Italy*

\*Correspondence: *Giuseppe Filomeni filomeni@bio.uniroma2.it*

#### Specialty section:

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

Received: *09 August 2018* Accepted: *04 September 2018* Published: *26 September 2018*

#### Citation:

*Rizza S, Rasola A, Townsend DM and Filomeni G (2018) Editorial: Redox and Metabolic Circuits in Cancer. Front. Oncol. 8:403. doi: 10.3389/fonc.2018.00403*

**6**

Zulato et al. provide evidence that down-regulation of the liver kinase B1 (LKB1) impacts on cancer cell redox signaling by perturbing the expression of several genes involved in ROS homeostasis. In particular, they found out that LKB1 loss induces NADPH oxidase 1 (NOX1) transcription, thus effecting cell redox state and sustaining tumorigenicity of LKB1-deficient tumors. Indeed, NOX1 inhibition is able to counteract ROS formation, angiogenesis and growth of LKB1-deficient tumor xenografts in mice. Another mechanism by which cancer cells adapt to changes in their redox homeostasis is described in the brief report by Piras et al. They demonstrate how, in aggressive undifferentiated neuroblastoma, the miRNA-494 is involved in the regulation of heme oxygenase 1 (HO-1), a crucial enzyme affecting cell adaptation to oxidative stress and playing an important role in cancer progression and resistance to therapies. In addition, Koundouros and Poulogiannis comprehensively report on the involvement of ROS metabolism and metabolic rewiring in tumorigenesis driven by phosphoinositide the 3 kinase (PI3K)/AKT axis, one of the most frequently deregulated signaling pathways in cancer. The Authors elaborate on different aspects, ranging from the activation of NADPH oxidases (NOXs) to the redox-dependent inactivation of the phosphatase and tensin homolog (PTEN); from the mechanisms through which PI3K/Akt activation helps maintaining redox adaptation of cancer, to the opportunities for therapeutically exploiting redox metabolism in hyperactive PI3K/Akt tumors. The interplay between metabolism rewiring of tumor cells and oncogenic driver mutations is further discussed by De Santis et al., who analyze the crosstalk among mutations in oncogenes (i.e., PI3K/AKT/mTOR, RAS pathway and MYC), tumor suppressors (i.e., p53 and LKB1), cancer cell metabolism and response to therapy.

From a different viewpoint, Stagni et al. focus on the emerging role of a renowned player in the DNA damage response, Ataxia Telangiectasia Kinase (ATM), in redox cancer biology. In this review article, the Authors highlight the complexity of the molecular circuits through which ATM modulates cancer progression by interfering with redox homeostasis and mitophagy in a DNA damage-independent way. ATM mutations, alongside the effects they produce on genome stability, affect mitochondria homeostasis and trigger ROS formation. On the other hand, ATM hyper-activation sustains survival of cancer stem cells by promoting autophagy. Regarding this last process, the role of autophagy in cancer is still debated. It can, indeed, act as a tumor-suppressor during the early stages of tumorigenesis whereas, in established tumors, it sustains the removal of damaged organelles, thus helping cell proliferation, and facilitating drug resistance. The review article from Ichimura and Komatsu discusses on the role of autophagy as major cellular defense mechanism against metabolic and oxidative stress in relation with the Kelch-like ECH-associated protein 1 (Keap1)/nuclear factor (erythroid-derived 2)-like 2 (Nrf2) system, the master regulator of the antioxidant transcriptional response. Autophagy and the Keap1/Nrf2 system are interconnected via the phosphorylation of the autophagy receptor protein p62/SQSTM1. The Authors provide an overview on recent findings indicating that p62-Keap1-Nrf2 axis drives cell growth and drug resistance in premalignant cells by promoting metabolic reprogramming.

Besides the established implication of ROS in neoplastic transformation and progression, in the last decades a prominent role for nitric oxide (NO) and S-nitrosylation in carcinogenesis has emerged. In this context, Rizza and Filomeni offer a new perspective on the role of denitrosylases, mainly Snitrosoglutathione reductase (GSNOR), in human cancer, whereas Papaleo's group provides an extensive review article Bignon et al. focusing on the mechanisms of S-nitrosylation from a structural and computational point of view, pointing to the main cancer-related targets of S-nitrosylation so far identified.

Since the pioneering studies of Otto Warburg, it is clear that most cancer cells preferentially use glycolysis to sustain their high rate of ATP production, even in the presence of normal oxygen tension. Such a metabolic rewiring is commonly referred to as aerobic fermentation, or "Warburg effect," and it is regulated by several transcription factors, among which the hypoxia-inducible factor 1α (HIF-1α) is one of the most relevant. Dysregulations of HIF-1α expression have been, indeed, implicated in processes such as angiogenesis, energy metabolism, cell survival, and tumor invasion. The oncogenic role of HIF-1α derives from his position at the crossroad between glucose metabolism, oxygen availability, redox stress, and gene transcription regulation. In this research topic, Laitala and Erler elaborate on a new role of HIFs in regulating the extracellular matrix. Cancer progression is, actually, controlled by tumor microenvironment, and hypoxic conditions seem to affect the tumor niche where stromal and cancer cells are in close contact. The original research article by Hlouschek et al. focuses on the resistance of lung cancer cells to radio- and chemotherapy due to chronic cycling hypoxia/reoxygenation stress. They show that this condition is able to trigger the up-regulation of the mitochondrial citrate carrier SLC25A1, impacting on glutathione levels and inducing radio-resistance. Coherently, they provide data suggesting that SLC25A1 inhibitors might sensitize tumor to radiotherapy.

As master regulators of metabolic fluxes, mitochondria play a crucial role in tumor biology. Cannino et al. performed an extensive analysis of the crosstalk of metabolic signals between mitochondria and rest of the cell, focusing on how mitochondrial bioenergetic circuitries can tune the metabolic requirements of cancer cells to the fluctuating environmental conditions. In the short review from Esparza-Moltó and Cuezva, the readers will find an overview on the role of H+-ATP synthase and its physiological inhibitor, the ATPase Inhibitory Factor 1 (IF1), whose over-activation in several human cancers is associated with energy metabolism reprogramming and mitochondrial ROS production. Mitochondrial homeostasis has been also investigated by Gibellini et al., who focus on the role of modulation of the mitochondrial Lon protease (LonP) in metastatic colon cancers. They show that LonP1 is poorly expressed in normal mucosa, while it increases gradually from aberrant crypt foci to adenoma, becoming highly abundant in established colorectal cancers. LonP1 expression seems to correlate with mitochondrial dysfunctions, the rate of glycolysis and pentose phosphate pathway, this seemingly enhancing the epithelial-mesenchymal transition.

In this Research Topic, we have also dealt with some therapeutic aspects of cancer cell redox balance and metabolism. Along this line, the original research by Mattarei et al. explores the utility of novel mitochondria-targeted furocoumarin derivatives as possible anti-cancer agents. The Authors synthesized and tested the efficacy of a neo-synthesized coumarin derivative that blocks the potassium channel Kv1.3, this inducing oxidative stress and cytotoxicity in several malignant cells. Lettieri-Barbato and Aquilano elaborate on the effects that diet can have on cancer cells sensitization to conventional cancer therapies, while simultaneously protecting normal cells from their side effects. The Authors review the recent advances in cancer therapy focusing on the effects of adjuvant dietary interventions, and theorize a novel nutritional approach based on moderate ketogenic diets that could be exploited for future pre-clinical research in cancer therapy.

# AUTHOR CONTRIBUTIONS

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

# FUNDING

Danish Cancer Society Grant Kræftens Bekæmpelse Videnskabelige Udvalg (KBVU) (grant R146-A9414 to GF); Associazione Italiana per la Ricerca sul Cancro (AIRC grant IG2018/20719 to GF and IG 2017/20749 to AR); University of Padua (to AR); Children's Tumor Foundation Drug Discovery Initiative (grant 2016A-05- 009, to AR); Neurofibromatosis Therapeutic Acceleration Program (to AR); Piano for Life Onlus and Linfa Onlus (to AR).

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

Copyright © 2018 Rizza, Rasola, Townsend and Filomeni. 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.

# Metabolic Plasticity of Tumor Cell Mitochondria

Giuseppe Cannino† , Francesco Ciscato† , Ionica Masgras, Carlos Sánchez-Martín and Andrea Rasola\*

Department of Biomedical Sciences, University of Padova, Padova, Italy

Mitochondria are dynamic organelles that exchange a multiplicity of signals with other cell compartments, in order to finely adjust key biological routines to the fluctuating metabolic needs of the cell. During neoplastic transformation, cells must provide an adequate supply of the anabolic building blocks required to meet a relentless proliferation pressure. This can occur in conditions of inconstant blood perfusion leading to variations in oxygen and nutrient levels. Mitochondria afford the bioenergetic plasticity that allows tumor cells to adapt and thrive in this ever changing and often unfavorable environment. Here we analyse how mitochondria orchestrate the profound metabolic rewiring required for neoplastic growth.

#### Edited by:

Saverio Marchi, University of Ferrara, Italy

#### Reviewed by:

Cesar Cardenas, Universidad de Chile, Chile Massimo Bonora, Department of Cell Biology, Albert Einstein College of Medicine, United States

#### \*Correspondence:

Andrea Rasola andrea.rasola@unipd.it; rasola@bio.unipd.it

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology

> Received: 01 June 2018 Accepted: 02 August 2018 Published: 24 August 2018

#### Citation:

Cannino G, Ciscato F, Masgras I, Sánchez-Martín C and Rasola A (2018) Metabolic Plasticity of Tumor Cell Mitochondria. Front. Oncol. 8:333. doi: 10.3389/fonc.2018.00333

Keywords: mitochondria, tumor metabolism, signal transduction, oxidative phosphorylation, neoplastic growth, oncometabolites, redox homeostasis, calcium

# INTRODUCTION

Mitochondria are metabolic hubs that harbor enzymes responsible for several biochemical circuitries, including tricarboxylic acid (TCA) cycle, oxidative phosphorylation (OXPHOS), fatty acid oxidation (FAO), biosynthesis of amino acids, lipids and nucleotides and maintenance of homeostatic levels of Ca2<sup>+</sup> and of reducing equivalent carriers. These bioenergetic, biosynthetic and signaling functions render mitochondria capable of rapidly sensing and integrating stress signals, in order to coordinate biochemical pathways required for the appropriate responses of the cell to environmental changes (1).

Mitochondria gained center stage in molecular oncology when Otto Warburg observed that tumor cells can ferment glucose to lactate even in the presence of oxygen, proposing that a failure in mitochondrial respiration was the cause of this metabolic trait, called aerobic glycolysis, and that this was in turn required for neoplastic growth (2, 3). Decades after this groundbreaking observation, we know that aerobic glycolysis is part of a wider metabolic rewiring that characterizes neoplastic growth. During this process, environmental conditions can rapidly fluctuate, following local changes in oxygen, pH or nutrient gradients, and can become extremely harsh for the transformed cell, which must become capable of tackling sudden shortages in blood supply or exposure to anti-neoplastic treatments.

Unlike Warburg's proposal, tumor cell mitochondria not only retain their functionality, but are also instrumental for integrating a variety of signals and adjusting the metabolic activity of the cell to such a demanding and stressful situation (4) (**Figure 1**). OXPHOS activity is downregulated, but not abolished, in many tumor cell types. Therefore, malignant cells start producing a large portion of their ATP through glycolysis rather than OXPHOS. Enhanced glucose utilization also increases the metabolic flux through pentose phosphate pathway (PPP) (5), which provides anabolic building blocks for nucleotide synthesis and NADPH for anti-oxidant defenses, whereas glycolytic intermediates are used for the biosynthesis of amino acids (4, 6–8). These metabolic changes down-regulate the TCA cycle, both because induction of PPP and of anabolic pathways that branch from glycolysis limit pyruvate availability, and because a low OXPHOS activity inhibits the formation of NAD<sup>+</sup> and FAD required for TCA cycle dehydrogenases. Thus, mitochondria must activate anaplerotic mechanisms in order to feed the TCA cycle, as its activity is required for fatty acid (FA) and amino acid biosynthesis and for the homeostatic maintenance of reducing equivalent carriers (6, 9–14); this is mainly achieved by increasing the pace of glutamine utilization (9, 15) (**Figure 2**).

Several recent lines of evidence suggest that mitochondria indeed play a key promoter role in tumor growth and progression (16). All along this process, mitochondrial biogenesis and quality control are often upregulated, and mitochondria can even retain a high level of OXPHOS in some tumor cell types. Rare human neoplasms with defective respiration caused by mutations in mitochondrial genome, such as oncocytomas (17, 18), are relatively benign, and mitochondrial DNA depletion impairs tumorigenicity in several tumor cell models (19). Altogether, these observations imply the existence of a negative selection for a loss of mitochondrial function in neoplastic transformation (20).

Mitochondrial bioenergetics is largely under the control of extra-mitochondrial biochemical pathways, whose activity is often altered by oncogenic mutations (21). Moreover, some metabolic alterations that directly originate from mitochondria are oncogenic per se (22). In certain tumor settings, mitochondria can act as neoplastic drivers by generating high levels of oncometabolites, i.e., metabolites that are able to change the genomic and epigenomic landscape of the cell, hence prompting the tumorigenic process (23, 24). Thus, the crosstalk between mitochondria and rest of the cell can amplify the metabolic drift of tumor cells away from their non-transformed counterparts during neoplastic progression.

In the present review, we analyse how the metabolic plasticity of tumor cell mitochondria contributes to the neoplastic process. However, any general consideration must be confronted with the real scenario of a tumor mass, where a myriad of factors influence metabolism. These include the tissue of origin of the neoplastic cell, its mutational and epigenomic profile and the local environmental conditions, which can dictate confined changes in the bioenergetic features of the cells, prompting metabolic heterogeneity even in different portions of the same tumor, or in different moments of its growth (25).

## ONCOGENIC SIGNALLING PATHWAYS AND MITOCHONDRIA

A complex network of signals moves back and forth between nucleus and mitochondria (**Figure 3**). This crosstalk constantly keeps under strict nuclear control any mitochondrial function, ensuring its proper harmonization with the metabolic status of the cell. Several major transduction pathways have a strong impact on mitochondrial function, including the transcriptional programs coordinated by HIF1, c-Myc and p53, as well as Ras and mTOR/AMPK signaling (4, 21, 26, 27). Consequently, pro-neoplastic dysregulation of any of these signaling axes strongly affects the mitochondrial metabolic machinery.

# Hypoxia-Inducible Factors (HIFs) and Mitochondrial Metabolism

HIFs induce transcription under low oxygen conditions and are active when their two subunits, aryl hydrocarbon receptor nuclear translocator (ARNT, or HIF-1β) and either HIF-1α or HIF-2α, bind hypoxia-responsive elements (HREs) in gene promoters. While ARNT is constitutively expressed, HIF-1α/2α undergo proteasomal degradation triggered by hydroxylation of specific proline residues. The prolyl-hydroxylases (PHDs) targeting HIF-1α/2α are dioxygenases inhibited in hypoxic or anoxic conditions, which leads to stabilization of HIF-1α and/or HIF-2α. HIF stabilization orchestrates a transcriptional program that equips tumor cells to sustain hypoxic stress by affecting several aspects of cancer biology, including angiogenesis, epithelial-to-mesenchymal transition, metastasis, resistance to anticancer therapies as well as metabolic reprogramming (28–30).

HIF-dependent metabolic rewiring embraces induction of glycolysis and FA synthesis together with OXPHOS downregulation, a key adaptation to low oxygen (31), and has profound effects on mitochondrial activity (**Figure 3**). One of the glycolytic enzymes induced by hypoxia is hexokinase type II (HK II), the most active hexokinase isoform whose expression is upregulated in many cancer types and contributes to their efficiency in glucose utilization (32, 33). In tumor cells, HK II is mainly anchored to the outer mitochondrial membrane, and its detachment from mitochondria rapidly induces cell death (34–36). Thus, mitochondrial binding of HK II has an important tumorigenic function (37) and displays a protective role for mitochondrial function and cell viability through mechanisms yet poorly defined, but involving autophagy regulation in conditions of glucose paucity (38).

The transcriptional program mastered by HIFs creates a bottleneck in funneling glycolysis toward the TCA cycle by slowing-down the conversion of pyruvate to acetyl-CoA (31). This is achieved both through induction of the M2 isoform of pyruvate kinase (PKM2), which is less active than the M1 counterpart in generating pyruvate from phosphoenolpyruvate, and by eliciting the expression of pyruvate dehydrogenase kinase 1 (PDK1), an inhibitor of the pyruvate dehydrogenase complex (PDC) (39, 40). In addition, HIFs promote lactate dehydrogenase (LDHA) expression, again pushing pyruvate away from the TCA cycle toward its conversion into lactate, using reducing equivalents provided by glycolysis-derived NADH and thus keeping the NAD<sup>+</sup> levels required for a sustained glycolytic activity (41). The parallel induction of monocarboxylic acid transporters (MCTs) causes lactate extrusion from the cell and contributes to acidification of the surrounding environment. As a combined result of these modulations, OXPHOS activity is down-modulated, and glycolytic intermediates upstream to pyruvate accumulate and can be diverted to anabolic routines (42).

In these conditions, tumor cells must use lipids and amino acids as main metabolic fuels (43), finding glucoseindependent sources for acetyl-CoA generation required for de novo FA synthesis and for acetylation reactions (see section Post Translational Regulation In Cancer Metabolism). In general, tumor cells increase FA synthesis and the intracellular levels of total FAs for membrane synthesis, lipid signaling or as energy source (when oxidized) (44, 45). HIF signaling increases lipid uptake and the induction of lipid kinases and oxidases, resulting in an overall dysregulation of lipid metabolism in cancer (46). To obtain high levels of acetyl-CoA, mitochondria of cells undergoing hypoxia boost reductive carboxylation of glutamine (47), which generates citrate via the TCA cycle enzymes isocitrate dehydrogenase (IDH) and aconitase. Citrate then moves to cytosol, where it can be cleaved into oxaloacetate and acetyl-CoA by ATP citrate lyase (ACLY), thus starting FA synthesis (**Figure 3**). HIF1α causes proteasomal degradation of a subunit of the α-ketoglutarate dehydrogenase (αKGDH) complex, a TCA component that is responsible for oxidative glutamine metabolism, by inducing the E3 ubiquitinligase SIAH2 (48). Thus, HIF-dependent transcription enhances reductive carboxylation of glutamine by inhibiting its oxidation. In parallel with induction of FA synthesis, HIF signaling downmodulates FAO both directly, by inhibiting the expression of the mitochondrial enzymes medium- and long-chain acetyl-CoA dehydrogenase (MCAD and LCAD) (49) and indirectly, by inducing PHD3, which activates acetyl-CoA carboxylase 2 (ACC2), thus prompting generation of the FAO repressor malonyl-CoA (50).

Mitochondria can also directly regulate HIF stability in a process termed pseudohypoxia that is independent of environmental oxygen levels and further adds flexibility to the metabolic responses of tumor cells (see section Mutations Of Mitochondrial Enzymes In Cancer Metabolism). Furthermore, at least in a model of renal carcinoma, HIF1α can repress the expression of PGC-1α (peroxisome proliferator-activated receptor gamma, coactivator-1α), a central regulator of mitochondrial biogenesis, which in turn stabilizes HIF1α (51). These observations highlight the existence of regulatory loops between mitochondria and the transcriptional program mastered by HIFs (52). Hypoxia also creates a redox stress in mitochondria, as oxygen is the final electron acceptor in OXPHOS and inadequate oxygen levels increase the leakage of electrons out of respiratory complexes, forming reactive

undergo an increase in ROS generation, and therefore increase their antioxidant defenses to avoid oxidative damage and maintain ROS homeostasis. GLUTs, Glucose

oxygen species (ROS). Therefore, HIF signaling is also involved in the maintenance of redox homeostasis, another complex bioenergetic adaptation required for neoplastic progression in which mitochondrial play a central role (see section Redox Homeostasis And Mitochondrial Metabolism In Tumors).

### c-Myc and Mitochondrial Metabolism

Transporters; MCTs, Lactate Transporters; SLCs, Solute Carriers.

c-Myc is one of the most frequently induced oncogenes in human cancers, where its transcriptional function becomes constitutively activated following deregulation of oncogenic pathways, gene amplification or chromosomal translocation (53). The effect of c-Myc activation is the orchestration of nutrient uptake and cell growth and proliferation, making its dysregulation a key oncogenic driver. These biological routines require a robust anabolic induction, and this is crucially supported by mitochondria. There are several ways by which c-Myc affects mitochondrial metabolism, thus sustaining growth of neoplastic cells in the unfavorable environment they must deal with. The transcriptional program mastered by c-Myc partially overlaps the metabolic effects of HIF-dependent signaling. Indeed, c-Myc upregulates the same set of glycolytic genes that are targeted by HIFs, including GLUT1, LDHA, MCTs, PKM2, and HK II, thus increasing glucose uptake and its utilization both in glycolysis and PPP (**Figure 3**). As discussed for HIFs, these changes cause a metabolic rewiring toward aerobic glycolysis, lowering in parallel pyruvate availability for the TCA cycle and OXPHOS (54).

At variance from HIFs, however, c-Myc is active under nonhypoxic conditions, and can stimulate mitochondrial biogenesis and respiration. c-Myc activates mitochondrial transcription factor A (TFAM), PGC1β and mitochondrial DNA polymerase gamma, which elicit the expression of hundreds of genes encoding for mitochondrial proteins (55). This could be relevant for the local adaptations of tumor cells to the microenvironmental heterogeneity they find in the tumor mass. It is possible to envision that c-Myc can prompt both glycolysis and OXPHOS in neoplastic cells located in the proximity of blood vessels, where high levels of oxygen are available. Instead, when cells encounter more hypoxic conditions, c-Myc could cooperate with HIFs in increasing glycolysis and attenuating mitochondrial OXPHOS, without inhibiting other mitochondrial metabolic activities (56).

Induction of mitochondrial serine hydroxymethyltransferase (SHMT2) by c-Myc provides an elegant example of this conditional cooperation between c-Myc and HIFs in regulating metabolic circuitries of tumor cell mitochondria. SHMT2 is the major source of the one-carbon unit required for folate metabolism and for the biosynthesis of nucleotides and amino acids (**Figure 3**). It utilizes serine, obtained from the glycolytic intermediate 3-phosphoglycerate, and tetrahydrofolate (THF) to catalyze the synthesis of glycine and 5,10-methylenetetrahydrofolate (5,10-CH2-THF). In turn, 5,10-CH2-THF can generate formate and the reducing equivalent donor NADPH in a reaction catalyzed

FIGURE 3 | Mitochondria at the crossroad of metabolic networks and signaling cascades. Several proteins with pro-neoplastic activity (in light blue boxes) alter the expression and/or the activity of metabolic enzymes and transporters, thus rewiring the metabolic status of cancer cells. Similarly, the possible loss of the tumor suppressor p53 (in light red boxes) impacts on cancer metabolism at several levels. Enzymes that catalyze metabolic reactions are shown in ovals. Yellow ovals indicate enzymes that are preferentially inhibited in tumors, while purple ovals indicate those that are mostly induced. G6P, glucose-6-phosphate; F6P, fructose-6-phosphate; F1,6BP, fructose-1,6-bisphosphate; G3P, glyceraldehyde-3-phosphate; 1,3BPG, 1,3-bisphosphoglycerate; 3PG, 3-phosphoglycerate; 2PG, 2-phosphoglycerate; PEP, phospho-enolpyruvate; R-5-P, ribose-5-phosphate; MCT, monocarboxylate transporter; GLUT, glucose transporter; PC, pyruvate carrier; ASCT2, alanine, serine, cysteine-preferring transporter 2; 5,10-CH2-THF, 5,10-methylenetetrahydrofolate; 10-CHO-THF, 10-formyl-THF; 5,10-CH-THF, 5,10 methenyl-THF; HIF, hypoxia-inducible factor; HK, hexokinase; PFK1, phosphofructokinase 1; PGAM1, phosphoglycerate mutase 1; PKM2, pyruvate kinase M2 isoform; LDHA, lactate dehydrogenase A; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; PDH, pyruvate dehydrogenase complex; PDK, pyruvate dehydrogenase kinase; αKGDH, α-ketoglutarate dehydrogenase; CS, citrate synthase; SDH, succinate dehydrogenase; FH, fumarate hydratase; IDH, isocitrate dehydrogenase; GLS, glutaminase; ACLY, ATP-citrate synthase; ACSS2, Acyl-coenzyme A synthetase short-chain family member 2; ACC, acetyl-CoA carboxylase; FASN, fatty acid synthase; SCD1, stearoyl-CoA desaturase 1; SHMT, serine hydroxymethyltransferase; MTHFD, methylenetetrahydrofolate dehydrogenase; CPT, carnitine O-palmitoyltransferase; MCAD, medium-chain acetyl-CoA dehydrogenase; LCAD, long-chain acetyl-CoA dehydrogenase; PGC1, peroxisome proliferator-activated receptor gamma, coactivator-1; TFAM, mitochondrial transcription factor A; PPP, pentose phosphate pathway; TCA, tricarboxylic acid; OXPHOS, oxidative phosphorylation.

by methylenetetrahydrofolate dehydrogenase 2 (MTHFD2). c-Myc-dependent induction of SHMT2 under normoxic conditions has important biosynthetic consequences: formate is released in the cytosol where it is involved in purine synthesis; glycine and 5,10-CH2-THF contribute to nucleotide synthesis, and NADPH is required for reductive biosynthesis of amino acids, deoxyribonucleotides and lipids (54, 55). When a tumor cell faces hypoxia, HIF stabilization further induces SHMT2. This is counterintuitive, as in low-oxygen conditions cells inhibit proliferation, thus reducing demand for anabolic fluxes. Nevertheless, under hypoxia SHMT2 is essential for survival of c-Myc-transformed cells as it protects them from oxidative stress. Indeed, MTHFD2 activity maintains a high NADPH:NADP<sup>+</sup> ratio, and NADPH is required for regeneration of the antioxidant tripeptide glutathione and hence for protection from ROS damage (57). Notably, SHMT2 is required for survival and proliferation of neoplastic cells in ischemic tumor zones (58).

c-Myc also promotes the glutamine addiction that characterizes several cancer cell types (54, 55) (**Figure 3**). Glutamine is both a nitrogen and carbon source essential for biomass accumulation, and a substrate used for bioenergetic purposes, and it is avidly consumed by neoplastic cells for proliferation and survival (**Figures 2**, **3**). c-Myc increases the expression of the plasma membrane glutamine transporter, ASCT2/SLC1A5, and of glutaminases (GLS) that convert glutamine to glutamate in cytosol (GLS1) or mitochondria (GLS2) as a first step of its oxidation (15, 59). Glutamate generates α-ketoglutarate (α-KG) either via glutamate dehydrogenase (GDH), in a reaction that releases ammonia, or via several aminotransferases that transfer glutamate nitrogen to α-keto acids, such as pyruvate, for producing other amino acids and α-KG. In turn, α-KG feeds the TCA cycle, which is therefore accelerated by c-Myc via glutaminolysis induction. c-Myc also increases the TCA cycle flux up to four-folds by eliciting the expression of most of its enzymes (55). When the TCA cycle flux is impaired, e.g., in low glucose conditions or by mutations in some of its components, α-KG can act as an anaplerotic substrate that provides carbon units to yield citrate by moving backwards through the TCA cycle through reductive carboxylation (9). In cytosol, citrate starts lipid synthesis. c-Myc induces the string of enzymes responsible for the first steps of lipidogenesis, whose upregulation occurs across most tumors (45). These enzymes include ACLY, which uses citrate to synthesize acetyl-CoA, Acetyl-CoA carboxylase, which generates malonyl-CoA, fatty acid synthase and stearoyl-CoA desaturase (55). Malonyl-CoA inhibits carnitine acyl transferase I (aka carnitine palmitoyltransferase I, CPT I), the carrier responsible for the uptake of fatty acids in mitochondria, thus acting in a feedback loop on mitochondrial metabolism to inhibit FAO (60).

Taken together, these observations demonstrate how tumor cell mitochondria can control the homeostatic balance of reducing equivalent donors, OXPHOS activity, lipid synthesis and oxidation in response to c-Myc and/or HIF activation, with crucial implications for chromatin remodeling, tuning of all major anabolic pathways and handling of oxidative insults.

#### p53 and Mitochondrial Metabolism

The transcription factor p53 is a key tumor suppressor activated by a set of stress signals, such as genotoxic damage, oncogene activation, nutrient or oxygen scarcity and loss of cell-to-cell contacts, all of which characterize malignant transformation (61, 62). A functional inactivation of p53 occurs in the majority of tumor types (63). p53 mainly exerts its activity in the nucleus by regulating the expression of genes or microRNAs (miRNAs), even though it can also act in cytosol and mitochondria to inhibit autophagy (64) or to promote cell death (65). p53 activation induces death or senescence when a sustained or intense stress causes irreversible cell damage. Conversely, mild stresses result in p53-dependent adaptive responses, consistent with its role in repairing or avoiding damage. During fluctuations in oxygen or nutrient availability, the effect of p53 is a global promotion of cell catabolism, associated with an inhibition of proliferation and growth (**Figure 3**). p53 interacts with mTOR (mammalian target of rapamycin) and AMPK (AMP-activated protein kinase), two master regulators of cellular metabolism. The mTORC1 complex is active in the presence of both adequate growth conditions and mitogens and coordinates the anabolic responses of the cell (66), whereas AMPK inhibits mTORC1 and allows cells to adapt to energetic stresses by utilizing glucose and FA for increasing ATP levels (see below) (67). A complex interplay exists between p53 and the mTORC1/AMPK circuitry: p53 both activates AMPK and elicits the expression of several negative regulators of mTORC1, whereas AMPK activates p53 through several means, including its phosphorylation and acetylation (62, 68). In this way, cells can respond in a balanced and flexible way to metabolic variations, different types of stress signals, growth and proliferation inputs (69).

p53 can also exert a more direct action on several metabolic effectors (**Figure 3**). In general, it dampens the glycolytic rate through a concerted downregulation of glucose transporters (GLUTs) (70) and of the glycolytic enzyme phosphoglycerate mutase (PGAM) (71). p53 also prompts the expression of TIGAR (TP53-induced glycolysis and apoptosis regulator), which indirectly inhibits phosphofructose kinase 1 (PFK1) (72), thus leading to a p53-dependent diversion of glycolytic intermediates into the PPP (62). However, the effect of p53 on glucose metabolism is highly context-dependent, as p53 can also negatively modulate PPP activity (73). In parallel to inhibiting glycolysis, p53 enhances mitochondrial bioenergetic activity in several ways (**Figure 3**). It promotes mitochondrial quality control by inducing mitophagy to substitute damaged mitochondria, and it increases mitochondrial DNA copy number and mitochondrial mass (68). p53 boosts TCA cycle via induction of the mitochondrial glutaminase GLS2 (74, 75), thus fueling glutamine to glutamate conversion and in turn α-KG generation through the TCA cycle enzyme α-KGDH. Moreover, p53 activates the PDC as it inhibits the negative PDC regulator pyruvate dehydrogenase kinase 2 (76), increasing pyruvate funneling into the TCA cycle through its conversion to acetyl-CoA. p53 also enhances OXPHOS activity by inducing the expression of subunit I of cytochrome c oxidase (COX), the complex IV of the electron transport chain, and of the COX assembly factor SCO2, as well as of apoptosisinducing factor (AIF), which is required for a proper OXPHOS functioning (68).

In general, p53 lifts lipid utilization inhibiting FA synthesis and increasing mitochondrial FAO in a coordinated manner (68). This results from p53-dependent modulation of glycolysis and PPP, as these routines supply the building blocks (acetyl-CoA and NADPH) needed for lipid synthesis. Furthermore, p53 directly stimulates FAO by inducing the mitochondrial membrane FA transporters carnitine acetyltransferases (CPTs) (77, 78), the transcriptional co-activator for FAO genes lipin 1 (79), and pantothenate kinase 1, which is essential in β-oxidation as it is involved in CoA biosynthesis (80). Moreover, p53 induces malonyl-CoA decarboxylase, which lowers the levels of the CPT allosteric inhibitor malonyl-CoA (81). Thus, p53 opposes the metabolic shift toward the induction of FA synthesis and glycolysis that characterizes many tumor cell types and supports mitochondrial FAO and OXPHOS. Notably, FAO induction further feeds OXPHOS via generation of FADH2 and NADH. Nonetheless, under conditions of extreme stress p53 can have an opposite effect on mitochondria, potently contributing to their dysfunction through repression of the master regulators of mitochondrial biogenesis PGC-1α/PGC-1β, thus leading to cell senescence or death (82).

## The AMPK/mTOR System and Mitochondrial Metabolism

AMPK acts as a homeostatic device whose purpose is the rapid restoration of energy balance through an orchestrated inhibition of ATP-consuming biosynthetic pathways. It activates following a drop in ATP levels, as it senses the AMP/ATP ratio. Therefore, AMPK induction is promoted by down-regulation of mitochondrial OXPHOS activity; in turn, AMPK induces glucose uptake and glycolysis while inhibiting storage of glucose, and it suppresses FA synthesis by phosphorylating and inhibiting ACCs (83) (**Figure 3**). ACC2, the isozyme associated to the outer mitochondrial membrane, generates malonyl-CoA that inhibits CPT1 and therefore mitochondrial import of FAs for their oxidation. Hence, ACC2 inhibition by AMPK induces mitochondrial FAO (67). Moreover, AMPK inhibits de novo lipid generation via direct phosphorylation of SREBP1 (sterol regulatory element binding protein 1), a master transcriptional regulator of lipid synthesis (84). AMPK also prompts mitophagy through phosphorylation of ULK kinases (1) and mitochondrial fragmentation through phosphorylation of mitochondrial fission factors (85) in order to restrict energy expenditure and to maintain cell viability in conditions of starvation or of OXPHOS dysfunction. When the bioenergetic stress lingers on, AMPK reprograms metabolism for allowing cells to tackle prolonged energy crises. Therefore, a sustained AMPK activation increases FAO as a bioenergetic source, while limiting glucose and lipid synthesis (67) and increasing mitochondrial biogenesis via PGC-1α, thus allowing a high degree of metabolic plasticity to cells undergoing energetic stress (1).

The protein complex mTORC1 is formed by the serine/threonine protein kinase mTOR, by the regulatory proteins Raptor, which facilitates substrate recruitment to mTORC1, and mLST8, which associates with the catalytic domain of mTORC1, and by the two inhibitory subunits PRAS40 and DEPTOR (86). mTORC1 promotes all major anabolic pathways, including protein, lipid and nucleotide synthesis, together with glucose metabolism and organelle biogenesis. It is activated by the Ras/ERK and by the PI3K/Akt signaling pathways, which are deregulated in most cancer types, whereas it is inhibited by the tumor suppressors p53 and LKB1. Thus, mTORC1 induction is a crucial event in the metabolic rewiring of tumor cells (66) (**Figure 3**). mTORC1 promotes a shift from OXPHOS to glycolysis, and it enhances the expression of many glycolytic genes by increasing HIF1α translation (87); in addition, it activates GDH, thus inducing glutaminolysis (88), and it upregulates MTHFD2 expression and therefore the mitochondrial folate pathway required for purine synthesis (89). The mTORC1 complex also activates SREBP, thus opposing the effect of AMPK on lipid metabolism and promoting lipidogenesis (90).

## Ras and Mitochondrial Metabolism

RAS genes encode a family of GTPases whose mutations are frequently causative of tumor development and are common in cancer. Hyperactivation of Ras-induced transduction pathways boost cell growth and proliferation, prompt resistance to death signals and promote the acquisition of invasive and metastatic properties (91, 92). Moreover, deregulated Ras signaling orchestrates pro-neoplastic changes in several components of the tumor microenvironment, including cancer-associated fibroblasts, endothelial, inflammatory and immune cells (93). Therefore, Ras mutations are among the most problematic oncogenic events, frequently associated with a dismal prognosis (94). Oncogenic activation of Ras signaling has a profound impact on the metabolic changes of tumor cells (**Figure 3**). It induces HIF1α both by upregulating the HIF1A gene and by enhancing translation of its mRNA via activation of the mTORC1 complex through the Ras downstream effectors ERK and PI3K/Akt (91). As described above, HIF1α in turn promotes both the transport and the glycolytic utilization of glucose and the funneling of glycolytic intermediates into the PPP (95).

In parallel, oncogenic Ras signaling affects mitochondrial bioenergetics in several ways. It prompts mitochondrial translocation of phosphoglycerate kinase I, which inhibits PDC via activation of the PDC inhibitor PDK1; as a result, less pyruvate is funneled in the TCA cycle, leading to a decrease in ROS levels together with a surge in lactate extrusion (96). K-Rasdependent transformation inhibits OXPHOS by down-regulating respiratory complex I content (97, 98). ERK, a crucial Ras effector that can locate in mitochondria (35, 99), decreases the activity of respiratory complex II, aka succinate dehydrogenase (SDH; see also section Post Translational Regulation In Cancer Metabolism) (100). Nonetheless, transformation by Ras does not abrogate OXPHOS activity, which still generates a large fraction of cellular ATP (101–103).

Several evidences underline the importance of mitochondrial activity in Ras-driven tumor cells. Together with loss of p53, hyperactive Ras elicits autophagy in a non-small-cell lung cancer setting, thus preserving a proper mitochondrial function that is required for lipid homeostasis and tumor growth (104). Inhibition of the mitochondrial transcription factor Tfam leads to mitochondrial depletion and impedes growth of K-Rasdependent lung tumors (105). Similarly to Myc-transformed cells, Ras-driven cancers strongly rely on glutamine for growth and survival (106, 107), and glutamine is the major carbon source for the TCA cycle when Ras is activated (95). In pancreatic cancer, which is characterized by activating mutations in K-Ras in more than 90% of cases, mitochondrial glutamate-oxaloacetate transaminase 2 (GOT2) utilizes glutamine-derived glutamate and oxaloacetate (OAA) to generate aspartate that is then exported to cytosol. Here, GOT isoform 1 converts back aspartate to OAA, which provides malate and finally pyruvate by malic enzyme (ME). This reaction produces NADPH, a key factor to reduce glutathione in order to avoid oxidative stress (108). Anchorageindependent growth of K-Ras-transformed cells requires α-KG generation by glutamine through glutaminase and alanine aminotransferase activity (105). Lung cancer cells exhibit a high channeling of glycolytic metabolites into the TCA cycle and glutathione biosynthesis, leading to protection from oxidative insults. This metabolic rewiring occurs when neoplastic cells are homozygous for K-RasG12D, but not in heterozygous K-RasG12D cells, which characterize early tumor stages, thus highlighting the importance of oncogenic Ras in shaping metabolic heterogeneity and adaptions of cancer cells (109).

#### MUTATIONS OF MITOCHONDRIAL ENZYMES IN CANCER METABOLISM

The oncogenic role played by changes in mitochondrial metabolism was first spotted by finding that mutations in subunits of SDH, an enzyme placed at the crossroad between OXPHOS and TCA cycle, as well as in the TCA cycle enzyme fumarate hydratase (FH), are causative of some human tumor types (110–113). These were paradigmshifting findings, as they demonstrated for the first time that mitochondria, and in particular core bioenergetic enzymes, could play an active role in the complex alterations of biochemical circuitries that lead to tumor onset (114). SDH and FH act as classical tumor suppressors (**Figure 4**). Inactivating mutations in SDH subunits have been identified in both genetic and sporadic cancer types, including familial paraganglioma/pheochromocytoma (PGL/PCC) (115), renal carcinoma, thyroid cancer, neuroblastoma, gastrointestinal stromal tumor, ovarian cancer and testicular seminoma (116), whereas FH loss-of-function hallmarks hereditary leiomyomatosis and renal cell cancer (HLRCC) and skin and uterine leiomyomas (22).

Research in the last few years allowed understanding that the metabolic origin of this peculiar subset of neoplasms illustrates concepts of general importance. Indeed, it emerged that there is a tight intertwining between metabolism and gene expression, as cells and tissues (and neoplasms make no exception) must strictly coordinate genome expression and metabolic state for the proper unfolding of their biological routines (117). Components of intermediary metabolism affect the activity of chromatinmodifying enzyme (**Figure 4**). This leads to epigenomic changes mastering pro-neoplastic molecular rewiring via gene expression regulation (118). Inactivating mutations of SDH and FH drive such mechanisms by causing accumulation of their substrates succinate and fumarate, respectively, which leads to inhibition of a class of enzymes called α-KG–dependent dioxygenases. To hydroxylate their substrates, these enzymes take up one oxygen atom by α-KG, which therefore acts as a co-substrate that is decarboxylated and releases carbon dioxide and succinate. Both succinate and fumarate competitively inhibit α-KG– dependent dioxygenases, including the JmjC domain-containing demethylases (KDMs), which hydroxylate lysine residue on histones, the TET (ten-eleven translocation) family of 5 methylcytosine hydroxylases, which induce DNA demethylation of CpG islands near gene promoters, and prolyl hydroxylases (PHDs), which prompt proteasomal degradation of HIF1α (119). These enzymes play central roles in epigenetic and transcriptional regulation, and their inhibition by high levels of succinate and fumarate suppresses differentiation and promotes proliferation and further metabolic changes (20), thus inducing tumorigenesis. As a consequences of these pro-neoplastic effects, both succinate and fumarate have been dubbed oncometabolites (23, 24) and similar patterns of epigenomic changes can be observed both in FH- and in SDH-deficient tumors (120). Oncometabolites also favor tumor cell motility by affecting extracellular matrix composition via inhibition of collagen prolyl-4-hydroxylases (119) and elicit angiogenesis through succinate-dependent transcriptional induction of VEGF (116).

Fumarate inactivates aconitase 2 (**Figure 4**), a TCA cycle enzyme containing an iron-sulfur group (121) and proteins involved in the biogenesis of iron-sulfur clusters (122), and it inhibits the enzymatic activity of SDH through a product inhibition effect, leading to down-modulation of both OXPHOS and TCA cycle function. In order to cope with the absence of a functional TCA cycle, FH-deficient cells utilize fumarate in a linear pathway starting with glutamine and ending with the biosynthesis and degradation of haem and eventually with bilirubin excretion. This pathway is crucial for mitochondrial NADH production, and makes FHdeficient cells dependent on heme oxygenase activity, thus creating a metabolic vulnerability (123). In addition, TET inhibition by fumarate causes inactivating hypermethylation of the anti-metastatic miRNA cluster miR-200, thus promoting the transcriptional program mastered by the miR-200 target ZEB1. ZEB1 induces epithelial-to-mesenchymal transition (EMT), a complex biological rearrangement by which tumor cells acquire invasive properties (124). Thus, fumarate-dependent inhibition of miR-200 boosts EMT (125) (**Figure 4**). Fumarate can also tune cell redox equilibrium via protein succination, i.e., the interaction between fumarate and cysteine residues to produce S-(2-succino)-cysteine, which hampers protein function (24). Fumarate both inactivates by succination Kelch-like ECHassociated protein 1 (KEAP1), which promotes proteasomal degradation of NRF2, the master transcriptional regulator of the cell response to oxidative stress (23, 126), and generates succinated glutathione, an alternative substrate to glutathione reductase that leads to an increase in mitochondrial ROS levels and to HIF-1 activation (127) (**Figure 4**).

FIGURE 4 | Biochemical mechanisms of oncometabolite accumulation. Inactivating mutations in genes encoding succinate dehydrogenase (SDH) and fumarate hydratase (FH), as well as oncogenic mutations in isocytrate dehydrogenase (IDH), lead to the accumulation of succinate, fumarate and D-2-hydroxyglutarate (D-2-HG). These oncometabolites inhibit α-KG-dependent dioxygenases, including prolyl hydroxylases (PHDs), the ten-eleven translocation (TET) family of methylcytosine hydroxylases and histone lysine demethylases (KDMs), leading to HIF1α stabilization and alterations in gene expression through epigenetic modifications. Separately, fumarate inactivates by succination both aconitase 2 (ACO2) and Kelch-like ECH associated protein 1 (KEAP1), which results in the activation of the antioxidant pathway mediated by NRF2, and also generates succinated glutathione (GSF), an alternative substrate to glutathione reductase (GSR). Additionally, tumor cells can use alternative ways to increase oncometabolite concentration, such as the upregulation of the mitochondrial chaperone TRAP1, which inhibits the activity of SDH, leading to the intracellular accumulation of succinate. 2SC, succination of cysteine residues; EMT, epithelial-to-mesenchymal transition; GSH, reduced glutathione; HO-1, heme oxygenase-1; IMS, intermembrane space, M, methylation; OAA, oxaloacetate; P, phosphorylation, Succ-CoA, succinyl-CoA; U, ubiquitination.

More recently a third oncometabolite was identified, D-2 hydroxyglutarate (D-2-HG). D-2-HG accumulates to millimolar concentrations in tumors with monoallelic mutations in IDH1 and IDH2, NADP+-dependent homodimers localized in cytoplasm and mitochondria, respectively, that convert isocitrate to α-KG (**Figure 4**). Mutant IDH1/2 dimerize with the wild-type protein and the heterodimeric enzyme acquires a neomorphic activity: the reduction of α-KG to D-2-HG in the presence of NADPH (128). As for fumarate and succinate, D-2-HG has a molecular structure that is similar to α-KG. Thus, D-2-HG acts as a competitive inhibitor of α-KG on the activity of α-KGdependent dioxygenases (129). This results in global remodeling

of DNA methylome, with an increase in methylation of both CpG islands and histones (130), and in inhibition of the PHDs that degrade the HIFs (131). Mutant versions of cytoplasmic and mitochondrial IDH isoforms are present in a fraction of acute myeloid leukemias and in the majority of glioblastomas, as well as in chondrosarcomas and cholangiocarcinomas (131). Cells harboring high levels of D-2-HG are endowed with glutamate depletion, probably because glutamate is utilized to produce α-KG and subsequently converted to D-2-HG. Both low glutamate levels and a higher NADP+/NADPH ratio caused by increased consumption of NADPH could suppress glutathione synthesis and regeneration, affecting the redox equilibrium of IDH mutant cells (9, 15, 24).

In addition to genetic mutations that affect the activity of enzymes involved in raising levels or generating oncometabolites, it is possible that tumor cells utilize more subtle ways to rapidly tune oncometabolite concentration, in order to afford rapid and flexible response to environmental changes. The molecular chaperone TRAP1 provides such an example. TRAP1 is a component of the HSP90 chaperone family whose expression is restricted to mitochondria and is increased in many malignancies (132). TRAP1 down-regulates the activity of both SDH (133) and respiratory complex IV (134), thus decreasing oxygencoupled ATP synthesis and shifting the burden of ATP production to glycolysis (**Figure 4**). Inhibition of SDH by TRAP1 leads to succinate accumulation (133). Such an inhibition is further enhanced when ERK phosphorylates TRAP1, as in cells lacking the Ras GTPase-activating protein neurofibromin (100) that are characterized by deregulated induction of the Ras/ERK1/2 signaling pathway and form tumors in patients with the genetic syndrome neurofibromatosis type I (135). An increase in succinate levels induced by TRAP1 drives HIF1α stabilization independently of oxygen levels (136), i.e., it generates conditions of pseudohypoxia, an adaptive feature of many tumors that allow them sustaining the neoplastic process even before hypoxic conditions are encountered by the growing malignancy. The importance of providing such a metabolic adaptableness is highlighted by the observation that abrogating TRAP1 expression ablates tumorigenicity when different cell types are xenografted in mice (100, 133), even though the specific importance of TRAP1 in neoplastic growth could be contextdependent (132).

#### REDOX HOMEOSTASIS AND MITOCHONDRIAL METABOLISM IN TUMORS

Mitochondria are the major source of intracellular reactive oxygen species (ROS), as about 1% of O<sup>2</sup> consumed by OXPHOS undergoes a one-electron reduction that forms a superoxide anion (137, 138). In addition, other mitochondrial metabolic enzymes, such as αKG dehydrogenase, PDH, the mitochondrial form of glycerol-3-phosphate dehydrogenase and acyl-CoA dehydrogenase can generate ROS (139–141). An excessive oxidant challenge damages biomolecules and leads to DNA mutations that eventually prompt cell senescence or death (142), but maintenance of a physiological redox equilibrium, or oxidative eustress (141), governs a variety of life processes and signal transduction pathways (137) (**Figure 5**).

Intracellular ROS levels are in general higher in tumor cells than in their non-transformed counterparts, and are involved in oncogene activation, tumor suppressor loss, metabolic rewiring, mutations in mitochondrial DNA (mtDNA) or hypoxia (142). ROS can reversibly target cysteine residues within the enzymatic sites of many phosphatases, such as the PI3K inhibitor PTEN, MAPK phosphatases and Tyr phosphatases (143), causing their inactivation. The consequent boost of kinase signaling pathways (144) affects mitochondrial metabolism, e.g. by tyrosine kinases that inhibit at multiple levels the PDC (145). In addition, many mitochondrial FAO enzymes contain ROS-sensitive Cys residues (**Figure 5**). Taken together these observations suggest that oxidative stress can tune mitochondrial metabolism by compromising both beta-oxidation of lipids and pyruvate entry into the TCA cycle (139). mtDNA mutations deregulate redox equilibrium by hampering respiration. Such mutations prompt in vitro and in vivo tumorigenicity, correlate with acquisition of metastatic potential and poor prognosis and can be used for cancer detection and determination of the degree of malignancy (20).

Hypoxia increases superoxide release from respiratory complex III, leading to PHD inhibition, possibly via oxidation of Fe2<sup>+</sup> that is required for PHD function, and to the ensuing stabilization of HIFα subunits (52). In turn, HIF activation decreases ROS production by (i) down-modulating OXPHOS activity, as it suppresses SDHB expression (146), induces NDUFA4L2 (NADH dehydrogenase 1 subcomplex, 4-like 2), which attenuates complex I activity (147), and prompts the substitution of the COX subunit 4-1 with COX4-2, hence optimizing COX activity in low oxygen conditions (148); (ii) up-regulating miR-210, which orchestrates inhibition of mitochondrial bioenergetics by targeting the SDHD transcript and by repressing the iron–sulfur cluster assembly proteins that are required for the incorporation of [4Fe-4S] and [2Fe-2S] groups in respiratory complexes I, II and III; (iii) downregulating mitochondrial biogenesis via c-Myc inhibition (147); (iv) inducing mitophagy through BNIP3, Bcl-2 and BN67IP3L/NIX induction (149). In keeping with this last point, absence of the mitophagy inducer Parkin enhances ROS generation by the persistence of dysfunctional mitochondria and increases tumorigenesis in multiple cancer models (150).

If activation of mitochondrial ROS generation remains below what triggers manifest cellular damage, it can contribute to the neoplastic process by causing DNA damage and genomic instability or by prompting dysregulated activation of crucial signaling pathways, eventually impacting on cell proliferation, angiogenesis and invasiveness (143). As an example, anchorageindependent growth of K-Ras-transformed cells requires an increase in mitochondrial ROS generated by respiratory complex III (105). Therefore, neoplastic cells must enhance their antioxidant devices, such as the tripeptide glutathione (lglutamyl-l-cysteinyl-glycine), in order not to reach a threshold of oxidative damage incompatible with their survival. Both glutamine-derived glutamate and glucose-derived glycine are

FIGURE 5 | Crosstalk between metabolism and redox homeostasis in cancer cell mitochondria. Neoplastic cells are characterized by high levels of mitochondrial ROS. Under a certain threshold, ROS facilitate tumor growth, but their excessive rise elicits oxidative damage and cell death. In cancer cell mitochondria, ROS levels are increased by respiratory chain complexes, αKGDH, mGPDH and ACADs and inhibit key enzymes of lipid metabolism and TCA cycle. In turn, ROS stimulate antioxidant defenses through stabilization of the transcription factors NRF2 and HIF1α. Proteins overexpressed or activated in cancer cells are indicated in purple, whereas proteins whose activity is down-regulated are shown in yellow. GSH, reduced glutathione; PDC, pyruvate dehydrogenase complex; PDH, pyruvate dehydrogenase; αKGDH, alpha-ketoglutarate dehydrogenase; SDH, succinate dehydrogenase; SDHB, succinate dehydrogenase subunit B; SDHD, succinate dehydrogenase subunit D; MDH, malate dehydrogenase; TRAP1, TNF receptor-associated protein 1; FAO, fatty acid oxidation; ACAD, acyl-CoA dehydrogenase; mGPDH, mitochondrial glycerol-3-phosphate dehydrogenase; NRF2, nuclear factor-E2-related factor 2; PHD prolyl hydroxilase; Keap1, Kelch-like ECH-associated protein 1; NDUFA4L2, NADH dehydrogenase 1 alpha subcomplex, 4-like 2; COX4-2, Cytochrome c OXidase subunit 4 isoform 2; ISCU, Iron–Sulfur cluster assembly proteins.

substrates for glutathione biosynthesis. NADPH, which is essential for the regeneration of reduced glutathione, is similarly obtained either by glucose through PPP and serine metabolism, or by glutamine via ME. Other anti-oxidant systems such as peroxiredoxins, which are induced by MYC, are also highly expressed in many cancer types (139).

The transcription factor Nrf2 (nuclear factor-E2-related factor 2) is a master regulator of cell response to oxidants that undergoes proteasomal degradation following ubiquitination by KEAP1. Under oxidative stress KEAP1 is inactivated, thus allowing nuclear accumulation and activation of Nrf2 (141) (**Figure 5**). Nrf2 induces enzymes that enhance carbon flux from glutamine toward GSH biosynthesis, utilization and regeneration, and stimulates NADPH production, e.g., by controlling ME, which boosts the oxidative decarboxylation of malate to pyruvate in order to feed the TCA cycle (142). Activation of Nrf2 increases mitochondrial membrane potential, FAO, ATP levels, rate of respiration and efficiency of oxidative phosphorylation (151). These protective functions against oxidative insults suggest that Nrf2 acts as a tumor suppressor, and indeed Nrf2 activation is beneficial in cancer chemoprevention and Nrf2-deficient mice are more sensitive to chemical carcinogenesis; in addition, the absence of Nrf2 has been related to a high metastatic potential (152). Nonetheless, the role of Nrf2 on tumorigenesis is highly contingent, as Nrf2 knockout mice are protected against tumor formation in the stomach, bladder, and skin (153) and activation of the Nrf2/KEAP1 system by somatic mutations is associated with a poor prognosis in patients (152) and has been observed in several cancer types. For instance, induction of the Nrf2/KEAP1 pathway occurs in the very early steps of hepatocarcinogenesis in the resistant hepatocyte rat model, where it associates with a metabolic rewiring toward increased glucose utilization, PPP activation and OXPHOS inhibition (154). In this model also the mitochondrial chaperone TRAP1 is highly expressed from the initial, pre-neoplastic lesions (154), and it probably contributes to the anti-oxidant mechanisms of tumor cells by decreasing SDH-generated ROS (155). TRAP1 has an oncogenic activity and its expression is induced in a variety of tumor types; however, TRAP1 levels decrease in the advanced stages of a small set of epithelial cancers (132). These contrasting observations on the role of Nrf2 and TRAP1 in neoplastic progression suggest that changes in cell redox equilibrium might have different effects on tumorigenesis, probably depending on tumor type and stage.

# MITOCHONDRIAL Ca2<sup>+</sup> AND METABOLIC PLASTICITY

Calcium ions are intracellular second messengers that tune a variety of fundamental cell processes (156). Mitochondria can accumulate high amounts of Ca2+, thus acting both as Ca2<sup>+</sup> stores that control the spatial and temporal shape of Ca2+-mediated cellular signals, and as effectors that utilize Ca2<sup>+</sup> to regulate cell survival, proliferation, redox state and metabolic changes. Mitochondrial Ca2<sup>+</sup> homeostasis requires an efficient interplay between endoplasmic reticulum (ER), where most intracellular Ca2<sup>+</sup> is stocked, and mitochondria in specialized microdomains called MAMs (Mitochondria-Associated Membranes) (157). In MAMs, Ca2<sup>+</sup> is released from ER through IP3Rs (Inositol 1,4,5-triPhosphate Receptors) and is taken up by Mitochondrial Calcium Uniporter (MCU) complex, thus increasing Ca2<sup>+</sup> concentration in mitochondrial matrix (158) (**Figure 6**).

Mitochondrial Ca2<sup>+</sup> homeostasis is dysregulated in most neoplastic cells and contributes to their adaptations to stressful conditions in a fast and flexible way. Transduction pathways deregulated in cancer, such as PI3K/Akt or Ras signaling, can limit Ca2<sup>+</sup> flux to mitochondria inhibiting IP3Rs, whereas tumor suppressors such as PTEN, BRCA1 or PML favor Ca2<sup>+</sup> release from IP3Rs and the subsequent increase in mitochondrial Ca2<sup>+</sup> levels (159, 160). Oncogenic mutations in p53 decrease the activity of SERCA (sarco/endoplasmic reticulum Ca2+- ATPase), which takes Ca2<sup>+</sup> up in ER, thus enhancing Ca2<sup>+</sup> transfer to mitochondria in MAMs (160). Rapid spikes of Ca2<sup>+</sup> levels in mitochondrial matrix induce the permeability transition pore (PTP), a mega-channel formed by ATP synthase whose prolonged opening elicits a sudden cell death (161). Thus, by slowing-down mitochondrial Ca2<sup>+</sup> entry through MAMs, tumor cells can avoid to succumb to several noxious stimuli (162, 163). Hence, a fine tuning of IP3Rs activity is crucial in preventing lethal matrix Ca2<sup>+</sup> overload. A complex interplay exists between IP3R regulation and Ca2<sup>+</sup> homeostasis in mitochondria in tumors. Indeed, inhibition of Ca2<sup>+</sup> transfer from ER to mitochondria decreases the viability of tumor cells compromising their bioenergetics. Notably, in these conditions neoplastic cells activate autophagy as a salvage mechanism, but this turns out to be insufficient for their survival (164).

More controlled raises in matrix Ca2<sup>+</sup> concentration have important metabolic effects, as Ca2<sup>+</sup> enhances the activity of mitochondrial dehydrogenases of the TCA cycle, IDH and αKGDH, and of PDH (165). These dehydrogenase reactions lead to formation of NADH that carries the reducing equivalents required for OXPHOS activity (158). Therefore, mitochondrial Ca2<sup>+</sup> stimulates respiration and increases ROS generation (166). Moreover, Ca2<sup>+</sup> stimulates the aspartate/glutamate exchanger in the inner mitochondrial membrane (167), further boosting TCA cycle activity by increasing matrix glutamate levels. Taken together, these observations suggest that lowering mitochondrial Ca2<sup>+</sup> concentration could play a key role in maintaining a "Warburg-like" phenotype in neoplastic cells, while protecting them from PTP opening. However, the few studies that have directly assessed the role of mitochondrial Ca2<sup>+</sup> in the tumorigenic process sketch elements of a more complex picture. For instance, the expression of MCU, whose activity can sharply increase mitochondrial Ca2<sup>+</sup> concentration (158), is unexpectedly increased and associated to poor prognosis, invasiveness and metastasis in models of breast cancer and hepatocellular carcinoma (168, 169). Indeed, these observations directly link MCU activity to the maintenance of redox homeostasis. MCU silencing in triple negative breast cancer models decreases ROS levels by lowering ATP production and NADH cellular content. This hampers HIF-1α stability and transcriptional activity, decreasing cell motility and invasiveness, tumor growth, lymph node infiltration

and lung metastasis (169). In HCC cells, a MCU-dependent increase in matrix Ca2<sup>+</sup> concentration stimulates TCA cycle activity and augments NADH/NAD<sup>+</sup> ratio. This inhibits a Sirtuin3/superoxide dismutase 2 axis that boosts mitochondrial ROS levels, which in turn sustain invasion and metastasis of hepatocellular carcinoma cells in an in vivo xenograft model (168).

Further studies are clearly needed to dissect how frequencies and amplitudes of mitochondrial Ca2<sup>+</sup> oscillations influence the metabolic changes that characterize tumorigenesis.

## POST TRANSLATIONAL REGULATION IN CANCER METABOLISM

Mitochondria can utilize post-translational modifications (PTMs) of their proteins in order to harmonize their activity to environmental conditions. A wide assortment of PTMs can lead to conformational changes in the tertiary structure of mitochondrial proteins, tuning their activity in response to changes in nutrient availability or redox conditions (170, 171) and furnishing cancer cells with a broad array of accurate and rapid metabolic adaptations (171). The investigation of these regulatory networks, and the functional connection with the metabolic changes that characterize neoplastic cells is complex and still in its infancy, and we will provide here only some general information.

The most prevalent mitochondrial PTM is acetylation of lysine residues, presumably because it requires acetyl-CoA that is highly compartmentalized in mitochondria (172). About 30% of mitochondrial proteins can undergo reversible acetylation. In general, this is an inhibitory mark for metabolic enzymes, as it would serve to sense the overproduction of acetyl-CoA, thus providing a negative feedback to mitochondrial metabolic circuitries that operate in an oxidative mode (173). Notably, hyperacetylation of mitochondrial proteins is observed in many diseases, including cancer (170). Acetylation is determined by the balance between the activity of acetyltransferases and deacetylases. Little is known on mitochondrial acetyltransferases. The only candidate is GCN5L1, which does not contain an acetyltransferase catalytic domain but promotes protein acetylation in the presence of acetyl-CoA, and its genetic disruption down-regulates acetylation of mitochondrial proteins. GCN5L1 is involved in lipid metabolism, as its induction promotes FAO, even if no data on tumor models are at present available (174).

Deacetylation is carried out by a class of enzymes called sirtuins, a protein family composed by 7 members, three of which (SIRT3–5) have a mitochondrial localization. Recent work has demonstrated that sirtuins are indeed deacylases, as they are able to transfer a variety of long acyl moieties including succinyl, malonyl, ADP-ribosyl and lipoyl groups, in addition to perform deacetylase reactions. All these reactions require NAD+, thus linking sirtuin enzymatic activity to the metabolic state of the cell and poising them as metabolic stress sensors (175). Mitochondrial sirtuins orchestrate the coordinated regulation of substrate clusters, in order to efficiently tackle conditions of metabolic stress. SIRT3 is the major mitochondrial deacetylase and is activated upon starvation and by increased NAD<sup>+</sup> levels. In these conditions, SIRT3 enhances oxidative metabolism of fatty acids, by activating LCAD, and of amino acids, by increasing the activity of GDH and GLS2 (176, 177). SIRT3 also activates the PDH complex (PDC), thus promoting the conversion of pyruvate to acetyl-CoA (178). In parallel, SIRT3 stimulates ROS-mitigating systems such as IDH2, a TCA cycle enzyme that generates NADPH required to reduce glutathione (175), and superoxide dismutase 2, which converts superoxide to hydrogen peroxide that is then neutralized by glutathione (173). In addition, SIRT3 activates by deacetylation all OXPHOS complexes, in particular complex I and SDHA, the entry point of electrons from NADH and FADH2, respectively, thus promoting an efficient respiration (175). Taken together, these observations indicate that SIRT3 opposes a Warburg-like metabolism, and cells lacking SIRT3 exhibit genomic instability and are prone to tumorigenesis in xenografts. Accordingly, Sirt3-knockout mice develop mammary tumors, and in many human cancer types SIRT3 is deleted or expressed at a very low level (178). Nonetheless, the functional connections between mitochondrial sirtuin activity and tumor growth are multifaceted and far from being fully understood. Indeed, SIRT4 seems to have a tumorsuppressor role similar to that of SIRT3, as SIRT4-null mice develop lung tumors, loss of SIRT4 accelerates tumor progression in a mouse Burkitt lymphoma model and SIRT4 expression is reduced in several types of human cancers. However, SIRT4 plays opposing roles to SIRT3 in the regulation of several metabolic pathways: it promotes lipogenesis and represses fatty acid oxidation by inhibiting malonyl-CoA decarboxylase (170), negatively regulates PDC and represses GDH (177, 178). SIRT5, the last mitochondrial sirtuin, primarily demalonylates and desuccinylates lysine residues in a NAD+-dependent way. Its functions in the metabolic rewiring of tumor cells are poorly understood, but it might be involved in glutamine metabolism as it inhibits GLS (177) and in OXPHOS and TCA regulation, as it decreases SDH activity by targeting both SDHA and SDHB and it inhibits PDC (170). Notably, SIRT5 could act as an oncogene, as it is overexpressed and associated with poor prognosis in human lung cancer (178). Finally, a further layer of metabolic regulation could be provided by sirtuin-directed PTMs such as phosphorylations (176).

Reversible phosphorylation at serine, threonine or tyrosine residues is emerging as an important mechanism regulating several aspects of mitochondrial metabolism. For instance, the inhibitory phosphorylation of PDC via enhanced expression of pyruvate dehydrogenase kinase-1 contributes to aerobic glycolysis and malignant phenotype, whereas PDH phosphatase exerts the opposite effect (145).

Adaptations to changes in nutrients and oxygen supply require a rapid OXPHOS regulation that can be achieved via reversible phosphorylations. The mitochondrial fraction of protein kinase A (PKA) activates in response to CO<sup>2</sup> generated in the TCA cycle. PKA increases the activity of respiratory complex I through phosphorylation of its NDUFS4 subunit (179). PKA also phosphorylates a subunit of cytochrome oxidase, preventing its allosteric inhibition by ATP and acting as a metabolic sensor to match OXPHOS activity with substrate availability and energy consumption requirements (180). Several phosphorylation sites are present on ATP synthase, but how they modulate the enzyme is still largely obscure. Preliminary data indicate that they could affect not only its activity, but also its assembly and dimerization (181). Activation of the tyrosine kinase Src increases the enzymatic activity of respiratory complex IV in isolated rat brain mitochondria (182); accordingly, the Src inhibitor dasatinib down-regulates the activity of respiratory complex IV in some tumor cell models, inhibiting their ROS-dependent invasiveness (134).

Tyrosine phosphorylation of SDH subunit A by the Srclike tyrosine kinase Fgr increases SDH activity, contributing to the capacity of mitochondria to modulate metabolism in conditions of nutrient restriction or hypoxia (183). Conversely, the kinase ERK1/2 decreases SDH activity. ERK1/2, SDH and the chaperone TRAP1 form a multimeric complex in mitochondria of neurofibromin-deficient cells. Mitochondrial ERK1/2 phosphorylates TRAP1, thus enhancing its inhibition of SDH. TRAP1 ablation or mutagenesis at the Ser residues targeted by ERK1/2 abrogates the tumorigenicity of cells lacking neurofibromin (100).

The mitochondrial fraction of the Ser/Thr kinase GSK-3 down-modulates the activity of PDH and of respiratory complex I (184). Moreover, GSK-3 phosphorylates the mitochondrial chaperone cyclophilin D (CyP-D), the best characterized proteinaceous regulator of the PTP, enhancing CyP-D-dependent PTP induction. In tumor cells, mitochondrial ERK1/2 inhibits by phosphorylation GSK-3, thus antagonizing PTP opening and cell death (99). A complex array of PTMs, including acetylations and nitrosylations in addition to phosphorylation, affects CyP-D activity (162). These PTMs could subtly tune the bioenergetic status of neoplastic cells, as CyP-D binds and down-regulates the enzymatic activity of ATP synthase (161).

# MITOCHONDRIAL DYNAMICS AND CANCER METABOLISM

Mitochondria are extremely dynamic organelles, undergoing the opposite processes of fusion and fission in a coordinated and balanced way. A comprehension of how mitochondrial dynamics contribute to the metabolic rewiring of cancer cells is still in its infancy, but several evidences are emerging that link decreased fusion and enhanced fission to neoplastic transformation, invasion and metastasis (185, 186). Signaling via oncogenic MAPK and PI3K promotes fission (187), and in certain tumor settings high levels of the fission protein DRP1, whose activity is increased by ERK-dependent phosphorylation, negatively correlate with survival of patients. Conversely, overexpression of the fusion proteins mitofusins decreases tumor growth, and their levels are directly related to OXPHOS activity and ATP production in several cell models (188).

However, the assumption that glycolytic cells have fragmented mitochondria, whereas OXPHOS is increased in cells with elongated mitochondria, appears as an oversimplification (188). Indeed, a prolonged DRP1 downregulation can inhibit respiration, suggesting that a proper OXPHOS modulation requires a balanced interplay between mitochondrial fission and fusion (189). Thus, it is difficult at present to make mechanistic correlations between mitochondrial dynamics and metabolism and to understand whether changes in the mitochondrial network of cancer cells are priming events or consequences of their metabolic rewiring. For instance, even if activation of several oncogenes increases fission, some others, such as Myc, promote mitochondrial fusion (187). Moreover, mitochondrial shape can impact on intramitochondrial Ca2<sup>+</sup> waves (185) and on Ca <sup>2</sup><sup>+</sup> fluxes at MAMs (189), thus playing a complex and probably context-dependent role on the metabolic adaptations of tumor cells (see section Mitochondrial Ca2<sup>+</sup> And Metabolic Plasticity).

Mitochondrial fission is strictly connected to mitophagy, a quality control process that maintains mitochondrial integrity and function through removal of damaged organelles, which must be isolated from the healthy network via DRP1-dependent sequestration (189, 190). Mitophagy is activated by a variety of stresses usually encountered by neoplastic cells, including hypoxia, nutrient deprivation, DNA damage and inflammation, which eventually cause mitochondrial membrane depolarization and decline in respiratory capability (149). Therefore, any impairment in the mitophagy process leads to accumulation of dysfunctional mitochondria, hence decreasing respiration and ATP production and increasing ROS levels. In general, defects in mitophagy affect the metabolic plasticity of mitochondria in response to environmental stresses such as altered Ca2<sup>+</sup> signaling, ROS generation and changes in nutrient availability, further amplifying their noxious effects on the cell. In cancer, a disruption of the homeostatic equilibrium between mitophagy and mitogenesis occurs (191, 192). It has been proposed that impairment of a correct mitophagy could be advantageous for the early phases of neoplastic growth, contributing to set a novel redox equilibrium, whereas later stages of tumor progression would be favored by mitophagy, as it would protect tumor cells from excessive mitochondrial damage, surge in ROS levels and apoptosis (189). Further work is certainly needed to draw a more complete picture of the functional interplay between mitochondrial fusion and fission, mitophagy and metabolic rewiring of cancer cells.

# CONCLUDING REMARKS

Metabolism is a multilevel process, encompassing and integrating a myriad of factors both at the organismal scale, such as age or lifestyle, and at the local level, including cellular composition of the microenvironment, nutrient supply and stiffness of the extracellular matrix. Accordingly, aberrant metabolic reprogramming in cancer is both the cause and the effect of alterations at multiple levels that reverberate on each other.

The (epi)genomic landscape of neoplastic cells, the intertwined molecular signaling between immune, stromal and other non-transformed cells with the malignant ones in tumor microenvironment (193, 194), as well as increases in hydrostatic forces and in the stiffness of the tumor milieu (195), are all factors that constantly tune cancer cell metabolism to fluctuating environmental conditions, leading to metabolic heterogeneity also across different areas of the same tumor. In neoplastic cells, mitochondria constitute a point of integration for many of these metabolic circuitries. For instance, reciprocal feedbacks exist between mechanosignaling, the process by which cells convert extracellular mechanical forces into biochemical outputs, and glutamine metabolism, as glutamine partly controls focal adhesions and actin stress fiber assembly, and in turn stiffness changes glutamine fluxes (196).

Moreover, mitochondria are at the heart of mechanisms that balance a variety of intracellular metabolic circuitries. For instance, NADPH homeostasis is maintained by several mitochondrial metabolic circuitries via TCA cycle intermediates and ATP generated by OXPHOS. These pathways include one carbon metabolism, PPP, whose oxidative branch is enhanced by citrate- or ATP-dependent inhibition of late glycolytic steps (5, 11), glutamine-derived carbons diverted out of the TCA cycle to convert malate into pyruvate via ME, and IDH (6, 197, 198). p53 represses the transcription of ME genes, thus inhibiting the usage of TCA cycle intermediates for NADPH production (68). Conversely, citrate inhibits PFK, pyruvate kinase and PDH, blocking pyruvate generation from glycolysis and leading to an increase in ME activity to maintain pyruvate levels, but also to provide NADPH (11). Another important source of NADPH is lipid β-oxidation, which becomes a major fuel for ATP synthesis during invasion and metastasis (199), further highlighting the tight connection between metabolic adaptations and biological conditions in cancer cells.

Citrate and acetyl-CoA provide other examples of multiple metabolic intersections. Mitochondrial citrate is a TCA cycle metabolite that originates by the condensation of OAA and acetyl-CoA, but it also forms via glutamine-fueled reductive carboxylation (47, 200), which is particularly important under hypoxic conditions and supports anchorage-independent growth of neoplastic cells by mitigating oxidative stress through a coordinated regulation of NADH/NADPH dependent IDH1 and 2 in cytosol and mitochondria (201). In mitochondria, citrate down-regulates SDH (11), i.e., the point of integration between TCA cycle and OXPHOS, contributing to raise the levels of the oncometabolite succinate. Alternatively, citrate can move to cytosol, where it both favors glucose usage in PPP and serine synthesis by inhibiting the late glycolytic steps and is converted into acetyl-CoA (11). In turn, acetyl-CoA starts lipid synthesis and sustains acetylation reactions (see section Post Translational Regulation In Cancer Metabolism) and influences a variety of biochemical circuitries involved in the neoplastic process, including histone acetylation and chromatin remodeling as well as redox homeostasis via acetylation of superoxide dismutase and IDH (1). Regulation of histone acetylation showcases the fundamental role played by metabolic enzymes and metabolites in gene expression control (117). One interesting example is provided by the PDC, which locates both in mitochondria and nucleus of prostate cancer cells. This compartmentalization is instrumental to orchestrate lipid biosynthesis both by providing cytosolic citrate and by inducing the transcription of genes for lipid synthesis by regulating histone acetylation (202).

As described in section Mutations Of Mitochondrial Enzymes In Cancer Metabolism, tumors with mutations in enzymes of the TCA cycle constitute an excellent model to study how mitochondrial metabolism causes pro-neoplastic (epi)genomic changes (203), and they also provide clues to identify molecular vulnerabilities that can be exploited for anti-tumor strategies. For instance, pyruvate carboxylase (PC) enables aspartate synthesis in SDH-deficient tumor cells, creating a metabolic vulnerability. lack of SDH activity commits cells to consume extracellular pyruvate, which sustains Warburg-like bioenergetic features. Pyruvate carboxylation diverts glucose-derived carbons into aspartate biosynthesis, thus sustaining cell growth (204, 205).

Further layers of complexity are provided by the heterogeneity of cellular components of the tumor mass. For instance, cancer stem cells (CSC) constitute a small population of self-renewal neoplastic cells that are per se capable of promoting tumor growth. The metabolic features of CSC differ from those of the bulky neoplasm, and recent evidences point toward an OXPHOS phenotype in CSC. In a sort of reverse Warburg metabolism, OXPHOS CSC might be fed by glycolytic tumor cells or by cells of the tumor microenvironment, such as cancer associated fibroblasts, and could switch to a glycolytic metabolism under hypoxia (206). Examples of symbiotic nutrient sharing between neoplastic cells and tumor microenvironment are observed in metastatic ovarian cancer cells, breast cancer cells or leukemic stem cells, which oxidize fatty acids supplied from surrounding adipocytes to sustain proliferation, survival and invasiveness and possibly to preserve cell redox balance (207, 208). Moreover, horizontal transfer of mitochondria from stromal to cancer cells

#### REFERENCES


can lead to an increase in OXPHOS metabolism of the latter (19, 209).

Metabolic plasticity of tumor cell mitochondria offers a high window of opportunity for efficient anti-cancer therapy, since transformed cells have metabolic needs that differ from their non-transformed counterparts, and molecules such PDH, IDH1/2 or glutaminase inhibitors are already in clinical trials. Nonetheless, such a plasticity can also be a hurdle when trying to develop selective therapeutic strategies. As an example, ovarian cancer cells gain resistance to antiangiogenic therapy by shifting their metabolic phenotype toward a highly glycolytic one (210). Importantly, tumor cell dependency on specific bioenergetic features in vitro can be extremely different from the in vivo situation, due to off-target effects, suboptimal pharmacokinetic properties of the compound or metabolic heterogeneity of the neoplastic mass.

A profound comprehension of the biochemical mechanisms that govern the bioenergetic flexibility of tumor cell mitochondria and its interplay with a multitude of extramitochondrial signals constitutes a central dowel to build an integrate model of the metabolic features that hallmark cancer. Incorporation of data obtained at different scales of analysis, from the organism to the organelle, remains a tremendous task. Nonetheless, huge advances have been recently made unveiling biochemical processes and therapeutic opportunities that were unimaginable even few years ago.

#### AUTHOR CONTRIBUTIONS

All authors contributed to writing the text and drawing the figures and approved the manuscript for publication.

#### ACKNOWLEDGMENTS

Research in the AR group is supported by grants from University of Padova, Children's Tumor Foundation Drug Discovery Initiative (grant 2016A-05-009), Neurofibromatosis Therapeutic Acceleration Program, Associazione Italiana Ricerca Cancro (AIRC grant IG 2017/20749), Piano for Life Onlus and Linfa Onlus. IM is supported by a post-doctoral Young Investigator Award Grant from the Children's Tumor Foundation.


and smooth muscle tumors. Oncotarget (2015) 6:38777–88. doi: 10.18632/oncotarget.6091


from endoplasmic reticulum to mitochondria. Cell Rep. (2016) 14:2313–24. doi: 10.1016/j.celrep.2016.02.030


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

Copyright © 2018 Cannino, Ciscato, Masgras, Sánchez-Martín and Rasola. 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.

# LonP1 Differently Modulates Mitochondrial Function and Bioenergetics of Primary Versus Metastatic Colon Cancer Cells

*Lara Gibellini1 , Lorena Losi2 , Sara De Biasi2 , Milena Nasi3 , Domenico Lo Tartaro1 , Simone Pecorini3 , Simone Patergnani4 , Paolo Pinton4 , Anna De Gaetano3 , Gianluca Carnevale3 , Alessandra Pisciotta3 , Francesco Mariani5 , Luca Roncucci5 , Anna Iannone5 , Andrea Cossarizza1 and Marcello Pinti <sup>2</sup> \**

#### *Edited by:*

*Andrea Rasola, Università degli Studi di Padova, Italy*

#### *Reviewed by:*

*Paolo E. Porporato, Università degli Studi di Torino, Italy Carlo Pucillo, Università degli Studi di Udine, Italy*

*\*Correspondence: Marcello Pinti marcello.pinti@unimore.it*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

> *Received: 01 February 2018 Accepted: 21 June 2018 Published: 09 July 2018*

#### *Citation:*

*Gibellini L, Losi L, De Biasi S, Nasi M, Lo Tartaro D, Pecorini S, Patergnani S, Pinton P, De Gaetano A, Carnevale G, Pisciotta A, Mariani F, Roncucci L, Iannone A, Cossarizza A and Pinti M (2018) LonP1 Differently Modulates Mitochondrial Function and Bioenergetics of Primary Versus Metastatic Colon Cancer Cells. Front. Oncol. 8:254. doi: 10.3389/fonc.2018.00254*

*1Department of Medical and Surgical Sciences for Children and Adults, University of Modena and Reggio Emilia, Modena, Italy, 2Department of Life Sciences, University of Modena and Reggio Emilia, Modena, Italy, 3Department of Surgery, Medicine, Dentistry and Morphological Sciences, University of Modena and Reggio Emilia, Modena, Italy, 4Department of Morphology, Surgery and Experimental Medicine, Section of Pathology, Oncology and Experimental Biology and LTTA Center, University of Ferrara, Ferrara, Italy, 5Department of Diagnostic, Clinical Medicine and Public Health, University of Modena and Reggio Emilia, Modena, Italy*

Mitochondrial Lon protease (LonP1) is a multi-function enzyme that regulates mitochondrial functions in several human malignancies, including colorectal cancer (CRC). The mechanism(s) by which LonP1 contributes to colorectal carcinogenesis is not fully understood. We found that silencing LonP1 leads to severe mitochondrial impairment and apoptosis in colon cancer cells. Here, we investigate the role of LonP1 in mitochondrial functions, metabolism, and epithelial–mesenchymal transition (EMT) in colon tumor cells and in metastasis. LonP1 was almost absent in normal mucosa, gradually increased from aberrant crypt foci to adenoma, and was most abundant in CRC. Moreover, LonP1 was preferentially upregulated in colorectal samples with mutated p53 or nuclear β-catenin, and its overexpression led to increased levels of β-catenin and decreased levels of E-cadherin, key proteins in EMT, *in vitro*. LonP1 upregulation also induced opposite changes in oxidative phosphorylation, glycolysis, and pentose pathway in SW480 primary colon tumor cells when compared to SW620 metastatic colon cancer cells. In conclusion, basal LonP1 expression is essential for normal mitochondrial function, and increased LonP1 levels in SW480 and SW620 cells induce a metabolic shift toward glycolysis, leading to EMT.

Keywords: LonP1, mitochondria, bioenergetics, beta-catenin, colorectal cancer, protease

# INTRODUCTION

LonP1 (also known as Lon or LonP) is one of the main proteases patrolling the mitochondrial matrix. Lon is a multi-function enzyme, exerting both proteolytic and chaperone activities, and also binds mitochondrial DNA and RNA [reviewed in Ref. (1)]. Main targets of LonP1 proteolytic activity are: (i) folded proteins, including 5-aminolevulinic acid synthase, steroidogenic acute regulatory protein, mitochondrial transcription factor A, and cytochrome c oxidase 4 isoform 1; (ii) misfolded proteins, including glutaminase C; (iii) oxidized proteins, including aconitase and cystathionine β-synthase

**30**

(2–8). Conversely, targets of LonP1 chaperone activity are still not known. Initial studies demonstrated that LonP1 was involved in mitochondrial maintenance and mitochondrial quality control during aging (9, 10). More recent evidence suggests that LonP1 is responsible for additional functions critical to tumor progression, including metabolic adaptation to hypoxia, protection against senescence, and resistance to apoptosis and oxidative stress (11–14). Recent data from our group demonstrated that LonP1 is upregulated in colorectal tumors, and that its downregulation or inhibition leads to severe mitochondrial dysfunction and to apoptosis (15, 16).

Colorectal cancer (CRC) is the third most common cancer worldwide (17). Despite several advances have been made in the diagnosis and treatment of CRC over the past decades, the overall prognosis still remains poor, and tumor metastasis represent a major obstacle to successful treatment (18). The epithelial–mesenchymal transition (EMT) is a biologic process that enables a polarized epithelial cell to assume a mesenchymal cell phenotype through biochemical changes and a number of distinct molecular processes (19). Several intracellular signaling pathways contribute to EMT, and involve ERK, MAPK, PI3K, Akt, Smads, β-catenin (β-ctn), Ras, c-fos, and among others (19). The Wnt/β-ctn signaling pathway has a crucial role in the negative regulation of E-cadherin, and in the development of EMT and CRC metastasis (20). In normal conditions, and in the absence of activated Wnt signals, β-ctn is phosphorylated by APC/Axin/ GSK-3β complex and then degraded by the proteasome. When Wnt ligands activate Frizzled and LPR receptors, β-ctn is no longer phosphorylated and translocates into the nucleus where it binds to transcription factors belonging to the family of T-cell factor (TCF) and lymphoid enhancer-binding protein, and then it activates transcription (21). In CRC, the vast majority of tumors have mutations in the key regulatory factors of the Wnt/β-ctn pathway and up to 80% present nuclear accumulation of β-ctn (20).

Starting from our previous observations, we aimed at investigating the precise role of LonP1 in colon primary tumor and metastasis. In this study, we show that elevated expression of LonP1 in CRC tissues is associated with nuclear localization of βctn, and that overexpression of LonP1, *in vitro*, differently affects β-ctn expression in SW480 and SW620 colon cancer cells. We also show that LonP1 overexpression differently affects mitochondrial functions and bioenergetics in SW480 and SW620 cells.

#### MATERIALS AND METHODS

#### Cell Culture

Four cancer cell lines were used for this study: I407 intestinal epithelial cells, SW480 and RKO colon carcinoma cells, and SW620 metastatic colon cancer cells. SW480 cells were cultured in DMEM high glucose supplemented with 10% fetal bovine serum (FBS) and gentamycin. I407, RKO, and SW620 cells were cultured in RPMI Glutamax supplemented with 10% FBS and gentamycin. Cells were maintained in 5% CO2 atmosphere at 37°C. Culture media and reagents were from ThermoFisher Scientific (Eugene, OR, USA).

#### Human Colorectal Tissues

We have studied a total of 45 patients who underwent surgical removal of CRC. Samples used for immunohistochemistry were formalin-fixed paraffin-embedded (FFPE) specimens, the other samples had been freshly frozen with a passage in liquid nitrogen, and then stored at −80°C until use. Tissues were obtained from the Department of Diagnostic and Clinical Medicine, and Public Health, University of Modena and Reggio Emilia, through an institutional review board-approved protocol. Demographic and clinical characteristics of patients are reported in **Tables 1** and **2**.

#### Retroviral Transduction

The pMSCV-Puro empty vector and the pMSCV containing the cDNA encoding for Lon protease (hereafter referred to as pLonP1) were used to transiently transfect amphotrophic Phoenix cell line (15). Cells infected with the empty vector will be indicated as pMSCV cells. Retroviral supernatants were used to stably transfect I407, RKO, SW480, and SW620 cells, and stable transfectants were selected by using 3–4 µg/ml puromycin (depending on the cell line), and then maintained in cell medium supplemented with 2 µg/ml puromycin.

#### RNA Interference

Cells were reverse transfected by using RNAiMAx (Life Technologies Corporation) and 10 nM s17901 small interfering RNAs (Life Technologies Corporation) against LonP1 mRNA. Then, cells were incubated for 72 h, trypsinized, and lysated by using RIPA buffer.



*M, male; F, female; NM, normal mucosa; ACF, aberrant crypt foci; Ad, adenoma; CRC, colorectal cancer.*


Table 2 | Demographic characteristics, tumor stage, Ki-67, β-catenin (β-ctn), E-cadherin (E-cad), p53, LonP1 expression in 21 colorectal cancer samples.

*M, male; F, female; Me, membrane; N, nuclear; P, present.*

#### Immunohistochemical Analyses

Tumor specimens were taken from patients who underwent surgical resection of the large bowel in the period 2010–2015. All slides were blindly reviewed by the pathologist (LL). For each case, a representative paraffin-embedded block containing tumor tissue and normal mucosa, as internal control, was sectioned at 4 µm. Immunoperoxidase staining was run with the Benchmark XT Automatic Staining System (Ventana Roche) with diaminobenzidine as chromogen and using the View DAB Detection Kit (Roche). At the end of the reaction, slides were counterstained with hematoxylin. The following antibodies were used: mouse monoclonal antibodies β-ctn (Roche, Basilea, Switzerland), p53 (Roche), and E-cadherin (Dako, Santa Clara, CA, USA) available as pre-diluted commercially preparations; mouse monoclonal Ki-67 antibody (MIB-1, Dako, used at 1:100 dilution), and rabbit polyclonal LonP1 antibody (Primm, Milan, Italy, used at a 1:500 dilution). According to previous reports, expression of β-ctn and E-cadherin in normal colon epithelium resulted to be restricted to cell membrane. Altered expression of β-ctn was contemplated when 10% of tumor cells or greater showed nuclear or cytoplasmic immunoreactivity. Loss of membrane expression of E-cadherin was considered in cases exhibiting either no immunoreactivity or 10% of tumor cells or less with positive membranous staining. Detection of proliferative activity was carried out using an anti-Ki-67 antibody, and basal cells of the normal colonic crypts were used as internal positive control. Ki-67 labeling index was determined by counting the number of positive nuclei for 1,000 neoplastic cells in 10 consecutive fields chosen randomly in nonnecrotic areas of the tumor.

#### Immunoblotting

Total protein lysate was obtained from cell lines by using RIPA buffer and from FFPE tissue as previously described (22). Protein concentration was determined by Bradford assay. Proteins were then separated in 4–12% or 12% Bolt Bis-Tris precast gels (Thermo Fisher Corporation) and transferred onto nitrocellulose membranes. Protein transfer was performed by using methanol transfer buffer or by using Trans-Blot Turbo cassettes (Bio-Rad Laboratories) with Trans-Blot Turbo blotting system (Bio-Rad Laboratories). Rabbit polyclonal anti-LonP1 was a custom antibody from Primm (Milan, Italy). Rabbit polyclonal anti-β-actin and rabbit polyclonal anti-E-cadherin were from Abcam (Cambridge, UK). Rabbit polyclonal to β-ctn, to lactate dehydrogenase A (LDHA), to glucose 6-phosphate dehydrogenase (G6PD), to N-cadherin, to Akt, to phospho-Akt (Ser473), to GSK-3β, to phospho-GSK-3β (Ser9), and to Twist were from Cell Signaling Technology (Danvers, MA, USA). Images were acquired by using ChemiDoc MP (Bio-Rad Laboratories) and Image Lab version 5.2.1 was used to perform densitometric analysis.

#### Immunofluorescence

Cells were grown on coverslips and fixed for 15 min in PBS containing 4% paraformaldehyde. Samples were then washed three times with PBS, incubated in PBS containing 0.1% Triton X-100 for 1 h at room temperature, and blocked by using PBS containing 3% bovine serum albumin for 1 h. Samples were then incubated with PBS containing primary antibody and secondary antibodies for 60 and 30 min, respectively. The following antibodies were used: rabbit polyclonal anti-β-ctn (Cell Signaling Technology) and anti-rabbit Alexa Fluor 647. Images were acquired by using a Nikon A1 confocal laser scanning microscope (Nikon, Tokyo, Japan).

#### Transmission Electron Microscopy

Cell pellets were fixed in 2.5% glutaraldehyde in Sorensen's Phosphate Buffer 0.1 M pH 7.4 (PB) for 1 h and post-fixed in 1% OsO4 in PB 0.1 M for 1 h, as previously described (23). Then, ultrathin sections were cut from Durcupan embedded samples, collected on nickel grids, stained with uranyl acetate and leadcitrate, and then analyzed by using a Zeiss EM 109 Transmission Electron Microscope (Zeiss AG, Jena, Germany).

#### Oxygen Consumption Rate (OCR)

The rate of oxygen consumption was assayed with the XF96 Extracellular Flux Analyzer (Seahorse Biosciences—Agilent Technologies, Santa Clara, CA USA). Cells were plated 2 days before the experiment, and experiments were performed on a confluent monolayer. The number of cells that were plated was: 5 × 104 /well for SW620-pMSCV and SW620-pLonP1, and 4 × 104 /well for SW480-pMSCV and SW480-pLonP1.

#### Cytofluorimetric Analyses

Flow cytometry was used to determine mitochondrial mass, mitochondrial membrane potential (MMP), mitochondrial reactive oxygen species (ROS), and glucose transporter (GLUT)-1 expression. For mitochondrial mass analysis, cells were incubated with MitoTracker Green FM (MTG, 200 nM, Thermo Fisher Corporation) for 30 min at 37°C. For MMP analysis, cells were incubated with tetramethylrhodamine, methyl ester (TMRM, 200 nM, Thermo Fisher Corporation) for 10 min at 37°C. Carbonyl cyanide m-chlorophenyl hydrazine (CCCP, 1 and 10 µM, Sigma Aldrich) was used to induce mitochondrial membrane depolarization. Mitochondrial superoxide production was assessed using MitoSox Red Mitochondrial Superoxide Indicator (mtSOX, 5 µM, ThermoFischer Corporation). GLUT-1 expression was assessed by staining cells with anti-GLUT-1 Alexa Fluor 488 (Abcam). Data were collected on an Attune NxT (Thermo Fisher Corporation) and analyzed with FlowJo software (Tree Star, Ashland, OR, USA).

#### Statistical Analysis

Quantitative variables were compared with non-parametric Mann–Whitney test. Statistical analyses were performed using GraphPad 5.0 (Prism, La Jolla, CA, USA). Error bars represent SD. A *p* value <0.05 was considered significant.

#### RESULTS

## LonP1 Is Upregulated in CRC and Is Associated With Mutated p53 or Nuclear **β**-ctn

We previously demonstrated that LonP1 silencing is associated with severe mitochondrial dysfunction and apoptosis susceptibility in colon cancer cell lines (22). Herein, we quantified the expression of LonP1 during colon cancer progression in fresh frozen tissues of ACF, adenoma (Ad), and CRC from a total of 24 patients, whose demographic and clinical characteristics are reported in **Table 1**. We found that the LonP1 was almost absent in normal mucosa, gradually increased from samples of ACF to Ad, and was most abundant in samples of CRC (**Figures 1A,B**). In addition, we quantified LonP1 levels in FFPE samples from 21 patients affected by CRC, whose demographic and clinical

Figure 1 | LonP1 expression during colon cancer progression. (A) Representative Western blot analysis of LonP1 expression in freshly frozen tissue samples from normal mucosa (NM), aberrant crypt foci (ACF), colonic adenoma (Ad), and colorectal cancer (CRC). (B) Relative protein level of LonP1 in NM, ACF, Ad, and CRC. Data are reported as mean ± SD; \**P* < 0.05 and \*\**P* < 0.01.

characteristics are reported in **Table 2**. Despite inter-individuals variations, LonP1 was upregulated in CRC samples compared with the normal mucosa counterpart (**Figures 2A–D**).

In the same FFPE samples, we explored the expression levels of EMT-related proteins together with mutated p53 and Ki-67, and their association with LonP1. Representative images of immunohistochemical staining are reported in **Figure 3A**. We found that LonP1 was preferentially upregulated in colorectal samples with mutated p53 or nuclear β-ctn (**Figure 3B**). No differences were observed regarding Ki-67 levels.

# LonP1 Modulation Is Associated With Changes in **β**-ctn Levels

To investigate the link between LonP1 and EMT in colon, we took advantage of four different cell lines, I407, RKO, SW480, and SW620 cells. I407 are intestinal epithelial immortalized cells. RKO and SW480 are colon carcinoma cells, whereas SW620 are metastatic colon cancer cells. Moreover, SW480 and SW620 represent a primary adenocarcinoma tumor and a lymph node metastasis from the same patient, respectively. We first established stable LonP1 overexpression in SW480 cells, and quantified the levels of several EMT-related proteins, including β-ctn, E-cadherin, and N-cadherin. Forced LonP1 expression led to increased β-ctn levels, reduced E-cadherin, and increased N-cadherin in primary adenocarcinoma cells SW480 (**Figure 4A**, left panels). Similar results were obtained by upregulating LonP1 in I407 and RKO cells (Figures S1A,C in Supplementary Material). In particular, forced LonP1 expression in these cells led to threefold decrease of E-cadherin levels and 2.5-fold increase of β-ctn levels in I407 and RKO cells, respectively. I407-pLonP1 cells also exhibited an elongated, mesenchymal-like morphology, which is typically observed in EMT (Figure S1B in Supplementary Material). Modulation of β-ctn in SW480-pLonP1 cells was also confirmed by fluorescence microscopy (**Figure 4B**). The

upregulation of LonP1 in SW620 cells, which are metastatic cells, led to opposite results (**Figure 4A**, right panels). Interestingly, downregulation of LonP1 in SW620 cells was associated with reduced β-ctn expression, increased E-cadherin expression, and unchanged N-cadherin in SW620 cells (**Figure 4C**). Considering that Twist is a key regulator of EMT, its expression was analyzed in SW480 cells overexpressing LonP1. Twist levels were almost twofold increased in SW480-pLonP1 cells, if compared to control SW480-pMSCV cells (**Figure 4D**).

Our study confirmed that LonP1 modulation is associated with changes in β-ctn levels and distribution, both *in vitro* and *ex vivo*, in samples from CRC patients. Considering the central role of Akt and GSK-3β in the regulation of β-ctn, we examined the phosphorylation status of Akt and GSK-3β, at serine 473 and serine 9, respectively. We found that LonP1 overexpression led to increased levels of phosphorylated Akt (p-Akt) and phosphorylated GSK-3β (p-GSK-3β), whereas LonP1 depletion led to a slight decrease in p-Akt and p-GSK-3β (**Figures 5A,B**).

#### LonP1 Impacts Glycolysis in Colon Cancer Cells

As Akt/GSK-3β signaling pathway plays a critical role in regulating glucose metabolism (24), in SW480 and SW620 cells we examined whether forced LonP1 expression had effects on the expression of GLUT-1, LDHA, and G6PD. Plasma membrane GLUT-1 levels increased in SW480 cells overexpressing LonP1 (**Figure 6A**). LDHA levels increased whereas G6PD was almost undetectable in SW480 cells overexpressing LonP1 (**Figure 6B**). In this model, glucose 6-phosphate is likely converted into fructose 6-phosphate, thus entering the glycolytic pathway, rather than being converted into glucose 6-phosphate-gluconolactone, and entering the pentose phosphate pathway. In SW620 cells, both LDHA and G6PD levels increased in the presence of high levels of LonP1. We, therefore, analyzed the glycolytic activity of these cells by performing a real-time analysis of the extracellular acidification rate (ECAR, **Figure 6C**). SW480-pLonP1 cells exhibited a higher maximal glycolysis compared with SW480-pMSCV control cells. Quantitative analysis showed no significant differences in basal glycolysis. No changes were observed in SW620 cells.

Considering that LonP1 has been involved in controlling tumor bioenergetics by reprogramming mitochondrial functions, we analyzed mitochondrial activity by monitoring OCR. Representative curves showed that a difference in mitochondrial respiration is present between SW480-pLonP1 and control cells (**Figure 7A**); quantitative analysis showed that basal OCR was higher in SW480-pLonP1 cells. Basal and coupled respiration as well as maximal respiration and spare respiratory capacity were lower in SW620-pLonP1 cells than in SW620-pMSCV cells (**Figure 7A**). Concerning OCR, slight differences were observed in I407 or RKO cells in the presence of higher levels of LonP1 (Figure S2A in Supplementary Material). Since MMP is required for production of ATP during oxidative phosphorylation (OXPHOS), we analyzed MMP in these cells, by using TMRM. MMP was unchanged in SW480-pLonP1 and I407-pLonP1 cells (**Figure 7A**; Figure S2B in Supplementary Material). However, both SW480-pLonP1 and control cells were sensitive to CCCPinduced MMP depolarization (**Figure 7B**). Interestingly,

SW620-pLonP1 and RKO-pLonP1 cells had depolarized mitochondria if compared to SW620-pMSCV and RKO-pMSCV cells, respectively (**Figure 7B**; Figure S2B in Supplementary Material). The reduction of MMP in SW620-pLonP1 was maintained when cells were treated with 10 µM CCCP. As mitochondrial activity is a critical source of ROS, and in particular of anion superoxide, we analyzed its levels by using MitoSOX Red Mitochondrial Superoxide indicator (MitoSOX). We found that SW480-pLonP1 cells produced higher levels of anion superoxide compared to control cells, at the basal level (**Figure 7C**). When challenged with hydrogen peroxide (H2O2), no differences were observed between cells overexpressing LonP1 and control cells (**Figure 7C**; Figure S2C in Supplementary Material). Mitochondrial superoxide decreased in SW620-pLonP1 if compared to SW620-pMSCV control cells, both at the basal level and after treatment with H2O2 as pro-oxidant stressor (**Figure 7C**).

Finally, we asked whether LonP1 modulation had impact on mitochondrial mass and mitochondrial ultrastructure. Mitochondrial mass was slightly decreased in pLonP1 cells (**Figure 7D**). Analysis of mitochondrial ultrastructure revealed that in SW480-pMSCV and SW480-pLonP1 cells mitochondria were numerous, occupied most of the cytoplasm and displayed *cristae* fragmentation together with the presence of vacuoles and vesicles within the mitochondrial matrix (**Figure 7E**). SW620-pLonP1 cells displayed mitochondrial alterations to a much lesser extent, both in number and in shape, with reduced *cristae* and increased vacuoles, and a noteworthy abundance of free ribosomes in the cytoplasm (**Figure 7E**), reasonably due to a massive synthesis of endogenous proteins, which correlates with particularly aggressive and undifferentiated neoplasms. The counterpart SW620-pMSCV cells did not show altered morphology, or number of mitochondria, whose *cristae* appeared preserved (**Figure 7E**).

Figure 4 | LonP1 modulates E-cadherin, N-cadherin, and β-ctn in SW480 colon primary tumor cells and SW620 metastatic tumor cells. (A) Representative Western blot analysis and relative protein level of LonP1, β-catenin (β-ctn), E-cadherin (E-cad), N-cadherin (N-cad) in SW480 and SW620 cells overexpressing LonP1 (namely pLonP1) and in control cells, i.e., cells stably transfected with empty vector (namely pMSCV). β-actin was used as loading control. Densitometries are reported in histograms, and data are reported as mean ± SD (*n* = 3). \**P* < 0.05 and \*\**P* < 0.01. (B) Representative confocal microscopy image showing β-ctn localization in SW480-pMSCV and SW480-pLonP1 cells. Bar, 10 μm. (C) Representative Western blot analysis and relative protein level of LonP1, β-ctn, and E-cad SW620, where LonP1 has been downregulated by using small interfering RNAs (siRNAs) against LonP1 mRNA (siLonP1). Control cells were transfected with scramble siRNAs and are indicated as siCTRL. Densitometries are reported in histograms, and data are reported as mean ± SD (*n* = 3). \**P* < 0.05 and \*\**P* < 0.01. (D) Representative Western blot analysis and relative protein level of Twist in SW480 cells overexpressing LonP1 (namely pLonP1) and in control cells, i.e., cells stably transfected with empty vector (namely pMSCV). β-actin was used as loading control. Densitometries are reported in histograms, and data are reported as mean ± SD (*n* = 3).

# DISCUSSION

Several studies have detected frequent alterations in the expression of mitochondrial proteases in a variety of human tumors, suggesting that these proteins may play a role as a novel class of tumor promoters or suppressors (14, 25–28). LonP1 is a mitochondrial protease and chaperone located in the mitochondrial matrix. Initial studies placed LonP1 among "stress response proteins," that is those proteins upregulated in response to cell stress (29, 30). More recently, LonP1 has been implicated in the control of metabolic networks in mitochondria in melanoma cells (14), and in hypoxic adaptation in glioma cells (11). We previously showed that: (i) LonP1 is upregulated in several cancer cell lines, including RKO, and in CRC tissues if compared to adjacent normal mucosa, and that (ii) colon cancer cells with low levels of LonP1 displayed reduced levels of OXPHOS complexes, reduced OCR, and increased mitochondrial ROS and highly fragmented and altered mitochondria (22). However, the

histograms, and data are reported as mean ± SD (*n* = 3). \**P* < 0.05.

precise role of LonP1 in colon cancer progression has not been clarified.

In this study, we analyzed the expression of LonP1 in different stages of colon cancer and the consequences of its upregulation at the mitochondrial and cellular level. The first observation is that LonP1 expression gradually increases in normal mucosa, ACF, Ad, and CRC. The fact that LonP1 levels are higher in CRC could suggest a potential role for LonP1 in EMT, an early step in the formation of metastasis (31). We observed high levels of LonP1 in CRC tissues with nuclear localization of β-ctn or in CRC tissues with mutated p53. It is noteworthy that both β-ctn and p53 are directly involved in EMT: β-ctn/ TCF4 complex induces EMT through transcription activation of ZEB1 (32), whereas p53 regulates the transcription of genes that are involved in pathways that suppress tumor metastasis, and mutations of p53 can precede metastasis (33). Although previous studies have reported that LonP1 overexpression is associated with decreased E-cadherin, increased N-cadherin and vimentin, a possible link between LonP1 and β-ctn has never been reported (14, 34).

β-ctn is a key component of the Wnt signaling pathway, and acts as negative regulator of E-cadherin in the induction of EMT. In the absence of Wnt stimulation, cytoplasmic β-ctn is phosphorylated by the APC/Axin/GSK-3β complex, and is degraded into the proteasome (35). We reported that LonP1 modulation led to important changes in total β-ctn levels, in several colon cancer cell lines. To understand which pathway was involved in LonP1 regulation of β-ctn, we investigated the phosphorylation status of Akt and GSK-3β, as Akt can phosphorylate, and thus inactivate GSK-3β (36). To explore the role of LonP1 in Akt/GSK-3β signaling pathway, SW480 and SW620 colon carcinoma cell lines were chosen as models since they are derived from primary and secondary tumors resected from the same patient, that makes them a valid tool to investigate changes in colon cancer progression (37). In SW480 cells, the upregulation of an oncoprotein or a protein-like LonP1 would lead to EMT. On the contrary, the downregulation of the same protein in SW620 cells, which are already metastatic, would leave unchanged the mesenchymal phenotype. Indeed, forced LonP1 expression in SW480 led to increased levels of mesenchymal markers, and LonP1 upregulation was associated with increased phosphorylation of Akt and GSK-3β, thus highlighting a role for LonP1 as a regulatory factor in the Wnt/β-ctn pathway. Accordingly, it was interesting to observe that Twist, a key promoter of cancer progression, was involved in EMT. A number of studies have suggested that Twist induces EMT *via* AKT/GSK-3β/β-catenin pathways, among others (38).

While previous studies regarding LonP1 and cancer reported that LonP1 induced EMT through ROS-dependent MAPK signaling, at least in 293 T cells, we found that mitochondrial ROS were not implicated in EMT induction (34). Rather, other mechanisms could be involved, including the alteration of pathways involved in the rearrangement of cellular metabolism. It has been reported that LonP1 controls tumor bioenergetics by remodeling subunits of electron transport chain (14). We found that in colon cancer cells LonP1 can influence glycolytic

activity together with mitochondrial activity. Alongside, effects of LonP1 overexpression have been observed in the initial steps of pentose phosphate pathway, which is required for the synthesis of ribonucleotides from glucose and is a major source of nicotinamide adenine dinucleotide phosphate. Pentose pathway is important for redox balance and anabolism, and seems indeed to be promoted in SW620-pLonP1 cells. In SW620-pLonP1 cells, glucose could be redistributed to alternative pathways, such as the pentose pathway, to support growth and survival of these cells, which were characterized by reduced OXPHOS. In agreement with this observation, the pentose phosphate pathway was strongly decreased in SW480-pLonP1, where OXPHOS was almost unchanged. Despite the fact that SW480 and SW620 have the same genetic background, characterized by mutations in KRAS and TP53, overexpression of LonP1 has dramatically different consequences on mitochondrial function, bioenergetics, and in malignant transformation. While LonP1 does not influence OCR or mitochondrial functions in SW480 cells, its overexpression in SW620 cells determines a reduction of OCR and depolarization of mitochondrial membrane. We would expect that mitochondrial membrane depolarization resulted in increased ROS production, but this was not case. We rather observed a decrease in mitochondrial anion superoxide in SW620-pLonP1 when compared with SW620-pMSCV cells. The analysis of morphology and ultrastructure of mitochondria confirmed that overexpression of LonP1 has a stronger impact on SW620 than SW480 cells.

In conclusion, our findings demonstrate a role for LonP1 in the regulation of EMT *via* GSK-3β/β-ctn and modifications in cellular metabolism, suggesting that changes in LonP1 expression and EMT in CRC are likely not two independent, concurrent phenomena, but might be functionally linked.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the Ethical Committee of the province of Modena (Italy). The protocol was approved by the Ethical Committee of the province of Modena. All subjects gave written informed consent in accordance with the Declaration of Helsinki.

# AUTHOR CONTRIBUTIONS

LG, AC, and MP conceived and designed the experiments. LG, SB, DT, GC, AP, AG, SPe, and SPa conducted the experiments. LL, FM, and LR provided samples and performed immunohistochemical stainings. MN reported and organized data. LG, LL, AI, PP, AC, and MP wrote and revised the paper.

# ACKNOWLEDGMENTS

This work was supported by grant from "Associazione Angela Serra per la Ricerca sul Cancro" (Modena) to AC and by "Associazione Italiana Ricerca sul Cancro" (AIRC, Investigator Grant number 19786) to MP. We gratefully acknowledged the "Associazione Italiana Mogli Medici" in Modena for the kind donation of the Trans-Blot Turbo Blotting System to AC. Sara De Biasi is an International Society for Advancement in Cytometry (ISAC) Marylou Ingram Scholar.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at https://www.frontiersin.org/articles/10.3389/fonc.2018.00254/ full#supplementary-material.

Figure S1 | LonP1 modulates E-cadherin, N-cadherin, and β-ctn in I407 and RKO cells. (A) Representative Western blot analysis and relative protein level of LonP1, β-catenin (β-ctn), E-cadherin (E-cad), N-cadherin (N-cad) in I407 cells overexpressing LonP1 (namely pLonP1) and in control cells, i.e., cells stably transfected with empty vector (namely pMSCV). β-actin was used as loading control. Densitometries are reported in histograms, and data are reported as mean ± SD (*n* = 3). \**P* < 0.05 and \*\**P* < 0.01. (B) Representative phase contrast images of I407 cells overexpressing LonP1 (pLonP1) and control cells (pMSCV). Scale bar, 10 µm. (C) Representative Western blot analysis and relative protein level of LonP1, β-catenin (β-ctn), E-cadherin (E-cad), and N-cadherin (N-cad) in RKO cells overexpressing LonP1 (namely pLonP1) and in control cells, i.e., cells stably transfected with empty vector (namely pMSCV). β-actin was used as loading control. Densitometries are reported in histograms, and data are reported as mean ± SD (*n* = 3). \**P* < 0.05 and \*\**P* < 0.01.

Figure S2 | LonP1 slightly modulates mitochondrial activity in I407 and RKO colon cancer cells. (A) Representative traces and quantitative analysis of the oxygen consumption rate in indicated cells. Subsequent injections of oligomycin (Oligo), mitochondrial decoupler carbonyl cyanide-4-(trifluoromethoxy) phenylhydrazone (FCCP), complex I inhibitor rotenone (Rot) and complex III inhibitor (AntA) were performed as indicated. (B) Mitochondrial membrane potential quantification as assayed by tetramethyl rodhamine methyl esther

#### REFERENCES


(TMRM) in the presence or absence of CCCP in indicated cells. Data are expressed as percentage of increase in median fluorescence intensity (MFI) and represented the mean ± SD (*n* = 3); \**P* < 0.05. (C) Mitochondrial anion superoxide quantification as assayed by MitoSOX Red Mitochondrial Superoxide Indicator (mitoSOX) in the presence or absence of hydrogen peroxide in indicated cells. Data are expressed as percentage of increase in MFI and represented the mean ± SD (n = 3); \**P* < 0.05.


38. Oh BY, Kim SY, Lee YS, Hong HK, Kim TW, Kim SH, et al. Twist1-induced epithelial-mesenchymal transition according to microsatellite instability status in colon cancer cells. *Oncotarget* (2016) 7:57066–76. doi:10.18632/ oncotarget.10974

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

*Copyright © 2018 Gibellini, Losi, De Biasi, Nasi, Lo Tartaro, Pecorini, Patergnani, Pinton, De Gaetano, Carnevale, Pisciotta, Mariani, Roncucci, Iannone, Cossarizza and Pinti. 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.*

# Signaling Pathways Regulating Redox Balance in Cancer Metabolism

#### *Maria Chiara De Santis1 , Paolo Ettore Porporato1 , Miriam Martini1 \* and Andrea Morandi2*

*1Department of Molecular Biotechnology and Health Science, Molecular Biotechnology Center, University of Torino, Torino, Italy, 2Department of Experimental and Clinical Biomedical Sciences, University of Florence, Florence, Italy*

The interplay between rewiring tumor metabolism and oncogenic driver mutations is only beginning to be appreciated. Metabolic deregulation has been described for decades as a bystander effect of genomic aberrations. However, for the biology of malignant cells, metabolic reprogramming is essential to tackle a harsh environment, including nutrient deprivation, reactive oxygen species production, and oxygen withdrawal. Besides the well-investigated glycolytic metabolism, it is emerging that several other metabolic fluxes are relevant for tumorigenesis in supporting redox balance, most notably pentose phosphate pathway, folate, and mitochondrial metabolism. The relationship between metabolic rewiring and mutant genes is still unclear and, therefore, we will discuss how metabolic needs and oncogene mutations influence each other to satisfy cancer cells' demands. Mutations in oncogenes, i.e., PI3K/AKT/mTOR, RAS pathway, and MYC, and tumor suppressors, i.e., p53 and liver kinase B1, result in metabolic flexibility and may influence response to therapy. Since metabolic rewiring is shaped by oncogenic driver mutations, understanding how specific alterations in signaling pathways affect different metabolic fluxes will be instrumental for the development of novel targeted therapies. In the era of personalized medicine, the combination of driver mutations, metabolite levels, and tissue of origins will pave the way to innovative therapeutic interventions.

Keywords: metabolic reprogramming, one-carbon metabolism, OXPHOS, oncometabolites, pentose phosphate pathway

## INTRODUCTION: METABOLIC DEREGULATION AND TUMOR PROGRESSION

Despite being considered for decades as a bystander effect of the genomic alterations that characterize a cancer cell, it is now established that the deregulation of key metabolic hubs can be a tumorigenic driver and that several cellular signaling alterations converge toward metabolic

#### *Edited by:*

*Andrea Rasola, Università degli Studi di Padova, Italy*

#### *Reviewed by:*

*Marco Sciacovelli, University of Cambridge, United Kingdom Stefano Indraccolo, Istituto Oncologico Veneto (IRCCS), Italy*

> *\*Correspondence: Miriam Martini*

*miriam.martini@unito.it*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

*Received: 31 January 2018 Accepted: 06 April 2018 Published: 23 April 2018*

#### *Citation:*

*De Santis MC, Porporato PE, Martini M and Morandi A (2018) Signaling Pathways Regulating Redox Balance in Cancer Metabolism. Front. Oncol. 8:126. doi: 10.3389/fonc.2018.00126*

**43**

**Abbreviations:** 1-CM, one-carbon metabolism; ALDH, aldehyde dehydrogenase 1 family member; αKG, alpha-ketoglutarate; DRP1, dynamin-related protein; F6P, fructose-6-phosphate; FH, fumarate hydratase; G3P, glyceraldehyde-3-phosphate; G6P, glucose-6-phosphate; G6PDH, glucose-6-phosphate dehydrogenase; IDH, isocitrate dehydrogenase; MDM2, mouse double minute 2 homolog; MOMP, mitochondrial outer membrane permeabilization; MTHFD, methylenetetrahydrofolate dehydrogenase (NADP + dependent); NADPH, nicotinamide adenine dinucleotide phosphate; OMM, outer mitochondria membrane; PGAM1, phosphoglycerate mutase 1; PGC-1α, peroxisome proliferator-activated receptor gamma coactivator-1 alpha; PHGDH, phosphoglycerate dehydrogenase; SDH, succinate dehydrogenase; SSP, serine synthesis pathway; TCA, tricarboxcylic acid; TIGAR, TP53-induced glycolysis and apoptosis regulator; PPP, pentose phosphate pathway; ROS, reactive oxygen species.

alterations and the accumulation of specific intermediates endowed with oncogenic potential (1). The altered metabolic scenario of a cancer cell inevitably affects its cellular redox homeostasis. Particularly, cancer cells are characterized by an altered redox status and enhanced reactive oxygen species (ROS) that have been shown to drive and sustain cancer cells proliferation. Higher ROS levels are compensated by an increase in antioxidant mechanism that allows the cancer cell to survive in a pro-oxidant environment. However, such pro-oxidant condition favors DNA damage and genomic instability, events that concur to enhance the malignant traits of cancer cells, including metabolic reprogramming. Metabolic reprogramming and redox homeostasis are intimately interconnected and, therefore, such a vicious loop is further promoted by genetic lesions that activate proto-oncogenes or repress onco-suppressors, leading to cancer progression (2).

This review gathers the recent findings on the role of oncogenicdependent metabolic reprogramming that cancer cells undergo during the different stages of tumor progression. Importantly, the main metabolic pathways concurring to redox homeostasis have been reviewed with a focus on the correlation between oncogenic lesions and ROS-dependent metabolic reprogramming.

#### PENTOSE PHOSPHATE PATHWAY (PPP)

The PPP, also termed phosphogluconate pathway, diverges from the glycolytic pathway after the initial phosphorylation of glucose catalyzed by the glycolytic enzyme hexokinase. Glucose-6-phosphate (G6P) is the principal substrate of the G6P dehydrogenase (G6PDH), the rate-limiting enzyme of the PPP. The primary endpoint of the PPP is to provide (i) phospho-pentoses, for the nucleotides and nucleic acids synthesis and (ii) reducing equivalents that are used for both reductive biosynthesis reactions and for redox homeostasis. Indeed, the production of reduced nicotinamide adenine dinucleotide phosphate (NADPH) in the oxidative phase of the pathway is essential for redox cellular homeostasis; directly, by buffering enhanced reactive oxygen intermediates and indirectly for the regeneration of the oxidized form of the glutathione (GSH), a pivotal molecule in neutralizing intracellular ROS levels (3). The non-oxidative phase of the PPP is characterized by a series of reversible reactions catalyzed by transaldolase (TALDO) and transketolase (TKT) enzymes that recruit additional glycolytic intermediates, i.e., fructose-6 phosphate (F6P) and glyceraldehyde-3-phosphate (G3P), to produce phospho-pentoses (4).

Despite leading to glucose oxidation, the main role of the PPP is anabolic rather than catabolic. Depending on the cellular requirements and response to exogenous stimuli, the PPP and the glycolysis cooperate to provide the needed metabolites. For instance, rapidly proliferating cells will boost the PPP to meet the requirements for pentoses to support nucleotide biosynthesis, both from G6P (oxidative phase) and from F6P and G3P (non-oxidative phase). While this condition is representative of a rapidly dividing cancer cell, a cell that has to maintain the redox cellular balance will sustain the oxidative branch for NADPH production and re-direct the non-oxidative branch toward F6P synthesis from the phospho-pentoses, which are then regenerating G6P to replenish the oxidative branch. ROS accumulation and subsequent oxidative stress result from the unbalance between ROS generation (e.g., from electron transport chain), and antioxidant mechanism (e.g., GSH, thioredoxin, and catalase) requiring NADPH to function as ROS scavengers. Therefore, not surprisingly, PPP deregulation has been linked to the pathogenesis of several diseases, such as G6PDH-deficiency. G6PDH-deficiency, and subsequent reduced activity, impairs the ability of erythrocytes to generate NADPH, hence exposing the cells' phospholipid bilayer to the detrimental effect of ROS, leading to hemolytic anemia. G6PDH-deficiency is not associated with acquired susceptibility to particular diseases and preventing oxidative stress-inducing situations (i.e*.,* certain drugs and food) leaves G6PDH-deficiency bearing individuals asymptomatic. Interestingly, G6PDH-deficiency is protective against malaria, heart, and cerebrovascular disease (5).

However, since the focus of this review is on cancer, we will discuss how PPP deregulation is affected by oncogenicdependent metabolic reprogramming and how this impacts on tumor progression and therapy resistance. Interestingly, G6PDH-deficiency has been reported to reduce cancer susceptibility and incidence. A retrospective observational study in ~4,000 patients that underwent colonoscopy with a 10 years follow-up from Sardinia region (Italy), where G6PDH-deficiency prevalence ranges between 12 and 24% and is often caused by the G6PDHC563T variant, shows that G6PDH-deficiency is associated with reduced colorectal cancer risk (5, 6). The potential explanation of such counterintuitive epidemiological data may be explained by the NADPH and PPP dependence that sustain lipid anabolism in cancer cells. Additionally, NADPH deficiency due to G6PDH alterations may alter the redox balance, hence reducing intracellular ROS levels, which play an important role in cancer initiation by increasing the DNA mutation rate and the synthesis of proinflammatory cytokines (7).

Glucose-6-phosphate dehydrogenase expression and activity have been proposed to be importantly influenced by genetic alterations that occur on oncogenes and/or oncosuppressor that ultimately lead to a pro-mitogenic signaling pathways activation of the cancer cells (**Figure 1**). Indeed, while the reprogramming of certain metabolic hubs seems to correlate with particular tumor phenotype and be associated with certain type of therapeutic option, the increased demand of NADPH, used for either antioxidant response and/or cellular anabolism than pentoses, is a prerequisite of almost any cancer (4). G6PDH is regulated by a plethora of extracellular stimuli, e.g., growth factors, that impacts on its expression and activity *via* the MAPK and PI3K signaling pathways. Since these signaling pathways are often hyper-activated in cancer due to oncogenes activation (e.g., K-RAS, MYC, and growth factor receptors), or oncosuppressors inactivation (e.g., p53 and PTEN loss-of-function and inactivation), any alterations that impact on these players may lead to G6PDH enhanced expression and activity (8, 9). Additionally, some of the aforementioned cancer-inducing genes can also regulate G6PDH function independently of the signaling cascade. For instance, loss of p53, which directly controls and inhibits glucose transporters expression (10), leads to enhanced glucose uptake that can be diverted into the PPP pathway. Similarly, glucose diversion

(ALDH1L) mediates the opposite reaction with concurrent generation of NADPH, similarly methylenetetrahydrofolate dehydrogenase (MTHDF) transform meTHF into 10-formyltetrahydrofolate (F-THF) with concurrent generation of nicotinamide adenine dinucleotide phosphate. The main chemotherapy agents are represented in blue. Dehydroepiandrosterone (DHEA), epigallocatechin gallate (EGCG), amino-5-(4-chlorophenyl)-7-(t-butyl)pyrazolo[3,4-d]pyrimidine (PP2), 6-aminonicotinamide (6-AN), and 5-fluorouracil (5-FU).

into the PPP can also be a consequence of the downregulation of TP53-induced glycolysis and apoptosis regulator (TIGAR), a p53-dependent gene known to inhibit the glycolytic enzyme phosphoglycerate mutase 1 (PGAM1) that leads to the increase of the oxidative PPP by reducing the amount of 3-phosphoglycerate (3PG), which has been reported to inhibit the PPP enzyme 6-phosphogluconate dehydrogenase (6PGDH) (11). Finally, p53-mediated G6PDH inhibition can also be mediated by direct protein-to-protein contact (3). DNA damage, telomeric instability, or oxidative stress can activate the ataxia telangiectasia-mutated (ATM) kinase, which in turn activates p53. Interestingly, ATM promotes the Hsp27 phosphorylation and binding to G6PDH, stimulating its activity and, therefore, increasing the PPP flux and the subsequent NADPH and phospho-pentoses production (12). However, others have reported that loss of p53 enhances NADPH production, therefore, by generating a negative feedback on the PPP. These reported contradictory results may be related to the different role that p53 can exert on either cell cycle or apoptosis (3). Finally, it has been recently demonstrated that polo-like kinase 1 (PLK1), a key player in cell mitosis, is able to directly phosphorylate and activate G6PDH (13).

Although G6PDH is the major rate-limiting step of the PPP, the amount of phosphorylated glucose trapped into the cells and the expression of enzymes regulating the glycolytic rate can also affect the PPP flux. Indeed, higher levels of G6P are required to sustain oxidative and non-oxidative branches of the PPP and this can be achieved by (i) enhancing the amount of glucose that is phosphorylated by the glycolytic-rate limiting enzyme hexokinase 2 (HK2), the main isoform expressed by cancer cells, and (ii) by engulfing the glycolytic pathway and subsequent enhancing the accumulation of glycolytic intermediates. HK2 enhanced expression, known to be regulated by oncogenic RAS activation, is essential to sustain the non-oxidative phase of the PPP in lung cancer (14). Additionally, pyruvate kinase (PK)-M2, an isoform of the PK that converts phosphoenolpyruvate (PEP) into pyruvate and that is expressed by many cancers, can be controlled by posttranslational modifications (e.g., oxidation, phosphorylation, and acetylation) that impair the PKM2 tetrameric-mediated metabolic function favoring the PKM2 dimer formation (15, 16). Particularly, an increase in the PKM2 dimer content induces an accumulation of the glycolytic intermediates, including that of G6P, that results in increased metabolic flux into the PPP and enhanced NADPH production (17–20).

6-Phosphogluconate dehydrogenase, the third enzyme involved in PPP, that catalyzes the oxidative decarboxylation of 6-phosphogluconate to ribulose 5-phosphate and CO2, with concomitant reduction of NADP to NADPH, has also been reported to be upregulated in many solid cancers and has been often correlated to G6PDH (4).

Transketolase and TALDO are the two key enzymes in the non-oxidative branch and divert glycolytic intermediates (e.g., F6P and G3P) into the PPP to fuel ribonucleotides biosynthesis, essential for fast proliferating cancer cells (21). Their expression levels have been found deregulated in cancer cells. Particularly, oncogenes activation that leads to hyper-proliferation may also have an impact on the non-oxidative branch of the PPP, favoring ribonucleotides biosynthesis. Indeed, the use of isogenic colorectal cancer cell lines that express either the mutant or the wild-type form of K-RAS or B-RAF shows that the enzymes of the non-oxidative branch of the PPP (ribose-5-phosphate isomerase and TKT) are upregulated in the K-RAS-mutant cell lines (20). These results are in line with a previous report showing that K-RAS mutations in pancreatic cancers enhance the nonoxidative branch of the PPP (22).

Due to the important role in maintaining the redox homeostasis, PPP is also importantly regulated by nuclear factor erythroid 2-related factor 2 (NRF2), a transcription factor that has an essential role in combating enhanced ROS levels and controlling ROS detoxification and homeostasis. NRF2, which is usually bound in its inactive form to the cytosolic Kelch-like ECH-associated protein 1 (KEAP1), is able to translocate into the nucleus upon NRF2-KEAP1 destabilization, oxidative stress, or protein succination driven by fumarate accumulation (23, 24). Once into the nucleus, NRF2 activates antioxidant response genes and this activation have been reported to play a major role in protecting cancer cells from the oxidative stress induced by antitumoral therapies. Activation of NRF2 or destabilization of NRF2–KEAP1 interaction caused by genetic modifications of NRF2 and KEAP1 has been reported in several cancer types, including those of liver, esophagus, intestine, lung, and breast (25, 26). Moreover, the increased expression of NRF2-dependent genes correlates with poor prognosis (27–29). NRF2 is tightly linked to metabolic reprogramming. In particular, NRF2 targeting impairs the activity and the expression of the enzymes involved in the PPP (e.g., G6PDH, TKT, and 6PGDH), a process that has been shown to be mediated by microRNAs (30). Recently, it has been shown that NRF2 enhances expression of the PPP enzymes. The consequent increase of the PPP flux leads to acquired proliferative advantage in the presence of constitutive activation of PI3K-AKT signaling pathway due to *PTEN* deletion (31). Oncogenic activation of K-RAS and B-RAF and overexpression of MYC, together with enhanced activation of the PI3K/AKT pathway induce NRF2 nuclear translocation, further reinforcing the link between oncogenes activation and NRF2 mediated PPP induction and redox homeostasis alterations.

An important driver of cancer cell growth and survival is the activation of the PI3K/AKT pathway that can occur as a consequence of external stimuli, e.g., growth factor receptor activation or as a consequence of a genetic lesion, e.g., *via* PTEN loss-of-function that ultimately leads to PI3K/AKT-mediated mTOR activation. It has been extensively reported that mTOR can control cell metabolism, including the glycolytic pathway and PPP (32). mTOR kinase can associate with different subunits leading to two signaling complexes, mTORC1 and mTORC2. Particularly, mTORC1 regulates cell proliferation and anabolism.

Indeed, it has been shown that mTORC1 activation sustains the metabolic flux through both glycolysis and the oxidative arm of PPP. Duvel and coworkers demonstrated that mTORC1 activation promotes G6PDH expression, a process that is in part mediated by SREBP1 (33). Additionally, it has been recently reported that mTOR-dependent G6PDH expression can be controlled by androgen receptor signaling in prostate cancer models (34). In the Duvel et al. manuscript it was also reported that mTOR activation induces HIF-1α expression and stabilization, leading to the transcriptional activation of a plethora of metabolic HIF-dependent genes in a hypoxia-independent mechanism (33). HIF-1α stabilization have been also reported to control the expression of TKT in pancreatic cancer cells, hence impacting on the non-oxidative branch of the PPP (35) and subsequent response and resistance to gemcitabine. However, also hypoxia has been reported to impact on PPP flux. Indeed, colon cancer cells subjected to 1% hypoxia increase their intracellular levels of ribose-phosphates and gluconic-acid, terminal and intermediate compounds of the oxidative branch of PPP, respectively (36).

# ONE-CARBON METABOLISM (1-CM): THE SERINE SYNTHESIS PATHWAY (SSP)

One-carbon (1-C) metabolism is responsible for the transfer of 1-C unit through folate intermediates, coupling the folate and the methionine cycle. This pathway is required for nucleic acid synthesis (purine and thymidine), amino acids homeostasis (methionine, serine, and glycine), antioxidant defense (NADPH production) and epigenetic maintenance (homocysteine re-methylation) (37–39). In particular, the methionine cycle produces the substrate for the S-adenosyl methionine (SAM) dependent methyl transferases. These enzymes are responsible for the addition of methyl groups to proteins, lipids, secondary metabolites, and nucleic acids and are, therefore, essential for epigenetic modifications.

In order to sustain the NADPH-mediated antioxidant defense, 1-CM supplements the major source of intracellular NADPH, which is produced by the oxidative branch of the PPP and by the malic enzyme. In the 1-C cycle, there are two steps that can lead to NADPH production: one catalyzed by the methyl dehydrogenases, mitochondrial 2-like (MTHFD2L) and cytosol MTHFD1, and the other catalyzed by the formyl dehydrogenases, cytosolic aldehyde dehydrogenase 1 family member L1 (ALDH1L1) and mitochondrial aldehyde dehydrogenase 1 family member L2 (ALDH1L2).

1-C units are mainly derived from serine, formate, and histidine, which are directly used in the cytosolic folate cycle or by glycine, sarcosine, and dimethylglycine, which are converted into folate and secreted into the cytosol (40, 41). Glycine has a prominent role because it can contribute to both the folate cycle and to serine production. In particular, the folate cycle is completed in the cytosol by serine hydroxymethyltransferase 2 (SHMT2), which catalyzes the concomitant conversion of l-serine to glycine and tetrahydrofolate to 5,10-methyleneTHF. Hence, glycine represents a precursor of glutathione and purines, required for antioxidant defense and proliferation, respectively. If necessary, glycine can be converted into serine with the concomitant production of 5,10-methylene-THF, although some of the specificity of serine-related metabolic functions are lost and this results, for instance, in decreased purine synthesis and reduced cell proliferation (42, 43).

The main donor of 1-C to the folate cycle is the non-essential amino acid serine that can be synthetized *de novo* by the cell (41), a process that requires three MYC-regulated enzymes, PHGDH, PSAT1, and PSPH. The backbone of serine is derived from the glycolytic or gluconeogenic pathway, with the production of the intermediate 3PG. The main transcriptional activator of the three SSP enzymes is ATF4, a cAMP-response element-mediated transcription factor that mediates oxidative stress response (*via* NRF2) (44), serine starvation (*via* mouse double minute 2 homolog MDM2) (45), and histone methylation (41).

At physiological levels, serine can act as an allosteric activator of the metabolic activity of PKM2, which catalyzes the last step of glycolysis, preventing the redirection of 3PG into serine synthesis and promoting glycolysis. Similarly, AKT-mediated phosphorylation of MDM2, a negative regulator of p53, on Ser166 induces the association of MDM2 to PKM2, in order to promote PKM2 activity. In condition of oxidative stress, serine deprivation, and the consequent decrease in the allosteric activation of PKM2, the non-phosphorylated form of MDM2 is recruited to the chromatin with ATF4, which activates a transcriptional program involved in amino acid metabolism and redox homeostasis (45–48). In particular, during oxidative stress, ATF4 has been reported to be transcriptionally activated by NRF2, which induces the transcription of SSP-related genes and regulates the antioxidant response (44). While during serine starvation, the inhibition of PKM2 metabolic activity leads to the accumulation of glycolytic intermediates that results in *de novo* serine synthesis activation, if G3P is used as a substrate for the PHGDH, or into oxidative PPP activation and/or if G6P is used as a substrate for G6PDH, as described above (27).

mTORC1 signaling, through ATF4, activates the transcription of MTHFD2 in both normal and cancer cells, increasing *de novo* purine synthesis necessary for nucleic acid production (48). PKM2-expressing cells can maintain mTORC1 activity and proliferate in serine-depleted medium (49).

It has been reported that highly proliferative cells require an exogenous amount of serine supply for their optimal growth to adjunct the *de novo* synthesis (50). Indeed, additional serine sources are derived from diet intake, from breakdown of intracellular proteins, and from the conversion of glycine. In serine starve condition, P53 null cells have an impaired proliferation rate and tumor growth is reduced in mice fed with serine-free diet (50).

Due to its role in maintaining amino acid and redox homeostasis and epigenetic regulation, it is not surprising that the SSP has been found altered in cancer. Since serine can become a limiting factor, cancer cells can adopt two strategies to bypass this issue: they can increase *de novo* serine synthesis to become dependent on some alternative amino acids, such as glutamine, or can promote exogenous serine intake (41). For instance, PHGDH is commonly altered in cancer cells, either upon gene amplification (51, 52), or activation of transcription factors promoting expression of SSP enzymes, e.g., MYC and NFR2 (44). Among the signaling pathways that promote the 1-C metabolism, it has been reported that HER2 amplification (53) increases the expression of PHGDH (**Figure 1**). The PHGDH promoter is positively regulated by specificity protein 1 (SP1) and nuclear transcription factor Y (NFY), two transcription factors that are often upregulated in cancer (54). PHGDH is overexpressed and associated with poor prognosis in triple negative breast cancer and melanoma (44, 52, 55). While genomic alterations are typical of PHGDH, the high expression of the folate enzyme is due to aberrant transcriptional regulation. In non-small cell lung cancer (NSCLC), ATF4 overexpression can be caused by KEAP1-mediated activation of NRF2 (44) or activation of the PI3K/mTOR signaling pathway (48, 56, 57).

Under hypoxia, HIF-1α and MYC can mediate SHMT2 increased levels that correlate with poor prognosis. Among the enzymes that catalyze the reactions coupled with the production of NADPH, THF dehydrogenases (cytosolic ALDH1L1 and mitochondrial ALDH1L2) and MTHF-dehydrogenases (mitochondria MTHFD2L and cytosolic MTHFD1) are involved in cancer. ALDH1L1 is usually underexpressed in cancer cells (57), as its overexpression would deplete the cytosolic 10-formyl-THF used for the synthesis of nucleotides. Conversely, ALDH1L2 is overexpressed to control mitochondrial redox homeostasis under hypoxia and to support melanoma cells metastatization (56). K-RAS activating mutation is associated with the high expression of the folate metabolism enzyme, MTHFD2 (47), and this leads to increased nucleotide synthesis, increased ATP and NADH production, and finally to enhanced mitochondrial NADPH production which is essential for struggling ROS increased levels. Conversely, during nutrient deprivation or hypoxia, when intracellular levels of ATP decline, 5′ AMP activated kinase (AMPK) activation represses the expression of MTHFD2 (50, 58).

# MITOCHONDRIAL METABOLISM

Reactive oxygen species mtROS are generated by mitochondria as a natural by-product of electron transport chain activity. Recently, it has been reported that mitochondria ROS (mtROS) can activate tumorigenic signaling and metabolic reprogramming. Increased production of ROS has long been observed to be a hallmark of many tumors and cancer cell lines. Consequently, this oncogenic signaling increases the expression of antioxidant proteins to balance the high production of ROS to maintain redox homeostasis.

Indeed, several aspects of mitochondrial biology, including biogenesis and turnover, fission and fusion, and mitochondriacontrolled metabolic signaling are controlling cellular transformation (59, 60).

Mitochondria are signaling hubs and bioenergetics organelles, which play an important role in cellular adaptation to environmental changes, directly responding to nutrient availability. For instance, during stress condition, such as extracellular acidosis or hypoxia (60, 61), mitochondria tune their metabolism to support fatty acid synthesis by upregulating the reductive carboxylation of α-ketoglutarate *via* citrate generation to support both production of lipid membranes and production of intermediates for protein acetylation (13).

An element controlling mitochondrial function is the mitochondrial mass, which can vary greatly in cell, depending on the microenvironmental context and oxidative fuels availability of certain nutrients. The mitochondrial mass is regulated by biogenesis and turnover, two different pathways that can act as positive and negative regulators of tumorigenesis (59). It is generally accepted that MYC is one of the key activator of mitochondrial biogenesis in cancer, and opposed effects are exerted by the activation of HIF-1α signaling pathway (62), as well as FOXO3a (62–64). In physiological conditions, MYC couples mitochondrial biogenesis with cell-cycle progression, whereas once deregulated in cancer it stimulates mitochondrial metabolism to support rapid cell growth (63). The major impact of MYC on mitochondria function depends on the direct regulation of the transcriptional coactivator peroxisome proliferator-activated receptor gamma coactivator-1 alpha (PPARGC1A, best known as PGC-1α) (58), that is responsible for the enhanced metabolic plasticity of aggressive cancer cells. In particular, PGC-1α levels have a dichotomous effect on tumors, with an upregulation that has been associated with pro- or anti-tumorigenic effects on the basis of tumor types and experimental conditions (65–67).

The discovery of the tumor suppressor liver kinase B1 (LKB1) and its major downstream effectors PGC-1α and AMPK established a central metabolic hub at the crossroad between energy regulation and cancer development (68). It has been already reported that LKB1 plays an important role in reducing intracellular ROS in response to oxidative stress (69, 70). Consequently, LKB1 loss increases oxidative DNA damage and mutations induced by the accumulation of ROS. In NSCLC, the redox imbalance caused by LKB1 inactivation modulates tumor plasticity and promotes tumor progression *via* metabolic adaptation (71).

In addition to aerobic glycolysis, cancer cells often rely on elevated glutaminolysis, supporting mitochondrial metabolism for cancer growth. In order to fuel the tricarboxcylic acid (TCA) cycle, glutamine is first converted to glutamate by glutaminase and then to α-ketoglutarate by glutamate dehydrogenase or aminotransferases. Therefore, glutamine supports the high proliferating rate of cancer cells by acting as substrate for the TCA cycle to produce building blocks, including ATP, lipids, nucleotides, and proteins. MYC oncogene transcriptionally promotes glutaminolysis and the use of glutamine as a bioenergetic substrate (72). The final effect of MYC-dependent glutaminolysis is a profound mitochondrial metabolism reprogramming due to glutamine catabolism dependency and TCA cycle anaplerosis. Noteworthy, reductive carboxylation under hypoxia has been shown to promote generation of 2-hydroxyglutarate (2HG) even in absence of isocitrate dehydrogenase (IDH) mutations (73, 74).

Recently, it has been reported that an association with the oncometabolite 2HG accumulations and MYC pathway activation in breast cancer patients (75). This analysis identified a subtype of tumors characterized by higher levels of 2HG that associate with poor prognosis and with a distinct DNA methylation pattern. These tumors tend to overexpress glutaminase, suggesting a functional relationship between MYC activation and glutamine dependence in breast cancer (75, 76).

2-Hydroxyglutarate is a TCA byproduct that can be generated in certain conditions and has been termed oncometabolite, since its production can drive oncogenesis (77). To date, various germline and somatic mutations in mitochondrial enzymes led to oncometabolites accumulation. For instance, succinate dehydrogenase (SDH), fumarate hydratase, and IDH (78) once mutated lead to cellular transformation and oncogenesis (79) *via* augmented levels of fumarate, succinate, and 2HG, respectively (79). Those molecules present various activity, altering epigenetic pattern, enzymatic activity, or the aforementioned protein succination induced by fumarate. Coherently with the oncogenic role of such metabolite, TRAP1 expression has been shown to induce cellular transformation by inhibiting SDH, thus promoting succinate build-up (80, 81).

Recently, Morita et al. reported a key role for mTORC1 signaling, stimulating mitochondrial biogenesis and activity, bolstering ATP production capacity (82). This effect is mediated by the activity of 4E-BP proteins that mediate mTORC1-driven translation of mitochondria-related mRNAs, mitochondrial respiration and biogenesis, and ATP production. These data reveal a feed-forward mechanism by which translation impacts mitochondrial function to maintain cellular energy homeostasis (82). Consistently, in muose Embryonic Stem cells has been shown that tumorigenicity and teratoma formation was bound to increased mitochondrial metabolism linked to higher mTOR activity (83).

Mitochondria are extremely dynamic structures and the balance of fission and fusion is responsible for their morphology (84). The mitochondrial division is controlled by dynaminrelated protein 1 (Drp1), a GTPase that is recruited to the outer mitochondria membrane. Drp1 mitochondrial translocation is regulated by different kinases that are activated by specific cell-cycle and stress conditions. Several studies described that cancer cells have an imbalance fission-fusion state, characterized by high fission and decreased fusion activities, causing a fragmentation of mitochondrial network (85). Disruption of mitochondrial dynamics are one of the key features of K-RASinduced cellular transformation, by stimulating mitochondrial fragmentation *via* ERK1/2-mediated phosphorylation of Drp1 (86). In addition, the mitochondrial remodeling caused by oncogenic K-RAS directly increases ROS generation and impacts the mitochondrial function itself (87). Similarly to what has been reported for the MYC oncogenic activation, the mitochondrial network remodeling caused by oncogenic K-RAS generates general tumorigenic stimuli that promote transformation. However, further studies are needed to understand the effects of oncogenic signaling pathways on mitochondrial dynamics. K-RAS-mediated metabolic remodeling drives also an augmented glutamine metabolism and malic enzyme activity, acting as an important source of NADPH (**Figure 2**) (88). Mitochondrial remodeling is also altered by ATPase inhibitory factor 1 (ATPIF), a factor exerting pro-tumorigenic function by preventing cristae remodeling and mitochondrial depolarization by inhibiting ATPase activity of complex V (89).

Redox homeostasis is an important process that can fuel anchorage-independent growth. In this stressful condition, mitochondrial IDH2 supports NADPH pool to decrease mtROS accumulation (41). Consequently, the limitation of excessive generation of mtROS is a converging pattern that can support metastasis formation, although a moderate degree of mtROS directly promotes metastasis formation (56, 90). Accordingly, inhibiting mitophagy by BNip3, hence degradation of dysfunctional mitochondria, promotes mtROS generation sufficient to induce metastasis in breast cancer (91).

Another important player in the regulation of mitochondrial metabolism and intracellular redox homeostasis is p53, a key regulator of several biological processes, including energy homeostasis, apoptosis, and cell-cycle arrest, through transcription-dependent and -independent mechanisms (92). To regulate energy metabolism, p53 represses glycolysis, by regulating the

expression of TIGAR, PGAM1, and PDK2 (pyruvate dehydrogenase kinase 2), and promotes oxidative phosphorylation process.

Beside on its nuclear role for gene expression, a transcriptionindependent role of p53 in inducing mitochondrial apoptosis has been suggested (93). The mitochondrial regulation of p53 is critical for the control of the cellular switch between survival and apopotosis. In fact, several studies have shown that after DNA damage, p53 is translocated to mitochondrial outer membrane, causing mitochondrial outer membrane permeabilization (MOMP), caspase-3 activation, and cytochrome c release (94).

Altogether these studies demonstrate that mitochondria are crucial players of tumorigenesis, given that this process requires the ability to adapt to cellular and environmental alterations. Therefore, a deep understanding of molecular mechanisms regulating the function of mitochondria will be critical for the next generation of anti-cancer agents.

#### TARGETING ONCOGENE-DEPENDENT METABOLISM: THERAPEUTIC OPPORTUNITIES

One of the biggest challenges in cancer management is to ensure a tailored therapeutic approach to the patients that will lead to the administration to the right drug (as either monotherapy or combination), at the right patient in the right moment. Target therapies have expanded the portfolio of therapeutic opportunities and are often used either alone or as adjuvant therapies with chemotherapeutic agents. Cancer patients are usually matched to targeted therapies depending on driver mutations profiled in their tumors. However, the metabolic landscape of the tumor profoundly affects the response to therapy (95) and should be considered for optimal therapeutic intervention. Additionally, independently of the oncogenic drivers that sustain cancer growth and progression, understanding the metabolic network and the requirements of a given cancer, will offer a series of additional potential targets that could be exploited therapeutically.

For instance, since the PPP plays a key role in ribonucleotides biosynthesis, necessary for duplicating tumor cells, and redox homeostasis, essential to handle the oxidative stress induced by anti-tumoral therapies, targeting key step of the PPP has been proposed as a potential therapeutic opportunity. However, few inhibitors targeting PPP enzymes are available and no clinical trials have been designed to investigate PPP targeting in solid cancers. However, in the preclinical setting PPP targeting has been effective. Due to the rate-limiting role of the G6PDH enzyme, its targeting has been proposed and a series of compounds able to inhibit its activity *in vitro* and/or *in vivo* have been reported. Among them, the best characterized is the adrenal hormone dehydroepiandrosterone (DHEA), shown to be effective in inhibiting proliferation and survival of different cancer models (96, 97). To date, 60 trials are (or have been) investigating the application of DHEA in different cancer types (clinicaltrials.gov). However, the majority of the studies have evaluated DHEA for its hormonal effect rather than for its ability to target PPP. In the concluded trials, DHEA is overall well tolerated and further studies should be designed to draw any conclusion on its anti-metabolic effect; accordingly, patients should be subgrouped based on their metabolic dependency, with a particular focus on G6PDH expression. Other compounds have been reported to target G6PDH, such as epigallocatechin gallate (78), chelerythrine (initially a PKC inhibitor), Amino-5-(4-chlorophenyl)-7-(t-butyl) pyrazolo[3,4-d]pyrimidine (PP2) (98), and the Bay 11–7082 (99) but these compounds will require additional characterization in relevant cancer models prior to enter into the clinical setting. Interestingly, other compounds have been investigated in high-throughput screening leading to additional promising G6PDH inhibitors (100). The PPP can also be blocked by the 6-aminonicotinamide (6-AN) that promotes the biosynthesis of the 6-AN adenine dinucleotide phosphate which is able to inhibit the 6PGDH enzyme, hence reducing the NADPH production (4). In summary, targeting the PPP pathway may be a successful approach because, in addition to the reduction of the ribonucleotide biosynthesis that ultimately impacts on cell division, it could potentiate the effect of other therapies by reducing the resistance to the oxidative stress induced by anti-cancer compounds.

Importantly, for an emerging group of therapies, Mayers and colleagues have proposed that the metabolism of a target tumor may be a key determinant of its response and may be determined by the tumor's tissue of origin and not exclusively by their mutation status [see the Perspective by Vousden and Yang (101)]. Indeed, branched-chain amino acids (BCAAs) destiny has been traced in lung and pancreatic cancer mouse models driven by the same mutations. Lung tumors showed increased BCAA nitrogen uptake and utilization for the biosynthesis of amino acid and nucleotides, while pancreatic tumors displayed decreased uptake of free BCAAs (102). These differences demonstrate that the tissue of origin shapes tumor metabolism and should be taken into account to ensure the optimal therapeutic approach.

It is established that p53 loss is a frequent aberration in many cancer cells and that one of the pro-tumoral effects of such genetic lesion is the increased serine dependency, indicating that limiting cellular serine availability may be a therapeutic option. Tumors with amplified SSP genes are only partially dependent on exogenous serine depletion and a potential therapeutic approach for this subset of cancer may come from impairing the *de novo* serine synthesis. For example, PHGDH inhibitors (21, 51, 103) could be useful in patients with PHGDH-amplified tumors (21, 51). Indeed, chemotherapy based on the inhibition of the folate metabolism and thymidylate synthesis has been the pioneer anti-cancer drugs that entered into the clinical management of cancer patients (104). For instance, methotrexate and 5-fluorouracil are still used in leukemias treatment, while pemetrexed, which acts by mimicking the folic acid and inhibiting the purines and pyrimidines biosynthesis, is administered to patients with mesothelioma and lung cancer (105, 106). Combination of a serine-free diet with metformin, an anti-diabetic biguanide that has been reported to target complex I of the electron transport chain among other targets, shows a synergistic antineoplastic effect (107). Side effects due to the inhibition of 1-CM in non-transformed cells mainly affect proliferating tissues, such as the intestinal epithelium and bone marrow, resulting in gastric damage, anemia, and immune deficiency. Circulating and intratumoral levels of serine and/or the expression of genes of the folate metabolism could be used as potential predictive biomarkers to identify subsets of cancers more likely to respond to anti-folate therapy (47, 108).

Mitochondria are gatekeepers of cell death and could, therefore, be considered as valuable targets for novel therapeutic avenues. The characteristic redox state of malignant cells, due to the altered electrochemical gradient across the inner mitochondrial membrane, renders mitochondria optimal target as shown by Leanza and coworkers (109). Metformin, as described above, has been recently proposed to be repurposed for cancer treatment in light of its activity as mitochondrial complex I inhibitor, promoting cancer specific cell death in a plethora of conditions, most notably in cancer stem cells (110, 111). In cancer tissues, the maintenance of mitochondrial structural integrity is essential to produce energy and overcome cellular stress, including nutrient deprivation and chemotherapy (112). Coherently, major efforts are ongoing for the development of novel molecules targeting directly targeting mitochondria in cancer (i.e., mitoCANS) (113).

# CONCLUDING REMARKS

Genomic aberrations play a major role in tumor development and targeting oncogenic drivers have been successful in many types of cancers. However, the advantages of the oncogenic activation in terms of enhancing proliferation and survival capacity, metastatic abilities and resistance to therapies have a profound effect on the metabolic network of a cancer. It is conceivable that, to prolong the efficacy of the current antitumoral therapeutic regimens or to combat the development of therapy resistance, a potential successful approach could be the targeting of the "engine" of the tumor growth, i.e., targeting its driving metabolism. Indeed, independently of the drivers that promote tumor progression and/or therapy resistance, metabolic reprogramming will occur, a process that has been proposed to converge according to the tissue of origin more than of the oncogenic drivers (114). Here, we have listed a series of metabolic pathways that are controlled by key oncogenes and oncosuppressors that are often deregulated in cancer and could be potentially targeted. The concomitant targeting of the drivers together with the engine (that sustain cancer growth and progression) will render the signaling and metabolic reprogramming more difficult to occur and could represent a valid therapeutic approach. Further preclinical investigations on synthetic lethality approach in different tumor types and on the potential side effects that such stringent combinatory approach could have on normal cells are required and, if successful, will represent a major contribution toward the introduction of the personalized therapy approach in cancer patients' management.

# AUTHOR CONTRIBUTIONS

MS, PP, MM, and AM selected the references and cowrote the text.

## REFERENCES


# FUNDING

PP is supported by Italian Ministry for University and Research (MIUR, Rita Levi-Montalcini program for young researchers, 2014). MM is funded by Fondazione Umberto Veronesi. AM is funded by Fondazione Umberto Veronesi and Associazione Italiana Ricerca sul Cancro and Fondazione Cassa di Risparmio di Firenze (grant 19515).


transport protein. *Metab Eng* (2017) 43(Pt B):198–207. doi:10.1016/j.ymben. 2016.11.004


114. Gaude E, Frezza C. Tissue-specific and convergent metabolic transformation of cancer correlates with metastatic potential and patient survival. *Nat Commun* (2016) 7:13041. doi:10.1038/ncomms13041

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

*Copyright © 2018 De Santis, Porporato, Martini and Morandi. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# The Role of Mitochondrial H**+**-ATP Synthase in Cancer

#### *Pau B. Esparza-Moltó and José M. Cuezva\**

*Departamento de Biología Molecular, Centro de Biología Molecular Severo Ochoa (CSIC-UAM), Centro de Investigación Biomédica en Red de Enfermedades Raras CIBERER-ISCIII, Instituto de Investigación Hospital 12 de Octubre (i+12), Universidad Autónoma de Madrid, Madrid, Spain*

Cancer cells reprogram energy metabolism by boosting aerobic glycolysis as a main pathway for the provision of metabolic energy and of precursors for anabolic purposes. Accordingly, the relative expression of the catalytic subunit of the mitochondrial H+-ATP synthase—the core hub of oxidative phosphorylation—is downregulated in human carcinomas when compared with its expression in normal tissues. Moreover, some prevalent carcinomas also upregulate the ATPase inhibitory factor 1 (IF1), which is the physiological inhibitor of the H+-ATP synthase. IF1 overexpression, both in cells in culture and in tissue-specific mouse models, is sufficient to reprogram energy metabolism to an enhanced glycolysis by limiting ATP production by the H+-ATP synthase. Furthermore, the IF1-mediated inhibition of the H+-ATP synthase promotes the production of mitochondrial ROS (mtROS). mtROS modulate signaling pathways favoring cellular proliferation and invasion, the activation of antioxidant defenses, resistance to cell death, and modulation of the tissue immune response, favoring the acquisition of several cancer traits. Consistently, IF1 expression is an independent marker of cancer prognosis. By contrast, inhibition of the H+-ATP synthase by α-ketoglutarate and the oncometabolite 2-hydroxyglutarate, reduces mTOR signaling, suppresses cancer cell growth, and contributes to lifespan extension in several model organisms. Hence, the H+-ATP synthase appears as a conserved hub in mitochondria-to-nucleus signaling controlling cell fate. Unraveling the molecular mechanisms responsible for IF1 upregulation in cancer and the signaling cascades that are modulated by the H+-ATP synthase are of utmost interest to decipher the metabolic and redox circuits contributing to cancer origin and progression.

#### *Edited by:*

*Andrea Rasola, Università degli Studi di Padova, Italy*

#### *Reviewed by:*

*Giovanna Lippe, University of Udine, Italy Valentina Giorgio, Institute of Neuroscience (CNR), Italy*

#### *\*Correspondence:*

*José M. Cuezva jmcuezva@cbm.csic.es*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

*Received: 15 January 2018 Accepted: 20 February 2018 Published: 07 March 2018*

#### *Citation:*

*Esparza-Moltó PB and Cuezva JM (2018) The Role of Mitochondrial H+-ATP Synthase in Cancer. Front. Oncol. 8:53. doi: 10.3389/fonc.2018.00053*

Keywords: oxidative phosphorylation, ATPase inhibitory factor 1, mitohormesis, metabolic reprogramming, hepatocarcinogenesis, inflammation

#### OVERVIEW OF METABOLIC REPROGRAMMING IN CANCER

Cancer cells experience a series of alterations during oncogenic transformation that confer them new features (1). Cellular metabolism is a central player in the acquisition of this new phenotype. Indeed, cancer cells are highly proliferative and readapt their metabolism to meet the demands imposed by the new phenotype, namely higher requirements of metabolic energy and of precursors for biosynthetic purposes (**Figure 1**) (2–5). One prominent feature of the metabolic reprogramming experienced by cancer cells is an enhanced glycolytic rate in the presence of oxygen, what is known as aerobic glycolysis (6) (**Figure 1**). Glycolysis provides cancer cells with various metabolic precursors that serve for the synthesis of amino acids, nucleotides, and lipids, as well as reducing power and ATP. In addition, mitochondrial function is readapted during oncogenic

transformation and mutations in genes encoding mitochondrial proteins contribute to cancer development (7, 8). In particular, the relative contribution of mitochondria to energy provision is reduced, the organelles becoming mostly dedicated to produce anabolic precursors through the tricarboxylic acid (TCA) cycle (9) (**Figure 1**). Likewise, mitochondria, which are crucial hubs in intracellular signaling (10), readapt this function in cancer (11). Intermediates of the TCA cycle contribute to signaling tumorigenesis (12) and mtROS, which are key mediators in mitochondrial communication (13), activate signaling pathways that promote cell proliferation and tumorigenesis (14, 15). Moreover, mitochondrial dynamics is also reprogrammed in cancer (16).

## THE H**+**-ATP SYNTHASE IS DOWNREGULATED IN CANCER

A critical enzyme complex within the mitochondria is the H<sup>+</sup>-ATP synthase, the rotatory engine of the inner mitochondrial membrane responsible for ATP synthesis by oxidative phosphorylation (OXPHOS) (17). The H<sup>+</sup>-ATP synthase consumes the proton electrochemical gradient generated across the inner mitochondrial membrane by the electron transport chain to drive ATP synthesis (17, 18). In addition, the H<sup>+</sup>-ATP synthase is a critical component of the mitochondrial permeability transition pore (PTP) whose prolonged opening triggers the execution of cell death (19–21). Although its mechanism of participation in PTP opening is currently debated (22–24), recent findings have mapped the residues in subunits of the H<sup>+</sup>-ATP synthase for Ca2<sup>+</sup>

activation (25) and pH inhibition (26) of the PTP, reinforcing the role of the H<sup>+</sup>-ATP synthase in PTP function and providing first evidence that single point mutations in the enzyme affect PTP modulation. Hence, the H+-ATP synthase integrates the bioenergetic and death-signaling functions of mitochondria, what makes it a relevant target for oncogenic transformation (27, 28). Actually, mutations in the mitochondrial-encoded subunit *a* of the H<sup>+</sup>-ATP synthase (MT-ATP6), which are found in different human carcinomas, promote tumor growth by restraining cell death (29, 30). Recent findings in yeast MT-ATP6 mutants have confirmed the role of these mutations in the PTP response to Ca2<sup>+</sup> (31), providing additional genetic evidence that supports the involvement of mutations in the H<sup>+</sup>-ATP synthase in PTP functioning during carcinogenesis. However, it should be noted that the two mutations in MT-ATP6 impact the PTP response only when the function of the outer mitochondrial membrane porin complex is perturbed (i.e., OM45-GFP background) (31). In fact, permeability transition has been documented in rho0 cells that lack mtDNA (32), highlighting the relevance of the genetic background of the cancer cell for the desensitization of the PTP.

Regardless of oncogenic mutations on the H<sup>+</sup>-ATP synthase, it has been documented that the relative expression of the catalytic subunit of the complex (β-F1-ATPase) is downregulated in most prevalent human carcinomas when compared with the corresponding normal tissues (33, 34) [for review, see Ref. (2)]. The relative expression of β-F1-ATPase in the tissue provides a "bioenergetic signature" of the carcinoma that informs of the overall capacity of mitochondria. The bioenergetic signature [also known as the bioenergetic cellular (BEC) index (2)], is assessed as the protein ratio between β-F1-ATPase and GAPDH and has been shown to be significantly reduced in colon, lung, breast, gastric, and renal carcinomas (2, 33). Interestingly, the quantification of these two proteins in carcinomas derived from different tissues (lung, esophagus, and breast) show similar quantities irrespective of the large differences found in their content in normal tissues (35). These findings support that during oncogenic transformation the tissue-specific differences in energy metabolism are abolished to converge on a similar phenotype to support tumor growth (35). In addition, the BEC index is a biomarker for cancer prognosis and response to therapy. In fact, a higher BEC index predicts a better overall survival and/or disease-free survival in acute myeloid leukemia patients and in colon, lung, breast, and ovarian cancer patients (36–42). These findings thus support that an impaired bioenergetic function of mitochondria favors recurrence and progression of the disease. Moreover, the BEC index also provides a tool for predicting the therapeutic response to various chemotherapeutic strategies aimed at combating tumor progression (43–46).

From a mechanistic view point the control of β-F1-ATPase expression is essentially exerted at post-transcriptional levels (47). In this regard, the translation of β-F1-ATPase mRNA (β-mRNA) both during development and in oncogenesis requires the specific activity of a *cis* element in the 3′ untranslated region of the mRNA that tightly controls its translation by RNA binding proteins (48–53) and miRNAs (54).

# THE DIVERSE ROLE OF INHIBITORY FACTOR 1 (IF1) IN HUMAN CARCINOMAS

Besides the lower BEC index found in tumors, some prevalent human carcinomas also upregulate the expression of the ATPase IF1, which is the physiological inhibitor of the H<sup>+</sup>-ATP synthase (55, 56). Classically, IF1 was thought to function only to prevent mitochondrial ATP consumption by the reverse activity of the H+-ATP synthase (ATP hydrolase), which happens when mitochondria become de-energized such as in ischemia or in hypoxia (57, 58). However, more recent findings indicate that IF1 can bind to the H+-ATP synthase under normal phosphorylating conditions, hence inhibiting also the forward ATP synthetic activity of the enzyme (59). It should be noted that when arguing about the inhibition exerted by IF1 on the H<sup>+</sup>-ATP synthase it is important to take into consideration the tissue content of IF1 and the molar ratio that exists between IF1 and the H<sup>+</sup>-ATP synthase because the tissue availability of the inhibitor affects, among other factors, its interaction with the enzyme by the mass–action ratio. Unfortunately, the information of the tissue content of these two proteins in human and mouse tissues is presently missing.

In addition, it should be stressed that IF1 binding to the H<sup>+</sup>-ATP synthase, and hence its activity as an inhibitor of the enzyme, is subjected to a stringent posttranslational regulation of the protein by phosphorylation (59). In this regard, we have shown that IF1 is phosphorylated in S39 by a mitochondrial cAMP-dependent protein kinase that renders IF1 unable to bind to the H<sup>+</sup>-ATP synthase and hence inactive as an inhibitor of the enzyme (59). Regulation of IF1 phosphorylation depends on the cellular metabolic state to allow the fine tuning of ATP production to the cellular metabolic demand (56, 59). In this regard, we should stress that IF1 is found dephosphorylated, and hence active as an inhibitor of the enzyme in colon, lung, and breast carcinomas as well as in hypoxic cells and in cells progressing through the S/G2/M phases of the cell cycle (59).

In addition, IF1 is sharply upregulated in colon, lung, breast, and ovarian carcinomas, which are tissues that under normal physiological conditions are essentially devoid of the inhibitory protein (60–62). Not surprisingly, IF1 is an independent prognostic marker of disease progression for patients bearing these carcinomas. In non-small cell lung cancer (63), bladder carcinomas (64), and gliomas (65), a high expression level of IF1 in the tumor predicts a worse patient prognosis. On the contrary, in colon and breast cancer patients, a high level of IF1 expression predicts a better outcome (60, 66), especially in the bad prognosis group of triple-negative breast cancer patients (67). In the case of breast cancer, lymph node metastases show a lower expression level of IF1 when compared with the primary tumors (68). This finding suggested that breast cancer cells expressing low levels of IF1 may have a higher metastatic potential, which is in full agreement with our recent finding that low IF1 expression in triple-negative breast cancer cells confers a more invasive phenotype (67).

By contrast, human tissues that express high levels of IF1 under basal physiological conditions such as endometrium, kidney, liver, and stomach do not experience a relevant increase in IF1 expression by oncogenesis (60, 62). Nevertheless, in hepatocarcinomas (69) and in gastric carcinomas (70), a higher tumor content of IF1 predicts a worse prognosis for the patients. However, it should be mentioned that a higher IF1-mRNA expression level is correlated with a better prognosis in patients bearing the intestinal subtype of gastric cancer (66). This apparent discrepancy might arise from the histological type of gastric carcinomas analyzed in both studies and/or because the expression of IF1 in human carcinomas is primarily exerted at posttranscriptional levels (60). Overall, these findings support that IF1 plays a relevant role in cancer origin and progression. However, it remains to be elucidated the differential role played by IF1 in favoring or repressing cancer progression in different types of carcinomas, strongly emphasizing the need for specific studies in cellular, xenograft, and genetically modified mouse models in which to address these issues (67, 71).

# THE IMPACT OF IF1-MEDIATED INHIBITION OF THE H**+**-ATP SYNTHASE IN CANCER

The role of the H<sup>+</sup>-ATP synthase in cancer and in signaling has been studied by developing cellular and mouse models with regulated expression of IF1. The overexpression of IF1 both in cultured cells (55, 60, 61, 67) and in different tissues *in vivo* (71–73) is sufficient to promote metabolic reprogramming to an enhanced aerobic glycolysis (**Figure 2**). Upregulation of glycolysis results from the limitation of cellular ATP availability as a result of the inhibition of the H<sup>+</sup>-ATP synthase, supporting that the rate of ATP production by OXPHOS defines the rate of glucose consumption by aerobic glycolysis (74, 75). Likewise, this metabolic situation triggers the activation of the energy sensor AMPK (71–73) (**Figure 2**). Conversely, the silencing of IF1 in cells that express high levels of the protein has the opposite metabolic effect (55, 60, 61).

The inhibition of the H<sup>+</sup>-ATP synthase by IF1 also reduces the backflow of protons into the mitochondrial matrix, triggering mitochondrial hyperpolarization and a mild increase in the production of mitochondrial ROS (mtROS) (**Figure 2**), both *in vitro* and *in vivo* (55, 60, 61, 67, 72, 73). In cells and tissues overexpressing IF1 mtROS trigger the carbonylation of some cellular proteins and signal the activation of the canonical nuclear factor kappa B (NFκB) pathway (71–73) (**Figure 2**). In colon cancer cells, NFκB induces a nuclear transcriptional program that favors cellular proliferation, invasion, and evasion of cell death (61). These results are in line with other findings reporting that mtROS are necessary for proliferation and tumorigenesis and that ROS scavenging with mitochondrial-targeted antioxidants reduces cancer cell growth and prosurvival pathways (14, 76, 77). A recent study also argues that IF1 overexpression in carcinomas might contribute to cancer progression by limiting the processing of the pro-fusion dynamin-related protein optic atrophy 1 and thus limiting cristae remodeling during apoptosis (78).

However, the phenotypic changes triggered by IF1 overexpression in colon cancer cells cannot be generalized to other cellular types. In fact, the transcriptional program triggered by IF1 overexpression in triple-negative breast cancer cells supports just the opposite, a less proliferative and invasive phenotype (67). This phenotype for breast cancer cells was confirmed by functional

FIGURE 2 | Main metabolic and redox circuits regulated by the inhibition of the H+-ATP synthase by inhibitory factor 1 (IF1). Dephosphorylated IF1 (green rod) inhibits the H+-ATP synthase (structure) when bound to the enzyme, while IF1 phosphorylation (yellow) prevents its binding and hence its inhibitory activity (59). The inhibition of a relevant pool of molecules of H+-ATP synthase by IF1 reduces cellular ATP availability, promoting metabolic reprogramming to an enhanced glycolysis and the activation of AMPK (71–73). Moreover, IF1 inhibition of the H+-ATP synthase triggers mitochondrial hyperpolarization because prevents H+ backflow through the enzyme enhancing the production of mitochondrial ROS (mtROS) that activate the canonical NFκB pathway (61, 72, 73). Activation of NFκB triggers the induction of different tissue-specific mitohormetic programs in the nucleus of the cell. In colon cancer cells overexpressing IF1, these programs favor proliferation, invasion, and resistance to cell death (61). In transgenic mice overexpressing IF1 in neurons or liver, promote the activation of survival pathways and antioxidant defenses (71, 72). In transgenic mice overexpressing IF1 in the intestine, the activated programs modulate the immune response of the tissue, favoring the development of an anti-inflammatory phenotype (73). Abbreviations: p-IF1, phosphorylated IF1; AMPK, AMP-activated protein kinase; NFκB, nuclear factor kappa B. The structure of the H+-ATP synthase was obtained from PDB.

analysis illustrating that IF1 overexpression promotes cell adhesion and maintenance of the extracellular matrix hampering epithelial to mesenchymal transition (67). Accordingly, the results may explain why breast cancer patients with high IF1 expression in the carcinoma have a better prognosis (60, 67).

In the mouse model overexpressing IF1-H49K (a constitutively active mutant of IF1) in the liver, we have shown that downregulation of OXPHOS triggers the induction of AMPK rendering a liver phenotype that is prone to cancer development (71). Indeed, transgenic mice when challenged with the carcinogen diethylnitrosamine develop more and bigger tumors than control mice because there is more extensive proliferation and diminished apoptosis of liver cells (71). Remarkably, IF1 overexpression in human hepatocarcinomas also triggers the activation of NFκB (**Figure 2**), which drives the promotion of angiogenesis and epithelial to mesenchymal transition (69). Not surprisingly, the expression of IF1 in liver cancer predicts a bad overall prognosis and the recurrence of the disease in these patients (69).

Mechanistically, although dimers of H<sup>+</sup>-ATP synthase are critical components of the PTP (19), and the overexpression of IF1 in the liver *in vivo* favors the formation of dimers of the enzyme (71), we have observed that cell death protection is not related to differential opening and regulation of the PTP (71). Actually, we support that the cell death protection afforded by IF1 overexpression in the liver is related to mitohormetic signaling through the induction of an antioxidant response guided by Nrf2 [nuclear factor (erythroid-derived 2)-like 2] (**Figure 2**) (71) because the metabolically preconditioned hepatocytes are more resistant to acetaminophen induced toxicity (71). Interestingly, Nrf2 upregulation is also a strategy deployed by cancer cells to detoxify the higher ROS levels that are produced in these cells (79).

## IF1-MEDIATED INHIBITION OF THE H**+**-ATP SYNTHASE MODULATES METABOLIC AND REDOX CIRCUITS

Besides the liver, overexpression of IF1-H49K in neurons *in vivo* also reprograms energy metabolism to an enhanced glycolysis and affords metabolic preconditioning (72). In fact, mice overexpressing IF1-H49K in forebrain neurons are partially protected from excitotoxic damage induced by striatal administration of quinolinic acid because preconditioning partially protects neurons from death, reducing the lesion area in the brain and improving motor performance of the transgenic mice (72). Preconditioning in neurons also involves AMPK activation and mtROS-mediated signaling to implement a Bcl-xL-mediated protection of neurons from apoptosis (72) (**Figure 2**). Other findings also support the neuroprotective role of IF1 in ischemia reoxygenation by promoting autophagy and maintaining mitochondrial bioenergetics (80), although the contribution of IF1 to promotion of autophagy in neurons remains to be studied in the *in vivo* model.

Interestingly, the signaling pathways triggered by the IF1 mediated inhibition of the H<sup>+</sup>-ATP synthase are not limited to the cells overexpressing IF1 but also implicate non-cell autonomous processes. The transgenic mice overexpressing IF1 in enterocytes also show metabolic reprogramming to an enhanced glycolysis and activation of mtROS–NFκB signaling pathway (73) (**Figure 2**). Transgenic mice are partially protected from intestinal inflammation after the administration of the inflammatory agent dextran sodium sulfate (DSS), due to increased recruitment of regulatory T cells and macrophages that are mainly polarized to the M2 phenotype (73). These findings are consistent with the anti-inflammatory phenotype afforded by the overexpression of IF1 as revealed by the higher levels of anti-inflammatory cytokines present in plasma and intestine of the transgenic mice (73). Colonocytes of control mice induce the oncogenic Akt/ mTOR/p70S6K and pro-inflammatory STAT3 pathways upon administration of DSS, something that is not observed in IF1 transgenic mice (73). Preconditioning and protection against stress in colon of IF1-transgenic mice clearly responds to the basal activation of the NFκB pathway due to mtROS production (73) (**Figure 2**), because protection from intestinal inflammation is blunted when an mtROS scavenger or an inhibitor of NFκB are administered (73).

Overall, transgenic mice overexpressing IF1 in liver, brain, or intestine reveal that by partial inhibition of OXPHOS and the production of mtROS the tissues acquire an advantageous phenotype against different forms of oxidative stress and inflammation (71–73) (**Figure 2**), stressing the role of the H<sup>+</sup>-ATP synthase as a therapeutic target in diverse human pathologies.

#### LONGEVITY AND THE INHIBITION OF THE H**+**-ATP SYNTHASE

Besides IF1, the H<sup>+</sup>-ATP synthase can be inhibited by some mitochondrial metabolites produced in the TCA cycle, such as α-ketoglutarate (α-KG) (81). Therefore, partial inhibition of the enzyme by α-KG also reduces ATP availability and TOR signaling, promoting autophagy in *Caenorhabditis elegans* (81). In addition, the oncometabolite (R)-2-hydroxyglutarate (2-HG), which is structurally similar to α-KG, also inhibits the H<sup>+</sup>-ATP synthase both in *C. elegans* and in mammalian cell lines (82). 2-HG is highly accumulated in some gliomas and acute myeloid leukemias that harbor mutations in the genes encoding the cytosolic and mitochondrial isocitrate dehydrogenases (IDH1 and IDH2, respectively). These mutations result in neomorphic enzymes with higher affinity for α-KG that catalyze its conversion into 2-HG (83, 84). Interestingly, the inhibition of the H<sup>+</sup>-ATP synthase by 2-HG or α-KG in glioblastoma cells triggers cell growth arrest and cell death under conditions of limited glucose (82). These results are consistent with previous findings that indicate that brain cancer patients with IDH mutations have a longer median overall survival than patients without mutations (83, 85). The suppression of cell growth may be ascribed to reduced ATP levels and mTOR signaling in the tumors (82).

Interestingly, both α-KG and 2-HG, by inhibiting the H<sup>+</sup>-ATP synthase, also extend lifespan in *C. elegans* (81, 82), and α-KG might play a role in longevity induced by dietary restriction (81). Other interventions targeting the H<sup>+</sup>-ATP synthase have also been shown to extend lifespan in several model organisms, as recently reviewed (62). For instance, silencing of subunits of the H<sup>+</sup>-ATP synthase in *C. elegans* and *Drosophila melanogaster* promotes longevity (86, 87). No contribution to longevity has been reported so far in mammals for the modulation of the H<sup>+</sup>-ATP synthase. However, recent findings indicate that the specific inhibition of the enzyme by the small molecule J147 prevents the age-associated drift of the hippocampal transcriptome and plasma metabolome in mice and extends lifespan in *D. melanogaster,* providing an additional link between the activity of the H<sup>+</sup>-ATP synthase, aging and age-associated pathologies such as dementia (88).

#### REFERENCES


Overall, these findings support the notion that the H<sup>+</sup>-ATP synthase is also a conserved hub in intracellular signaling that plays a key role in signaling mitohormesis contributing to cell fate decisions and longevity.

# CONCLUDING REMARKS

Reprogramming cellular metabolism is a hallmark of cancer that is necessary to fulfill the metabolic demands imposed by the oncogenic process. In this context, the mitochondrial H<sup>+</sup>-ATP synthase is a main hub in rewiring energy metabolism and in retrograde signaling to the nucleus programs required for cancer progression. In this regard, most prevalent carcinomas show reduced expression of the catalytic subunit of the H<sup>+</sup>-ATP synthase (β-F1-ATPase) relative to the glycolytic GAPDH, what provides a protein signature of energy metabolism of clinical relevance in oncogenesis. Moreover, some prevalent carcinomas also show an increased expression of IF1, the physiological inhibitor of the enzyme. As revealed in different *in vitro* and *in vivo* systems, IF1 overexpression is sufficient to rewire energy metabolism to an enhanced glycolysis and to trigger an mtROS signal that promotes nuclear reprogramming. IF1-mediated reprogramming is mainly geared by the activation of AMPK and NFκB pathways resulting in the induction of tissue-specific programs aimed at preventing cell death, oxidative damage or inflammation. The precise molecular events that lead to the upregulation of IF1 in cancer and its role in cancer progression in different carcinomas remain to be established. The H<sup>+</sup>-ATP synthase, engine of OXPHOS, is also a crucial hub in mitohormetic signaling to modulate cytoprotective defenses that contribute to longevity in several organisms. Therefore, deciphering the metabolic and redox circuits controlled by the H<sup>+</sup>-ATP synthase and IF1 are of utmost importance to understand how they contribute to oncogenesis and thus providing new targets for cancer and age-associated diseases.

#### AUTHOR CONTRIBUTIONS

PE-M and JC wrote the paper. All the authors read, contributed, and approved the final manuscript.

# ACKNOWLEDGMENTS

PE-M was supported by a predoctoral fellowship from Fundacio´n La Caixa (Obra Social La Caixa). The work was supported by grants from MINECO (SAF2016-75916-R), CIBERER-ISCIII, and Fundación Ramón Areces, Spain.


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

*Copyright © 2018 Esparza-Moltó and Cuezva. 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 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.*

# Phosphoinositide 3-Kinase/Akt Signaling and Redox Metabolism in Cancer

#### *Nikos Koundouros1,2 and George Poulogiannis1,2\**

*1Department of Cancer Biology, Institute of Cancer Research, London, United Kingdom, 2Division of Computational and Systems Medicine, Department of Surgery and Cancer, Imperial College London, London, United Kingdom*

Metabolic rewiring and the consequent production of reactive oxygen species (ROS) are necessary to promote tumorigenesis. At the nexus of these cellular processes is the aberrant regulation of oncogenic signaling cascades such as the phosphoinositide 3-kinase and AKT (PI3K/Akt) pathway, which is one of the most frequently dysregulated pathways in cancer. In this review, we examine the regulation of ROS metabolism in the context of PI3K-driven tumors with particular emphasis on four main areas of research. (1) Stimulation of ROS production through direct modulation of mitochondrial bioenergetics, activation of NADPH oxidases (NOXs), and metabolic byproducts associated with hyperactive PI3K/Akt signaling. (2) The induction of pro-tumorigenic signaling cascades by ROS as a consequence of phosphatase and tensin homolog and receptor tyrosine phosphatase redox-dependent inactivation. (3) The mechanisms through which PI3K/Akt activation confers a selective advantage to cancer cells by maintaining redox homeostasis. (4) Opportunities for therapeutically exploiting redox metabolism in *PIK3CA* mutant tumors and the potential for implementing novel combinatorial therapies to suppress tumor growth and overcome drug resistance. Further research focusing on the multi-faceted interactions between PI3K/Akt signaling and ROS metabolism will undoubtedly contribute to novel insights into the extensive pro-oncogenic effects of this pathway, and the identification of exploitable vulnerabilities for the treatment of hyperactive PI3K/Akt tumors.

#### *Edited by:*

*Giuseppe Filomeni, Danish Cancer Society, Denmark*

#### *Reviewed by:*

*Paolo Armando Gagliardi, Universität Bern, Switzerland Antonella Papa, Monash University, Australia*

*\*Correspondence: George Poulogiannis george.poulogiannis@icr.ac.uk*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

> *Received: 11 March 2018 Accepted: 26 April 2018 Published: 15 May 2018*

#### *Citation:*

*Koundouros N and Poulogiannis G (2018) Phosphoinositide 3-Kinase/Akt Signaling and Redox Metabolism in Cancer. Front. Oncol. 8:160. doi: 10.3389/fonc.2018.00160*

Keywords: reactive oxygen species, cancer, phosphoinositide 3-kinase/Akt signaling, oxidative stress, metabolism

# INTRODUCTION

Tumorigenesis is a multi-step process involving the complex interplay of several biological processes that are highly dependent on the activation of pro-proliferative and pro-survival signaling cascades, accumulation of genetic aberrations, and adaptation to various microenvironmental stress conditions (1, 2). Underpinning these tumorigenic processes is the ability of cancer cells to alter their metabolism to promote nutrient synthesis or scavenging to support their high proliferation demands (3, 4). A major consequence of such extensive metabolic rewiring is the production of reactive oxygen species (ROS), which include both free radical molecules such as superoxide (O2 •− ) and hydroxyl radicals (<sup>⋅</sup> OH), as well as non-free radical species, of which hydrogen peroxide (H2O2) is among the most prominent (5). Although elevated ROS levels can have detrimental consequences on cell viability through extensive damage of proteins, DNA, and organelles, the concomitant increase in antioxidant and detoxification capacities in cancer cells allows for redox homeostasis, thus generating a tightly regulated system whereby ROS can promote tumorigenesis and cancer progression (6, 7). Specifically, there is renewed interest in examining the implications of redox balance on cancer cell proliferation and survival through the regulation of key signaling cascades such as the phosphoinositide 3-kinase and AKT (PI3K/Akt) pathway, which is of particular relevance as it controls many hallmarks of cancer (1, 8, 9). In this review, we will examine the emerging relationship between redox homeostasis and PI3K/Akt signaling, and discuss how their cross-regulation may promote cancer pathogenesis. In addition, we present and look with optimism opportunities toward a future therapeutic exploitation of redox homeostasis in tumors with enhanced PI3K/Akt activation.

#### THE PI3K/AKT PATHWAY AT A GLANCE

The intricacies of PI3K/Akt signaling have been extensively reviewed previously (10–12), and will only be briefly introduced here. Activation of receptor tyrosine kinases (RTKs) or G-protein-coupled receptors facilitate the recruitment of class I PI3Ks which phosphorylate phosphatidylinositol-(4,5) bisphosphate (PIP2) to phosphatidylinositol (3,4,5)-trisphosphate (PIP3). Oncogenic mutations in *PIK3CA*—the gene encoding the class I PI3K catalytic subunit p110α—are found in approximately onethird of human cancers and 40% of breast cancers (13, 14). The accumulation of PIP3 allows for the localization of AKT to the plasma membrane and its subsequent activation following phosphorylation on serine 473 and threonine 308 by mTORC2 and PDK1, respectively. Negative regulation of the PI3K/Akt pathway is largely mediated by the phosphatase and tensin homolog (PTEN) through dephosphorylation of PIP3 to PIP2. AKT is an important downstream effector of oncogenic PI3K signaling and regulates several pathways, including inhibition of apoptosis, stimulation of mTORC1-dependent cell growth, and modulation of cellular metabolism (11). Glucose and glutamine are considered as essential nutrients for cancer cell proliferation, and PI3K/ Akt signaling has been shown to regulate the cellular metabolism of both nutrients. Specifically, AKT may directly increase glucose uptake through activation of the glucose transport receptor GLUT1, and stimulate glycolysis by phosphorylating hexokinase 2 (11, 15). Moreover, activating mutations in *PIK3CA* render colorectal cancer cells more dependent on glutamine anaplerosis to replenish TCA cycle intermediates through upregulation of glutamate pyruvate transaminase 2, and glutamine deprivation significantly reduces the proliferation of *PIK3CA* mutant, but not wild type, cancer cells (16). Later work has also demonstrated the importance of Akt-independent signaling cascades associated with PI3K activation, with a particular focus on serum and glucocorticoid-regulated kinases (SGKs) (17).

#### REGULATION OF ROS PRODUCTION BY PI3K/AKT SIGNALING

Aberrant PI3K/Akt signaling drives many of the molecular mechanisms contributing to increased ROS levels through direct modulation of mitochondrial bioenergetics and activation of NADPH oxidases (NOXs), or indirectly through the production of ROS as a metabolic by-product (**Figure 1**) (6, 18). Mitochondria are a major source of cellular ROS, and these are largely derived from electron leakage at complexes I and III of the electron transport chain (6, 19). Complex I in particular has been shown to generate ROS through two mechanisms: the first involves reduction of O2 by flavin mononucleotide following binding of NADH induced by high NADH/NAD<sup>+</sup> ratios, and the second is through reverse electron transport whereby excessive NADH is produced following reduction of NAD+ from ubiquinol (6, 20). Interestingly, AKT has been shown to translocate to the mitochondrial matrix and inner membrane in a PI3Kdependent manner following IGF-1 stimulation (21). AKT may directly phosphorylate mitochondrial GSK-3β, reducing its activity and thus alleviating the negative regulation imposed on pyruvate dehydrogenase and α-ketoglutarate dehydrogenase complexes, which have been reported to generate superoxide and H2O2 (19, 22, 23).

The activation of NOXs by PI3K/Akt signaling also contributes to higher ROS levels in cancer cells (**Figure 1**) (18). There are seven members of the NOX enzyme family (NOX1-5, DUOX1, and DUOX2) that are expressed across several tissue types including the liver, lung, and gastrointestinal tract (24). NOXs are composed of different subunits including p67phox, p47phox, p40phox, p22phox, RAC1, RAC2, and NOXO1/NOXA1 which facilitate the localization of the enzyme complex to the cytoplasm or plasma membrane (24). The main function of NOX enzymes was first ascertained in phagocytes as contributing to the respiratory burst through production of superoxide following electron transfer from NADPH to oxygen (25). The activity of these enzymes is also becoming increasingly associated with various hallmarks of cancer including angiogenesis and metastasis (26). The importance of PI3K/Akt signaling in activating NOXs has been demonstrated by specifically ablating NOX activity following treatment with the PI3K inhibitor wortmannin or knockout of *Akt1* (18, 27). In terms of cancer progression, PI3K-mediated activation of NOX is necessary to promote cell migration and chemotaxis in response to stimulants such as hepatocyte growth factor (HGF) and platelet-derived growth factor (PDGF) (28, 29). Migration of lung endothelial cells requires HGF-mediated activation of c-MET, which signals through the PI3K/Akt pathway and results in the accumulation of p47phox/Rac1 in lamellipodia and localized production of H2O2 to the leading edge of the cell (29). Consequently, inhibition of the PI3K/Akt pathway also inhibits the translocation of NOX subunits and reduces ROS, hindering the metastatic potential of lung cancer cells, at least in part, by reducing expression of the metalloprotease MMP9 and the prometastatic miRNA miR-21 (29–31). Angiogenesis is necessary for tumor growth especially following metastasis, and NOX isoforms have been implicated in re-vascularization particularly in PI3K/ Akt-hyperactive tumors (32). Vascular endothelial growth factor is the most potent stimulant of angiogenesis and can activate NOX isoforms either directly or indirectly through PI3K/Akt induction (32–34). The subsequent production of superoxide and H2O2 are necessary for the regulation of transcription factors, which promote angiogenesis, including NF-κB, MMPs, COX-2, and HIF-1α (32).

It is well established that hyperactive PI3K/Akt signaling contributes to increased glycolytic rate, and recent studies have identified dependencies on specific metabolic enzymes that, when perturbed, impact glycolytic flux and consequently cancer cell viability (35). Notably, *PIK3CA* mutant cancer cell lines have elevated 2-oxoglutarate dehydrogenase (OGDH) activity, and inhibition of this enzyme leads to reduced tumor growth *in vivo* (35). The sensitivity to OGDH inhibition can be explained by the observed decrease in the NAD+/NADH ratio following accumulation of 2-oxoglutarate, thus limiting the available NAD<sup>+</sup> needed for glycolysis (35). Interestingly, OGDH is a potent source of ROS and this, together with the need to maintain a stable NAD<sup>+</sup>/NADH ratio, suggests that maintaining redox homeostasis is essential for the proliferation of *PIK3CA* mutant tumors (35, 36). Another important ROS-generating metabolic pathway that is regulated by PI3K/Akt signaling is the production of prostaglandins (PGE2) from arachidonic acid by the cyclooxegenases COX-1 and COX-2 (37). The peroxidase activity of the COX enzymes results in the generation of superoxide as a by-product (38). Recent clinical studies have also highlighted the importance of PGE2 metabolism in PI3K-driven tumors, with *PIK3CA* mutation status being used as a biomarker for aspirin treatment in colorectal cancer (39). Thus, in addition to the direct effects of PI3K/Akt pathway in augmenting ROS levels, several indirect mechanisms associated with the intrinsic metabolic dependencies of *PIK3CA* mutant tumors exist, which could also contribute to the cellular pool of ROS levels.

# ACTIVATION OF PI3K/AKT AND PRO-TUMORIGENIC SIGNALING BY ROS

Elevated pro-proliferative signaling cascades and inhibition of growth suppressors are necessary for tumorigenesis, and high ROS levels affect both, by potentiating activation of PI3K/Akt signaling mainly through inhibition of phosphatases such as PTEN or direct activation of oncogenes including AKT (40, 41). In breast cancer, exposure to hormones including 17-β estradiol and its derivative 4-OH-E2 lead to a dose-dependent increase in ROS levels and consequent malignant transformation of MCF10A cells *in vitro* and *in vivo* (42). Mechanistically, this is dependent on ROS-mediated hyper-phosphorylation of PI3K and subsequent downstream PDK1-mediated activation of AKT, culminating in the upregulation of cell cycle promoting genes such as *CDK1* and *PCNA* (42). Interestingly, treatment with ROS scavengers such N-acetylcysteine (NAC) or knockdown of *AKT1* rescues the malignant transformation induced by 4-OH-E2 exposure, indicating an important regulatory role of ROS on PI3K and AKT (42). This regulation has also been demonstrated in T-cell acute lymphoblastic leukemia (T-ALL) where interleukin-7 (IL-7) enhances NOX and mitochondrial complex I activity, thus contributing to elevated ROS in T-ALL cells (43). AKT phosphorylation and activation is induced by the IL-7-dependent increase in ROS levels and promotes glucose uptake through upregulation of the GLUT1 receptor (43).

Excessive oxidative stress also activates PI3K/Akt signaling by inhibiting the activity of its negative regulator PTEN, one of the most frequently altered tumor suppressor genes in cancer (44). High ROS levels modulate PTEN through direct oxidation leading to reduced phosphatase activity, or indirectly through phosphorylation, which increases its stability and prevents its recruitment to the membrane (45, 46). For example, H2O2 mediated oxidation of cysteine residues 124 and 71 on PTEN leads to the reversible formation of disulfide bridges and reduction of its catalytic activity (47, 48). PTEN oxidation and subsequent inactivation can also occur in the absence of H2O2 treatment. This has been demonstrated in the COX-2- and LOX-5-dependent synthesis of prostaglandins from arachidonic acid in pancreatic cancer cells (49). Oxidized PTEN as a result of arachidonic acid metabolism does not migrate at a lower molecular weight under non-reducing conditions, whereas a migratory shift is seen following the formation of a disulfide bond between Cys-124 and Cys-71 upon H2O2 treatment (49). As COX-2 expression and activity is regulated by AKT, a model could exist in pancreatic tumors whereby arachidonic acid metabolism perpetuates PI3K/ Akt activity through PTEN inactivation, consequently resulting in persistent prostaglandin production and associated inflammation (50, 51).

Oxidative stress can also induce posttranslational modifications, which facilitate PTEN ubiquitylation and subsequent degradation. One such redox-dependent modification, which contributes to PTEN inactivation and enhanced ubiquitinproteasome degradation is S-nitrosylation (SNO) (52, 53). Mitochondrial dysfunction, which is triggered by loss of the tumor suppressor *PARK2* leads to a reduction in ATP levels and concomitant activation of AMPK (52, 54). AMPK-mediated activation of nitric oxide synthase 3 (NOS3/eNOS) leads to enhanced nitric oxide (NO) levels and NO-derived reactive nitrogen species, contributing to PTEN SNO on Cys-83 (52, 53, 55). This modification is important for the proliferation of PTEN proficient cells under energy-deprived conditions (52). Of note, the AKT kinase can also signal directly to activate eNOS, contributing to high NO levels (56). Nitrosative stress-induced posttranslational modifications, unlike in the case of PTEN, are not always limited to inhibitory effects, and can result in enhanced stabilization and activation of oncogenes including EGFR and Src that also contribute to the activation of the PI3K/ Akt pathway (57). Taken together, these findings suggest a model whereby multiple oncogenic signaling cascades, in particular PI3K/Akt signaling, can be potentiated through elevated nitrosative stress to promote pro-survival adaptations during nutrient deprivation.

Reactive oxygen species can also modulate PI3K/Akt signaling by regulating protein tyrosine phosphatases (PTPs), which inhibit RTKs such as EGFR and PDGFR through dephosphorylation. PTPs can be broadly classified into four main groups based on their specific substrates and all contain an essential cysteine residue in the catalytic domain (58). ROS—and H2O2 in particular—have been shown to reversibly oxidize this cysteine residue, leading to reduced phosphatase activity of PTPs and sustained RTK activation (59). Notably, detoxification of H2O2 by overexpressing catalase or inhibition of NOXs contributes to a marked reduction in tyrosine phosphorylation of PDGFR, EGFR, and the insulin receptor (60, 61). Perhaps the most potent activator of PI3K/Akt signaling is insulin, and stimulation of IRS-1 following insulin binding, increases H2O2 generation and NOX4 activity (62). PTP1B and SHP-2 are the main PTPs that dephosphorylate IRS-1; therefore, their inhibition in response to insulin-induced ROS levels facilitates persistent activation of many downstream signaling cascades including the PI3K/ Akt pathway (62). The implications of this regulation in the cancer context are not only limited to increased proliferation and metastatic potential, but also play a major role in promoting drug resistance (63, 64). The development of drug resistance is particularly relevant in breast cancer, where dysregulated IRS-1/ IGF-1R signaling decreases the sensitivity to tamoxifen and trastuzumab in estrogen receptor (ER)- and HER2-positive tumors, respectively, by potentiating ERK and PI3K/Akt signaling (63, 65, 66). Thus, by inhibiting several key negative regulators, redox stress has a significant role in the activation of PI3K/Akt signaling and associated pro-tumorigenic phenotypes.

## PI3K/AKT SIGNALING MAINTAINS REDOX BALANCE IN CANCER CELLS

In order for ROS to confer a selective advantage to cancer cells, there must be a concomitant increase in antioxidant responses to prevent adverse effects on cell viability. These responses include upregulation of the Keap1–Nrf2 pathway and modulation of glutathione metabolism, both of which are regulated by PI3K/Akt signaling (**Figure 2**). The transcription factor Nrf2 is recognized as a central mediator of ROS detoxification by inducing the expression of several enzymes that are involved in the antioxidant response, including glutathione *S*-transferase, superoxide dismutases, and NAD(P)H:quinone oxidoreductase 1 (Nqo1) (67). Under low concentrations of cellular ROS, Nrf2 is bound to the E3 ubiquitin ligase Keap1 in the cytosol and degraded by the proteasome (68). This interaction is inhibited under high concentrations of ROS, which oxidize cysteine residues on Keap1, thus allowing Nrf2 to translocate to the nucleus and induce the expression of genes containing antioxidant response elements (69). PI3K/Akt activation has been shown to be essential for the nuclear translocation of Nrf2, and accordingly treatment of neuroblastoma SH-SY5Y cells with PI3K inhibitors LY294002 and wortmannin, but not MAPK inhibitors, reduces

dehydrogenase; 6-PGDL, 6-phosphogluconolactone; GL, gluconolactonase; 6-PG, 6-phosphogluconate; 6-PGD, 6-phosphogluconate dehydrogenase; R-5-P, ribulose-5-phosphate; GR, glutathione reductase; GPx, glutathione peroxidase; SGK, serum and glucocorticoid-regulated kinase; PI3K, phosphoinositide 3-kinase.

Nrf2 transcriptional activation of antioxidant genes (70). One of the most important implications of PI3K-mediated upregulation of Nrf2 signaling is promoting cancer cell survival by conferring protection against excessive oxidative stress (71, 72). This is particularly relevant in BRCA1-deficient breast cancers, which lack effective DNA repair mechanisms through homologous recombination and are therefore susceptible to genetic modification induced by ROS (73). Interestingly, although up to 80% of BRCA1 mutant breast cancers are ER negative, estrogen positively regulates Nrf2 transcriptional activity through activation of PI3K/Akt signaling allowing BRCA1-null cells to detoxify high ROS levels and accumulate additional genetic aberrations that may contribute to tumorigenesis (73, 74). Inhibition of PI3K/Akt signaling either by PTEN overexpression or BKM120 treatment hinders estrogen-mediated Nrf2 activation, suggesting that targeting this

pathway might be beneficial in treating BRCA1-deficient tumors by re-sensitizing them to elevated ROS levels (73).

Phosphoinositide 3-kinase/Akt activation has both Nrf2 dependent and -independent effects on cellular metabolism that contribute to ROS detoxification. Several antioxidant pathways rely on the reducing power of NADPH, which is predominantly generated by 6-phosphogluconate dehydrogenase (6PGD) and glucose-6-phosphate dehydrogenase (G6PD) from the pentose phosphate pathway (PPP) (75, 76). Active Nrf2 induces the expression of the aforementioned PPP enzymes through an AKTdependent mechanism, as well as enzymes directly involved in NADPH synthesis such as malic enzyme 1 (ME1) and isocitrate dehydrogenase IDH (77). As AKT is a potent activator of glucose uptake and glycolysis, tumors dependent on PI3K/Akt signaling could shunt the glucose-6-phosphate (G6P) generated during glycolysis to the PPP activated by Nrf2, thus maintaining a stable pool of NADPH, which could be used for anabolic processes to sustain tumor growth and proliferation, or ROS detoxification (77, 78) (**Figure 2**).

Phosphoinositide 3-kinase/Akt and Nrf2 signaling, as well as the associated increase in NADPH are key regulators of the synthesis of glutathione, which exists in either reduced (GSH) or oxidized (GSSG) form (79). Under conditions of oxidative stress, enzymes such as GSH transferase (GSH-Tr) and GSH peroxidase (GPx) facilitate the reduction of ROS, including H2O2, by oxidizing GSH to GSSG (79). In order to completely detoxify ROS, oxidized GSSG must then be reduced back to GSH by NADPH-dependent glutathione reductase (GSSG-Rx) (79). The induction of glutathione synthesis through upregulation of glutamate-l-cysteine ligase, as well as transcriptional activation of GSH-Tr and GPx is dependent on a PI3K/Akt/Nrf2 signaling axis (80, 81). AKT can also increase the stability of Nrf2 by activating p21Cip1/WAF1 which disrupts the interaction between Keap1 and Nrf2, and by inhibiting GSK-3β that leads to the reduction of Nrf2 phosphorylation, preventing its nuclear export and ubiquitination (82, 83). Importantly, glutathione biosynthesis has been shown to be a metabolic vulnerability in *PIK3CA* mutant breast cancer cells, as treatment of *PIK3CA* or *AKT* mutant MCF10A cells with buthionine sulfoximine (BSO) significantly reduces anchorage-independent growth and inhibits cell proliferation in 3D, but not 2D culture (83). Furthermore, cisplatin treatment when given in combination with BSO leads to tumor regression of *PIK3CA* mutant, but not wild type, cell line-derived xenograft models indicating that disrupting redox homeostasis through GSH metabolism could repurpose existing therapies for the treatment of *PIK3CA* mutant breast cancers (83).

Although AKT is perhaps the most extensively characterized downstream effector of PI3K signaling, recent studies have highlighted the importance of an Akt-independent axis which relies on other members of the AGC serine/threonine kinase family such as PDK1, RSK, and SGK1 (84, 85). In particular, SGK1 has been shown to promote antioxidant responses during pregnancy that are essential for fetal development, and exert protective effects in Parkinson's disease (86). SGK1 can negatively regulate JNK signaling, an important inducer of superoxide species by modulating complex I activity of the electron transport chain (87, 88). In addition, SGK1-mediated inactivation of GSK3-β facilitates MCL1 localization to the mitochondria to significantly reduce mitochondrial ROS production through inhibition of NOX4 expression, adversely affecting the response to chemotherapy (88, 89). It is important to note, however, that in models of lung cancer, MCL1 may also promote ROS production from the mitochondria by binding to voltage-dependent anion channels and increasing the mitochondrial flux of calcium (90).

#### THERAPEUTICALLY EXPLOITING REDOX HOMEOSTASIS IN PI3K-DEPENDENT TUMORS

In terms of redox homeostasis, it is clear that PI3K/Akt signaling is unique in that it activates both ROS generating and detoxifying processes, thus creating a stable presence of ROS, which can exert pro-tumorigenic effects. This raises the attractive therapeutic prospect of perturbing redox homeostasis in tumors with hyperactive PI3K/Akt signaling either alone or in combination with existing treatments. Chemotherapy and radiotherapy are among the most common therapeutic interventions for cancer patients, and one mechanism through which they induce apoptosis in cancer cells is by upregulating ROS (91). Anthracyclins, such as doxorubicin, and platinum-based therapies including cisplatin, can directly induce ROS production through modulation of the electron transport chain, and elevated ROS levels subsequently activate caspases, cytochrome *c* release, and DNA damage leading to apoptosis (92, 93). Previous clinical studies have demonstrated that *PIK3CA* mutant breast cancers display decreased sensitivity to anthracyclines and cisplatin, however, as these drugs exert anti-cancer effects through ROS upregulation, it is conceivable that the antioxidant pathways, which are elevated by PI3K/Akt signaling, could counteract these therapies (91, 94). Indeed, active Nrf2 protects cancer cells from ROS induced cell death, and inhibition of this transcription factor by brusatol treatment enhances the response to cisplatin (95, 96). In addition, targeting glutathione biosynthesis with BSO selectively sensitizes *PIK3CA* mutant breast tumors to cisplatin as compared to wild-type ones, indicating that impairing the antioxidant response could be an exploitable vulnerability in PI3K-driven tumors (83).

The apparent anti-cancer effects of excessive ROS accumulation are also important in targeting tumors, which have become resistant to PI3K/Akt inhibitors. A common strategy to overcome resistance is to place patients on "drug holidays" and subsequently re-challenge them with the inhibitor (97). In PI3K inhibitor resistant breast cancer cells, substantial metabolic rewiring occurs following removal of the class IA PI3K inhibitor GDC-0941 that is characterized by increased glycolysis and mitochondrial respiration (98). During drug holidays, the metabolic phenotype of resistant cancer cells is altered by an Akt-independent PI3K/ mTORC1 signaling axis, which drives excessive production of ROS and inhibits the proliferation of these cells (98). Notably, this proliferative defect is rescued following treatment of resistant cells with ROS scavengers such as NAC, thus demonstrating the importance of regulating ROS homeostasis in prohibiting the expansion of a resistant cell population emerging through therapies (98). These findings present an opportunity to improve responses by specifically exploiting the unique metabolic profile of PI3K inhibitor resistant cells, either through glucose deprivation or by further increasing oxidative stress (98). Developing resistance to AKT-specific inhibitors such as MK2206 is also a significant clinical problem, and studies into the identification of synthetic lethal interactions between AKT and antioxidant inhibitors have shown promising results in overcoming drug resistance (99). In non-small cell lung carcinoma (NSCLC) models, dual treatment with the thioredoxin reductase-1 (TXNRD1) antioxidant inhibitor auranofin and MK2206 induced cancer cell-specific apoptosis through ROS-stimulated JNK signaling (99). Importantly, this synthetic lethality was observed in lung tumors with functional NRF2–KEAP1 signaling and overexpression of TXNRD1, indicating that the activity of the anti-oxidative response in NSCLC could be used as a biomarker for determining which patients may benefit from dual AKT/ TXNRD1 inhibition (99).

Although it seems that enhancing ROS production may have inhibitory effects on cancer cells, it is important to note that finding the balance between antioxidant and ROS-generating mechanisms is complex and so is their therapeutic exploitation (100). Notably, there is still significant debate regarding the administration of antioxidants during cancer therapy as a means of limiting drug toxicity and whether their use adversely affects the patient's response, and/or the potential systemic consequences of deliberately elevating ROS levels (101). It is, therefore, necessary to consider all of these implications and to ensure that tumors are well characterized at the genetic and metabolic levels to determine if targeting redox homeostasis is a suitable treatment option.

#### CONCLUSION AND FUTURE PERSPECTIVES

Several pro-tumorigenic processes converge on hyperactive PI3K/ Akt signaling, and it is becoming increasingly evident that ROS metabolism is no exception to this. The capacity for the PI3K/ Akt cascade to directly activate both ROS generating and various antioxidant pathways suggests a tight regulation on cellular redox

#### REFERENCES


homeostasis, the intricacies of which merit further investigation. Understanding the complex interplay between ROS and PI3K/ Akt signaling is particularly relevant for developing therapeutic strategies to target tumors dependent on this pathway, especially since recent clinical trials have demonstrated only modest responses to PI3K/Akt pathway inhibitors and development of resistance (102). While ROS metabolism certainly adds another layer of complexity to PI3K/Akt signaling, this area of research holds great promise, not only for the potential identification of novel biomarkers and metabolic dependencies but also the prospect of implementing more potent therapeutic combinations, which perturb redox homeostasis and effectively target PI3K-driven tumors.

#### AUTHOR CONTRIBUTIONS

NK and GP wrote the manuscript and designed the figures.

#### FUNDING

NK is supported by an Institute of Cancer Research PhD studentship. Work in the GP lab is supported by the Institute of Cancer Research and a Cancer Research UK Grand Challenge award (C59824/A25044).


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

*Copyright © 2018 Koundouros and Poulogiannis. 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 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.*

# Computational Structural Biology of S-nitrosylation of Cancer Targets

Emmanuelle Bignon<sup>1</sup> , Maria Francesca Allega<sup>1</sup> , Marta Lucchetta<sup>1</sup> , Matteo Tiberti <sup>1</sup> and Elena Papaleo1,2 \*

*<sup>1</sup> Computational Biology Laboratory, Danish Cancer Society Research Center, Copenhagen, Denmark, <sup>2</sup> Translational Disease Systems Biology, Faculty of Health and Medical Sciences, Novo Nordisk Foundation Center for Protein Research, University of Copenhagen, Copenhagen, Denmark*

Nitric oxide (NO) plays an essential role in redox signaling in normal and pathological cellular conditions. In particular, it is well known to react *in vivo* with cysteines by the so-called *S*-nitrosylation reaction. *S*-nitrosylation is a selective and reversible post-translational modification that exerts a myriad of different effects, such as the modulation of protein conformation, activity, stability, and biological interaction networks. We have appreciated, over the last years, the role of *S*-nitrosylation in normal and disease conditions. In this context, structural and computational studies can help to dissect the complex and multifaceted role of this redox post-translational modification. In this review article, we summarized the current state-of-the-art on the mechanism of *S*-nitrosylation, along with the structural and computational studies that have helped to unveil its effects and biological roles. We also discussed the need to move new steps forward especially in the direction of employing computational structural biology to address the molecular and atomistic details of *S*-nitrosylation. Indeed, this redox modification has been so far an underappreciated redox post-translational modification by the computational biochemistry community. In our review, we primarily focus on *S*-nitrosylated proteins that are attractive cancer targets due to the emerging relevance of this redox modification in a cancer setting.

Keywords: S-nitrosylation, (de)nitrosylating enzymes, redox modifications, molecular dynamics simulations, cysteine, redox cancer biology

#### INTRODUCTION

Despite being amino acids (Cys) play diverse roles in biology. In fact, they represent a special class of residues due to the thiol moiety of their side chain (**Figure 1**). The thiol group can undergo a plethora of different biological modifications affecting protein structure, reactivity, stability and function (1). Cysteines are thus unique molecular switches (2). These modifications include, for example, disulfide bridge formation, high oxidation states, sulfenylation, persulfidation, metalation, S-nitrosylation, glutathionylation, sulfhydratation, among others. Cysteines can also be impacted by lipid modifications? including palmitoylation and prenylation or they can be the coordinating residues for metal ions such as zinc, iron, or copper in metallo-proteins.

#### Edited by:

*Michael Breitenbach, University of Salzburg, Austria*

#### Reviewed by:

*Andrey V. Kozlov, Ludwig Boltzmann Gesellschaft, Austria Mario Schubert, University of Salzburg, Austria*

> \*Correspondence: *Elena Papaleo elenap@cancer.dk*

#### Specialty section:

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

> Received: *18 April 2018* Accepted: *02 July 2018* Published: *14 August 2018*

#### Citation:

*Bignon E, Allega MF, Lucchetta M, Tiberti M and Papaleo E (2018) Computational Structural Biology of S-nitrosylation of Cancer Targets. Front. Oncol. 8:272. doi: 10.3389/fonc.2018.00272*

**72**

Oxidative modifications of cysteine thiols can be a reversible or irreversible processes. Examples of reversible Cys modifications include Cys sulfenylation, S-glutathionylation and S-nitrosylation (SNO). With regards to the latter, NO is a reactive gas produced by NO synthases (NOS) 1–3 using the substrate arginine. NO produced inside the cell can be efficiently and quickly consumed through reactions with (bio)molecules that are in proximity to the NO source (3). A significant amount of the NO signal can be stored and propagated as nitrosyl adducts at specific cysteine sites of proteins via S-nitrosylation.

In this review, we will focus on computational structural and chemical studies that helped the understanding of the complex mechanisms induced by S-nitrosylation, as well as possible future directions for the computational studies of Snitrosylation, which is a rather underappreciated modification compared to other well-known and more investigated posttranslational modifications (PTMs), such as phosphorylation.

At first, we will provide the general background of what Snitrosylation is, what is known on the enzymes regulating this PTM, as well as which are the biological effects so far discovered to be triggered by S-nitrosylation. We will then discuss the recent approaches and findings from bioinformatics, computational chemistry, and biochemistry methods applied to S-nitrosylation with particular attention to targets of interest for cancer research.

#### S-NITROSYLATION OF CYSTEINE RESIDUES

Among the broad spectrum of cysteine redox modifications mentioned above, S-nitrosylation (SNO) accounts for the

*S*-nitrosylation and possible pathways of interconversion. The figure is meant to be illustrative and does not depict the complete reaction schemes. R groups are part of the protein main chain.

oxidative modification of cysteines by nitric oxide (NO) to form S-nitrosothiols (4). SNO is, by far, the reversible Cys modification with the most significant prevalence and cellular functions, providing a ubiquitous mechanism for cellular signaling mediated by thiols. NO exerts its primary biological functions through protein S-nitrosylation so that it can be considered the prototype of redox-based signals (4, 5).

S-nitrosylation is considered a reversible and ubiquitous PTM and many studies demonstrated its role in protein activity, stability, localization, and protein-protein interactions (see section Biological Mechanisms Promoted by S-nitrosylation) in a myriad of cellular processes (6). S-nitrosylation emerged in the last decades as a new paradigm in signal transduction and regulation of proteins (7). More than 3,000 proteins are known to be affected by this redox PTM (3).

Alterations of S-nitrosylated protein targets or the enzymes regulating SNO have been also associated with different pathologies, including cancer, cardiovascular, respiratory, and neurodegenerative disorders (4, 6, 8, 9).

Protein S-nitrosylation features tight spatiotemporal specificity for certain protein Cys residue (10, 11). If physiological amounts of NO are present, only one or few Cys residues of a protein are targeted. These modifications are generally sufficient to change the protein function, activity or specificity for interaction partners (11–13). S-nitrosylation, as well as other alternative S-oxidative modifications mediated by reactive oxygen species, can target separated populations of Cys residues. Thus, depending on the Cys that is modified, the functional effects that are triggered can be very diverse (14). S-nitrosylation has also been shown to target distinct Cys residues with the final goal of exerting a coordinated effect (4). The known and possible determinants of specificity of S-nitrosylation toward certain Cys residues are discussed more in details in section Computational Structural and Chemical Studies of S-nitrosylation. Computational studies allowed to disclose them and define sequence or structural motifs around SNO sites.

One major mechanism promoted by S-nitrosylation of proteins is based on S-transnitrosylation, i.e., the capability of modifying with SNO other protein targets thus allowing the propagation of the SNO-based signals (4), similarly to the wellknown kinase-based signaling cascade.

Different degrees of S-nitrosylation can be observed in protein targets, spanning from cases of mono-SNO (single cysteine) to multiple cysteines (multi-SNO), depending on the availability of NO, as well as on the properties of the target proteins in the proximity of the Cys sites (3). NO production is tightly controlled in normal conditions, and this leads to a basal S-nitrosylation level. In this context, a subset of SNO targets (such as CD40 and pro-caspases) is in their resting state. Another subset of proteins, on the contrary, is constitutively S-nitrosylated. Indeed, they require this modification to be active, as exemplified by caveolin-1 or connexin-43. In cases in which these targets have multiple SNO sites, an incremental degree of S-nitrosylation can be observed with transitions from mono- to poly-SNO which allow a progressive activation in response to stress (3). The degree of SNO and the switch from mono- to multi-SNO can also strongly depend on the availability and accessibility of cysteines in the target protein. Cysteines are, for example, enriched in proteins that are on the cell surface or actively involved in cellto-cell communication and signal transduction, such as CD40, other TNFRs, receptor tyrosine kinases, integrins, and connexins (3).

Due to the high SNO reactivity and the propensity for S-transnitrosylation, to fully appreciate the mechanisms and consequences of S-nitrosylation in normal and cancer cellular contexts is particularly challenging. Hence, it becomes crucial to assess the role of SNO-proteins as both targets of this redox modification and transducers of the SNO signal.

#### ENZYMATIC REGULATION OF S-NITROSYLATION

Cellular S-nitrosylation is dynamically governed by the equilibrium between S-nitrosylated proteins and low-molecularweight S-nitrosothiols, which in turns are tightly controlled by several enzymes, such as S-nitrosylases and denitrosylases (4). NO is generally the product of three main isoforms of NO synthase (NOS) in mammalian cells (see section Nitric Oxide Synthases). S-nitrosothiols (RSNOs) are initially formed via different chemical routes that involve a one-electron oxidation, such as reaction of NO with thiyl radicals, transfer of the NO group from metal-NO complexes to a Cys thiolate, or the reaction of a Cys thiolate with species generated by NO autooxidation (4). Recent evidence suggested, however, an important role for metalloproteins in catalyzing de novo S-nitrosylation [see ref. (4) for more details]. The NO group is then transferred from a donor to an acceptor Cys thiol via S-transnitrosylation, (see section S-nitrosylation of Cysteine Residues). S-nitrosylation occurs not only in proteins but also in low molecular weight (LMW) thiols such as glutathione (GSH) and coenzyme A. SNO-proteins and SNO-LMW thiols exist in thermodynamic equilibria, which are regulated by the removal of the SNO moiety by the direct action of denitrosylases on the protein targets, as well as by the action of GSNOR on SNO-LMW molecules (see sections Denitrosylation Systems, Thioredoxin System and GSNOR System). These enzymes, described in the next sections, are crucial to control steady-state levels of SNO and to ensure the regulation of the NO-based signal cascade.

#### Nitric Oxide Synthases

The compartmentalization of SNO targets with NOS enzymes favors the interaction between the enzymes and the Snitrosylation substrate which can, in turn, occur directly, or through scaffolding proteins (4). For example, the CAPON protein acts as a scaffold to mediate the interaction between nNOS and the S-nitrosylation target Dexras1 (15). Another example is the complex formed by iNOS and S100A8/A9, for which targeted S-transnitrosylation is favored by S100A9 on multiple protein targets that share a short linear motif I/L-X-C-X2-D/E where C represents the SNO site (4).

Each NOS isoform can also be S-nitrosylated, generally through an auto-catalytic mechanism, which involves a metal center. NOSs can then act as S-transnitrosylating partners for scaffolding proteins with or without the intervention of a LMW-SNO (4).

The different NOS isoforms are expressed in various organs, tissues, cells, or even subcellular compartments (16). For example, the localization of eNOS in the Golgi apparatus allows the generation of a local NO pool that specifically targets the compartmentalized proteins for S-nitrosylation (17). An example of another finely tuned compartmentalization is attested by the binding of different NOS variants to separate regions of the same target protein, which result in S-nitrosylation of different Cys residues (4).

#### Denitrosylation Systems

As mentioned in section S-nitrosylation of Cysteine Residues, Snitrosylation is a PTM consisting of a covalent bond formation between of nitric oxide (NO) to a cysteine residue to form a S-nitrosothiol (5). In response to a specific stimulus, Snitrosylated proteins (SNO-proteins) can undergo the reverse reaction of denitrosylation (i.e., reducing SNO back to SH). Two mechanisms of denitrosylation have been identified (18): the S-nitrosoglutathione reductase (GSNOR) system, comprising GSH and GSNOR, and the thioredoxin (Trx) system, comprising Trx and Trx reductase (TrxR) as shown in **Figure 2**. Both mechanisms use intermediate molecules (GSH and Trx) to remove the NO group from S-nitrosylated proteins. In the next paragraphs, we will illustrate these two denitrosylating systems and the cellular processes that they affect.

## Thioredoxin System

Thioredoxins (Trx) are a family of small redox proteins involved in multiple cellular processes.

The Trx system was originally identified as a key player in the cellular redox homeostasis thanks to its role as a disulfide reductase (19). There are two distinct mammalian thioredoxins, Trx1- mainly localized in the cytosol but possibly translocated to the nucleus-and Trx2, mainly located in the mitochondria.

The thioredoxin system is composed of a thioredoxin (Trx), a thioredoxin reductase (TrxR), and NADPH. It includes the dithiol Cys-X-X-Cys active site that is essential for their oxidoreductase function (20). In the denitrosylation process, the reduced Trx [Trx(SH)2] denitrosylates S-nitrosylated proteins forming a reduced protein thiol (-SH) and producing nitroxyl (HNO) or free NO groups and oxidized Trx [Trx(S-S)]; the latter is then reduced by TrxR at the expense of NADPH (21). For example, Trx1 reduces the target protein disulfide bond with concomitant oxidation of its Cys32 and Cys35 residues. TrxR then reduces the oxidized Trx1 thanks to NADPH consumption to regenerate the reduced Trx1 form.

Apart from its canonical role as a disulfide reductase, the Trx system has an important role in denitrosylation of S-nitrosylated proteins (21, 22). The Trx-dependent denitrosylation requires a multi-step process that entails: (i) the formation of mixed disulfide bridges through the attack of the nucleophilic Cys (Cys32 in Trx1) on the sulfur atom of the SNO moiety, (ii) the release of HNO, (iii) the resolution of the mixed disulfide bridge through the action of the second reactive Cys (Cys35 in Trx1) so

(Prot-SNO), releasing Prot-SH and the oxidized Trx (Trx(*S*2)), which is further reduced back to (Trx SH 2 ) by Thioredoxin Reductase TrxR in a NADPH-dependent reaction. GSNOR-system (bottom): Prot-SNO is denitrosylated by *S*-transnitrosylation with glutathione (GSH), forming a reduced protein thiol (Prot-SH) and GSNO. The latter is then reduced by GSNOR to GSSG at the expenses of NADH and one GSH molecule. The resulting GSSG is further reduced by the Glutaredoxin enzyme (Grx) to give back the reduced GSH molecules—this step is not depicted here.

that the oxidized variant of Trx can be formed and (iv) the final reduction step by the action of TrxR (19). The knowledge of this mechanism was also exploited experimentally to entrap a protein target in its S-nitrosylated state mutating the resolving Cys of Trx (Cys35 in Trx1) (23, 24).

An alternative mechanism for Trx-mediated S-denitrosylation has also been proposed in which the initial step is postulated to be a S-transnitrosylation reaction from the target protein to the active site cysteine of Trx, with the subsequent release of HNO (24).

Another protein, i.e., the thioredoxin-interacting protein (TXNIP), regulates the Trx denitrosylating activity. In a feedback loop, NO controls this inhibition by repressing TXNIP activity, providing a dynamic regulation of Trx-mediated denitrosylation in response to NO levels (25).

Moreover, attesting the versatility of the Trx system, Trx1 has been identified as a S-transnitrosylase (26). Trx1, in the oxidized state, can be modified by S-nitrosylation on its Cys73, which in turn mediates the S-nitrosylation of caspase-3 and other protein targets (19). The SNO-Trx form can be then regulated by either the Trx system or a GSH-mediated denitrosylation (21). In agreement with the existence of an acid-base motif for S-nitrosylation (see section S-nitrosylation Sites Specificity), it has been shown that charged residues in the proximity of Cys73 are required for Trx transnitrosylase activity (27). The stability of SNO-Trx is regulated by both Trx- and GSH-mediated denitrosylation (21). Additional Trx1 S-transnitrosylation motifs have been proposed that involve proximal alanine residues (28).

A thioredoxin-related protein of 14 kDa called Trp14 was also recently identified to act as a denitrosylase for the other master regulator of S-nitrosylation, i.e., GSNOR, as well as certain SNOproteins, including caspase-3 and cathepsin B (24). Trp14 activity tightly depends on both TrxR1 and NADPH even if further studies are needed to better clarify Trp14 role in vivo.

Studies on targets of Trx-mediated denitrosylation allowed to identify two motifs, C-X5-K and C-X6-K, within the SNO targets that are modulated by the Trx system (28). These motifs could be used to predict and identify new targets where the Trx system is the major regulator of the S-nitrosylated state.

#### GSNOR System

GSNOR (i.e., the GSNO reductase) is present in all mammals and ubiquitously expressed across different tissues (29). It is known as a class III alcohol dehydrogenase and it is encoded by the ADH5 human gene. GSNOR is a homodimer composed by two identical subunits (chains A and B) containing two bound zinc ions each (30).

The GSNOR-mediated protein denitrosylation requires the tripeptide GSH to form a reduced protein thiol (Prot-SH) and GSNO (18). GSNO is reduced by GSNOR to GSSG in a NADHdependent reaction (using NADH as an electron donor) (31) involving hydride transfer. GSSG is further reduced by the Glutaredoxin enzyme (Grx) to give GSH back.

GSNOR expression and activity can be regulated by different biomolecules in a very context-dependent manner. For example, VEGF or IL-13 can induce GSNOR mRNA expression in lungs. Sp1 can transcriptionally regulate GSNOR levels in hepatocytes. MicroRNAs (miRs) such as miR-342-3p can downregulate GSNOR expression (4). Moreover, as also mentioned above, GSNOR is post-translationally regulated by S-nitrosylation and this modification has been suggested to regulate allosterically the enzyme activity, as attested by an enhanced GSNOR activity upon S-nitrosylation in mouse models (4).

The S-nitrosylated sites of GSNOR in vivo have not been identified yet and further studies will be required to fully understand and appreciate the mechanisms of this redox modification of the major regulator of S-nitrosylation. Prediction methods such as the ones described in section Prediction and Annotation of S-nitrosylated Proteins and the usage of molecular dynamics (MD) simulations could help in the identification of GSNO SNO-sites and in unveiling the determinants of its allosteric regulation. For example, the integration of computational techniques inspired by network theory and all-atom simulations hold promise to study longrange effects induced by PTMs (32–35). These methods could thus be translated to the study of GSNOR S-nitrosylation as soon as accurate physical models for S-nitrosylation will be available and validated.

As mentioned before, GSNOR has been found in different kind of tissues, particularly in liver, brain and kidney. GSNOR deficiency might positively or negatively affect physiology. GSNOR is the only ADH enzyme in the brain, highlighting its importance in this organ. As a result of its ubiquitous expression in different tissues, GSNOR controls several molecular processes and is likely to be involved in disease conditions onset.

#### GSNOR in Cancer

Nitric oxide (NO) regulates protein functions, as well as the activity of many enzymes. S-nitrosylation is a key mechanism in the transmission of NO-based cellular signals in vital cellular processes such as DNA repair (36), apoptosis (37), cell proliferation, and cell cycle regulation. These are all processes related to cancer onset.

As a key enzyme of denitrosylation, GSNOR controls the intracellular levels of S-nitrosylated proteins and the reduction of its expression or stability has been shown to result in dysfunctional S-nitrosylation signaling and, eventually, in pathological states such as cancer (38). In particular, GSNOR deregulation has been observed to be involved in some pathways representing the hallmarks of cancer like DNA damage repair, energetic metabolism and cell death. Noteworthy, GSNOR tumor suppressor role has been recently proposed (38).

It was demonstrated that GSNOR-deficiency is sufficient to induce spontaneous formation of hepatocellular carcinoma (HCC) (39) through S-nitrosylation and proteasomal degradation of the key DNA repair protein O(6)-alkylguanine-DNA alkyltransferase (AGT). The AGT enzyme removes the alkyl group from guanine bases, repairing the highly mutagenic and cytotoxic O6-alkylguanines which can be generated by carcinogenic compounds such as the diethylnitrosamine (DEN). Alkylation affects the stability of AGT resulting in its irreversible inactivation through degradation via proteasome. A similar effect is induced by S-nitrosylation of AGT on its Cys145; S-nitrosylated AGT is still degraded via the proteasome, but in this case before the repair of O6-alkylguanines. By GSNO catabolism, GSNOR maintains low levels of S-nitrosylated AGT. Therefore, GSNOR deficiency inactivates AGT-dependent DNA repair and may critically contribute to hepatocarcinogenesis (38).

Besides DNA alteration, tumor cells reorganize their core metabolism to sustain their growth and proliferation and GSNOR involvement in metabolic pathways has been observed. In particular, GSNOR deficiency in hepatocytes is characterized by mitochondrial alteration and by increases in succinate dehydrogenase (SDH) levels (40). Succinate dehydrogenase - or respiratory complex II - participates in both the citric acid cycle and the electron transport chain and catalyzes the oxidation of succinate to fumarate, regulating the levels of these two metabolites that are included in the class of "oncometabolites." More precisely, the mechanism through which GSNOR ablation modulates SDH involves the mitochondrial molecular chaperone TNF receptor-associated protein 1 (TRAP1) which, if Snitrosylated at Cys501, undergoes proteasomal degradation and is not able to interact with SDH, loosing the ability of inhibiting it (41). Besides, it was demonstrated that GSNOR ablation makes HCC cells more sensitive to SDH-targeting drugs (40), suggesting a new potential therapeutic target.

Different subtypes of breast cancers are also linked to the GSNOR expression modulation. In particular, the HER2 breast cancer subtype—characterized by high human epidermal growth factor receptor 2 (HER2) expression—is associated with lower expression of GSNOR which leads to a poor prognosis (42). Cañas et al. demonstrated that the antiproliferative effect of trastuzumab, a monoclonal antibody used for HER2 breast cancer, is suppressed by inhibition of GSNOR. Indeed, GSNOR restores the activation of survival signaling pathways, representing a possible reason for drug resistance observed in many patients on whom this treatment was used (43). Due to the central role of GSNOR in S-nitrosylation regulation and cancer, more insight will be gained in the future by comprehensive analyses of genomics and proteomics profiling of samples from cancer patients using biostatistics and bioinformatics to unveil the complex interplay between GSNOR, its regulators and targets in a cancer context.

# BIOLOGICAL MECHANISMS PROMOTED BY S-NITROSYLATION

As mentioned above, S-nitrosylation exerts a plethora of functions inside the cell. Indeed, S-nitrosylation gained attention as a PTM, but it is still less understood at the molecular level compared to other well-known PTMs. In 1992, Stamler et al. proposed for the first time that the formation of biologically active S-nitrosothiols—more stable than NO itself could represent an important mechanism through which NO is involved in the regulation of cellular activities (44).

In the context of tumor biology, we know that nitric oxide (NO) has different effects on cellular ability to survive and proliferate depending on its concentration. In fact, high concentrations of NO (>500 nM) appear to be toxic for cancer cells—causing cytostasis and apoptosis—whereas low concentrations (<100 nM) are able to induce the activation of cancer-promoting pathways (45). Thus, NO concentration depending on the balance between NOSs activity and denitrosylation (see section GSNOR System)—determines also the extent of S-nitrosylation inside the cell. Another important source of NO is nitrite (46), especially in ischemic conditions but it is not the main focus of this section. In this section, we will try to summarize the polyhedral/multifaceted consequences of this PTM on protein function (**Figure 3**).

# Influence of S-nitrosylation on Catalytic Activity of Enzymes

Probably one of the most common outcomes of S-nitrosylation is the regulation of catalytic activity of enzymes, whose modification can have activatory or inhibitory effects. In this way,

the cellular redox state—highly variable/deregulated in cancer cells—exerts its role in controlling enzymatic activity.

S-nitrosylation can alter well-known cancer-related proteins, such as kinases. It has been observed in breast cancer cells that estrogens are able to work in synergy with NO to induce their proliferation and migration. In fact, in MCF7 cells, βestradiol induces NOS expression and NO production, thus promoting the activation of c-Src through S-nitrosylation of Cys498 (47). c-Src is a tyrosine kinase, whose activation is responsible for disrupting E-cadherin junctions, promoting cell invasion and its catalytic activity is well-known to be regulated by phosphorylation. Recent evidence brought to light the existence of a tight regulation of c-Src based on S-nitrosylation and its cross-talk with phosphorylation (47).

The increase of iNOS (inducible NOS, also known as NOS2) is correlated with a decreased survival of ER negative and basallike breast cancer (BC) patients. Indeed, this correlates with high EGFR phosphorylation levels. A study from 2012 showed that, in the context of basal-like BC, S-nitrosylation can stimulate EGFR and Src proteins (both membrane-associated proteins). The Snitrosylated variants of EGFR and Src are then responsible for activating the oncogenic signaling based on c-Myc, Akt, STAT3, and β-catenin, while inhibiting the tumor suppressor PP2A (48). Moreover, NO levels required to activate these proteins are the same levels promoting an aggressive cellular phenotype (49).

More recently, it has been discovered that the NO-dependent EGFR activation can induce Extracellular signal-Regulated Kinase (ERK) phosphorylation—whose abnormal elevation has been described in tumor cells—thus activating it in basal-like triple negative breast cancer (50). ERK belongs to the MAPK superfamily and its aberrant upregulation and activation, which frequently occur in human tumors, are responsible for the acquisition of a malignant phenotype (51). In particular, there is a subset of basal-like BC, i.e., the BL2 molecular subtype, that is highly dependent on growth factor signaling (EGFR included) (52). In the BL2 subtype, Garrido et al. (50) demonstrated that NO-mediated activation of both EGFR and ERK was responsible for the increased migration and invasion abilities of cancer cells, this being accompanied by NF-kB activation and the increased secretion of pro-inflammatory cytokines.

ERK S-nitrosylation also links NO signaling to apoptosis. Indeed, ERK1 and ERK2 exert anti-apoptotic functions and are able, on the one hand, to induce the activity of other apoptosis antagonists, as for instance Bcl-2 and IAP. On the other hand, they can repress pro-apoptotic proteins, such as Bad and BIM (53). More in details, in a cellular model of NO-induced apoptosis, it has been demonstrated that NO decreases the levels of p-ERK, suggesting the S-nitrosylation of the kinase as an essential way to trigger cell death of tumor cells (54).

Recently, Gupta et al. (55) showed that S-nitrosylation plays a role also in the PI3K/Akt pathway, which is often dysregulated in cancer. In conditions of energy deprivation and in the presence of a signal able to activate the AMPK kinase, eNOS activation promotes inhibition of PTEN through its S-nitrosylation and degradation mediated by the ubiquitinproteasome system. Intriguingly, PTEN inactivation upon S-nitrosylation was originally identified in the context of neurodegenerative diseases (56), whereas Gupta and coworkers, for the first time, demonstrated the role of this redox-dependent modification in supporting proliferation and survival of cancer cells through the activation of PI3K/Akt signaling and the subsequent stimulation of mTOR activity (55).

Heat Shock Protein 90 (Hsp90)—the cytosolic molecular chaperone—represents a co-activator of eNOS associating to the NO synthase together with Akt. Indeed, to exert this role, Hsp90 needs to work as an ATPase, but this activity was discovered to be inhibited by S-nitrosylation, occurring on Cys597, localized in the C-terminal domain in the region interacting with eNOS. In fact, it was proposed that the PTM induces a conformational change able to disrupt the interaction between eNOS and Hsp90. This may represent a mechanism to react to overproduction of NO inside the cells (57, 58). The S-nitrosylation of Cys597 also regulates ATP hydrolysis and chaperone activity of Hsp90 and shifts the conformational equilibrium within the ATPase cycle (58). Future studies in which the SNO modification of Cys597 can be properly modeled, with the related conformational changes assessed, will provide more details on the mechanism and interplay between this redox modification and the activation of the chaperone.

The class of Ras GTPases—HRas, NRas, and KRas, i.e., the founding members of a large superfamily of small GTPases regulates several cytoplasmic signaling networks that govern cell growth, survival, and differentiation and act upstream to several pathways mentioned above (59). The three Ras proteins are over-activated by somatic mutations in 33% of human cancers, contributing to excessive growth, invasion, and ability to metastasize (60). The mutated variants of these proteins are associated with a constitutively active GTP-bound state, which makes cancer cells addicted to the expression of oncogenic Ras proteins. Some years ago, it has been also demonstrated that HRas undergoes S-nitrosylation on Cys118 with the effect of stabilizing the GTP-bound HRas form by enhancing the dissociation of guanine nucleotides (61). For this reason, it has been proposed that this PTM could represent a way to diversify the Ras-dependent oncogenic signaling beyond that of the mutated Ras.

In summary, many examples have been provided of regulation of enzymatic activity by S-nitrosylation, and they are especially important in the context of kinases, G-coupled proteins and chaperones i.e., three usual suspects in a cancer setting.

## Nuclear Translocation/Protein Redistribution

A large body of literature investigates and explains the multifaceted role of ion channels, both potassium and chloride, in the process of tumorigenesis. Indeed, they are involved in many of the main processes leading the transformation of a normal cell into a neoplastic one and, among all the ion channels, several chloride intracellular channels (CLICs) have been demonstrated to be overexpressed in different cancer types (62). Among the latter, CLIC4 has been originally identified as a p53- and cMycresponsive protein with proapoptotic functions, most of which are dependent on its translocation to the nucleus. Moreover, this channel undergoes structural conformational changes upon cellular redox changes (63, 64). In 2010, Malik et al. (65) have demonstrated that S-nitrosylation affects CLIC4 nuclear translocation, enhancing the association of the channel protein with two proteins responsible for the nuclear import, i.e., importin α and Ran, and thus increasing the nuclear levels of CLIC4. It has been proposed that through this mechanism CLIC4 would be able to induce apoptosis when NO cellular levels overcome the denitrosylating capability of the cell. Moreover, this would explain the mislocalization of CLIC4 in tumor cells where the redox balance is altered (65, 66).

S-nitrosylation can be responsible not only for the protein nuclear translocation but also for protein redistribution inside the cell. This is the case of the Tumor necrosis factor receptor superfamily member 6 (also known as Fas/APO-1/CD95), a cell surface receptor able to induce apoptosis when stimulated by the ligand FasL/CD95L or agonistic antibodies. After stimulation, the receptor recruits a number of proapoptotic factors - including caspase-8/10 and procaspase 8/10 to name a few—to assemble the death-inducing signaling complex (DISC) (67). Cancer cells evolved several ways to elude the possible apoptotic induction by CD95, for instance regulating the expression of the receptor (68) or inhibiting the interactions between the members of the DISC complex (69). A few years ago, it has been demonstrated that Fas undergoes S-nitrosylation at Cys199 in the transmembrane domain and Cys304 in the death domain, with both the events involved in determining Fas membrane localization, the latter also favoring Fas aggregation and translocation in plasma membrane lipid rafts (70, 71). As already mentioned, cancer cells can lose sensitivity to Fas-mediated apoptosis because of a decreased Fas expression, but S-nitrosylation (and inducers of this PTM) may be able to recover this sensitivity (70).

# Transcription Factors in NO-Dependent Signaling

As mentioned above, the Ras family of proteins—that owes its fame to the fact of being very often mutated in most of the cancer types—is a known target of S-nitrosylation (see section Influence of S-nitrosylation on Catalytic Activity of Enzymes). To fully appreciate the regulation of Ras family members and Ras-dependent tumorigenesis, another player needs to be considered, adding an extra layer of complexity to an already intricate scenario. In 2012 Ras was shown to activate Ets-1 transcriptional activity in human ER-negative (ER-) breast tumors (72). Ets-1 is a proto-oncoprotein member of the Ets family of transcription factors sharing a unique DNA binding domain. Being expressed in a large variety of cellular types, Ets-1 has a role both in physiological and pathological conditions but its role in carcinogenesis is due to its ability to regulate the expression of angiogenic and extracellular matrix remodeling factors promoting an invasive phenotype (73, 74). When performing the promoter region analysis of genes upregulated in ER- breast tumors showing high levels of NOS2 expression, Switzer et al. observed that the common denominator of the promoters of these genes was the presence of the Etsbinding sequence, pointing at the role of this transcription factor in the NOS2 (and thus NO)-dependent oncogenic signaling (72). In fact, Ets-1 activation following phosphorylation by MEK/ERK—in turn, activated by SNO-Ras—resulted in the expression of basal-like markers, as P-cadherin and S100A8 to name a few (75), as well as metastasis-related proteins such as CTSB and MMP-7 (76). Since breast tumors, differently from other cancer types, more rarely harbor Ras mutations (77), the Snitrosylation of Ras and thus the activation of Ets1 signaling axis may indeed explain the wild-type Ras-mediated tumorigenesis of cancers overexpressing NOS2.

## Protein Stability

S-nitrosylation can also directly affect protein stability and turnover in different ways, including cases in which protein stability and/or the propensity for proteasomal degradation is enhanced or reduced upon the S-nitrosylation of target cysteines.

An example is the caveolin-1 (Cav-1) protein that is enriched in 50–100 nm sized cell membrane invaginations with a structural role. These invaginations are implicated in many cellular pathways, such as endocytosis, lipid homeostasis, and signal transduction (78). Cav-1 has been described as having a controversial role in cancer development, with pro- and antitumorigenic effect depending on the context and the specific cancer type (79, 80). Interestingly, a relation between Cav-1 and NO has emerged in the context of anoikis - i.e., the detachmentinduced apoptosis—in lung cancer cells, where it has been demonstrated that Cav-1 is rapidly ubiquitinated and degraded by the proteasome after cell detachment and anoikis (81). In the same work, the authors observed that, upon exposure of the cell to NO donors, S-nitrosylation of Cav-1 was able to stabilize the protein—inhibiting its proteasomal degradation—pointing at a crucial role of the NO-mediated stabilization of Cav-1 in anoikis regulation.

Solid tumors are often characterized by a hypoxic (or even anoxic) microenvironment caused by scarce oxygen supply. Cancer cells activate a series of pathways promoting angiogenesis and other survival pathways to overcome hypoxia (82, 83). The master regulator of the activation response to low oxygen tensions is represented by the hypoxia-inducible factor 1 (HIF-1), a transcriptional complex made up of the constitutively expressed HIF-1β/ARNT and HIF-1α subunits, continuously synthesized, and degraded under normoxic conditions to be finally stabilized by hypoxic conditions (84). Because of this tangled role in cancer development, HIF-1 has been (and still is) at the heart of many scientific studies that have highlighted the existence of a multitude of post-transcriptional and post-translational mechanisms for regulating the activity of this protein (85). The degradation pathway of HIF-1α subunit in normoxic conditions starts with its hydroxylation by prolyl-hydroxylases (PHDs) in the oxygen-dependent degradation (ODD) domain. This event then targets the protein for ubiquitylation and the subsequent rapid degradation by the proteasome (86, 87). With regard to this, S-nitrosylation plays an active role in controlling the stability of the HIF-1α subunit. Indeed, in murine cancer cell line models, Li et al. (88) demonstrated that exposure to ionizing radiation stimulated NO generation in tumor-associated macrophages (TAMs). As a consequence, HIF-1α underwent S-nitrosylation on Cys533 (corresponding to the human Cys520), in turn inhibiting its binding with the E3 ubiquitin protein ligase part of the von Hippel-Lindau (vHL) tumor suppressor protein complex responsible of HIF-1 α degradation. This mechanism is of interest not only in relation to the already established role of HIF-1α in cancer therapy (89) but also for other diseases involving the immune system and inflammation (90).

We already mentioned the role of molecular chaperones, such as Hsp90, in cancer development and the regulation of its activity through S-nitrosylation (see section Influence of S-nitrosylation on Catalytic Activity of Enzymes). Hsp90 is highly conserved and with up to four different homologs in higher eukaryotes, such as the mitochondrial tumor necrosis factor receptor-associated protein 1 (TRAP1) (91). Recently, TRAP1 has been reported as a target of S-nitrosylation on its Cys501 with the effect of making the chaperone more prone to proteasomal degradation (40). In fact, levels of TRAP1 were reduced in human hepatocellular carcinoma cell lines depleted of S-nitrosoglutathione reductase (GSNOR), representing the best documented denitrosylase involved in regulating the levels of SNO-proteins in the cell (see section Enzymatic Regulation of S-nitrosylation). TRAP1 degradation was indicated being the causative event underlying the increase of succinate dehydrogenase (SDH) levels and activity. A few years ago it has been shown that TRAP1 is able to support cancer growth decreasing SDH activity, this leading to HIF-1α stabilization through the increase of succinate levels (41).

Recently, we and others showed that structural computational methods efficiently predict when mutations are likely to destabilize the protein in the context of disease-related mutations (including cancer) (92–95). Our measurements correlate with the rate of proteasomal degradation (95). In principle, similar approaches could be applied to predict how S-nitrosylation of Cys could alter protein stability and proteasomal degradation, as well as if the effect is likely to be structurally destabilizing or not. A major bottleneck in this context is the current lack of proper parameters for this uncanonical modification (see section Physical Models for S-nitrosothiols). Another computational approach useful to this scope could be the estimate of solvent accessibility of the ubiquitination sites for proteasomal degradation of the target proteins on structural ensembles collected by molecular simulations of S-nitrosylated and unmodified protein structures. A SNO-induced propensity to faster rates of proteasomal degradation might be also associated with higher solvent accessibility of the ubiquitination sites induced by the PTM.

## Protein-Protein Interactions

An additional way used by S-nitrosylation to regulate protein function is through the control of protein-protein interactions. In the context of cancer progression, integrins represent a family of cell adhesion receptors that are attractive targets due to their ability to regulate cell morphology, cell-cell interaction, and signal transduction in the extracellular matrix (ECM). These processes are potentially linked to all the stages of tumor development (initiation, progression and, metastasis) (96). In 2012, Isaac et al. have demonstrated that in prostate cancer (PCs) cell lines several cysteines of integrin α6 (ITGα6)—a subunit of integrin α6β1, usually overexpressed in PC cells and the corresponding lymph node metastases - are targets of S-nitrosylation. Particularly, Cys86 S-nitrosylation of ITGα6 enhances its binding to ITGβ1 (overexpressed in PC cells too), decreasing the extent of cell adhesion and potentially explaining the ability of iNOS and NO to promote migration of the cancer cells (97, 98).

In conclusion, despite the fact that we illustrated separately the different strategies used by S-nitrosylation to impact on the cellular proteome, it is important to bear in mind that many of the processes individually discussed are however highly interconnected, as well as they can occur for the same SNO target. The relation between S-nitrosylation and cancer can be seen as a tangled map of connections, many of which are still poorly defined.

# PREDICTION AND ANNOTATION OF S-NITROSYLATED PROTEINS

#### The Database of S-nitrosylated Proteins: dbSNO

Technical advances in mass spectrometry-based proteomics have improved the identification of S-nitrosylation sites (14, 99–103) and contributed overcoming the challenges due to the labile and highly dynamic nature of the thiol redox forms. This has been made possible by the availability of chemically-selective approaches to detect thiol redox modifications in concert with mass spectrometry-based proteomics (103).

These advances made now available a large list of experimentally identified SNO sites that require proper annotations and curations in publicly available repositories.

dbSNO2.0 (http://dbSNO.mbc.nctu.edu.tw) has been developed as a freely available resource to collect and explore the SNO sites of experimentally-identified S-nitrosylated proteins (104, 105). Moreover, regulatory networks are annotated for the S-nitrosylated proteins in the database, along with annotation of relevance in disease. dbSNO is the first manually curated repository for SNO peptides accounting for more than 4,000 SNO sites in almost 3,000 proteins. Annotation on targets with known 3D structure are also included together with information on solvent accessibility, neighboring residues and side-chain orientation for up to 298 substrate cysteine residues.

dbSNO2.0 can provide a useful framework for the redox biology community thanks to its availability of structural annotation for the computational and modeling studies on Snitrosylated proteins. The bottleneck will be a regular update and curation of this research. We could expect that the number of characterized SNO sites will grow at a fast rate if we consider that the computational analyses of annotated protein databases using predicted SNO motifs indicate that up to 70% of the proteome may be targeted by S-nitrosylation (see section Prediction of S-nitrosylated Targets and SNO Sites). In the future, the SNO community could benefit of more collaborative and interdisciplinary efforts to develop and maintain a broader upto-date repository and where also the experimental datasets from SNO-proteomics can be deposited for re-analysis, inspired by the efforts that the genomic community already established.

# Prediction of S-nitrosylated Targets and SNO Sites

In the last few years, as the topic of S-nitrosylation gained importance and recognition among researchers, several attempts have been made to predict S-nitrosylation sites on proteins based on experimental data as a reference, using machinelearning algorithms on protein sequence data. Nonetheless, predicting SNO-sites has proven challenging for a number of reasons. First of all, a simple sequence linear motif that clearly identifies S-nitrosylation sites does not exist, as detailed in section Computational Structural and Chemical Studies of Snitrosylation. The chemical-physical features of the environment surrounding the SNO site are crucial determinants for SNO specificity, and the pK<sup>a</sup> of the thiol moiety of a cysteine greatly influences its propensity to be S-nitrosylated. Accurate methods for the prediction of pK<sup>a</sup> from the 3D structure of proteins are currently available, such as PROPKA (106). Thus, the prediction of cysteine pK<sup>a</sup> would be in principle a viable alternative when the 3D structure is available, or, even better, when an ensemble is available and protein flexibility can be taken into account (107). However, an analysis of structural determinants for Snitrosylation showed that, at least in a limited data set of 55 proteins, features such as the acid–base motif flanking NO-Cys, hydrophobic content, predicted pK<sup>a</sup> and solvent accessibility do not distinguish SNO-sites from non-S-nitrosylable cysteines (108). This study also reports that the presence of a SNO-site is correlated with the presence in the surrounding area of an acid-base motif, possibly with different functions than the mere activation of Cys or stabilization of SNO, such as in facilitating protein-protein interactions that would, in turn, induce Snitrosylation. These results indicate that, while the local chemical environment around the SNO sites is certainly influencing reaction rates, predicting S-nitrosylation could benefit from a less reductionist approach.

Despite what just described, machine learning approaches tested so far have used protein sequences and sequence-derived features only; furthermore, even considering these, the set of experimental data on which machine algorithms can be trained is limited, at least respect to other PTMs such as phosphorylation, and typically unbalanced. This problem has been partially relieved by the publication of curated collections of SNO sites, such as the dbSNO database (105) and other manually curated datasets, as for example those by Xue et al. (109) and Li et al. (110). These resources do not alleviate the issue of using sequence data exclusively. Approaches based on protein structure would probably be more viable or at least include features that are more relevant to the problem at hand.

Given the limitations above, it is not surprising that the performance of SNO-site predictors has been so far consistently lower than predictors for other types of PTMs, with a Matthews correlation coefficient typically between 0.2 and 0.4. It should be emphasized, however, that no systematic review of performance on a common and independent test set has been performed so far. The fact that some of the methods have not been released as implementation or source code and that some of the web servers for SNO-site prediction are unreachable at the time of writing also makes fair comparisons difficult.

The different approaches tested so far for the prediction of SNO-sites are summarized in **Table 1**. Most of them relied on machine learning algorithms such as Support Vector Machines or (SVM) k-Nearest Neighbours (kNN) to classify Cys residues as S-nitrosylation sites or not. As feature space, all the methods considered amino-acid composition of the residues flanking the putative SNO-cysteine, often together with more structurally-oriented predictions such as physicochemical properties of amino acids, predicted secondary structure, predicted disorder propensity. This led, in turn, to a relatively high number of features among which the most discriminating and uncorrelated ones are selected, ranking them using methods as the minimal redundancy maximal relevance (mRMR) or the relative entropy selection. From a historical perspective, computational prediction of SNO sites has been tackled since 2006, when Hao et al. trained a support vector machines (SVM) classifier on a small dataset of SNO-proteins experimental data. The model turned out to barely outperform random chance (111). The first released algorithm designed to classify cysteine residues is named GPS-SNO (Group-based prediction system) (109), which uses an optimized scoring function to discriminate between SNO and not-SNO sites. Another prediction tool named SNOSite (112) was released in 2011, based on a collection of SVM models, each derived from sequences clustered according to a maximal dependence decomposition scheme to obtain statistically significant conserved motifs. CPR-SNO used a SVM



*<sup>a</sup>Link to the website not working at the time we wrote this review article.*

classifier in which different encoding schemes for the protein sequence flanking the potential SNO-Cys were tested (113). In 2012, Li et al. (110) used the minimal-redundancy maximalrelevance (mRMR) method to determine the importance of several features of the sequence flanking C-sites to predict SNOsites. The features included amino acid type, physicochemical features and conservation, disorder and secondary-structure propensity as well as solvent accessibility. Incremental feature selection was used to determine the set of features that would give the best performance. In 2013, two different predictors by Xu and colleagues were also released. The first, iSNO-PseACC (114) uses position-specific amino acid propensity into the form of pseudo amino acid composition with conditional random field models to predict the presence of SNO-sites, while the second, iSNO-AAPair (115) takes into account the coupling effects between residues close in sequence. In 2014, a paper published by Zhang et al. (116) used a scheme similar to Li et al. (110), in which a subset of features was calculated from the flanking sequence, feature selection was operated on them using relative entropy selection and incremental feature selection, while classification was done through the kNN algorithm. The final model was made available through the PSNO web server. In 2014 again, a paper by Huang et al. (117) used a similar approach to Li et al. (110) which also uses the same feature set, using the kernel sparse representation classification together with mRMR. Finally, another method has been implemented in iSNO-ANBPB and released in 2014, which uses an adapted normal distribution bi-profile Bayes (ANBPB) for feature extraction together with SVM (118).

#### COMPUTATIONAL STRUCTURAL AND CHEMICAL STUDIES OF S-NITROSYLATION

#### S-nitrosylation Sites Specificity

Many mechanisms can lead to the S-nitrosylation of proteins: reaction with NO2, S-transnitrosylation, thyil radical recombination, and transition metal catalyzed pathways (see also sections S-nitrosylation of Cysteine Residues and Enzymatic Regulation of S-nitrosylation). These different pathways do not necessarily take place in the same cellular and protein environment. Thus, it is especially difficult to describe the microenvironment properties driving the specificity of S-nitrosylation sites (4, 14, 119, 120).

Albeit no systematic environment features have been found to be predictive descriptors for this PTM, the structural microenvironment would favor the susceptibility of cysteines to react with NO or undergo S-transnitrosylation. However, one has to keep in mind that these features could also favor cysteine reactivity toward other PTMs, hence it is really difficult to depict a precise panel of conditions that have to be fulfilled for S-nitrosylation to take place (4, 119).

Intensive investigations on the subject led to the conclusion that several environment features might have an influence on the specificity of cysteine targeting mechanisms associated to Snitrosylation. The latter are certain characteristics that are subtler than a simple primary sequence effect, so that no discernable consensus sequence to accurately predict Cys SNO sites has been found so far (14, 28, 111, 121). In the 90's, Stamler and coworkers were the first to report the role of basic and acidic side-chains in the proximity of a cysteine in promoting Snitrosylation, highlighted by their work on the S-nitrosylated hemoglobin protein (122, 123). In Stamler's model, a histidine residue proximal to the Cysβ93 in the oxygenated R-state of hemoglobin facilitates a base-catalyzed S-nitrosylation whereas a proximal aspartic acid in the deoxygenated T-state favors acid-catalyzed denitrosylation, coupling the oxygenation status of the hemoglobin to the SNO formation and release during the respiratory cycle. In another work, the acid-base motif in SNO-proteins has been expanded to include residues −6 and +6 from the target Cys and the notion of hydrophobic regions surrounding the SNO-Cys (121).

Acid-base motifs have been also reported at the tertiary structure level, i.e., not as short-linear motifs in the sequence space. In this case, acidic and basic residues in the proximity of the SNO site are contributed by different regions of the protein that are brought together in the 3D structure, as first found in the methionine adenosyltransferase (MAT) protein (124). The Snitrosylation of the Cys121 of MAT is indeed promoted by two arginines which are distal in the primary sequence but close in the tertiary structure and lower the Cys pKa. Similarly, 3D acid-base motifs have been found in the case of caspase-3 and aquaporin-1 proteins (12).

Since these first works on hemoglobin and MAT, several data consolidated the hypothesis of a 3D acid-base motif (12, 108, 125–129). Overall, the charged acidic and basic side chains in proximity of the SNO site regulate S-nitrosylation by influencing both thiol pKa (and as a consequence their reactivity) and SNO stability. As a general rule, the presence of an acid-base motif within 8Å of the target cysteine could facilitate interactions that make it to S-nitrosylation—e.g., protein-protein interactions in the case of S-transnitrosylation (130–132).

Besides the acid-base motif, other structural characteristics appeared to drive the specificity of cysteine reactivity toward CysNO. For instance, the presence of hydrophobic residues in the surroundings of the S-nitrosylation sites has also been proposed. As a matter of fact, S-transnitrosylation and NO oxidation reactions may preferentially take place on cysteine residues located in a hydrophobic pocket because of their low pK<sup>a</sup> that, in turn, make them prone to react (10, 131, 133– 138). Hydrophobic environments favoring S-nitrosylation are not limited to hydrophobic residues in the protein target. Indeed, they can also be contributed by membranes or interactions with other proteins (12).

This specific environment may also favor interactions with transition metal nitrosyls (127, 128, 139), as well as stabilize radical nitrosating species in the case of thiyl radical recombination (128, 140). Interestingly, studies suggested that acid-base motifs would be correlated with location of the target cysteines on α-helices, whereas hydrophobic environments would be more associated with S-nitrosylation on Cys sites located in β-sheets (130). Further factors that govern S-nitrosylation specificity are steric hindrance and solvent accessibility, that have also been proven to be key-features for S-nitrosylation promotion (3, 127). These two determinants are especially important in the case of S-transnitrosylation, in which protein-protein interactions are crucial. However, bioinformatic investigations revealed that, surprisingly, around 50% of the Snitrosylation and S-glutathionylation sites are actually deeply buried in the native structure of the protein (141). Several studies revealed that buried cysteines may still be susceptible to undergo S-nitrosylation when surrounded by hydrophobic residues, by channeling of NO to target thiols (6, 10, 131, 133– 138). This might be the case of the recent buried SNO site discovered in c-Src kinase where S-nitrosylation has a clear effect on protein activity and relevance in a cancer context (see section Computational Structural and Chemical Studies of S-nitrosylation). NO oxidation might also be favored upon co-localization of the target protein with NOS isoforms (3, 119), indeed enhancement of S-nitrosylation has been observed especially in the presence of surrounding eNOS in the Golgi apparatus (17, 142, 143). Compartmentalization of NO is induced by its quick consumption after production (<0.1 s) through reaction with molecules that colocalize with NOS (17). Furthermore, several NOS and NOS-interacting proteins are well-known to undergo S-nitrosylation—e.g., S100A8/A9, Dlg-4 and Cav-1. Thus, co-compartmentalization of the target protein with NOS could be considered as one of the many factors driving S-nitrosylation specificity, with the nitrosyl group being transported to a distant cellular zone through Stransnitrosylation.

In the analysis of cysteine SNO sites, we should also take into account the capability of the cysteine side chain to be involved in hydrogen bonds and how the substitution of the SH group with SNO can affect the native hydrogen bond network. Cys can serve as a hydrogen bond (HB) donor when protonated (SH) as well as a HB acceptor in both protonated and deprotonated states (S−) (144).

Hence, acid-base motif and hydrophobic pocket proximity, appropriate redox potential, solvent accessibility, steric hindrance and colocalization with NOS have been highlighted as significant factors driving S-nitrosylation site specificity. Based on these considerations, machine learning techniques have been used in order to investigate S-nitrosylation sites. Several web servers have been designed to predict cysteines susceptible to be S-nitrosylated based on the protein sequence, and the dbSNO database is also available online to probe CysNO environment in PDB entries and predict potential SNO-proteins that may play a role in NO signal regulation in cancer cells - see details in section Prediction and Annotation of S-nitrosylated proteins.

## SNO-Induced Long Range and Allosteric Effects

Little is known about the potential distal effects induced by S-nitrosylation. Nevertheless, allostery has been proposed to be an important factor for the promotion of Snitrosylation/denitrosylation (128), and allosteric effects due to S-nitrosylation have been observed to play an important role in the regulation of the activity of several proteins (97, 145–147). Allosteric SNO-induced effects have been also postulated for the denitrosylase GSNOR, pointing out an intriguing feedback regulatory mechanism of NO signaling (see sections GSNOR System and GSNOR in Cancer).

Another striking example is the inactivation of the inducible nitric-oxide synthase (iNOS) upon auto-S-nitrosylation. NOSs (iNOS, eNOS, and nNOS) are well known to work as homodimers (148), with—at the interface between the two monomers—a zinc atom coordinated to four cysteines in a conserved ZnS<sup>4</sup> motif (149, 150). The latter is of utmost importance for the protein-protein interactions and the integrity of the homodimer, hence its activity. Therefore, the regulatory effects linked to the allosteric disruption of the interactions at the dimer interface have been extensively investigated for the design of new NOSs ZnS4-related therapeutics (151–155). Interestingly, this site is specifically S-nitrosylated at the Cys109 position in iNOS. The latter induces the release of Zn2+, coupled to a strong destabilization of the dimer, resulting in its disruption and iNOS inactivation (145). Upon high NO concentration, Snitrosylation of iNOS limits its activity (i.e., NO production) through an allosteric mechanism driving the dimer/monomer balance (145, 156). Thus, the Cys109 specific S-nitrosylation plays an important role in the regulation of iNOS activity,

whose dysfunction has been shown to play an important role in tumor growth in several cancer types (157). Noteworthy, eNOS dimer dissociation upon treatment with NO donors has also been reported (158), suggesting a similar equivalent autoinhibitory mechanism. Similar structural impact, although less pronounced, has been reported for the NMDA receptor, for which S-nitrosylation has been suggested to allosterically regulate the ligand-binding, regulatory and linker regions (146, 159, 160). Likewise, the cyclic nucleotide-gated channels (CNGcs), a class of ionic channels that permeate cations and are especially important in sensory-receptor cells, might be allosterically regulated by Snitrosylation of a specific cysteine located in their ligand-binding region (147).

Besides, relatively small distal effects of S-nitrosylation have been observed in certain X-ray structures of SNO-proteins (161– 164). This behavior nicely fits in the context of allosteric effects that can occur without a marked change in protein shape, as attested in many other biological cases (33, 165). For instance, hemoglobin (Hb) S-nitrosylation leads to local reorganization but no large changes in the quaternary structure of the tetramer in the crystals (162). However, the authors suggested that larger effects might occur in solution, with a shift in the balance between Hb R and R2 states. In a similar fashion, S100A1 S-nitrosylation leads to distal reorganizations of the linker region and the two helices III and IV from the C-terminal EF-hand, that are known to be important for target recognition (164).

Nevertheless, the paucity of experimental structures of SNOproteins does not allow to efficiently probe the distal effects that could be induced by S-nitrosylation. In this context, methods based on MD simulations using reliable force field parameters for Cys-NO would constitute a considerable asset to explore such reorganizations. Indeed, many methods to study distal conformational changes, including methods based on MDderived ensemble, have been proposed and successfully applied to the study of long-range effects induced by protein PTMs such as phosphorylation or other perturbation of the protein native structure (32–35, 166–168) and they can be naturally translated to the study of allosteric effects promoted by S-nitrosylation.

## A Case Study: Effects of S-nitrosylation on Src Kinases

Kinases are usual suspects in the context of cancer research (169– 171). Intriguingly, S-nitrosylation has been recently pointed out as regulatory mechanisms for kinases (141, 172), which are enzymes with the main regulatory role for another PTM, i.e., phosphorylation, highlighting an interesting cross-talk between different PTM cellular signals.

As mentioned in the previous sections, Src kinases are a family of kinases responsible for cellular proliferation, differentiation and survival (172). Disregulation of Src kinases has been linked to different cancer types. They are multi-domain proteins, including four different domains of which one carries out is the catalytic activity. Two tyrosines (Tyr416 and 527) are regulated by phosphorylation and their phosphorylation is responsible of either activation or deactivation of the kinase, respectively.

Apart from phosphorylation, another layer of posttranslational regulation—relying on S-nitrosylation—has been demonstrated for the c-Src kinase. The SNO site is the Cys498, which is one of the nine cysteines of the human Src kinase (47). c-Src is known to promote cancer cell invasion and metastasis and its S-nitrosylation enhances the protein activity but the structural mechanisms behind this have been poorly understood. Cys498 is also conserved in other kinases of the Src family, suggesting a common mechanism, as attested by the fact that the c-Yes kinase activation is also mediated by NO (47). These data overall link NO-dependent activation of c-Src to cancer cell invasion and metastasis but the structural and molecular details are still elusive.

In a recent computational work, Rando and coworkers applied different computational structural analyses accounting for electrostatic, steric and hydrophobic properties to compare the Cys498 selective SNO site with the other three c-Src cysteines that are not affected by S-nitrosylation. Their data pointed in the direction of a rather buried and highly nucleophilic Cys with a highly hydrophobic environment in which NO can be more prone to undergo decomposition into the electrophilic intermediates (173).

As a future direction, structural studies to assess conformational changes in the S-nitrosylated and non-Snitrosylated protein could shed new light on the NO-mediated activation of c-Src and other similar kinases, as well as properly assess the accessibility of the SNO site in the native conformational ensemble of the protein. Also, in this context, the main limitation is related to the poor availability of force field parameters to describe SNO proteins.

#### Physical Models for S-nitrosothiols (RSNOs)

The impact of S-nitrosylation on protein structure, function and stability can be different from one protein to the other (see section Biological Mechanisms Promoted by S-nitrosylation). Therefore, one could not describe a systematic structural and reactivity behavior of S-nitrosylated proteins. There is a real need of investigations by both experimental and theoretical means, which represent a colossal yet of utmost importance work with the aim at gaining knowledge about the important phenomena driving RSNOs formation and reactivity.

S-nitrosothiols exhibit a highly complex chemistry, which represents a real challenge for theoretical studies. In the last decade, efforts have been dedicated to the development of an accurate structural and electronic description of the -SNO group, by both experimental and theoretical means (174–176). Concerning theoretical investigations, a certain amount of highlevel quantum chemistry studies has been performed on small RSNOs (mainly with R=H,Me). The main part of the theoretical studies focused on the description of the thionitrous acid HSNO, the smallest RSNO (128, 143, 177–180). Indeed, the high complexity of the -SNO moiety requires the use of timedemanding calculations. Thus, the size of the system is rapidly limited by the computational resources. The nature of the S-N bond is especially challenging to investigate, and lots of efforts have been made to gain information about the properties of this bond. Values of the bond dissociation energy (BDE) and the activation energy corresponding to the transition between the cis and trans isomers have been computed using diverse levels of theory (181–184). One of the most recent computational studies using high level CCSD(T) coupled to CBS extrapolation methods (179) suggested that the cis - trans interconversion requires an activation energy up to ∼9 kcal/mol, with the cis isomer slightly less stable than the trans one, roughly by 0.1 kcal/mol. Investigations on larger RSNOs moieties have shown that the nature of R may have a strong influence on RSNO chemistry, hence CysNO structure might exhibit different features than HSNO (128, 185–189). However, there is still a lack of theoretical studies about CysNO properties, with only few investigations being reported concerning models structurally closer to the Snitrosylated cysteine than HSNO.

Electronic properties of the -SNO group have been highly investigated by computational means. Especially, a lot of work is available on HSNO structure description by high-level QM methods and even Car-Parrinello metadynamics (179, 181, 184, 190–192), which shed light on the multi-reference character of the -SNO group. Indeed, the complexity of the S-N bond nature might result from the combination of three different resonance structures: the neutral -S-N=O, the zwitterionic - S <sup>+</sup> =N-O−, and the RS−/NO<sup>+</sup> ion pair. Further investigations on the larger CH3SNO model confirmed this feature, yet it seems less pronounced for CH3SNO than HSNO (186). Unfortunately, no further evidences about RSNOs multi-reference character has been reported for larger models of RSNOs, the computational cost of such high-level calculations being prohibitive.

Nevertheless, these investigations underlined the difficulty to obtain an accurate description of S-nitrosothiols structure, with the S-N bond exhibiting very complex chemical properties. Hence, one should pay a particular attention while dealing with such systems, by starting with a careful choice of the level of theory.

The unusual electronic structure of RSNO and its multireference character makes it a difficult system to accurately model. Hence, high level ab initio QM methods should be used in order to obtain a reliable and accurate description of the complex electronic density of the SNO moiety. However, the use of such quantum calculations is computationally very demanding when it comes to study models larger than HSNO or CH3SNO. Thus, benchmark studies have been led to assess the capacity of the less time-consuming DFT and TD-DFT methods to reproduce structural and electronic features obtained by high level ab initio methods and experiments. Overall, several of these theoretical investigations highlighted the reliability and robustness of the B3P86 functional with large basis sets for the calculation of S-N bond energy dissociation, spectroscopic and structural properties of small RSNO (182, 189, 193). Likewise, the PBE0 functional with large basis sets has also been revealed as a good compromise between computational time and accuracy in describing RSNO properties (179, 186, 188, 194). However, it is always strongly recommended to verify results obtained by DFT methods using more computationally demanding higher-level methods (186). Likewise, experimental data on RSNOs characteristics, mainly from NMR studies, have been reported and can validate values predicted by computational chemistry (44, 195–197).

The electronic and structural properties of RSNO are highly modulated by the molecular environment, especially in the presence of proximal charges (e.g., upon coordination to metal ions), with large fluctuations of the S-N bond stability being observed in several experimental and theoretical works (182, 185, 188, 198, 199). Studies of RSNO interacting with metals especially highlighted the dramatic influence of surroundings on the S-N bond nature. For instance, coordination to Cu<sup>I</sup> , which is known to play important roles in NO-release regulation by catalyzing the decomposition of RSNOs, tends to weaken the S-N bond upon Scoordination while N-coordination was predicted to strengthen it (182, 200).

An interesting property of the -SNO moiety is the multireference character induced by the dramatic difference between its resonance structures (**Figure 4)**, as highlighted in several theoretical works published by Timerghazin's group (184–186, 188, 201). Indeed, the use of an external electric field (EEF) in QM calculations brought out the high polarizability of the– SNO moiety, with modulation between two minor resonance structures, which exhibit opposite charge distribution and reactivity. The first one exhibits an ionic structure, with the charge located mainly on the sulfur atom, interacting with the electrophilic NO<sup>+</sup> moiety through a long and weak S– N bond. Upon opposite polarization of the EEF, the -SNO adopts a totally different configuration, with the charge located on the NO double bond, the sulfur atom being in this case prone to undergo nucleophilic attacks. The balance between

the zwitterionic, neutral, or ionic conformations might thus be highly influenced by the electric field induced by the -SNO chemical environment and lead to a fluctuating reactivity of this moiety (185). The coexistence of such antagonist structures brings explanation to the contradictory observations that have been published concerning RSNO structure and reactivity (193, 202–205).

Nowadays the computational resources available do not allow to model the dynamics of an entire protein at the QM level of theory. Thus, investigations about the S-nitrosylated cysteine electronic structure have been performed on reduced models only. The largest one reported so far being the portion of an α-helix S-nitrosylated in silico, on which Talipov et al. performed QM/MM calculations to probe the effect of proximal charged amino acids on CysNO electronic structure (188). Nevertheless, QM/MM calculations are way too time-consuming to be systematically used to investigate influence of the protein environment in realistic models, since the dynamics of the system might also have a strong influence on the cysteine environment. Furthermore, Talipov et al. suggested that even small allosteric effects could induce dramatic effects on the CysNO reactivity. Thus, the dynamics of the system should be taken into account when studying the S-nitrosylated cysteine behavior in a protein environment. The use of all-atom MM-MD simulations can help toward this goal. However, attention should be devoted to the choice of the force field parameters used to describe the highly complex S-nitrosylated cysteine structure.

So far, AMBER and GROMOS force field parameters have been developed to describe CysNO and few case studies of proteins harboring CysNO by classical molecular dynamics simulations have been reported. On the one hand, the AMBER parameters developed by Han et al. have been generated using quantum mechanics (206) and validated using as a model system a S-nitrosylated thioredoxin crystal structure on Cys69 (207). On the other hand, Petrov et al. (208) parameterized CysNO by deriving GROMOS force field values. Moreover, online tools have been developed to insert PTMs in silico using GROMOS force field parameters, including the cysteine S-nitrosylation: the Vienna-PTM and the Automated Topology Builder (ATB) servers (209–211). Nevertheless, an in-depth look into the values of these two different sets of parameters highlights not negligible differences (**Tables 2**, **3**). Thus, these parameters have to be extensively validated against experimental data to assess their efficiency in describing a such complex chemical structure. The major bottleneck in this field is related to the fact that only a minority of experimental structures is available by NMR and X-Ray crystallography (161–164, 207, 212–216) and classical MD simulations using reliable force fields usually provide predictive data about structural and dynamical behavior of biosystems. However, in the CysNO case, it would be difficult to model with standard force fields using point charges, given that the polarization of the—SNO moiety is likely to undergo a dramatic variation of the reactivity depending on its micro-environment. To overcome this issue, one might consider to use polarizable force fields, which are undergoing marked improvements (217).

Overall, investigations that have been led so far on small RSNO models provide a solid basis toward the understanding TABLE 2 | Comparison between GROMOS and AMBER parameters for *S*-nitrosylated cysteine with regards to atomic charges.


of S-nitrosylated cysteines structural properties within a protein environment. The system size and the simulation time scale are known to be very rapidly prohibitive while using QM and hybrid QM/MM (-MD) methods. Besides, classical MD simulations are nowadays a method of choice for the theoretical investigation of dynamics of proteins and macro-molecules in general (32, 33, 218), but its reliability relies on the force field accuracy. Hence, a particular effort should be realized in order to bypass limitations of standard force fields and find a sustainable solution in order to unravel the complex mechanisms underlying proteins S-nitrosylation.

# REACTIVITY OF S-NITROSOTHIOLS

The unusual structure of S-nitrosothiols (RSNOs), balancing between three different mesomeric forms, implies a complex chemistry of these moieties. The biological relevance of RSNOs relies on their important role in NO storage and transport, as well as their function as HNO donors in vivo (219–224). As RSNOs are relevant to a broad spectrum of diseases, ranging from asthma to cancer (6, 8, 140, 225–227), extensive works have been reported with the final goal of developing new SNO-related therapeutical strategies. Efforts have been made especially for the design of new RSNO-inspired NO-releasing biomaterials (228–230). In this framework, a deep knowledge of the mechanisms driving the RSNOs reactivity is of utmost importance for the design of therapeutics, with implications for the treatment of a large range of diseases. Theoretical investigations have been led in order to unveil the electronic mechanisms ruling RSNO denitrosylation. The latter can take place through several reaction pathways: Stransnitrosylation, disulfide bridges formation (S-thiolation), and homolytic cleavage of the S-N bond—see **Figure 5**.

#### RSNO-Thiol Interactions

Albeit the S-transnitrosylation and S-thiolation have been studied extensively by experimental means (22, 26, 28, 131, 231, 232), there is only a little amount of theoretical data available. A proper electronic description of the S-nitrosylated cysteine requires the use of high-level quantum methods, whereas the current computational resources do not allow to study such a reaction in a protein model, meaning that current studies are limited to small RSNO models. Moreover, only one work has been published about the electronic mechanisms driving Sthiolation by Ivanova et al. using MeSNO and MeSH models (194). According to their DFT calculations on this simplified model system, the reaction takes place in three phases. First, the thiol proton is transferred to the—SNO nitrogen atom,


TABLE 3 | Comparison between GROMOS and AMBER force field (FF) parameters for *S*-nitrosylated cysteine (i.e., bonds, angles, dihedral angles).

leading to a thiolate MeS<sup>−</sup> and the MeSNHO<sup>+</sup> moiety. The latter undergoes a strong delocalization of the electronic density on the oxygen atom, increasing the electrophilicity of the sulfur group. Meanwhile, the thiol sulfur moves out of plane to initiate the nucleophilic attack on the -SNO sulfur atom. The second phase is the formation of the S-S bond, which results in a zwitterionic species involving a tri-coordinated sulfur. The third and last phase consists in the release of HNO following the S-N bond cleavage. A similar reaction mechanism has been proposed by Moran et al. (233) for the hydrolysis of O-protonated RSNO.

The addition of explicit water molecules in the model system highlighted the possibility of a reaction assistance, with an activation energy decrease up to 20 kcal/mol. The water molecules participate to the proton transfer, but also have a stabilizing effect on the charge-separated intermediates. This stabilization is even more pronounced when moving from gas phase to a polarizable aqueous environment by using an implicit solvent model, suggesting that the environment provided by the protein embedding might also offer a similar stabilization. This hypothesis is supported by investigations highlighting the effect of external electric fields on the RSNO electronic density, that can lead to its dramatic polarization (185, 186, 188). Noteworthy, the biologically relevant competition between the S-thiolation and the S-transnitrosylation processes might then be driven by the polarization of the -SNO by its surroundings. The ionic conformation would favor thiolate attack on the NO<sup>+</sup> (i.e., S-transnitrosylation) while its antagonist structure might be prone to undergo nucleophilic attack of thiolate on the sulfur atom with release of a HNO molecule—i.e., S-thiolation and disulfide bridges formation (188). Very recently, a work by Wolhuter et al. suggested that S-nitrosothiols could be reactive intermediates leading to disulfide bridges formation and Sthiolation (234). Their work highlighted the strong propensity of S-nitrosothiols to induce disulfide bond formation, pointing out the larger instability of S-nitrosothiols compared to disulfides. According to their results, S-nitrosothiols might be only transient

moieties favoring the formation of S-S bonds, rather than the putative stable redox regulators. The study of cysteines nitrosothiols reactivity, involved in the promotion of disulfide bridges formation, is of major importance for the understanding of the complex mechanisms ruling the cell redox regulation. We thus envision that further theoretical and experimental efforts are still necessary to strengthen our knowledge in this area.

#### Homolytic Cleavage of RSNO

Theoretical and experimental investigations about the Snitrosothiols S-N bond dissociation stressed out the ease of this reaction, both thermal, and photochemical decomposition happening at room temperature (199, 235–240). The latter reaction induces the homolytic loss of NO, eventually leading to the formation of a stable S-S bond, which involves the reaction between two newly formed thyil moieties. The weakness of the S-N bond (values from 15 to 35 kcal/mol) (184, 186, 189, 193, 203, 219, 241) may favor its rapid decomposition, although the S-nitrosothiol chemical environment can strongly enhance its half-life (198, 199, 204).

Recent theoretical work about the S-N photo-dissociation in model compounds (small RSNOs) shed light on the barrierless character of the process upon irradiation in the visible and UV regions as well as upon exogenous photosensitization, with population of S1, S2, and T<sup>1</sup> states respectively (189). Experimental investigations showed that irradiation of Snitrosothiols at UV-visible wavelengths induces the release of nitric oxide and a thyil radical (236, 242). Likewise, photodecomposition of RSNOs has been observed in human skin upon UVA irradiation (243). Similar wavelengths have been reported for light-induced release of NO from NO-metal complexes (mostly iron, but also Mb, Co, or Ru among others), which has been widely investigated for its biological relevance and has been observed in the UV-vis-NIR regions depending on the nature of the complex (244–247).

Thermal decomposition of small RSNOs has also been investigated using high-level computational methods (184, 186, 201). In the most recent of them, Khomyakov and Timerghazin underlined the difficulty to obtain accurate values of the S-N bond dissociation energy, the convergence of the S-N bond features being especially slow due to the unusual, multireference character of the -SNO moiety (186). Their calculations suggested a value of 32.2 kcal/mol for the CH3SNO model S-N bond cleavage in implicit aqueous solvent, with a very small stabilization of energy when shifting from water to the nonpolar diethylether solvent—often used to mimic the protein environment. Noteworthy, metal ions and especially copper ions can efficiently catalyze the S-N bond thermo-dissociation, making it an ultrafast process (182, 236, 248, 249).

NO transport and storage are mainly provided by ironnitrosyl (e.g., in myoglobin and hemoglobin) and nitrosothiols compounds. These processes are known to be of utmost importance for the modulation of important cellular mechanisms such as the mitochondrial respiration (250) or vasodilation/cardioprotection (251). As mentioned previously in this review, nitric oxide is also known to have a multifaceted role in cancer biology, with both tumor-suppressing and tumorpromoting effects reported (252, 253). The redox chemistry of S-nitrosothiols is as rich as complex and is related to a large variety of biochemical processes. It is notably balanced by pH, temperature, UV-vis irradiations, RSNO micro-environment (influencing its pKa), chemical nature of RSNO, and presence of reactive compounds such as metal ions (254, 255). The dysregulation of NO flux in the tumor micro-environment has also been found to modulate the redox signaling pathways during cancer progression (256), which might consequently influence RSNOs reactivity. Several compounds are known to react with S-nitrosothiols such as phosphine derivatives, sulfenic acids, and a large variety of nucleophiles. Some of them have biological relevance (e.g., thiols and seleno compounds), but an important aspect of this broad reactivity is to provide perspectives for the development of efficient RSNOs detection methods and RSNObased therapies—e.g., biotin labeling, phosphine compounds, and metal complexes (257–261). Noteworthy, low-molecular weight S-nitrosothiols derivatives such as S-nitroso-N-acetyl penicillamine (SNAP), GSNO and L-/D-CysNO are commonly used in biological experiments. For an in-depth description of RSNOs chemistry, very good articles, and reviews are available in the literature (254, 262, 263).

Considering their high biological relevance, mechanisms driving RSNO chemistry are matter of intensive investigations, and the ease of RSNOs dissociation through several pathways at room temperature, leading to the efficient release of nitric oxide, is an interesting property very often used in studies aiming at developing therapies against cancer and other diseases (264–271).

# CONCLUSIONS

The discovery of S-nitrosylation opened new venues in the context of the cellular signaling induced by post-translational modifications since NO relies on this modification to transmit its redox signaling. The mechanisms involved in its regulation and the effects caused by it are very complex and diverse, as described above. S-nitrosylation has been emerging also as a key mechanism in many diseases, such as cancer. As a result of extensive efforts by cellular and proteomics studies, we now know several enzymatic regulators of S-nitrosylation, as well as a myriad of protein targets that are modulated by this post-translational modification. However, structural studies that can help in understanding the mechanisms induced by this modification and its reactivity are still not as in-depth as the investigations carried out so far for other more conventional PTMs, such as phosphorylation. The complexity of this redox modification challenges experimental and computational structural and biophysical studies. Nevertheless, advances in computational biochemistry hold promise to both generate new mechanistic hypotheses that can be experimentally tested and rationalize at the molecular and atom level the experimental results collected so far. Several efforts are still required to experimentally solve new structures of S-nitrosylated proteins, using techniques such as X-ray crystallography and NMR, as well as to collect experimental data probes the conformational changes induced by this PTM on both structure and dynamics at local and distal sites of the proteins. Once more information will be available, we will be capable of overcoming the limitations of standard force fields to model the effects induced by Snitrosylation on protein structure and dynamics, as well as to unveil the long-range allosteric effects triggered by this redox PTM. Similar studies can, for example, shed new light in the context of identifying and characterizing SNO sites that are buried in the native structure and that can become available for modification upon transient conformational changes of the protein. Moreover, in the context of reactivity, theoretical studies will need to account for more complete models that can account for the structural environment of the S-nitrosylated cysteine. Prediction algorithms for SNO-sites would also largely benefit from more available experimental and structural data, hopefully making the current predictions more accurate. The redox community should also dedicate more comprehensive, collaborative, and organized efforts toward the development of a common publicly available repository for the S-nitrosyloma with both sequence, structural and experimental information.

# AUTHOR CONTRIBUTIONS

ML contributed to the writing of chapter 3, MA to the writing of chapter 4, EB contributed to the writing of chapter 6 and 7 and MT to the writing of chapter 5. EP wrote the introduction, chapter 2, conclusions and contributed to the writing of all the other chapters. All the authors revised the final version of the manuscript and contributed with figures and tables.

#### REFERENCES


#### FUNDING

Our research in the context of redox modifications of protein is supported by an Alfred Benzon Principal Investigator Fellowship 2017–2019 to EP and EB, a PRACE-DECI-13th grant CHARPEDO (for computational access to the UK HPC infrastructure ARCHER), an ISCRA-LISA HPC grant RedoxC-HPL13PEJU2 (for calculations on Marconi) and a DeiC Pilot grant 2015– 2016 (for calculations on the Danish HPC Infrastructure Computerome).


death in parkinson's disease models. J Neurosci. (2014) 34:15123–31. doi: 10.1523/JNEUROSCI.4751-13.2014


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

The reviewer MS and handling Editor declared their shared affiliation.

Copyright © 2018 Bignon, Allega, Lucchetta, Tiberti and Papaleo. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Role, Targets and Regulation of (de)nitrosylation in Malignancy

Salvatore Rizza<sup>1</sup> \* and Giuseppe Filomeni 1,2 \*

*<sup>1</sup> Redox Signaling and Oxidative Stress Research Group, Cell Stress and Survival Unit, Center for Autophagy, Recycling and Disease, Danish Cancer Society Research Center, Copenhagen, Denmark, <sup>2</sup> Department of Biology, University of Rome Tor Vergata, Rome, Italy*

Keywords: ADH5, GSNOR, nitric oxide, NOS, nitrosylation, cancer, FAK1, HIF-1α

# NITRIC OXIDE INVOLVEMENT IN CANCER

Nitric oxide (NO) is a free radical that can target cellular biomolecules directly, or by means of the activity of its metabolites (RNS) generated upon reaction with transition metals (e.g., NO+), oxygen (e.g., N2O3), or superoxide (ONOO−). For instance, it is well-documented that NO and RNS affect DNA integrity and mitochondrial physiology, this leading to genetic mutations (1) and damage to the mitochondrial respiratory chain (2, 3), respectively. Processes ranging from apoptosis, angiogenesis, immunity, and neuronal physiology, all show seemingly contradictory behavior in response to NO. Indeed, the relevance of the steady-state NO concentrations represents a key determinant of its biological function. In support to this assumption, it has been demonstrated that cGMP-mediated processes occur at the low nM range, whereas higher NO concentrations cause protein kinase B (PKB)/Akt phosphorylation; stabilization of hypoxia inducible factor (HIF)-1α; phosphorylation of p53 and, at the µM range, they can generate detrimental conditions usually referred as to nitrosative stress (**Figure 1**). Likewise, in tumor biology, it is now commonly accepted that high NO concentrations mediate apoptosis and cancer growth inhibition, whereas (relatively) low concentrations usually promote tumor growth and proliferation, this supporting the nature of "doubled-edged sword" molecule for NO (4, 5). This dichotomy originates from the observations that the inducible form of NO synthase (iNOS or NOS2) was implicated in the macrophage-mediated tumor killing process (6, 7) (**Figure 1**). NOS2−/<sup>−</sup> mice develop intestinal tumors (8), thereby substantiating the protective role of NOS2 within host defense mechanisms (9, 10). In accordance, a growing body of evidence pointed out that NO-releasing drugs can be toxic for cancer cells.

On the other hand, low rate of NO production can promote tumor growth rather than killing. In line with this assumption, the overexpression of NOS isoforms has been detected in a wide range of human tumors. In particular, NOS2 has been found to be upregulated in melanoma, estrogen receptor-negative (ER−)-breast cancer, as well as in pancreatic, cervical liver and ovarian cancers (10). Moreover, NOS2 seems to be involved in maintaining physiologically relevant levels of NO to sustain the progression phase of carcinogenesis; mainly it is required to promote angiogenesis and to enhance the ability of cancer cells to counteract nutrient paucity in solid tumors and to metastasize (10, 11). NOS2 is also overexpressed in glioma stem cells, and its activity is required for the expression of the cell cycle inhibitor cell division autoantigen-1 (CDA1), which sustains growth and tumorigenicity (12). NOS2 has been also found to be upregulated in hepatocellular carcinoma (HCC), and is often increased in the hepatocytes of patients with chronic hepatitis and alcoholic cirrhosis, conditions that predispose to HCC (13–15). Notwithstanding all these lines of

#### Edited by:

*Saverio Marchi, University of Ferrara, Italy*

Reviewed by: *Francesco Moccia, University of Pavia, Italy*

> \*Correspondence: *Giuseppe Filomeni giufil@cancer.dk Salvatore Rizza rizza@cancer.dk*

#### Specialty section:

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

Received: *13 June 2018* Accepted: *02 August 2018* Published: *04 September 2018*

#### Citation:

*Rizza S and Filomeni G (2018) Role, Targets and Regulation of (de)nitrosylation in Malignancy. Front. Oncol. 8:334. doi: 10.3389/fonc.2018.00334*

**95**

evidence, investigations on NOS2−/<sup>−</sup> mice, in spontaneous and fibrosis-associated models of HCC, reveal little effect of NOS2 derived NO on hepatocarcinogenesis (16), meaning that other players are also involved.

#### S-nitrosylation and Cancer

Redox signal underlying both pro-survival and death pathways, is a molecular information transduced by means of reactive cysteine residues that can undergo S-hydroxylation (SOH), upon reaction with ROS (i.e., H2O2) or S-nitrosylation (SNO), the posttranslational modification induced by NO, which is now emerging to underlie NO bioactivity (17). In the presence of a sulfhydryl group in their close proximity, both these modifications can resolve in a more stable disulfide bridge (S-thiolation, SS) (18–20). Actually, it has been recently questioned whether S-nitrosylation—given its nature of instable posttranslational modification—is able to convey the NOmediated signal, or just acts as mere intermediate for disulfide bridge formation (21). Whatever is the end effector (if directly the SNO group or, indirectly, the SS adduct), the extent of Snitrosylation is determined by a delicate balance between: (i) the rate of NO production, which is catalyzed by NOSs (22, 23), (ii) the activity of a recently discovered class of enzymes termed nitrosylases (24, 25), and (iii) the efficiency of SNO removal, that is mediated by denitrosylases. S-nitrosoglutathione reductase (GSNOR) represents the prototype of this class of oxidoreductases and, so far, the only denitrosylase able to completely reduce NO moiety, reason why it has been also termed GSNO terminase (26–28). Notwithstanding current literature offers still conflicting lines of evidence about the role of NOS/NO system in cancer biology, even less is known on the role played by GSNOR and denitrosylation. In this scenario, it has been reported that GSNOR-ablated (GSNOR-KO) mice show predilection to hepatocellular carcinoma (HCC) in association with S-nitrosylation and proteasomal degradation of the DNA damage repair enzyme O 6 -alkylguanine-DNA alkyltransferase (AGT) (29). As a result, the repair of carcinogenic O 6 alkylguanines is significantly impaired with a consequence increase in tumorigenesis (29, 30). Analyses performed on human HCC patients showed a significant decrease of GSNOR protein levels and activity in the 50% of cases (30), arguing for a functional link between GSNOR-dependent S-nitrosylation and HCC. Although this evidence supports a driving role for GSNOR and excessive S-nitrosylation in HCC ontogenesis, it is still unknown whether they are also implicated in the other phases of carcinogenesis, e.g., tumor promotion and progression (31).

In this regard, it has been published that GSNOR deficient HCC cells have a compromised mitochondrial electron transport chain characterized by the upregulation of succinate dehydrogenase (SDH), likely as an adapting response to the general impairment of the mitochondrial respiratory machinery (32) derived from excessive nitrosative stress. The hyper-nitrosylation of the mitochondrial chaperone TNF receptor–associated protein 1 (Trap1) has been identified as the molecular event responsible for such a rearrangement and, in turn, for the enhanced sensitivity of GSNOR-downregulating HCC to SDH-targeting mitochondrial drugs (32). Nevertheless, it is worth to note that the mean size of GSNOR-deficient tumor xenografts is larger (approximately the double) than parental (GSNOR-proficient) HCC (32), suggesting that excessive S-nitrosylation arising from GSNOR loss, might promote tumor progression and growth in vivo. This hypothesis finds support in a recent study correlating GSNOR downregulation with HER2<sup>+</sup> breast cancer resistance to trastuzumab and poor patient prognosis (33). Altogether, these pieces of evidence argue for a new role of GSNOR in malignancy and resistant phenotypes of breast cancer. However, no evidence about the molecular mechanisms underneath has been provided so far.

# Known and Supposed Targets of S-nitrosylation in Aggressive Cancer

Based on what above reported, it is plausible that impairments of denitrosylation capacity (e.g., upon GSNOR deficiency) modulates the function/activity of oncoproteins susceptible to S-nitrosylation. These defects, more specifically than a general increase of NO production (that could impact on a plethora of different targets and to a different extent) might account for a deregulated NO-signaling in carcinogenesis. This hypothesis is further sustained by a very recent study indicating that GSNORdeficiency (and excessive S-nitrosylation deriving from it), is a condition associated with aging (34, 35), which represents a major risk factor for cancer development. Actually, cancer might count as an aging disease, and shares with aging some common features (e.g., genomic instability, telomere shortening, oxidative stress, deregulation of nutrient sensing) that, indeed, characterize both disorders (36).

Besides those previously mentioned, and others well documented to play a role in apoptosis (e.g., p53, Bcl2, and Fas), many oncoproteins have been discovered in the last decades to undergo S-nitrosylation. The modification of oncoproteins and tumor suppressors by NO—independently on the effects induced, whether gain- or loss-of-function (23, 37)—is emerging as a critical phenomenon associated with neoplastic transformation. Some of these NO-modified oncoproteins participate to signal transduction and are found mutated or modulated in cancer. Within this class of proteins in which S-nitrosylation has been identified as pro-oncogenic modification, we can list: (i) the GTPase Ras (nitrosylated at Cys118) (23, 38), which underlies cancer cell growth downstream of receptor-associated tyrosine kinases; (ii) the phosphatase and tensin homolog PTEN (nitrosylated at Cys83) (39), which regulates the levels of phosphatidyl inositole-3-phosphate/Aktdependent pathway; (iii) the protein kinase c-Src, which represents one of the master regulators of tumor proliferation, invasion and metastatic phenotype, and has been found to be nitrosylated at Cys498 (40) (**Figure 1**). Interestingly, this residue is conserved throughout the Src family of protein tyrosine kinases (SFKs) and, at least for other two members, i.e., Yes and Fyn, has been also reported to stimulate their activation (40). Focal adhesion kinase 1 (FAK1) is also comprised in the

the main cellular denitrosylase. Counteracting the effects induced by NOS, GSNOR finely modulates protein *S*-nitrosylation (second panel from the top), which is establishing as the main posttranslational modification underlying NO bioactivity. A disbalance in NO signaling can promote tumor induction, survival and progression. *(Continued)*

FIGURE 1 | NOS2 deficiency impairs the capability of macrophages to kill cancer cells (Top). Conversely, in conditions of normal (or induced) NOS activity, GSNOR decrease has been linked to many cancer hallmarks, such as: (i) apoptosis and anoikis resistance (due to caspases and, reasonably, FAK-1 *S*-nitrosylation); (ii) genomic instability (DNA repair impairment, due to AGT *S*-nitrosylation and degradation); (iii) cells hyperproliferation (*via* the NO-mediated activation of oncoproteins, such as AKT, RAS, and Src); (iv) angiogenesis (putatively regulated by HIF-1α and TRPs *S*-nitrosylation). Extreme nitrosative stress conditions—induced, for instance, by NOS overexpression or by the use of NO-donors—activate cell death and are implemented (or physiologically activated in macrophages) to destroy cancer cells (Bottom). NO, nitric oxide; GSNOR, *S*-nitrosoglutathione reductase; NOS, nitric oxide synthase; FAK1, focal adhesion kinase 1; AGT, *O*<sup>6</sup> -methylguanine-DNA methyltransferase; HIF-1α, hypoxia-inducible factor-1α; TRP, Transient receptor potential channel.

SFKs family. It is phosphorylated upon integrin engagement in a Src-dependent or independent (auto-phosphorylation) fashion, thus initiating multiple downstream signaling pathways responsible for aggressive and metastatic phenotype (e.g., resistance to anoikis and cell migration) (41). Similarly to Src, Yes and Fyn have been reported to act as FAK1-interacting kinases and to be involved in FAK1 activation as well (42, 43). Notwithstanding this tight relationship, triple-KO cells in which Src, Yes, and Fyn expression is suppressed (SYF cells), still show phospho-active levels of FAK1 upon treatment with NO donors (40). This unexpected evidence clearly indicates that S-nitrosylation of Src, Yes, and Fyn is dispensable for NO-driven phosphorylation of FAK1 and, interestingly, suggests that FAK1 might represent a direct target of S-nitrosylation (**Figure 1**), with this modification driving its oncogenic function.

Another oncoprotein, which has been identified to be crucial in cancer cell survival and growth, especially under low-oxygen tension (hypoxia), is HIF-1α. HIF-1α deregulation has been deeply implicated in different aspects of cancer biology, such as angiogenesis, cell resistance, and tumor invasion (44–47). From a metabolic point of view, HIF-1α aberrant activation underpins the so-called "Warburg effect": the preferential glycolytic consumption of glucose in cancer cells, which takes place also under normal oxygen tension. HIF1α has been found nitrosylated at Cys533 (48), with this being relevant in stroke and cardiovascular disease. However, if S-nitrosylation might somehow induce HIF1α oncogenic activity still remains neglected (**Figure 1**) and would deserve to be investigated in the future.

Among the various classes of proteins that have been identified in the last decades as being activated by S-nitrosylation, the transient receptor potential (TRP) ion channels (49), which represent a huge family of proteins underpinning, among others, warm, taste, and pain sensory transduction, are worth to be mentioned. Besides their well-documented role in the nervous system as mediators of sensations, in the last years it is emerging that many TRPs, such as those belonging to the "melastatin" (TRPM), "vanilloid" (TRPV), and "ankyrin" (TRPA) subfamilies, are overexpressed in many cancer types, this being pivotal for calcium signaling-dependent control of tumor-promoting processes, e.g., vascularization and metastasis (50, 51). In particular, it has been proposed that, by modulating intracellular Ca2<sup>+</sup> concentrations, TRPs are deeply involved in tumor initiation, progression and resistance (52). In this context, it has been very recently found out that TRPA1 is upregulated in breast and lung cancer downstream of the activation of Nrf2, the master regulator of antioxidant response, this conferring non-canonical resistance to tumor cells against oxidative stress and ROS-producing chemotherapeutics (53). Many other observations argue for TRPs inhibition being a promising tool to eradicate cancer (54–56). However, notwithstanding the evidence that S-nitrosylation interferes with TRPs activity and calcium signaling, to date there's still no indication supporting a direct involvement of TRPs Snitrosylation in carcinogenesis (**Figure 1**). Mostly, there's still no study aimed at understanding whether TRPs targeting on nitrosylable cysteines might represent a novel line of intervention in cancer treatment.

## Putative Mechanisms That Affect Denitrosylation in Cancer

The above reported evidence points out that defects in GSNOR expression and denitrosylation are pivotal for sustaining the tumorigenic effects of NO, namely, its role in the progression phase of cancer (31). A recent report on the epigenetic regulation of GSNOR might be of help to understand how this condition can be established in cancer. In particular, it has been demonstrated that GSNOR expression is controlled by the activity of the demethylase Ten-eleven translocation protein 1 (Tet1), a member of the 2-oxoglutarate–dependent dioxygenases that regulates transcription by removing methyl groups from CpG islands located in the promoter regions of genes (34). Remarkably, Tet1 expression has been found to be reduced in a wide range of solid cancers, such as melanoma, prostate, lung, and liver tumors (57, 58)—where also GSNOR mRNA seems to be downregulated—and to correlate with advanced cancer stage, nodal metastases, and poor survival rate in breast cancer patients (59). These lines of evidence suggest that GSNOR might be epigenetically downregulated in aggressive cancer as a consequence of Tet1 reduction, thus providing a new link between epigenetics and redox signaling. This hypothesis can be even extended to further mechanisms of epigenetic regulation. Indeed, given the complex structure of GSNOR mRNA, it has been proposed that GSNOR expression might be also regulated via microRNAs (miRs) (27). However, no putative miRs, able to target GSNOR transcript, has been so far identified to be upregulated in cancer, or hypothesized acting as additional modulators of Snitrosylation.

# CONCLUSION

The role of GSNOR-mediated denitrosylation in carcinogenesis has been capturing the interest of many researchers working on cancer biology, as many lines of evidence indicate that this process is frequently deregulated in cancer cells. In this article, we have tried to summarize what has been discovered in the last years and provide some hints on possible aspects that are still overlooked. Understanding how GSNOR expression is deregulated in may cancer histotypes, as well as the mechanisms underlying the modification of new protein targets involved in cancer resistance and aggressiveness, are, indeed, issues that deserve to be investigated in the future, since they could set the stage for new anticancer approaches interfering with the redox adaptation distinctive of many cancer cells.

#### REFERENCES


# AUTHOR CONTRIBUTIONS

GF conceived the paper. GF and SR wrote the paper. SR drew the figure.

## ACKNOWLEDGMENTS

This work has been supported by Danish Cancer Society Grant KBVU R146-A9414, and Associazione Italiana per la Ricerca sul Cancro (AIRC) Grant IG20719.


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

Copyright © 2018 Rizza and Filomeni. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Mitochondrial citrate carrier (slc25a1) sustains redox homeostasis and Mitochondrial Metabolism supporting radioresistance of cancer cells With Tolerance to cycling severe hypoxia

*Julian Hlouschek, Christine Hansel, Verena Jendrossek and Johann Matschke\**

*Institute of Cell Biology (Cancer Research), University Hospital Essen, University of Duisburg-Essen, Essen, Germany*

#### *Edited by:*

*Andrea Rasola, Università degli Studi di Padova, Italy*

#### *Reviewed by:*

*Alessandra Castegna, University of Bari Medical School, Italy Margaret Ashcroft, University of Cambridge, United Kingdom*

*\*Correspondence: Johann Matschke johann.matschke@uk-essen.de*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

> *Received: 22 February 2018 Accepted: 01 May 2018 Published: 25 May 2018*

#### *Citation:*

*Hlouschek J, Hansel C, Jendrossek V and Matschke J (2018) The Mitochondrial Citrate Carrier (SLC25A1) Sustains Redox Homeostasis and Mitochondrial Metabolism Supporting Radioresistance of Cancer Cells With Tolerance to Cycling Severe Hypoxia. Front. Oncol. 8:170. doi: 10.3389/fonc.2018.00170*

Pronounced resistance of lung cancer cells to radiotherapy and chemotherapy is a major barrier to successful treatment. Herein, both tumor hypoxia and the upregulation of the cellular antioxidant defense systems observed during malignant progression can contribute to radioresistance. We recently found that exposure to chronic cycling severe hypoxia/reoxygenation stress results in glutamine-dependent upregulation of cellular glutathione (GSH) levels and associated radiation resistance opening novel routes for tumor cell-specific radiosensitization. Here, we explored the role of the mitochondrial citrate carrier (SLC25A1) for the improved antioxidant defense of cancer cells with tolerance to acute and chronic severe hypoxia/reoxygenation stress and the use of pharmacologic SLC25A1 inhibition for tumor cell radiosensitization. Exposure to acute or chronic cycling severe hypoxia/reoxygenation stress triggered upregulated expression of SLC25A1 in lung cancer, prostate cancer, and glioblastoma cells *in vitro*. Interestingly, exposure to ionizing radiation (IR) further promoted SLC25A1 expression. Inhibition of SLC25A1 by 1,2,3-benzene-tricarboxylic acid (BTA) disturbed cellular and mitochondrial redox homeostasis, lowered mitochondrial metabolism, and reduced metabolic flexibility of cancer cells. Even more important, combining IR with BTA was able to overcome increased radioresistance induced by adaptation to chronic cycling severe hypoxia/reoxygenation stress. This radiosensitizing effect of BTA-treated cells was linked to increased reactive oxygen species and reduced DNA repair capacity. Of note, key findings could be reproduced when using the SLC25A1-inhibitor 4-Chloro-3-[[(3 nitrophenyl)amino]sulfonyl]-benzoic acid (CNASB). Moreover, *in silico* analysis of publically available databases applying the Kaplan–Meier plotter tool (kmplot.com) revealed that overexpression of SLC25A1 was associated with reduced survival of lung cancer patients suggesting a potential link to aggressive cancers. We show that SLC25A1 can contribute to the increased antioxidant defense of cancer cells allowing them to escape the cytotoxic effects of IR. Since upregulation of SLC25A1 is induced by adverse conditions in the tumor environment, exposure to IR, or both pharmacologic inhibition of SLC25A1 might be an effective strategy for radiosensitization of cancer cells particularly in chronically hypoxic tumor fractions.

Keywords: redox homeostasis, SLC25A1, radiation resistance, chronic hypoxia, cell metabolism, mitochondria, DNA repair

# INTRODUCTION

More than 50% of NSCLC patients receive radiotherapy (RT) or radiochemotherapy (RCT) as part of their treatment. Recent meta-analysis revealed that tumor hypoxia is major biological barrier to successful chemotherapy, RT, and potentially some targeted therapies, promoting treatment failure and poor prognosis of patients suffering from non-small cell lung cancer (NSCLC) (1). Though targeting hypoxia-mediated therapy resistance is considered as an attractive approach to improve therapy outcome in solid human tumors including NSCLC, so far clinical trials evaluating the use of hypoxia-targeting agents did not meet the expectations as they failed to reveal a benefit for the patients (1). This might at least be partially due to the lack of appropriate predictive biomarkers for patient selection but emphasizes the need for the definition of mechanism-based more effective novel therapeutic strategies to overcome hypoxia-induced therapy resistance and the co-development of predictive biomarkers and improved imaging of heterogeneous tumor hypoxia to guide RT protocols (1).

Tumors form a complex microenvironment by co-opting various normal tissue cells and immune cells to support their growth and survival (2). However, the imbalance between cell growth and tumor vascularization limits not only the availability of nutrients and oxygen (O2) but also the removal of secreted potentially toxic metabolites such as lactic acid. As a consequence of rapid proliferation, poor blood supply and altered metabolism tumor hypoxia is frequently linked to lactic acidosis and thus a low pH in the tumor (3, 4). The resulting O2-deprived and nutrient-deprived acidic microenvironment exerts a selection pressure on the tumor cells thereby directing the acquisition of genetic and epigenetic alterations that allow the cancer cells to survive and to adapt to these adverse conditions during multistep carcinogenesis thereby promoting tumor growth and even metastasis (5–8). Moreover, accumulating evidence indicates that the adverse microenvironment also impacts the therapy response at multiple levels and promotes the resistance of solid tumors to chemotherapy and RT (9, 10).

Herein limited availability of O2 known as "tumor hypoxia" is considered as a major environmental factor driving genomic instability, malignant progression, and resistance of solid tumors to RT and chemotherapy (9–11). The cytotoxic efficacy of RT and certain DNA-damaging anticancer drugs relies on the formation of reactive oxygen species (ROS) and thus on local availability of molecular O2 in the tumor tissue during treatment delivery; therefore, an acute decrease in O2 levels as a consequence of insufficient O2 supply confers direct resistance by decreasing oxidative stress and therapy-induced cell killing, the so-called "oxygeneffect" (10, 12, 13). Moreover, cancer cells dispose of multiple survival pathways that allow them to adapt to acute hypoxia and survive these adverse conditions [for a review, see Ref. (14)].

But tumor hypoxia is highly dynamic with respect to its duration (short term to long term) and schedule (transient, chronic, or intermittent), and also fluctuates regionally presumably as a result of the instability and chaotic organization of the tumor vasculature (15–18). Thus, considerable fractions of human vascularized solid tumors are exposed to dynamic changes between hypoxia and intermittent reoxygenation. Chronic changes between hypoxia and intermittent reoxygenation ("cycling hypoxia") constitute a major driving force in the development of malignant progression, tumor heterogeneity, and clonal evolution of therapy-resistant cells (15, 19–24). Acute hypoxia/reoxygenation stress and tumor cell adaptation to chronic cycling severe hypoxia/reoxygenation stress also impact the outcome of cancer RT (23–26). A detailed understanding of the processes that allow cancer cells to escape the cytotoxic effects of RT in the adverse tumor microenvironment is required if we aim to develop effective therapeutic strategies to improve therapy outcome.

In this context, others and we demonstrated that adaptation to chronic hypoxia/reoxygenation stress drives upregulation of cellular GSH levels to avoid excessive ROS damage and promote death resistance (24, 27, 28). Altered nutrient and energy metabolism is one of the emerging hallmarks of cancer cells (29, 30). Moreover, a progressive upregulation of cellular antioxidant systems has been associated with malignant progression (31), suggesting a broader relevance of the above findings for understanding tumor progression and clonal evolution of therapy-resistant cells in tumors with heterogeneous environments. Importantly, as a proof of principle others and we demonstrated that targeting metabolic reprogramming associated with increased GSH levels is a promising strategy for radiosensitization (24, 32).

However, the above studies also demonstrated that cancer cells use different metabolic adaptation strategies to increase their cellular GSH levels, avoid ROS-dependent damage, and escape genotoxic therapies, presumably depending on the genetic background (24, 28, 31–33). Thus, targeting altered glutamine usage will be a viable therapeutic strategy to reduce glutathione levels in some but not all cancer models (24). We therefore reasoned that targeting the increased GSH-based antioxidant capacity and thus the common phenotype of aggressive cancer cells with increased stress tolerance, might be an effective strategy for therapeutic intervention with broader relevance.

Generally, increased GSH levels can be targeted by using drugs interfering with the regeneration of glutathione, the provision of reduction equivalents, increased glutathione synthesis, or glutathione transport and uptake (24, 34, 35). In an effort to define novel ways to specifically target increased GSH levels in aggressive cancer cells the observation about a link between the mitochondrial citrate carrier (CIC, also known as mitochondrial citrate transport protein, CTP) and the maintenance of cytosolic and mitochondrial NADPH pools and the mitochondrial redox homeostasis attracted our attention (36, 37).

The CIC is encoded by the *SLC25A1* gene located on chromosome 22q11.2. Besides citrate, SLC25A1 is also responsible for the electroneutral transport of isocitrate, malate, and phosphoenolpyruvate (38). Furthermore, SLC25A1—together with cytosolic isocitrate dehydrogenase 1 (IDH1) and mitochondrial isocitrate dehydrogenase 2 (IDH2)—takes part in the transport of NADPH derived from reductive carboxylation over the mitochondrial membrane (36) and might thus play a role in GSH regeneration. Overall, SLC25A1 is important for the maintenance of mitochondrial homeostasis and its overexpression was shown to drive tumorigenesis in various types of cancer (39).

Though the authors linked SLC25A1 expression to anchorageindependent growth of NCI-H460 cancer cells (36), it was tempting to speculate that the function of SLC25A1 regarding maintenance of redox homeostasis and mitochondrial function might contribute to the increased radioresistance of lung cancer cells with tolerance to chronic hypoxia/reoxygenation stress. However, the role of SLC25A1 for the cellular radiation response has not yet been investigated. Therefore, in the present study we aimed to explore the role of SLC25A1 for the increased antioxidant capacity of cancer cells adapted to chronic cycling severe hypoxia/reoxygenation stress and the use of SLC25A1 inhibition as novel strategy for radiosensitization of NCI-H460 lung adenocarcinoma cells exposed to acute or chronic cycling severe hypoxia.

#### RESULTS

#### Acute and Chronic-Cycling Hypoxia Increase Expression of *SLC25A1* and *IDH2*

To gain insight into a potential relevance of SLC25A1 for tolerance of lung cancer cells to chronic cycling severe hypoxia, we used our established cell model of so-called "anoxia-tolerant" NCI-H460 lung cancer cells and the respective control cells termed "oxic" NCI-H460 cells. These cells had been exposed to 25 cycles of severe hypoxia (48 h) and reoxygenation stress (120 h) as described earlier (24). Quantitative real-time PCR (qRT-PCR) analysis revealed a significant upregulation of *SLC25A1* in the anoxia-tolerant NCI-H460 cells as compared with the oxic control cells under standard culturing conditions, suggesting that basal upregulation of *SLC25A1* might be a consequence of adaptation to chronic cycling severe hypoxia (**Figure 1A**). *SLC25A1* upregulation was associated with upregulation of *IDH2*, whereas *IDH1* expression was not altered (**Figure 1A**). To test a more general relevance of these findings, we additionally examined the expression of the respective genes in similarly generated anoxia-tolerant DU145 and T98G cells and again observed an upregulated basal *SLC25A1* and *IDH2* expression in the anoxia-tolerant cells as compared to the respective oxic control cells (**Figures 1B,C**), whereas *IDH1* expression was not altered. Interestingly, exposure to acute severe hypoxia (0.2% O2) was also able to trigger increased expression of *SLC25A1* and *IDH2* in the lung cancer cells and this effect was observed in both, oxic and anoxia-tolerant NCI-H460 cancer cells, compared to the oxic NCI-H460 control cells under normoxic (Nx) conditions (**Figures 1D,E**). However, the apparent upregulation of *SLC25A1* and *IDH2* expression induced by acute hypoxia was not significant for anoxiatolerant NCI-H460 cells as the major increase over the levels of oxic NCI-H460 control cells in normoxia was already caused by the adaptation to chronic cycling severe hypoxia, whereas exposure to acute hypoxia had only a minor addition effect (Figures S1D,E in Supplementary Material). Similar observations about a significant upregulation of SLC25A1 expression upon exposure of NCI-H460 cells to acute or chronic cycling severe hypoxia were made using Western blot analysis (Figures S1A,B in Supplementary Material).

# Overexpression of SLC25A1 in Lung Cancer Is Associated With Reduced Overall Survival of Lung Cancer Patients

To investigate whether upregulation of SLC25A1 in an adverse microenvironment *in vitro* might be relevant for the clinical situation, we searched for and analyzed the data of Kaplan–Meier plotter tool (kmplot.com) (40–42), about SLC25A1 expression in lung cancer patients (**Figures 2A,C**) and normal lung tissue by an *in silico* analysis, respectively (**Figure 2B**). The patient cohort has been described in detail in Ref. (40), whereas the parameters used for our *in silico* analysis are given in Tables S1 and S2 in Supplementary Material. We split the lung cancer patient cohort by median of SLC25A1 expression into "High" and "Low," respectively. Our *in silico* analysis revealed that SLC25A1 overexpression was associated with significantly reduced overall survival and median survival in lung cancer patients. Interestingly, this effect regarding overall and median survival was enhanced in the cohort of patients with successful surgery with tumor-free margins (R0-resection) (**Figure 2C**). Of note, SLC25A1 displayed a higher expression in lung cancer patients compared to normal lung tissue (**Figure 2B**) suggesting that SLC25A1 might be a relevant target in lung cancer.

## Pharmacologic Inhibition of SLC25A1 Sensitizes Cancer Cells to the Cytotoxic Action of Ionizing Radiation (IR) and Overcomes Increased Radioresistance Induced by Chronic Cycling Severe Hypoxia

So far, our data demonstrated that acute and chronic cycling severe hypoxia alter the expression levels of SLC25A1. Moreover, our earlier data revealed that anoxia-tolerant NCI-H460 cells are more resistant to IR and chemotherapeutic agents compared to non-selected oxic control cells and this increased radioresistance could be linked to improved antioxidant defense of anoxiatolerant cancer cells (22–24). Since SLC25A1 has recently been linked to cellular redox homeostasis (36) we wondered whether SLC25A1 might contribute to increased radioresistance of cancer cells in acute or chronic cycling severe hypoxia. To test a potential influence of SLC25A1 on radiosensitivity, we treated the anoxiatolerant NCI-H460 cells and the oxic control cells with the pharmacological SLC25A1-inhibitor 1,2,3-benzene-tricarboxylic acid (BTA) (38, 39) 2 h before irradiation with 0–5 Gy under standard Nx conditions (20% O2) and determined the effects of single and combined treatment on cell survival in standard long-term colony survival assays upon removal of BTA 24 h after treatment (delayed plating) (experimental timeline, **Figure 3A**). To test the impact of drug-treatment in acute hypoxia, we additionally performed similar experiments in cells that had been adapted to acute severe hypoxia (0.2% O2) for 2 h prior to inhibitor treatment (**Figure 3A**).

Our results revealed that treatment with the SLC25A1-inhibitor BTA sensitized both oxic and anoxia-tolerant NCI-H460 cancer cells to the cytotoxic action of IR when treatment was performed under Nx conditions (**Figures 3B,D**). As expected, the effect was more pronounced in the anoxia-tolerant cells with increased

SLC25A1 expression and higher radioresistance. Interestingly, radiosensitization was also observed when treatment was performed in acute severe hypoxia (**Figures 3C,E**). BTA treatment was even able to partially compensate the reduced efficacy of IR in acute hypoxia (**Figure 3E**). Remarkably, we observed these effects even though BTA was already removed 24 h after IR so that long-term incubation was performed in inhibitor-free media. To confirm the relevance of SLC25A1 inhibition for the radiorsensitizing effects of BTA, we performed additional experiments with a chemically distinct SLC25A1-inhibitor 4-Chloro-3-[[(3-nitrophenyl)amino]sulfonyl]-benzoic acid (CNASB), with higher specificity for SLC25A1 (43). These experiments revealed that combined treatment of CNASB and IR similarly sensitized both, oxic and anoxia-tolerant NCI-H460 cancer cells, to the cytotoxic action of IR under Nx and hypoxic (Hx) conditions in short-term proliferation (Figure S4A in Supplementary Material) and long-term survival assays (Figure S4C in Supplementary Material) corroborating the radisensitizing action of BTA at the level of SLC25A1.

# Cellular and Mitochondrial Redox Homeostasis Is Impaired After Inhibition of SLC25A1

As demonstrated in NCI-H460 cells IDH1, IDH2, and SCL25A1 are all part of a pathway responsible for the bidirectional

transport of NADPH over the mitochondrial inner membrane (36). We therefore investigated whether the cytotoxic and radiosensitizing effects of the acute BTA treatment might be linked to alteration of cellular redox homeostasis. To this end, we measured total NADPH level (NADPH + NADP<sup>+</sup>) by using a luciferase-coupled enzymatic reaction. As shown in **Figure 4A,** anoxia-tolerant NCI-H460 cells were characterized by a significant increase of total NADPH, presumably as a consequence of adaptation to chronic cycling severe hypoxia, whereas treatment with BTA for 2 h reduced the total NADPH level to the levels of the untreated oxic control cells (**Figure 4A**). In contrast, BTA had no significant effect on total NADPH levels in the oxic control cells (**Figure 4A**). We observed less pronounced effects regarding the BTA-mediated depletion of total NADPH levels in acute hypoxia (**Figure 4B**). Interestingly,

respectively). Median survival (in months) is indicated for each survival curve.

the NADP<sup>+</sup>/NADPH ratios were slightly increased after BTA treatment especially in anoxia-tolerant NCI-H460 cancer cells compared to respective controls in normoxia and hypoxia (**Figures 4C,D**).

NADPH is required amongst others for the regeneration of the reduced form of glutathione (GSH). We therefore next examined the effect of BTA treatment on cellular glutathione levels (**Figures 4E,F**). In line with our previous findings (24), anoxia-tolerant NCI-H460 cancer cells had elevated GSH levels compared to oxic control cells (**Figure 4E**). As expected, BTA treatment significantly reduced GSH levels in oxic and anoxiatolerant NCI-H460 cancer cells when BTA treatment was performed under Nx conditions (**Figure 4E**). Similar observations were made when cells were treated with BTA in acute severe hypoxia (**Figure 4F**). Of note, BTA decreased the GSH levels of

(Hx, 0.2% O2). Twenty-four hours after treatment cells were collected, plated at different cell numbers (200–3,200) in full medium without the inhibitor and grown under Nx for 9 days. (A) Schematic representation of the experimental timeline in normoxia (Nx, left panel) or severe Hx (right panel). Photomicrographs in (B) and (C) show representative pictures of colony formation upon treatment with IR without or with prior BTA treatment for Nx (B) or severely hypoxic conditions (C). Survival curves shown in (D,E) depict quantification of colony formation upon treatment in normoxia (D) and severe hypoxia (E). Colonies were scanned and counted using GelCount. Survival fractions were calculated to the plating efficiency of unirradiated cells under Nx or Hx conditions. Mean values ± SEM are shown, *n* = 3 (\**p* ≤ 0.05, \*\**p* < 0.01, and \*\*\**p* ≤ 0.001; two-way ANOVA with Tukey post-test.) Comparisons between treatment and corresponding controls at the same dose are indicated.

the anoxia-tolerant cancer cells to the levels of the oxic control cells (**Figures 4E,F**).

To further specify the impact of the SLC25A1-inhibitor BTA on cellular antioxidant capacity, particularly mitochondrial and cellular redox homeostasis, we measured mitochondrial and cellular ROS after BTA treatment in normoxia and acute severe hypoxia (**Figures 4G,H**; Figure S2 in Supplementary Material). Importantly, BTA treatment increased the levels of mitochondrial ROS in oxic and anoxia-tolerant NCI-H460 cancer cells under Nx conditions (**Figure 4G**). This effect was even enhanced when BTA treatment was performed in acute hypoxia (**Figure 4H**; Figure S2 in Supplementary Material). Additionally, cellular ROS-levels were increased in both, oxic and anoxia-tolerant cells, upon BTA treatment in normoxia and acute severe hypoxia (Figures S2E,F in Supplementary Material). Nevertheless, BTA-induced cellular ROS levels were lower compared to the mitochondrial ROS levels (Figures S2 in Supplementary Material).

Taken together, the SLC25A1-inhibitor BTA efficiently decreased cellular NADPH and GSH levels resulting in increased mitochondrial and cellular ROS. These findings point to a role of SLC25A1 in the regulation of cellular antioxidant capacity and mitochondrial redox homeostasis.

# SLC25A1 Inhibition Lowers Mitochondrial Metabolism and Alters Metabolic Demand

So far, our data indicated that increased production of mitochondrial ROS upon SLC25A1 inhibition by BTA treatment might play a role in BTA-mediated radiosensitization. However, SLC25A1 is of broader relevance for the transport of metabolites between mitochondrial intermembrane space and mitochondrial matrix as it shuttles isocitrate, malate and phosphoenolpyruvate, in addition to citrate (38, 44). We therefore wondered whether BTA might have a more general effect on mitochondrial metabolism and therefore additionally measured the effects of BTA on mitochondrial metabolism by using extracellular flux measurements (**Figure 5**).

Pretreatment of cells with BTA for 24 h led to decreased basal mitochondrial respiration and lowered ATP-Production in both, oxic and anoxia-tolerant NCI-H460 cells, as determined by using the Mito Stress Test (**Figure 5A**). Additionally, we observed a decreased ability of cells treated with the SLC25A1 inhibitor to respond to forced mitochondrial respiration by uncoupling the electronic transport chain, an effect that is termed reduced spare respiratory capacity (**Figure 5A**, left panel). As shown in **Figure 5A,** the inhibitory effect was more pronounced in the anoxia-tolerant NCI-H460 cells as these cells were found to dispose of an increased spare respiratory capacity (**Figure 5A**, right panel) in comparison to oxic control cells (**Figure 5A**, middle

treatment under Hx conditions. Mean values ± SEM are shown, *n* = 3. (ns *p* > 0.05, \**p* ≤ 0.05, \*\**p* < 0.01, and \*\*\**p* ≤ 0.001; two-way ANOVA with Tukey post-test.)

panel). Similar effects on cell metabolism were observed when using CNASB, another more potent small molecule inhibitor of SLC25A1 (Figure S4B in Supplementary Material).

But inhibition of SLC25A1 might also impact the metabolic demands of oxic and anoxia-tolerant NCI-H460 cells to maintain mitochondrial function. To identify the metabolic pathways needed to maintain mitochondrial respiration under BTA treatment we performed the Fuel Flex Test. In this test the glucose/ pyruvate pathway was inhibited by UK5099, the glutamine pathway by using BPTES and the fatty acid (FA) pathway by etomoxir treatment, respectively. Overall, BTA treatment induced a higher dependency of oxic and anoxia-tolerant NCI-H460 cells on the final glycolytic product pyruvate as well as glutamine to maintain mitochondrial respiration (**Figure 5B**). Furthermore, combined inhibition of two of these pathways reduced the capacity to maintain mitochondrial respiration only for glucose/pyruvate and FAs but not for glutamine when the other two pathways were inhibited (**Figure 5B**). This suggests that BTA reduces the capacity and flexibility of oxic and anoxiatolerant cells to oxidize other fuels, when the pathway of interest is inhibited. Since BTA treatment lowered the flexibility of oxic and anoxia-tolerant cell to oxidize FAs, when glucose/pyruvate or glutamine pathways were inhibited, we measured FA uptake upon BTA treatment. We observed a reduction in the uptake of labeled FAs following BTA treatment (**Figure 5C**). Taken together, BTA treatment lowered the metabolic flexibility for all metabolic pathways examined in both oxic and anoxia-tolerant NCI-H460 cells, as defined by the difference between metabolic dependency and capacity (**Figure 5B**) and lead to reduced FA uptake (**Figure 5C**).

These findings revealed a role of SLC25A1 in conserving mitochondrial metabolism. Importantly, the enhanced flexibility of mitochondrial metabolism as a consequence of upregulated SLC25A1 expression might contribute to increased stresstolerance of the NCI-H460 cancer cells.

#### Inhibition of SLC25A1 Affects the Repair of IR-Induced Double-Strand Breaks (DSB)

So far, our data indicated that pharmacologic inhibition of SLC25A1 by BTA or CNASB increases radiosensitivity of oxic and anoxia-tolerant NCI-H460 cancer cells and that disturbance of the redox homeostasis and of metabolic flexibility might participate in the radiosensitizing effects. Herein, one important aspect of radiosensitivity is the ability of the cells to repair IR-induced DNA DSB. Therefore, we finally examined if BTA treatment would affect the time-dependent induction and resolution of DNA-repair foci positive for Histone H2A.X phosphorylated at serine 139 (γH2AX), a marker for DNA DSB (45). BTA treatment alone (without IR) did not induce additional DSB, ruling out a direct effect of BTA on DNA-damage induction and repair (**Figures 6A,B**). Interestingly, the presence of BTA during exposure to IR slowed the resolution of IR-induced γH2AX-foci and led to higher residual amount of DNA lesions 24 h after IR-treatment in Nx (**Figures 6A,C**) and even in Hx conditions (**Figures 6B,D**). Surprisingly, despite the differences in SLC25A1 expression between oxic and anoxia-tolerant NCI-H460 cells, BTA treatment had similar effects on DSB repair in both cell lines. Even more important, despite the obvious differences in radiosensitivity, the kinetics of DSB repair as determined by the resolution of IR-induced γH2AX-foci were rather similar in oxic and anoxia-tolerant NCI-H460 cells. This suggests that other parameters of the DNA damage response might dictate the differences in radiosensitivity, presumably the differences in

the ability to cope with radiation-induced ROS. Nevertheless, it was an interesting observation that BTA was able to retard the resolution of IR-induced γH2AX-foci and cause a higher amount of residual DNA lesions 24 h after irradiation.

Interestingly, it had recently been shown that a knockout of SLC25A1 not only impairs mitochondrial function, resulting in higher glycolysis rate and the usage of glutamine for compensation of impaired FA synthesis, but also leads to the accumulation of

Figure 6 | Inhibition of SLC25A1 affects the repair of radiation-induced DNA damage. NCI-H460 oxic and anoxia-tolerant cells were left untreated or exposed to ionizing radiation (IR) with a single dose of 5 Gy with or without prior preincubation of the cells for 2 h with 5 mM 1,2,3-benzene-tricarboxylic acid (BTA) under normoxic (Nx, 20% O2) and hypoxic (Hx, 0,2% O2) conditions. (A) Time-dependent accumulation of γH2AX foci in irradiated NCI-H460 oxic and anoxia-tolerant cells under Nx conditions was evaluated by fluorescence microscopy without and with additional treatment with 5 mM BTA at indicated time points after IR. (B) Time-dependent accumulation of γH2AX foci in irradiated NCI-H460 oxic and anoxia-tolerant cells under Hx conditions was evaluated by fluorescence microscopy without and with additional treatment with 5-mM BTA at indicated time points after IR. (C) Mean number of γH2AX foci per cell after IR without and with additional BTA treatment in Nx was calculated with the Focinator v2 software and normalized to oxic control cells 30 min after IR. (D) Mean number of γH2AX foci per cell after IR without and with additional BTA treatment in Hx was calculated with the Focinator v2 software and normalized to oxic control cells 30 min after IR. (E) D-2 hydroxyglutarate levels (D-2-HG) were determined in lysed cells 6 h after treatment with BTA in Nx. (F) Schematic representation of the suggested mechanisms of the actions of BTA and IR on cancer cell survival. Mean values ± SEM are shown, *n* = 3 (\**p* ≤ 0.05, \*\**p* < 0.01, and \*\*\**p* ≤ 0.001; two-way ANOVA with Tukey post-test. Comparisons between treatment and corresponding controls at the same time point are indicated). MIM, mitochondrial inner membrane; α-KG, α-Ketoglutarate; HR, homologous recombination.

2-hyodroxyglutarate (2-HG) in NCI-H460 cells (44). Of note, accumulation of the oncometabolite 2-HG has been linked to DNA-repair defects (46). We therefore wondered whether BTAmediated inhibition of SLC25A1 might have similar effects and measured the effects of BTA treatment on the level of D-2-HG. For this, we used an enzymatic reaction detecting a colored product photometrically at 450 nm. As shown in **Figure 6E**, BTA treatment indeed significantly increased the levels of D-2- Hydroxyglutarate (D-2-HG) 6 h after BTA treatment in both oxic and anoxia-tolerant NCI-H460 cells (**Figure 6E**). This increase in 2-HG might thus be responsible for the observed delay in the kinetics of the repair of radiation-induced DNA DSB observed in both cell lines (**Figures 6A–D**). The finding that there was no significant difference in the absolute D-2-HG levels upon BTA treatment between oxic control and anoxia-tolerant NCI-H460 might explain why the BTA-induced retardation in DNA repair was similar in both cell lines. Taken together our novel findings suggest that BTA exerts a dual effect on the cellular radiation response by targeting both DNA repair and metabolic escape mechanisms of aggressive cancers making it a promising radiosensitizer.

#### DISCUSSION

Tumor hypoxia is an important biological factor causing poor therapy outcome and worse prognosis in NSCLC patients. However, so far hypoxia-targeting strategies have not been translated into clinical practice highlighting the need for an improved understanding of the underlying mechanisms if we aim to develop more effective strategies for therapeutic intervention. In our work, we focus on mechanisms of radiation resistance caused by adaptation of cancer to an adverse Hx environment.

Our previous work identified enhanced antioxidant capacity based on glutamine-dependent glutathione-regeneration as a mechanism favoring stress tolerance and radioresistance of cancer cells with tolerance to chronic cycling severe hypoxia (23, 24). Here, we reveal important novel mechanistic aspects of increased antioxidant capacity in cancer cells and demonstrate a therapeutic potential for SLC25A1 inhibition to overcome radioresistance in chronically Hx lung cancer cells: (i) Exposure of cancer cells to acute or chronic cycling severe hypoxia was associated with upregulated expression of the citrate transporter SLC25A1. (ii) Pharmacologic inhibition of SLC25A1 by BTA reduced cellular antioxidant capacity, enhanced the generation of mitochondrial ROS and abrogated the increase in radioresistance of lung cancer cells induced by adaptation to chronic cycling severe hypoxia. (iii) BTA treatment was also associated with a pronounced disturbance of mitochondrial metabolism, accumulation of the oncometabolite 2-HG, and impaired repair of radiation-induced DNA DSB. These findings point to a yet unknown role of SLC25A1 in radioresistance that might be linked to regeneration of GSH for ROS-detoxification and metabolic regulation of DNA repair. Remarkably, we confirmed the metabolic and radiosensitizing effects of SLC25A1 inhibition also with CNASB, another potent small-molecule inhibitor with documented higher specificity to SLC25A1 (43). In line with our observations using BTA the radiosensitizing effects of CNASB were more prominent in the anoxia-tolerant cancer cells. This further corroborates our assumption that chronic cycling hypoxia/reoxygenation stress increases not only the expression but also the reliance of cancer cells on the SLC25A1 mediated redox-homeostasis for survival under stress conditions.

Interestingly, high SLC25A1 expression in lung cancer patients has already been linked with poorer overall survival compared to lung cancer patients with low SLC25A1 expression in an earlier report (42). However, here we expand these findings with respect to the importance of SLC25A1 in terms of recurrence and progression of lung cancer suggesting that SLC25A1 may be an attractive therapeutic target for tumor cell-specific radiosensitization at least in patients with high SLC25A1 expression. Of note, mutant p53 can promote transcriptional activation of SLC25A1 eventually by an interaction with the FOXO-1 transcription factor (42). This suggests that SLC25A1 might mediate some of the oncogenic activities of mutant p53—an important prognostic marker predictive for relapse and resistance to chemotherapy and RT (47).

In addition, upregulated expression of SLC25A1 was always associated with upregulation of IDH2. Thus, SLC25A1 and IDH2 might cooperate in adaptive metabolic processes that allow cancer cells to cope with the adverse conditions in the tumor microenvironment. Others have described a crucial role of the presence of active IDH2 for proliferation and survival of glioblastoma cancer cells in hypoxia (48).

However, our primary goal was to evaluate the use of SLC25A1 as a therapeutic target to overcome radioresistance in chronically Hx lung cancer cells. It has been shown earlier that inhibition of SLC25A1 by BTA reduces the growth of xenograft tumors of breast, bladder or lung cancer cells without any evidence for additional normal tissue toxicity (39). But here we demonstrate for the first time that SLC25A1 has a role in lung cancer cell radiation resistance. For this we measured the effects of the SLC25A1-inhibitor BTA on radiation-induced eradication of clonogenic tumor cells in long-term colony formation assays (delayed plating) upon treatment in combination with IR. Our results revealed that treatment with BTA for 24 h was sufficient to significantly reduce the survival fraction of both, anoxiatolerant NCI-H460 cells and the respective oxic NCI-H460 control cells. However, the radiosensitizing effect was more pronounced in the anoxia-tolerant NCI-H460 cells and could be further enhanced by treatment in acute severe hypoxia. Of note, we made similar observations when using a chemically distinct small molecule SLC25A1 inhibitor, CNASB, thereby corroborating the action of BTA at the level of SLC25A1. Moreover, the latter finding underlines the importance of SLC25A1-controlled metabolic and redox related alterations for adaptation to acute and chronic cycling hypoxia.

Mechanistically, we discovered that BTA exerts a dual effect on the cellular radiation response by targeting both, cell metabolism and DNA repair, making it a promising radiosensitizer. On the one hand, radiosensitizing effect of the SLC25A1-inhibitor BTA was associated with a reduction of the cellular antioxidant capacity: BTA treatment significantly reduced cellular total NADPH levels and increased NADP+/NADPH ratios in the anoxiatolerant NCI-H460 cells. Moreover, BTA reduced GSH levels in both oxic and anoxia-tolerant NCI-H460 cancer cells, even in acute severe hypoxia. The reduction in these major determinants of cellular antioxidant capacity resulted in increased production of mitochondrial ROS and this effect on cellular redox-balance was even more pronounced when treatment had been performed in acute severe hypoxia. Our findings strongly suggest that SLC25A1 plays a role for maintenance of the antioxidant defense of lung cancer cells with tolerance to acute or chronic cycling severe hypoxia with potential impact on the sensitivity of the cells to ROS-induced damage. Our findings are in line with a report about a role of SLC25A1 for redox homeostasis in NCI-H460 cancer cells obtained by a stable knockout of SLC25A1 (36). Thus, our pharmacologic approach reproduces the findings obtained by the genetic approach, but shows in addition that such effects can also be obtained under conditions of acute severe hypoxia and in cells with tolerance to chronic cycling severe hypoxia. Importantly, the above-mentioned effects were neither due to direct toxic effects of BTA neither under Nx nor under Hx conditions as shown by the lack of a relevant increase in apoptosis (Figure S3 in Supplementary Material) or total cell death (not shown) upon acute treatment up to 72 h, nor to BTA-induced alterations in *SLC25A1* expression (Figure S1F in Supplementary Material).

Though SLC25A1 inhibition by BTA led to a significant induction of mitochondrial ROS compared to the respective untreated control cells, the percentage of cells stained positive for mitochondrial ROS was below 20% (Figure S2 in Supplementary Material). We therefore assume that inhibition of SLC25A1 might have effects in addition to the alterations in the redox balance that contribute to the cytotoxic effects BTA.

SLC25A1 is responsible for the bidirectional shuttling of citrate between the mitochondria and cytosol thereby supporting redox homeostasis by delivery of isocitrate to IDH2, which in turn regenerates NADPH. But SLC25A1 also impacts biosynthetic processes such as lipid biosynthesis (42) and other cellular processes. For example, it has been demonstrated that SLC25A1 and its function in citrate export to cytosol is central for cytokine-induced inflammatory signals and maintenance of NADPH redox state in macrophage activation (37, 49). Moreover, knockout of SLC25A1 has been associated with dysregulation of mitochondrial metabolism in cancer cells (44). Therefore, we further tested the effects of adaptation to chronic cycling severe hypoxia and BTA treatment on energy metabolism. Interestingly, anoxia-tolerant NCI-H460 cells displayed enhanced basal respiration and increased ability to respond to forced mitochondrial respiration (spare respiratory capacity) when compared to oxic NCI-H460 control cells. These findings demonstrate that the mitochondrial changes caused by adaptation to chronic cycling severe hypoxia are not restricted to components of the mitochondrial apoptosis signaling cascade as described earlier (22) but extend to alterations in mitochondrial oxidative metabolism. We also found that pharmacologic inhibition of SLC25A1 by BTA or CNASB significantly decreased mitochondrial respiration and ATP production. Interestingly, treatment with BTA or CNASB particularly lowered the spare respiratory capacity of anoxia-tolerant NCI-H460 cancer cells, pointing to a possible involvement of SLC25A1 in adaptation of oxidative metabolism to metabolic stress induced by chronic cycling severe hypoxia.

In addition, BTA treatment rendered NCI-H460 oxic and anoxia-tolerant cells more dependent on glycolysis-derived pyruvate and glutamine. It has been discussed that enhanced glucose uptake and a shift toward the pentose phosphate pathway (PPP) to generate more NADPH might be a strategy of cancer cells to counteract increased ROS (50). However, cancer cells use diverse strategies to increase their antioxidant capacity (cellular GSH) (31). For example, serine catabolism can participate in mitochondrial redox control under Hx conditions (51) whereas in our hands, altered glutamine usage contributed to the regeneration of glutathione and improved ROS defense of cancer cells with tolerance to cycling severe hypoxia (24). Excessive ROS-production caused for example by oncogene-induced proliferation—causes oxidative damage of cellular macromolecules such as nuclear and mitochondrial DNA, membranes and proteins, thereby affecting major cell functions and cell survival (52–55). Moreover, there is an intimate link between adaptive changes in cell metabolism, generation of ROS, and the death threshold at the mitochondria by interconnected metabolic and redox sensitive pathways (56, 57). Therefore, the ability of the SLC25A1-inhibitor BTA to disrupt redox homeostasis makes it an attractive approach to enhance the cytotoxic effects of ROS-dependent treatments such as IR.

Finally, pharmacologic inhibition of SLC25A1 reduced the flexibility of oxic and anoxia-tolerant NCI-H460 cells to oxidize major metabolic fuels, which could in turn reduce their capability to meet altered nutrient availability after hypoxia-induced microenvironmental changes. Furthermore, BTA treatment decreased the capacity of NCI-H460 cells to oxidize FAs and also reduced FA uptake, particularly in anoxia-tolerant cells. Thereby our findings corroborate data obtained by genetic knockout of SLC25A1 in NCI-H460 cells revealing increased glycolysis and the usage of glutamine to compensate for the loss of FA synthesis as a consequence of reduced citrate transport (44). The same group further highlighted the complex functional role of SLC25A1 in glycolysis, redox homeostasis, and lipogenesis by quantitative metabolic flux analysis (44). Of note, dependency on the uptake of FAs has been recognized as a specific metabolic vulnerability of Hx cancer cells (58, 59).

Our metabolic investigations demonstrate that in addition to ensure the export of citrate from the mitochondria to the cytosol, e.g., for FA synthesis, SLC25A1 seems to be crucial for mitochondrial homeostasis and increased metabolic flexibility of the anoxia-tolerant NCI-H460 cells. Thereby our data confirm the assumption that SLC25A1 might be crucial for metabolic plasticity of cancer cells enabling adaptation and survival under adverse conditions such as nutrient starvation (glucose) or oxidative stress suggested by others (39, 47). But our data extend the stress conditions to acute and chronic cycling severe hypoxia. It appears that inhibition of SLC25A1 leads to a massive disturbance of mitochondrial metabolism and this might severely affect the ability of cancer cells to cope with the toxic effects of IR leading to radiosensitization and enhanced clonogenic cell death.

However, radiosensitivity is largely determined by the ability of the cells to repair radiation-induced DNA DSB. Analyzing the time-dependent formation and resolution of γH2AX –positive DNA repair foci as a mean of DNA-DSB we found that BTA treatment did not alter the initial amount of radiation-induced γH2AX -foci at 0.5–1 h after irradiation. However, BTA-mediated SLC25A1 inhibition slowed the kinetics of γH2AX-foci resolution, and this resulted in significantly higher levels of residual DNA damage in oxic and anoxia-tolerant NCI-H460 cells exposed to combined treatment when compared to irradiation alone at 2, 4, 6, and 24 h after irradiation.

Unexpectedly, we found that despite the obvious differences in radiosensitivity between anoxia-tolerant and oxic control NCI-H460 cells the kinetics in formation and resolution of γH2AXfoci indicative were rather similar in both cell lines. Therefore, it was not surprising that both cell lines responded similarly to combined treatment with BTA and IR with respect to the kinetics of the induction and repair of DNA DSB. From these data we conclude that other parameters of the DNA damage response such as their ability to cope with radiation-induced ROS might dictate the adaptation-induced differences in radiosensitivity and that the resulting differences in the extent of radiosensitization between oxic control and anoxia-tolerant cells might be due to drug-induced interference with redox homeostasis and disturbance of mitochondrial metabolism to meet cellular energy demands during the radiation response.

Nevertheless, these observations revealed that inhibition of SLC25A1 impacts DNA repair. Analyzing the underlying mechanisms we found that BTA-mediated inhibition of DSB repair correlated with an increase in D2-hydroxyglutarate levels induced by BTA treatment. In this context, BTA treatment led to elevated levels of D-2-HG in both NCI-H460 oxic and anoxia-tolerant cancer cells. Thereby, pharmacologic inhibition of SLC25A1 with BTA reproduces another metabolic effect of genetic inhibition of SLC25A1 where somatic loss of SLC25A1 induced a broad dysregulation of mitochondrial metabolism with accumulation of L- and D-enantiomers of 2-HG in NCI-H460 cells (44). Moreover, deletion of the SLC25A1 gene in humans causes the neurometabolic disorder D- and L-2-hydroxyglutaric aciduria, which is also characterized by increased accumulation of 2-HG (60, 61). These observations make it highly likely that elevated D-2-HG levels observed in response to BTA treatment are a direct and acute consequence of SLC25A1 inhibition.

Both enantiomers of 2-HG have been shown to promote malignant progression by their inhibitory action on α-ketoglutarate (αKG)-dependent dioxygenases; they are either synthesized as so-called "oncometabolite" as a result of gain-of-function mutations in IDH1/IDH2 (62, 63) or as pathologic metabolites in Hx cancer cells (64, 65).

Of note, accumulation of 2-HG has recently been linked to inhibition of DNA repair by inducing a homologous recombination repair defect thereby sensitizing cancer cells to poly (ADP-ribose) polymerase (PARP) inhibitors (46, 66). Furthermore, 2-HG was also shown to inhibit ALKBH DNA repair enzymes, leading to enhanced sensitivity to alkylating agents (67). Our new findings strongly suggest that BTA-induced accumulation of 2-HG might contribute to BTA-mediated radiosensitization and this finding could be exploited therapeutically in combination treatments.

Taken together, we identified a role of SLC25A1-regulated redox homeostasis in the tolerance of cancer cells to acute and chronic cycling hypoxia and increased radioresistance caused by adaptation to these stress conditions. In addition, our results clearly demonstrate that inhibition of SLC25A1 by the small molecule inhibitors BTA and CNASB is suited to increase radiosensitivity of NCI-H460 cancer cells exposed to acute or chronic cycling hypoxia. Mechanistically, treatment with the SLC25A1 inhibitor BTA disturbed cellular redox homeostasis by decreasing NADPH/GSH levels but also affected mitochondrial metabolism and cellular energy provision thereby reducing cancer cell survival (summarized in **Figure 6F**). Importantly, we demonstrate that BTA treatment impaired the repair of IR-induced DNA DSB and this was associated with elevation of D-2-HG levels. Since the method allowed us only to detect D-2-HG we cannot exclude that BTA treatment might also increase L-2-HG levels.

The pronounced effects of the SLC25A1-inhibitor BTA on cellular antioxidant capacity, cell metabolism and DNA repair make SLC25A1 inhibitors such as BTA interesting leads for the development of metabolic inhibitors of radioresistance. However, the development of small molecule inhibitors specifically targeting SLC25A1 at nanomolar drug concentrations is required for clinical translation of this approach.

Importantly, high expression of SLC25A1 in lung cancer patients was associated with a poor outcome, particularly after successful surgery (R0-resection), pointing to a potential clinical relevance of SLC25A1 particularly in terms of tumor recurrence. Moreover, exposure to IR, acute hypoxia, and chronic cycling severe hypoxia increased SLC25A1 expression in our *in vitro* models of anoxia-tolerant cancer cells. These findings suggest that SLC25A1 could be relevant as a potential biomarker of increased antioxidant capacity and metabolic flexibility and thus of a high potential for metabolic escape from radio (chemo)therapy.

We conclude that targeting the increased metabolic flexibility of cancer cells with tolerance to environmental stress, or of metabolic escape mechanisms ensuring DNA repair and cell survival under therapy, e.g., by pharmacologic inhibition of SLC25A1, represent attractive strategies to enhance vulnerability to genotoxic treatments and overcome microenvironmentmediated resistance to radio(chemo)therapy in advanced cancers.

## MATERIALS AND METHODS

#### Reagents and Cell Lines

If not stated otherwise, all chemicals were purchased from Sigma Aldrich (St. Louis, MO, USA). NCI-H460 lung adenocarcinoma cells, DU145 prostate cancer cells and T98G glioblastoma cells were obtained from ATCC (Bethesda, MD, USA) and were routinely tested for mycoplasma. NCI-H460, DU145 or T98G cells with tolerance to cycling severe hypoxia/reoxygenation stress were generated by exposure to 16 cycles (T98G) or 25 cycles (NCI-H460, DU145) of severe hypoxia (48 h, <0.1% O2) and reoxygenation (120-h air plus 5% CO2 referred as 20% O2) as described earlier (24). These cells hypoxia/reoxygenationtolerant cells will be termed "anoxia-tolerant cells" throughout the manuscript. Control cells were cultured in parallel under standard ambient O2 conditions (20% O2 plus 5% CO2; the control cells will be termed "oxic cells" throughout the manuscript) (24, 59). Upon selection, cancer cells were routinely grown in RPMI 1640 medium supplemented with 10% (v/v) fetal calf serum (Gibco/Life Technologies, Carlsbad, CA, USA) and maintained in a humidified incubator at 37°C and 5% CO2 (referred to as "normoxia" or "Nx conditions,"). For severely Hx conditions cells were grown in a humidified hypoxia work station (*In vivo* 400, Ruskinn Technology Ltd., Bridgend, Great Britain) at 37°C, 0.2% O2, and 5% CO2 (referred to as "hypoxia" or "Hx conditions,").

#### Patient Survival Data

Patient array data were obtained from and analyzed by Kaplan– Meier plotter tool (kmplot.com) as described elsewhere (40, 41). The cohort was split by median of SLC25A1 expression ("High" and "Low," respectively). Analysis was performed in the cohort once without additional restrictions and once including only patients which had successful surgery (only surgical margins negative), as described in detail elsewhere (40). For further details of used settings please refer to Tables S1 and S2 in Supplementary Material.

#### qRT-PCR Analysis

cDNA was synthesized from 1 µg of total RNA using QuantiTect Reverse Transcription Kit (Qiagen, Hilden, Germany) according to the manufacturer's protocol. Specific primers were synthesized based on available sequences for each listened gene. Primer sets for qRT-PCR were designed with the Blast web tool (U.S. National Center for Biotechnology Information, Bethesda, MD, USA). To exclude cross-reaction of primers with the genes the sequence of interest was compared as well with the Blast database. PCR products were 150–200 bp in size. We used published β2-microglobulin (B2M) primer sequences as housekeeping gene (68). qRT-PCR and cycling conditions were performed using specific oligonucleotide primers (*B2M* forward: TGCTGTCTCCATGTTTGATGTATCT; reverse: TCTCTGCTCCCCACCTCTAAGT; *SLC25A1* forward: CAACGGGGTGAGGGCAT; reverse: CTCGGTGGGGAAGGT GATG; *IDH1* forward: CTCTGTGGCCCAAGGGTATG; reverse: GGATTGGTGGACGTCTCCTG; IDH2 forward: CCTGCTCG TTCGCTCTCC; reverse: GCTTCGCCACCTTGATCCT) and using qPCR kit for SYBR® Green I, 6-Carboxyl-X-Rhodamine (ROX) (Eurogentec, Cologne, Germany) according to the manufacturers protocol. Reactions were carried out on an ABI Prism 7900HT using MicroAmp Optical 384 well Reaction plate (Applied Biosystems by Life Technologies, Bleijswijk, Netherlands) and BIO-RAD PCR Sealers Microseal "B" Film Adhesive seal (optically clear; BIORAD, Munich, Germany). Melting curves were obtained after each PCR run and showed single PCR products. cDNAs were run in triplicate, without reverse transcriptase and no-template controls were run in duplicates. Expression levels for the genes of interest and for the housekeeping gene *B2M* were measured in three independent PCR runs. Expression ratios were calculated using the geometric mean expression of the housekeeping gene *B2M* to normalize the expression data for the genes of interest according to the 2-ΔΔCt—method as described by others (69).

#### Western Blot Analysis

Anti-rabbit SLC25A1 Polyclonal antibody (Thermo Fisher, Rockford, IL, USA) and anti-mouse β-actin from Sigma Aldrich (St. Louis, MO, USA) were used for Western blot analysis. After harvesting, cells were lysed in 75 µL of RIPA buffer containing 0,5% Sodiumdesoxycholate, 1% NP-40 (Nonidet), 0,1% Sodiumdodecylsulfate (SDS), 50-mM Tris-HCl, 150-mM NaCl, 5-µg/mL aprotinin, 5-µg/mL leupeptin, and 3-µg/mL pepstatin. Protein was separated by 12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) under reducing conditions and transferred onto polyvinylidene fluoride (PVDF) membranes (Roth, Karlsruhe, Germany). Blots were blocked in RotiBlock (Roth, Karlsruhe, Germany) for 1 h at room temperature. The membranes were incubated overnight at 4°C with the respective primary antibodies. The secondary antibody was incubated for 1 h at room temperature. Detection of antibody binding was performed by enhanced chemiluminescence (ECL Western Blotting Analysis System; GE Healthcare/Amersham Biosciences, Freiburg, Germany). Equal loading was verified by antibodies against β-actin. Densitometry analysis was performed using ImageJ 2.00 (National Institutes of Health, Bethesda, MD, USA).

#### Determination of Redox Homeostasis

NADPH levels were measured using NADPH Glo Assay (Promega, Madison, WI, USA) according to manufacturer's protocol. Briefly, adherent cells in 96-well plates were lysed and heated under acidic and basic conditions to measure NADP+ and NAPH individually using a luciferase-coupled enzymatic reaction. In parallel, technical replicates were fixed with 4% paraformaldehyde and stained with 10-µg/mL solution of fluorescent dye Hoechst 33342 (Thermo Scientific, Waltham, MA, USA) for normalization to DNA content. Luminescence and Flourescence were measured in triplicates using a BioTek Synergy Microplate reader (BioTek, Winooski, NH, USA).

Levels of reduced Glutathione (GSH) were determined by using Monochlorobimane (MCB) which was described to be specific for GSH metabolized by Glutathione-S-Transferase, leading to a fluorescent adduct (70). The published protocol for measurement of GSH in a Microplate Reader (71) was adapted for using 10-µM MCB with 15 min of preincubation. To rule out alterations in the speed of dye metabolism instead of alterations in absolute cellular GSH levels, kinetic measurements of fluorescence were performed for 15 min at 37°C. Each assay contained cells treated with H2O2 to deplete GSH as a negative control.

To quantify mitochondrial ROS production, cells were stained for 30 min at 37°C with 5 µM of MitoSox (Molecular Probes/ Invitrogen, Carlsbad, CA, USA). To quantify cellular ROS production, cells were stained for 30 min at 37°C with 5 µM of Dihydroethidium (DHE) (Molecular Probes/Invitrogen, Carlsbad, CA, USA). MitoSox- or DHE-positive cells were detected by flow cytometry (BD Accuri C6, Becton Dickinson, Heidelberg, Germany; FL-2). Fractions of gated positive cells (at least 10,000) with higher fluorescence were evaluated. Fold changes were quantified to the corresponding controls.

#### Extracellular Flux Analysis

NCI-H460 cells were plated at 7.500 cells/well in XF96 microplates (Seahorse Bioscience, Billerica, MA, USA) in RPMI Medium with 10% FCS according to manufacturer's recommendations 48 h prior to the assay. Treatment with 5 mM BTA was performed for 24 h. One hour prior to the assay, medium was exchanged to XF base medium (Seahorse Bioscience, Billerica, MA, USA) with 1-mM Pyruvate, 2-mM Glutamine and 10-mM Glucose and incubated at 37°C without CO2. During assays, OCR was measured using a Seahorse XFe 96 analyzer. Mito Stress Test Kit containing 1-µM Oligomycin, 2-µM Carbonyl cyanide-*4*-(trifluoromethoxy)phenylhydrazone (FCCP), 0.5-µM Rotenone and 0.5-µM Antimycin A was perfomed according to manufacturer's protocol. For individual normalization to DNA content, fluorescence was measured after cells were fixed with 4% PFA and stained with 10-µg/mL Hoechst 33342 solution after each assay. Fuel Flex Test Kit containing 3-µM BPTES, an inhibitor of glutaminase (GLS1), 4-µM etomoxir, an inhibitor of carnitine palmitoyl-transferase 1 A (CPT1A) and 2-µM UK5099, an inhibitor of the mitochondrial pyruvate carrier (MPC), was also performed according to manufacturer's protocol. Data were analyzed using Wave 2.4 software (Seahorse Bioscience, Billerica, MA, USA).

#### Determination of Fatty-Acid Uptake

The uptake of FA was quantified by using fluorescent FA analog C1-BODIPY® 500/510 C12. In brief, fluorescent FA (5 µM) were added 24 h after treatment with 5-mM BTA to serum-free media. We quenched the fluorescence of FA in media by adding trypan blue (0.33 mM) to the media. The uptake of fluorescent FA was measured after 1 h, at 37°C spectrophotometrically at 485/528 nm (59). For individual normalization to DNA content, Hoechst 33342 fluorescence was measured after the assay. Cells were fixed with 4% PFA and stained with 10-µg/mL Hoechst 33342 solution.

#### Quantification of D-2-Hydroxyglutarate

Levels of D-2-HG were measured using colorimetric D-2- Hydroxyglutarate Assay Kit (BioVision, Milpitas, CA, USA) according to manufacturer's protocol. At a glance, 107 cells were lysed, spun down and supernatant was transferred into a 96-well plate to quantify the enzymatic conversion of D-2- Hydroxyglutarate to α-Ketoglutarate leading to colored product which is detected photometrically at 450 nm using a BioTek Synergy Microplate reader (BioTek, Winooski, NH, USA).

#### Irradiation and Treatment

1,2,3-benzene-tricarboxylic acid and CNASB were purchased from Sigma-Aldrich (St. Louis, MO, USA). Stock solutions of 200 mM BTA were made with phosphate-buffered aqua bidest and 14-mM CNASB was solved in DMSO. Irradiation was performed at room temperature with an X-ray machine (Precision X-Ray Inc., North Branford, CT, USA) operated at 320 kV, 12.5 mA with a 1.65-mm Al filter, at a distance of 50 cm and a dose rate of 3.71 Gy/min. For an irradiation dose of 2 Gy, the cells were irradiated approximately 32 s, for 5 Gy approximately 81 s. Cells were returned to the incubator immediately after exposure to IR. For irradiation under Hx conditions, Hx cell dishes were kept in BD GasPak EZ Pouch System (Becton Dickinson, Heidelberg, Germany). In terms of combined treatments, inhibitors were added 2 h prior to irradiation.

#### Immunofluorescence: **γ**H2AX Assays

Cells were seeded on glass coverslips placed in 12-well plates and irradiated 24 h later with 5 Gy. Next, cells were fixed and permeabilized in 3% PFA/0.2% Triton-X100 for 15 min and incubated in blocking solution including 2% goat serum at room temperature for 1 h. γH2AX foci were stained for 1 h at room temperature (RT) with Alexa Fluor® 647 mouse anti-H2A.X (pS139) (BD Biosciences, San Jose, CA, USA) diluted 1:50 in blocking solution. DNA was stained with Hoechst33342 (3 µM in PBS) for 30 min at RT. Coverslips were mounted onto glass slides with DAKO mounting medium (Dako NA Inc., Carpinteria, CA, USA). Slides were analyzed with a Zeiss Axiovert 200 fluorescence microscope with ApoTome and ZEN imaging software (Carl Zeiss, Goettingen, Germany). γH2AX foci in at least 50 cells per slide were counted with the Focinator software developed in our laboratory (72, 73).

#### Colony Formation Assays

Clonogenic cell survival in response to the respective treatments was determined comparing the clonogenic survival of cells cultured under Nx and severely Hx conditions. For treatment in normoxia, exponentially growing cells were seeded in tissue culture flasks, incubated under standard culturing conditions (20% O2, 5% CO2, 37°C) and irradiated 24 h later (0 to 5 Gy) without or with prior BTA treatment (5 mM). BTA treatment was performed 2 h prior to irradiation. For treatment in hypoxia, tissue culture flasks of exponentially growing cells were exposed to severe hypoxia (0.2% O2) 2 h prior to BTA treatment and 4 h prior to irradiation, respectively. After completion of the treatments, cells were incubated for 24 h under Nx or Hx conditions, respectively, then washed, collected (0.05% Trypsin, 0.01% EDTA), and plated to 6-well plates at densities of 200–3,200 cells per well (delayed plating). The cell viability was checked before plating the cells by using CASY COUNT (Omni Life Science, Bremen, Germany) and only the number of viable cells was plated. Plates were subsequently incubated for 9 days under standard Nx conditions before quantification of colony formation. For this, cells were fixed in 3.7% formaldehyde and 70% ethanol, stained with 0.05% Coomassie blue, and colonies of at least 50 cells were counted by GelCount (Oxford Optronix, Oxfordshire, Great Britain). The plating efficiency and surviving fraction (SF) to corresponding Nx and Hx controls were calculated as described elsewhere (74).

#### Toxicity Testing

For quantification of apoptotic DNA-fragmentation (sub-G1 population), cells were incubated for 30 min at room temperature with a staining solution containing 50-µg/mL PI in a hypotonic citrate buffer 0.1% (w/v) sodium citrate and 0.05% (v/v) Triton X-100 and subsequently analyzed by flow cytometry (BD Accuri C6, Becton Dickinson, Heidelberg, Germany; FL-2) (75).

For determination of cell proliferation and viability, cells were washed with PBS (1x), fixed with Glutaraldehyde (0.1% in PBS), and stained with crystal violet (0.1% in PBS). The dye was released by TritonX-100 (0.2% in PBS) and measured spectrophotometrically at 540 nm as described elsewhere (76).

#### Statistics

Data represent mean values of at least three independent experiments ± SEM except for **Figures 5A,B** which show data from *n* = 12–18 wells from two independent experiments. Data analysis was performed either by two-way ANOVA test using parametric methods and employing Tukey multiple comparison

#### REFERENCES


post-test where appropriate or by unpaired Student's *t*-test using Prism6 software (Graph Pad Inc., La Jolla, CA, USA). The values of *P* ≤ 0.05 were considered significant.

# AUTHOR CONTRIBUTIONS

JM, JH, and VJ designed and conceptualized the research. JH, CH, and JM performed experiments, analyzed, validated, and visualized the results. JM, JH, and VJ wrote the original manuscript draft. JM and VJ supervised the work. VJ acquired the funding. All authors critically revised, edited, and approved the final version of the manuscript.

## ACKNOWLEDGMENTS

We thank A. Warda for excellent technical support, K. Al-Refae and A. Krysztofiak for support with DNA DSB foci analysis, and the Focinator-Software and the Brigitte und Dr. Konstanze Wegener-Stiftung for funding of the Seahorse XFe 96 analyzer to VJ.

#### FUNDING

The work was supported by grants of the German Research Association (grant number DFG GRK1739/2) and the Deutsche Krebshilfe/Mildred-Scheel-Stiftung (grant numbers 110355 and 70112711) to VJ.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at https://www.frontiersin.org/articles/10.3389/fonc.2018.00170/ full#supplementary-material.


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

*Copyright © 2018 Hlouschek, Hansel, Jendrossek and Matschke. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner 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.*

# Hypoxic Signalling in Tumour Stroma

#### *Anu Laitala and Janine T. Erler\**

*Biotech Research and Innovation Centre (BRIC), University of Copenhagen (UCPH), Copenhagen, Denmark*

Hypoxia is a common feature in solid tumors and is associated with cancer progression. The main regulators of the hypoxic response are hypoxia-inducible transcription factors (HIFs) that guide the cellular adaptation to hypoxia by gene activation. The actual oxygen sensing is performed by HIF prolyl hydroxylases (PHDs) that under normoxic conditions mark the HIF-α subunit for degradation. Cancer progression is not regulated only by the cancer cells themselves but also by the whole tumor microenvironment, which consists of cellular and extracellular components. Hypoxic conditions also affect the stromal compartment, where stromal cells are in close contact with the cancer cells. The important function of HIF in cancer cells has been shown by many animal models and described in hundreds of reviews, but less in known about PHDs and even less PHDs in stromal cells. Here, we review hypoxic signaling in tumors, mainly in the tumor stroma, with a focus on HIFs and PHDs.

#### *Edited by:*

*Giuseppe Filomeni, Danish Cancer Society, Denmark*

#### *Reviewed by:*

*Kasper Rouschop, Maastricht University, Netherlands Simone Cardaci, University of Glasgow, United Kingdom*

*\*Correspondence: Janine T. Erler janine.erler@bric.ku.dk*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

*Received: 14 February 2018 Accepted: 14 May 2018 Published: 29 May 2018*

#### *Citation:*

*Laitala A and Erler JT (2018) Hypoxic Signalling in Tumour Stroma. Front. Oncol. 8:189. doi: 10.3389/fonc.2018.00189*

Keywords: hypoxia, HIF, PHDs, tumour stroma, metastasis

#### INTRODUCTION

Solid tumors are most often partly hypoxic. During the rapid growth of cancer cells, the surrounding vasculature becomes inadequate and is unable to meet the high demand of oxygen creating heterogeneously distributed hypoxic areas within the tumor. Hypoxia in tumors promotes abnormal angiogenesis, desmoplasia, and inflammation. It also boosts the selection of cancer cells that have a more malignant phenotype promoting tumor progression and metastasis, and thereby serving as an indicator for disease outcome. Hypoxic tumor cells are also resistant to radiotherapy and most chemotherapies. Tumors include stromal cells and extracellular matrix (ECM) in addition to cancer cells, which are also affected by the hypoxic environment (1, 2). An ever-increasing number of studies have highlighted the importance of the tumor microenvironment (TME) in regulating tumor progression and dissemination (3).

The cellular response to the drop in oxygen concentration leads to stabilization of hypoxiainducible transcription factor (HIF) in all cell types, which is one of the key regulators of hypoxia response. The HIF prolyl hydroxylases (PHDs) are oxygen-dependent enzymes that target HIF for degradation in air. Thus, in hypoxia, PHDs are inactive, and HIF is stabilized. HIF regulates the expression of many genes that help cells to adapt to hypoxic conditions by decreasing oxygen consumption and increasing its supply. Generally, this includes shifting the energy metabolism to the less oxygen requiring glycolytic pathway and stimulation of the angiogenic genes to increase vascular flow to the hypoxic regions (4).

In addition to HIF activation, responses to hypoxia are also mediated through HIF-independent pathways. These signaling mechanisms include the unfolded protein response (UPR) and mammalian target of rapamycin (mTOR) signaling (5). They work as independent pathways, but in many cases, their signaling is integrated with HIF activation as well as with each other (6).

Unfolded protein response is activated during endoplasmic reticulum (ER) stress. ER stress arises from accumulation of unfolded proteins, which can be contributed by several factors such

**119**

as nutrient and calcium depletion and hypoxia. It consists of different and complex signal-transduction cascades, where pancreatic ER kinase (PERK), inositol-requiring enzyme 1, and activating transcription factor 6 act as sensor proteins, situated in the ER membrane. These mediate pro-survival signaling aiming to counteract ER stress by limiting the production of unfolded proteins and degrading the misfolded proteins preventing cellular damage (5, 7). However, if UPR is not able to re-establish ER homeostasis, it can also lead to pro-death signaling leading to apoptosis. Many publications have shown UPR to be activated in numerous tumor types where hypoxia is one of the promoting factors. Cancer cells have, however, found ways to avoid the ER stress-induced apoptosis (7). UPR signaling has been shown to target vascular endothelial growth factor (VEGF) in cancer cells. Activation of transcription factor 4, which is downstream of the PERK, was shown to induce VEGF mRNA levels, but to lesser extent than HIF signaling. UPR was, nevertheless, also shown to phosphorylate HIF-1α enhancing its activity, which resulted in increased VEGF expression (8). URP signaling has also been shown to have a role in tumor infiltrating immune cells, which can promote both immunosurveillance and immune escape mechanisms (9).

The mTOR kinase is a part of the mTOR complexes 1 and 2 (mTORC1 and 2), and it regulates cell survival, growth, and metabolism. mTOR mediates signals from different intracellular and extracellular stimuli such as growth factors, nutrients, stress, and oxygen levels (10). As a result, mTOR is phosphorylated at multiple sites, which in turn leads to direct phosphorylation of downstream targets such as eukaryotic translation initiation factor 4E binding protein 1 and ribosomal protein S6 kinase. Many of mTOR upstream signaling pathways go through the tumor suppressor tuberous sclerosis 1 and 2 (TSC1/2) complex, and it is also subject to hypoxic regulation. TSC1/2 complex negatively regulates the small GTPase ras-homolog-enriched-in-brain (Rheb). During mTOR activating signaling, TSC1/2 complex is inhibited by phosphorylation, leading to mTOR activation through active Rheb (6, 11). Hypoxia is known to increase the ratio of AMP/ ATP, which leads to activation of AMP-activated protein kinase (AMPK). AMPK is capable of phosphorylating TSC2, but this phosphorylation on the contrary activates the TSC1/2 complex, which inhibits Rheb. Inactive Rheb in turn cannot phosphorylate mTOR, which attenuates the downstream signaling (12). mTOR signaling also intercepts with HIF-dependent hypoxic signaling. HIF upregulates the expression of REDD1 (regulated in development and DNA damage responses), which also activates TSC1/2 complex decreasing mTOR signaling (13). However, this inhibitory mechanism seems to have cell type-dependent differences (14). Oncogenic mTOR has also been reported to promote HIF signaling as well as HIF target gene expression such as lysyl oxidase (LOX) (15) in an HIF-dependent way. VEGF, however, has been shown to be regulated by both HIF-dependent and HIF-independent mechanisms (16). This was also seen in cancer-associated fibroblasts (CAFs) isolated from human breast cancers where tumor suppressor p16INK4A downregulation led to increased Akt/mTOR signaling also affecting HIF-α positively and further increasing the VEGF-A secretion. Enhanced Akt/ mTOR signaling in CAFs was shown also to increase their invasion and migration capabilities. However, the role of HIF was not addressed (17).

Of these three different hypoxia response pathways mentioned, we will focus on the HIF signaling pathway that influences both cancer and stromal cells. Little is known about how HIF/PHDs activities regulate stromal cell function in the TME. Here, we review how HIF/PHD pathways impact on cancer progression in the different cellular and non-cellular compartments of TME.

### HIFs AND PHDs: KEY PLAYERS IN OXYGEN SENSING AND HYPOXIA SIGNALING

Hypoxic signaling is mediated *via* the hypoxia-inducible transcription factor (HIF), which is a dimer consisting of two subunits, HIF-α and HIF-β. HIF-β (also known as ARNT) is constitutively expressed and stabile, but HIF-α is targeted to proteasomal degradation under normoxic conditions. Both subunits contain basic helix-loop-helix and PER-ARNT-SIM domains that are important for DNA binding and dimerization, but only HIF-α contains an oxygen-dependent degradation (ODD) domain that is important for the oxygen-dependent regulation. Under hypoxic conditions, HIF-α is able to bind with HIF-β forming a dimer that specifically binds to hypoxia response elements and so activates the transcription of over hundreds of genes that help the cells to adapt to low oxygen levels (18–21).

HIF-α has three different isoforms, HIF-1α, HIF-2α, and HIF-3α, of which the two first are more widely studied. HIF-1α and HIF-2α resemble each other, but the main differences are in their N-terminal transactivation domain (N-TAD). They both have a similar C-terminal transactivation domain (C-TAD), which contributes to the transcription of their shared targets, whereas the N-TAD differences confer the different binding capabilities and specificity among their targets (22). HIF-3α is the most different from the three and is subject to extensive alternative splicing resulting in many splicing variants. It has only one specific proline residue in the ODD domain and is lacking the C-TAD. Its function only as a transcription factor has been doubted and in fact different HIF-3α variants have been shown to have inductive and suppressive effects on HIF targets (23, 24).

The ODD domain contains two specific proline residues (Pro402 and 564 in human HIF-1α) that are hydroxylated under normoxic conditions by the HIF prolyl hydroxylase 1–3 (also known as EGLN2, 1, and 3 or HIF-P4H1-3) (**Figure 1**). These enzymes belong to the 2-oxoglutarate (2-OG)-dependent dioxygenases and require oxygen, iron, ascorbate, and 2-OG for the hydroxylation reaction. The hydroxylated prolines act as recognition sites for the von Hippel-Lindau (pVHL)–E3 ubiquitin ligase complex, which tags HIF-α for proteasomal degradation (18, 25, 26). HIF is also regulated by factor-inhibiting HIF (FIH) in normoxia. FIH belongs to the same 2-OG family as PHDs. It hydroxylates the asparagine residue within the C-TAD blocking HIF from binding to the p300-CBP coactivators and inhibits the transcriptional activation of HIF (27).

HIF prolyl hydroxylases are fast in responding are fast in responding to changes in O2 concentration due to their much higher *K*m values than the physiological O2 concentrations (28). They are widely expressed in different tissues, but at varying levels. PHD2 is the main enzyme to regulate HIF. Inactivation of PHD2 alone leads to HIF stabilization in normoxia. Both PHD1 and PHD3 are generally regarded to have a complementary role as inactivation of these alone does not lead to HIF stabilization (29). Global inactivation of PHDs in mice supports the finding that PHD2 is a key modulator in HIF regulation. Inactivation of PHD2 led to embryonic lethality, whereas phenotypes resulting from PHD1 or PHD3 inactivation were viable and grossly normal (30). PHDs have specificity toward different HIF-α hydroxylations. PHD2 preferentially hydroxylates HIF-1α, whereas PHD3 has a preference toward HIF-2α (31). Both PHD2 and PHD3 are hypoxia inducible, and it seems PHD2 is upregulated more by HIF-1 and PHD3 by HIF-2 (32). PHD2 has been reported to bind HIF-1α also in hypoxia inhibiting the HIF-1α N-terminal transcriptional activity without affecting its proteolysis (33).

and FIH-stimulating transcriptional activation of HIF-1α, leading to increased expression under normoxic conditions.

HIF prolyl hydroxylases have been reported to have additional targets and to participate in HIF-independent signaling. PHD2 has been shown to inhibit Akt by hydroxylation and binding of pVHL. As hypoxia inhibits PHD2, it activates and promotes tumor growth *via* Akt (34). PHD3 has been reported to interact with activating transcription factor-4, which has been shown to activate genes involved in redox balance, apoptosis, and general cell survival under conditions of compromised nutrition (35, 36). PHD3 has been associated with apoptosis *via* different mediators. For example, PHD3 induces a neuronal regulator kinesin family member 1B β (KIF1β) that in turn induces apoptosis. For this interaction, PHD3 hydroxylation activity was required, but it is still not known whether there is direct hydroxylation of KIF1β (37). PHD3 has been found to hydroxylate and activate the human homolog of the *Caenorhabditis elegans* biological clock protein CLK-2 (HCLK2), which is an important player in DNA damage response. This was required for the activation of the ATR/CHK1/p53 pathway, so that inhibition of PHD3 attenuated DNA damage and the consequent apoptosis (38). PHD3 also directly inactivates the β2-adrenergic receptor (β2AR) by hydroxylating the intracellular domain, which leads to binding of the pVHL and subsequent degradation. β2AR activation is also known to play a role in apoptosis (39, 40).

#### HIF AND PHDs IN TUMORIGENESIS: BAD COPS?

Extensive literature exists regarding the role of HIF and PHDs expressed by tumor cells. We will only touch on a few aspects here, and we recommend more in-depth reviews such as Rankin et al. (41) and Schito and Semenza (4) for more detailed information. Generally, HIF activation in cancer cells has been regarded as tumor promoting. In solid tumors, this usually results from oxygen deprivation leading to hypoxia. However, HIF signaling can also be activated by other oncogenic or tumor suppressor pathways, but mutations directly affecting HIF activation are not common in cancer, with the exception of pVHL deficiency (42). As pVHL mediates the proteolysis of HIF-α subunits, inactivation of this protein leads to constitutively active HIF also in normoxic conditions. However, Von Hippel-Lindau mutations are responsible only for a limited set of cancers such as clear cell renal cancer (ccRCC), pheochromocytoma, and pancreatic neuroendocrine tumors. HIF-2α has been shown to be the driving force in ccRCC (43). HIF activation can also be controlled through PI3K/PTEN/ AKT and RAS/RAF/MAPK signaling pathways. The former regulates the mTOR signaling that increases the translation of HIF protein, as described earlier. The latter has been reported to promote the HIF-coactivator p300 complex and subsequent transcription of target genes (42, 44, 45).

The metabolic components have also been shown to affect HIF stabilization through oxygen-independent regulation of PHD activity. The citric acid cycle (CAC) intermediate 2-OG is an absolute requirement for the enzymatic activity of PHD enzymes, whereas excess amount of other CAC metabolites fumarate, succinate, and oxaloacetate competitively inhibit PHDs (46). Inactivating mutations in enzymes processing fumarate and succinate leading to their accumulation, have been shown to induce HIF, and are associated with rare hereditary cancers, such as papillary renal cell carcinoma and paraganglioma (47–49). Mutations driving abnormal activation of isocitrate dehydrogenase (IDH) 3a were shown to decrease 2-OG amount causing HIF stabilization in cancer cells (50). Other types of IDH mutations lead to the production of 2-hydroxyglutarate (2-HG), which can competitively inhibit PHDs or even increase the activity depending on the 2-HG stereoisomerism (51, 52). However, the mechanisms for the PHD activation by 2-HG is debated (53). In triple negative breast cancer cells, the xCT glutamate-cystine antiporter has been shown to be inhibited by increased glutamate secretion. This in turn decreases the cellular cysteine level. Free cysteine prevents the oxidation of the specific cysteine residues in PHD2, but when the level decreases it leads to self-inactivation of PHD2 and subsequent HIF stabilization (54). Cancer cellsecreted lactate from glycolysis was shown to stabilize HIF in tumor-associated macrophages (TAMs) leading to induction in VEGF and arginase 1 (ARG1), markers associated with an M2 phenotype (55).

Multiple reports have showed that HIFs in cancer cells play a role in tumor progression. Inactivation of either HIF-1α or HIF-2α in *in vivo* breast cancer cell models decreases tumor growth and metastasis to the axillary lymph nodes (56), lung (57–61), and bone (62, 63). Hypoxia promotes cancer cell survival in multiple different ways and these have been reviewed in many papers (64, 65) These ways include the switch to anaerobic metabolism and neovascularization, which will be discussed further below. Hypoxia also promotes growth factor signaling, epithelial–mesenchymal transition, decrease in the apoptotic potential and evasion from the immune system. Cancer cells secrete growth factors to promote their own growth and survival as they generally also have membrane receptors for these factors, resulting in autocrine signaling. These growth factors also stimulate the surrounding stromal cells to change into a tumor promoting phenotype (65). HIF-1 has been shown to directly inhibit apoptosis by decreasing for instance the expression of the proapoptotic Bcl-2 family protein Bid (66) and to induce the expression of the apoptosis inhibitor survivin (67). In this way, cancer cells are protected from the harsh hypoxic environment leading also to a decrease in drug responsiveness (66). Hypoxia also modifies the cell surface proteins, which can shield the cells from immune system in many ways. For example, HIF-1 has been shown to directly upregulate programmed death-ligand 1 (PD-L1) (68) that suppresses T cell activation and CD47, which prevents phagocytosis by macrophages (69). HIF-1 induces a set of microRNAs that are small, non-coding RNA molecules known as hypoxamiRs (70). One of these is miR-210, which has been shown also to regulate cancer cell sensitivity toward cytotoxic T lymphocyte (CTL)-mediated lysis. miR-210 promotes cancer cell immune evasion without affecting cell surface proteins or CTL reactivity (71). In addition, HIF-induced miR-210 was shown also to promote angiogenesis. In cancer cells, miR-210 inhibits the expression and secretion of fibroblast growth factor receptor-like 1, which in turn was found to be a negative regulator of the angiogenesis (72).

In addition, hypoxia promotes cancer cell invasion and metastasis. For example, hypoxia enhances the motility and invasiveness of the cancer cells by promoting epithelial to mesenchymal transition. This happens by upregulating the expression of the transcription factors Snail1, Snail2, and Twist that downregulate expression of E-cadherin that is an important component of the adherens junctions. HIF also mediates this indirectly *via* Notch signaling (59, 73). Generally, CAFs are responsible for the remodeling of the ECM that assists in cancer cell invasion. However, numerous publications have proved that hypoxia induces cancer cells to secrete ECM remodeling enzymes that have essential roles in invasion, metastasis and premetastatic niche formation (60, 74–80). HIF-1 also mediates expression levels of angiopoietin-like 4 and L1 cell adhesion molecule. The former hampers vascular permeability and the latter cancer cell adhesion to the vasculature. Both are needed in intravasation and extravasation during the metastatic process (57).

Hypoxia also promotes the cancer stem cell phenotype. These cells are capable of unlimited cell division, differentiation to other cell types and, most importantly, can initiate cancer (59, 81). The hostile hypoxic microenvironment in the primary tumor primes the cancer cells to survive in the metastatic organ. There is some evidence that hypoxia plays a role in setting a dormancy phenotype by upregulating the main dormancy genes (NR2F1, DEC2, and p27) in breast cancer cells, which persist post-hypoxia helping the cancer cells to become therapy resistance (82).

Experiments have also been performed targeting PHDs that regulate HIF stability. Cancer cell-specific PHD2 haplodeficiency in the MMTV-PyMT breast cancer model led to increased HIF-1α and HIF-2α stabilization, but that did not have an effect on tumor growth or directly on the cancer cells' invasive behavior. However, reduced lung metastasis was seen (83). In the MDA-MB-231 breast cancer model, PHD2 inactivation in cancer cells leads also to attenuate tumor growth. This resulted from decreased transforming growth factor β (TGF-β) processing leading to decreased expression of the extracellular protein osteopontin (SPP1), which has been associated with breast cancer malignancy (84). However, opposing results have been reported with the MCF-7 breast cancer model where PHD2 inactivation in cancer cells promoted tumor growth (84, 85) by upregulation of interleukin-8 (IL-8), VEGF, and the growth factor amphiregulin leading to increased vasculature formation (85). PHD2 inactivation led to HIF-1α stabilization and increased tumor growth in human colon carcinoma (HCT116), colorectal carcinoma (HT29 and RKO), and pancreatic carcinoma (SU.86.86) models. Generally, this was due to increased number of blood vessels. Further experiments with HCT116 cells revealed that the phenotype was not only HIF dependent. PHD2 was also shown to negatively regulate the transcription factor NF-κB. Hence, PHD2 inactivation led to increased secretion of angiogenic factors IL-8 and angiogenin (ANG) *via* NF-κB. These recruited bone marrow-derived cells (BMDCs) and promoted angiogenesis (86). PHD2 inactivation in murine osteosarcoma LM8, Lewis lung carcinoma (LLC), and B16BL6 melanoma cell lines also decreased tumor growth *in vivo*. The effect was claimed to be mainly HIF-independent since inactivation of HIF-1α led to increase in the tumor growth, which could be abolished when HIF-1α was inactivated together with PHD2. After more validation in the LM8 cell line, the PHD2 inactivation was shown to increase the number of vessels in the tumor but with decreased number of circulating cancer cells leading to reduced lung metastasis. LM8 cells were found to have increased expression of matrix metalloproteinases MMP2 and MT1MMP, which increased TGF-β signaling leading to decreased cancer cell proliferation (87, 88). PHD2 overexpression in the pancreatic cancer cells MIA PaCa-2 and PANC-1 decreased tumor growth by suppressing tumor vasculature in an HIF-dependent way resulting from decreases seen in VEGF and angiopoietin-1 (ANGPT1). However, this did not affect metastasis (89).

HIF prolyl hydroxylase 1 has been shown to play a role in cancer progression, but mainly in HIF-independent ways. PHD1 inactivation in ZR75-1, T47D, and MCF7 breast carcinoma cell lines decreased the tumor growth *in vivo* by HIF independently decreasing the cyclin D levels that are involved in cell cycling processes (90). Later PHD1 was shown to hydroxylate Forkhead box O3 (FOXO3), which hampers its interaction with the USP9x deubiquitinase leading to degradation. Decrease in FOXO3a decreases the cyclin D levels (91). Inhibition of PHD1 in HCT116 human colon carcinoma cells *in vitro* sensitized the cells for chemotherapy by decreasing p53 activation during the treatment and inhibited DNA repair increasing cell death (92).

## HIF AND PHDs IN TUMOR STROMA: GOOD COPS?

It has become more evident that cancer progression is not only regulated by the cancer cells but also by the surrounding cancer stroma (93). In addition to the cancer cells, the TME includes different cells such as CAFs, endothelial cells (ECs), immune cells, growth factors, cytokines, and ECM (94). As described earlier, HIF activation in cancer cells has generally been regarded as a tumor-promoting factor, but opposing results have been shown with stromal cells. Global PHD2 haplodeficiency leading to increased stromal HIF-1α and HIF-2α stabilization did not have an effect on primary tumor growth in LLC and pancreatic carcinoma tumor models as well as in the MMTV-PyMT breast cancer metastasis model. Surprisingly, PHD2 haplodeficiency led to decreased lung metastasis, and this clearly demonstrated the importance of stromal hypoxic signaling on tumor progression (83, 95). Recent publications show that stromal factors together regulate the growth and metastatic capabilities of the cancer cells in a complex manner, but nevertheless revealing possible insights into new therapeutic targets.

#### HYPOXIC RESPONSE OF CAFs

Cancer cells are known to harness normal quiescent fibroblast to promote cancer progression and one way they achieve this is by secreting TGF-β. This activates the fibroblasts that become CAFs with increased capacity to contract and remodel the ECM (96). CAFs are the most abundant stromal cell type and are generally thought to promote cancer progression. Many coculture experiments with CAFs and cancer cells show increased tumor growth when compared with normal fibroblast together with cancer cells (97). CAFs can promote cancer cell invasion by secreting growth factors and cytokines or remodeling the ECM—either making tracks or aligning fibrils for cancer cells to travel along (98). When fibroblasts become CAFs, their gene expression changes, and one of the most commonly used marker for CAFs is α-smooth muscle actin (αSMA). In addition to αSMA, CAFs have been shown to express other markers such as fibroblast-specific protein 1 (FSP1), platelet-derived growth factor receptor (PDGFR), fibroblast activation protein, and periostin, for example. Recently, it has become clear that CAF populations are heterogeneous with diverse markers, which may give them specific functions during tumorigenesis (99).

HIF-1 has been shown to play a role in normal fibroblast and CAF behavior. One experiment showed HIF-1α as a tumor suppressor when its inactivation in FSP1-expressing cells in MMTV-PyMT transgenic mice led to enhanced tumor growth by reducing the tumor vascular density with less leaky vessels together with decreased infiltration of TAMs. The molecular mechanisms behind these results were not characterized. No difference was seen, however, in tumor growth in the same model with HIF-2α-deficient fibroblasts (100). Another experiment with *in vitro* data indicated that hypoxic CAFs are involved in formation of new blood vessels in a manner requiring both HIF-1α and G-protein estrogen receptor to upregulate VEGF in hypoxia (101). HIF-1α has been shown to work as a tumor promoting factor as well. Experiments using the active form of HIF-1α in human skin fibroblast showed increased tumor growth when coinjected together with MDA-MB-231 cells, whereas HIF-2α did not have an effect. It was suggested that HIF-1α activation led to autophagy and aerobic glycolysis, which would produce nutrients to surrounding cancer cells promoting their growth (102, 103). PHD2 modulation in CAFs also affects tumor progression, but mainly *via* HIF. Hypoxia, PHD2 depletion, or PHD2 haplodeficiency was shown to inactivate CAFs by reducing their contraction and ECM remodeling capabilities (83, 104). Yes-associated protein 1 oncoprotein has been demonstrated to be important for CAF activation. It mediates the contraction *via* myosin light chain (MLC), which in turn regulates the contractile actomyosin function (105). Hypoxia and dimethyloxalylglycine (DMOG) decreased the levels of the phosphorylated MLC in CAFs (104). CAF inactivation was further demonstrated *in vivo* when CAFs with either PHD2 haplodeficiency or shRNA inactivation were orthotopically transplanted together with breast cancer cells leading to decreased lung metastases but without any changes in primary tumor growth (83, 104). Also, systemic PHD inhibition with DMOG decreased lung and liver metastases in a 4T1 breast cancer model. DMOG treatment also reduced the stiffness and αSMA levels within the tumors without affecting the number of αSMA positive cells (104). However, PHD2 haplodeficiency in platelet-derived growth factor receptor α (BDGFRα)-positive CAFs in MMTV-PyMT transgenic mice did not have an effect on the lung metastases (83). As a side note, targeting only one CAF marker might not ensure wide inactivation in a heterogeneous CAF population (99).

#### HYPOXIC RESPONSE OF THE TUMOR VASCULATURE

The tumor vasculature is essential in supplying nutrients and oxygen to the fast dividing cancer cells but becomes quickly inadequate to meet the need of a fast expanding tumor. Oxygen deprivation in the tumor causes HIF stabilization and upregulation of proangiogenic factors both from cancer cells and tumor stromal cells to promote the vessel formation needed for tumor progression and metastasis (106). ECM proteins have also been shown to be involved in angiogenesis. Hypoxia-inducible LOX promotes the expression of VEGF in cancer cells *via* PDGFRβmediated Akt activation (107). Tumor vasculature, however, has been found to be abnormal with morphological changes, leakiness, and aberrant pericytes, which result in poorly perfused vessels that are not able to rescue the hypoxic condition. This results from excessively secreted pro- and antiangiogenic factors leading to disorganized vessel formation. This can hinder, for example, the delivery of anticancer drugs and also benefit cancer cells during their extravasation process (108).

Inactivation of HIF-1α in Tie2 positive ECs lead to decreased tumor growth in a subcutaneous LLC model, caused by reduced density of tumor vessels. ECs had decreased proliferation and migration under hypoxia as well as decreased expression levels of VEGF and VEGF receptor-1 (VEGFR-1). This demonstrated the importance of EC autocrine expression of these factors during vessel formation (109). Similar results were obtained with the MMTV-PyMT breast cancer model and the LLC model. HIF-1α inactivation in Tie2-positive ECs drastically decreased lung metastasis, but only had moderate effect on primary tumor growth. In addition, HIF-1α inactivation decreased the expression of inducible nitric oxide synthase (iNOS), further decreasing the amount of nitric oxide (NO) and leading to less permeable ECs for cancer cell migration. HIF-2α inactivation in ECs on the other hand led to decreased levels of arginase 1 (ARG1), which in turn increased the NO levels making the EC layer more accessible for cancer cells. Supporting this finding, *in vivo* tail vein injection of LLC cells showed more cancer cell colonization in the lung of HIF-2α EC null mice (110). However, HIF-2α inactivation from vascular endothelial (VE)-cadherin positive ECs decreased tumor growth in LLC and B16F1 models, as well as in carcinogen-induced skin epithelial tumors. HIF-2α inactivation was shown to increase the number of vessels and their branching but decreasing the number of mature functional tumor vessels. This resulted from disrupted delta-like ligand 4/Notch (DII4/Notch) signaling leading to decreased expression of angiopoietin-2 (111, 112). These results show different roles for HIF-1 and HIF-2 in angiogenesis. Mice with PHD2 haplodeficiency in Tie2-positive ECs did not have an effect on the primary tumor growth, but the amount of liver metastases in a pancreatic cancer model was reduced. This resulted from HIF-2-induced expression of sVEGFR-1 and VE cadherin that normalized ECs with less permeable vessels leading to a less hypoxic tumor without any changes in the actual number of vessels (83, 95).

Pericytes are mural cells that cover the ECs and provide vascular stability and are known to signal to each other during angiogenesis (108). Pericyte depletion was shown to disrupt primary tumor growth by decreasing the microvessel density and increasing hypoxia, but this, however, led to increased metastasis showing their importance in tumor vasculature (113). Later, the increased hypoxia was shown to induce Ang-2 expression within the tumor which resulted in vascular instability and lung metastasis. This could be restored by inhibiting Ang-2 signaling (114).

# HYPOXIC RESPONSE OF IMMUNE CELLS

The microenvironment of solid tumors also contains tumor infiltrating immune cells. These include lymphoid cells and myeloid cells, such as T- and B-cells and natural killer cells, TAMs, myeloid-derived suppressor cells (MDSCs), and neutrophils, which all have both pro- and antitumorigenic effects. Hypoxia has been shown to alter immune resistance and suppression, which helps tumor cells to survive immune surveillance (2). TAMs are the most common immune cells found within the tumor and are generally associated with tumor progression and also have been shown to promote cancer cell invasion and angiogenesis (115). Hypoxic tumor regions have been shown to recruit TAMs *via* VEGF and semaphoring 3A expression and once in a low oxygen condition their gene expression profile changes to a more cancer promoting phenotype (116, 117). Macrophages are known to have plasticity, and their phenotype can change based on external cues. Once they get activated they can have tumor suppressing M1 or tumor promoting M2 polarization. Stimulation driving the M1 phenotype also activates HIF-1 which upregulates iNOS levels and drives NO synthesis. During M2 polarization HIF-2 is induced leading to an increase in arginase-1, which in turn decreases the NO synthesis (118). *In vitro* assays suggest that depletion of HIF-1α in TAMs induces more M2 phenotype (119). By using PHD2-haplodeficient mice, which ensured less hypoxic tumors, Laoui et al. showed, however, that the oxygenation state in the tumor did not significantly affect the differentiation of TAMs into M1 or M2 like (120). They saw changes in hypoxia-regulated genes in M2 like TAMs, but not in M1 like TAMs. Overall the results suggest that hypoxia itself does not influence the activation of the TAMs into M1 nor M2 (120).

HIF-1α- and HIF-2α-expressing macrophages have been shown to promote tumor progression. Inactivation of HIF-1α from lysozyme 2 (LysM)-positive neutrophils and macrophages in MMTV-PyMT mice led to hindered tumor progression and smaller tumors. This was shown to result from an increased number of cytotoxic T-cells (121). Mice lacking HIF-2α from LysM-positive macrophages in inducible hepatocellular carcinoma and colitis-associated cancer had reduced number of TAMs in tumor regions which associated with a delayed tumor progression tendency (122). However, opposing results have been shown with PHD inactivation. One experiment showed that PHD2 inactivation in macrophages together with CD4<sup>+</sup> T-cells decreased the tumor progression by overall downregulation of protumoral and antitumoral cytokines eventually leading to increased tumor cell death in an LLC model, and this was in part because of HIF-1α stabilization (123). Mice with a melanoma tumor model were treated with granulocyte–macrophage colony-stimulating factor together with systemic inhibition of PHD3 by AKB-6899 leading to decreased tumor size and lung metastasis. This resulted from HIF-2α stabilization in TAMs, which increased the expression of VEGF sequestering the soluble VEGFR-1 which decreased the vasculature within the tumor (124).

Hypoxia and HIF-1α stabilization regulates the MDSCs within the tumor enhancing their role as T-cell suppressors (125). Tumor-infiltrating MDSCs, as well as macrophages and tumor cells, were shown to have increased levels of the cell surface protein PD-L1, which was shown to be an HIF-1α target and upregulated by hypoxia. PD-L1 upregulation in MDSCs suppressed T-cell activation, whereas PD-L1 blockade under hypoxia enhanced MDSC-mediated T-cell proliferation and function and was associated with decreased interleukin-6 and interleukin-10 expression (68). Hypoxia has been shown to decrease the expression of the cell surface protein major histocompatibility complex class I *in vivo* and *in vitro* in renal cell carcinoma cells providing them a mechanism to evade immune surveillance by cytotoxic T-cells (126).

#### HYPOXIC RESPONSE OF THE ECM

In addition to the different cell types within the TME, the ECM composition and structure also contributes to tumor progression (94). The ECM does not only provide an architectural scaffold defining tissue boundaries but it also regulates cellular behavior, such as cell adhesion and migration by mechanical and biochemical cues. The ECM can bind and store growth factors or directly interact with receptors on the cell surfaces. The ECM is a dynamic meshwork of proteins that is constantly being remodeled and it is rich in proteoglycans and fibrillar collagens. ECM proteins have also been shown to directly induce cancer cell proliferation and metastasis. LOX is upregulated by hypoxia, and in a colorectal cancer model, LOX induced cancer progression *via* activation of SRC signaling (127). The ECM is mainly secreted by CAFs as it is one of the main functions of fibroblast to maintain the homeostasis of the ECM. CAFs, however, have abnormal ECM regulation activity, and they have been shown to secrete more ECM proteins than normal fibroblasts (98). In addition to CAFs, cancer cells are known to play a part in ECM remodeling (128). A high collagen content together with a stiff ECM is generally associated with malignant metastatic cancers (75, 129). The biosynthesis of the fibrillar collagens requires posttranslational modifications that take place both inside and outside the cell with the help of collagen modifying enzymes, many of which are known to be regulated by HIF (130). Collagen modifying enzymes have been shown to have important roles in cancer progression. Inactivation of collagen prolyl 4-hydroxylase P4HA1 or P4HA2 in MDA-MB-123 breast cancer cells decreased collagen deposition and attenuated tumor growth and resulted in as much as 99% decrease in lung metastasis (79). Inactivation of lysyl hydroxylase PLOD2 in the same cancer model decreased the metastasis to lungs resulting from inefficient formation of collagen fibers and decreased tumor stiffness (78). Inhibiting LOX and lysyl oxidase-like 2 (LOXL2) and LOXL4 decreased metastasis in several cancer cell types (60, 74, 76). LOX, LOXL2, and LOXL4 deficiencies decreased collagen cross-linking and stiffness which was also needed to recruit CD11b<sup>+</sup> BMDCs into the premetastatic niches to promote tumor cell colonization (60, 77, 80). LOXL2 in turn upregulated the expression of matrix remodeling enzymes such as tissue inhibitor of metalloproteinase-1 and matrix metalloproteinase-9 (MMP9) (76). LOXL2 also activated CAFs *via* integrin β3 and promoted their collagen contraction, a phenomenon known to increase matrix stiffness (131).

Cancer-associated fibroblasts are able to mechanically remodel the ECM by active contraction resulting in matrix stiffening (105). CAFs with PHD2 haplodeficiency were less active than wild-type CAFs which was demonstrated by decreased expression levels of markers associated with CAFs (αSMA, FSP1, and PDGFRα) and decreased contraction capabilities (83). PHD2 haplodeficiency in CAFs also decreased matrix production, and this was accompanied by decreased expression of collagen modifying enzymes LOX, P4HA1, P4HA2, PLOD2 as well as matrix metalloproteinases MMP2 and MMP9. This is partly surprising, since hypoxic fibroblast and cancer cells generally upregulate the expression of the same genes (132). Similar results were seen with human head and neck carcinoma-associated fibroblasts and human vulvar carcinoma-associated fibroblasts. PHD2 inactivation or 1% oxygen decreased αSMA expression, and they had decreased contraction capabilities (104).

## HYPOXIC RESPONSE OF CANCER METABOLISM

One of the most common characteristics for cancer cells is rapid proliferation, which is accompanied by high demands of energy. In normoxia, cells generally produce energy as ATP by mitochondrial oxidative phosphorylation, which is an O2-dependent pathway. As tumors grow, cells face the limitation of decreased oxygen concentration and switch to inefficient glycolysis to ensure ATP production (46). HIF-1 has been shown to be an important regulator in the metabolic switch as it induces the expression of genes that adjust the cellular metabolism away from oxidative phosphorylation toward increased glycolysis. This mechanism is not just to ensure the ATP production under hypoxia but also to prevent excessive formation of reactive oxygen species (ROS). Most importantly HIF-1 induces pyruvate dehydrogenase kinase 1 that inhibits the activity of the pyruvate dehydrogenase complex (PDC). PDC regulates the first step needed in oxidative phosphorylation where pyruvate is converted to acetyl coenzyme A (133). HIF-1α also upregulates lactate dehydrogenase A and monocarboxylate transporter 4 which promotes conversion of pyruvate to lactate (end product of glycolysis) and transports it out of the cell, respectively (134, 135). HIF activation also upregulates the expression of glucose transporters GLUT1 and GLUT3, which are responsible for the glucose import into the cells (136, 137). Lactate from glycolysis is considered as a major reason for the tumor acidification. Hypoxia upregulates carbonic anhydrase IX, which regulates the tumor pH and has an important role in the survival of tumor cells in hypoxic regions of tumors and metastasis (138). In addition, HIF indirectly targets the metabolic genes that contribute to the glycolytic shift by upregulating miR-210. miR-210 in turn inhibits multiple targets such as iron-sulfur cluster assembly proteins (ISCU), cytochrome *c* oxidase assembly protein (COX10), NADH dehydrogenase (ubiquinone) 1 alpha subcomplex 4 (NDUFA4), and succinate dehydrogenase complex subunit D (SDHD), all of which play a role in mitochondrial function (139–141). PHD3 was in turn shown to induce the activity of PDC HIF independently by binding to the complex. PHD3 deficient cells were seen to have clear decreased PDC activity, which promoted survival in prolonged hypoxia together with a lower amount of ROS (142). HIF also suppresses mitochondrial function by downregulating multiple components in the electron transport chain and also by direct mitochondrial autophagy mediated by increased BCL2 interacting protein 3 (BNIP3) levels. In this way, hypoxia also decreases the levels of ROS, which are harmful in increased quantities (143). There are reports stating that mitochondrial ROS itself also regulates HIF, but there is also many opposing results [reviewed in Ref. (144)]. Increased ROS levels have been shown to promote HIF stabilization during hypoxia, by further inhibiting PHD enzymes. Contradicting results claim it is not ROS regulating HIF stabilization, but rather the mitochondria, which monitor the intracellular oxygen availability. Under normoxic conditions ROS are able to stabilize HIF, but the exact mechanism is not clear. It has been suggested to be *via* inhibition of PHDs or pVHL, but another signaling pathway has been proposed. HIF-1α gene expression was shown to be upregulated in oral squamous cell carcinoma and MCF-7 breast cancer cells by ROS *via* induced ERK and P13K/AKT signaling (145, 146). Overall, HIF-regulated glucose metabolism supplies sufficient glucose import for less efficient glycolysis and removes the excess end products out of the cell. It defends the cells against increased

(TAM) infiltration, which support tumor growth. Hypoxic cancer cells corrupt fibroblasts by secreting, for example, transforming growth factor β, which turns fibroblast into cancer-associated fibroblasts (CAFs) to drive the cancer progression. CAFs promote desomplasia and together with cancer cells they secrete extracellular matrix (ECM) components such as collagens. One important secreted factor is the hypoxia-inducible enzyme lysyl oxidase, which cross-links fibrillar collagen at the primary tumor and prepares the premetastatic niche for the recruitment of bone marrow derived cells (BMDCs) that are needed for cancer cell colonization at the metastatic site.

ROS production, which is generated in the oxidative phosphorylation pathway under too low oxygen concentration and gives the cancer cells protection to proliferate and grow under hypoxic conditions (147).

Within the solid tumor there are regions that are not under hypoxia, but still the cancer cells have a tendency to switch to inefficient glycolysis even in presence of oxygen. However, the metabolic state in the different cells within the tumor is heterogeneous and there has been evidence of supplying metabolites between cells (148). When fibroblasts become CAFs, they have been shown to decrease the expression of isocitrate dehydrogenase 3α. This leads to a decreased amount of effective 2-oxoglutarate, which is needed as a co-substrate in the hydroxylation reaction, and increasing levels of succinate and fumarate. This results to PHD inhibition, which stabilizes HIF-1α and promotes glycolysis in the CAFs. Metabolites from the normoxic glycolysis in CAFs was hypothesized to feed the surrounding cancer cells (149).

#### CONCLUSION AND FUTURE PERSPECTIVES

Hypoxia is a condition that plays a role in normal physiology, such as during embryonic development, but it is more commonly associated with pathological states. Oxygen deprivation is a common feature of all solid tumors and affects all components of the TME (**Figure 2**). Hypoxia has clear effects on both cancer cells and surrounding stromal cells, but it does not promote tumor progression in every case. Different cell types seem to be differentially regulated by the hypoxic environment. In cancer cells, HIF stabilization generally promotes tumor progression. Inactivation of HIF-1 or HIF-2 in cancer cells in mouse cancer models decreases the tumor growth and the formation of metastases in many ways, practically affecting all the hallmarks of cancer.

#### REFERENCES


When cancer cells are targeted with PHD2 inactivation, which enables HIF stabilization, the results are more variable. PHD enzymes do also have HIF-independent targets and functions that can complicate their role in cancer promotion. However, PHD2 haplodeficiency in ECs leads to less tumor hypoxia resulting from HIF-2 activation in the ECs themselves stimulating EC intrinsic proangiogenic signaling leading to EC normalization. Global haplodeficient inactivation of PHD2 or treatment with a PHD inhibitor does not have an effect on primary tumor growth, but it rather decreases metastasis. Targeting PHD2 in CAFs leads to their deactivation and reduces their capabilities to promote cancer cell invasion and metastasis both *in vitro* and *in vivo*. These findings suggest that global targeting of PHD and HIF activities is sufficient to reprogramme the TME to suppress tumor progression. However, we are only beginning to understand how hypoxia, HIFs and PHDs alter the TME, and further research is required to gain insight. Thus, while hypoxia research has been performed for decades, recent data suggests that there is much more to learn, and that this may have important clinical implications regarding the use of agents that target hypoxia pathways such as HIF or PHD inhibitors.

## AUTHOR CONTRIBUTIONS

The review was written by AL and edited by JTE.

#### ACKNOWLEDGMENTS

We would like to thank Kamilla Westarp Zornhagen for the helpful comments for the manuscript. We thank the KBVU grant (R146-A9250) from the Danish Cancer Society that supports AL and the Novo Nordisk Foundation for the Hallas Møller Stipendum that supports JTE.


promotes lymphatic metastasis of hypoxic breast cancer cells. *Proc Natl Acad Sci U S A* (2012) 109:E2707–16. doi:10.1073/pnas.1214019109


conditions by inducing P4HA1, P4HA2, and PLOD2 expression in fibroblasts. *J Biol Chem* (2013) 288:10819–29. doi:10.1074/jbc.M112.442939


activity. *Biochem Biophys Res Commun* (2014) 451:288–94. doi:10.1016/j. bbrc.2014.07.114


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

*Copyright © 2018 Laitala and Erler. 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 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.*

# Activation of p62/SQSTM1–Keap1– Nuclear Factor erythroid 2-Related Factor 2 Pathway in Cancer

#### *Yoshinobu Ichimura\* and Masaaki Komatsu*

*Department of Biochemistry, Niigata University Graduate School of Medical and Dental Sciences, Niigata, Japan*

Autophagy and the Keap1–Nrf2 system are major cellular defense mechanisms against metabolic and oxidative stress. These two systems are linked *via* phosphorylation of the ubiquitin binding autophagy receptor protein p62/SQSTM1 in the p62–Keap1–Nrf2 pathway. The p62–Keap1–Nrf2 pathway plays a protective role in normal cells; however, recent studies indicate that this pathway induces tumorigenesis of pre-malignant cells, and promotes the growth and drug resistance of tumor cells *via* metabolic reprogramming mediated by Nrf2 activation. These findings suggest that impairment of autophagy is involved in the acquisition of malignancy and maintenance of tumors, and furthermore, that p62/SQSTM1 could be a potential target for chemotherapy in cancers that harbor excess p62.

#### *Edited by:*

*Giuseppe Filomeni, Danish Cancer Society, Denmark*

#### *Reviewed by:*

*Manuela D'Eletto, Università degli Studi di Roma Tor Vergata, Italy Hilary Ann Coller, University of California, Los Angeles, United States Gianmaria Fimia, University of Salento, Italy*

> *\*Correspondence: Yoshinobu Ichimura yichimu@med.niigata-u.ac.jp*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

> *Received: 22 March 2018 Accepted: 24 May 2018 Published: 07 June 2018*

#### *Citation:*

*Ichimura Y and Komatsu M (2018) Activation of p62/SQSTM1–Keap1– Nuclear Factor Erythroid 2-Related Factor 2 Pathway in Cancer. Front. Oncol. 8:210. doi: 10.3389/fonc.2018.00210*

Keywords: selective autophagy, p62/SQSTM1, Keap1–nuclear factor erythroid 2-related factor 2 system, metabolic reprogramming, cancer

# INTRODUCTION

Autophagy is a bulk degradation process in which cytoplasmic components are sequestered in a double-membrane structure to form an autophagosome. The contents of the autophagosome are subsequently degraded after fusion with a lysosome (**Figure 1**) (1). Under starvation conditions, autophagy provides amino acids essential for protein synthesis in response to metabolic stress. Basal autophagy, on the other hand, removes specific substrates, including protein aggregates, damaged organelles, and invading bacteria. This selective pathway uses autophagic receptor proteins for efficient degradation. Autophagic receptors are typically categorized into two different types: ubiquitin binding proteins and organelle membrane proteins. Ubiquitin binding receptors include p62/SQSTM1, neighbor of BRCA1 gene (NBR1), optineurin, nuclear dot protein 52 kDa/calcium binding and coiled-coil domain2 (CALCOCO2), Tax1 binding protein 1, and toll-interacting protein; membrane binding receptors include BCL2/adenovirus E1B 19 kDa interacting protein 3-like (NIX/BNIP3L), BNIP3, FUN14 domain containing 1, and family with sequence similarity 134, member B (FAM134B) (2). These receptors recognize and sort substrates, and recruit core autophagic machinery to the target at existing autophagosome formation sites.

Conserved among metazoans, the ubiquitin binding autophagy receptor p62/SQSTM1 (hereafter referred to as p62) acts as a hub protein in various cellular signaling pathways, including NF-κB, mechanistic target of rapamycin (mTOR), Caspase 8, and nuclear factor erythroid 2-related factor 2 (Nrf2) (3–7). p62 mediates the autophagic degradation of polyubiquitinated substrates *via* direct interaction with microtubule-associated protein light chain 3 (LC3) on the autophagosome (2) (**Figure 1**). p62 forms aggregates by self-oligomerization of its N-terminal Phox and Bem1 (PB1) domain, and associates with ubiquitin through its C-terminal ubiquitin-associated (UBA) domain (**Figure 2**). NBR1 has a remarkable similarity with p62 in the domain architecture, which consists of PB1, zinc finger, LC3-interacting region (LIR), and UBA domains (8). PB1 domain of p62 forms

Figure 1 | p62-mediated selective autophagy. p62 interacts with ubiquitinated proteins through its C-terminal ubiquitin-associated (UBA) domain, and forms aggregates through its N-terminal Phox and Bem1 domain. The resulting protein aggregates are tethered to the autophagosome by direct interaction of p62 and light chain 3 localized on the isolation membrane. The isolation membrane elongates and sequesters p62 and ubiquitin into the autophagosome. The autophagosome fuses with lysosome to form an autolysosome.

self-oligomer or hetero-oligomer with other PB1 proteins including NBR1, while PB1 domain of NBR1 forms hetero-oligomer only. Thus, NBR1 self-interacts through their coiled-coil domain, and cooperates with p62 oligomer in selective autophagy of ubiquitinated substrates (9). There are several examples of posttranslational modifications of p62 for selective autophagy. For example, phosphorylation of the p62 UBA domain at serine 407 by unc-51-like kinase 1 (ULK1) destabilizes the UBA dimer of p62, while sequential phosphorylation of serine 403 by casein kinase 2 (CK2), TANK-binding kinase 1 (TBK1), or ULK1 increases the affinity of UBA for ubiquitin chains (10–12). Ubiquitination of p62 at lysine 420 by the Keap1/Cul3 E3 ligase complex inhibits the dimerization of the UBA domain (13), while ubiquitination at lysine 7 in the PB1 domain by the E3 ligase TRIM21 suppresses the oligomerization of p62 (14).

Upon various cellular stress conditions, p62 functions as a signaling hub *via* characteristic domains, such as the zinc finger (ZZ) domain, TRAF6 binding (TB) domain, LIR, and Keap1-interacting region (KIR), which interacts with RIP kinase, TRAF6, Raptor, LC3, and Keap1, respectively (**Figure 2**). Transcription of p62 is activated by cellular stresses, such as oxidative, metabolic, and pathogenic conditions, and aberrant expression of p62 or defective autophagy causes appearance of large number of p62 aggregates into the cytoplasm (15, 16). Importantly, the aggregate of p62 has been found in a common hallmark in some of serious diseases, such as cancer, alcoholic hepatitis, and neurodegenerative disease (17). The aforementioned diseases are thought to result from loss or dysfunction of p62-regulated cell signaling.

#### PATHOPHYSIOLOGICAL ROLE OF AUTOPHAGY IN CANCER

Early research has indicated a role of autophagy in tumor suppression. Beclin 1, the mammalian ortholog of yeast Atg6, was reported as autophagy-related protein involved in tumorigenesis (18). Specifically, beclin 1 heterozygous-deficient mice were shown to exhibit increased frequency of spontaneous tumorigenesis in the liver, lung, and lymphomas (19, 20). Because systemic knockout of *Atg* genes in mice exhibit neonatal lethality due to deprivation of amino acids in plasma and tissue (21), conditional knockouts of *Atg* genes in mice have been used as a physiological model of autophagy. Intriguingly, mosaic knockouts of *Atg5* or liver-specific knockouts of *Atg7* in mice have demonstrated that autophagy deficiency generates benign tumors in the liver in an age-dependent manner (22, 23). Impairment of autophagy results in accumulation of damaged mitochondria, which are major sources of reactive oxygen species (ROS) resulting in DNA damage, thus contributing to malignant progression (24). Other studies report that reduction of autophagy results in growth suppression of various cancers, such as hepatocyte, pancreatic cancer, lung cancer, breast cancer, colon cancer driven by K-Ras or BRAF mutations (25). On the other hand, studies have found that overexpression of transcription factor EB, a critical regulator of autophagy, promotes cancer growth (26, 27). Autophagydeficient tumors re-growth and form large tumors in allograft, suggesting that growth arrest of autophagy-deficient tumors is canceled through nutrient-generating autophagy in an ectopic environment (28). It has been reported recently that oxidation of p62 promotes its oligomerization *via* disulfide-linked conjugates, followed by activation of autophagy (29). In this pathway, p62 senses the ROS and induces autophagy for cellular homeostasis and cell survival even under the oxidative stress conditions in aging or cancer. Moreover, increased p62 in autophagy-defective cells inhibits RNF168, an E3 ligase for histone H2A activated in response to DNA damage (30). These findings suggest that autophagy-deficient cells abolish DNA repair activity thereby

resulting in tumorigenesis. Autophagy plays a complex role in cancer, and its function can be dependent on the stage, environment, or type of cancer. Taken together, the above findings indicate that while autophagy plays a role in the inhibition of tumorigenesis, it could also facilitate the tumor growth once established, at least in mouse models.

# Keap1–Nrf2 PATHWAY

Keap1–Nrf2 pathway is a critical cytoprotective response mediated by the activation of transcription factor Nrf2 during oxidative or electrophilic stress. Under normal conditions, Nrf2 is constitutively degraded in ubiquitin-proteasome system *via* the interaction with the E3 ubiquitin ligase adaptor protein Kelch-like erythroid cell-derived protein with CNC homology [ECH]-associated protein 1 (Keap1) (31). Keap1 binds to Nrf2 through direct interaction between the double glycine repeat or Kelch repeat (DGR) and the C-terminal region (CTR) of Keap1 (Keap1DC), and the ETGE and DLGex motifs of Nrf2. Known as the hinge and latch model (32–34), Keap1 forms a homodimer with its N-terminal BTB domain in which one protein interacts with high affinity to ETGE and the other with low affinity to DLGex (**Figure 3A**). Oxidative stress or electrophiles trigger a conformational change of Keap1 by modification of certain cysteine residues in Keap1,

proteasome.

higher affinity than phosphor-p62, whereas the DLGex motif of Nrf2 binds to Keap1DC with covering K67. Nrf2 is eventually ubiquitinated and degraded by the

which leads to its dissociation from Nrf2 (35). The released Nrf2 translocates into the nucleus to induce the transcription of antioxidant-responsive element-regulated genes, such as the cytoprotective or metabolic-related genes *NQO1*, *HO-1*, *GCLC*, *GSTm*, which help to protect the cells from oxidative and metabolic stress.

#### p62–Keap1–Nrf2 PATHWAY

In previous reports, we demonstrated that phosphorylation of p62 at serine 349 results in Nrf2 activation (6). p62 also has a KIR motif (349-STGE-352) (36), which allows binding to Keap1DC, but its affinity to Keap1 is significantly lower compared to the ETGE motif of Nrf2. During selective autophagy, p62 is translocated to ubiquitinated targets, such as protein aggregates, depolarized mitochondria, and pathogens, through the phosphorylation of the p62 UBA domain at serine 403 and serine 407 by CK2 and/or TBK1 (10, 11). Subsequently, mTOR complex 1 (mTORC1) phosphorylates p62 at serine 349, which dramatically enhances its interaction with Keap1, since the Keap1 binding affinity of phospho-p62 is higher than that of DLGex motif of Nrf2 (**Figure 3B**). Finally, the DLGex of Nrf2 is competitively displaced by phospho-p62, which results in the dissociation of Nrf2 from Keap1 and robust activation of Nrf2 (p62–Keap1–Nrf2 pathway) (**Figure 3B**) (6). p62–Keap1 and ubiquitinated cargos are eventually removed by selective autophagy. These results indicate that p62-mediated selective autophagy is coupled with the Keap1-Nrf2 system in normal cells. In p62–Keap1– Nrf2 pathway, the cytoprotective effects of Nrf2 activation could be enhanced in concert with the selective degradation of phosphorylated p62 and Keap1 complex.

# DYSREGULATION OF THE p62–Keap1–Nrf 2 PATHWAY IN CANCER

Notably, persistent phosphorylation of mouse p62 at serine 351 (corresponding to serine 349 in human p62) has been found in hepatic adenoma in autophagy-deficient livers and in hepatitis C virus-positive human hepatocellular carcinoma (HCC) (22, 37). Knockout of *p62* in an HCC cell line markedly abrogates tumor growth, whereas forced expression of a phosphorylation-mimic

regulates c-Myc activity *via* the activation of mechanistic target of rapamycin complex 1 (5). Persistent activation of Nrf2 occurs due to the positive feedback loop of Nrf2 in cancer cells (6). Nrf2 activation induces metabolic reprogramming, including the glucuronate pathway (7) and glutathione synthesis (8), while c-Myc enhances glucose and glutamine metabolism (9). The activation of multiple metabolic pathways permits the drug resistance and cell proliferation of cancer.

allele of p62, but not a phosphorylation defective mutant, resulted in recovery of the growth defect (6, 22). These results indicate that the persistent activation of Nrf2 through phosphorylation of p62 is involved in the development of HCC. Importantly, Nrf2 also induces *p62* expression, resulting in the persistent activation of Nrf2 *via* a positive feedback loop in the p62–Nrf2–Keap1 pathway (16). Consistently, an amplified copy number of *p62* on chromosome 5q has been identified in renal cancer, suggesting that *p62* is an oncogene. Furthermore, accumulating evidence demonstrates that the abnormal expression of p62 is associated with malignancy in various cancers, including liver (22, 37), kidney (38, 39), lung (40), breast (41, 42), pancreatic (43), prostate (44, 45), head and neck (46, 47), ovarian (48, 49), oral (50, 51), colon (52, 53), endometrial (54), skin (55, 56), and gastric cancers (57). Indeed, accumulation of phosphorylated p62 has been observed in about half of HCC patients in our studies (6). Somatic mutations of Nrf2 and Keap1 have also been found in cancers at high frequency (58–60); these mutations could cause persistent activation of Nrf2 *via* disrupting the interaction between Nrf2 and Keap1. The above-mentioned lines of evidence suggest that dysregulation of p62–Keap1–Nrf2 pathway is involved in cancer development.

# METABOLIC REPROGRAMMING BY THE p62–Keap1–Nrf2 PATHWAY

Nuclear factor erythroid 2-related factor 2 has been shown to regulate the expression of antioxidant proteins, detoxification enzymes, proteasome subunits, and autophagy-related proteins for oxidative stress response and proteostasis (protein homeostasis) in normal cells. Recent studies demonstrate that Nrf2 increases the expression of multiple enzymes involved in the pentose phosphate pathway, purine nucleotide synthesis, as well as glutathione synthesis, and glutaminolysis in lung cancer, which activates the phosphatidylinositol 3-kinase-Akt pathway (61). In support of these findings, we also found that Nrf2 activation provided metabolic reprogramming of glucose and glutamine through the activation of Nrf2 in HCC harboring phosphorylated p62 (p-S349) (**Figure 3B**), which led to increased cell proliferation and resistance to anti-cancer agents of HCC (37) (**Figure 4**). Taken together, these findings suggest that molecular targeting of p62 represents a potential chemotherapeutic approach against HCC.

By chemical screening, we have identified an inhibitor for the Keap1-phosphorylated p62 (p-S349) protein–protein interaction the acetonyl naphthalene derivative K67 (37) (**Figure 3C**). Structural analysis demonstrated that K67 binds to a Keap1DC pocket, which is the binding site of phosphorylated p62, Nrf2- ETGE, and Nrf2-DLGex. Treatment of HCC with K67 suppressed proliferation and reduced tolerance to cisplatin or sorafenib (37). Levels of p62 accumulation and c-Myc expression are reportedly associated with high risk for tumor recurrence and poor prognosis of HCC patients (62). Further, it has been demonstrated that high p62 expression in non-tumor tissue is required for transformation to HCC, which was caused by the activation of Nrf2, mTORC1, and c-Myc (62) (**Figure 4**). These results are consistent with the anti-proliferative and anti-malignant effects found in autophagy-deficient tumor cells (63). More recently, Karin and colleagues reported that p62-mediated activation of Nrf2 triggers mouse double minute 2 homolog (MDM2) expression in premalignant pancreatic intraepithelial neoplasia 1, resulting in development and malignancy of pancreatic ductal adenocarcinoma (43) (**Figure 5**).

pre-malignant tumor cells (1). p62 is phosphorylated (2), and subsequently nuclear factor erythroid 2-related factor 2 (Nrf2) is activated (3). Mouse double minute 2 homolog (MDM2) induced by Nrf2 results in the increase of Notch signaling (4) and the reduction of p53, followed by cell malignancy (5). Abbreviations: PanIN, premalignant pancreatic intraepithelial neoplasia; PDAC, pancreatic ductal adenocarcinoma.

## CONCLUSION AND FUTURE PERSPECTIVES

In recent years, there are many reports showing that p62– Keap1–Nrf2 pathway plays a protective role in oxidative and stress conditions. For example, ER stress-induced apoptosis is prevented by the activation of p62–Keap1–Nrf2 pathway (64). Quercetin attenuates the hepatotoxicity induced by a various hepatotoxicants, through the activation of p62–Keap1–Nrf2 pathway (65). Licochalcone A activates p62–Keap1–Nrf2 pathway and suppresses arthritis in a collagen-induced arthritis mouse model (66). Trehalose induces p62 expression and activates p62–Keap1–Nrf2-mediated antioxidant response during oxidative stress (67). Meanwhile, persistent activation of p62– Keap1–Nrf2 pathway has been shown to be involved in liver tumorigenesis in mice (6, 22). More recently, we demonstrated that overexpression of a p62 variant lacking KIR mitigates Nrf2 activation (68). These results suggest that phopsho-p62 could be a novel target for cancer therapy as described in

## REFERENCES


Ref. (37). Many kinase inhibitors have been used in cancer therapy. However, the protein kinases responsible for the phosphorylation of p62 in tumor have not yet been identified. Further studies are required to identify the regulatory factors involved in the p62–Keap1–Nrf2 pathway.

## AUTHOR CONTRIBUTIONS

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

## FUNDING

YI is supported by a Grant-in-Aid for Scientific Research (B) (16H05137, to YI). MK is supported by a Grant-in-Aid for Scientific Research on Innovative Areas (25111006, 25111001, and 15K21749 to MK), a Japan Society for the Promotion of Science (an A3 foresight program, to MK), and Japan Agency for Medical Research and Development (AMED) (17cm0106101h0002 to MK).


sensitivity to phosphatidylinositol 3-kinase pathway inhibitors. *PLoS One* (2014) 9(3):e90171. doi:10.1371/journal.pone.0090171


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

*Copyright © 2018 Ichimura and Komatsu. 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 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.*

# Ataxia-Telangiectasia Mutated Kinase in the Control of Oxidative Stress, Mitochondria, and Autophagy in Cancer: A Maestro with a Large Orchestra

*Venturina Stagni1,2\*, Claudia Cirotti1,2 and Daniela Barilà1,2\**

*1Department of Biology, University of Rome "Tor Vergata", Rome, Italy, 2 Laboratory of Cell Signaling, Istituto di Ricovero e Cura a Carattere Scientifico (IRCCS) Fondazione Santa Lucia, Rome, Italy*

#### *Edited by:*

*Giuseppe Filomeni, Danish Cancer Society, Denmark*

#### *Reviewed by:*

*Apolinar Maya-Mendoza, Danish Cancer Society, Denmark Shiaw-Yih Lin, University of Texas MD Anderson Cancer Center, United States*

#### *\*Correspondence:*

*Venturina Stagni venturina.stagni@gmail.com; Daniela Barilà daniela.barila@uniroma2.it*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

*Received: 01 February 2018 Accepted: 02 March 2018 Published: 16 March 2018*

#### *Citation:*

*Stagni V, Cirotti C and Barilà D (2018) Ataxia-Telangiectasia Mutated Kinase in the Control of Oxidative Stress, Mitochondria, and Autophagy in Cancer: A Maestro With a Large Orchestra. Front. Oncol. 8:73. doi: 10.3389/fonc.2018.00073*

Ataxia-telangiectasia mutated kinase (ATM) plays a central role in the DNA damage response (DDR) and mutations in its gene lead to the development of a rare autosomic genetic disorder, ataxia telangiectasia (A-T) characterized by neurodegeneration, premature aging, defects in the immune response, and higher incidence of lymphoma development. The ability of ATM to control genome stability several pointed to ATM as tumor suppressor gene. Growing evidence clearly support a significant role of ATM, in addition to its master ability to control the DDR, as principle modulator of oxidative stress response and mitochondrial homeostasis, as well as in the regulation of autophagy, hypoxia, and cancer stem cell survival. Consistently, A-T is strongly characterized by aberrant oxidative stress, significant inability to remove damaged organelles such as mitochondria. These findings raise the question whether ATM may contribute to a more general hijack of signaling networks in cancer, therefore, playing a dual role in this context. Indeed, an unexpected tumorigenic role for ATM, in particular, tumor contexts has been demonstrated. Genetic inactivation of Beclin-1, an autophagy regulator, significantly reverses mitochondrial abnormalities and tumor development in ATM-null mice, independently of DDR. Furthermore, ATM sustains cancer stem cells survival by promoting the autophagic flux and ATM kinase activity is enhanced in HER2-dependent tumors. This mini-review aims to shed new light on the complexity of these new molecular circuits through which ATM may modulate cancer progression and to highlight a novel role of ATM in the control of proteostasis.

Keywords: ataxia-telangiectasia mutated kinase, oxidative stress, mitophagy/autophagy, cancer, proteostasis

# INTRODUCTION

#### Redox Homeostasis

Reactive oxygen species (ROS) are physiologically by-products of cellular metabolism and play a central role in many physiological and pathological processes including inflammation and chronic diseases such as atherosclerosis and cancer, underscoring the importance of investigating cellular pathways involved in redox homeostasis (1, 2).

Main sources of ROS are enzymes and organelles such as mitochondria (3). About 2–4% of oxygen consumed by mitochondrial oxidative phosphorylation is partially reduced and flows through membranes to activate signaling pathways that have then to be promptly turned off. Intracellular enzymatic and non-enzymatic antioxidant defense is responsible for redox homeostasis, preventing ROS accumulation (4). Together with ROS, reactive nitrogen species (RNS) are harmful molecules mostly generated by spontaneous reaction between ROS and nitric oxide signaling molecule (5).

Reactive oxygen species and RNS damage proteins as well as cellular organelles; therefore, several systems evolved to regulate and preserve a functional cellular protein pool, to ensure the quality and functionality of cellular organelles, and to finally guarantee the maintenance of proteostasis (3, 6). The autophagylysosomal machinery (7), the ubiquitin–proteasomal system (8), and molecular chaperones, including heat shock proteins (HSPs) (9, 10), cooperate to this aim and, indeed, they are all finely regulated by oxidative stress, which augments their functionality in order to support proteostasis and organelle quality control in challenging conditions (6).

#### Ataxia-Telangiectasia Mutated Kinase (ATM) and Oxidative Stress Response

Ataxia-telangiectasia mutated kinase is a serine/threonine protein kinase, and it is a well-characterized tumor suppressor gene, which plays a central role in the nucleus in the DNA damage response (DDR). In humans, loss of function in ATM results in ataxia telangiectasia (A-T), a pleiotropic disease whose hallmarks include neurodegeneration, cancer-proneness, premature aging, radio-sensitivity, metabolic, and immune dysfunctions (11). For many years, the defect in DNA-damage response has been considered the solely responsible for A-T phenotype.

Increasing numbers of reports have described elevated readouts of oxidative stress in plasma of A-T patients, in cultured A-T fibroblasts and lymphocytes, and in tissues and cultured cells from Atm-deficient mice (12, 13). Notably, the response of A-T fibroblasts to induced oxidative stress was found defective [reviewed in Ref. (14)].

Consistently with the loss of redox homeostasis, mitochondria, are severely compromised in A-T appearing swollen and with disrupted cristae structure; as a consequence, A-T cells display mitochondrial ROS overproduction and decreased ATP levels (15). Interestingly, some of the pathological phenotypes identified in A-T, including insulin resistance, premature aging, and neurodegeneration cannot be easily connected to the well-known role of ATM in DDR, while conversely, they could be linked to the interplay between ATM and ROS (16, 17). More importantly, the administration of antioxidants to *Atm*<sup>−</sup>/<sup>−</sup> mice ameliorates the disease progression and delayed cancer development (thymic lymphomas), by reducing ROS and restoring mitochondrial membrane potential (18).

These observations were at first puzzling and, more recently, they could be linked to a role of ATM in regulating cellular oxidative stress signaling. In particular, ATM is activated in the cytosol by ROS through the formation of ATM dimers *via* disulfide bonds (16, 19). Downstream to oxidative stressdependent activation, ATM regulates a number of processes to promote restoration of redox homeostasis including adjustment of glutathione levels and activation of pentose phosphate pathway (20), regulation of mitochondrial mass, function and turnover (15, 21, 22), removal of peroxisomes *via* autophagy (23). More recently, ATM activation in response to oxidative stress has been shown to be involved in the control of proteostasis, preventing protein aggregation through a still unknown mechanism (24).

## ATM AND AUTOPHAGY

The autophagy system is a finely regulated catabolic process responsible for the selective removal of cytoplasmic components (i.e., proteins, aggregates, or whole organelles) properly targeted by posttranslational modifications (ubiquitination). Basal autophagy physiologically occurs to ensure proteins turnover, maintaining intracellular homeostasis. Moreover, the autophagy system is activated by oxidative stress triggered by endogenous and exogenous stressors including nutrient starvation, hypoxia, and mitochondria and peroxisome dysfunction (25).

Ataxia-telangiectasia mutated kinase is activated in the cytosol by all the conditions listed above (16, 26); moreover, it has a role in autophagy induction (22, 27). It has been clearly demonstrated that ATM sustains autophagic pathway by inhibiting the negative regulator mTOR complex 1 (mTORC1). At the molecular level, ATM activation upon oxidative and/or nitrosative stress is responsible for the activation of LKB1/AMPK/TSC2 signaling axis, culminating with mTORC1 inhibition and relieving its repression on ULK1, which is the key protein responsible for the nucleation and formation of the autophagosome membrane, further activated by AMPK-mediated phosphorylation. This signaling pathway starting from ATM culminates in autophagy flux induction (22, 27).

The same pathway is also activated by ATM upon ROS induction under hypoxia (28). In this context, ATM promotes HIF1a stabilization by direct phosphorylation on Ser696, culminating on mTORC1 inhibition (28). Consistently, under hypoxic conditions, ATM-deficient cells fail to activate HIF1a and to inhibit mTORC1, further supporting the requirement for ATM in this pathway (28). Evidence for a role of ATM in the modulation of HIF-1a basal expression has also been provided (29, 30).

Finally, a recent work suggested that ATM regulates autophagy also by sustaining the levels and activity of ATG4C protease in cancer cells grown as mammospheres (31), characterized by low ROS levels (32). Interestingly, ATG4 proteases are the only ATG members that act as oxidative stress sensors (33). It has been demonstrated that oxidative signal leads to inactivation of ATG4s by oxidation of essential cystein residues on these proteins, at the site of autophagosome formation, thereby promoting lipidation of ATG8, an essential step in the process of autophagy (33). These data suggest that the ATM–ATG4C axis may represent a new molecular link that connects ROS, ATM, and autophagy signaling (31).

Overall, these publications suggest a role of ATM in the cytosol in regulating autophagosome formation upon exogenous and endogenous oxidative stress.

# ATM IN SELECTIVE AUTOPHAGY: MITOPHAGY AND PEXOPHAGY

The main source of intracellular ROS are metabolically active organelles, such as mitochondria and peroxisomes (34, 35). Not surprisingly, ATM localizes to both these compartments to sense ROS increase and to activate pro-survival or pro-death intracellular pathways, depending on the intensity of the stimuli (15, 23, 36). The role of ATM in preserving mitochondrial functionality is well documented since many years. *In vivo,* loss of ATM results in mitochondria abnormalities causing ROS overproduction, strong decrease in ATP levels, and ultrastructural alterations. Moreover, the selective removal of damaged mitochondria, process known as mitophagy, is strongly impaired causing the accumulation of dysfunctional organelles (15). More recently, these evidences have been recapitulated also in neuroblastoma cells: ATM depletion results in a similar mitochondrial phenotype and mitophagy alteration, partially rescued by NAD+ cofactor replenishment (37). Taken together, these papers demonstrate the relation between ATM and mitochondria.

Although the molecular mechanism responsible for ATM function in the control of mitochondrial homeostasis deserves further investigation, it has been demonstrated that ATM activation upon mitochondrial stress or ROS increase protects cells from damage; indeed, ATM-mediated modulation of the wellcharacterized PINK1–Parkin pathway promotes the elimination *via* mitophagy of altered mitochondria (38).

Very recently, ATM localization to peroxisomes and its role in peroxisomes selective removal, named pexophagy, has been described. As for mitochondria, ATM localizes to peroxisomes probably to sense ROS increase and prevent damage. ATM localization in peroxisomes outer membrane is mediated by its interaction with PEX5, a peroxisome import receptor. Upon peroxisomal ROS increase, ATM-mediated PEX5 phosphorylation targets PEX5 for mono-ubiquitination and recognition by autophagic-adaptor protein (such as p62), incorporating dysfunctional organelles into autophagic vescicles (23). Very interestingly, pexophagy defects observed in ATM-deficient cells are rescued by reconstitution of ATM expression, confirming the direct role of ATM in this response (23).

The removal of damaged organelles described, so far, is also sustained by ATM-dependent induction of general autophagy, as ATM inhibits the autophagy negative regulator mTORC1, sustaining ULK1 pro-autophagic protein activation as described above (22).

Taken all together, these evidences highlight a relevant role of ATM in the cytosol: ATM ensures a prompt reply to ROS increase by activating autophagy, mitophagy, and pexophagy in order to preserve proteostasis and cellular homeostasis.

#### ROS-DEPENDENT ATM ACTIVATION AND CANCER

Elevated rates of ROS have been detected in almost all cancers, where they promote many aspects of tumor development and progression (39, 40). In cancer cells, high levels of ROS can result from increased metabolic activity, mitochondrial dysfunction, peroxisome activity, increased cellular receptor signaling, oncogene activity, increased activity of oxidases, cyclooxygenases, lipoxygenases, and thymidine phosphorylase, or through crosstalk with infiltrating immune cells (41). Moreover, ROS deregulation in low oxygen tension or hypoxia condition is a common feature of all solid tumors, it is strongly associated with tumor development, malignant progression, metastatic outgrowth, and resistance to therapy and it is considered an independent prognostic indicator for poor patient prognosis in various tumor types (42). It has been largely demonstrated that ROS increase leads to proteome oxidation and instability, and alteration of the proteostasis control machine (9). More interestingly, in order to survive under stress conditions (i.e., ROS increase/hypoxia condition/starvation), many cancer cells adapt their proteostasis network and become uniquely dependent on it, an example of non-oncogene addiction (43). Individual nodes of the proteostasis network, such as Hsp90 and other HSP chaperones involved in the protein quality control networks, are currently exploited as drug targets in cancer and entered in clinical trials (44, 45).

The identification of new cytoplasmic signaling mediated by ATM in response to oxidative stress (46) and the finding that ATM can regulate networks that ensure proteins and organelles quality open the question whether these networks may contribute to A-T pathogenesis and to cancer progression (16).

It has been hypothesized for a long time that higher cancer predisposition of A-T patients depends exclusively on defects in ATM-dependent-DDR, which leads to genomic instability (11). Unexpectedly, allelic loss of the autophagy regulator Beclin-1, significantly delayed tumor development in ATM-null mice. This effect was not associated to the rescue of DNA damage signaling but rather to a significant reversal of the mitochondrial abnormalities (15). Accordingly, it has been also demonstrated that Rapamycin (mTOR inhibitor) and antioxidant treatments rescue ATM-dependent lymphomagenesis, suggesting that the dysregulation of mTORC1 and ROS contribute to A-T pathology (22). Moreover, suppression of ATM may significantly contribute to the activation of mTORC1 observed in hypoxic tumors and can promote tumor cell survival through autophagy regulation (28). Importantly, autophagy is a dichotomous phenomenon, involved in cell growth as well as in cell death, depending on its magnitude and on the cell context (47). Autophagy, as DNA damage, has been proposed to play a tumor-suppressive role in the early stages of tumorigenesis and, indeed, it is upregulated by several tumor suppressor genes; however, above a certain threshold, autophagy can also induce cell death and, if triggered appropriately, can be used as a means of killing cancer cells (48). Paradoxically, it was recently published that autophagy promotes the stem-like phenotype in breast cancer, suggesting a controversial role in cancer of autophagy (49). Interestingly, it has been reported that ATG4A and Beclin1 autophagic genes are upregulated in breast cancer stem cells (BCSCs) and are essential genes involved in BCSCs formation and maintenance (50, 51). Overall, these papers support the idea that BCSCs utilize autophagy for survival and growth, suggesting that, in this context, autophagy promotes tumor progression and tumor relapse acting as a tumor-promoting signaling. Interestingly, it was recently demonstrated that ATM kinase

could have a pro-survival role in BCSCs through regulation of ATG4C gene and autophagy (31).

Finally, the expression of HSP90, a central player in the control of proteostasis, increases under stress conditions (as ROS accumulation upon oxidative stress) and, it is exploited by cancer cells to support the stability and the aberrant activity of oncoproteins overexpressed or mutated in malignancy including HER2, BCR-ABL, and EGFR (45, 52). According to this observation, HSP90 is one of the most actively pursued cancer drug targets and several different HSP90 inhibitors entered in clinical trials so far (52). Growing evidences support the idea that ATM could regulate HSP90 activity. ATM kinase can directly phosphorylate HSP90 (53, 54) although the significance of these posttranslational modifications is still largely unknown. More interestingly, we recently demonstrated that ATM activity sustains HSP90 interaction with its client protein HER2, promoting its stabilization and, therefore, sustaining HER2-dependent tumorigenicity (55, 56). These data suggest a new connection between ATM kinase and HSP90 chaperone: ATM may contribute to the control of protein quality and stability and could also modulate tumor progression *via* the regulation of this heat shock protein.

# CONCLUSION

In conclusion, although the canonical role of ATM in the management of DNA damage defines ATM as a tumor suppressor gene, the identification of several novel functions of ATM, mostly related to its activation in response to oxidative stress and to its ability to modulate the cellular response to this insult, support multiple roles of ATM in cancer (**Figure 1**). ATM-dependent regulation of autophagy, mitophagy, pexophagy, and proteostasis suggest the idea that the effect of ATM expression and activity in cancer may be the result of its multiple functions in several signaling pathways and may, therefore, be strictly dependent on the specific cellular context. More studies are urgently needed to ascertain the molecular mechanisms through which this panel of cytosolic functions of ATM could modulate cancer development and therapy.

# AUTHOR CONTRIBUTIONS

VS and DB came up with the topic for this mini-review; VS, DB, and CC wrote and edited the text.

# FUNDING

This work has been supported by research grants to Associazione Italiana per la Ricerca sul Cancro-AIRC (IG8706), by Ministero dell'Istruzione, dell'Università e della Ricerca, JPI HDHL NUTRICOG MiTyrAge, and PRIN\_2015LZE9944\_005 to DB; VS has been supported by AIRC (IG8706) and CC by JPI HDHL NUTRICOG MiTyrAge project.

1. Jones DP. Redefining oxidative stress. *Antioxid Redox Signal* (2006) 10:1865–79.


REFERENCES

doi:10.1089/ars.2006.8.1865


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

The reviewer AM-M and the handling editor declared their shared affiliation.

*Copyright © 2018 Stagni, Cirotti and Barilà. 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 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.*

# Involvement of NADPH Oxidase 1 in Liver Kinase B1-Mediated Effects on Tumor Angiogenesis and Growth

*Elisabetta Zulato1 , Francesco Ciccarese2 , Giorgia Nardo1 , Marica Pinazza1 , Valentina Agnusdei1 , Micol Silic-Benussi1 , Vincenzo Ciminale1,2 and Stefano Indraccolo1 \**

*<sup>1</sup> Istituto Oncologico Veneto IOV – IRCCS, Padova, Italy, 2Department of Surgery, Oncology and Gastroenterology, University of Padova, Padova, Italy*

The liver kinase B1 (*LKB1*) gene is a tumor suppressor with an established role in the control of cell metabolism and oxidative stress. However, whether dis-regulated oxidative stress promotes growth of LKB1-deficient tumors remains substantially unknown. Through *in vitro* studies, we observed that loss of LKB1 perturbed expression of several genes involved in reactive oxygen species (ROS) homeostasis. In particular, this analysis evidenced strongly up-modulated NADPH oxidase 1 (*NOX1*) transcript levels in tumor cells lacking LKB1. NOX1 accounted in part for enhanced cytotoxic effects of H2O2 induced oxidative stress in A549 LKB1-deficient tumor cells. Notably, genetic and pharmacologic inhibition of NOX1 activity reduced angiogenesis and growth of A549 tumors in mice. These results suggest that NOX1 inhibitors could counteract ROS production and the angiogenic switch in LKB1-deficient tumors.

#### *Edited by:*

*Giuseppe Filomeni, Danish Cancer Society, Denmark*

#### *Reviewed by:*

*Alfredo Criollo, Universidad de Chile, Chile Zhi-Xiang Xu, University of Alabama at Birmingham, United States*

*\*Correspondence:*

*Stefano Indraccolo stefano.indraccolo@unipd.it*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

*Received: 09 February 2018 Accepted: 14 May 2018 Published: 04 June 2018*

#### *Citation:*

*Zulato E, Ciccarese F, Nardo G, Pinazza M, Agnusdei V, Silic-Benussi M, Ciminale V and Indraccolo S (2018) Involvement of NADPH Oxidase 1 in Liver Kinase B1-Mediated Effects on Tumor Angiogenesis and Growth. Front. Oncol. 8:195. doi: 10.3389/fonc.2018.00195*

Keywords: NADPH oxidase 1, liver kinase B1, angiogenesis, lung cancer, reactive oxygen species, oxidative stress

#### INTRODUCTION

Increased oxidative stress is common in cancer cells. In fact, uncontrolled proliferation is connected with profound metabolic remodeling accompanied by altered redox status. Reactive oxygen species (ROS) are important biological messengers involved in cell signaling, tumor growth, and metastasis (1). Recent studies suggest that increased oxidative stress could be exploited for therapeutic purposes (2, 3), as tumor cells might be particularly vulnerable to drugs which further increase oxidative stress or otherwise block biochemical buffers of ROS (4).

The liver kinase B1 (LKB1) is a serine threonine kinase, whose hereditable mutations are associated with the Peutz-Jeghers syndrome and is somatically mutated or deleted in certain tumor types, with relatively high frequencies in non-small cell lung cancer (NSCLC) and cervical carcinoma (5). LKB1 is involved in energy homeostasis control, as well as control of cell proliferation and polarity (6). LKB1 and its downstream effector AMP-activated protein kinase (AMPK) have also been recently involved in metabolic adaptation and in the regulation of tumor cell response to oxidative stress by various mechanisms including control of NADPH homeostasis (7) and enhancement of the activity of superoxide dismutase-2 and catalase by a p38-mediated mechanism (8). These studies largely utilized LKB1-deficient tumor cell lines and their matched LKB1-proficient counterpart to

**Abbreviations:** LKB1, liver kinase B1; NOX1, NADPH oxidase 1; ROS, reactive oxygen species; NSCLC, non-small cell lung cancer; AMPK, AMP-activated protein kinase; ACC, acetyl-CoA carboxylase; CCLE, cancer cell line encyclopedia; NAC, *N*-acetyl cysteine; MVD, microvessel density; NRF2, nuclear respiratory factor 2; bFGF, basic fibroblast growth factor; VEGF, vascular endothelial growth factor; VEGFR, vascular endothelial growth factor receptor.

investigate effects of this kinase on oxidative stress. Although several features and mechanisms associated with altered ROS homeostasis in LKB1-mutant tumor cells have been uncovered by using this experimental system, previous studies did not analyze the transcriptional perturbations of genes involved in redox control in LKB1-deficient cell lines.

Here, we investigated at transcriptional level three pairs of LKB1-deficient cell lines and observed that LKB1 is involved in the regulation of NADPH oxidase 1 (*NOX1*) transcript expression. Inhibition of NOX1 activity attenuated cytotoxic effects of oxidative stress in LKB1-deficient tumor cells *in vitro* and was an effective strategy to prevent the angiogenic switch and growth of LKB1-deficient A549 tumors *in vivo*. Although marked heterogeneity in the expression levels of *NOX1* among LKB1-deficient lung and cervical cancer cell lines does not allow us to generalize these findings, our results hint at NOX1 as a candidate mediator of angiogenesis in a subset of LKB1-deficient tumors.

#### MATERIALS AND METHODS

#### Cell Culture and *In Vitro* Treatments

A549 cells were provided by Dr. Reuben J. Shaw (Salk Institute for Biological Studies, La Jolla, CA, USA). H460, H1650, H1975, HeLa, SiHa, and HEK293T cells were purchased from ATCC (Manassas, VA, USA). H23, H1437, H2228, H2009, and H1299 cell lines were kindly provided by Dr. Massimo Broggini (Mario Negri Institute, Milan, Italy). HeLa, SiHa, and HEK293T cells were cultured in Dulbecco modified Eagle medium (DMEM; EuroClone, Milan, Italy), supplemented with 10% FCS (EuroClone), 10 mM HEPES, and 1% antibiotic–antimycotic mix (Thermo Fisher, Waltham, MA, USA). H460, H1650, H1975, H23, H1437, H2228, H2009, and H1299 cells were grown in RPMI1640 medium (EuroClone) supplemented with 10% FCS, 1% HEPES, 1% Na pyruvate, 2 mM l-Glutamine (Lonza, Basel, Switzerland), and 1% antibiotic–antimycotic mix. A549 cells were cultured in DMEM-F12 medium (Gibco-BRL, Grand Island, NY, USA) supplemented with 10% FCS, 2 mM l-Glutamine, and 1% antibiotic–antimycotic mix. In some experiments, cell lines were treated with the appropriate vehicle or with the following substances: 2 mM hydrogen peroxide (H2O2, Sigma-Aldrich, Saint Luis, MO, USA), 2 mM *N*-acetylcysteine (NAC, Sigma-Aldrich), 100 µM apocynin (Sigma-Aldrich), for the indicated time points.

#### Retrovirus Production and Transduction With Viral Particles

The retroviral vector overexpressing human LKB1wt (RV-LKB1wt; plasmid # 8592) and the corresponding control vector (pBABE, plasmid # 1764) used to transduce LKB1-mutated cells were purchased from Addgene (Cambridge, MA, USA). Viral particles were produced by transfecting HEK293T cells with second-generation packaging plasmids. Virus was harvested 24 and 48 h after transfection and concentrated by ultracentrifugation at 24,000 rpm for 2 h. A549, H460, and HeLa cells were transduced with a VSV-G-pseudotyped retroviral vector encoding LKB1 cDNA (LKB1wt) or with an empty vector (pBABE). Briefly, 0.5 × 106 cells were incubated at 37°C overnight and 200 µl of concentrated vector-containing supernatant were layered over target cells. The day after, cells were washed and cultured in fresh medium. Cells expressing the constructs were selected in puromycincontaining medium (8 µg/ml for A549 and H460 cells, 4 µg/ml for HeLa cells) for 10–14 days prior to subsequent analysis.

#### Reverse Transcription-PCR and Quantitative PCR

Total RNA was isolated using RNAeasy® Mini Kit (Qiagen, Venlo, Netherlands) according to manufacturer's instructions. cDNA was synthesized from 0.3 to 1 µg of total RNA using High Capacity RNA-to-cDNA Kit (Thermo Fisher). Expression levels of 20 oxidative metabolism-related genes were analyzed by real-time Ready custom panels (Roche Diagnostics, Basel, Switzerland). Real-time PCRs were performed using Platinum® SYBR® Green (Thermo Fisher) in an ABI Prism 7900 HT Sequence Detection System (Thermo Fisher). Results were analyzed using the ΔΔCt method with normalization against HMBS gene expression.

Genes included in Custom Panels were selected based on their role in oxidative metabolism and are reported in **Table 1**. Primers used for qRT-PCR are reported in **Table 2**.




#### Annexin-V/Propidium Iodide Assay

Cells were incubated with Annexin-V-FITC in Hepes buffer containing propidium iodide (PI), using the Annexin-V Fluos Staining Kit (Roche Diagnostics, Mannheim, Germany). Labeled cells were analyzed on EPICS-XL cytofluorimeter using Expo32 software (Beckman Coulter, Fullerton, CA, USA). Dot plot images were obtained by using the Kalusa analysis 1.3 software (Beckman Coulter).

#### Measurement of Mitochondrial Membrane Potential (**Δψ**)

Cells were incubated with tissue culture medium containing 15 nM tetramethylrhodamine, methyl ester (TMRM; Thermo Fisher) and 20 ng/ml Verapamil (efflux pumps inhibitor; Sigma-Aldrich) for 15 min at room temperature. Labeled cells were analyzed with an LSRII flow cytometer using the 560 nm laser and the 585/15 nm filter.

#### RNAi Experiments

Cells (3 × 105 ) were seeded in six-well tissue culture plates. The following day, Opti-MEM® I Medium with GlutaMAX™ (Thermo Fisher) was mixed separately with 20 nM Stealth RNAi™ siRNA (Thermo Fisher) and with Lipofectamine® RNAiMAX Transfection Reagent (Thermo Fisher) and incubated 5 min at room temperature. Separated mix were combined and, following additional 20 min of incubation, 500 µl of final mix were added, drop by drop, to adherent cells, along with 1.5 ml of culture medium. Lipofection was blocked after 6 h, through removal of medium with liposomes and addition of culture medium. Sequences of Stealth RNAi™ siRNAs are reported in **Table 3**.

#### *In Vivo* Experiments

Procedures involving animals and their care conformed to institutional guidelines that comply with national and international laws and policies (EEC Council Directive 86/609, OJ L 358, 12 December, 1987) and were authorized by the ethical committee of the University of Padova and the Italian Ministry of Health (Authorization no. 143/2012-B). For tumor establishment, both dorsolateral flanks of 8-week-old SCID mice (Charles River, Wilmington, MA, USA) were injected subcutaneously with 0.5 × 106 LKB1mut A549 tumor cells mixed at 4°C with liquid Matrigel (Becton-Dickinson). Tumor volume (mm3 ) was measured by caliber and calculated according to the following formula: *L* × *l* <sup>2</sup> × 0.5, where *L* is the longest diameter, *l* is the shortest diameter, and 0.5 is a constant to calculate the volume of an ellipsoid. In the experiment with NAC, LKB1mut A549 cells were 16 h pretreated with NAC 2 mM before injection and during tumor growth NAC was administered intraperitoneally Table 3 | Stealth siRNAs used in transfection experiments.


at 150 mg/kg every day. Control mice received intraperitoneal injections of PBS. In the experiment in which expression level of *NOX1* was reduced by siRNA-mediated silencing, LKB1mut A549 cells were injected 48 h after lipofection. In the experiment with apocynin, LKB1mut A549 cells were 16 h pretreated with apocynin at 100 µM before injection and during tumor growth apocynin was administered *via* oral gavage at 16 mg/kg every day.

At the end of the experiment, mice were killed by cervical dislocation. For all experiments, five to six mice per group were used. The tumors were harvested by dissection and snap-frozen.

#### Immunofluorescence Analysis (IFA)

Tumor vessels were labeled with a rat anti-CD31 mAb (1:50 dilution; Becton-Dickinson) followed by staining with a goat anti-rat 546 secondary antibody (Thermo Fisher). Sections of 5 µm were cut from frozen biopsies and microvessel density (MVD) was quantified by screening the CD31-stained sections for the areas of highest vascularity, as previously reported (9). Briefly, nuclei were stained with TO-PRO-3 Iodide (1:1,000 dilution; Thermo Fisher). Immunofluorescence signals were visualized on a Zeiss Axiovert 100 M confocal microscope (Carl Zeiss AG; Oberkochen, Germany), using argon (488 nm) and helium-neon (543–633 nm) laser sources. For each sample, the number of fields analyzed varied between 5 and 10 per sample, depending on the size of sections; at least 4 samples per group were analyzed. Images were collected at a magnification of 200×. MVD was quantified by screening for the areas of highest vascularity. Two independent operators blindly evaluated the images: morphologically identifiable vessels were counted as an individual vessel.

#### Statistical Analysis

Results were expressed as mean value ± SD. Student's *t*-test or Mann–Whitney test were used depending on the distribution of values. Differences were considered statistically significant when *P* < 0.05. For each cell line, data are expressed as fold change arbitrarily setting controls at 1. If not specified otherwise, the symbol § was used for comparison between LKB1mut and LKB1wt cells, the symbol \* for comparison between treated and untreated cells and the symbol # for comparison between treated cells and cells pre-treated with other drugs (NAC, Apocynin) or siRNA.

#### RESULTS

## LKB1 Regulates Expression of Genes Involved in ROS Metabolism in Tumor Cells

To investigate whether LKB1 modulated expression of genes involved in the response to cellular redox stress, we compared expression levels of 20 relevant genes (**Table 1**) by real-time PCR in human tumor cell lines isogenic for LKB1 *status* and grown *in vitro* under standard conditions. We performed these experiments using the LKB1-deficient NSCLC cell lines A549 and H460, and cervical carcinoma cell line HeLa transduced with a control retroviral vector (hereafter referred to as LKB1mut) and their LKB1-proficient counterparts obtained by retroviral transduction of wild-type LKB1 cDNA (hereafter referred to as LKB1wt). As expected, LKB1 reconstitution fully restored AMPK and acetyl-CoA carboxylase phosphorylation under glucose starvation—a condition known to trigger LKB1-dependent AMPK activation and associated with oxidative stress [(10) and data not shown]. Remarkably, in all cell lines tested, several antioxidant genes were downregulated and others were upregulated following reconstitution of LKB1 (**Figure 1**). Most of these variations were, however, not shared by the various cell lines analyzed. The most consistently downregulated gene in LKB1wt cells was *NOX1*, a homolog of the macrophage *NOX2*, accounting for the respiratory burst in response to pathogens (11). NADPH oxidases catalyze the transfer of one electron from NADPH to oxygen, generating superoxide or H2O2, thus further increasing oxidative stress (12).

Further analysis of *NOX1* mRNA levels in nine NSCLC and two cervical cancer cell lines, with known *LKB1* mutational status, showed that almost all cell lines bearing *LKB1* mutations presented increased *NOX1* expression compared with *LKB1* wildtype cells (**Figure 2A**). However, broad heterogeneity in *NOX1* expression was detected in LKB1-mutant cell lines, being *NOX1* levels particularly high in A549 cells (**Figure 2A**). An additional *in silico* analysis was performed by matching the *LKB1* mutational status and *NOX1* mRNA transcript data of tumor cell lines, available at the COSMIC and Cancer Cell Line Encyclopedia web sites, respectively. Results showed that *NOX1* transcripts were significantly increased in LKB1-mutated NSCLC cell lines, thus strengthening our observations (**Figure 2B**). The relatively high expression of *NOX1* in A549 cells led us to focus on these cells to investigate the possible involvement of NOX1 in the regulation of oxidative stress.

#### NOX1 Accounts for Increased Sensitivity to Exogenous Oxidative Stress in A549 Cells

Gene silencing and pharmacologic inhibition approaches were used to analyze whether NOX1 could account for increased sensitivity to oxidative stress in LKB1mut cells. Interestingly, siRNA-mediated *NOX1* silencing in A549 cells increased resistance of LKB1mut A549 cells to H2O2-induced cell death, while leaving substantially unaffected response to oxidative stress in LKB1wt cells (**Figure 3A**). Consistent with this, measurement of mitochondrial membrane potential showed a significant increase in the number of TMRM negative LKB1mut A549 cells upon H2O2 treatment (**Figure 3B**). This phenomenon in the mitochondrial membrane potential was partially rescued by *NOX1* silencing. Similar results were obtained following NOX1 inhibition with apocynin (**Figure 3C**). It should be noted that despite interference with NOX1 activity, LKB1mut cells showed increased H2O2-induced cell death compared with LKB1wt cells (**Figures 3A–C**), probably

Figure 1 | Liver kinase B1 (LKB1) *status* affects expression of oxidative metabolism-related genes. Expression profile of selected genes in isogenic pairs of A549 (top), H460 (middle), and HeLa (bottom) cells. Transcription levels of each gene were quantified by real-time Ready Custom Panels and normalized to *HMBS* mRNA content. Results are reported as mean value ± SD of three independent experiments and are expressed as fold change with respect to LKB1mut cells, arbitrary set as 1 for each analyzed transcript (§ *P* < 0.05, §§*P* < 0.01, §§§*P* < 0.001). ND = expression level not determined.

Figure 2 | Liver kinase B1 (LKB1) controls NADPH oxidase 1 (*NOX1*) expression in human cell lines. *NOX1* expression levels in LKB1-deficient (LKB1mut) and LKB1-proficient (LKB1wt) non-small cell lung cancer (NSCLC) and cervical cancer cell lines. (A) Expression levels of *NOX1* mRNA in NSCLC and cervical cancer cell lines, with defined *LKB1* and *KRAS* mutational *statu*s. Results are reported as mean value ± SD of three independent experiments and are expressed as fold change compared to H2009 cells, arbitrary set at 1. (B) *In silico* analysis of *NOX1* levels in NSCLC cell lines. Expression level of *NOX1* as Robust Multi-array Average (RMA) values (source: Cancer Cell Line Encyclopedia). The dispersion of *NOX1* expression values in *n* = 13 *LKB1* mutated and in *n* = 12 *LKB1* wild-type cancer cell lines is represented as box-plots, indicating the first and the third quartiles and the median. The list of *LKB1*-mutated and non-mutated cancer cell lines is reported in the table (§ *P* < 0.05). nd = mutational status not determined.

due to either incomplete blockade of NOX1 or multifactorial mechanisms of control of oxidative stress by LKB1.

# NOX1 Promotes Angiogenesis and Growth of LKB1-Deficient A549 Tumors

High levels of ROS have been involved in induction of angiogenesis and tumor growth (13). To investigate whether oxidative stress could play a role in LKB1mut tumor formation, we co-injected LKB1mut A549 cells along with administration of the antioxidant NAC. LKB1mut A549 cells formed tumors (**Figure 4A**), and co-injection with NAC resulted in significant reduction of tumor growth and reduction of tumor vascularization, assessed 7 days after tumor cell injection (**Figure 4A**). Notably, chronic administration of apocynin significantly reduced tumor growth and MVD of LKB1mut tumors (**Figure 4A**), suggesting that blockade of NOX

Figure 3 | NADPH oxidase 1 (NOX1) increases sensitivity to exogenous oxidative stress *in vitro*. (A) A549 LKB1mut and LKB1wt cells were transiently transfected with si*RNA* targeting *NOX1* and cell death was measured upon H2O2 treatment (only statistically significant differences between silenced and control cells are indicated, # *P* < 0.05, ###*P* < 0.001). A representative dot plot diagram of Annexin-V/PI stained cells is shown (right panel). (B) Measurement of mitochondrial membrane potential in A549 LKB1mut and LKB1wt cells transiently silenced for *NOX1* expression (only statistically significant differences between silenced and control cells are indicated, # *P* < 0.05, ##*P* < 0.001, ###*P* < 0.001). (C) Induction of cell death in A549 LKB1mut or LKB1wt cells treated with the NOX pharmacological inhibitor apocynin 100 µM upon H2O2 treatment (only statistically significant differences between apocynin-treated cells and control cells are indicated, # *P* < 0.05).

activity impairs angiogenesis and tumor growth. These results were confirmed and strengthened by siRNA-mediated silencing of *NOX1* in A549 cells prior to their injection in mice (**Figure 4B**). Analysis of expression levels of *NOX1* in samples recovered 7 days after tumor cell injection showed, as expected, marginal silencing of the target gene, in contrast to the >80% reduction of *NOX1*

Figure 4 | NADPH oxidase 1 (NOX1) impacts on tumor growth and angiogenesis *in vivo*. (A) A549 LKB1mut cells were s.c. injected in SCID mice and treated daily with either *N*-acetyl cysteine (NAC) (150 mg/kg/day, gray triangles), or apocynin (16 mg/kg/day, white dots), or vehicle (black dots) (left panel). The image of two representative masses/group shows a difference in vascularization of treated compared to untreated tumors 7 days after tumor cells injection, as indicated by CD31 staining (magnification 200×) and quantification of the microvessel density (MVD) (right panel) (*n* = 3–6 fields from five samples/group; \**P* < 0.05, \*\**P* < 0.01, \*\*\* *P* < 0.001 A549 LKB1mut NAC-treated versus untreated tumors; ##*P* < 0.01, ###*P* < 0.001 A549 LKB1mut apocynin-treated versus untreated tumors). (B) Growth curves of s.c. tumors originated from A549 LKB1mut cells transiently transfected with si*RNA* control (black dots) or with a si*RNA* targeting *NOX1* (white triangles) (left panel). Seven days after tumor cells injection, LKB1mut si*RNA* tumors appear more vascularized compared to LKB1mut si*NOX1* tumors, as shown by the representative image. Immunofluorescence analysis of CD31+ cells and quantification of microvasculature in tumors at day 7 (right panel) (*n* = 3–6 fields from five samples/group; # *P* < 0.05*,* ###*P* < 0.001 A549 LKB1mut si*RNA* versus si*NOX1*). (C) Expression levels of *NOX1* in LKB1-deficient (LKB1mut) cells were analyzed prior to their injection in mice (bottom panel, at left) and in samples obtained from experiment shown in B (*n* = 3 and *n* = 6 for si*RNA* and si*NOX1* samples, respectively) and recovered 1 week post injection (bottom panel, at right).

levels at time of injection (**Figure 4C**). This suggests that NOX1 could play a key role in the initial phases of the angiogenic switch in the A549 LKB1mut tumor xenograft model.

#### DISCUSSION

This study disclosed for the first time an effect of LKB1 on *NOX1* expression in tumor cells. According to our results, LKB1 acts as a suppressor of *NOX1* expression at the RNA level. Among a set of 20 genes involved in redox homeostasis, *NOX1* transcript was consistently down-regulated following forced expression of LKB1 cDNA in LKB1-deficient cancer cells. Other genes were also perturbed, presumably as a direct consequence of LKB1 loss or, alternatively, as adaptive response to compensate for loss of protective mechanisms. In this respect, Kaufman et al., reported that LKB1 loss leads to activation of the nuclear respiratory factor 2 transcription factor, which regulates oxidative stress response genes (14).

NADPH oxidase 1 belongs to a family of NADPH oxidase (11) and cancer cells generally express more NOX isoforms. In the case of A549 cells utilized in our study, *NOX2* was expressed at a much lower baseline level compared to *NOX1* and *NOX4* was not expressed (data not shown). Therefore, although apocynin is not an isoform-selective NOX inhibitor (11), effects measured following treatment of A549 cells with this compound are likely to be accounted for by NOX1. A second consideration regards *NOX1* levels in tumor cells. In fact, real-time PCR experiments showed marked differences in the relative expression levels of *NOX1* among the NSCLC and cervical cancer cell lines analyzed (**Figure 2A**). It can, therefore, be speculated that the biological consequences of NOX1 inhibition might also depend on the amount of NOX1 in tumor cells. In line with this hypothesis, we indeed observed marked reduction of H2O2-induced cell death following apocynin treatment of A549 cells but not HeLa cells, which express substantially less NOX1 transcripts (**Figure 2** and data not shown). Therefore, although LKB1 acted as suppressor of *NOX1* expression in all cell lines analyzed, other endogenous factors are likely to modulate *NOX1* expression in tumor cells. According to previous studies, one of these factors could be KRAS, as its activating mutations were shown to correlate with *NOX1* overexpression in human colon cancer samples (15) and in cancer cells (16). In this regard, analysis of NSCLC cancer cell lines showed increased *NOX1* mRNA levels in cell lines bearing *LKB1* and *KRAS* mutations (**Figure 2**). This heterogeneity may also have precluded the inclusion of NOX1 among top ranking genes regulated by LKB1 in other genome-wide studies which previously investigated transcriptional profiles in LKB1-mutant cells and clinical samples (5, 14, 17, 18).

Although a detailed explanation of the mechanism connecting LKB1 to NOX1 is lacking and goes beyond the aim of our study, some hypothesis can be advanced. First, although LKB1 is not a transcription factor it is localized both in the cytoplasm and the nucleus of cells (17, 19). Therefore, suppression of NOX1 by LKB1 could be accounted for by its still poorly characterized nuclear functions. In this regard, it has been demonstrated that LKB1 forms a nuclear complex with the transcription regulatory factor LMO4, the transcription factor GATA-6 and Ldb1, and it has been suggested that this complex is involved in the regulation of GATA-mediated gene expression (20). Intriguingly, GATAbinding sites have been identified within the *NOX1* promoter (21). It is thus possible that LKB1 could be involved in the regulation of *NOX1* mRNA transcript through a GATA-6 mediated mechanism. Second, among indirect mechanisms, it has been shown that loss of LKB1 especially combined with KRAS mutations (as is the case in A549 cells) triggers the serine–glycine-one-carbon pathway, leading to increased S-adenosylmethionine generation, increased activity of DNA methyltransferases and elevation in DNA methylation (22). Hypothetically, this might silence some unknown repressor and thus unleash *NOX1* expression. Hence, alterations of NOX1 expression might occur as consequence of changes in epigenetic marks following vector-mediated reexpression of LKB1 in tumor cells. Further studies are required to validate these hypothesis.

The second innovative observation of this study regards the anti-angiogenic effects of drugs affecting ROS homeostasis in the context of LKB1-deficient tumors. Previous studies reported a link between angiogenesis and LKB1 expression in tumor models. Zhuang et al. described reduction of MVD accompanied by reduced expression of the angiogenic factors basic fibroblast growth factor and vascular endothelial growth factor (VEGF) in LKB1-deficient MDA-MB435 xenografts when LKB1 cDNA was added back, but further mechanistic insights were not provided (23). Along this line, it was known that NOX1 can promote the angiogenic switch in NIH-3T3 fibroblasts converting them into tumorigenic cells by a mechanism involving generation of ROS and increased expression of VEGF mRNA (24). Therefore, in view of our findings, it is conceivable that some of the effects on VEGF expression that follow interventions on LKB1 might depend on modulation of *NOX1* levels in tumor cells. Interestingly, Okon et al. reported that LKB1 functions as a RAB7 effector and suppresses angiogenesis by promoting the cellular trafficking of neuropilin-1 from the RAB7 vesicles to the lysosomes for degradation (25). Neuropilin-1 is an angiogenic receptor that can be expressed both by endothelial and tumor cells (26). When expressed by tumor cells, neuropilin-1 seems to increase VEGF-mediated angiogenesis. Therefore, it can be speculated that adding LKB1 back to tumor cells triggers complex changes in the tumor microenvironment involving NOX1, ROS, and neuropilin-1 that impact on VEGF/VEGFR-2 signaling and angiogenesis with negative rebounds on tumor growth that largely exceed the modest effects on tumor cell proliferation, which can be detected *in vitro*. As the contribution of each component of this envisioned model could vary in individual tumors, additional studies are required to validate this hypothesis both in preclinical models and in patients.

In conclusion, our study disclosed for the first time that LKB1 controls *NOX1* expression *in vitro*. Perturbing ROS homeostasis or inhibition of NOX1 reduced tumor growth and angiogenesis *in vivo* and could represent a new possible therapeutic approach in LKB1-deficient KRAS-mutated tumors.

#### REFERENCES


#### ETHICS STATEMENT

Procedures involving animals and their care conformed to institutional guidelines that comply with national and international laws and policies (EEC Council Directive 86/609, OJ L 358, 12 December, 1987) and were authorized by the ethical committee of the University of Padova and the Italian Ministry of Health (Authorization no. 143/2012-B).

#### AUTHOR CONTRIBUTIONS

SI conceived the study, designed experiments, analyzed data, and wrote the paper. EZ and FC designed and performed experiments, analyzed data, and wrote the paper. GN, MP, VA, and MS-B performed experiments and analyzed data. VC analyzed data and corrected the paper.

# FUNDING

This work was supported by grants from the AIRC (IG18803 to SI).


26. Miao HQ, Lee P, Lin H, Soker S, Klagsbrun M. Neuropilin-1 expression by tumor cells promotes tumor angiogenesis and progression. *FASEB J* (2000) 14:2532–9. doi:10.1096/fj.00-0250com

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

*Copyright © 2018 Zulato, Ciccarese, Nardo, Pinazza, Agnusdei, Silic-Benussi, Ciminale and Indraccolo. 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 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.*

# microrNa-494 favors ho-1 expression in Neuroblastoma cells exposed to oxidative stress in a Bach1-independent Way

*Sabrina Piras 1†, Anna L. Furfaro1†, Rocco Caggiano2 , Lorenzo Brondolo1 , Silvano Garibaldi <sup>3</sup> , Caterina Ivaldo1 , Umberto M. Marinari <sup>1</sup> , Maria A. Pronzato1 , Raffaella Faraonio2,4\* and Mariapaola Nitti <sup>1</sup> \**

#### *Edited by:*

*Giuseppe Filomeni, Danish Cancer Society, Denmark*

#### *Reviewed by:*

*Luisa Rossi, Università di Roma Tor Vergata, Italy Sara Baldelli, Università telematica San Raffaele, Italy Keith R. Laderoute, SRI International, United States*

#### *\*Correspondence:*

*Raffaella Faraonio raffaella.faraonio@unina.it; Mariapaola Nitti mariapaola.nitti@unige.it*

*† These authors have contributed equally to this work.*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

*Received: 27 February 2018 Accepted: 17 May 2018 Published: 13 June 2018*

#### *Citation:*

*Piras S, Furfaro AL, Caggiano R, Brondolo L, Garibaldi S, Ivaldo C, Marinari UM, Pronzato MA, Faraonio R and Nitti M (2018) microRNA-494 Favors HO-1 Expression in Neuroblastoma Cells Exposed to Oxidative Stress in a Bach1-Independent Way. Front. Oncol. 8:199. doi: 10.3389/fonc.2018.00199*

*1Department of Experimental Medicine, University of Genoa, Genoa, Italy, 2Department of Molecular Medicine and Medical Biotechnologies, University of Naples "Federico II", Naples, Italy, 3Department of Internal Medicine, Cardiology, Ospedale Policlinico San Martino, University of Genoa, Genoa, Italy, 4CEINGE-Biotecnologie Avanzate, Naples, Italy*

Heme oxygenase 1 (HO-1) is crucially involved in cell adaptation to oxidative stress and has been demonstrated to play an important role in cancer progression and resistance to therapies. We recently highlighted that undifferentiated neuroblastoma (NB) cells are prone to counteract oxidative stress through the induction of HO-1. Conversely, differentiated NB cells were more sensitive to oxidative stress since HO-1 was scarcely upregulated. In this work, we investigated the role played by miR-494, which has been proved to be involved in cancer biology and in the modulation of oxidative stress, in the upregulation of HO-1. We showed that NB differentiation downregulates miR-494 level. In addition, endogenous miR-494 inhibition in undifferentiated cells impairs HO-1 induction in response to exposure to 500 µM H2O2, reducing the number of viable cells. The analysis of Bach1 expression did not reveal any significant modifications in any experimental conditions tested, proving that the impairment of HO-1 induction observed in cells treated with miR-494 inhibitor and exposed to H2O2 is independent from Bach1. Our results underline the role played by miR-494 in favoring HO-1 induction and cell adaptation to oxidative stress and contribute to the discovery of new potential pharmacological targets to improve anticancer therapies.

#### Keywords: heme oxygenase 1, neuroblastoma, miR-494, oxidative stress, Bach1

# INTRODUCTION

Heme oxygenase 1 (HO-1) is a 32-kDa inducible enzyme belonging to the HO system, which catalyzes the degradation of the iron-containing molecule heme, leading to the generation of free iron (Fe2<sup>+</sup>), carbon monoxide (CO), and biliverdin. Biliverdin reductase converts biliverdin into bilirubin (1) and ferritin quenches free iron (2). Overall, ferritin, CO, and bilirubin exert strong antioxidant, anti-apoptotic, and anti-inflammatory effects (3). Different activators are involved in HO-1 induction and the nuclear factor erythroid 2-related factor 2 (Nrf2) is considered the most important (4, 5). Moreover, Keap1 by favoring Nrf2 proteasomal degradation, and Bach1 by preventing Nrf2 binding to the promoter region of HO-1, work as HO-1 repressors (6–8).

A sustained HO-1 expression in cancer correlates with a high degree of malignancy (e.g., aggressiveness, metastatic, and angiogenetic potential), although the pro-tumorigenic role of HO-1 seems to be tumor specific and tissue specific (9, 10). In the treatment of highly aggressive neuroblastoma (NB), the upregulation of HO-1 limits the efficacy of bortezomib (11, 12) suggesting HO-1 inhibition may represent a molecular target in the clinical strategies against NB (13, 14).

By modulating the expression of many different proteins, microRNAs (miRs) supervise and integrate numerous signaling pathways and their involvement has been postulated in various physiological and pathophysiological processes, from differentiation to senescence or oncogenesis (15–17). Strong evidence supports the notion that miRs can behave as oncogenes or tumor suppressor genes (18), and given the important role of oxidative stress response in tumorigenesis, understanding miR regulation in this condition is of major interest. Since a role played by miR-494 in the modulation of oxidative stress has been demonstrated in other contexts (19, 20), but no studies have been conducted in NB cells so far. In this work, we aimed at investigating the functional role of miR-494 in NB cell response to oxidative stress, focusing on its involvement in HO-1 induction.

# MATERIALS AND METHODS

#### Cell Culture and Differentiation

SH-SY5Y and SK-N-BE(2C) NB cells were cultured in RPMI 1640 medium (Euroclone, Italy) supplemented with 10% fetal bovine serum (Euroclone), 2 mM glutamine (Sigma-Aldrich, Italy), 1% amphotericin B (Sigma-Aldrich), and 1% penicillin/streptomycin (Sigma-Aldrich). Cells were differentiated by growth in the same medium supplemented with 10 µM all-trans retinoic acid (ATRA) (Sigma-Aldrich) for 4 days, up to 8 days. Differentiation was monitored by checking morphological changes such as neurite elongation and biochemical markers such as MAP2 and NeuroD1 expression (21, 22).

# RNA Extraction and microRNA Level Evaluation

Total RNA was extracted using TRIZOL reagent (Life Technologies, Carlsbad, CA, USA) according to the manufacturer's instructions. The cDNA templates for evaluation of mature miR levels were obtained from input RNAs (10 ng) using TaqMan™ Advanced miR cDNA Synthesis Kit (Thermo Fisher Scientific, USA, Cat. No. A28007) following the manufacturer's protocol. Real-time quantitative PCR for hsa-miR-494, hsa-miR-128, hsamiR-425-5p, and hsa-let7g-5p was performed in triplicate on diluted cDNA templates (1:10) by using the TaqMan® Advanced miR Assays (Thermo Fisher Scientific, Cat. No. A25576). hsamiR-425-5p and hsa-let7g-5p were used as endogenous reference miRs. Relative quantification of miR expression levels was performed according to the ΔΔCt method.

# Transfection of microRNA Inhibitors

SH-SY5Y cells were transiently transfected with 100 pmol of miR-494 inhibitor (miRCURY LNA miR inhibitor—hsa-miR-494-3p, QIAGEN, Hilder, Germany) and miR inhibitor Control (mirCURY LNA miR inhibitor Control—negative Control A, QIAGEN) by using Lipofectamine 2000 (Life Technologies) following the manufacturer's instruction. 96 h after transfection, cells were treated with 500 µM H2O2 for further 6 h to assess Bach1 levels, or for further 24 h to assess HO-1 levels and cell viability.

# Cell Viability Assay

Cell viability was evaluated by Trypan blue assay as previously described (12).

# Reactive Oxygen Species (ROS) Evaluation

Evaluation of ROS was performed by using 2′,7′-dichlorofluorescein diacetate (DCFH-DA; Sigma-Aldrich) assay. After treatments, cells stained with 5 µM DCFH-DA for 30 min at 37°C were analyzed by FACS (Attune™ Acoustic Focusing Flow Cytometer, Thermo Fisher Scientific). Values are expressed as arbitrary units of fluorescence.

#### Immunoblotting

Total protein lysates, prepared by using RIPA buffer (13), were subjected to electrophoresis on SDS-polyacrylamide gel (Mini protean precast TGX gel, Bio-Rad, Milan, Italy) (22). Immunodetection was performed using mouse anti-hnRNPQ (1:1,000, Santa Cruz), rabbit anti-PTEN (1:1,000, Cell Signaling Technology, MA, USA), rabbit anti-Bach1 (1:4,000, Bethyl Lab, Montgomery, TX, USA), and rabbit anti-HO-1 (1:2,000, Origene, Herford, Germany) and specific secondary antibodies (GE Healthcare). The membranes were re-probed with the loading control antibodies, rabbit anti-GAPDH (1:1,000, Santa Cruz) or mouse anti-tubulin (1:2,000, AbCam). The bands were detected by means of an enhanced chemiluminescence system (GE Healthcare) and developed films analyzed using a specific software (GelDoc, Bio-Rad).

#### Statistical Analyses

Statistical analysis of the differences among mean values ± SEM from three or more experiments was performed by using *t*-test to compare two groups or one-way ANOVA followed by Dunnett's post-test to compare more groups.

# RESULTS

#### ATRA-Induced NB Cell Differentiation Is Associated With Reduced Levels of miR-494

The analysis of miR-494 expression, performed after 4-day exposure to 10 µM ATRA, showed a significant reduction in both SH-SY5Y and SK-N-BE(2C) cell lines (**Figure 1A**). SH-SY5Y cells increase the expression of differentiation markers already after 4-day exposure to ATRA, as widely proved (21–23). However, since SK-N-BE(2C) need more time to complete ATRA-induced differentiation (24), miR evaluation was also performed after 6- and 8-day exposure to 10 µM ATRA on SK-N-BE(2C) cells. In these conditions, only a small further decrease of miR-494 expression has been observed (Figure S1a in Supplementary

reference miRs. Results are reported as relative to the values obtained in untreated control cells, which was set equal to 1. Statistical analysis: *n* = 3; \**p* < 0.05 vs undifferentiated. (B) Expression levels of mature miR-128 in undifferentiated or ATRA-differentiated SH-SY5Y and SK-N-BE(2C) NB cells. hsa-miR-425-5p and hsa-let7g-5p were used as endogenous reference miRs. Results are reported as relative to the values obtained in untreated control cells, which was set equal to 1. Statistical analysis: *n* = 3. No significant differences. (C) WB analysis of PTEN. GAPDH expression has been used as loading control. 40 µg of protein has been loaded. The bands show the most representative experiment. Statistical analysis: *n* = 3; \**p* < 0.05 vs NegC. (D) WB analysis of hnRNPQ. GAPDH expression has been used as loading control. 40 µg of protein has been loaded. The bands show the most representative experiment. Statistical analysis: *n* = 2; \**p* < 0.05 vs NegC. (E) Percentage of viable cells (Trypan blue analysis) after miR-494 inhibition and 24 h exposure to 500 µM H2O2. Statistical analysis: *n* = 4; \**p* < 0.05 vs NegC and miRNA 494 inhibitor.

Material). The expression of miR-128 has been also analyzed due to its involvement in stress response and differentiation (25, 26), but no changes were observed in both cell lines after differentiation (**Figure 1B**). The following experiments have been carried out on SH-SY5Y NB cells which strongly downregulated miR-494 (−10-folds vs undifferentiated) in the shortest experimental time (4 days).

#### miR-494 Inhibition Modifies Cell Responses to Oxidative Stress

To evaluate whether the reduction of endogenous miR-494 could modify NB cell sensitivity to oxidative stress, cells were transfected with a specific miR-494 inhibitor and then exposed to 500 µM H2O2. The effectiveness of miR-494 inhibition was checked by evaluating the protein levels of two miR-494 targets, namely, PTEN and hnRNPQ that resulted upregulated of about 50% (**Figures 1C,D**). The analysis of viable cells revealed no changes induced by miR-494 inhibition itself, in comparison to cells transfected with a NegC. Conversely, miR-494 inhibition significantly decreased the percentage of viable cells after the exposure to 500 µM H2O2 (**Figure 1E**). The analysis of markers of apoptosis such as BAX and PARP did not show any changes (Figures S1b and S1c in Supplementary Material) and this led us to rule out the occurrence of apoptosis. These results indicate that the expression of miR-494 in undifferentiated NB cells favors cell adaptation/response to oxidative stress.

#### miR-494 Inhibition Impairs HO-1 Induction in Response to Oxidative Stress

To investigate whether a reduced expression of miR-494 influences oxidative stress response, NB cells transfected with miR-494 inhibitor or NegC were treated with H2O2 and ROS levels and HO-1 expression were evaluated. ROS levels were increased only in cells treated with miR-494 inhibitor and exposed to H2O2 (**Figure 2A**). In this experimental condition, no significant induction of HO-1 has been observed (**Figure 2B**). Conversely, in NB cells transfected with NegC, the exposure to H2O2 was able to significantly increase the expression of HO-1, and the level of ROS was not significantly modified.

The ubiquitination pattern was also analyzed but no changes were detected in any experimental conditions (Figure S1d in Supplementary Material).

*In silico* analyses predicted Bach1 as a target of miR-494 with two putative sites within its 3′UTR (**Figure 2C**). Thus, we checked the protein levels of Bach1. WB analysis showed that miR-494 inhibition did not modify Bach1 expression after H2O2 exposure (**Figure 2D**). Different Bach1 post-translational modifications have also been analyzed; ubiquitination and sumoylation were not detected and Bach1 acetylation was not modified in any experimental conditions (Figure S1e in Supplementary Material). These results show no involvement of Bach1 in the miR-494 dependent HO-1 regulation.

Furthermore, Keap1 levels have been also checked but no changes were detected in any experimental conditions (Figure S1f in Supplementary Material), proving no involvement of Nrf2 in this context.

#### DISCUSSION

In this work, we pointed out the involvement of miR-494 in the upregulation of HO-1 in NB cell response to oxidative stress. We took into consideration two miRs, such as miR-128 and miR-494. Indeed, miR-128 has been demonstrated to be involved in NB differentiation (26) and response to oxidative stress (25) but we did not observe any modification of miR-128 levels in the different experimental conditions we tested. Thus, we evaluated miR-494 which, from bioinformatics analyses, was predicted to have two putative binding sites on Bach1 3′UTR, the main repressor of HO-1 transcription.

In numerous contexts, miR-494 functions as tumor suppressor gene and has been linked to the induction of senescent phenotype in normal cells (19, 27) but in other contexts it correlates with tumor aggressiveness and progression (28). To the best of our knowledge, there has been no evidence of miR-494 expression in NB so far. We demonstrated that miR-494 is expressed in two undifferentiated NB cell lines and undergoes a significant downregulation after ATRA-induced differentiation. The reduction is dramatic for SH-SY5Y cells that easily differentiate in response to ATRA and minor but always significant in SK-N-BE(2C) which shown medium sensitivity to ATRA (24). There is only a paper in literature showing that the expression of miR-494 is upregulated by ATRA in the acute myeloid leukemia cell line HL-60 (29), and this lets us hypothesize that there may be a cell-type-specific regulation for miR-494. Thus, we further analyzed SH-SY5Y cells which, from our previous works, have been proved to increase their sensitivity to oxidative stress after differentiation (22, 30), investigating a possible correlation with the miR-494 downregulation. We observed that miR-494 inhibition in undifferentiated cells significantly reduced the number of viable cells after exposure to H2O2. The role of miR-494 in cell survival is controversial, depending on the cellular context where miR operates and on the accessibility of its targets. As also shown in our work, miR-494 inhibition is able to increase PTEN expression and, potentially, to antagonize the AKT survival pathway, as proved in other contexts (31, 32). However, the modulation of miR-494-PTEN signaling under stress condition has not yet been investigated.

Next, we provided evidence that endogenous miR-494 inhibition impairs HO-1 upregulation in response to oxidative stress, similar to what we have already shown in differentiated cells exposed to H2O2 (22). In addition, we showed that the lack of HO-1 upregulation correlates with higher ROS levels, highlighting the importance of HO-1 induction in quenching ROS. However, the analysis of Bach1 expression revealed that there are no significant modifications in the level of Bach1 in response to H2O2 in NB cells treated with miR-494 inhibitor compared with cells transfected with NegC. Moreover, there are no changes in Bach1 ubiquitination, sumoylation, and acetylation in any experimental conditions examined. Thus, miR-494 could contribute through a Bach1-independent mechanism to modulate HO-1 expression under stress response. It has been already demonstrated that HO-1 transcription can be controlled *via* Bach1 turnover in the absence or presence of oxidative stress and can also be

been loaded. The bands show the most representative experiment. Statistical analysis: *n* = 4; \**p* < 0.05 vs NegC.

insensitive to Bach1-mediated repression (33). Moreover, AKT-dependent HO-1 induction has been already proved (34), and miR-494-PTEN might crucially modulate it. To validate this hypothesis, further analyses are needed.

# AUTHOR CONTRIBUTIONS

SP, ALF, SG, RF, and MN conceived and designed the experiments. SP, ALF, RC, LB, SG, and CI conducted the experiments. SP, ALF, SG, RF, and MN analyzed the results. UMM, MAP, RF, and MN contributed reagents/materials/analysis. SP, ALF, UMM, RF, and MN wrote the paper. All the authors reviewed the manuscript.

# FUNDING

This study was supported by grants from University of Genoa (to MN).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at https://www.frontiersin.org/articles/10.3389/fonc.2018.00199/ full#supplementary-material.

FIGURE S1 | (A) Expression levels of mature miR-494 and miR-128 in undifferentiated and 6- or 8-day-differentiated SK-N-BE(2C) NB cells. hsa-miR-425-5p and hsa-let7g-5p have been used as endogenous reference miRs. Results are reported as relative to the values obtained in untreated undifferentiated cells which was set equal to 1. Statistical analysis: *n* = 3; *\*p* < 0.05 vs undifferentiated. (B) WB analysis of BAX in SH-SY5Y cells treated with miR-494 inhibitor and exposed to 500mM H2O2, as indicated. GAPDH expression has been used as loading control. 10 mg of proteins was loaded. The blots show one representative experiment. Statistical analysis: *n* = 3; no significant differences. (C) WB analysis of PARP in

#### REFERENCES


SH-SY5Y cells treated with miR-494 inhibitor and exposed to 500mM H2O2, as indicated. GAPDH expression has been used as loading control. 50 mg of proteins was loaded. The blots show one representative experiment. Statistical analysis: *n* = 2; no significant differences. (D) WB analysis of ubiquitination in SH-SY5Y cells treated with miR-494 inhibitor and exposed to 500mM H2O2, as indicated. 20 mg of proteins was loaded. The blot shows one representative experiment. (E) Analysis of Bach1 post-translational modifications in SH-SY5Y cells treated with miR-494 inhibitor and exposed to 500mM H2O2 for 6 h. 300 mg of protein lysate was immunoprecipitated using anti Bach1 and loaded in electrophoresis (ip). An aliquot of supernatant collected after the first step of immunoprecipitation was loaded in electrophoresis (sn). WB detection was performed as indicated. The blots show the most representative experiment. (F) WB analysis of Keap1 in SH-SY5Y cells treated with miR-494 inhibitor and exposed to 500mM H2O2 as indicated. GAPDH expression has been used as loading control. 40 mg of proteins was loaded. The blots show one representative experiment. Statistical analysis: *n* = 3; no significant differences.


pathway in heart. *PLoS One* (2016) 11:e0148545. doi:10.1371/journal.pone. 0148545

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

*Copyright © 2018 Piras, Furfaro, Caggiano, Brondolo, Garibaldi, Ivaldo, Marinari, Pronzato, Faraonio and Nitti. 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 are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Novel Mitochondria-Targeted Furocoumarin Derivatives as Possible Anti-Cancer Agents

*Andrea Mattarei1 , Matteo Romio1 , Antonella Managò2 , Mario Zoratti3,4, Cristina Paradisi1 , Ildikò Szabò2 , Luigi Leanza2 \* and Lucia Biasutto3,4\**

*1Department of Chemical Sciences, University of Padova, Padova, Italy, 2Department of Biology, University of Padova, Padova, Italy, 3CNR Neuroscience Institute, Padova, Italy, 4Department of Biomedical Sciences, University of Padova, Padova, Italy*

Targeting small molecules to appropriate subcellular compartments is a way to increase their selectivity and effectiveness while minimizing side effects. This can be accomplished either by stably incorporating specific "homing" properties into the structure of the active principle, or by attaching to it a targeting moiety *via* a labile linker, i.e., by producing a "targeting pro-drug." Mitochondria are a recognized therapeutic target in oncology, and blocking the population of the potassium channel Kv1.3 residing in the inner mitochondrial membrane (mtKv1.3) has been shown to cause apoptosis of cancerous cells expressing it. These concepts have led us to devise novel, mitochondria-targeted, membrane-permeant drug candidates containing the furocoumarin (psoralenic) ring system and the triphenylphosphonium (TPP) lipophilic cation. The strategy has proven effective in various cancer models, including pancreatic ductal adenocarcinoma, melanoma, and glioblastoma, stimulating us to devise further novel molecules to extend and diversify the range of available drugs of this type. New compounds were synthesized and tested *in vitro*; one of them—a prodrug in which the coumarinic moiety and the TPP group are linked by a bridge comprising a labile carbonate bond system—proved quite effective in *in vitro* cytotoxicity assays. Selective death induction is attributed to inhibition of mtKv1.3. This results in oxidative stress, which is fatal for the already-stressed malignant cells. This compound may thus be a candidate drug for the mtKv1.3-targeting therapeutic approach.

Keywords: Kv1.3, mitochondria-targeting, cancer, triphenylphosphonium, psoralens, pancreatic duct adenocarcinoma

#### INTRODUCTION

In therapeutic oncology, the ultimate goal is to cause the death of all cancerous and cancer stem cells while inflicting negligible damage to healthy cells and organs. Several strategies have been adopted in an effort to achieve this difficult result. One is to selectively hit only the unwanted cells with an effective death-inducing treatment. Examples of this approach include focused radiation therapy, the many attempts at delivering drugs selectively to tumoral cells, and immunotherapy. Another is aiming a drug at a molecular target which is expressed specifically by cancer cells and/or whose function is more cogently needed by cancer cells than by healthy ones, including of course malfunctioning oncogenes. An example may be the development of small molecule inhibitors of deregulated oncogenic kinases. A third one may be to exploit an intrinsic characteristic of cancer cells, such as increased aerobic glycolysis (the Warburg effect) or their rapid growth and the associated redox stress

#### *Edited by:*

*Giuseppe Filomeni, Danish Cancer Society, Denmark*

#### *Reviewed by:*

*Jiri Neuzil, Griffith University, Australia Vincenzo Ciminale, Università degli Studi di Padova, Italy*

#### *\*Correspondence:*

*Luigi Leanza luigi.leanza@unipd.it; Lucia Biasutto lucia.biasutto@cnr.it*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

*Received: 01 February 2018 Accepted: 04 April 2018 Published: 23 April 2018*

#### *Citation:*

*Mattarei A, Romio M, Managò A, Zoratti M, Paradisi C, Szabò I, Leanza L and Biasutto L (2018) Novel Mitochondria-Targeted Furocoumarin Derivatives as Possible Anti-Cancer Agents. Front. Oncol. 8:122. doi: 10.3389/fonc.2018.00122*

**163**

they experience (1). This latter feature is of particular interest for this paper. An elevated production of ROS is a characteristic of rapidly proliferating cells (2). In cancerous cells, it is induced by various mechanisms (1). For example, K-ras hyperactivity has been reported to lead to suppression of respiratory chain (RC) complex-I, ROS generation, mitochondrial dysfunction, and switch to glycolytic metabolism (3, 4). Tumor suppressor p53 on the other hand acts *via* Bcl-2 family proteins to reduce mitochondrial ROS, and its loss or malfunction may thus lead to an increase in their production (5, 6). In fact, anti-apoptotic proteins of the Bcl-2 family, such as Bcl-XL and Bcl-w, reportedly increase mt ROS production. They do this by binding and neutralizing Bax, which reduces ROS by interacting with complex I (7, 8).

This stressed state can be exploited to induce cell death (1, 9–11). Excessive ROS can cause cell death by processes such as apoptosis (12) [mediated, for example, by redox-sensitive apoptosis signal-regulating kinase family members (13)], necrosis (14), and ferroptosis (15). An oxidative stress exceeding the "death threshold" can be achieved either by weakening the cellular antioxidant defenses which keep it within "safe" limits (16)—for example, by inhibiting a member of the peroxiredoxin system (17, 18)—or by increasing it. In turn, this latter option can be achieved either by using drugs which are themselves redox-active [e.g., Q-7BTPI (19)], or by stimulating the cells' ROS-producing apparatuses, such as the mitochondrial RC [e.g., Ref. (20)].

Given their key role in cancer metabolism, progression, and survival (9, 21–26) and in apoptosis, mitochondria are a focus of anti-cancer chemotherapy (27–29). Of relevance here, mitochondrial ion channels are potential targets of strategies aiming to stress cancer cells to death. They influence mitochondrial membrane potential ΔΨ, ROS production, volume, and ion homeostasis (30). Pharmacological manipulation of mitochondrial ion channels can lead to cell death bypassing the upstream players of intrinsic apoptosis (p53 status, Bax/Bak/Bcl-2 expression and alterations of cytosolic signaling pathways) (31).

In particular, our group has uncovered (32, 33) a crucial role of mitochondrial potassium-selective channel mtKv1.3 blockage by pro-apoptotic Bax in the apoptotic death of cells expressing mtKv1.3, which include many cancer cell lines (34). The other finding this line of research descends from is the observation (35) that 5-(4-phenylbutoxy)psoralen (Psora-4) (**Figure 1**), a membrane-permeant molecule, blocked Kv1.3 with an EC50 of 3 nM. A derivative, PAP-1, was less effective but more selective for Kv1.3 vs Kv1.5, which is also often expressed in the mitochondria of cancer cells. We used these compounds to show that pharmacological inhibition of mtKv1.3 could cause the same outcome as inhibition by Bax, i.e., death by apoptosis (36, 37). This outcome is currently understood to result from the following chain of events: stopping the depolarizing K<sup>+</sup> influx causes inner mitochondrial membrane (IMM) hyperpolarization, with ensuing increased ROS level, activation of the mitochondrial permeability transition pore, mitochondrial swelling, loss of transmembrane potential, loss of cytochrome c, and further ROS release (36).

5-(4-phenylbutoxy)psoralen and PAP-1, however, had only a modest effect on cancerous cells when used at pharmacologically meaningful concentrations. To improve their effectiveness we sought to target the drugs to the IMM and mitochondrial

matrix. The most effective and popular strategy for mitochondrial targeting relies on conjugating the drug to a lipophilic, membrane-permeant cation, most often triphenylphosphonium (TPP) (38–41). Various drugs based on this design and producing cytotoxic oxidative stress in cancerous cells have already been produced: mitochondria-targeted vitamin E succinate (MitoVES) (42–44), a construct interacting with RC complex-II; MitoMets (45, 46), metformin derivatives inhibiting RC complex I and inducing ROS production; MitoTam (47), based on tamoxifen and likewise acting *via* RC complex-I; mitochondriotropic derivatives of the polyphenols resveratrol (20, 48) and quercetin (19, 49), also causing deadly redox stress in cultured cells *via* the RC or concentration-enhanced autoxidation, respectively. We thus synthesized two TPP-comprising PAP-1 derivatives, PAPTP and PCARBTP, both of which turned out to be promising chemotherapeutic agents, selectively eliminating cancerous cells *in vitro* and in *in vivo* oncological models, including orthotopic melanoma and pancreatic ductal adenocarcinoma (PDAC) (50). These compounds caused the death of pathological cells independently of the expression levels of key pro- or anti-apoptosis proteins, such as Bax, Bak, Bcl-2, or p53. Importantly, they had no significant impact on healthy tissues and cells, including the immune system of mice and humans (50). The structure of PCARBTP combines two concepts: mitochondrial targeting, conferred by TPP, and prodrug function, provided by the carbamate link connecting the two parts of the molecule, PAP-1 and TPP. Since the carbamate group is hydrolyzed over several minutes in a physiological environment, this device allows the delivery of the essentially unmodified active agent to mitochondria.

Given the promising results, we have extended the search for anti-tumoral agents combining the Kv1.3-inhibiting furocoumarin structure and the mitochondriotropic TPP group.

### MATERIALS AND METHODS

#### Chemistry

Details of the synthetic procedures used in this study are provided in the Supplementary Material.

#### Cell Lines

B16F10 cells (ATCC) were grown in Minimum Essential Media (MEM, Thermo Fisher Scientific) supplemented with 10 mM HEPES buffer (pH 7.4), 10% (v/v) fetal bovine serum (FBS), 100 U/mL penicillin G, 0.1 mg/mL streptomycin, and 1% nonessential amino acids (100× solution; Thermo Fisher Scientific). Lymphocytes (Jurkat, CTLL-2, and K562) were grown in RPMI 1640 (Thermo Fisher Scientific), supplemented as MEM. Medium for CTLL-2 was further supplemented with four units/mL/day of mouse interleukin-2 (IL-2). A panel of pancreatic cancer cell lines (51) was used: BxPC3, AsPC1, Capan-1, and PANC-1. BxPC3 derived from the body of the pancreas of a patient with adenocarcinoma. These cells are not prone to give metastasis and they are poorly differentiated. The other three lines were originally obtained from metastases (AsPC-1, Capan-1) or have considerable metastatic potential (PANC-1). All were provided by ATCC. AsPC1 and BxPC3 were cultured in RPMI-1640 supplemented with 10% FBS "GOLD" (PAA Laboratories/GE Healthcare Life Sciences), 1 mM GlutaMAX, and 1 mM sodium pyruvate (Thermo Fisher Scientific). PANC-1 were cultured in DMEM (4.5 g/L d-glucose) supplemented with 10% FBS "GOLD", 1 mM GlutaMAX and 1 mM sodium pyruvate. Capan-1 cells were grown in IMEM supplemented with 20% FBS "GOLD," 1 mM GlutaMAX, and 1 mM sodium pyruvate. The HPV16-E6E7—immortalized human pancreatic duct epithelial cells (HPDE), kindly provided by Dr. Ming-Sound Tsao (Ontario Cancer Institute, Toronto, ON, Canada) (52) were used as a model for benign pancreatic ductal epithelium. The complete HPDE growth medium was a mixture of 50% RPMI 1640, supplemented with 10% FCS and 1 mM GlutaMAX, and 50% keratinocyte medium SFM (Thermo Fisher Scientific) supplemented with 0.025% bovine pituitary extract, 2.5 mg/L epidermal growth factor (Thermo Fisher Scientific).

#### Cell Viability and Cell Death Assays

For cell growth/viability MTT assays we used a protocol previously described (36, 50). Briefly, cells were seeded (5–10 × 103 cells/ well) in standard 96-well plates and allowed to grow in medium (200 µL) for 24 h to ensure attachment. The growth medium was then replaced in the dark with a medium that contained the desired compound (from a stock solution in DMSO) at the final concentration. The final concentration of DMSO was 0.1% or lower in all cases (including controls). To inhibit multiple drug resistance (MDR) "pumps," where indicated we used non-toxic concentrations of cyclosporine H (CSH) (1 µM; Sequoia) and Probenecid (100 µM; Sigma Aldrich) in the case of CTLL-2 cells, and of CSH only (4 µM) for the other cell lines. After incubation for 24 h, CellTiter 96 AQUEOUS One solution (Promega, Italy) was added to each well as indicated by the supplier. Absorbance was measured at 490 nm to detect formazan formation using a Packard Spectra Count 96-well plate reader.

For cell death assays of non-adherent cells, cells were incubated with the test substances for 24 h, washed in HBSS, and resuspended in DMEM without serum and Phenol Red and incubated for 30 min at 37°C in the dark with Annexin-V FLUOS (Roche) (1 µL/200 μL sample). DMSO concentration was <0.1% in all cases. Flow cytometry analysis was carried out after the labeling period with a Becton Dickinson FACS Canto II flow cytometer and data were processed by quadrant statistics using BD VISTA software.

## Downregulation of Kv1.3 Expression by siRNA

The sequences for the siRNA targeting human Kv1.3 were coupled to Alexa Fluo 555 (Hs\_KCNA3\_1 Flexi tube siRNA for Kv1.3 and All-star negative control siRNA as scramble/control; Qiagen). Jurkat cells were transfected by electroporation, as previously reported (36). After 48 h from transfection, cells were treated for 24 h with the various compounds as indicated. Cell death was then evaluated by the binding of fluorescein isothiocyanate (FITC)-labeled Annexin-V and FACS analysis.

# Mitochondrial Morphology, Membrane Potential, and ROS Production

Mitochondrial morphology was studied in melanoma B16F10 cells. 1 × 105 cells were seeded in a 6-well plate with 2 mL of complete medium. After 24 h medium was replaced with 1 mL HBSS supplemented with 500 nM Mitotracker green (Thermo Scientific). Cells were incubated at 37°C in the dark for 20 min and then the mitochondrial network was observed by confocal microscopy using a Leica DMI6000 fluorescence microscope with confocal settings (Leica Microsystem, Wetzalar, Germany).

Mitochondrial membrane potential and ROS production were measured in leukemic Jurkat T cells. 5 × 105 cells were resuspended in 300 µL of HBSS supplemented either with 20 nM TMRM or 1 µM MitoSOX. Cells were incubated for 20 min at 37°C in the dark. Then, cells were diluted by the addition of further 1.2 mL of HBSS and analyzed by FACS (FACSanto II, Beckton Dickson).

# Western Blot

Kv1.3 protein expression was assessed after transfection with control ("scramble") and anti-Kv1.3 siRNA. Jurkat cells from parallel siRNA experiments were lysed overnight refrigeration at −80°C in lysis buffer (25 mM Tris pH 7.8 + 2.5 mM EDTA + 10% glycerol + 1% NP-40 + 2 mM DTT). After thawing, debris was centrifuged off at 20,000 × *g* for 10 min at 4°C. Supernatants were collected and protein concentration was determined using the BCA method in a 96-well plate (200 μL total volume for each well) incubating at 37°C in the dark for 30 min. Absorbance at 540 nm was measured by a Packard Spectra Count 96-well plate reader. Proteins were separated by SDS-PAGE in a 10% polyacrylamide gel. After separation by electrophoresis, gels were blotted overnight at 4°C onto polyvinylidene fluoride membranes and then membranes were blocked with a 10% solution of defatted milk and were incubated with the following primary antibodies overnight at 4°C: anti-Kv1.3 (1:200, rabbit polyclonal, Alomone Labs APC-101); anti-GAPDH (1:1,000, mouse monoclonal, Millipore MAB374). After washing, the membranes were developed using corresponding anti-mouse or anti-rabbit secondary antibodies (Calbiochem). The antibody signal was detected with enhanced chemiluminescence substrate (SuperSignal West Pico Chemiluminescent Substrate, Thermo Scientific).

#### Statistics

Statistical significance of the effects was assessed by paired *t*-test or two-way ANOVA analysis.

# RESULTS AND DISCUSSION

One of the concepts we tested was to minimize the changes to the structure of the "parent" drug, Psora-4, while still turning it into

lymphocyte CTLL-2 cells were transfected either with the empty vector (CTLL-2 pJK) or with the expression vector for Kv1.3 (CTLL-2 Kv1.3). Cell viability was assessed by MTT assay after 24 h of incubation with Psora-4, P5TP (A), PAP-1, or PCTP (B) at different concentrations either without or with the addition of 1 µM cyclosporine H and 100 µM probenecid as multidrug resistance pumps inhibitors. DMSO 10% was used as positive control, since it is toxic at this concentration. Results are reported as mean percentage of viable cells normalized with respect to untreated cells (*n* = 3; \*\*\**p* < 0.001).

a mitochondria-targeted drug. Thus, we simply substituted the distal phenyl ring with the TPP group (P5TP, **Figure 1**).

A second approach was that of attaching a mitochondriatargeting group to PAP-1. This strategy had proved successful when attaching the TPP moiety *via* a stable bond system or *via* a labile linker comprising a carbamate group (50). We thus tested another labile "joint," the carbonate group, producing PCTP (**Figure 1**).

The rationale for the preparation of these new derivatives was based on previous studies which underlined the importance of not altering the planar furocoumarin system. The modification at position five of the psoralen scaffold did not affect the ability of

Figure 3 | PCTP is specific in inducing cell death by Kv1.3 inhibition. (A) Jurkat T lymphocytes and leukemic K562 cells were treated for 24 h with PAP-1, PCTP, or PCARBTP at the indicated concentrations with or without the addition of 4 µM cyclosporine H as multidrug resistance pumps inhibitor. Cell death by apoptosis was then determined by incubation with fluorescein isothiocyanate-labeled annexin-V for 20 min at 37°C in the dark. Annexin-V positive cells were measured by FACS analysis (*n* = 3; \*\*\**p* < 0.001 vs control). (B) Jurkat cells were either transfected with a control siRNA (Scramble) or siRNA against Kv1.3 (siRNA) and after 48 h from the transfection they were treated as in (A) with PAP-1 and PCTP (*n* = 3; \*\*\**p* < 0.001 vs control). Insert: Kv1.3 downregulation was assessed by Western blot after siRNA transfection. A representative image is shown of three independent observations. GAPDH was used as loading control.

the molecule to block the potassium cation inside the cavity of the channel as proposed by Zimin and coworkers (53).

For both derivatives the synthesis started with the natural compound bergapten (5-methoxypsoralen) and utilized the key intermediate PSBI (Figure S1 in Supplementary Material). The conversion of bergapten to PSBI was achieved by demethylation to bergaptol (2) promoted by BBr3, followed by alkylation with 1-bromo-4-chlorobutane to obtain the chloro-intermediate 3 and substitution of chloride with iodide *via* the Finkelstein reaction. Details are provided in Supplementary Material.

PCTP underwent a slow hydrolysis in DMEM, at pH 7.4 and 37°C, yielding PAP-OH and 4-TPP-butan-1-ol iodide with a *t*1/2 of about 17 h (data not shown).

The new derivatives were first screened for their cytotoxic activity on murine CTLL-2 lymphocytes. These cells do not express Kv1.3, and were either transfected with an empty vector to provide a control (CTLL-2/pJK) or stably transfected with an expression vector for Kv1.3 (CTLL-2/Kv1.3) (54) (**Figure 2**). Cell survival was assessed with the MTT assay. P5TP did not represent an improvement over Psora-4 or PAP-1 (**Figure 2A**). On the other hand, PCTP proved remarkably effective and exhibited selectivity toward cells expressing Kv1.3: while viability of CTLL-2/Kv1.3 cells decreased in a dose-dependent manner, that of CTLL-2/pJK cells was only slightly affected (**Figure 2B**).

Since MDR pumps may play a crucial role in extruding drugs from cells, MTT assays were also performed in the presence of MDR inhibitors (cyclosporin H and probenecid). In this specific case, MDR inhibition sensitized CTLL-2/Kv1.3 cells to Psora-4 and PAP-1, as previously reported (36), while there were no differences in the activity observed with PCTP. This observation suggests that the positively charged compound may escape MDR action due to a rapid "electrophoretic" transport through the plasma membrane, due to the negative-inside electrical potential difference. This has been already proposed for mitoVES. Contrary to VES, mitoVES was not a substrate for the ABCA1 pump in non-small cell lung carcinoma H1299 cells (55).

We then tested PCTP on two human leukemic cell lines: Jurkat leukemia T cells and K562 chronic myelogenous leukemia cells (**Figure 3A**). Cell death was determined by Annexin-V-FITC staining and FACS analysis. As expected, PCTP induced apoptosis only in Kv1.3-expressing Jurkat cells (54), while it was quite ineffective in killing K562 cells, which lack Kv1.3 (34, 56).

To further demonstrate Kv1.3 involvement in apoptosis induction by PCTP, Jurkat cells were transiently transfected with siRNA targeting Kv1.3 to reduce its expression (36). These cells have the peculiarity that they express only Kv1.3 among the potassium channels of the Kv family (54). Experiments confirmed that Kv1.3 expression is crucial for cell death induction by PCTP, since Kv1.3 silencing protected the cells from death (**Figure 3B**).

We proceeded testing PCTP also with other Kv1.3-expressing cancer cell lines. We took advantage of a mouse B16F10 melanoma cell line (**Figure 4A**), which also expresses Kv1.3 in mitochondria, as we have shown before (36). In this case the presence of MDR inhibitors was crucial, as already observed with PAP-1. Nevertheless, PCTP is more powerful than the precursor in triggering cell death.

non tumoral human pancreatic duct epithelial line (HPDE, used as a negative control) were treated for 24 h with PAP-1 (B), PCTP (C), or PCARBTP (D), both in absence or presence of 4 µM CSH. Mean percentage of viable cells normalized with respect to untreated cells (*n* = 3; \*\*\**p* < 0.001 vs control).

Figure 5 | PCTP induces mitochondrial swelling, decrease in membrane potential, and increase in ROS production in Kv1.3-expressing cells. (A) Mitochondrial morphology was observed in B16F10 cells after staining the mitochondrial network by incubation for 20 min at 37°C with 500 nM Mitotracker green. The effects on mitochondria have been observed by confocal microscopy after 30 min of incubation either with or without 10 µM PCTP. The images are representative of three independent experiments (*n* = 3; Bars = 15 µm). (B) Mitochondrial membrane potential was measured by FACS analysis of TMRM fluorescence in leukemic Jurkat T cells. Values are reported as percentage of the initial fluorescence (*n* = 3; \*\*\**p* < 0.001 with respect to untreated, all other conditions). (C) Mitochondrial ROS production was measured by FACS analysis of the increase in the fluorescence of MitoSOX in leukemic Jurkat T cells. Values are reported as percentage of the initial fluorescence (*n* = 3; \*\*\**p* < 0.001 with respect to untreated, all other conditions).

The derivative was finally tested on various PDAC cell lines, which have been shown to express Kv1.3 (51). All have been previously characterized, and were found to be mutated in p53, to express variable but robust levels of Bcl2-family anti-apoptotic proteins and to be largely resistant to standard chemotherapeutics (57–59). Most of them (with exception of Bx PC-3 cells) are also mutated in K-ras (57). These cell lines provide an *in vitro* model of one of the most feared and untreatable human cancers, for which a viable pharmacological approach is much needed.

Interestingly, also in this case PCTP proved remarkably effective in inducing cell death (**Figure 4C**) while its precursor, PAP-1, was essentially inactive (**Figure 4B**). Cytotoxicity varied somewhat from cell line to cell line (**Figure 4C**). PCTP was confirmed to induce apoptosis (not shown). These results again show that mitochondriotropic mitoKv1.3 inhibitors can overcome chemoresistance, exerting their cytotoxic effects despite alterations of the cellular anti-apoptotic apparatus.

We investigated the impact of PCTP on the mitochondria of intact cells. To monitor morphological changes we used B16F10 cells labeled with the permanent mitochondrial marker MitoTracker Green (**Figure 5A**). The exemplary images in **Figure 5A** show that the mitochondrial network underwent fragmentation. Mitochondrial fission has been firmly associated with the process of apoptosis (60–62). Mitochondrial depolarization (TMRM staining), and ROS generation (MitoSOX™ Red staining) were observed using Jurkat cells in FACS experiments (**Figures 5B,C**, respectively).

These observations are fully coherent with the mechanistic model deduced from the data obtained studying apoptosis (32) and using PAPTP to induce it (50): the initial event is channel inhibition, with consequent production of ROS. In turn, ROS promote the onset of the permeability transition, resulting in mitochondrial depolarization and further ROS release. The effects of PCTP on cancerous cells *in vitro* are comparable to those of PAPT and PCARBTP [**Figures 4** and **5**; (50)]. PCTP might conceivably even outperform these latter compounds *in vivo*, depending on factors such as pharmacokinetics and the rate of hydrolysis of the carbonate bond system. The mitochondrial effects suggest that the compound might have significant undesirable effects on healthy cells. It is, however, of relevance that non-tumoral, fast-growing, Kv1.3-expressing HPDE (51, 63) were not affected by PCTP (**Figure 4C**), as was the case also for PCARBTP (**Figure 4D**). While *in vivo* work is needed to investigate this crucial point, this observation suggests that PCTP might resemble PAPTP and PCARBTP in acting specifically on cancerous cells, sparing others.

#### CONCLUSION

After PCARB, we have identified another mitochondriotropic prodrug of PAP-OH, PCTP, with marked pro-apoptotic effects on Kv1.3-expressing cancerous cells, including four PDAC lines. As is also the case for other mitochondriotropic psoralenic derivatives, its administration *in vitro* causes mitochondrial dysfunction and ROS generation. The results definitely warrant further testing in *in vivo* oncological models.

#### AUTHOR CONTRIBUTIONS

AMat, MR, AMan, LB, and LL performed experiments. AMat, LL, IS, MZ, and CP designed research. LL, IS, MZ, and CP analyzed results. LL, LB, MZ, AMat, and IS wrote the article.

### ACKNOWLEDGMENTS

The authors are grateful to Prof. H. Kalthoff for the gift of cell lines.

#### FUNDING

The work was supported by the Italian Association for Cancer Research (AIRC IG grant 15544 to IS), by the Italian Ministry of University and Education (PRIN 20107Z8XBW\_004 to MZ, PRONAT project to MZ and LB; PRIN 2015795S5W to IS), and

#### REFERENCES


by the CNR (Project of Special Interest on Aging and InterOmics; to MZ and LB). LL is the recipient of a young researcher grant of the University of Padova (no. GRIC12NN5G) and a PRID 2017 (no. BIRD162511) from the University of Padova. LL is grateful to EMBO for a short-term fellowship (ASTF 233-2014).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at https://www.frontiersin.org/articles/10.3389/fonc.2018.00122/ full#supplementary-material.


kills tumor cells in vivo. *Cancer Cell* (2017) 31(4):516–31.e10. doi:10.1016/j. ccell.2017.03.003


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

The reviewer VC declared a shared affiliation, with no collaboration, with the authors to the handling Editor.

*Copyright © 2018 Mattarei, Romio, Managò, Zoratti, Paradisi, Szabò, Leanza and Biasutto. 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 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.*

# Pushing the Limits of Cancer Therapy: The Nutrient Game

*Daniele Lettieri-Barbato\* and Katia Aquilano*

*Department of Biology, University of Rome Tor Vergata, Rome, Italy*

The standard cancer treatments include chemotherapy, radiotherapy, or their combination, which are generally associated with a multitude of side effects ranging from discomfort to the development of secondary tumors and severe toxicity to multiple systems including immune system. Mounting evidence has highlighted that the finetuning of nutrients may selectively sensitize cancer cells to conventional cancer therapies, while simultaneously protecting normal cells from their side effects. Nutrient modulation through diet also improves cancer immunesurveillance in a way that severe immunosuppression could be avoided or even the immune response or immune-based cancer therapies be potentiated also through patient microbiota remodeling. Here, we review recent advances in cancer therapy focusing on the effects of adjuvant dietary interventions (e.g., ketogenic diets, fasting) on the metabolic pathways within cancer cells and tumor environment (e.g., microbiota, immune system, tumor microenvironment) that are involved in cancer progression and resistance as well as cancer cell death. Finally, based on the overall literature data, we designed a nutritional intervention consisting in a plantbased moderate ketogenic diet that could be exploited for future preclinical research in cancer therapy.

#### *Edited by:*

*Salvatore Rizza, Danish Cancer Society, Denmark*

#### *Reviewed by:*

*Colin E. Champ, University of Pittsburgh Cancer Institute, United States Enrico Desideri, Universität Wien, Austria*

#### *\*Correspondence:*

*Daniele Lettieri-Barbato d.lettieribarbato@hotmail.it*

#### *Specialty section:*

*This article was submitted to Molecular and Cellular Oncology, a section of the journal Frontiers in Oncology*

> *Received: 12 February 2018 Accepted: 23 April 2018 Published: 08 May 2018*

#### *Citation:*

*Lettieri-Barbato D and Aquilano K (2018) Pushing the Limits of Cancer Therapy: The Nutrient Game. Front. Oncol. 8:148. doi: 10.3389/fonc.2018.00148*

Keywords: mitochondria, diet, fasting, immunemodulation, microbiota and immunity

# AN OVERVIEW ON THE CONTROL OF TUMOR PROGRESSION BY DIETARY INTERVENTIONS

A plethora of epidemiological and experimental data demonstrated the efficacy of geroprotective dietary regimens (e.g., fasting, calorie, proteins, or single amino acids restrictions) in cancer prevention (1–3). Furthermore, such dietary patterns are emerging to be effective in selectively killing cancer cells, whereas increasing resistance of normal cells to the toxic effects of the anticancer therapeutics.

Calorie restriction (CR), defined as 30–60% less of daily calorie requirement without malnutrition, is known to extend healthy life span from yeast to mammals (4). The anticancer effects of CR are known since several years (5). CR is particularly effective in reducing the incidence, mass, and metastasis of breast cancer cells (6, 7). Remarkably, applying CR in combination with radiotherapy enhanced the radiotherapy efficacy inducing a more pronounced apoptosis of breast cancer cells than radiotherapy alone (7). In human, however, CR requires high compliance challenges to be maintained for adequate therapeutic period. For these reasons, short period of fasting without malnutrition have been proposed as potentially safe interventions to be associated with cancer treatments (8).

Fasting is commonly defined as a time-controlled deprivation of all kinds of foods and dietary nutrients. Differently to nocturnal fasting, time-controlled fasting leads to a profound metabolic reprogramming building up adaptive stress responses that are involved in life and health span

**171**

extension (9–13). However, the adaptive stress responses induced by fasting occurring in normal cells differ from those activated by cancer cells because oncogenes might limit the activation of nutrient-sensing pathways while increasing chemotherapy vulnerability (8). Notably, proto-oncogenes such as IGF1R, PI3K, and AKT activate growth signaling and addict cancer cells to nutrient such as glucose and amino acids to meet their high proliferative rate (8). It has been shown that different cycles of fasting are effective in limiting tumor progression in several murine cancer models (14–17). However, the greatest effects were observed when fasting was combined with the conventional chemotherapy or radiotherapy (14–18). Interestingly, in these studies, fasting interventions alone do not cause clear signs of discomfort, but rather improve the animal condition. When fasting was combined with conventional therapies (e.g., temozolomide), most of the mice appeared healthy with the tumor-size below the controls, indicating that the combination of both treatments is well tolerated and improve tumor-bearing survival (14). The protective role of fasting against the side effects of anticancer therapy was confirmed in another study in which fasting was able to improve the overall cardiac response (maintenance of diastolic/systolic volumes and left ventricle wall thickness) to high-dose of doxorubicin (19). Fasting also exerted a significant protection against reduced mobility, ruffled hair, and hunched back posture caused by high dose of etoposide in mice (20). The anticancer effects of fasting might also rely on ketone bodies increase (21, 22). In support of this assumption, meta-analysis on ketogenic diets (KD), low in carbohydrates and high in fats, suggested a salutary impact on survival in animal models, with benefits prospectively linked to the magnitude of ketosis, time of diet initiation, and tumor location (23). Other evidence also demonstrated that KD might be safely used as adjuvant therapies to conventional radiation and chemotherapies (24). In particular, KD together with conventional radiotherapy led to increased radiation sensitivity in pancreatic cancer xenografts in mice (25). Similar results were obtained in mice bearing lung cancer xenografts (26). However, patients have demonstrated difficulty to comply with a KD while receiving concurrent radiation and chemotherapy in advanced lung and pancreatic cancer (25). Therefore, as better tolerated with respect to CR and KD, fasting appears to be more promising as adjuvant treatment in cancer therapy. Finally, it has been demonstrated that fasting could be replaced by the administration of CR mimetics, which showed the capability to improve the efficacy of chemotherapy as well. However, the objective response rates with metformin (27–30) or rapalogs (31) in clinical trials are still unclear and comparative analyses delineating a selective effectiveness of these drugs in cancer treatment and patient tolerability have to be more deeply elucidated.

#### NUTRIENT MODULATION IN PROLIFERATING/RESILIENT CANCER CELLS: A MOLECULAR VIEW

The reduced levels of nutrients and growth factors observed during fasting led to hypothesize their mandatory role in governing the differential stress responses in normal and cancer cells (10, 14, 16, 18). The different responses of normal and cancer cells to fasting shed light on their different sensitivity to nutrients and growth factors (18).

IGF-1/IGF-1R signaling is strongly dependent on nutrient availability and involves intensification of cancer cell proliferation, through the direct effects on PI3K/Akt signaling, and resistance to cell death imposed by chemotherapeutics and radiotherapy (**Figure 1**) (32). Indeed, fasting reduces circulating IGF-1 levels and this event protects mice deficient in the liver production of IGF-1 against chemotherapy drugs (16). Accordingly, restoration of IGF-1 was sufficient to reverse the protective effect of fasting (16). Reducing IGF-1 protects primary glia, but not glioma cells, against cyclophosphamide and mouse embryonic fibroblasts against doxorubicin (16). In the opposite manner, IGF-1 supplementation in starved breast cancer cells reversed drug sensitization. Overall, these findings strongly indicate that the fasting-mediated sensitization of cancer cells to chemotherapeutic drugs is conferred by the decrease of IGF-1 levels (15).

Nutrient shortage *per se* is able to increase mitochondrial reactive oxygen species (ROS) production in cancer cells arguing that limiting nutrient availability could enhance the effectiveness of redox-based cancer therapeutics (**Figure 1**) (33, 34). Actually, in breast cancer and melanoma cells, nutrient starvation was found to increase superoxide levels and aggravate oxidative stress caused by cyclophosphamide and cisplatin (15, 35). When applied in combination, fasting and chemotherapy act in synergy in elevating ROS levels and triggering DNA damage also in *in vivo* models of cancer (36). Micro-PET analyses in murine models of colon cancer cells revealed that fasting is effective as oxaliplatin (OXP) in reducing the average tumor glucose consumption and the lowest values were achieved by coupling fasting with OXP. In colon cancer cells, nutrient starvation upregulates oxidative phosphorylation with a significant production in mitochondrial superoxide caused by electron leakage. Consequently, starvation or OXP alone markedly increased ROS generation and their combination (starvation plus OXP) exacerbated ROS production in colon cancer cells (36). The hypothesis that cytotoxicity induced by glucose deprivation in cancer cells is mediated by mitochondrial superoxide and H2O2 was confirmed by exposing glucose-deprived transformed human fibroblasts to electron transport chain blockers (ETCBs), known to increase mitochondrial superoxide and H2O2 production (37). Glucose deprivation in the presence of ETCBs enhanced oxidative stress as well as cell death in several different human cancer cell lines (PC-3, DU145, MDA-MB231, and HT-29). In addition, human osteosarcoma cells lacking functional mitochondrial electron transport chain [rho(0)] were resistant to glucose deprivation-induced cytotoxicity and oxidative stress in the presence of antimycin A (complex III inhibitor), thus highlighting the role of mitochondrial ROS as mediators of cancer cell death (37).

The mechanisms by which KDs act as adjuvants in cancer therapy also seem to be associated with increased oxidative stress within cancer cells (24). Indeed, upon KD, the high level of circulating fatty acids limits the availability of glucose for glycolysis (Randle's Cycle) (38). This reduces the formation of pyruvate and glucose-6-phosphate and in turn the synthesis of

released fatty acids (4). The concomitant activation of OxPHOS metabolism and reduction in GSH levels are causative of oxidative stress (5) culminating in a massive lipid peroxidation (LOOH) (6) and ferroptosis in cancer cells (7). Diet low in starch, sugars, and proteins but rich in fatty acids also increases ketone bodies and modulates gut microbiota features by producing short-chain fatty acids (SCFAs). Ketone bodies and SCFAs affect PD-L1, nuclear factor-kb, and DNA repair enzymes genes transcription (8) promoting the chemo/radiotherapy vulnerability of cancer cells (9).

NADPH through the pentose phosphate pathway (PPP) (39). NADPH is necessary for buffering hydroperoxides (LOOH) production *via* the NADPH-dependent glutathione/glutathione peroxidase (GSH/GPX) system (40, 41). As consequence, an increase of LOOH is likely elicited (24) (**Figure 1**). Accordingly, hyperketotic diabetic patients have a higher level of lipid peroxidation in erythrocytes membrane and a significant decrease in cellular GSH levels than normal ketonic diabetic patients (42). Treatments with the ketone body acetoacetate elevated the levels of lipid peroxidation in human endothelial cells inhibiting their proliferation (42). This evidence suggests a direct role of ketone bodies in directly affecting GSH levels.

The main non-enzymatic cellular antioxidant GSH acts as an electron donor to reduce oxidized macromolecules, becoming itself oxidized in the process. Oxidized glutathione (GSSG) may then be restored in GSH through the action of the NADPHdependent glutathione reductase (43). This enzymatic process generates NADP<sup>+</sup>, which may be reconverted to NADPH using electrons obtained from different biochemical pathways (44). Thus, proliferating cancer cells develop a peculiar metabolic flexibility to maintain a functional redox threshold by regulating NADPH levels through glycolytic flux modulation (33). Indeed, glucose-addicted human cancer cells cultured in a low-glucose medium without serum and amino acids are able to reprogram their metabolism by shifting toward PPP, which sustains the production of NADPH to dampen oxidative stress (33). However, during the initial stages of solid tumor development, when cells migrate to the lumen of lymphatic or blood vessels by loss of attachment (LOA) to the extracellular matrix, the glucose availability could not be sufficient to produce an adequate amount of NADPH and proliferation is inhibited (45). Upon such environmental stress, cancer cells induce adaptive responses consisting in the activation of AMPK signaling that inhibits fatty acid synthesis and triggers fatty acids oxidation to maintain energy production and NADPH levels (46, 47). Although cancer cells build up such adaptive responses, it has been observed that during LOA, cancer cells undergo ATP and NADPH drop and increase ROS production (48). Several papers demonstrated that cancer cells experiencing glucose shortage might maintain their proliferative capacity and membrane biogenesis by the uptake of extracellular lipids (49). Accordingly, extracellular saturated fatty acids supplementation supports the proliferative demand for biomass synthesis of proliferating cells (50, 51). Otherwise, supplementation with polyunsaturated fatty acids (PUFA) induced a significant cytotoxic effect on cancer cells either alone (52–54) or in combination with conventional anticancer therapies (55, 56). Differently to saturated fatty acids, PUFA are strongly susceptible to peroxidation (lipid peroxidation) in *in vivo* systems (57, 58). This appears to be a key mechanism triggering cancer cell death (59). With all this in mind, forcing the changes in the membrane lipids composition by dietary/nutrient enrichment in PUFA might promote an intrinsic sensitivity toward lipid peroxidation (57, 58, 60) and cancer cell death (**Figure 1**).

#### NUTRIENT-MEDIATED COMMITMENT TO FERROPTOSIS IN CANCER CELLS

By preserving NADPH levels, cancer cells sustain GPX/GSH activity during nutrient limitation, and this may confer resistance to redox-based chemotherapeutics (61–63). Indeed, many rebel cancer cells use a common trick to evade annihilation; they enter into what is known as a mesenchymal state that is "epithelial-to-mesenchymal" transition, which provides cancer cell resistance to conventional therapeutic regimens (64). It has been demonstrated that high therapy-resistant mesenchymal cancer cells strictly rely on the selenium-dependent GPX4 for survival (65). By using the reducing power of GSH, GPX4 converts potentially toxic L-OOH to non-toxic lipid alcohols (L-OH) (**Figure 1**) (66–68). Accordingly, inactivation of GPX4 through GSH depletion with erastin, or with a direct GPX4 inhibitor, ultimately results in lipid peroxidation in cancer cells (69). It is thus provocative to hypothesize that the evolutionary pressure to maintain the selenium protein GPX4 might correlate with an organism's requirement for an increased PUFA content, which, in turn, renders complex biological activities possible (70).

Uncontrolled lipid peroxidation is causative of the onset of a metabolically regulated cell death called "ferroptosis," which is characterized by the iron-dependent formation of LOOH leading to cell death (**Figure 1**) (71). Sulfur amino acids play a key role in ferroptosis. In particular, agents that inhibit cystine uptake *via* the cystine/glutamate antiporter (XC system), such as sulfasalazine or erastin, arrest tumor growth and induce ferroptosis (72, 73). The uptake of cystine is followed by its NADPH-dependent conversion in cysteine, the ratelimiting amino acid precursor for the GSH biosynthesis (74). Direct depletion of cystine from plasma using an engineered cystine-degrading enzyme conjugate arrests tumor growth and triggers cell death (75). Agents that conjugate to GSH, as well as chemical or genetic inhibition of GSH biosynthesis, disrupt tumor cell growth and induce a ferroptosis-like form of cell death (76). Ferroptosis appears to be an effective cell death mechanism in cancer cells, since lipophilic antioxidant α-tocopherol or iron chelators, such as deferoxamine, efficiently dampen it (77). Hence, the presence of extracellular cysteine and cystine are crucial for growth and proliferation of various types of cancer, as these amino acids maintain GSH levels and prevent oxidative stress (**Figure 1**) (78–80). Because cysteine is limiting in the biosynthesis of GSH, some cancer cells, under cysteine unavailability, make use of the transsulfuration pathway to biosynthesize cysteine from methionine (Met), a dietary essential sulfur amino acid (81, 82). The essentiality of Met in cancer is supported by the evidence that some cancer cells display a higher sensitivity to Met shortage with respect to normal cells (83–87). The first steps of the transsulfuration pathway are the conversion to *S*-adenosylmethionine (SAM) and transfer of the methyl group of SAM to a large variety of methyl acceptors with formation of *S*-adenosylhomocysteine (SAH) (88), which can be then converted to homocysteine (Hcy) by SAH hydrolase (AHCY) (89). Alternatively, Hcy is converted to cystathionine by cystathionine β-synthase (CBS). CBS catalyzes the condensation of Hcy and serine, thereby forming cystathionine, which is subsequently cleaved to cysteine. Furthermore, exogenous cysteine is also essential for several cancer types (glioma, prostate, and pancreatic), as blocking its uptake through the cystine/glutamate antiporter reduces viability due to the cell death caused by uncontrolled oxidative stress (90–92). Similarly, CBS blockage reduces cancer cell proliferation and attenuates growth of patient-derived colon cancer xenografts models (93). Although these findings suggest that fasting or selective nutrient modulation could trigger ferroptotic cell death in cancer cells, a clear evidence linking nutrient availability to ferroptosis is still lacking. Several works demonstrated that starved cancer cells (mainly in amino acids) as well as cells lacking the enzyme producing NADPH from glucose (glucose-6-phosphate dehydrogenase) experience massive ROS production and autophagy-dependent cell death (33, 94, 95). Autophagy is a process described as intracellular removal of damaged organelles by self-degradative process (96). Interestingly, a tight relationship between autophagic cell death and ferroptosis is emerging (97–99). Indeed, it seems that autophagy activation leads to a degradation of ferritin (ferritinophagy) (97), thus increasing the intracellular free iron levels promoting ROS production and ferroptosis (**Figure 1**) (99).

# DIETARY STRATEGIES TO BOOST THE IMMUNOMETABOLIC RESPONSES IN CANCER THERAPY

Short-term fasting has a beneficial impact on cancer immunosurveillance (100). In particular, Pietrocola and co-workers demonstrated that fasting or CR-mimicking drugs, induce the depletion of regulatory T cells (which dampen anticancer immunity), thus igniting autophagic flux in murine models of KRAS-induced lung cancers. Accordingly, the inhibitory effect of fasting on tumor growth is lost in cancers that have been rendered autophagy deficient (100). Recently, also, isocaloric diet with protein restriction has been demonstrated to induce an IRE1α-dependent UPR in cancer cells, enhancing cytotoxic CD8<sup>+</sup> T cell (a type of effector T lymphocyte)-mediated response against tumors (101).

Similarly to what observed with prolonged fasting (102), cycles of a fasting-mimicking diet (FMD) are effective in increasing hematopoietic cells proliferation and promoting immune system regeneration and modulation (103). Importantly, FMD has stimulatory effect on common lymphoid progenitor cells and CD8<sup>+</sup> T cell-dependent cytotoxicity on breast cancer and melanoma cells (**Figure 2**) (17, 102). The presence of cytotoxic CD8+ T cells in the tumor environment [tumor infiltrating lymphocytes (TIL)] is considered a positive outcome of the cancer treatment (104, 105).

CD8+ T cells are influenced by nutrients and other supportive signals that are generally available in their environment. Generally, tumor cells inactivate CD8<sup>+</sup> T cells. The suppression of oxidative phosphorylation and an upregulated glycolytic flux

of proliferating cancer cells create an immunosuppressive microenvironment (106). Indeed, the glucose-dependent CD8<sup>+</sup> TIL could undergo a competitive disadvantage for nutrients, and this would negatively affect their immune function. The immunosuppressive metabolic environment could be further enhanced by tumor expression of inhibitory ligands for programmed death 1 receptor (PD-1) which, when bound to their cognate receptors on T cells, limits T cell-intrinsic glucose uptake and glycolysis (107, 108). It has been reported that KD significantly reduces the expression of the inhibitory ligand PD-1 (PD-L1) on CD8<sup>+</sup> TIL (109). Additionally, mice fed with KD have reduced expression of PD-L1 on the cancer cells that notoriously inhibits CD8<sup>+</sup> T cells activity (109). This suggests that KD may alter tumor-mediated T cell suppression by reducing the number of cells that are susceptible to inhibition through the PD-1 inhibitory pathway (**Figure 2**).

Nowadays, there has been intense interest in developing adoptive T cells immunotherapy (ACI), which consists in reintroducing into a patient T cells that are previously activated and expanded *in vitro* (110, 111). The success of the ACI depends on the replicative capacity of implanted T cells. A large amount of research has been directed in optimizing T cell activation and using appropriate adjuvants for ACI. However, few experimental studies have been focused on manipulating metabolic pathways that could potentially enhance immunotherapy efficacy. When posed in culture, T cells dispose of a high glucose availability, which is far from the glucose physiological levels especially in the tumor environment (112, 113). Thus, once reintroduced in patients, T cells suffer from low glucose levels and show a moderate survival and replicative capacity. It has been reported that limiting glycolysis in cultured T cells can increase their longevity without inhibiting proliferative capacity (114, 115) (**Figure 2**). Another potential way to enhance the replicative capacity and longevity of ACI cells is promoting oxidative phosphorylation and mitochondrial biogenesis *via* the inhibition of glucose-related signaling pathway that ultimately leads to *in vivo* persistence and improved antitumor immunity (116). The metabolic reprogramming of infiltrating glycolytic lymphocytes toward a catabolic state reliant on fatty acid oxidation appears to assure the success of immunotherapy (113). In line with this assumption, it was recently demonstrated that the enhancement of lipid catabolism in CD8<sup>+</sup> T cells increases the efficacy of immunotherapy within a tumor microenvironment low in glucose (117). In a mouse model of malignant glioma, an enhanced cytolysis *via* tumorreactive CD8<sup>+</sup> T cells was also achieved by ketogenic diet (109). The immunometabolic reprogramming necessary for CD8<sup>+</sup> TIL could at least partially explain the mechanism by which KD or fasting enhances cytotoxic effect against cancer cells. Such diets are indeed powerful in inducing a cellular metabolic shift from glycolysis toward FAO.

It is now emerging that CD8<sup>+</sup> TIL response to immune checkpoint blockade inhibitor PD1 can be also modulated by gut microbiota (118–120). A very recent paper has revealed that fecal microbiota from patients affected with metastatic melanoma and responsive to anti-PD1 therapy display increased abundance of *Akkermansia muciniphila*. *A. muciniphila* introduction into mice receiving human nonresponder fecal microbiota transplant improved antitumor immune CD8<sup>+</sup> T cell infiltration and activity and increased anti-PD1 therapy efficacy (120, 121). Another intriguing observation is that *Faecalibacterium* and *Bifidobacterium* are associated with anti-inflammatory responses, a regulatory arm of the immune system that aims to prevent overactivation of the immune response and restores host homeostasis (120). Given that changes in host metabolism and microbiota can occur in tandem, it was hypothesized that gut microbial diversity and composition are predictors of the response to

environment to cancer cells.

cancer therapy (121) (**Figure 2**). Accordingly, germ-free mice implanted with human tumor cells and transplanted with feces from chemotherapy responders showed an ameliorated response to chemotherapy than mice colonized with microbiota from nonresponder patients (119).

The diet has a strong capacity to rapidly and reproducibly reshape the gut microbiome (122). Indeed, fasting or plant-based diet remodels microbial community structure and overwhelms interindividual differences in microbial gene expression. The animal-based diets are known to increase the abundance of biletolerant microorganisms (*Alistipes*, *Bilophila*, and *Bacteroides*) and decrease the levels of the high fermentative *Firmicutes* that metabolize dietary plant polysaccharides (*Roseburia, Eubacterium rectale*, and *Ruminococcus bromii*) (122). More recently, it has been demonstrated that alternate day fasting shifts the gut microbiota composition from *Bacteroides* to *Firmicutes* leading to elevation of the fermentation products (123). Plant-based foods are mainly characterized by resistant starches and dietary fibers and promote their gut microbiota-mediated fermentation and decomposition. These processes provide additional amount of short chain fatty acids (SCFAs) to the host (124) (**Figure 2**). The major SCFAs, i.e., acetate, propionate, and butyrate, have different production ratios and physiological activities. Through 1 H NMRbased metabolomics, it was revealed that mice treated with alternate day fasting increased acetate levels both in the cecum and sera (123). Acetate, when ligated to coenzyme A (acetyl-CoA), is among the most central and dynamic metabolites in intermediary metabolism. Under stressful circumstance (e.g., fasting-like conditions), cancer cells may convert extracellular acetate to acetyl-CoA, thus promoting the biogenesis of membrane building blocks that sustain the high proliferative rate. This adaptive response involves the cytosolic form of short-chain acyl-CoA synthetases (ACC2). Accordingly, increased ACC2 protein levels were detected in a subset of human triple negative breast cancer samples, and such an elevation correlates with poor survival (125). Differently to acetate, butyrate shows many regulatory properties including the inhibition of histone deacetylases. Histone deacetylase inhibitors (HDACi's) are emerging as promising anticancer drugs when administered alone or in combination with chemotherapeutic agents or radiotherapy. Previous research suggests that HDACi's have a high degree of selectivity for killing cancer cells. For instance, the HDACi sodium butyrate suppresses DNA double strand break repair induced by etoposide more efficiently in MCF-7 cells than in HEK293 cells (126). Sodium butyrate alone also resulted in accumulation of ROS, DNA double-strand breaks, and apoptosis in HCT-116 colon cancer cell lines; when combined with mitomycin C or radiotherapy, sodium butyrate increases sensitivity of cancer cells to the drug (127, 128). In animal models of gastric carcinoma, sodium butyrate was found to inhibit tumor mass formation and increase tumor infiltration by CD8<sup>+</sup> TIL (129). Finally, several studies also demonstrated a strong effectiveness of SCFA to inactivate nuclear factor-kb by downregulating the production of the pro-inflammatory cytokine TNFα (130–134), which is commonly activated to promote a pro-carcinogenic environmental milieu (135) (**Figure 1**).

#### CONCLUSION AND PERSPECTIVE

Despite recent advances have been made in cancer therapy, the prognosis for many cancer patients remains poor, and current treatments still show severe adverse events. Thus, finding complementary treatments that have limited patient toxicity and simultaneously enhance therapy responses in cancer versus normal cells is urgent. Diet has a strong capacity to modulate cell responses to environmental stimuli and shows great potential in improving cancer prognosis. The mechanisms by which dietary nutrients enhance anticancer effects of standard anticancer therapies (chemotherapy, radiotherapy, immunotherapy) has not been fully elucidated yet. Preclinical studies have demonstrated the safety and efficacy of specific dietary interventions in counteracting tumor progression during anticancer therapy in murine models. However, most of the data present in the literature take advantage of the use of mice and this may limit the translation to clinical research. Therefore, a huge amount of work is now necessary to confirm these very promising results in humans.

Deprivation of nutrients (e.g., glucose, sulfur amino acids) as well as of nutrient-responsive growth factors (e.g., IGF-1) seems to selectively kill high proliferative/resilient cancer cells by forcing their glycolytic asset toward an oxidative metabolism (i.e., fatty acids and ketone bodies as energy sources) and limiting GPX activity as consequence of reduced GSH levels. Nutrient scarcity also improves immunometabolism enhancing cytotoxic efficiency of CD8<sup>+</sup> TIL within the tumor mass through, probably, the concomitant gut microbiota and immunometabolic rearrangements (**Figure 3**).

Herein, we propose weekly cycles of 4 days of a plant-based moderate ketogenic diet (k-PBD) that could reprogram systemic metabolism conferring a hostile environment to cancer cells (**Figure 3**). In particular, k-PBD should be low in proteins (mainly vegetable proteins low in sulfur amino acids and selenium),

#### REFERENCES


carbohydrates (non-starchy vegetables), and high in lipids (mainly unprocessed vegetable oils rich in PUFA). Remarkably, even though not supported by experimental data, it is highly expected that this diet could be able to increase ketonemia as it contains high amounts of fats concomitantly to reduced calories. This diet could increase the efficiency of CD8+ TIL, by reprogramming their metabolism (fat-dependent metabolism) to better counteract the metabolic features of proliferating cancer cells (glucose-dependent metabolism) and sensitize cancer cells to the therapy. The k-PBD could be consumed prior to conventional cancer therapies (e.g., prior each cycle of chemotherapy or prior a single fraction of radiation therapy). With this composition and time of treatment, k-PBD could be effective in: (i) changing the membrane chemistry by PUFA enrichment (high peroxidation index); (ii) reducing the sulfur-dependent antioxidant power (lowering NADPH, GSH, GPX4); (iii) forcing the metabolic shift toward mitochondrial metabolism in cancer cells. Furthermore, the high fermentative fibers of k-PBD could induce a functional microbiota reshaping improving immunotherapy efficacy (e.g., anti-PD1 therapy) (**Figure 4**).

## AUTHOR CONTRIBUTIONS

DL-B conceptualized and wrote the manuscript. KA performed critical revision of the manuscript for intellectual content.

L-proline catabolism to induce a transient ROS signal. *Cell Metab* (2012) 15:451–65. doi:10.1016/j.cmet.2012.02.013


by distinct mechanisms. *Mol Cell Biol* (2005) 25:9543–53. doi:10.1128/ MCB.25.21.9543-9553.2005


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

*Copyright © 2018 Lettieri-Barbato and Aquilano. 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 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.*