# METABOLISM MEETS FUNCTION: UNTANGLING THE CROSS-TALK BETWEEN SIGNALLING AND METABOLISM

EDITED BY : Alessandra Castegna, Paolo E. Porporato and Daniel McVicar PUBLISHED IN : Frontiers in Oncology

#### Frontiers eBook Copyright Statement

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

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

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

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

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

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

ISSN 1664-8714 ISBN 978-2-88966-262-3 DOI 10.3389/978-2-88966-262-3

### 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

# METABOLISM MEETS FUNCTION: UNTANGLING THE CROSS-TALK BETWEEN SIGNALLING AND METABOLISM

Topic Editors:

Alessandra Castegna, University of Bari Aldo Moro, Italy Paolo E. Porporato, University of Turin, Italy Daniel McVicar, National Cancer Institute (NCI), United States

Citation: Castegna, A., Porporato, P. E., McVicar, D., eds. (2020). Metabolism Meets Function: Untangling the Cross-Talk Between Signalling and Metabolism. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-262-3

# Table of Contents

*05 Editorial: Metabolism Meets Function: Untangling the Cross-Talk Between Signaling and Metabolism*

Alessandra Castegna, Daniel W. McVicar, Annalisa Campanella, Erika M. Palmieri, Alessio Menga and Paolo E. Porporato


Iñigo San-Millán, Colleen G. Julian, Christopher Matarazzo, Janel Martinez and George A. Brooks


Christine M. Heske

*54 Non-invasive Investigation of Tumor Metabolism and Acidosis by MRI-CEST Imaging*

Lorena Consolino, Annasofia Anemone, Martina Capozza, Antonella Carella, Pietro Irrera, Alessia Corrado, Chetan Dhakan, Martina Bracesco and Dario Livio Longo

*63 Cross-Talk Between the Tumor Microenvironment, Extracellular Matrix, and Cell Metabolism in Cancer*

Mona Nazemi and Elena Rainero


Fátima Baltazar, Julieta Afonso, Marta Costa and Sara Granja

*100 NAMPT and NAPRT: Two Metabolic Enzymes With Key Roles in Inflammation*

Valentina Audrito, Vincenzo Gianluca Messana and Silvia Deaglio

*117 Nutritional Exchanges Within Tumor Microenvironment: Impact for Cancer Aggressiveness*

Giuseppina Comito, Luigi Ippolito, Paola Chiarugi and Paolo Cirri


Kshama Gupta, Ivan Vuckovic, Song Zhang, Yuning Xiong, Brett L. Carlson, Joshua Jacobs, Ian Olson, Xuan-Mai Petterson, Slobodan I. Macura, Jann Sarkaria and Terry C. Burns

*196 TGF*β *Signaling Increases Net Acid Extrusion, Proliferation and Invasion in Panc-1 Pancreatic Cancer Cells: SMAD4 Dependence and Link to Merlin/NF2 Signaling*

Raj R. Malinda, Katrine Zeeberg, Patricia C. Sharku, Mette Q. Ludwig, Lotte B. Pedersen, Søren T. Christensen and Stine F. Pedersen

*211 Dynamically Shaping Chaperones. Allosteric Modulators of HSP90 Family as Regulatory Tools of Cell Metabolism in Neoplastic Progression* Carlos Sanchez-Martin, Stefano A. Serapian, Giorgio Colombo and Andrea Rasola

# Editorial: Metabolism Meets Function: Untangling the Cross-Talk Between Signaling and Metabolism

Alessandra Castegna1,2\*, Daniel W. McVicar 3\*, Annalisa Campanella1 , Erika M. Palmieri <sup>3</sup> , Alessio Menga<sup>4</sup> and Paolo E. Porporato4\*

<sup>1</sup> Department of Biosciences, Biotechnologies and Biopharmaceutics, University of Bari, Bari, Italy, <sup>2</sup> IBIOM-CNR, Institute of Biomembranes, Bioenergetics and Molecular Biotechnologies, National Research Council, Bari, Italy, <sup>3</sup> Laboratory of Cancer Immunometabolism, Center for Cancer Research, National Cancer Institute (NCI), Frederick, MD, United States, <sup>4</sup> Department of Molecular Biotechnology and Health Science, Molecular Biotechnology Center, University of Torino, Torino, Italy

Keywords: metabolism, cancer, Warburg effect, mitochondria, oncometabolite

Editorial on the Research Topic

### Metabolism Meets Function: Untangling the Cross-Talk Between Signaling and Metabolism

#### Edited and reviewed by:

Michael P. Lisanti, University of Salford, United Kingdom

#### \*Correspondence:

Paolo E. Porporato paolo.porporato@unito.it Alessandra Castegna alessandra.castegna@uniba.it Daniel W. McVicar mcvicard@mail.nih.gov

#### Specialty section:

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

Received: 17 September 2020 Accepted: 30 September 2020 Published: 20 October 2020

#### Citation:

Castegna A, McVicar DW, Campanella A, Palmieri EM, Menga A and Porporato PE (2020) Editorial: Metabolism Meets Function: Untangling the Cross-Talk Between Signaling and Metabolism. Front. Oncol. 10:607511. doi: 10.3389/fonc.2020.607511 Tumor metabolism is a long established field in cancer biology, as the seminal findings of OttoWarburg date back to the 1920s. Since then, the discovery that oncogenes, besides promoting the Warburg effect, modulate anabolic pathways, has prompted scientists to re-evaluate the role that tumor metabolism plays in the neoplastic process. Today, metabolic reprogramming of neoplastic cells is considered a hallmark of cancer, with the discovery that flexibility in the acquisition of various cellular characteristics is supported by specific metabolic pathways. Clinical and pharmacological advances, for example the application of FDG-PET in the clinical setting (1) and the development of novel pharmacological strategies based on antimetabolites (2), provide further support and validation of the role of metabolism in cancer. Here, we present a collection of works with the aim of bringing together work from a variety of scientists across the field of tumor metabolism toward an understanding of how different metabolic pathways are activated in neoplastic and surrounding cells, the mechanisms linking altered metabolism to tumorigenesis and the potential for pharmacological applications.

One of the most prominent metabolic adaptations typical of cancer cells is sustained aerobic glycolysis resulting in the consumption of high amounts of glucose even in the presence of oxygen. For some time, it has been known that highly glycolytic cells typically accumulate what was characterized initially as a by-product, lactate. However, many researchers are now identifying novel properties of lactate, including roles in the cancer–cancer and cancer–stromal shuttles, and as a signaling oncometabolite, as nicely reviewed by Baltazar et al. Similarly, we present work by San-Millán et al. further elaborating on the potential effects of lactate overload by describing its role in supporting tumor aggressiveness, regulating transcriptional signatures associated with proliferation and upregulating oncogenes in breast cancer cells. Lactic acid accumulation also results in a drop in extracellular pH, a feature commonly described in cancer and known to promote aggressiveness (3). Not surprisingly, several signaling pathways independently converge to trigger net acid extrusion decreasing extracellular pH. Work by Malinda et al., reported in our collection, details an example of a signaling pathway regulated this way in cancer, the TGF-b signaling.

To date, despite the well-established knowledge that acidosis is a recurring issue in cancer, limited tools are available to map tumor pH in vivo and reveal the spatial distribution of acidic areas within the tumor. In an effort to close this knowledge gap, in our collection Consolino et al. present the use of MRI-CEST imaging to map tumor metabolism. In parallel, Cavallari et al. address the promise of

**5**

exploiting glycolytic metabolism for clinical purposes by describing novel methods to obtain Hyperpolarized [1-13C] Pyruvate for metabolic imaging.

In part as a result of their increased glycolytic flux, tumors require high levels of NAD+. Nicotinamide phosphoribosyltransferase (NAMPT), the rate-limiting enzyme in the NAD+ salvage pathway, is upregulated in many cancers and, as such, pharmacological targeting of NAMPT represents an interesting approach to block cancer growth. In our collection, Heske highlights recent findings describing the effects of NAMPT inhibitors on the non-metabolic functions of malignant cells, supporting utilization of co-targeted therapies consisting of NAMPT inhibitors and other drugs to fully exploit the multiple functions of this enzyme.

The impact of NAMPT biology is not limited to the cancer cells themselves. In a companion contribution in our collection, Audrito et al. describe the peculiar role of NAMPT and nicotinate phosphoribosyltransferase (NAPRT) in cells of the immune system during inflammation. These enzymes are released as soluble factors with cytokine/adipokine/DAMP-like actions in inflammatory settings making them possible "two hit" targets in inflammatory cancers. The authors review the available data concerning the interesting and unique dual roles of this family of enzymes in inflammation.

Despite their extraordinary rates of aerobic glycolysis, tumors are not simply glycolytic cells. As Ordway et al. remind us in this collection, tumors display heterogeneous metabolism as an essential component of their physiologic robustness. Indeed, metabolic plasticity is an essential component of tumor resistance to stress, including stress derived from exposure to chemotherapeutics as described by Desbats et al., or radiation as detailed by Gupta et al. herein. Mitochondria play an important role in conferring this metabolic plasticity, as nicely addressed in our collection by Fiorito et al., who describe the role of iron, and in particular of heme, in cancer. Furthermore, we present work by Sanchez-Martin et al. unraveling the importance of mitochondrial metabolism in cancers by addressing the role of TRAP1, a mitochondrial chaperone protein belonging to the HSP90 family, in controlling cancer metabolism, and defining its role as a potential pharmacological target.

A tumor consists not only of transformed cells, but also on nontransformed stromal cells, such as endothelial cells, fibroblasts and macrophages, that can be recruited and hijacked by cancer cells, promoting tumor progression. Because of the important role they play in tumor malignancy, it is crucial to unravel and understand the complexity of the mechanistic relationships between the various cell types within the tumor that constitute the so-called tumor microenvironment (TME) (4). In our collection, Comito et al. describe the metabolic remodeling of the different cell populations within the TME, focusing on reciprocal re-education through the symbiotic sharing of metabolites, acting both as nutrients and transcriptional regulators, and evaluating their impact on tumor growth and metastasis. In addition, Nazemi and Rainero describe how cancer-associated fibroblasts of the TME dictate cancer cell metabolism, describing the impact of nutrient scavenging from the microenvironment in cancer cell growth. Their contribution also focuses on the cross-talk between nutrient signaling and the trafficking of the extracellular matrix (ECM) receptors of the integrin family; critical aspects of tumor aggressiveness.

An important stromal component of the TME are tumorassociated macrophages (TAMs) (5, 6), which contribute to several steps in the formation of metastasis (7, 8), and are recruited through mechanisms that can be mediated by functionally relevant metabolic reprogramming. Identification of the metabolic checkpoints regulating macrophage function, which might be targeted to improve cancer specific immune responses is now a promising strategy for therapeutic intervention. The metabolic mechanisms of macrophage polarization and functional skewing are starting to emerge (9–11), but more needs to be unraveled with respect to trace elements. Metal ions are involved in various biological processes. In our collection, Serra et al. describe new findings regarding the role of these micronutrients in metabolic and cellular signaling mechanisms in macrophages. The processes they describe are components of the "metallic" cross-talk between macrophages and cancer cells and may represent opportunities for innovative pharmaceutical or dietary interventions in cancer therapy.

Amino acid metabolism is also crucial for cancer development. Activation of signaling pathways associated with proliferation can lead to amino acid depletion. Furthermore, amino acid deprivation occurring in cancer tissues might be overcome by the ability of cancer cells to synthesize the given specific amino acid. Herein, Chiu et al. nicely address the role of asparagine synthase (ASNS) and asparagine in cancer. Asparagine biochemistry is gaining more attention in the scientific community with the revelation that ASNS is overexpressed in some cancers, promoting cell proliferation, chemoresistance, and metastasis formation. During proliferation, amino acid metabolism is upregulated via the mTOR pathway, a signaling network known to contribute to cancer progression. Different mechanisms and factors regulate mTOR function. In our collection Gozzelino et al. elaborate on the role of one of these, phosphatidylinositol 3-4bisphosphate (PI(3,4)P2), showing it to be a novel emerging signaling molecule that regulates biological functions, and acts as an effector of metabolic reprogramming events relevant in cancer development.

Taken together, the work in the present collection represent a unique contribution to our understanding of the mechanisms, and the functional outcomes, associated with metabolic reprogramming in cancer.

### AUTHOR CONTRIBUTIONS

AC and PEP drafted the paper. AM, EP, and ALC critically read and edited. AC and DMV finalized the paper. All authors listed have made a substantial, direct, and intellectual contribution to the work, and approved it for publication.

### FUNDING

This work was funded, in part, by the intramural research program of the NIH, Center for Cancer Research of the National Cancer Institute (DMV) and from AIRC (MFAG 21564-PEP).

### REFERENCES


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

Copyright © 2020 Castegna, McVicar, Campanella, Palmieri, Menga and Porporato. 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.

# Asparagine Synthetase in Cancer: Beyond Acute Lymphoblastic Leukemia

Martina Chiu<sup>1</sup> , Giuseppe Taurino<sup>1</sup> , Massimiliano G. Bianchi <sup>1</sup> , Michael S. Kilberg<sup>2</sup> and Ovidio Bussolati <sup>1</sup> \*

*<sup>1</sup> Laboratory of General Pathology, Department of Medicine and Surgery, University of Parma, Parma, Italy, <sup>2</sup> Department of Biochemistry and Molecular Biology, University of Florida College of Medicine, Gainesville, FL, United States*

Asparagine Synthetase (ASNS) catalyzes the synthesis of the non-essential amino acid asparagine (Asn) from aspartate (Asp) and glutamine (Gln). ASNS expression is highly regulated at the transcriptional level, being induced by both the Amino Acid Response (AAR) and the Unfolded Protein Response (UPR) pathways. Lack of ASNS protein expression is a hallmark of Acute Lymphoblastic Leukemia (ALL) blasts, which, therefore, are auxotrophic for Asn. This peculiarity is the rationale for the use of bacterial L-Asparaginase (ASNase) for ALL therapy, the first example of anti-cancer treatment targeting a tumor-specific metabolic feature. Other hematological and solid cancers express low levels of ASNS and, therefore, should also be Asn auxotrophs and ASNase sensitive. Conversely, in the last few years, several reports indicate that in some cancer types ASNS is overexpressed, promoting cell proliferation, chemoresistance, and a metastatic behavior. However, enhanced ASNS activity may constitute a metabolic vulnerability in selected cancer models, suggesting a variable and tumor-specific role of the enzyme in cancer. Recent evidence indicates that, beyond its canonical role in protein synthesis, Asn may have additional regulatory functions. These observations prompt a re-appreciation of ASNS activity in the biology of normal and cancer tissues, with particular attention to the fueling of Asn exchange between cancer cells and the tumor microenvironment.

### Edited by:

*Paolo E. Porporato, University of Turin, Italy*

#### Reviewed by:

*Nigel Richards, Cardiff University, United Kingdom Cesare Indiveri, University of Calabria, Italy*

> \*Correspondence: *Ovidio Bussolati ovidio.bussolati@unipr.it*

#### Specialty section:

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

Received: *21 October 2019* Accepted: *10 December 2019* Published: *09 January 2020*

#### Citation:

*Chiu M, Taurino G, Bianchi MG, Kilberg MS and Bussolati O (2020) Asparagine Synthetase in Cancer: Beyond Acute Lymphoblastic Leukemia. Front. Oncol. 9:1480. doi: 10.3389/fonc.2019.01480* Keywords: asparagine synthetase, acute lymphoblastic leukemia, asparagine, glutamine, cancer

## INTRODUCTION

Asparagine Synthetase (asparagine synthase (glutamine-hydrolysing) or glutamine-dependent asparagine synthetase, E.C. 6.3.5.4, ASNS) catalyzes the biosynthesis of asparagine (Asn) from aspartate through an ATP-dependent reaction that exploits the amido-N of glutamine (Gln) to form the amido group of Asn.

The human ASNS gene is located at chromosome 7q21.3 and is 35 kb long with 13 exons (1). The ASNS protein (561 aa) has two primary domains, termed the N- and C-terminal domains, and is expressed in many tissues, with a wide range of expression levels. Particularly high levels of expression are detected in the pancreas, brain, thyroid and testes, while the liver has low expression of ASNS. Several transcript varieties and putative isoforms of human ASNS have been described although information on their role in physiology and pathology is lacking.

**8**

ASNS deficiency (ASNSD, OMIM 615574) is an autosomal recessive, rare, severe disorder associated with congenital microcephaly, cognitive impairment, progressive cerebral atrophy, intractable seizures, and early death (2, 3). The prevalence of neurologic symptoms suggests that ASNS plays a unique role in brain development. Interestingly, plasma and cerebral spinal fluid Asn levels are lowered only in some of the patients tested thus far, preventing diagnosis on biochemical bases (4). For more detailed information on ASNS structure, enzymatic mechanism, and mutations associated with ASNSD, the reader is referred to recent reviews and original articles (5–7). In particular, the high-resolution crystal structure of human ASNS recently provided by Zhu et al. (7) indicates that the enzyme is composed of two domains, with the C-terminal synthetase domain more similar to ASNS in other organisms than the N-terminal glutaminase domain. The glutaminase domain has a topology similar to that of other amidotransferases and other conserved amino acid residues are present at the interface of the two domains where substrate recognition occurs. Also the amino acids in the synthetase site are for the most part conserved in human and bacterial ASNS.

### ASNS REGULATION

Numerous studies have placed ASNS at the center of the cell response to amino acid deprivation and other forms of cellular stress [reviewed in (5, 8–10)]. Through transcriptional regulation, the ASNS gene is a target of two signaling pathways aimed to ensure cell survival under conditions of imbalanced amino acid availability, named the Amino Acid Response (AAR) (9), and of increased endoplasmic reticulum stress, the Unfolded Protein Response (UPR) (10). Through the activation of, respectively, the GCN2 and the PERK kinases, both these stress-response pathways converge on the phosphorylation of the α-subunit of the initiation factor eIF2, which provokes the attenuation of global protein synthesis and, at the same time, the preferential translation of a selected population of mRNAs, including the transcription factor ATF4. ATF4 is the major factor for ASNS induction, working as a trans-activator through the binding to an enhancer element within ASNS promoter (8). A very recent contribution (11) demonstrates that in Asn-depleted cancer cells a translational reprogramming, dependent on the increase of MAPK-interacting kinase 1 (MNK1) and eukaryotic translation initiation factor 4E (eIF4E), promotes enhanced ATF4 translation and, hence, ASNS expression. The role of other components of the UPR, such as IRE and ATF6, seems less important (12). However, ASNS transcription is also influenced by factors such as p53, which can serve as a negative regulator of the gene (13).

### LOW ASNS EXPRESSION IN ACUTE LYMPHOBLASTIC LEUKEMIA: OLD OBSERVATIONS AND NEW PERSPECTIVES

Interest in the role of ASNS in cancer was initially due to the observation of low synthetic activity for Asn in malignant tissues (14, 15), which were, therefore, auxotrophic for Asn, thus accounting for sensitivity to bacterial L-asparaginases (ASNase). The widespread clinical use of ASNase in acute lymphoblastic leukemia (ALL) began in the 1970s and today is a cornerstone of multi-drug therapy for this hematological cancer (16, 17). Thus, ASNase represents the first, and until now uniquely successful, example of a therapeutic approach targeting a metabolic feature of a specific form of cancer. Moreover, the strict requirement for extracellular Asn of ALL blasts (and of some lymphoma models), due to low levels of ASNS protein expression, was the first example of a cancer-specific auxotrophy for a non-essential amino acid (18). More recently, other examples have been described in human cancers, such as the loss of argininosuccinate synthetase in hepatocellular carcinomas, metastatic melanomas, and other cancers, leading to auxotrophy for arginine (19), and the absence of glutamine synthetase expression in multiple myeloma (20) and oligodendroglioma (21), leading to Gln auxotrophy.

Given the low expression of ASNS, the incubation of ALL blasts with ASNase is rapidly followed by the fall of intracellular Asn and by a prolonged nutritional stress, which causes proliferative arrest and, eventually, apoptotic death of leukemia cells. In most normal and cancer cell types investigated thus far, ASNS mRNA and protein expression is rapidly increased upon Asn deprivation, as a result of the transcriptional response to the AAR (see below) but, while the fast increase in mRNA occurs also in ALL cells (22), the increase in protein is severely delayed, suggesting the existence of an active translational silencing mechanism. It is this delay in the increase of ASNS protein expression that renders ALL cells sensitive to ASNase (23). Possible translational control may explain the numerous clinical reports that showed no correlation between ASNS mRNA and ASNase sensitivity (24). Recently, Jiang et al. have demonstrated that methylation status of the ASNS promoter is not the same in different ALL models and that hypermethylation inversely correlates not only with the basal ASNS expression but also with the capacity to trigger the ATF4-dependent increase in ASNS expression upon following Asn depletion (25). However, although, intuitively, ASNS induction has been correlated with resistance to ASNase, it has been known for many years that ASNase-resistant ALL cells present a complex phenotype. Indeed, if ASNS overexpression is sufficient to induce the ASNase-resistant phenotype in specific ALL cell models (22), adequate availability of the ASNS substrates Gln and aspartate (Asp) requires multiple adaptation mechanisms (26, 27) (**Figure 1**).

The relationship between ASNS expression and ASNase sensitivity/resistance has also been complicated by the fact that both the bacterial ASNases exploited in therapy, derived from Escherichia coli or from Erwinia chrysantemi [now Dickeya dadantii (28)], are endowed with a low level of glutaminase activity (29). Therefore, after ASNase infusion, the depletion of both Asn and Gln ensues, although at different levels of severity and with different kinetics. The capacity of counteracting glutamine depletion is obviously also relevant for the cellular adaptation to ASNase-dependent nutritional stress. Indeed, ASNS protein induction would be functionally less effective in conditions of severe cell depletion of Gln, since

FIGURE 1 | Mechanisms involved in resistance to L-asparaginase. Upper panel, ASNase catalyzes the hydrolysis of asparagine (Asn) into aspartate (Asp) and of glutamine (Gln) into glutamate (Glu), driving low-ASNS cells to cell death. Central panel, ASNS induction and increase in GS protein expression are not able to rescue ASNase-induced apoptotis due to poor availability of their substrates Asp and Glu. Lower panel, the overexpression of EAAT1 or EAAT3 anionic amino acid transporters provides Glu (for the synthesis of Gln, through Glutamine Syntethase) and Asp (27). Both Gln and Asp are needed for Asn synthesis via ASNS and for an effective cell rescue. The model is mainly based on data obtained with prostate cancer cells by Sun et al. (27) but it may apply to other low-ASNS cancers.

human ASNS requires Gln as its obliged ammonia-donating substrate (**Figure 1**).

The issue of the relevance of the glutaminase activity for the antileukemic effects of ASNase has been widely debated [see for review (30)]. In the last few years, importance has been attributed to residual ASNS protein expression in ALL blasts. Chan et al. obtained a mutant E. coli ASNase, devoid of glutaminase activity, which is fully effective toward ASNS-null ALL blasts, but not toward ALL blasts with a residual expression of ASNS protein (31). Unfortunately, no attempt was made to correlate ASNS protein expression to enzymatic activity. However, more recent results from the same group, obtained with a murine leukemia model of ASNS-null ALL, indicate that, actually, glutaminase activity was needed for a durable suppression of the tumor (32). Further investigation is needed to fully understand the role of cellular Gln levels on ASNase sensitivity and glutaminase action in ALL progression.

### LOW EXPRESSION OF ASNS AS A MARKER OF SENSITIVITY TO ASNASE IN OTHER CANCERS

The assumption that low ASNS expression represents the major hallmark for sensitivity to ASNase prompted the research of other Asn-auxotroph cancers. As far as hematological cancers are concerned, ASNase has been proposed for the therapy of several conditions [see for review (33)]. Several years ago it was demonstrated that the M5 subgroup of acute myeloid leukemias (AML) is characterized by low ASNS expression and, hence, high sensitivity to ASNase (34). A more complete attempt to categorize AML subgroups on the basis of ASNase sensitivity indicated that M1 and M0 were the most sensitive, while M3 and M7 were poorly sensitive and M4-M5 were confirmed to have a moderate sensitivity (35). Although no correlation was made between ASNase sensitivity and ASNS protein expression in that paper, a good response to therapy associated with low ASNS mRNA expression was later reported, at least for M0 (36). More recently, since chromosome 7 monosomy (-7) is frequently detected in adverse-risk AML and therapy-related myeloid neoplasms in children, the hypothesis that this aberration correlates with sensitivity to ASNase was investigated (37). Monosomic cells were indeed more sensitive to ASNase and exhibited significantly lowered ASNS mRNA and protein expression (37). However, the correlation between ASNase-sensitivity and ASNS expression of AML was not considered strong (33), consistently with the importance attributed to Gln, rather than Asn depletion, in the mechanism of the cytotoxic effects of bacterial ASNases on AML cells (38, 39).

ASNase has greatly improved the therapy of Natural Killer (NK)/T cell lymphoma, an aggressive lymphoid tumor associated with a poor prognosis (40, 41). Using a panel of 7 lymphoma cell lines and a retrospective analysis of patient samples, Li et al. demonstrated that ASNS expression inversely correlated with sensitivity to ASNase and positive clinical outcome (42). These data have been substantially confirmed in a more recent study (43).

ASNS has been investigated in solid tumors for many years, and the emerging picture is quite complex. In more than 50% of sporadic pancreatic ductal adenocarcinomas (PDAC), ASNS protein expression is very low (44), an observation that should be considered in light of the fact that normal exocrine pancreas has the highest basal ASNS expression of any tissue in the body (8, 45). Moreover, pancreatitis is one of the primary clinical complications exhibited by ALL patients treated with ASNase (46), suggesting that pancreatic exocrine cells are particularly sensitive to Asn depletion. Consistent with low ASNS expression, PDAC cell lines were sensitive to ASNase, and the most sensitive expressed the lowest levels of ASNS (44). Furthermore, in pancreatic cancer cells, ASNS induction is caused by glucose deprivation and is associated with increased resistance to cisplatin-induced apoptosis (47). Collectively, these data support the possible exploitation of ASNase in selected cases of low-ASNS PDAC, although they must also be interpreted in the context of the complex metabolic peculiarities of pancreatic cancer (48). Recently, it has been demonstrated that ASNS hypermethylation leads to the lack of ASNS protein expression in gastric and liver cancer cells, making them sensitive to E. coli ASNase treatment both in vitro and in vivo (49). Thus, patients could be stratified for ASNase trials on the basis of ASNS protein expression level.

### HIGH ASNS EXPRESSION IN CANCER: A PRO-TUMOR ENZYME?

In several models of human solid cancers ASNS expression has been found to be positively correlated with tumor growth and, in some cases, chemo-resistance, especially if cis-platinum-derived drugs are involved (47, 50, 51). Interestingly, a recent report indicates that ASNS may be itself an additional target of platinum(II) compounds, and that these drugs cause a decrease in cell Asn as a consequence of ASNS inhibition (52). However, independently of effects on chemoresistance, ASNS overexpression has been linked to unfavorable clinical outcomes in multiple cancers (53). For example, as a possible extension of ASNase exploitation to solid tumors, Lorenzi et al. reported that the sensitivity to ASNase of cell lines derived from human ovarian carcinomas was inversely correlated with ASNS mRNA abundance (54) and, even more strongly, with ASNS protein levels (55). In other cases, direct genetic targeting of ASNS expression has been used to document enzyme effects on cancer cells. For instance, ASNS silencing lowers proliferation of human gastric cancer cells either in vitro or in vivo and synergizes the cytotoxicity of cisplatin (50). ASNS mRNA is significantly overexpressed in human gastric cancer samples compared with normal gastric tissue, and its expression inversely correlates with patient survival (50). ASNS knockdown hinders growth of melanoma cells and epidermoid carcinoma cells, inducing cell cycle, down-regulation of CDK4, CDK6, and Cyclin D1, and induction of p21WAF (56). With a similar approach, Xu et al. reported a role for ASNS in the growth and colony formation ability of lung cancer (NSCLC) cells and demonstrated higher ASNS expression in lung cancer tissues than in normal lung tissue (57). A pharmacological approach was instead adopted by Hettmer et al. who demonstrated that an adenylated ASNS inhibitor inhibits the growth of murine and human sarcoma cell lines (58). In the same contribution, ASNS silencing lowered the portion of cells in S phase, an effect rescued by exogenous asparagine, and ASNS expression was found in a substantial portion of human rhabdomyosarcomas (over 70%) and in a smaller, but significant percentage of human leiomyosarcomas (more than 40%).

In breast cancer cells, ASNS is a target of IGF1/IGF2 dependent anabolic signaling (59), and, consistently, ASNS silencing depressed cell proliferation in two distinct cell lines, one of which derives from a triple negative tumor (60). Moreover, ASNS expression and Asn availability have been found to be strongly correlated with the metastatic behavior of breast cancer (61). Interestingly, in this study ASNS knock down did not affect the growth of the primary tumor but its metastatic behavior, which was significantly promoted, together with epithelial-tomesenchymal transition, by enforced ASNS expression (61). From xenografts of the triple negative breast cancer cell line MDA-MB-231 Ameri et al. (62) obtained circulating tumor cells (CTC), which exhibit an increased capability of inducing ATF3 and ATF4 under hypoxic conditions, higher ASNS expression and a more aggressive phenotype in vitro and in vivo. Mining publicly available datasets, Lin et al. demonstrated that, among the breast cancer subtypes, triple negative has the highest ASNS protein expression (53).

As far as prostate cancer is concerned, data on possible derangements of ASNS expression in cells derived from this tumor have been known since several years. ASNS was included in a group of over-expressed genes in prostate cancer cells adapted to grow in suspension (63). More recently, ASNS mRNA overexpression, due to increased copy number of the gene, was detected in surgical specimens of castration-resistant prostate cancer and correlated with ASNS protein abundance (64). Moreover, ASNS protein expression was associated with progression to a therapy-resistant disease state (64). Interestingly, the effects of ASNase on the PC3 prostate cancer cell line and ASNS induction have been used to validate a detection system for measuring restrictive amino acids in tumors based on ribosome profiling (diricore, a procedure for **di**fferential **ri**bosome measurements of **co**don **re**ading (65).

Somewhat contradictory findings have been obtained for the role of ASNS in human hepatocellular carcinoma (HCC). Indeed, although ASNS was overexpressed in HCC, low expression has been found to be a negative outcome marker, at least in terms of overall survival, and experiments with HCC cell lines indicated that ASNS hinders cell proliferation, migration, and tumorigenicity (66). On the contrary, Li and Dong have reported that ASNS levels, along with those of the ER stress-related transcription factor ATF6, are lower in HCC than in either control subjects or patients affected by chronic hepatitis B (67). While the reasons for the discrepancy between these two studies are unclear, it should be noted that Zhang et al. studied ASNS protein expression (66), whereas only mRNA was measured by Li and Dong (67). As noted above for ALL and ovarian cancer, there can be a lack of correlation between ASNS mRNA and protein expression. Interestingly, Li and Dong discovered an ASNS polymorphism (rs34050735), corresponding to the 5' UTR region of the mRNA, that was significantly associated with HCC (67).

In colorectal cancer ASNS expression may also have protumor or anti-tumor roles. ASNS has been found up-regulated in several human cell lines and clinical specimens derived from colon carcinoma with mutated KRAS (68). In particular, in a series of 93 patients, ASNS protein was high in over 70% of the KRAS-mutated cases but only in 30% of those with wild-type KRAS. ASNS expression was induced by KRAS-activated signaling, in particular through the PI3K-AKTmTOR pathway, and repressed upon KRAS-silencing. Moreover, ASNS knockdown in vivo suppressed the growth of KRASmutant colon cancers, suggesting a tumor-favoring role of the enzyme in these cancers (68). However, the situation in vivo may be more complex. Indeed, Lin et al. have reported that low ASNS expression could constitute a negative prognostic factor, using both transcriptional data from public databases and immunocytochemical analysis of a cohort of 172 patients. In particular, ASNS low expression was significantly associated with advanced post-treatment tumor, nodal status, inferior tumor regression grade, shorter local recurrence-free survival, metastasis-free survival and disease-specific survival, and was predictive of worse outcomes and poor therapeutic response to neo-adjuvant therapy (69). It is tempting to attribute these discrepancies to differences in the mutational status of the tumors but further data are needed to confirm this hypothesis.

### ASNS IN CANCER: BEYOND ASN SYNTHESIS?

A direct link between the effects of ASNS expression on cancer cells and Asn production has been recently demonstrated. Looking for regulators of the metastatic behavior in human colon cancer, Duquet et al. (70) identified the sex-determining region Y (SRY)-box, member 12 (SOX12), as a suppressor of metastatic behavior of HT29 xenografts. However, in a more recent paper, Du et al. (71) demonstrate that in two independent, large colorectal cancer cohorts HIF-1α-mediated SOX12 overexpression is not only associated with a metastatic behavior, but also with a poor prognosis. Moreover, it also enhanced cell proliferation in vitro. Investigating the underlying mechanisms, they discovered that Asn synthesis was greatly favored by SOX12 through the coordinated induction of glutaminase, glutamic oxaloacetic transaminase 2, and ASNS. These observations were confirmed in samples from patients. Consistent with the observations in the patients, downregulation or overexpression of the three enzymes had opposite effects on cell proliferation and metastasis development. The role of Asn in these effects was confirmed by the inhibition of tumor growth and metastasis by ASNase (71).

However, although it is tempting to attribute ASNS effects on cancer growth to the Asn synthesizing activity of ASNS, its role may not be simply ensuring Asn availability for protein synthesis. In this case, very small amounts of Asn would be sufficient for cell viability and growth. Moreover, increased ASNS activity may not produce large effects on the intracellular levels of Asn, since Asn exerts a product inhibition on the enzyme acting on the recognition site for glutamine.

In fact, Asn could have additional roles, as recently suggested by Krall et al. (72), who demonstrated that Asn, either produced by the cell through ASNS activity or imported from the medium, is used as an exchange factor to promote entry and consumption of other amino acids, such as serine, arginine and histidine, and consequently, activate mTORC1 activity and protein synthesis. The transport routes responsible for the exchange were not identified, although the authors suggest that the ubiquitous exchange transporter for neutral amino acids LAT1 may be involved (72). However, earlier characterization work on LAT transporters would instead suggest that LAT2, rather than LAT1, mediates the efflux of amino acids with amido-side chain, such as Gln and Asn (73, 74). Moreover, as a determinant of serine uptake, Asn may modulate both serine metabolism and nucleotide synthesis (72).

Recent results would indicate that the relationships between ASNS, Asn, and mTORC1 activity may be more complex than envisaged. Indeed, ASNS silencing in melanoma and colon carcinoma cells causes the activation of the MAPK cascade and the activation of mTORC1 that, in turn, potentiates ATF4 dependent ASNS induction (11). Under the conditions adopted by Pathria et al., intracellular Asn is lowered by 30% by ASNS silencing, but the hypothesis that this decrease accounts for MAPK and mTORC1 activation was not directly verified, leaving open the question if these effects are due to changes in intracellular Asn or to some other ASNS-dependent mechanisms. However, these important contributions provide insight into how Asn can influence protein synthesis and cell viability well-beyond its role of proteinogenic amino acid, explaining why, in some instances, it can compensate for Gln starvation (75).

ASNS may also have other roles in cancer cell metabolism, possibly related to its participation in the response to cell stress. In NSCLC, for example, KRAS promotes ATF4 pathway activation during nutrient depletion, promoting amino acid uptake, and Asn biosynthesis (76). In the same cell model, ASNS contributes to apoptotic suppression, protein biosynthesis, and mTORC1 activation, while ASNS repression due to the inhibition of AKT had an anti-tumor effect, which is enhanced by the depletion of extracellular Asn (76). KRAS-mediated overexpression of ASNS has been also described in colon cancer in the context of adaptation to nutritional stress upon Gln starvation (68). That study found that mutated KRAS caused Asp decrease and Asn increase and that these changes were associated, both in cancer cell lines and primary tumors, with increased ASNS expression through the PI3K-AKT-mTOR pathway. These cells were resistant to Gln depletion, a behavior suppressed by ASNS knockdown but rescued if ASNS-silenced cells are incubated in Asn-supplemented medium. Moreover, both ASNS knock-down and the combined treatment with rapamycin and ASNase inhibited the growth of KRAS-mutant colon cancer xenografts in vivo (68). The relationship between KRAS mutations and ASNS expression may underlie a specific role of Asn in autophagy regulation. It is known that the knockout of Atg5, a gene needed for the autophagic response, significantly extends the survival of a murine model of salivary duct carcinoma (SDC) driven by oncogenic KRASG12V, while it causes a specific Asn deficiency and a compensatory ASNS overexpression (53). Consistently, autophagy or ASNS inhibition reduced KRAS-driven tumor cell proliferation, migration, and invasion, all effects rescued by Asn supplementation. Finally, these observations were reflected in human cancer-derived data, linking ASNS expression and malignancy (53).

A role for ASNS in the cell response to nutritional stress has been also shown by Ye et al. (77), considering that its master activator ATF4 is also overexpressed in human tumors. Overexpression of ASNS or Asn supplementation, but not of other non-essential amino acids, counteracts the proliferative block and cytotoxicity due to ATF4 silencing in human fibrosarcoma and colorectal adenocarcinoma cells. The knockdown of ATF4, or the suppression of its induction by GCN2 silencing, inhibited tumor growth in vivo (77). These results have been further extended by Tameire et al. (78), demonstrating that MYC upregulates ATF4 through GCN2 activation and that, subsequently, ATF4 induces several genes that are also MYC targets, many of which involved in amino acid transport (such as SLC1A5) and metabolism (such as ASNS). This group of genes also includes 4E-BP1, leading the authors to hypothesize that, through ATF4-mediated gene induction, tumor cells couple enhanced translation rates with survival. In several human tumors, such as diffuse large B-cell lymphoma, colorectal cancer, breast cancer and sarcoma, 4E-BP1 levels were positively correlated with ATF4-target genes, including ASNS (78). These results potentially link ASNS induction and the successful response to oncogene-dependent proteotoxic stress and hence cancer cell survival, although the precise role played by ASNS in these complex mechanisms awaits further investigation.

### ASNS EXPRESSION AS A METABOLIC VULNERABILITY

Together with Asn auxotrophy, associated with ASNS silencing, arginine auxotrophy, which depends on absent expression of argininosuccinate synthase 1 (ASS1), represents another, widely investigated metabolic vulnerability in human cancers. In arginine-auxotroph human breast cancer cell lines, arginine depletion induces ASNS, provoking a depletion of Asp that hinders malate-aspartate shuttle and promotes cell death (79). Thus, in this particular model, ASNS induction, rather than constituting a pro-survival mechanism, would promote cytotoxicity through Asp depletion.

The importance of Asp metabolism in cancer has been increasingly recognized in the last few years (80–82). Since human cells cannot use Asn as a source of Asp, due to lack of sizable expression of enzymes with asparaginase activity, the metabolic relationship between the two amino acids is a oneway pathway, where Asp can be used a Asn source, while the reverse is not possible. Interestingly, if the expression of guinea pig ASNase is forced in human cancer cells, Asn uptake can fuel the intracellular pool of Asp, and cell growth is stimulated (83), providing a proof-of-principle demonstration of the importance of an adequate Asp availability for fast cell proliferation. Membrane transport can limit cell availability of Asp, which, at the levels present in human plasma, relies on the activity of high-affinity, sodium-dependent EAAT transporters (84, 85). A member of the family, EAAT1, coded by SLC1A3, has been identified as an important contributor of resistance to ASNase in several lines of prostate cancer cells (27). In one of these models, although ASNS is heavily induced upon ASNase treatment, cell death is not prevented if EAAT1 is pharmacologically inhibited (27). Interestingly, prostate cancer cell lines endowed with low expression of EAAT1 exhibit sizable levels of other EAATs, such the ubiquitous EAAT3, which is regulated at transcriptional level under various stress conditions (85).

Another example of possible ASNS-mediated vulnerability comes from the studies of Wong et al. on KRAS-mutated colorectal cancers. SLC25A22, which encodes a mitochondrial glutamate transporter, is one of the genes up-regulated in these tumors. Increased SLC25A22 protein was observed in colorectal cancer tissues and was associated with shorter survival, while transporter knock-down hindered cancer cell proliferation, migration, invasion in vitro and tumor formation and metastasis in vivo. The biochemical alteration attributable to SLC25A22 knockdown and accounting for the anti-proliferative effects is the inhibition of Asp biosynthesis and the consequent depletion of oxaloacetate leading to hampered regeneration of NAD<sup>+</sup> and NADP+, glycolysis hindrance and energetic crisis. In this context, the inhibition of ASNS-mediated Asn synthesis would be another effect of Asp depletion, specifically leading to hindered cell migration (86). One would wonder what are the links between the two KRAS-mediated effects on Asn-Asp metabolism. Although SLC25A22 would be permissive for Asp synthesis, ASNS induction would promote its consumption, suggesting that a dysregulated ASNS expression would be, in fact, a menace for the energetic equilibrium of the cancer cell.

### DISCUSSION

The examples discussed in the last paragraph indicate that, in some cancers, low ASNS expression may be advantageous. However, most of the epidemiological and experimental evidence gathered thus far suggest a pro-cancer role of the enzyme, pointing to a metabolic advantage for high-ASNS cancer cells. Thus, both low- and high-ASNS expression may imply metabolic advantages in particular cancer models (**Figure 1**). It should be remarked that either situation also implies some potential metabolic vulnerabilities, such as Asp depletion, for high-ASNS tumors, and Asn auxotrophy, for low-ASNS cancers.

If high ASNS expression really confers marked metabolic advantages, one wonders what is the significance of ASNS silencing in the majority of ALL blasts and in the other examples of Asn-auxotroph tumors, discussed above. For these cancers, the maintenance of the intracellular pool of Asp seems more important than ensuring an intracellular source of Asn. In these cells, blocking the expression of ASNS would indeed leave most cell Asp available to other metabolic pathways, such as nucleotide and non-essential amino acid synthesis or energy production (**Figure 2**). On the other hand, Asn auxotrophy not only has the obvious consequence of an increased sensitivity to Asn depletion and, hence, to ASNase treatment, but also entails a strict dependence of the cancer cells on extracellular sources of the amino acid even under normal growth conditions.

synthesis and cell growth by activating the mammalian target of rapamycin complex 1 (mTORC1) through the influx of essential amino acids mediated by exchange through a LAT transporter (72), tentatively identified as LAT2. Other transporters have been omitted for clarity. However, other mechanisms, such as direct effects of Asn or Asp on mTORC1, should not be excluded but the information available (11, 68, 72) does not allow generalizable conclusions. See text for discussion.

Since Asn plasma levels are much lower than those of Gln, and the transport systems for neutral amino acids are usually endowed with fairly high Km values, it is expected that Asn auxotroph tumors establish close relationships with their microenvironment to exploit neighboring cells as an efficient Asn source. Actually, metabolic support to ASNase-treated ALL blasts by ASNS-expressing mesenchymal stromal cells has been reported (87), although mechanisms underlying the putative Asn fluxes have not been investigated.

At variance with the transport of other amino acids, in particular Gln, which shares many structural similarities with Asn and is its metabolic precursor, the characteristics of Asn transport have not been extensively studied in Asn-auxotroph cancer cells. It is known that Gln, Asn and His are substrates of the so called "N system" transporters (88), such as SNAT3, SNAT5 and SNAT7 (89, 90), but little information is available on the expression of these transporters in cancer tissues, and no attempt has been made thus far to correlate their expression with that of ASNS. Also other transporters of the SLC38 family, such as the System A carriers SNAT1 and SNAT2 (90), and the product of SLC1A5 (ASCT2) (91) accept Asn as a substrate. However, lack of a comprehensive knowledge of Asn transporters in cancer constitutes an important gap given that Asn membrane fluxes are obviously essential for the survival of Asn-auxotroph cancer cells. The definition of transport mechanisms involved in Asn transmembrane fluxes would be highly valuable also for the biology of high-ASNS cancers, which are thought to export sizable amounts of the amino acid into the extracellular medium (72). In these tumors, Asn may work as a modulator of the behavior of normal cells within the cancer microenvironment, as recently suggested for endothelial cells (92).

The results recounted in this contribution indicate that the role played by ASNS may be cancer-specific and should be assessed on an individual basis. Therefore, to better define the role of ASNS expression and activity in human cancers, specific and potent inhibitors would be extremely important and have been actively searched for many years (93, 94). Many classes of compounds have been proposed thus far (29), in some cases with high potency (95) and promising results in vitro (96, 97), but no specific ASNS inhibitor is yet in clinical experimentation or even commercially available. As a consequence, experimental ASNS inhibition still relies on genetic manipulation. However, most recently, Zhu et al. (7) have described a slow-onset inhibitor, which binds to a negatively charged cluster of side chains in the synthetase domain of human ASNS with nanomolar affinity and a good specificity in vitro and may be the basis for novel anticancer compounds targeting ASNS.

### AUTHOR CONTRIBUTIONS

MC, GT, MB, MK, and OB drafted the manuscript and approved it.

### REFERENCES


### FUNDING

This work has been funded by the University of Parma (OB) and by The National Institutes of Health, The National Cancer Institute, Grant No.: CA203565 (MK). MC was supported by a fellowship of the Associazione Italiana per la Ricerca sul Cancro (AIRC No.19272).


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

Copyright © 2020 Chiu, Taurino, Bianchi, Kilberg and Bussolati. 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.

# Is Lactate an Oncometabolite? Evidence Supporting a Role for Lactate in the Regulation of Transcriptional Activity of Cancer-Related Genes in MCF7 Breast Cancer Cells

Iñigo San-Millán1,2 \*, Colleen G. Julian<sup>3</sup> , Christopher Matarazzo<sup>3</sup> , Janel Martinez <sup>1</sup> and George A. Brooks <sup>4</sup>

<sup>1</sup> Department of Medicine, Division of Endocrinology, Metabolism and Diabetes, University of Colorado School of Medicine, Aurora, CO, United States, <sup>2</sup> Department of Human Physiology and Nutrition, University of Colorado, Colorado Springs, CO, United States, <sup>3</sup> Department of Medicine, University of Colorado School of Medicine, Aurora, CO, United States, <sup>4</sup> Department of Integrative Biology, University of California, Berkeley, Berkeley, CA, United States

#### Edited by:

Paolo E. Porporato, University of Turin, Italy

#### Reviewed by:

Pierre Sonveaux, Catholic University of Louvain, Belgium Fatima Baltazar, University of Minho, Portugal

\*Correspondence: Iñigo San-Millán inigo.sanmillan@cuanschutz.edu

#### Specialty section:

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

Received: 26 October 2019 Accepted: 19 December 2019 Published: 14 January 2020

#### Citation:

San-Millán I, Julian CG, Matarazzo C, Martinez J and Brooks GA (2020) Is Lactate an Oncometabolite? Evidence Supporting a Role for Lactate in the Regulation of Transcriptional Activity of Cancer-Related Genes in MCF7 Breast Cancer Cells. Front. Oncol. 9:1536. doi: 10.3389/fonc.2019.01536 Lactate is a ubiquitous molecule in cancer. In this exploratory study, our aim was to test the hypothesis that lactate could function as an oncometabolite by evaluating whether lactate exposure modifies the expression of oncogenes, or genes encoding transcription factors, cell division, and cell proliferation in MCF7 cells, a human breast cancer cell line. Gene transcription was compared between MCF7 cells incubated in (a) glucose/glutamine-free media (control), (b) glucose-containing media to stimulate endogenous lactate production (replicating some of the original Warburg studies), and (c) glucose-containing media supplemented with L-lactate (10 and 20 mM). We found that both endogenous, glucose-derived lactate and exogenous, lactate supplementation significantly affected the transcription of key oncogenes (MYC, RAS, and PI3KCA), transcription factors (HIF1A and E2F1), tumor suppressors (BRCA1, BRCA2) as well as cell cycle and proliferation genes involved in breast cancer (AKT1, ATM, CCND1, CDK4, CDKN1A, CDK2B) (0.001 < p < 0.05 for all genes). Our findings support the hypothesis that lactate acts as an oncometabolite in MCF7 cells. Further research is necessary on other cell lines and biopsy cultures to show generality of the findings and reveal the mechanisms by which dysregulated lactate metabolism could act as an oncometabolite in carcinogenesis.

Keywords: lactate, cancer, oncogenes, transcription factors, cell cycle genes

## INTRODUCTION

A role of lactate in cancer metabolism was first described almost a century ago when Otto Warburg and associates discovered that cancer cells were not only characterized by accelerated glucose consumption, but also by a marked increase in lactate production (1). They exposed tumor cells to amino acids, fatty acids, and glucose, expecting the highest rate of respiration in glucose-exposed

**18**

cancer cells. Contrary to expectations, glucose brought respiration to a standstill. "In trying to discover why this happened, it was found that lactic acid appeared in the Ringer's solution, produced by glycolysis, and that this inhibited the respiration," asserted Warburg (2). Warburg also found that arterial glucose uptake in tumor cells was about 47–70%, compared to 2–18% in normal tissues, and that tumor cells converted 66% of glucose uptake to lactate (3). Based on these observations, Warburg concluded that increased glycolytic activity was integral to carcinogenesis, a phenomenon subsequently termed the "Warburg Effect" (4).

In the last decade there has been a "renaissance" of cancer metabolism and the knowledge acquired has been significant. It is now known that carcinogenesis involves complex metabolic processes characterized by tumor heterogenicity, involving different metabolic activities and regulations necessary for tumor growth, survival and progression (5). Increasing body of literature implicates the involvement of lactate for carcinogenesis. Sonveaux et al. showed that lactate is a major fuel for biomass and anabolic necessities of cancer cells (6, 7).

Given the unexplained causes and consequences of the Warburg effect in cancer, recently, we articulated the "lactagenesis hypothesis" (8). According to our hypothesis, the predominant role of lactate in cancer is not only for fuel or cancer biomass purposes but also for carcinogenic signaling properties. Lactate is involved in the main biological processes that are known to drive or sustain carcinogenesis, specifically: angiogenesis, immune escape, cell migration, metastasis, and self-sufficient metabolism (8).

While lactate has been the subject of intense investigation since at least the nineteenth century, until recently it was believed that lactate was solely a byproduct of oxygen-limited, anaerobic metabolism. However, in the mid 1980's George Brooks proposed the "Lactate Shuttle Hypothesis" based on results of isotope tracer studies in rodents and humans (9–13). His studies showed for the first time that lactate production and exchange could also occur under fully aerobic conditions debunking the belief that lactate was a "waste" product of anaerobic glucose metabolism (14). Specifically to cancer cells, it is estimated that in cancer cells, lactate production accounts ∼70% from aerobic glycolysis (15).

Furthermore, it is now recognized that lactate is a major energy source (16–19), the major gluconeogenic precursor (19), a signaling molecule and a "lactormone" (13) that also influences gene expression (20–23). Exogenous L-lactate exposure, for instance, has been reported to upregulate the transcriptional activity of 673 genes in L6 cells (20). Hussien and Brooks later showed that both lactate dehydrogenase A (LDHA) and LDHB as well as monocarboxylate transporters (MCTs) were expressed in breast cancer cells, including MCF7 (24).

Most recently, Zhang et al. (25) have shown that "lactylation" of histone lysine residues serves as an epigenetic modification that directly stimulates gene transcription from chromatin in human and mouse cells. They also showed that lysine lactylation (Kla) levels increased in a dose-dependent fashion in response to exogenous L-lactate and that endogenous production of lactate is a key determinant of histone Kla levels (25). Furthermore, Verma's group has recently and elegantly demonstrated that tumor-mediated lactate can elicit epigenomic reprograming of cancer-associated fibroblasts from pancreatic ductal adenocarcinoma (26).

Moreover, there has been growing interest in blocking lactate transport and exchange among and within cancer cells to decrease tumor growth and carcinogenesis (6, 23, 27–29).

Renewed interest in understanding the causes and consequences of the Warburg Effect has shown that lactate can be produced from glutaminolysis. That observation is of consequence because glutaminolysis is considered a hallmark of cancer metabolism (30). It has been known since the 1970's that glutamine is a major energy source for mitochondria in HeLa cells (31) as well as biomass precursor for the proliferation of cancer cells (32). In pediatric glioblastoma cells (SF-188), MYC overexpresses glutaminolysis to elicit a mitochondrial metabolic reprograming favoring glutamine for energy purposes (33). Beyond purposes of bioenergetics and biomass, glutaminolysis can also be a major source of lactate in cancer (34). During high rates of glutaminolysis (a typical characteristic in many cancers) oxidative phosphorylation of glucose is decreased while glutamine fermentation to lactate is increased (34). This concept is important as it could explain why therapies targeting glycolysis may not be very efficient if lactate is derived from glutamine. Furthermore, recent research studies have focused on blocking glutamine metabolism in cancer. In particular one recent study in mouse cancer cells showed that a glutamine agonist JHU082 both shut down oxidative phosphorylation and glycolysis as well as enhanced oxidative phosphorylation and immune response in T-Cells (35).

In breast cancer, there is an average of about 33 somatic mutations (36). Within the multiple somatic mutations in different cancers, there are a few key driver genes that confer a selective growth advantage to "drive" tumorigenesis (36, 37). The driver genes involved in selective growth advantage are referred to as "mut-driver" or "epi-driver" genes (36). Although the epi-driver genes are not yet well-identified or understood, 125 mut-driver genes involved in multiple tumors have been identified (36).

In the present study, we sought to determine whether endogenously produced lactate or exogenously added L-lactate (Sodium L-lactate) exposure could act as an oncometabolite affecting the transcription of key driver genes recognized to play a central role in breast cancer (specifically in MCF7 cells).

### MATERIALS AND METHODS

### Overview

We tested our lactagenesis hypothesis by exposing MCF7 cells to glucose, which resulted in endogenously produced lactate, or added, exogenous sodium L-lactate and observing the transcription of key driver genes breast cancer, some of which are involved in the majority of cancers (38–44). We took this approach because of the orchestrated action of oncogenes, tumor suppression genes, transcription factors and cell cycle genes that activate an array of pathways leading to increased cell proliferation and the metabolic reprograming of cancer cells from oxidative phosphorylation (OXPHOS) to glycolysis and lactate production.

### Cell Culture Experiments

For both study objectives MCF7 cells (ATCC) were maintained in Dulbecco Modified Eagle Medium (DMEM) with 10% Fetal Bovine Serum (FBS) and Penicillin 100 units/mL-Streptomycin 100 ug/mL (Invitrogen) [DMEM 10% FBS-Pen-Strep] and cultured in a 5% CO<sup>2</sup> atmosphere at 37◦C. Briefly, MCF7 cells (1 × 10<sup>6</sup> ) were plated in cell culture dishes using DMEM 10% FBS-PenStrep. Once the cells reached 80% confluence, the cells were incubated in DMEM (high glucose; 4,500 mg/L) containing 0, 10, or 20 mM sodium, L-lactate (Sigma) with 10% NuSerum (BD Biosciences), and Pen-Strep. Controls were MCF7 cells incubated in DMEM without glucose, glutamine, or lactate. Cells were maintained for either 6 or 48 h before being harvested for gene expression profiling. Cell pellets were stored in RLT buffer (Qiagen) with beta mercaptoethanol added, and stored at −80◦C. The concentration of lactate and glucose in the cell culture media at the time of harvest was determined using the L-lactate Assay Kit (AbCam) or Glucose Assay Kit (Cayman Chemical), respectively, following manufacturer specifications.

### Cell Line Authentication

MCF7 (ATCC <sup>R</sup> HTB22TM) cells were authenticated by ATCC using morphology, karyotyping, and PCR based approaches to confirm identity of human cells and to rule out both intra- and inter-species contamination. These included an assay to detect species specific variants of the cytochrome C oxidase I gene (COI), and short tandem repeat (STR) profiling. The cell line was obtained in March of 2018 and the last test was done in September of 2018.

### RNA Isolation and Assessment of MCF7 Gene Expression

Total RNA was extracted using the AllPrep DNA/RNA Mini Kit (Qiagen) and the quantity assessed using Nanodrop spectrophotometry. RNA was reverse transcribed using a cDNA conversion kit. The cDNA in combination with RT<sup>2</sup> SYBR <sup>R</sup> Green qPCR Mastermix was used on a Custom RT2 Profiler PCR Array (Qiagen). RT2 Profiler PCR Arrays were used to compare gene expression profiles between MCF7 cells cultured in glucosefree media, glucose-supplemented media (leading to Warburg Effect and lactate production) and glucose-supplemented media with exogenous lactate added. Targeted genes typical of MCF-7 cells included on the array were: oncogenes (MYC, NRAS, and PIK3CA), transcription factors (HIF1A and E2F1), tumor suppressor genes (BRCA1 and BRCA2) as well as cell cycle genes, and proliferation genes (AKT1, ATM, CCND1, CDK4, CDKN1A, CDK2b, and MIF) (**Table 1**). Genes were categorized according to NIH Genetics Home Reference (www.ghr.nlm.nih. gov). Each condition was run in triplicate, and from each of the three biological replicates duplicate samples were run on the expression array to ensure accuracy and reproducibility.

### Statistical Analyses

Lactate and glucose concentrations were compared between groups by analysis of variance (ANOVA) in GraphPad Prism (v. 5.01; GraphPad Software).

TABLE 1 | Fold changes in expression for cancer-related genes between glucose-starved MCF7 cells vs. MCF7 cells exposed to 0, 10, or 20 mM lactate for 6 or 48 h.


Results are shown for genes showing a 1.5-fold or greater change in expression and a p-value ≤0.05. <sup>+</sup>Genes classified according to NIH Genetics Home Reference (www.ghr.nlm.nih.gov).

Cycle threshold (Ct) values, the number of PCR cycles required for florescent signal to exceed background levels, are inversely proportional to the amount of target nucleic acid present in the sample. C<sup>t</sup> values were exported and then uploaded onto GeneGlobe, Qiagen's data analysis web portal (https://geneglobe.qiagen.com). Within the GeneGlobe platform, 1C<sup>t</sup> values were calculated by subtracting the C<sup>t</sup> value for the reference gene (GAPDH) from the C<sup>t</sup> value for target genes for each sample. For all experiments, controls were MCF7 cells cultured in glucose/glutamine-free media with 0 mM lactate for 6 h to eliminate the impact of glucosederived endogenous lactate production (Warburg Effect) on gene expression. Statistical tests were performed on raw 1C<sup>t</sup> values for each group. Fold change was then calculated using 2−11CT formula. Gene expression differences between experimental groups and controls are expressed as foldregulation. Criteria for reporting gene expression differences include: fold-regulation of ≥2.0 with a p-value of ≤0.05. Lactate and glucose concentrations were compared between groups by analysis of variance (ANOVA) in GraphPad Prism (v. 5.01; GraphPad Software).

### RESULTS

### Glucose-Derived Lactate Production

As expected, glucose incubation resulted in high concentrations of lactate (28.8 ± 2.9 and 21 ± 6.8, mM (p < 0.001) both 6 and 48 h, respectively post-incubation in glucose (**Figure 1**). As no lactate was added in this experiment in glutamine-free media, lactate accumulation was considered to be glucose-derived as a result of the Warburg Effect.

### Glucose-Derived Endogenous Lactate (Warburg Effect) Upregulates the Transcriptional Activity of Oncogenes, Transcription Factors, Tumor Suppressor Factors as Well as Cell Cycle and Proliferation Genes

Compared to controls, the expression of three key oncogenes, MYC, RAS, and PIK3CA, was between 2- and 3.3-fold greater in MCF7 cells cultured for 48 h in glucose-containing media leading to lactate accumulation (p < 0.05) (**Table 1**, **Figure 2**). Endogenous lactate production also affected the expression of transcription factors known to be involved in MCF7 cancer cells. Specifically, compared to controls, MCF7 cells cultured in glucose-containing media for 48 h showed 2.9 and 2.3-fold increases in mRNA expression of transcription factors HIF1A and E2F1, respectively (p < 0.01) (**Table 1**, **Figure 2**). After 48 h exposure, the transcriptional activity of tumor suppressor factors BRCA1 and BCRA2 increased 3.4- to 4.9-fold, respectively (p < 0.01) (**Table 1**, **Figure 2**). Finally, transcriptional activities of cell cycle and proliferation genes (except for MIF) increased 2.1- to 8.1-fold (p < 0.05) (**Table 1**, **Figure 2**) after 48 h exposure between.

### Exogenous Lactate Exposure Upregulates the Transcriptional Activity of Oncogenes, Transcription Factors, Tumor Suppressor Factors as Well as Cell Cycle and Proliferation Genes 10 mM Lactate Exposures

Exposing MCF7 cells to 10 mM lactate for 6 h increased the expression of oncogenes MYC, NRAS, and PIK3CA between 3.6- and 7.8-fold (p < 0.05). After 48 h transcriptional activity was slightly weaker for these oncogenes (2.6 and 2.8, p < 0.05) (**Table 1**, **Figure 3A**). After 6 h, transcription factors, HIF1A and E2F1, were overexpressed by 4.4- and 3.4-fold, respectively (p < 0.01) (**Table 1**, **Figure 3A**). Expression was similar after the 48 h exposure for HIF1A 4.1-fold, p < 0.001), and slightly reduced for E2F1 (2.6-fold, p < 0.001) (**Table 1**, **Figure 3A**). After 6 h, transcriptional activity of tumor suppressor factors BRCA1 and BRCA2 was increased between 3.4- and 6.1-fold, respectively (p < 0.05). After 48 h exposure, transcriptional activities of BCRA1 and BCRA2 increased 4.3- to 4.9-fold, respectively (p < 0.001) (**Table 1**, **Figure 3A**). Compared to controls, the transcriptional activity of proliferation and cell cycle genes (except for ATM) was significantly greater after 6 h exposure to 10 mM lactate with values ranging from 2.6- to 7.5-fold (all, p ≤ 0.05), while a slightly weaker response was observed after the 48 h 10 mM lactate exposure with values (except MIF) ranging from 1.4- to 4-fold (all, p ≤ 0.05) (**Table 1**, **Figure 3A**).

### 20 mM Lactate Exposures

In contrast to a 10 mM lactate exposure, neither NRAS nor PIK3CA gene expressions were increased compared to controls in MCF7 cells, while MYC was increased by 6.3-fold (p < 0.05) when exposed to 20 mM for 6 h (**Table 1**, **Figure 3B**). However, 48 h exposure to 20 mM lactate induced modest increases in NRAS and PIK3CA gene expression (1.9- and 2.0 fold, respectively, p < 0.01), and a slight decrease in expression of MYC (2.8-fold, p < 0.01) (**Table 1**, **Figure 3B**).

Exposure to 20 mM lactate also upregulated the expression of several transcription factors after exposure of 6 or 48 h. At

6 h, gene expression values for HIF1A and E2F1 were 4.8- and 3.3-fold, respectively greater than control (p < 0.05) while at 48 h exposure transcriptional activity for HIF1A and E2F1 was slightly lower 2.9- to 1.7-fold, respectively (p < 0.05) (**Table 1**, **Figure 3B**). Transcriptional activity of tumor suppressor factors after 6 h exposures for BCRA1 and BCRA2 was increased by 3.7 and 5.6-fold, respectively (p < 0.001) and was slightly lower after the 48 exposure (2.4- to 3.3-fold, respectively, p < 0.001) (**Table 1**, **Figure 3B**).

After 6 h exposures, the transcriptional activity of all cell cycle genes except for ATM and CDK2b was overexpressed ranging from 1.5- to 5.6-fold (all, p ≤ 0.05) (**Table 1**, **Figure 3B**). After 48 h, we observed increased transcriptional activity in all cell cycle and proliferation genes studied ranging from 1.7- to 4.2-fold (p < 0.05). Forty-eight hours exposures shown weaker response than at 6 h, but still significant upregulation of cell cycle genes (**Table 1**, **Figure 3B**).

### DISCUSSION

Our findings demonstrate that in the MCF7 human breast cancer cell line, lactate alters the transcriptional activity of several key oncogenes as well as other driver genes involved in metabolic reprograming as well as the regulation of cell cycle and proliferation. In the aggregate, these observations are in line with our "lactagenesis hypothesis" (8) positing augmented lactate production for signaling carcinogenesis as one essential purpose of the Warburg Effect.

After both 6 and 48 h exposures there was a high presence of glucose-derived lactate in the cells incubated in glucose without added lactate (or glutamine), which replicated the Warburg studies (**Figure 1**). Previously, we have shown that lactate is oxidized in mitochondrial preparations from non-transformed tissues (14, 45, 46), and recently it has been confirmed that lactate is also oxidized by mitochondria of cancer cells (6, 24) purportedly for energetics (7, 24). Beyond lactate bioenergetics and biomass properties, our study suggests that glucose-derived lactate is sufficient to alter the transcriptional activity of key oncogenes, transcription factor genes, tumor suppressor genes as well as cell cycle, and proliferation genes, all of which are known to be involved in the development of MCF7 breast cancer cells (**Table 1**, **Figure 2**). The experiments, adding 10 and 20 mM of Lactate to MCF7 cells, augmented the transcriptional properties of lactate (**Table 1**, **Figures 3A,B**) which supports the hypothesis that lactate could be an oncogenic regulator, an oncometabolite.

Although lactate is the obligatory product of glycolysis under fully aerobic conditions (13), and our findings indicate that the addition of L-lactate to glucose (glutamine-free) media increases the transcriptional activity of the candidate genes studied herein, it is certainly possible that lactate and other metabolites involved in glycolysis, the pentose phosphate pathway or the TCA cycle could also influence the transcriptional activities of various genes in tumorigenesis. For example, Damiani et al. have observed that TCA intermediates that are not used for biomass purposes can be disposed via lactate production (34). Hence, while lactate is a metabolic intermediate, it has numerous downstream effects as known to occur via cell redox changes (14), allosteric binding

FIGURE 3 | Treatment with exogenous lactate (10 or 20 mM) upregulates the expression of several key genes in MCF7 cells. The fold upregulation of expression with 6 and 48 h exposure to 10 mM (A) or 20 mM lactate (B) relative to controls (MCF7 cells cultured for 6 h in glucose/glutamine-free media) (p's ≤ 0.05–0.001).

(47), metabolic reprograming (26), and lactylation (25). The downstream effects of lactate in cancer remain to be determined.

PIK3CA is considered to be the most mutated oncogene in breast cancer (48, 49). Furthermore, PIK3CA mutations are key drivers of breast cancer and its upregulation is associated with poor prognosis (50). Noteworthy, PIK3CA mutations are more frequent in estrogen receptor cancer cells, such as like MCF7 (51). In this study we demonstrate that lactate exposure to MCF7 cells is able to increase the transcriptional activity of PIK3CA between 2.2- and 4.3-fold (p < 0.05–0.001) (**Table 1**, **Figures 2**, **3A,B**). Furthermore, the PIK3/AKT/mTOR pathway is key and important intracellular pathway with major role regulating cell cycle, tumor growth, and proliferation (52, 53), one of the most activated signaling pathways in breast cancer (52) as well as required for survival of MCF7 (54).

In our study, we found that AKT1 transcriptional activity was upregulated between 2- and 3.35-fold. The significant increase in transcriptional activity elicited by lactate in both PIK3CA and AKT1 implicates lactate as a signaling oncometabolite of the key PIK3/AKT/mTOR pathway involved in the development of many cancers.

Another major oncogene, MYC, is known to have multiple roles in metabolic regulation including cellular adaptations following endurance exercise training (55), but is frequently overexpressed in breast cancer cells (56, 57), including MCF7 cells (58), and associated with poor prognosis (57). In our study, we found that MYC is highly expressed across all our experiments between 2.8- and 7.7-fold (p < 0.01) (**Table 1**, **Figures 2**, **3A,B**).

Hypoxia inducible factor 1 (HIF1α) as a major transcription factor in cancer (39, 42). HIF1α increases the transcription of genes regulating glucose transport and glycolytic enzymes (42), eliciting a metabolic reprogramming, leading to the Warburg Effect and lactate production. Furthermore, the overexpression of HIF1A, the gene encoding HIF1α, plays an important role in breast cancer tumor growth and metastasis as well as being related to aggressiveness and poor prognosis (59–61). In all of our lactate exposures experiments HIF1A transcriptional activity was overexpressed (between 2.9- and 4.8-fold, p < 0.001) (**Table 1**, **Figures 2**, **3A,B**), a finding that is not novel, as others have previously found similar results (22). Still, our present results corroborate those of others showing an important effect of lactate on transcriptional activities of this key transcription factor.

Our results showing an effect of lactate on expressions of MYC and HIF1A genes are consistent results of others showing an upregulation of the glycolytic pathway in cancer (62, 63). Hence, our results obtained on transcription of MYC and HIFA are supportive our lactagenesis hypothesis.

BRCA1 and BRCA2 are tumor suppressor genes typically mutated in breast cancer and highly connected with cancer aggressiveness and survival (64–66). BRCA1 contributes to the regulation of DNA repair, chromosomal remodeling, apoptosis, cell-cycle control, and transcriptional activity (67). While the loss or reduced expression of nuclear BRCA1 is prevalent in basallike breast cancers with negative estrogen, progesterone, and epidermal growth factor receptors (triple negative), its cytosolic expression is observed in estrogen-positive receptor breast cancers (68). In estrogen-positive receptor breast cancer cells (the characteristic of MCF7 cells), cytosolic BRCA1 expression is inversely related to survival (68). Furthermore, transcriptional activity of BRCA1 and BRCA2 genes has been observed in multiple breast cancers (including MCF-7 cells) (69–71). In our study, we found that lactate exposure is a potent regulator of their transcriptional activity with increases in mRNA expression between 3.3- and 6.1-fold (p < 0.001) (**Table 1**, **Figures 2**, **3A,B**).

Increased cell cycle and proliferation is a characteristic of cancer cells where all different phases in cell cycle are affected in cancer mainly by cyclin-dependent kinases (CDKs) (72). Among the significant results, we found that all CDKs were overexpressed by lactate exposure in a range from 2- to 6.7-fold (p < 0.01–0.05) (**Table 1**, **Figures 2**, **3A,B**). While the trigger of this genetic dysregulation hasn't been elucidated, our data show that most genes involved in the different phases of cell cycle are overexpressed by glucose-derived lactate alone as well as exogenous lactate; again, implicating lactate as a regulator of CDKs, thus shedding new possible light in cancer cell division and proliferation as well as therapeutics.

The traditional view of dysregulated downstream signaling pathways in cancer is hierarchically mediated by somatic mutations mainly due to dysregulation of oncogenes and tumor suppressors (73, 74). Our results show that, at least in MCF7 cells, lactate doesn't obey a hierarchical order of signaling, and also that in MCF7 cells, lactate signals multiple key steps essential in carcinogenesis, including cell proliferation.

It has been estimated that each gene driver mutation confers only a small selective growth advantage, about 0.4% increase in the difference between cell birth and death (75). However, this small difference over many years can result in significant production and accumulation of tumor cells leading to cancer (36). Likewise, we believe that a similar phenomenon can hold true for the constant transcriptional activity elicited by dysregulated lactate on the main key driver genes over the years.

Furthermore, Marticorena et al. (76) have recently shown that genetic mutation alone could not be a necessary element for cancer development as in their study, they found that both non-cancerous and cancerous esophagus cells shared cancerassociated genetic mutations.

Beyond the roles of oncogenes and tumor suppressor factors, others have speculated that Epi-drivers, like epigenetic changes affecting DNA and chromatin proteins could also be involved in carcinogenesis (36). As mentioned above, a remarkable new study by Zhang et al. has shown the regulation of gene expression by histone "lactylation," where both exogenous and endogenous lactate levels stimulate gene transcription from chromatin in human and mouse (25). From the extensive work of others, it is known that many mechanisms are also involved in histone acetylation in cancer. A classic example is the retinoblastoma pathway. Once hypophosphorylated at the beginning of G1 phase, retinoblastoma protein (pRb) is hyperphosphorilated at the end of G1 phase and the E2F1/pRB complex breaks off, allowing transcriptional activity of E2F1 at the end of G1 phase. E2F1 can then recruit histone acetylase for acetylation allowing chromatin transcription of genes to facilitate cell cycle moving passed the restriction (R) point into the G1/S transition and Sphase of cell cycle. In our study we show that lactate increases E2F1 mRNA transcription between 1.6- and 4.1-fold (p < 0.05– 0.001). Again, results support our lactagenesis hypothesis.

In concert, here we show that at least in MCF7 cells, lactate acts as an oncometabolite capable of regulating transcriptional activities of key oncogenes, transcription factors, tumor suppressors, and cell cycle genes involved in breast cancer.

An imperative question we pose now is what cell-specific properties, and mechanisms allow lactate to induce candidate cells toward a cancer phenotype. We have a plethora of knowledge and expertise about muscle (77, 78) and wholebody lactate metabolism during exercise (14) and as mentioned vide supra, we have known for decades that lactate is a major source of cellular energy, especially for mitochondria. In normal physiology, there is a dynamic, order of magnitude, range of muscle lactate production, and accumulation (14). However, as a tissue, muscle is resistant to carcinogenesis. In fact, rhabdomyosarcoma, historically thought to be a rare form of muscle cancer, has been recently proven to raise from endothelial progenitor cells following metabolic reprogramming and myogenic transdifferentiation, but not being originated from myocytes in the tissue itself (79). As well, from epidemiology, we know that regular exercise reduces the incidences of some forms of cancers in addition to other chronic diseases (80). Although lactate has been historically associated to exercise, it is noteworthy to differentiate between effects of transient increases in exercise-derived lactate and chronic lactate elevation in cancer. During and after exercise, lactate is ultimately cleared from muscle fibers with the clearance rate depending on mitochondrial function and cardiometabolic fitness level of the person. In contrast, in cancer, lactate is not rapidly cleared, and is highly concentrated in the tumor and its microenvironment; an effect of which could be to promote carcinogenesis.

### Study Limitations

We acknowledge that the present exploratory study has been conducted on a cancer cell line (MCF7). Hence, for further testing of the lactagenesis hypothesis will be important to reproduce this study with other cancer cell lines as well as with tumor biopsy cultures to show generality of the findings and reveal the mechanisms by which dysregulated lactate metabolism could act as an oncometabolite in carcinogenesis.

### REFERENCES


As well, we acknowledge that effects of treatment on gene transcription are not perfect predictors of protein synthesis and circulating protein levels. However, mRNA levels do show a positive correlation with protein expression (81) with a significant amount of protein (40%) being correlated to mRNA levels (82). Likewise, protein expression alone is a perfect predictor of the ultimate biological action, as the completion of a biological action is due to a compendium of multiple epigenetic effectors including the tumor microenvironment in the case of cancer. In this exploratory study, our objective was to determine the impact of lactate exposure on the expression of key genes known to be involved in the pathogenesis of cancer in MCF7 cells. Future work will be expanded to include the assessment of protein levels for differentially expressed genes.

In summary, our study supports the hypothesis that lactate has the potential to serve as an oncometabolite, regulating transcriptional activities of different key cancer-related genes involved in metabolic reprograming as well as cell cycle and proliferation (p's < 0.05–0.001). Beyond present results with MCF-7 cells additional studies on different cancer cell lines and cultured tumor biopsy cells will be needed to further support the lactagenesis hypothesis and to better understand the role of lactate in carcinogenesis.

### DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

### AUTHOR CONTRIBUTIONS

IS-M and GB contributed to the hypothesis and experiments design as well as the preparation of the manuscript. CJ contributed to the experiments and also to the manuscript. CM contributed to the experiments. JM, IS-M, and GB contributed to the preparation of the manuscript.

### FUNDING

Funding for this study came from IS-M Laboratory funds and supplementary support from NIH 1 R01 AG059715-01 to GB.


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

Copyright © 2020 San-Millán, Julian, Matarazzo, Martinez and Brooks. 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 Multifaceted Role of Heme in Cancer

Veronica Fiorito\*, Deborah Chiabrando, Sara Petrillo, Francesca Bertino and Emanuela Tolosano

*Department of Molecular Biotechnology and Health Sciences, Molecular Biotechnology Center, University of Torino, Turin, Italy*

Heme, an iron-containing porphyrin, is of vital importance for cells due to its involvement in several biological processes, including oxygen transport, energy production and drug metabolism. Besides these vital functions, heme also bears toxic properties and, therefore, the amount of heme inside the cells must be tightly regulated. Similarly, heme intake from dietary sources is strictly controlled to meet body requirements. The multifaceted nature of heme renders it a best candidate molecule exploited/controlled by tumor cells in order to modulate their energetic metabolism, to interact with the microenvironment and to sustain proliferation and survival. The present review summarizes the literature on heme and cancer, emphasizing the importance to consider heme as a prominent player in different aspects of tumor onset and progression.

#### Edited by:

*Alessandra Castegna, University of Bari Aldo Moro, Italy*

#### Reviewed by:

*Cinzia Domenicotti, University of Genoa, Italy Daniel Pereira Bezerra, Oswaldo Cruz Foundation (Fiocruz), Brazil*

> \*Correspondence: *Veronica Fiorito veronica.fiorito@unito.it*

#### Specialty section:

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

Received: *27 October 2019* Accepted: *19 December 2019* Published: *15 January 2020*

#### Citation:

*Fiorito V, Chiabrando D, Petrillo S, Bertino F and Tolosano E (2020) The Multifaceted Role of Heme in Cancer. Front. Oncol. 9:1540. doi: 10.3389/fonc.2019.01540*

Keywords: heme, cancer, iron, metabolism, microenvironment (ME)

### INTRODUCTION

The onset and progression of cancer rely on the ability of tumor cells to channel different biological processes toward the promotion of cell proliferation, the escape from immunosurveillance and the resistance to drugs. Inorganic iron has been reported to play pivotal roles in several aspects related to cancer metabolic adaptation and tumor microenvironment reprogramming (1–3). Similarly, organic iron, in the form of the iron-containing porphyrin heme, is potentially a best candidate molecule exploited/controlled by tumor cells in order to modulate the energetic metabolism, to interact with the microenvironment and to sustain proliferation and survival. Heme bears a series of functions that are far beyond those mediated by its iron atom, including oxygen transport and storage, drug and steroid metabolism, transcriptional and translational regulation, signal transduction and microRNA processing (4, 5). In addition, heme synthesis is a cataplerotic pathway for the tricarboxylic acid cycle, as it consumes succynil-CoA. However, among the different metabolic processes that cancer cells can regulate to meet their specific demands, heme metabolism has been so far marginally studied, and frequently the importance of heme for cancer has been attributed to the iron atom contained in heme, rather than to specific functions mediated by the entire heme molecule itself. The present review will summarize the literature on heme and cancer highlighting both positive and negative effects of heme on cancer cells and on components of the tumor microenvironment.

### DIETARY HEME AND CANCER

Historically, the role of heme in cancer has been studied focusing on the effects mediated by exogenous dietary heme on the organism. These studies contributed to the current notion that

**28**

dietary heme is a risk factor for cancer. Heme is an iron coordinating porphyrin contained predominantly in red and processed meat in the form of hemoglobin and myoglobin. Red meat refers to unprocessed mammalian muscle meat, while processed meat refers to meat that has been transformed through salting, curing, fermentation, smoking, or other processes to enhance flavor or improve preservation. Recently, the International Agency for Research on Cancer (IARC), following an assessment of over 800 studies performed world-wide, classified processed meat as group 1 "carcinogenic to humans" and fresh red meat as group 2A "probably carcinogenic to humans" (6). Conversely, no link between white meat and fish and cancer has been found (7–9). For this reason, heme has been proposed as the key molecule contributing to tumorigenesis upon red and processed meat intake.

The role of dietary heme in cancer has been highlighted in different types of carcinomas. Indeed, high consumption of red and processed meat has been associated with increased incidence of esophageal, gastric, breast, endometrial, pancreas and lung tumor (10–15), while no clear link was found for bladder and prostate cancer (16–18). However, the majority of studies focused on the role of dietary heme in the pathogenesis of colorectal cancer (CRC), still a leading cause of cancer deaths in Western Countries (19–21). Dietary heme is absorbed mostly in the upper part of the small intestine. Once absorbed, heme is degraded by the action of the enzymes heme oxygenases (HMOXs) into biliverdin, carbon monoxide (CO), and iron (Fe2+), that is then scavenged by the protein ferritin (4). However, if red/processed meat is assumed in large amounts, all the ingested heme cannot be absorbed by the small intestine and it accumulates for a considerable time in the large intestine (22, 23). In presence of high free heme levels both ferritin and HMOXs are saturated and cells accumulate free heme and labile iron that exert a variety of cytotoxic effects on intestinal mucosa (24). For example, heme is able to induce cytotoxic damage to surface epithelial cells that changes surface to crypt signaling, resulting in hyperproliferation and finally hyperplasia of crypt cells in heme-fed mice (25). Furthermore, free heme and labile iron accumulation result in the production of reactive oxygen species (ROS) that pathologically oxidize DNA, lipids and proteins. It has been well-demonstrated that ROS-induced DNA damage and gene mutations cause CRC and that proteins involved in CRC development are redox-sensitive (26). Additionally, ROS are able to induce lipid peroxidation of intestinal cells. Reactive lipid peroxides, formed by the action of ROS, covalently bind to the protoporphyrin ring of heme giving rise to an extremely lipophilic molecule, named cytotoxic heme factor (CHF) that induces cytotoxic damage on intestinal epithelial cells (27). To note, processed meat when in contact with gastric acid can also give rise to lipid hydroxiperoxide (LOOHs). LOOH is then broken down by the iron released by heme to produce free radicals and, subsequently, aldehyde molecules like malondialdehyde (MDA) and 4-hydroxynonenal (4-HNE) (28). MDA is toxic and it is able to bind DNA forming mutagenic adducts. 4-HNE induces apoptosis and kills normal cells, but not precancerous cells that are mutated on Adenomatous Polyposis Coli (APC) gene (29).

In addition to the processes described above, heme is also able to catalyze the formation of N-nitroso compounds (NOC) in the gastrointestinal tract (30). NOC are known carcinogens that can determine DNA mutation through alkylation, and high NOC concentrations have been associated to increased red meat consumption (31, 32). However, it is important to underline that NOC found after ingestion of red meat in humans consist mainly of nitrosyl iron and nitrosothiols, products that have profoundly different chemistries as compared to some tumorigenic N-nitroso species (33). Therefore, more studies are required to clarify the real involvement of heme in NOC pro-tumor effects.

Finally, it has been demonstrated that heme alters the normal intestinal bacterial flora especially by decreasing the number of gram-positive bacteria (34) leading to a state of dysbiosis (microbial imbalance or maladaptation) that exacerbates colitis and adenoma formation in mice (35) and is correlated to the insurgence of CRC (35–37). Moreover, the gut microbiota can induce per se hyperproliferation via mechanisms occurring in the colon lumen, such as modulation of oxidative and cytotoxic stress or by influencing the mucus barrier, and these effects are intensified in presence of heme (38). Indeed, a recent study showed that mice receiving a diet with heme show an increased population of mucolytic bacteria in their colon. These bacteria, synergistically with heme-produced CHF, damage gut epithelium and lead to a compensatory aberrant hyperproliferation. Conversely, mice receiving heme together with antibiotics do not show this phenotype (38).

Overall, the studies on dietary heme and cancer support the idea that heme contained in food can sustain cancer by different mechanisms (**Figure 1**). However, it must be recognized that

**Abbreviations:** IARC, International Agency for Research on Cancer; CRC, colorectal cancer; HMOXs, heme oxygenases; CO, carbon monoxide; Fe2+, ferrous iron; ROS, reactive oxygen species; CHF, cytotoxic heme factor; LOOHs, lipid hydroxiperoxide; MDA, malondialdehyde; 4-HNE, 4-hydroxynonenal; APC, Adenomatous Polyposis Coli; NOC, N-nitroso compounds; ALA, 5 aminolevulinic acid; ALAS1, 5-Amilolevulinate Synthase 1; ALAD, ALA dehydratase; PBGD, porphobilinogen deaminase; UROS, uroporphyrinogen III synthase; UROD, uroporphyrinogen decarboxylase; CPOX, coproporphyrinogen oxidase; PPOX, protoporphyrinogen oxidase; FECH, ferrochelatase; PpIX, protoporphyrin IX; FGS, tumor fluorescence-guided surgery; PDT, photodynamic therapy; ETC, electron transport chain; OXPHOS, oxidative phosphorylation; ATP, adenosine triphosphate; HLRCC, hereditary leiomyomatosis and renal-cell cancer; TCA cycle, tricarboxylic acid cycle; ATPIF1, ATPase Inhibitory Factor 1; PDH, pyruvate dehydrogenase; ANT, adenine nucleotide transporter; GRBOX, glycolysis regulated box; TDO, tryptophan 2,3-dioxygenase; IDO1/2, indoleamine-2,3-dioxygenase 1 and 2; CYP1B1, cytochrome P450 family 1 subfamily B member 1; COX2, prostaglandin-endoperoxide synthase 2; NSCLC, non-small cell lung cancer; NR1D1 or Rev-erbα, nuclear receptor subfamily 1 group D member 1; NPAS2, neuronal PAS domain protein 2; PER2, period circadian regulator 2; HRM, heme-responsive motif; SLC46A1 or HCP1, solute carrier family 46 member 1; FLVCR2, feline leukemia virus subgroup C receptor family member 2; SLC48A1 or HRG1, solute carrier family 48 member 1; GLUT1, glucose transporter 1; FLVCR1, feline leukemia virus subgroup C receptor family member 1; ABCG2, ATP binding cassette subfamily G member 2; HER2, human epidermal growth factor receptor 2; TME, tumor microenvironment; TAMs, tumor-associated macrophages; RBCs, red blood cells; TECs, tumor-associated endothelial cells; ECM, extracellular matrix; MMPs, matrix metalloproteinases; CAFs, cancer-associated fibroblasts; MPO, myeloperoxidase; EPO, eosinophil peroxidase; NGF, nerve growth factor; DGCR8, DiGeorge critical region-8; miRNA, microRNA; CLOCK, clock circadian regulator; NCOR, nuclear receptor corepressor; HDAC3, histone deacetylase 3; GIS1, histone demethylase GIS1; jmjd-2/KDM4, lysine demethylase 4; AML, acute myeloid leukemia; CCs, cancer cells; Ac, acetylation; Me, methylation.

methodologies employed in some of these studies have been challenged. Indeed, some animal studies took advantage of diets low in calcium and high in fat, combined with the exposure of heme frequently at doses higher than that expected with a normal dietary consumption of red meat. Moreover, pork meat shows low heme levels, but it has been associated with CRC. Finally, it cannot be excluded that carcinogenesis could be ascribed to other molecules contained mostly in red meat, not related to heme (33). Therefore, further research is required to clarify the role of heme present in red and processed meat in cancer.

### HEME SYNTHESIS AND CANCER

Heme can be acquired by dietary sources, but in addition, all the cells in the organism are able to synthesize heme. Heme synthesis includes eight different reactions occurring partly in mitochondria and partly in the cytosol. The first rate limiting step is based on the condensation of succynil-CoA and glycine to produce 5-aminolevulinic acid (ALA), a reaction catalyzed by the enzyme 5- Amilolevulinate Synthase 1 (ALAS1). Then, by subsequent reactions involving the enzymes ALA dehydratase (ALAD), porphobilinogen deaminase (PBGD), uroporphyrinogen III synthase (UROS), uroporphyrinogen decarboxylase (UROD), coproporphyrinogen oxidase (CPOX), protoporphyrinogen oxidase (PPOX) and ferrochelatase (FECH), heme is finally produced (4, 39).

The study of heme synthesis in tumors has raised interest since many years. Indeed, in nineties, it was discovered that tumors, upon ALA administration, are able to accumulate remarkably higher amount of protoporphyrin IX (PpIX) as compared to normal tissues, and this property was demonstrated to be exploitable for tumor fluorescenceguided surgery (FGS) and to kill cancer cells by photodynamic therapy (PDT) (40–42). Since then, extensive research has been performed to determine the molecular mechanism involved in enhanced ALA-PpIX accumulation in tumor cells. Particularly, the rate of heme biosynthesis in different kinds of tumor was dissected in several works, leading to accumulation of conflicting results. However, ALAS1, PBGD and UROD expression and/or activity were frequently found up-regulated in cancer (43). Consistently, repression of heme biosynthesis by the ALAD inhibitor succinylacetone was shown to reduce tumor cell survival and proliferation (44–46). Conversely, FECH levels were found often down-modulated in tumor cells as compared to normal cells (43, 47).

Taking together these discoveries, it can be concluded that tumors are characterized by high porphyrins synthesis, not necessarily associated to final heme production. This conclusion, however, is controversial. Indeed, cancer cells have been shown to display high heme levels (45, 48), increased activity of heme containing proteins (45, 46, 49) and enhanced expression of heme exporters (45, 50), suggesting that heme, and not only its precursors, is produced and that the entire heme biosynthetic pathway is promoted in tumors. Therefore, the reason why tumors accumulate more ALA-mediated PpIX than surrounding normal tissues and why tumors enhance heme synthesis remains a fundamental question to be answered (**Figure 2**).

Considering the fact that heme is a crucial cofactor for complexes of the electron transport chain (ETC), one possible explanation for increased heme biosynthesis in tumors is that it could be exploited by cells to sustain oxidative phosphorylation (OXPHOS). Indeed, despite the well-accepted model proposed by Otto Warburg and co-workers, pointing to increased glycolysis in cancer cells, many lines of experimental evidence have shown that the function of mitochondrial OXPHOS in most tumors is intact and that the vast majority of tumor cells generate adenosine triphosphate (ATP) via oxidative phosphorylation (51). Data obtained on the human lung carcinoma cell line A549 showed that induction of heme biosynthesis by ALA enhances OXPHOS in these cells (52). Similarly, in additional studies on lung cancer (46) and in an in vitro model of myeloid leukemia (45), it was demonstrated that increased heme synthesis is associated to higher oxygen consumption, an indicator of OXPHOS, and the effect was prevented by cell treatment with succynilacetone.

Nevertheless, tumors like hereditary leiomyomatosis and renal-cell cancer (HLRCC), due to mutations in the tricarboxylic acid (TCA) cycle enzyme fumarate hydratase, show reduced OXPHOS and sustained glycolysis but increased heme synthesis (53). Similarly, human colon carcinomas overexpress the mitochondrial ATPase Inhibitory Factor 1 (ATPIF1), an inhibitor of the mitochondrial H+-ATP synthase (54) and OXPHOS, as well as a promoter of iron incorporation into PpIX by FECH (55). These studies support the idea that suppression of OXPHOS and enhanced glycolysis in tumors could be associated to increased heme biosynthesis, suggesting that heme can mediate additional functions in cancer, unrelated to its role as a cofactor for ETC complexes.

A complex interplay between heme and energy metabolism is further supported by studies indicating that ALAS1 expression is suppressed by glucose (56). Moreover, it has been recently reported that heme can negatively regulate the activity of pyruvate dehydrogenase (PDH) by binding the PDHA1 subunit, favoring a switch from pyruvate oxidation in the mitochondria to its glycolytic conversion into lactate (57). In addition, heme can control the trafficking of ADP and ATP between mitochondria and cytosol. The translocase involved in the ADP/ATP exchange is the adenine nucleotide transporter (ANT), an integral protein located in the inner mitochondrial membrane. While in rodents only three ANTs exist (ANT1, 2, and 4), in humans there are four ANT isoforms (ANT1, 2, 3, and 4), encoded by four distinct genes with different promoter sequences, supporting distinct regulatory control. In humans, ANT1 is mainly expressed in heart, skeletal muscle and brain, while ANT3 is ubiquitous (58) and ANT4 is expressed in testis and germ cells (58). Conversely, ANT2 is poorly detectable in tissues, but its expression can dramatically increase in cancer and proliferating cells with high glycolytic rates and/or low oxygen (58, 59). ANT1 and ANT3 export ATP synthesized in mitochondria toward the cytosol, while several evidences suggest that ANT2 may realize an inverse exchange, translocating the glycolytic ATP synthesized in the cytosol toward the mitochondrial matrix (59). Moreover, ANT1 and 3 are crucial component of the mitochondrial permeability transition pore complex and play a major role in mitochondria-mediated cell death (60–62). Heme binds ANT1, 2, and 3 isoforms and ANTs are believed to contribute to heme biosynthesis by transporting heme precursors into mitochondria (63). Particularly for ANT1, it has been demonstrated that heme binds to the center pore domain and to the residues that associate to ADP, and binding of heme, or of the heme precursors PpIX and coproporphyrin III, inhibits the ADP uptake in a competitive manner (63). Moreover, heme can transcriptionally repress the expression of ATP/ADP carrier (AAC) 3, the yeast ANT2 orthologous, through the ROX1 factor (64). Interestingly, human ANT2 promoter shows a negative regulatory motif, called glycolysis regulated box (GRBOX) that is a ROX1-like motif (59), suggesting possible heme-dependent regulation also of human ANT2. This could be particularly relevant in cancer, because ANT2 has been postulated to contribute to the rapid metabolic adaptation of tumor cells to glycolysis and hypoxic stress by preserving the mitochondrial ATP content necessary to maintain the mitochondrial membrane potential and intramitochondrial metabolic pathways when OXPHOS is impaired (58, 65). Moreover, additional studies have highlighted an antiapoptotic role for ANT2 (66), its implication in the PI3K/Akt pathway (67), frequently activated in cancer, and in the regulation of cancer-related microRNAs (67). Finally, ANT2 is considered a promising therapeutic target to control tumor cell growth, migration, invasion and chemosensitization (68). Therefore, the modulation of heme synthesis, by affecting ANTs activity and expression, could have consequences on the transport of both OXPHOS-derived ATP from mitochondria to cytosol or of glycolytic ATP from the cytosol to the mitochondria.

Overall, the role of heme synthesis in the modulation of cancer energy metabolism remains controversial.

Another possibility is that increased heme synthesis in tumors is intended to support the activity of specific hemoproteins. Cells are equipped with several hemoproteins (69) and for some of them a role in cancer has been reported, including myoglobin (70), tryptophan 2,3-dioxygenase (TDO) and indoleamine-2,3-dioxygenase 1 and 2 (IDO1/2) (71–73), mitochondrial cytochromes (74, 75), cytochrome P450 (76, 77), and cyclooxygenases (78, 79). Studies on lung cancer demonstrated that the amount of oxygen-utilizing hemoproteins, such as cytoglobin, cytochrome c, cytochrome P450 family 1 subfamily B member 1 (CYP1B1) and prostaglandinendoperoxide synthase 2 (COX2) are higher in tumor cells as compared to normal cells and that, at least for cytoglobin and cytochrome c, their levels depend on the rate of heme synthesis

activity and stability, as well as the activity of hemoproteins and of heme binding proteins involved in the circadian clock machinery.

and cellular heme content (46). A similar phenotype was observed in human non-small cell lung cancer (NSCLC) tissues (49). Moreover, in colorectal cancer cells ALA administration results in the down-modulation of cyclooxygenase 2 expression, although its overall enzymatic activity is maintained (80). These hemoproteins can control oxygen availability or participate in crucial metabolic processes tightly modulated in cancer cells to allow/promote cell survival and proliferation, as well as to escape tumor immunosurveillance. Furthermore, heme can interact and regulate the activity of additional proteins with an already reported role in cancer. For example, heme is a crucial regulator of the nuclear receptor subfamily 1 group D member 1 (NR1D1 or Rev-erbα) (81), as well as an interactor for neuronal PAS domain protein 2 (NPAS2) (82) and period circadian regulator 2 (PER2) (83, 84). All these proteins are involved in the circadian clock mechanism, and alterations in circadian rhythm is typically observed in cancer, so that some of the clock machinery components have been proposed as targets in cancer therapy (85). Therefore, heme synthesis promotion in tumor cells could also partly contribute to modulate this system in order to sustain cell growth.

In addition, increased heme synthesis in cancer cells may be promoted in order to regulate the tumor suppressor P53, a transcription factor that controls a broad and flexible network of biological processes, including DNA damage response, autophagy, cellular metabolism, epigenetics, inflammation, just to cite the most relevant ones (86). Interestingly, Shen et al. (87) demonstrated that the stability of P53 is directly regulated by heme. Indeed, heme binds to a C-terminal heme-responsive motif (HRM) of P53. The heme-P53 interaction interferes with the P53 DNA binding activity in vitro, thus suggesting that heme may modulate the transcription of P53 target genes. Furthermore, the binding of heme to P53 promotes its nuclear export and degradation via the ubiquitin-proteasome system. These findings provide mechanistic insights into the process of tumorigenesis associated with iron excess. It is well-established that iron excess is a hallmark of several tumor types (88). A positive correlation between iron and heme levels in vivo has been reported (87), leading to the hypothesis that iron excess in cancer sustains the synthesis of heme, that in turn directly affects P53 stability and function. Considering the increasing amount of data showing that tumors reprogram heme metabolism (not only iron metabolism) to achieve advantages in terms of proliferation and survival, it is tempting to propose that the alteration of heme metabolism that occurs during tumor onset and cancer progression may contribute to the dysregulation of P53 expression.

In the end, it has been proposed that heme synthesis is enhanced in cancer in order to mediate TCA cycle cataplerosis. By this view, enhanced heme biosynthesis is primarily intended to consume succynil-CoA rather than to produce heme. In HLRCC it has been demonstrated that increased heme biosynthesis is crucial to avoid the accumulation of toxic TCA cycle intermediates, as a consequence of mutations in fumarate hydratase. Heme produced in this system is then addressed to degradation by the enzyme HMOX1, indicating that its production is not intended to increase the intracellular bioavailable heme pool (53).

Overall, it seems clear that heme biosynthesis is frequently enhanced in tumors and that this phenomenon could serve different and sometimes conflicting purposes. Additional studies are required to fully elucidate the underlying mechanism that can reconcile all these aspects.

## HEME IMPORT/EXPORT/DEGRADATION AND CANCER

Other than being synthesized, heme can also be imported inside the cell by three main importers: (1) the solute carrier family 46 member 1 (SLC46A1 or HCP1), a folate importer able to transport also heme, (2) the feline leukemia virus subgroup C receptor family member 2 (FLVCR2), and (3) the solute carrier family 48 member 1 (SLC48A1 or HRG1). Among them, FLVCR2 and HCP1 are exposed on the cell plasma membrane, while HRG1 is localized on endolysosomes.

The role of FLVCR2 in normal cells is not well-studied and to our knowledge there is only one paper to date in which FLVCR2 was analyzed in cancer. In this paper, interestingly, it was observed that FLVCR2 is overexpressed in bovine papillomavirus-associated urinary bladder cancer (89).

Regarding HCP1 and HRG1, their levels were reported to be dramatically increased in lung cancer cells/tissues (46, 49), where they contribute to import heme in order to ensure proper activity of hemoproteins. Indeed, depletion of heme in the culture medium of tumor cells or the use of heme-sequestering peptides, to avoid heme uptake, result in the down-modulation of hemoproteins activity (46, 49). In addition, high expression of HCP1 was detected in gastric cancer cells as compared to normal gastric cells (90). Moreover, HRG1 overexpression was observed in HeLa cells and in highly invasive and migratory cancer cell lines, where it can be detected not only in endolysosomes but also on the cell plasma membrane. Its silencing in these cells was demonstrated to result in reduced survival and migratory capacity, while the opposite phenotype was observed upon its over-expression (91, 92). In HeLa cells it was demonstrated that HRG1 interacts with the vacuolar H+-ATPase and regulates its activity, thus modulating the acidification of endosomes (91). By this way, HRG1 has a unique role in regulating pHdependent endocytic pathway, with an impact on the ability of the cells to acquire nutrients, to mediate signaling in response to growth factor receptor activation, and to internalize and traffic integrins and other proteins, which is necessary for cell survival, migration, and proliferation (91). Moreover, in additional highly invasive and migratory cancer cell lines, it was shown that, by the interaction with the vacuolar H+-ATPase, HRG1 participates to the regulation of cytosolic/extracellular pH gradient. Indeed, HRG1 indirectly favors the alkalinisation of cell cytosol and the acidification of the extracellular environment, a condition that enhances extracellular matrix degrading enzymes expression and activity, facilitating a more invasive phenotype of cancer cells (92). In addition HRG1 expression in these cells was associated to a pH-dependent promotion of glucose transporter 1 (GLUT1) trafficking to the plasma membrane, leading to increased glucose uptake and glycolysis, thus favoring cell growth (92).

Altogether, these studies indicate that heme import is enhanced in cancer with the aim to promote the activity of specific hemoproteins and to modulate pH gradient and cell metabolism.

Together with heme import, also heme export has been shown to be increased in cancer. Heme export is mediated by the specific heme exporter feline leukemia virus subgroup C receptor family member 1 (FLVCR1). Moreover, the ATP binding cassette subfamily G member 2 (ABCG2), a known exporter for a broad range of molecules and xenobiotics, is also involved in heme/porphyrins export.

The expression of FLVCR1 has been reported up-regulated in different kinds of tumors, including bovine papillomavirusassociated urinary bladder cancer (89), synovial sarcoma (50), and hepatocellular carcinoma (93). For hepatocellular carcinoma, the high expression of FLVCR1 was associated to higher neoplasm disease staging, adjacent tissue inflammation, vascular invasion and neoplasm histologic grade, as well as to reduced overall survival and disease-free status (93). In synovial sarcoma cells, the silencing of FLVCR1 was associated to reduced cell proliferation, survival and tumorigenicity in vitro and in vivo (50). Similarly, FLVCR1 silencing impairs the survival of neuroblastoma cells in vitro, particularly upon ALA administration (94).

The literature on ABCG2 and cancer is very rich; however, it is important to distinguish between its role as a heme/porphyrins exporter as compared to its role in the efflux of additional unrelated substrates and xenobiotics. Focusing on heme, it has been shown that triple negative breast cancer cell lines have significantly reduced ALA-PpIX levels as compared with estrogen receptor (ER) positive and human epidermal growth factor receptor 2 (HER2) positive breast cancer cell lines because of elevated ABCG2 activity (95). In line with this, high ABCG2 expression was considered a major cause of failure of ALAphotodynamic therapy in different kinds of tumors, as it can induce resistance to this kind of treatment by preventing the accumulation of the photosensitizer PpIX inside the cells (96, 97). Moreover, in an in vitro model of myeloid leukemia, it has been shown that MYCN drives increased heme synthesis and that, in this system, the excess of PpIX produced is promptly exported out of the cells by ABCG2 to avoid cell toxicity (45). Therefore, in these cells the forced stimulation of heme synthesis, associated with the imposed reduction of ABCG2 expression, leads to tumor cell death.

Overall, the literature on heme import and export in cancer looks counterintuitive, indicating that both processes are enhanced in tumor cells. Although an explanation for these apparently conflicting phenomena does not exist, it could be postulated that the two systems are both promoted in order to establish a new equilibrium in heme homeostasis. Another possibility is that import and export target two different pools of heme, and that cellular heme compartmentalization could play a role in this context. Interestingly, it has been shown that heme export by FLVCR1 is highly associated to heme synthesis (98), thus supporting the idea that the trafficking of endogenously synthesized heme could be managed differently as compared to exogenous heme imported into the cells. Additional studies are required to fully elucidate the biological significance for these alterations of heme trafficking in cancer.

Finally, intracellular heme homeostasis benefits of an additional system to control heme levels, which is the degradation of the molecule. As described above, HMOXs, the ratelimiting enzymes in heme catabolism, catalyze the stereospecific degradation of heme to biliverdin, with the concurrent release of ferrous iron ions and CO. There are two main HMOXs isoforms, encoded by two distinct genes: HMOX1 and HMOX2. HMOX2 is a constitutively expressed protein, while HMOX1 can be strongly induced in many tissues in response to cellular stress caused by a wide spectrum of stimuli including, but not restricted to, heme. The literature on HMOXs and cancer is wide, and a comprehensive summary of works on this topic can be found in several reviews (99–101). Nevertheless, also in this case it is important to distinguish between HMOXs role as heme degrading enzymes as compared to their noncatalytic role. Indeed, it has been shown that enzymatically inactive HMOX1 could translocate to the cell nucleus and exerts gene expression regulatory functions (102, 103). The role of HMOX1 in tumor cell proliferation is controversial: its expression in tumors was found very frequently up-regulated and a role for HMOX1 as a mediator of ROS-promoted cell proliferation was established; on the other hand, however, in other kinds of tumors an HMOX1 anti-proliferative action was demonstrated, often associated to effects mediated by CO and biliverdin (99, 101). However, most experiments support a permissive role of HMOX1 in tumor growth. What seems to be clear is that HMOX1 can affect different aspects of tumorigenesis, encompassing regulation of cell proliferation and differentiation, promotion of cytoprotection and inhibition of apoptosis, induction of angiogenesis and metastatization, as well as immunosuppression (99–101). Moreover, HMOX1 activity may affect anti-tumor therapies, as its expression is further elevated in response to radio-, chemo-, or photodynamic therapy and is involved in resistance to them (100, 101). Summarizing, the expression of HMOX1 and the activity of its byproducts can provide the selective advantage for tumor cells to overcome the increased oxidative stress occurring during tumorigenesis and/or as a consequence of anti-tumor therapies.

### HEME AND TUMOR MICROENVIRONMENT

Other than acting directly on tumor cells, heme can exerts its functions in cancer by modulating the tumor microenvironment (TME) (**Figure 3**). The TME is composed by tumor cells, endothelial cells of surrounding blood vessels, myofibroblasts, stellate cells, adipose cells, peritumor nerve cells, immune cells, endocrine cells, fibroblasts, and the extracellular matrix (104, 105). All the cell types in the TME contribute to tumor progression mainly by releasing factors, which establish a favorable environment for the cancer cell and promote tumor cell survival and migration, metastasis formation, chemo-resistance, and the ability to evade the immune system responses (104).

Heme has been reported to modulate the activity of tumor-associated macrophages (TAMs). Macrophages have heterogeneous phenotypes that range from the "classicallyactivated" pro-inflammatory M1-cells to the "alternativelyactivated" anti-inflammatory M2-cells (106). If on one hand M1 like macrophages have anti-tumor properties, on the other hand macrophage polarization toward a M2-like phenotype correlates with pro-tumor activities, such as enhanced angiogenesis, matrix remodeling, and immune suppression (107, 108). Due to the

specific cues that characterize the tumor microenvironment, TAMs are induced to preferentially acquire a M2-like specialized phenotype that protects cancer cells from targeted immune responses (109). Notably, M. Costa da Silva and colleagues demonstrated that TAMs exposed to haemolytic red blood cells (RBCs), a condition sometimes observed in cancer due to the extravasation of RBCs from the abnormal tumor-associated vessels (110), accumulate iron intracellularly and acquire a M1 pro-inflammatory phenotype, which in turn promotes tumor cell death (111). This is in line with studies on haemolytic diseases, such as sickle cell disease, where it has been observed that macrophage exposure to free heme, released by damaged erythrocytes, leads to M1-reprogramming (112). Moreover, these findings are in agreement with additional studies reporting the concept that iron can drive macrophages polarization toward a pro-inflammatory M1-phenotype (113, 114). The effects on TAMs described can be mainly ascribed to the iron atom contained in heme. However, another important aspect that should be taken into account is the role exerted by CO in the control of myeloid cell differentiation. Indeed, the CO produced upon heme degradation by HMOXs has been reported to cause tumor regression and increased drug sensitivity in prostate and lung cancer models (115), and this can be partly attributed to CO effects on macrophages within the TME. Specifically, it has been shown that CO treatment in mice leads to an increased number of M1-like macrophages and reduced tumor growth (116). Taken together, these works have led to the implementation of strategies able to modulate the exposure of macrophages to heme or iron, as the use of iron oxide nanoparticles (109, 111), for cancer therapy.

Another fundamental component of the tumor microenvironment is represented by tumor-associated endothelial cells (TECs). TECs strongly differ from their normal counterparts (117) as they display a high pro-angiogenic potential that mostly relies on their enhanced ability to proliferate and migrate. The switch of TECs from quiescence to a highly active state is favored by a specific metabolic reprogramming (118–120). The cross-talk between cancer cells and TECs promotes aberrant neo-angiogenesis, which is required to sustain tumor growth, by providing oxygen and nutrients, and to favor metastatization (117, 121, 122). To our knowledge, there are no studies in literature analyzing the role of heme in TECs. However, we demonstrated that alterations in endothelial intracellular heme metabolism strongly affect the angiogenic process during development (98, 123). Consistently, another study highlighted the critical role of the heme biosynthetic pathway in supporting endothelial functions (124). In addition, tumor angiogenesis is also promoted by the increased stiffness of the extracellular matrix (ECM) found in TME (125, 126). High ECM stiffness is mainly due to increased collagen deposition and increased cross-linking within the tumor stroma. Matrix remodeling is primarily controlled by the activity of a group of zinc-dependent endopeptidases, named matrix metalloproteinases (MMPs), which are mainly released by cancer-associated fibroblasts (CAFs). However, recent studies highlighted the involvement in this process of peroxidases released by immune cells, such as myeloperoxidase (MPO) and eosinophil peroxidase (EPO). In particular, MPOs and EPOs have been reported to be able to directly induce the secretion of collagen I and collagen VI by CAFs (127), thus increasing matrix stiffness and promoting tumor angiogenesis and metastatization (128). Notably, MPOs and EPOs are heme-containing enzymes, thus suggesting that changes in the amount of available heme within the cells could affect the activity of these enzymes in matrix remodeling. Taking together all these considerations, it is tempting to speculate that heme could affect different processes involved in tumor angiogenesis and future studies will help to verify this hypothesis.

Finally, an additional promising aspect to dissect in the future is the possible implication of heme in the control of tumor innervation. The peripheral nervous system is nowadays gaining growing interest in cancer research due to its role in modulating both cancer cells and TME. This is achieved through the reciprocal interaction between nerves and cancer cells (nerve-cancer cell cross-talk), as well as between nerves and the TME (129–131). Indeed, recent data clearly indicate that tumor onset and progression is accompanied by increased innervation, through a mechanism largely dependent on the secretion of neurotrophic factors by cancer cells. Furthermore, nerves influence tumor onset, progression and metastasis formation, mainly through the secretion of neuropeptides and neurotransmitters in TME, where they interact with receptors expressed by cancer cells and by other cells of the TME (131– 136). Heme is required for the survival of different types of neuronal cells (137, 138); however, the specific role of heme in the peripheral nervous system and its potential implication in tumor innervation is completely unknown. Several evidences suggest that heme is crucial for the maintenance of the peripheral nervous system, particularly for sensory neurons, one of the types of nerves that actively innervates tumors (139–141). Notably, mutations in genes encoding proteins involved in heme synthesis and export have been reported in diseases characterized by the degeneration of sensory neurons (94, 142– 145). Finally, heme may also regulates pathways important for nerve-cancer cell cross-talk. Indeed, heme is involved in the regulation of gene expression in neurons via nerve growth factor (NGF) signaling (146), thus suggesting that heme may modulate nerve outgrowth in the tumor microenvironment. Furthermore, heme also regulates the metabolism of some neurosteroids and neurotransmitters (137), with a potential implication in nerve-cancer cross-talk. In addition, other than sustaining cancer, tumor innervation is also one of the main cause of chronic pain in oncologic patients, particularly those in the advanced stage of the disease (147, 148). Interestingly, mutations in the heme exporter FLVCR1 have been reported in patients with peripheral sensory neuropathy (94, 149, 150), thus indicating an involvement of heme in pain perception. Furthermore, CO produced upon heme catabolism can act as an atypical neurotransmitter or neuromodulator in the nervous system and is involved in nociception regulation (151). Overall, these evidences support the idea that heme could be implicated in different aspects of tumor innervation. Future targeted in vitro and in vivo experiments will definitively verify whether this hypothesis is correct.

### ADDITIONAL HEME FUNCTIONS POTENTIALLY RELEVANT FOR CANCER

In the previous paragraphs, we discussed the works that contributed to explore the role of heme in tumor growth and metastatization. However, we believe that heme could be involved in additional aspects of tumor biology, not investigated so far. Indeed, emerging evidences indicate that heme is required for the processing of microRNA (miRNA) that, by regulating more than 50% of the mammalian genome are implicated in several pathways crucial for cancer onset, growth and metastatization (152). Specifically, the RNA-binding protein DiGeorge critical region-8 (DGCR8), which is essential for the first processing step of pri-miRNAs, is a heme-binding protein (153–156). Heme binding to DGCR8 is required for its dimerization and activation (153) and the modulation of heme availability was reported to affect pri-miRNA processing in vitro (157). A correlation between alterations of heme and miRNAs expression was also reported. For instance, ALA-mediated sonodynamic therapy or PDT is associated with the altered expression of selected miRNAs (158–161). Although, further studies are required to fully understand the physio-pathological implications of these findings, these data suggest that mysregulation of miRNAs may represent an additional mechanism through which alterations of heme metabolism sustain and promote cancer progression.

In addition, several data support a potential involvement of heme in epigenetic modifications, that control multiple processes essential for cancer cells (162, 163). This is suggested by the observation that heme regulates the transcriptional and demethylase activity of the yeast histone demethylase GIS1 (GIS1), that belongs to the lysine demethylase 4 (jmjd-2/KDM4) subfamily of demethylases implicated in histone methylation, cellular signaling and tumorigenesis (164). The yeast GIS1 protein is conserved from yeast to mammals, suggesting a possible role for heme in the regulation of this protein also in mammals. Furthermore, several heme-regulated proteins of the circadian rhythms machinery are epigenetic modifying enzymes themself. For instance, clock circadian regulator (CLOCK) is a histone acetyltransferase and Rev-erbα functions as a transcriptional repressor by forming a complex with the nuclear receptor corepressor (NCOR) and the histone deacetylase 3 (HDAC3) (165). Finally, the activity of epigenetic modifying enzymes relies on the availability of specific metabolites (like αketoglutarate) and cofactors (acetyl-CoA, S-adenosylmethionine, and nicotinamide adenine dinucleotide). Most of them are produced by the TCA cycle. Since heme biosynthesis is a TCA cycle cataplerotic pathway (166), it is reasonable that alterations of heme homeostasis may affect the availability of metabolites to epigenetic modifying enzymes. Based on the role of heme in

the control of cellular metabolism (166) and circadian rhythms (81), we propose that the alterations of heme metabolism observed in cancer may also contribute to the mysregulation of cancer epigenetics.

Finally, heme metabolism has been reported as an apoptosis modifying pathway in acute myeloid leukemia (AML) (167). This is relevant, because it suggests the possibility to target heme metabolism in order to increase drug sensitivity in cancer cells. In addition, we observed that the modulation of heme metabolism induces paraptosis (123), at least in endothelial cells. This suggests the possibility to potentially exploit the regulation of heme metabolism in cancer therapy, as the availability of compounds inducing alternative forms of programmed cell death could be very useful to counteract the resistance to apoptosis in tumors (168). Similarly, the identification of heme as an inhibitor of the proteasome (169) appears as a promising property to be exploited for therapeutic purposes, particularly considering that proteosome inhibitors are successfully currently used in cancer therapy (170).

### CONCLUSIONS

By the present review, we attempted to provide a comprehensive overview of the literature on heme in cancer, highlighting heme participation in multiple processes that sustain tumor growth and metastatization, encompassing the control of mitochondrial metabolism, the function of hemoproteins and P53 signaling.

Summarizing, the studies on dietary heme and cancer, although affected by some limitations, support the idea that heme contained in food can sustain cancer by different mechanisms. Conversely, the impact of endogenous heme in cancer is much more complex to envisage. On the one hand, heme biosynthesis is frequently enhanced in tumors, but this phenomenon could serve different and sometimes conflicting purposes. Moreover, both heme import and export are increased in tumor cells, but the reason is unclear. In addition, heme degradation by HMOX1 in tumor cells seems to support cancer by counteracting oxidative stress during tumorigenesis and upon anti-tumor therapies, but concomitantly to promote TAMs acquisition of a M1-like phenotype, favoring tumor regression and increased drug sensitivity. Finally, heme-containing enzymes like MPOs and EPOs can promote tumor angiogenesis and metastatization, and heme could also potentially affect cancer epigenetics, miRNAs and tumor innervation. Therefore, targeting heme metabolism is promising because it could have a broad impact on different aspects of cancer. For this reason, we envision that future work should be directed to the development of novel therapeutic strategies based on heme (**Figure 4**). However, the choice of the appropriate strategy is challenging, due to possible conflicting effects obtained by the block or the promotion of heme-related processes. To overcome these problems, further work is required in order to classify the precise tumor subtypes that can benefit of each single strategy.

Anyhow, according to present literature, the targeting of heme metabolism has already been exploited for cancer therapy. In particular, the stimulation of heme synthesis with ALA has been widely used for PDT (171). In addition, the indirect inhibition of heme synthesis through iron chelation therapy has been recently proposed for selected cancer types. Iron chelation therapy has emerged as an important chemotherapeutic strategy, because of the strong link between iron excess and tumorigenesis. However, iron-deprivation therapy was successful only on selected tumor types. The discovery that heme directly regulates P53 stability (87) explained the selective therapeutic efficacy of iron deprivation-based chemotherapy. Indeed, Shen et al. demonstrated that this selectivity was due to the P53 status of the tumor types (87). Specifically, iron chelation therapy, by decreasing heme levels, leads to the stabilization of P53 proteins only in tumors with wild-type P53, and not in case of P53 mutations. As already suggested by Shen et al., these findings will allow the discrimination of the types of tumor that should benefit of iron chelation therapy (87). The targeting of heme synthesis through the deprivation of iron required for heme synthesis (iron chelation therapy) remains the best therapeutic strategy to date. However, because iron is essential for multiple processes beyond heme synthesis, a key challenge of chelation therapy is to balance iron levels in order to avoid excessive iron chelation. Moreover, it has to be underlined that emerging studies demonstrate that some kinds of tumor could be counteracted by iron supplementation, rather than by iron chelation therapy (172). Therefore, more specific strategies to blunt heme synthesis are required, not based on iron. The discovery that the beneficial effects of iron-chelation therapy on the growth of certain tumor types depends on the regulation of P53 by heme raises the possibility to directly target heme synthesis to counteract tumor growth. Several compounds, like succinylacetone or N-methyl protoporphyrin, are used to inhibit heme synthesis in vitro. Although these compounds are not yet used in the clinic, we cannot exclude that new drugs based on them will be developed. Similarly, it could be possible that, in the future, drugs aimed at blocking/stimulating heme importer/exporter/degrading proteins will be identified, in order to perturb heme-related mechanisms in cancer.

In conclusion, we hope that, in the next future, the growing awareness on heme role in processes relevant for cancer will stimulate research aimed at implementing innovative therapeutic approaches and at identifying the tumor subtypes sensitive to these treatments.

### AUTHOR CONTRIBUTIONS

VF, DC, SP, and FB wrote the manuscript. ET critically revised the manuscript intellectual content.

### FUNDING

This work was supported by the Italian Association for Cancer Research (AIRC) IG18857 to ET.

### REFERENCES


effects on rat colonic epithelium. Carcinogenesis. (2000) 21:1909–15. doi: 10.1093/carcin/21.10.1909


microRNA-143 and suppression of the Bcl-2/Bax signaling pathway. Mol Med Rep. (2016) 14:544–50. doi: 10.3892/mmr.2016.5248


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

Copyright © 2020 Fiorito, Chiabrando, Petrillo, Bertino and Tolosano. 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.

# Beyond Energy Metabolism: Exploiting the Additional Roles of NAMPT for Cancer Therapy

### Christine M. Heske\*

*Pediatric Oncology Branch, National Cancer Institute, National Institutes of Health, Bethesda, MD, United States*

Tumor cells have increased requirements for NAD+. Thus, many cancers exhibit an increased reliance on NAD<sup>+</sup> production pathways. This dependence may be exploited therapeutically through pharmacological targeting of NAMPT, the rate-limiting enzyme in the NAD<sup>+</sup> salvage pathway. Despite promising preclinical data using NAMPT inhibitors in cancer models, early NAMPT inhibitors showed limited efficacy in several early phase clinical trials, necessitating the identification of strategies, such as drug combinations, to enhance their efficacy. While the effect of NAMPT inhibitors on impairment of energy metabolism in cancer cells has been well-described, more recent insights have uncovered a number of additional targetable cellular processes that are impacted by inhibition of NAMPT. These include sirtuin function, DNA repair machinery, redox homeostasis, molecular signaling, cellular stemness, and immune processes. This review highlights the recent findings describing the effects of NAMPT inhibitors on the non-metabolic functions of malignant cells, with a focus on how this information can be leveraged clinically. Combining NAMPT inhibitors with other therapies that target NAD+-dependent processes or selecting tumors with specific vulnerabilities that can be co-targeted with NAMPT inhibitors may represent opportunities to exploit the multiple functions of this enzyme for greater therapeutic benefit.

#### Edited by:

*Paolo E. Porporato, University of Turin, Italy*

#### Reviewed by:

*Valentina Audrito, University of Turin, Italy Cyril Corbet, Catholic University of Louvain, Belgium*

\*Correspondence: *Christine M. Heske*

*christine.heske@nih.gov*

#### Specialty section:

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

Received: *29 October 2019* Accepted: *16 December 2019* Published: *17 January 2020*

#### Citation:

*Heske CM (2020) Beyond Energy Metabolism: Exploiting the Additional Roles of NAMPT for Cancer Therapy. Front. Oncol. 9:1514. doi: 10.3389/fonc.2019.01514*

Keywords: NAD+, NAMPT, sirtuins, PARP, ROS, cancer

### INTRODUCTION

Cancer cells have altered metabolic needs, including an accelerated rate of nicotinamide adenine dinucleotide (NAD+) cycling relative to normal cells (1). To maintain this, NAD<sup>+</sup> metabolism is altered in cancer cells, many of which have an increased dependence on certain NAD<sup>+</sup> production enzymes (2, 3) (**Figure 1**). Several redundant NAD<sup>+</sup> production pathways exist. In the de novo pathway, tryptophan is first converted to quinolinic acid (QA) through a series of steps; QA is converted to nicotinic acid mononucleotide (NAMN) via quinolinate phosphoribosyltransferase (QPRT) and is then converted to NAD<sup>+</sup> via nicotinamide nucleotide adenylyltransferase (NMNAT) and NAD synthetase (NADS). In normal cells, QPRT expression follows a tissue-specific distribution; more recent insights have revealed that QPRT expression is altered in some cancer cells (4–7). The Preiss-Handler pathway converts nicotinic acid (NA) to NAMN through nicotinate phosphoribosyltransferase (NAPRT), an enzyme that is widely expressed in normal tissues but variably expressed in cancer cells (8–11). NAMN is then converted to NAD<sup>+</sup> through the activity of NMNAT and NADS, as in the de novo pathway. The salvage pathway, of which nicotinamide phosphoribosyltransferase (NAMPT) is the rate-limiting enzyme,

**43**

converts nicotinamide (NAM) to nicotinamide mononucleotide (NMN), which is then converted to NAD<sup>+</sup> through NMNAT. This pathway is of major importance to cancer cells, as it recycles NAM, the product of NAD+-consuming enzymes, back to NAD+. In fact, many types of cancer cells have been shown to highly express NAMPT, reflecting potentially increased dependence on this pathway due to high NAD<sup>+</sup> utilization and in some cases, loss of expression of other key NAD<sup>+</sup> biosynthetic enzymes (3, 9, 12, 13). Among the types of cancers reported to have high NAMPT expression are colorectal cancer (CRC), breast cancer, osteosarcoma, chondrosarcoma, pancreatic ductal adenocarcinoma, oral squamous cell carcinoma, prostate cancer, rhabdomyosarcoma, leiomyosarcoma, esophagogastric junction adenocarcinomas, thyroid cancer, leukemia, lymphoma, ovarian cancer, and some renal cancer, and in many of these, higher expression correlated with worse outcomes (14–30). Of note, NMN may also be produced from nicotinamide riboside via nicotinamide riboside kinase (9). Currently, however, NAMPT is the only NAD<sup>+</sup> production enzyme that has been targeted in the clinic (2, 31, 32).

Clinical NAMPT inhibitors have investigated in a number of early phase clinical trials (**Table 1**). Published results on the early phase experience with first generation clinical NAMPT inhibitors describe a disease control rate of ∼25% and few objective

### TABLE 1 | Summary of clinical trials testing NAMPT inhibitors.


*NHL, Non-Hodgkins lymphoma; RP2D, recommended phase 2 dose; OR, objective response; SD, stable disease.*

responses (33–38). Given the limited efficacy seen in these small studies, efforts to optimize the use of NAMPT inhibitors in the clinic are necessary. These include strategies such as drug combinations or selection of specific patient subsets more likely to be sensitive to these agents. Several new NAMPT inhibitors have recently entered early phase testing and preclinical efforts are focusing on use of these potential strategies to enhance activity and minimize toxicities (2, 3, 31, 32, 39).

The role of NAD<sup>+</sup> in supporting the energy metabolism of cancer cells has been well-established as NAD<sup>+</sup> is an important co-factor for a number of metabolic enzymes (10). Accordingly, NAMPT inhibitors have been shown to impair energy metabolism through disruption of specific metabolic pathways, decreased ATP production, and increased energetic stress (40–44). In cancer cells, NAMPT inhibitors affect glycolysis (40, 45–55), oxidative phosphorylation (45, 49, 53, 56–59), serine biosynthesis and one-carbon metabolism (60), the pentose phosphate pathway (40, 53, 61), amino acid metabolism (53), purine and pyrimidine metabolism (53), and fatty acid and lipid synthesis (45, 62). In addition, cancer types harboring mutations in metabolic pathways, such as isocitrate dehydrogenase (IDH), have been shown to be exquisitely sensitive to loss of NAMPT activity (63–65).

Beyond its role in energy metabolism, NAD<sup>+</sup> plays a vital role in many other cellular functions which may similarly be targeted with NAMPT inhibitors (13, 32, 66–70). An understanding of the non-metabolic implications of NAMPT inhibition may uncover additional targetable vulnerabilities, providing combinatorial opportunities for therapeutic intervention in cancer. The purpose of this review is to describe the impact of NAMPT and NAMPT inhibition on the non-energetic cellular functions of NAD<sup>+</sup> in cancer cells.

### SIRTUIN FUNCTION

The sirtuins are a family of NAD+-dependent deacylases and ADP-ribosyltransferases that are responsible for a significant amount of cellular NAD<sup>+</sup> consumption (71). The role of sirtuins in cancer is increasingly being described and consequently, there is a growing understanding of how targeting sirtuin function may be beneficial in certain cancer types (72). Among these insights are an appreciation of the role of NAMPT in sirtuin expression (73). In CRC models, several groups have reported that regulation of SIRT1 activity is mediated by NAMPT (74–77). In some cases, NAMPT was found to be a direct transcriptional target of c-Myc, resulting in a positive feedback loop between c-Myc, NAMPT, and SIRT1 that drove tumor cell proliferation and progression (74–76). These studies showed that the use of NAMPT inhibitors resulted in a loss of SIRT1 expression, derepression of TP53, and decreased tumor cell growth in CRC models (74, 75, 77). A similar effect has been observed in prostate (19) and gastric cancer models (78).

Regulation of the sirtuins by NAMPT has additionally been reported in other cancer types. In melanoma cells, NAMPT regulation of the E2F family member 2 was shown to impact transcription and translation of SIRT1, and genetic or pharmacologic loss of NAMPT activity resulted in activation of TP53 and apoptotic cell death (79). A similar regulatory network has been described in breast cancer cells where increased SIRT1 activity and induction of deacetylation of TP53 were observed upon exposure to the extracellular form of NAMPT (80).

In cancers where a direct regulatory link between NAMPT and the sirtuins has not yet been elucidated, experimental induction of NAMPT has been shown to have a corresponding effect on SIRT1 activity (81), while genetic or pharmacologic inhibition of NAMPT has resulted in decreased SIRT1 (82–85), SIRT2 (86), and SIRT3 (87). Interestingly, the functional significance of NAMPT on sirtuins is likely cancer cell type specific, since changes in SIRT1 expression with NAMPT inhibition have not been observed in all NAMPT inhibitor-sensitive cells (88). These differences may be important for clinical translation of NAMPT inhibitors, as they may play a role in determining which cancer types are more susceptible to NAMPT inhibitors (89).

### DNA DAMAGE REPAIR RESPONSE

Poly-ADP ribose polymerases (PARPs) represent another group of NAD+-dependent proteins that consume a large proportion of cellular NAD<sup>+</sup> (90). Among their functions, PARPs play a key role in DNA damage detection and repair (91). Thus, an expected consequence of NAD<sup>+</sup> depletion is impaired DNA damage repair. Decreased PARP activity with NAMPT inhibitor use has been reported in several cancer models, forming the basis for preclinical testing of NAMPT inhibitors plus PARP inhibitors in Ewing sarcoma and triple-negative breast cancer (92, 93). In both studies, synergy between NAMPT inhibitors and PARP inhibitors was observed. In the breast cancer study, the effect was noted to be greatest in BRCA-deficient models, suggesting that underlying defects in homologous recombination (HR) may further enhance the efficacy of NAMPT inhibition (93). This is supported by data indicating that NAMPT inhibition impairs non-homologous end joining and increases cellular dependence on HR (94). In ovarian cancer models, a regulatory relationship between BRCA1 and NAMPT has been described (95). Although it is not BRCA-deficient, Ewing sarcoma is also characterized by defective HR (96), further supporting the idea that cancers with defective DNA repair mechanisms may have increased susceptibility to NAMPT inhibition. Interestingly, results from a recent study in preclinical leukemia models revealed functional antagonism between NAMPT and PARP inhibitors, suggesting that cell type specific differences in how these pathways interact may be present (97).

Other cancer types with DNA repair deficiencies have been identified as selectively sensitive to NAMPT inhibitors. Nonsmall cell lung cancers (NSCLC) with excision repair crosscomplementation group 1 (ERCC1) deficiency were exquisitely sensitive to NAMPT inhibitors, in vitro and in vivo (98). ERCC1 deficiency is also associated with mitochondrial defects, suggesting that additional factors may contribute to NAMPT inhibitor sensitivity in this cancer subtype. In ovarian cancer, expression of NAPRT, the key enzyme in the Preiss-Handler pathway, correlated with a BRCA-ness gene expression signature, and cells carrying these features were more sensitive to NAMPT inhibitors (99). Mechanistic studies of the downstream consequences of NAMPT inhibitors on DNA damage repair have described a variety of effects including decreased PARylation (92, 99) decreased RAD51, and impaired double-strand break repair by the HR pathway (100).

Given these insights, a number of studies have sought to determine the efficacy of NAMPT inhibitors when combined with DNA damaging agents. Enhanced antitumor activity has been reported across multiple malignancies when genetic or pharmacological inhibition of NAMPT has been combined with radiation (101), DNA alkylating agents (63, 99, 100, 102, 103), topoisomerase inhibitors (19, 46, 86), or other classes of chemotherapy known to augment the effects of impaired DNA repair (19, 20, 43, 46, 104–106). Surprisingly, in some cancers, an improvement in efficacy was restricted to combinations with only certain chemotherapeutic agents, as in preclinical studies in pancreatic cancer models which revealed that gemcitabine, but not 5-fluorouracil (5-FU) or oxaliplatin, enhanced the antiproliferative effect of NAMPT inhibitors (107). In other studies, the combinatorial effects of NAMPT inhibitors with drugs from across chemotherapeutic classes was similar (46). The mechanisms for these differences are not known. Lastly, a study characterizing the effects of resistance to NAMPT inhibitors in CRC cell lines reported changes in expression of genes involved in DNA repair and an increased sensitivity to DNA damaging agents, further suggesting an intimate connection between NAMPT dependency and DNA damage repair (108). Taken together, these insights have clinical implications as they suggest that tumors with certain defects in DNA repair mechanisms may be selectively sensitive to NAMPT inhibitors, and that rational combinations with chemotherapies may enhance the efficacy of this class of agents, particularly in selected tumor types.

### REDOX HOMEOSTASIS

Maintenance of intracellular redox homeostasis is a critical cellular process requiring a balance between reactive oxygen species (ROS) generation and elimination, as excessive levels of ROS can result in cell death (109). NAD<sup>+</sup> is an important regulator of cellular ROS levels which can accumulate upon depletion of NAD<sup>+</sup> (110). This is particularly true in cancer cells which generally have increased ROS production and require very tight control of ROS balance (111). Hence, an additional consequence of NAMPT inhibition is disruption of ROS homeostasis.

NAMPT has been shown to contribute to the cellular capacity to tolerate oxidative stress in a number of studies. In CRC cell lines, NAMPT functioned to increase NADH pools, protecting cells against oxidative stress (112). In breast cancer models, NAMPT increased the pool of NAD<sup>+</sup> that could be converted to NADPH through the pentose phosphate pathway, thus maintaining glutathione in the reduced state. This was of particular importance in cells undergoing glucose-deprivation, for which high levels of NAMPT decreased mitochondrial ROS levels (61). Additionally, in several studies, NAMPT inhibition enhanced cancer cell susceptibility to oxidative stress through a reduction in antioxidative capacity via downregulation of antioxidant proteins (102, 113).

Depletion orinhibition of NAMPT has been shown to increase ROS in models of NSCLC (40), leukemia (43, 97), prostate cancer (19), breast cancer (61), glioblastoma (102), CRC (112), and others (85). In all cases, this was associated with a loss of cancer cell viability. In addition, there was a differential effect noted between the induction of ROS in cancer cells compared to normal cells treated with NAMPT inhibitors, suggesting the existence of a therapeutic window for ROS induction with these agents (114). Interestingly, not all cancer cells exhibit an increase in ROS upon inhibition of NAMPT. In Ewing sarcoma and some NSCLC models, cells were able to maintain ROS balance in the presence of a NAMPT inhibitor, suggesting that there may be cell-type dependent differences in these effects (40, 92). This may be the result of active compensatory NAD<sup>+</sup> production pathways, such as the Preiss-Handler pathway, in certain cancers (114). A comprehensive understanding of these differences will be important to clinical translation of this class of agents as they may have implications for patient selection.

Finally, several studies have reported on the efficacy of combining NAMPT inhibitors with ROS inducing agents. Use of NAMPT inhibitors plus β-lapachone, an agent that targets the NADPH quinone oxidoreductase-1 (NQO1) and generates ROS, resulted in excessive ROS production and had enhanced efficacy against growth of pancreatic adenocarcinoma cells, particularly those which overexpressed NQO1 (52, 115). A similar effect was seen with this combination in NSCLC models (83). In addition, adding a NAMPT inhibitor to ROS-containing plasma-activated medium resulted in increased ROS production, decreased intracellular reduced glutathione, and cell death of breast cancer cells (116).

### ONCOGENIC SIGNALING

Crosstalk between NAMPT and oncogenic signaling pathways has been reported in several cancer models. From a clinical perspective, co-targeting NAMPT with these pathways may be a beneficial strategy. In some cases, oncogenic factors regulate expression and activity of NAMPT, such as in Ewing sarcoma, where the oncogenic transcription factor EWS-FLI1 has been shown to regulate NAMPT expression (49) and in breast cancer, where FOXO1, a tumor suppressor, negatively regulates the expression of NAMPT while AKT positively regulates it (117). In other cases, NAMPT regulates the activity of oncogenic signaling pathways. For example, NAMPT overexpression in breast cancer cells and extracellular NAMPT (eNAMPT) released by melanoma cell have both been associated with AKT phosphorylation (118, 119). Exogenous eNAMPT was also found to induce phosphorylation of AKT and ERK1/2 and increase proliferation of breast cancer cells, and the use of AKT and ERK1/2 inhibitors could abrogate these effects (120). In multiple cancer models, a decrease in phospho-ERK1/2 was observed with NAMPT inhibition (44, 121, 122) and combining NAMPT inhibitors with ERK1/2 blockade enhanced cell death (121).

An interaction between NAMPT and mTOR has also been described in a number of malignancies. In hepatocellular carcinoma cells, NAMPT inhibition was associated with loss of activation of mTOR and its downstream targets. A corresponding increase in AMPKα activation was also noted (41). A similar effect was observed in leukemia cells (42), pancreatic ductal adenocarcinoma cells (46), and pancreatic neuroendocrine tumor cells (123). In both pancreatic cancer subtypes, the antiproliferative effect of NAMPT inhibition could be potentiated with concurrent mTOR inhibitor treatment (46, 123). In multiple myeloma models, NAMPT inhibition was also associated with loss of mTOR activation which was thought to contribute to autophagic death (121). In contrast, changes in AMPKα and mTOR were not observed in studies of non-cancerous cells treated with NAMPT inhibitors (41).

A number of studies have investigated changes in NAMPT expression that occur with development of drug resistance to targeted therapies. Both in clinical samples and experimental models, BRAF inhibitor resistant melanoma cells expressed higher levels of both intra- and extracellular NAMPT than their sensitive counterparts (57, 124). Remarkably, BRAF inhibitor resistance could be overcome with addition of NAMPT inhibitors (57). In addition, induced expression of NAMPT was able to render melanoma cells resistant to BRAF inhibitors while BRAF inhibition in sensitive cells resulted in transcriptional downregulation of NAMPT (125).

In addition to the oncogenic signaling molecules already described, correlative studies have also proposed a link between NAMPT expression and EGFR (44, 126), HER2, and estrogen receptor positivity (127). Furthermore, based on the data supporting crosstalk between oncogenic signaling and NAMPT, co-targeting NAMPT along with other signaling pathway molecules, as has been described with the BTK inhibitor ibrutinib in Waldenstrom macroglobulinemia cells, could be a promising therapeutic strategy (128).

### EPITHELIAL-MESENCHYMAL TRANSFORMATION AND STEMNESS

NAMPT has been described as a mediator of cancer cell stemness (129). Studies in clinical CRC tumor samples revealed that high NAMPT expression was associated with the presence of a high proportion of cancer-initiating cells. Mechanistically, this was the result of the influence of NAMPT on transcriptional regulation of stem cell signaling pathways and was mediated by SIRT1 and PARP (126, 130). In glioblastoma tumors and patient-derived stem-like cells, high NAMPT expression was observed (131). NAMPT overexpression in experimental models of glioblastoma resulted in a cellular phenotype consistent with that of a cancer stem-like cell (126), while pharmacological and genetic inhibition of NAMPT decreased the ability of glioblastoma stem cells to self-renew and form in vivo tumors (131). In one study, the loss of cancer stem cell pluripotency upon inhibition of NAMPT was the result of an excess of autophagy, a well-described consequence of NAMPT inhibition (15, 58, 64, 121, 132–135), which disrupted the maintenance of cancer cell stemness (136). NAMPT inhibition has also been shown to reverse the ability of cancer cells to dedifferentiate (137).

NAMPT inhibition also affects epithelial-mesenchymal transition (EMT) in cancer cells. In hepatocellular carcinoma cells, pharmacological NAMPT inhibition resulted in changes in EMT marker proteins indicating a reversal of EMT, as well as a reduction in cellular capacity for invasion and metastasis formation, through a decrease in SIRT1 (84). Similarly, data showing that both NAMPT overexpression and exogenous eNAMPT induced EMT in breast cancer cell lines (127), and that eNAMPT promoted osteosarcoma cell migration and invasion (138). Furthermore, NAMPT inhibition diminished motility in glioma cells (54). In contrast, in lung cancer cell lines, NAMPT inhibition activated EMT and increased cellular invasiveness also through decreased SIRT1 (85), and in breast cancer models, NAMPT inhibition enhanced metastatic behavior (139), suggesting the impact of NAMPT inhibition on EMT may be cell-type specific. Interestingly, expression of NAPRT, which differs across cancer cell lines, correlates with EMT status (140), and may be related to the differential effects of NAMPT inhibition on EMT. Thus, an understanding of the effect of NAMPT inhibition on metastasis is important as it may differ for different malignancies, impacting optimal clinical translation of these agents.

### IMMUNE REGULATION OF TUMOR MICROENVIRONMENT

In addition to the effect of NAMPT on primary tumor cells, recent insights have begun to elucidate the impact of NAMPT on the immune suppressive characteristics of the tumor microenvironment (26, 141). In murine cancer models, macrophage colony stimulating factor was shown to increase NAMPT expression in myeloid cells which, in turn, negatively regulated CXCR4 expression in hematopoietic cells in the bone marrow. Consequently, low CXCR4 resulted in mobilization of immature myeloid-derived suppressor cells (MDSCs), contributing to tumor immunosuppression. Importantly, pharmacologic inhibition of NAMPT resulted in a decrease in MDSC mobilization, reversing the immunosuppression and re-sensitizing tumor cells to immunotherapeutic agents in preclinical models (142).

NAMPT has also been shown to be upregulated in tumor associated neutrophils (TANs) in patients with melanoma and head and neck cancer, and in murine cancer models. Inhibition of NAMPT in ex vivo TANs followed by adoptive transfer of the TANs into tumor-bearing mice reduced tumor angiogenesis and proliferation through suppression of SIRT1 and resultant transcriptional blockade of proangiogenic genes (143). While more studies are required to better understand the role NAMPT inhibition plays in the microenvironmental immune milieu, these and other preliminary reports suggest NAMPT inhibitors could be used to enhance immunotherapies in the clinic (144). In addition, correlative studies describing the effects of NAMPT inhibition on tumor microenvironmental factors are currently lacking in the clinical literature but would be informative to further clinical development of this class of agents and should be pursued in future studies.

### DISCUSSION

Given the critical role that NAD<sup>+</sup> plays in the growth and survival of malignant cells, NAMPT is an attractive therapeutic target in cancer. In addition to its function in cellular energy metabolism, NAMPT is involved in sirtuin function, support of DNA repair mechanisms, maintenance of redox balance, molecular signaling, determination of cellular states, and tumorrelated immune suppression. Depending on the cancer cell type, NAMPT inhibitors may be able to impair many of these additional functions. Furthermore, combination therapies with agents that target these functions in a complementary manner have the potential to dramatically improve the efficacy of NAMPT inhibitors. There are an increasing number of reports describing additive or synergistic effects of NAMPT inhibitors being used in combination with other agents in the preclinical setting. With newer generation NAMPT inhibitors currently undergoing phase 1 evaluation, clinical translation of these rational combinations is a logical next step.

In addition to developing rational combination regimens using NAMPT inhibitors, careful patient selection represents an additional opportunity to maximize the efficacy of these agents. For example, IDH mutant cancers have been shown to have exquisite sensitivity to NAMPT inhibitors (63–65), as have tumors deficient in NAPRT (9, 145–151). Patient selection may also be guided by recognition of specific vulnerabilities in the non-metabolic pathways supported by NAMPT, such as HR-deficiency or EMT targeting for metastatic disease. In conclusion, it is critical to understand the impact of NAMPT and NAMPT inhibitors on both the energetic and the non-energetic cellular functions of NAD<sup>+</sup> in cancer as these insights may be key to future development of this class of agents.

## AUTHOR CONTRIBUTIONS

CH researched, wrote, and edited the manuscript.

### REFERENCES


## FUNDING

The author was supported by grants from the Intramural Research Program of the National Institutes of Health (NIH), National Cancer Institute, Center for Cancer Research.

## ACKNOWLEDGMENTS

The author would like to thank Mr. Choh Yeung, Dr. Marielle Yohe, and Dr. Sameer Issaq for critical reading of the manuscript.


sensitivity to DNA-damaging agents in waldenstrom macroglobulinemia. Clin Cancer Res. (2019) 25:369–77. doi: 10.1158/1078-0432.CCR-18-1776


apoptosis activator TRAIL in human leukemia cells. Exp Hematol. (2010) 38:979–88. doi: 10.1016/j.exphem.2010.07.013


NAMPT inhibitor sensitivity in glioma. Nat Commun. (2019) 10:3790. doi: 10.1038/s41467-019-11732-6


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

Copyright © 2020 Heske. 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.

# Non-invasive Investigation of Tumor Metabolism and Acidosis by MRI-CEST Imaging

Lorena Consolino1,2, Annasofia Anemone<sup>2</sup> , Martina Capozza<sup>2</sup> , Antonella Carella<sup>3</sup> , Pietro Irrera<sup>4</sup> , Alessia Corrado<sup>3</sup> , Chetan Dhakan3,4, Martina Bracesco<sup>2</sup> and Dario Livio Longo<sup>3</sup> \*

*<sup>1</sup> Department of Nanomedicines and Theranostics, Institute for Experimental Molecular Imaging, RWTH Aachen University, Aachen, Germany, <sup>2</sup> Department of Molecular Biotechnology and Health Sciences, Molecular Imaging Center, University of Torino, Turin, Italy, <sup>3</sup> Institute of Biostructures and Bioimaging (IBB), Italian National Research Council (CNR), Turin, Italy, <sup>4</sup> University of Campania "Luigi Vanvitelli," Naples, Italy*

Altered metabolism is considered a core hallmark of cancer. By monitoring *in vivo* metabolites changes or characterizing the tumor microenvironment, non-invasive imaging approaches play a fundamental role in elucidating several aspects of tumor biology. Within the magnetic resonance imaging (MRI) modality, the chemical exchange saturation transfer (CEST) approach has emerged as a new technique that provides high spatial resolution and sensitivity for *in vivo* imaging of tumor metabolism and acidosis. This mini-review describes CEST-based methods to non-invasively investigate tumor metabolism and important metabolites involved, such as glucose and lactate, as well as measurement of tumor acidosis. Approaches that have been exploited to assess response to anticancer therapies will also be reported for each specific technique.

### Edited by:

*Alessandra Castegna, University of Bari Aldo Moro, Italy*

#### Reviewed by:

*Phillip Zhe Sun, Emory University, United States Barbara Marengo, University of Genoa, Italy*

\*Correspondence:

*Dario Livio Longo dariolivio.longo@cnr.it; dario.longo@unito.it*

#### Specialty section:

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

Received: *13 December 2019* Accepted: *29 January 2020* Published: *18 February 2020*

#### Citation:

*Consolino L, Anemone A, Capozza M, Carella A, Irrera P, Corrado A, Dhakan C, Bracesco M and Longo DL (2020) Non-invasive Investigation of Tumor Metabolism and Acidosis by MRI-CEST Imaging. Front. Oncol. 10:161. doi: 10.3389/fonc.2020.00161* Keywords: tumor metabolism, tumor acidosis, CEST-MRI, imaging, therapy, tumor pH

### INTRODUCTION

Outgrowing tumor mass typically displays an abnormal and disorganized vascular network, with poor functional vessels and extended hypoxic region (1, 2). Hypoxia is considered one of the major driving forces of tumorigenesis through the activation of the hypoxia-inducible factor 1 (HIF-1), that directly alters the expression of genes related to cell metabolism and proliferation (3). The induced metabolic modification markedly responds to tumor requirement for survival and expansion. On one side, the upregulation of the transmembrane receptor GLUT-1 ensures increased glucose avidity as a metabolic source of proliferation (4). On the other side, the metabolic switch to the glycolytic pathway exposes tumors to the paradoxically accumulation of acidic metabolites, as lactic acid and hydrogen ions, that results to be toxic for cancer cells. Therefore, the upregulation of dedicated proton transporters allows the extrusion of acidic products on the extracellular microenvironment, guarantees the maintenance of an aberrant pH gradient and induces the adaptation and clonal expansion of the most aggressive cells able to survive in such a hostile environment (5–7).

Considering the strategic role of metabolism on tumorigenesis, several targeting therapies have been developed to interfere with tumor expansion, alone or in combination with standard therapeutic treatments (8–13). Therefore, approaches for in vivo assessing the response to treatments and for improving tumor diagnosis are strongly required. In the clinical setting, positron-emission tomography (PET) technique is routinely exploited for measuring glucose

**54**

uptake via 18F-fluorodeoxyglucose (FDG) injection, although radiation exposure limits repeated longitudinal studies (14– 16). Furthermore, magnetic resonance imaging (MRI) offers a wide panel of approaches, by combining an optimal tissue contrast and good spatial information with acceptable sensitivity, to quantitatively interrogate several aspects of tumor microenvironment, including tumor metabolism and acidosis (17–20). One of the most promising and emerging technique for investigating tumor metabolism is the chemical exchange saturation transfer (CEST)-MRI (21, 22). CEST-MRI allows the detection of molecules endowed with mobile protons in chemical exchange with water. The application of radiofrequency (RF) pulses at specific offsets, corresponding to the absorbance peak of the mobile protons, nullifies the magnetization of the mobile protons, that become "saturated." The exchange of the saturated protons with those of water molecules results in a transfer of reduced magnetization, hence in a decrease of the water signal, generating a (negative) contrast that can be detected by MRI. Consequently, many endogenous (proteins, peptides, sugars) or exogenous molecules owing exchangeable mobile protons can be imaged by CEST-MRI (23–25).

In this mini review, we will focus on CEST-MRI as a novel tool for imaging several aspects of tumor metabolism in both preclinical and clinical settings.

### IMAGING MOBILE PROTEINS (AMIDE PROTON TRANSFER: APT)

Amide proton transfer (APT) imaging is a CEST-MRI approach that can detect the amide protons of endogenous mobile proteins and peptides that resonate at 3.5 ppm (26). APT imaging has been initially exploited for studies of ischemic stroke, neurologic disorders and brain tumors (27–32). Tumors exhibit a close relationship between unregulated proliferation and concentrations of mobile proteins, that may accumulate as defective products (33). Especially in high grade malignant brain tumors, the level of peptides and mobile proteins is substantially elevated (34). In Yan et al. the APT signal was compared between normal brain tissue and tumor in rats implanted with gliosarcoma. This study demonstrated that higher APT contrast in brain tumor correlated with an increased concentration of cytosolic proteins (35). In addition, APT imaging has been used for tumor characterization and diagnosis of brain tumors in patients (36–39). Furthermore, it is possible to use this innovative technique to differentiate between malignant gliomas and malignant lymphoma (40), to discriminate solitary brain metastases from glioblastoma (41) and to predict genetic mutations in gliomas, in particular the isocitrate dehydrogenase (IDH) mutation status (42, 43). Another feature that makes APT particularly interesting is its ability to differentiate between treatment-induced effects and true tumor progression (44, 45), providing a unique and non-invasive MRI biomarker for distinguishing viable malignancy from radiation necrosis and for predicting tumor response to therapy (46). In addition to brain tumors, APT imaging has been investigated in breast and prostate cancer. As it was demonstrated in brain tumors, APT imaging is able to discriminate between prostate cancer and noncancer tissues, reporting an increase of cell proliferation rate and cellular density in tumor regions (47). Furthermore, variations in the APT signal have been observed in breast tumors, likely reporting about therapeutic effects and transformation of breast parenchyma (48, 49). In summary, APT imaging represents a promising biomarker for monitoring tumor progression and response to treatment and can be easily implemented in existing clinical scanners, despite further work is needed to remove confounding effects (protein concentration, pH, etc.) to the observed APT contrast (50–54).

## IMAGING GLUCOSE

Tumors typically display upregulated glucose uptake and glycolytic metabolism (55). In the clinical setting, PET imaging with the glucose analog FDG is considered the gold standard technique for non-invasively mapping glucose uptake and for assessing tumor response to conventional therapy (56). However, high maintenance costs and side effects related to radioactivity exposure of patients strongly limit the repeated applications of radionuclide techniques (57). Therefore, the idea of exploiting unlabeled D-glucose as an MRI contrast agent may represent a cheaper and potential alternative to FDG without involving ionizing radiations. Glucose molecules own five hydroxylic groups in fast exchange rate (500–6,000 Hz) with bulk water protons that can provide CEST contrast at 1–1.2 ppm from the water resonance (58, 59). The feasibility of imaging glucose uptake with the CEST-MRI technique was demonstrated in colorectal tumor xenograft murine models, with glucose contrast (GlucoCEST) correlated to FDG accumulation as measured by autoradiography (60). A different GlucoCEST contrast was also reported between two human breast tumor models characterized by different metabolic activity (58). In addition, the dynamic measurement of GlucoCEST contrast enhancements upon time (Dynamic Glucose Enhanced—DGE) following glucose injection showed increased penetration in brain tumors compared to the contralateral regions, demonstrating interesting application for brain tumors due to the reduced permeability of the blood brain barrier (61). One limitation of the GlucoCEST approach is the fast metabolism of native glucose that results in CEST contrast disappearance. Therefore, non-metabolizable glucose derivatives have been investigated for achieving prolonged contrast (=detectability) inside the tumor regions. Once phosphorylated by hexokinase enzymes, 2-Deoxy-D-glucose (2DG) remains entrapped in tumor cells and provides CEST contrast for long time, up to 90 min post injection (62, 63). However, the high doses required to generate enough contrast are not feasible for toxicity issues. A more promising molecule that has been intensively studied is the non-metabolizable 3- O-methyl-D-glucose (3OMG), that is considered non-toxic. Several studies tested 3OMG in different breast cancer models and showed higher uptake and CEST contrast in the more aggressive tumors, in according with the results obtained by FDG-PET (64–66). Beyond 3OMG, glucosamine (GlcN) and Nacetyl glucosamine (GlcNAc) can accumulate in tumors that overexpress the glucose transporters GLUT1 and GLUT2. These molecules were exploited as CEST contrast agents in breast and melanoma murine cancer models with different aggressiveness showing diverse accumulation inside the tumor (67, 68). Interesting results have been also obtained with low-calorie sweeteners, like sucralose, that was shown to provide CEST contrast in glioma tumor regions, and maltitol, that showed increased enhancement in brain tumors with compromised blood brain barrier (BBB) (69, 70).

Due to the high safety profile of glucose, its first use in patients was reported as early as 2015 in a glioma patients by using a high-field (7T) scanner (71). In comparison with the conventional small molecular weight Gd-based contrast agent, different areas of contrast enhancement were detected, suggesting that D-glucose may highlight tumor regions with different perfusion or permeability properties (**Figures 1A,B**). In addition, GlucoCEST contrast time curves highlighted potentially distinct biological areas of the brain tumor 10 min after D-glucose bolus infusion (**Figures 1B,C**). Another study investigated the GlucoCEST approach in head and neck cancer patients with a 3T scanner (72). Increased GlucoCEST contrast was registered in the tumor regions compared to muscle tissue and GlucoCEST enhancements were moderately correlated with FDG-PET results, despite a spatial mismatch likely reflecting the different metabolism between FDG and glucose. To improve the sensitivity of GlucoCEST, a similar approach that exploits the chemical exchange of mobile protons based on the Spin Lock method (dubbed CESL or chemical exchange spin lock) has been proposed for detecting glucose (73, 74). First results were obtained at high fields (9.4T) with a dynamic acquisition following glucose injection in glioma patients, demonstrating the feasibility of this approach for monitoring glucose accumulation in human brain tumors. Other studies showed a different glucose uptake in tumor brain regions in comparison to normal gray matter ones at lower magnetic fields (75), thus demonstrating its translational application at clinical level (76).

Overall, these results suggest that GlucoCEST could represent a valid alternative to FDG-PET for tumor diagnosis and staging, still several limitations, including reduced detectability at low field and origin of the glucose-based contrast arising from different compartments need to be tackled in the next years (77).

### IMAGING TUMOR ACIDOSIS

### Intracellular Tumor pH Imaging

The amine and amide concentration-independent detection (AACID) approach is a recently developed CEST contrast mechanism that has been shown to be sensitive to intracellular pH changes (pHi). AACID CEST technique uses the ratio of the CEST effects generated by amide (1ω = 3.50 ppm) and amine

(1ω = 2.75 ppm) protons from endogenous tissue proteins, which are predominantly from the intracellular space, for removing the concentration dependence. As a consequence, the measured CEST effect is only pH dependent, allowing to measure tumor intracellular pH (pHi) (78). McVicar et al. exploited the AACID CEST technique in a glioblastoma murine model to detect the selective acidification and decrease of pHi following the treatment with lonidamide, an anticancer drug that inhibits the monocarboxylic transporters (78). Similar results were obtained in glioblastoma murine models upon the administration of several pH-modulators such as topiramate, dichloroacetate and cariporide (79–82).

Another non-invasive pH-weighted imaging technique is the amine CEST approach, in which the amine protons (resonating at 3 ppm) of glutamine or glutamate molecules provide a pHdependent (but not concentration independent) CEST contrast for mapping acidic tumor regions. Harris et al. applied this approach in both glioma murine models and in glioblastoma patients to detect acidic tumors and response to bevacizumab treatment (83, 84). Although the high translational potential of these endogenous approaches, concerns related to their capability to distinguish between intra- and extracellular pH contribution are still under consideration. In addition, variation of amide protons concentrations might be responsible of confounding effects resulting in less reliable pH estimations.

A recent approach to uncouple the contribution of concentration and exchange rate to the measured CEST contrast is that based on the omega-plot technique, initially developed to assess chemical exchange rates in paramagnetic contrast agents (85). Such approach has been improved and exploited for diamagnetic molecules in vitro (simulating complex endogenous systems) by simultaneous determination of labile proton ratio and exchange rate (that is dependent on pH) (86, 87). Although not yet demonstrated, the omega plot approach may provide useful information for intracellular pH, but further technical advancements are needed to translate it in vivo.

### Extracellular Tumor pH Imaging

To overcome the limitations of endogenous CEST-MRI techniques, exogenous molecules have been exploited as extracellular tumor pH reporters for CEST-MRI applications. In the last decade, great expectations surrounded the class of the X-ray FDA-approved iodinated contrast media, considering their high safety profile and translational potential (88). Due to their hydrophilic chemical structure, iodinated agents remain confined outside the cells and can be visualized as perfusion agents in tumor by CEST-MRI (89, 90). Their first application as pH CEST-MRI agents involved the use of iopamidol (Isovue <sup>R</sup> , Bracco Diagnostic), possessing two amide proton pools that can be saturated at 4.2 and 5.5 ppm (91, 92). The set-up of a ratiometric procedure allows to accurately measure extracellular tissue pH (pHe) in the pH range of 5.5–7.9, independently of the contrast agent concentration, with an accuracy of 0.1 units at several magnetic fields (93–95). CEST-MRI tumor pH imaging was combined to FDG-PET to elucidate the deregulation of tumor metabolism in a breast cancer model (96). This work evidenced that tumor regions with more acidic pHe show increased FDG uptake and demonstrated in vivo, for the first time, the relationship between tumor acidosis and high glycolytic rate. In addition, it provided evidence of the feasibility of measuring tumor pH heterogeneity at the clinical field of 3T (**Figures 2A,B**). The combination of CEST pH-imaging and FDG-PET was then exploited for predicting the early therapeutic efficacy of metformin in a preclinical model of pancreatic cancer (98). In addition, the possibility to measure tumor pHe opened new routes for monitoring the effect of novel anticancer treatments that can reverse the glycolytic tumor phenotype (97). Anemone et al. showed that this approach can monitor early pH changes in a breast murine cancer model upon the treatment with dichloroacetate, a small compound targeting mitochondria, and that can be exploited to detect the onset of the resistance, hence providing useful insights about the therapeutic efficacy (**Figures 2C,D**).

Another iodinated agent used for pH mapping is iopromide (Ultravist <sup>R</sup> , Bayer Healthcare), that has two amide pools resonating at 4.6 and 5.6 ppm that can be exploited to measure tumor pH within the 6.5–7.2 range (99). CEST-MRI with iopromide revealed that breast cancer models with different histopathological features show significant differences in pHe values and that tumor acidosis is associated with metabolic biomarkers in B-lymphoma xenografts (100, 101). In addition, a comparative study between iopromide and iopamidol showed that although these agents measured similar pH values in vivo, iopamidol reveals more accurate pH measurement (102).

One of the main advantages of this class of agents relies in their very high safety profile for administration in patients. Consequently, CEST-MRI pH imaging with iopamidol was initially translated for measuring kidney and bladder pH in healthy volunteers (103–105). Later on, the capability to provide accurate tumor pH maps was demonstrated with iopamidol in both breast and ovarian cancer patients showing acidic tumor pH values (106). These preliminary results pointed out that efficient translation still requires optimization of several aspects, including acquisition protocol and data analysis to further evaluate the diagnostic and therapeutic utility of tumor pH mapping in the clinical setting. To this purpose, different studies aimed to optimize RF irradiation, reduce respiration artifacts and enlarge the body coverage acquisition have been performed (107–109). In addition, new ratiometric approaches have been formulated to extend the use of iodinated agents even with a single resonating protons for pH measurements (110, 111). Promising results have been obtained with iobitridol (Xenetix <sup>R</sup> , Guerbet), showing accurate pH measurement in murine tumors once irradiated with different power levels (112).

PARACEST pH-responsive agents are characterized by a large chemical shift of the mobile protons from the water peak that should improve their detectability in comparison to DIACEST molecules, as iodinated agents or glucose (23, 113). The Yb-HPDO3A contrast agent has been exploited for measuring tumor pHe in both melanoma and in glioma murine models (114, 115). Interestingly, in the melanoma model changes in tumor pHe were observed and correlated with the tumor progression stage. Similar approaches based on other PARACEST agents allowed to measure tumor pHe in rat brain tumor models,

although direct injection of the contrast agent in the tumor and renal ligation were needed to maintain high concentrations of the agent for measuring pH (116–118). Currently, the high saturation power needed to generate enough CEST contrast limits a wider applicability of these pH responsive PARACEST agents, however molecules with optimal exchange rates have been recently proposed (119).

### IMAGING LACTATE

The preferential ATP production via glycolysis of glucose to lactate leads to high lactate levels that some cancer cells can even exploit as a metabolic fuel (120–122). Conventionally, lactate can be observed and quantified by Magnetic Resonance Spectroscopy (MRS) or by the recently developed hyperpolarization technique (123–129). However, these methods are limited by low spatial resolution and long acquisition times. The chemical shift of the hydroxylic proton of the lactate is very close to the water signal and renders quite difficult to directly detect lactate in vivo by CEST imaging. However, correlation of the signal arising from lactate between CEST and MR spectroscopy has been performed in a lymphoma murine tumor upon lactate infusion (130) or in a mitochondrial disease model (131). Other approaches exploited lactate-responsive PARACEST contrast agents for taking advantage of the larger chemical shift difference of these molecules and the CEST contrast dependence with lactate concentration (132, 133). Zhang et al. (134) demonstrated the feasibility of this approach by measuring lactate excreted from lung cancer cells in tissue culture.

### CONCLUSION AND FUTURE PERSPECTIVES

In summary, CEST-MRI imaging is a fast-expanding field with enormous potential to assess several aspects of tumor metabolism. Moreover, since tumor acidosis is a general feature in all tumors, imaging tumor pH might become a powerful and wide tool for oncological imaging at both preclinical and clinical level. First studies in patients demonstrated the feasibility of these novel imaging approaches for imaging human tumors. Further improvements in fast acquisition sequences, post-processing and standardization set-up are mandatory for the widespread use of CEST-MRI in the clinical settings. Despite the fundamental insights that imaging tumor acidosis with iopamidol can provide, additional studies are needed to validate it in comparison to established clinical approaches and to demonstrate that it can be exploited for monitoring treatment response to (novel) anticancer therapies.

### AUTHOR CONTRIBUTIONS

DL and LC conceived, structured, and edited the mini review article. DL, LC, AA, MC, ACa, PI, CD, ACo, and MB each wrote individual sections of the mini review article and critically revised it for intellectual content. All authors provided final approval of the version of the article submitted for publication.

### REFERENCES


### FUNDING

We gratefully acknowledge the support of the Associazione Italiana Ricerca Cancro (AIRC MFAG #20153 to DL) and Compagnia San Paolo project (Regione Piemonte, grant #CSTO165925) and from the European Union's Horizon 2020 research and innovation programme (Grant Agreement No. 667510) funding. LC was supported by the AIRC fellowship for abroad Monica Broggi. The Italian Ministry for Education and Research (MIUR) is gratefully acknowledged for yearly FOE funding to the Euro-BioImaging Multi-Modal Molecular Imaging Italian Node (MMMI).


CEST-MRI–application to pH-weighted MRI of acute stroke. NMR Biomed. (2014) 27:240–52. doi: 10.1002/nbm.3054


enhanced MRI at ultrahigh fields. Magn Reson Med. (2017) 78:215–25. doi: 10.1002/mrm.26370


contrast agent for pH mapping of kidneys: in vivo studies in mice at 7 T. Magn Reson Med. (2011) 65:202–11. doi: 10.1002/mrm.22608


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

Copyright © 2020 Consolino, Anemone, Capozza, Carella, Irrera, Corrado, Dhakan, Bracesco and Longo. 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.

# Cross-Talk Between the Tumor Microenvironment, Extracellular Matrix, and Cell Metabolism in Cancer

### Mona Nazemi and Elena Rainero\*

*Biomedical Science Department, The University of Sheffield, Sheffield, United Kingdom*

The extracellular matrix (ECM) is a complex network of secreted proteins which provides support for tissues and organs. Additionally, the ECM controls a plethora of cell functions, including cell polarity, migration, proliferation, and oncogenic transformation. One of the hallmarks of cancer is altered cell metabolism, which is currently being exploited to develop anti-cancer therapies. Several pieces of evidence indicate that the tumor microenvironment and the ECM impinge on tumor cell metabolism. Therefore, it is essential to understand the contribution of the complex 3D microenvironment in controlling metabolic plasticity and responsiveness to therapies targeting cell metabolism. In this mini-review, we will describe how the tumor microenvironment and cancer-associated fibroblasts dictate cancer cell metabolism, resulting in increased tumor progression. Moreover, we will define the cross-talk between nutrient signaling and the trafficking of the ECM receptors of the integrin family. Finally, we will present recent data highlighting the contribution of nutrient scavenging from the microenvironment to support cancer cells growth under nutrient starvation conditions.

#### Edited by:

*Paolo E. Porporato, University of Turin, Italy*

#### Reviewed by:

*Stefano Indraccolo, Istituto Oncologico Veneto (IRCCS), Italy Krishna Beer Singh, University of Pittsburgh, United States*

> \*Correspondence: *Elena Rainero e.rainero@sheffield.ac.uk*

#### Specialty section:

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

Received: *22 November 2019* Accepted: *12 February 2020* Published: *26 February 2020*

#### Citation:

*Nazemi M and Rainero E (2020) Cross-Talk Between the Tumor Microenvironment, Extracellular Matrix, and Cell Metabolism in Cancer. Front. Oncol. 10:239. doi: 10.3389/fonc.2020.00239* Keywords: extracellular matrix, cell metabolism, cancer associated fibroblasts, nutrient scavenging, nutrient signaling

### INTRODUCTION

Cancer-associated fibroblasts (CAFs) are a heterogenous and plastic population of activated fibroblasts, which represent a significant proportion of the tumor microenvironment. For instance, CAFs can account up to 70 and 90% of breast and pancreatic cancer tumor mass, respectively (1, 2). The role of CAFs in tumorigenesis is widely established and they have been shown to contribute to tumor growth, metastasis and resistance to therapy (3). Moreover, CAFs play an important role in the regulation of cancer metabolism, primarily through the secretion of metabolites and the generation of a stiffer and fibrotic ECM, which in turn affects cancer cell metabolism.

Cells interact with the ECM through plasma membrane receptors. Integrins are transmembrane receptors that regulate cell adhesion, migration, and mechanotransduction through mediating cell-ECM interaction (4, 5). Integrins can trigger different intracellular signaling pathways promoting cell growth, survival, and proliferation (6). Furthermore, ligand bound-integrin trafficking has recently been shown to directly or indirectly affect nutrient signaling (7, 8). Mechanistic target of Rapamycin (mTOR) signaling pathway is the key regulator of anabolic and catabolic processes of the cells. During nutrient availability, mTOR induces anabolic processes such as protein, nucleotide, and lipid biosynthesis and inhibits cellular autophagy and lysosomal biogenesis (9–11). mTOR forms two independent complexes, mTORC1 and mTORC2. mTORC1 adjusts cell growth and proliferation in response to growth factors and amino acids, while mTORC2 has a role in actin organization. Furthermore, mTORC2 can control cell proliferation and survival through AKT activation downstream of growth factor signaling (12). During nutrient starvation, the activity of mTORC1 is restrained, allowing the cells use other sources of nutrient acquisition, such as autophagy (13).

In cancer cells, the activation of oncogenic signaling has a profound effect on cell metabolism. Indeed, aberrant activation of phosphatidyl inositol 3 kinase (PI3K)/AKT and Ras signaling pathways facilitates constant glucose uptake through GLUT1 receptor (14, 15). In parallel, derailed function of the oncogenic protein c-myc induces the expression of glutamine transporters and glutamine-utilizing enzymes (16, 17). In addition, high rate of tumor growth and insufficient vasculature forming deep inside the tumors significantly increase the chance of nutrient scarcity in the tumor microenvironment (TME) (18). Indeed, the TME has been shown to be depleted of amino acids and glucose (19, 20).

In this mini-review, we will highlight how CAFs control cancer cell metabolism, through metabolite secretion and the generation of a stiffer microenvironment. We will then summarize the cross-talk between integrin trafficking, metabolism and nutrient signaling and we will describe the different mechanisms through which cancer cells can exploit unconventional nutrient sources.

### THE TUMOR MICROENVIRONMENT AND CAFs DICTATE CHANGES IN CANCER CELL METABOLISM

The activation of fibroblasts to CAFs is coupled with a metabolic shift to glycolysis, increased catabolic activity and autophagy (21). As a result of this, CAFs secrete a variety of metabolites that can support cancer cell growth and metabolism.

In prostate cancer, CAFs promote the so-called reverse Warburg effect, whereby lactate secretion and mitochondria transfer from CAFs to cancer cells (CCs) lead to an increase in mitochondrial activity and oxidative phosphorylation (OXPHOS) in CCs (**Figure 1A**). This promotes epithelial-to-mesenchymal transition, metastatic burden, and chemotherapy resistance (22). Similarly, the secretion of pyruvate by CAFs has been shown to promote lymphoma cell survival through the upregulation of the tricarboxylic acid (TCA) cycle (23) (**Figure 1B**), as well as ECM remodeling and lung metastasis in breast cancer, through the upregulation of alpha ketoglutarate production (24).

In ovarian cancer, CAFs have been shown to promote glycogen metabolism and glycolysis in CCs, through the production of cytokines, which is in turn promoted by CCderived TGFβ. This CAFs-CC signaling loop results in increase CC proliferation, invasion and metastasis (25) (**Figure 1C**). It is important to note that all the afore-mentioned studies were performed using conditioned media from CAFs or shortterm co-culture experiments (a few hours). Despite providing useful insight in the contribution of secreted factors, these fail to recapitulate the complex 3D structure of the TME, whereby changes in ECM composition and properties could play a role in the metabolic reprogramming observed in these studies. Indeed, the stiffness of the TME has been shown to have profound effects on both CAF and CC metabolism in squamous cell carcinoma, leading to an increase in glycolysis and glutamine consumption in both cell populations. In particular, the activation of the YAP/TAZ signaling pathway in stiffer environments promotes the expression of glycolytic enzymes (26), as well as glutamine uptake and conversion to glutamate in CCs. Glutamate is secreted by CCs and utilized by CAFs for maintaining redox homeostasis through the glutathione pathway. CAFs, on the other hand, secrete aspartate, which is used by CCs to promote nucleotide synthesis (27) (**Figure 1D**). Similar results have been observed in vivo in an orthotopic mouse model of breast cancer, where ECM stiffness is coupled with increased glycolysis, glutamine metabolism and aspartate production. Importantly, a reduction in intratumoral levels of aspartate and glutamine leads to a decrease in cell proliferation (27). Interestingly, glutamate secretion by CCs is also involved in the invasive phenotype of breast cancer cells. Indeed, by binding to its receptor GRM3, glutamate promotes the trafficking of the matrix metalloprotease MT1-MMP, leading to CC invasive migration (28).

CAFs have a pivotal role in determining the composition and organization of the ECM within the TME, although other cell types have been shown to contribute to ECM secretion and deposition. Increased collagen density is observed during carcinoma progression and has been shown to be mainly produced by CAFs (29). Importantly, changes in collagen density have been linked with alterations in breast cancer cell metabolism. High density collagen gels are associated with a reduction of oxygen consumption and the amount of glucose that is metabolized through the TCA cycle, with a concomitant increase in the use of glutamine to fuel the TCA cycle. This is associated with a reduction in the expression of glycolysis genes and an increase in oxidative glutamine metabolism, serine synthesis and one carbon metabolism enzymes (30) (**Figure 1E**). This is in contrast with the data presented above by Bertero et al. (27). This could be due to differences in the type of cells analyzed (ovarian cancer vs. breast cancer), or the experimental settings (hydrogels with different stiffnesses vs. collagen I gels with different densities). Moreover, independent sets of metabolic enzymes where shown to be modulated in the two studies, suggesting that separate mechanotransduction pathways could impinge on cell metabolism. Within the TME, the ECM is constantly remodeled through the action of matrix degrading enzymes. Recent data indicate that the degradation of hyaluronan, a ubiquitous ECM component, promotes glucose uptake in several cancer cell lines, by increasing the plasma membrane localization of the glucose transporter GLUT1. This in turns fosters glycolysis and promotes cancer cell migration (31). It is unclear whether the degradation of other ECM components is also linked to changes in cell metabolism. Moreover, future work will be needed to characterize how ECM degradation in more complex 3D environments impinges on CC metabolism.

### THE CROSSTALK BETWEEN CELL METABOLISM AND THE TRAFFICKING AND FUNCTION OF THE ECM RECEPTORS OF THE INTEGRIN FAMILY

mTOR is one of the signaling pathways that has been shown to be regulated by integrin trafficking. It has been demonstrated that in ovarian cancer cells, glucose starvation induces the translocation of α5β1 integrin from peripheral focal adhesions to a centrally located patch of fibrillar adhesions. This has been identified as the main internalization site for fibronectin (FN) bound α5β1, resulting in tensin and Arf4-dependent endocytosis and lysosomal delivery of FN-bound α5β1. This internalization pathway leads to mTORC1 lysosomal recruitment and activation (**Figure 2A**). Mimicking nutrient depleted environments by inhibiting mTORC1 activity leads to higher ligand-bound α5β1 integrin endocytosis and tensin-dependent sub-nuclear adhesion formation, showing that tensin-dependent α5β1 internalization is controlled by nutrient availability (7). Serum and growth factor starvation also induce normal mammary epithelial cells to internalize soluble laminin through β4 integrin; this mediates an increase in the cellular amino acid content through laminin lysosomal degradation and amino acid extraction, which in turn returns mTORC1 activity to its normal condition to avoid excess uptake of extracellular proteins (**Figure 2A**). In vivo studies also reveal that mammary epithelial cells in dietary restricted mice increase laminin uptake from the basement membrane and fibroblast ECM secretion (8). Altogether, these data indicate that the uptake of ECM components, regulated by integrin trafficking under nutrient deficient conditions, could provide a source of nutrients for cancer cells. However, more studies are required to investigate whether different integrin heterodimers are involved in the internalization of other ECM components and whether this is required for nutrient signaling as well.

AMP-activated protein kinase (AMPK) is a serine/threonine kinase that also works as a sensor for the availability of nutrients. Lack of energy and cell exposure to certain levels of stress activate AMPK, which inhibits anabolic pathways and induces catabolic pathways to provide more energy or source of nutrition for the cells (32). It has been revealed that AMPK can inhibit the activation of α5β1 integrins in fibroblasts. In particular, AMPK reduces tensin3 expression, thus preventing integrin activation (**Figure 2A**). In AMPK KO mouse embryonic fibroblasts (MEFs), the activity of α5β1 is increased, leading to more fibrillar adhesion formation and fibronectin biogenesis (33). More evidence is required to evaluate the effect of higher fibronectin biogenesis on nutrient signaling/availability for cells, since previous studies highlighted that active integrins in fibrillar adhesion would lead to higher fibronectin endocytosis under starvation (7). In addition, since cancer cells are more prone to face nutrient scarcity, more investigations are required to elucidate the effect of AMPK activity in those cells. Even though studies on mTORC1 activity suggest that nutrient deficiency assist cells to benefit from alternative food sources such as fibronectin and laminin through receptor dependent (integrin)-endocytosis, this study shows that starvation could also have an inhibitory effect on integrin membrane transport and trafficking. As it has been reviewed by Georgiadou and Ivaska (34), AMPK activity in cancer cells and cancer associated fibroblast cells (CAFs) lead to a cross-talk between them. Low activity of AMPK in CAFs induces

tensin expression leading to upregulation of ECM secretion and formation of fibrillar adhesion. In contrast, high AMPK activity in cancer cells under starvation inhibits mTORC1 activity which results in ECM internalization and lysosomal degradation to provide nutrient for the cells.

In addition to signaling pathways that directly deal with nutrient availability around the cells, a link between integrin mechanosensing and cell metabolism has also been proposed. CD98hc is an integrin co-receptor and amino acid transporter sensing mechanical forces and causing cell stiffening through integrin and RhoA activation. It has been demonstrated that there is a cross-talk between mechanosensing interruption and sphingolipid synthesis in fibroblasts bearing C330S mutation in CD98hc. Sphingolipids are essential membrane components that can regulate transmembrane protein dynamics. Loss of CD98hc reduces sphingolipid availability, leading to defects in movement, recruitment and activation of regulatory proteins upstream of RhoA (**Figure 2A**). Therefore, it is suggested that there is a crosstalk between sensing the mechanical forces that cells are exposed to and their metabolic state. This suggests that cancer cells could benefit from different environmental mechanical forces through metabolic adaptation. Indeed, it has been demonstrated that circulating tumor cells display an oxidative switch, while the cells in the primary tumor are mostly glycolytic (35). An intriguing explanation for this phenotype could be the loss of mechanical forces present in the primary tumor. Further studies are needed to address this point, as invasive cancer cells experience different mechanical forces during their migration (36).

### STRATEGIES UNDERTAKEN BY CANCER CELLS TO SCAVENGE NUTRIENTS FROM THE MICROENVIRONMENT

Even though blood pressure inside the tumors is low due to high interstitial pressure, lymphatic deficiency and leaky blood vessels increase the accessibility of cancer cells to the blood serum or its main protein, albumin (37, 38). Rasdriven cancer cells have a higher rate of macropinocytosis which helps them to internalize extracellular proteins (ECPs). In Ras-driven pancreatic ductal adenocarcinoma cells (PDACs) and Ras-transformed MEFs, amino acid starvation induces albumin macropinocytosis followed by lysosomal degradation and amino acid extraction (**Figure 2B**). It has also been revealed that mimicking nutrient starvation by inhibiting mTORC1 signaling pathway induces cells to rely on extracellular macromolecules rather than amino acids (19, 39, 40). Lack of nutrient and oxygen delivery leads to cell death at the center of tumors. In PTEN-deficient prostate cancer cell lines and KRas-driven pancreatic cancer cells, AMPK activation and mTORC1 suppression under amino acid and glucose starvation assist cell proliferation through inducing cell debris scavenging. It was revealed that amino acids extracted from cell debris can participate in building cell biomass (41). In addition to the access to albumin and cell debris, PDAC cells are also surrounded by a dense network of ECM containing collagen I and collagen IV. PDAC cells are able to internalize collagen I and IV through macropinocytosis under glucose starvation and receptor-dependent endocytosis under low glutamine conditions (**Figure 2B**). PDAC cells degrade the internalized collagens in the lysosomes, providing a source of proline. Proline is then fed into the TCA cycle, leading to ATP production and cell survival, via the activation of the ERK1/2 pathway. Contrary to albumin, collagen uptake in this context does not affect mTORC1 signaling pathway, and the role of mTOR inhibition has not been addressed (42). It is not clear whether other amino acids from the collagen also contribute to cancer cell survival. Even though the role of Ras has been highlighted in terms of inducing macropinocytosis under starvation conditions, its role could be dispensable in the presence of growth factors. Indeed, it has been demonstrated that MEFs are be able to induce ECP uptake through growth-factor (GF)

dependent macropinocytosis, in a PI3 kinase-dependent manner (**Figure 2B**). Under amino acid starvation, activation of Rac1 and PLCγ as an PI3K effector induces ECP macropinocytosis, while availability of glucose and amino acids induces AKT-dependent activation of mTORC1 signaling. This allows cells to use free amino acid from transporters instead of macropinocytosis to uptake macromolecules (43). PDAC cells are heterogenous in terms of macropinocytosis potential. A subset of PDAC tumors have been shown to upregulate macropinocytosis under glutamine starvation conditions, while other tumors display constitutive macropinocytosis which is not affected by the level of nutrients. Mechanistically, glutamine starvation potentiates EGFR signaling, leading to the activation of the Ras/Pak signaling pathway. This in turn induces ECP macropinocytosis, followed by lysosomal degradation (44). Along with in vitro studies showing that cancer cells can use albumin as an alternative source of nutrients, hypoalbuminemia was also seen in some cancer patients (45). An in vivo study directly showed that PDAC cells inside the tumor are able to internalize albumin through macropinocytosis and use albumin derived amino acids for further metabolic pathways, while adjacent normal cells do not have this ability (46). In addition to taking advantage of unusual protein sources, cancer cells are able to use extracellular lipids for their survival under hypoxia or starvation conditions. Cancer cells need lipid and fatty acids to reproduce their membrane and proliferate. In normal oxygen conditions (normoxia), cancer cells can produce non-essential fatty acids. In contrast, hypoxic or Ras- driven cancer cells rely on scavenging fatty acids from TME. Ras-driven cancer cells can internalize serum lipids with one fatty acid tail (lysophospholipid) to compensate the lack of lipid production (47). Under glucose and amino acid starvation lipid droplet (LP) content of cancer cells declines. PTEN-deficient prostate cancer cells rely on lipids extracted from cell debris through macropinocytosis to replenish their LP storage containing fatty acids and cholesterol (41). Invasive breast cancer cells also show higher proliferation and migration in co-culture with obese adipocytes. Transferring of fatty acids from adipocytes to invasive breast cancer cells induces adipose triglyceride lipase (ATGL)-mediated lipolysis and oxidation of fatty acids in mitochondria (48).

Since cancer cells could face different nutrient conditions in various niches during their development, further work is needed to elucidate how ECP and ECM could affect nutrient availability and how Ras and PI3K pathways would interact and affect each other in response to different nutrient conditions.

### REFERENCES


### CONCLUSIONS AND FUTURE DIRECTIONS

Cancer cells display an elevated metabolic plasticity, allowing them to adapt to different environments and nutrient levels. The TME, in particular CAFs and cell-ECM interaction, are key in controlling this metabolic switch. CAFs secrete several metabolites, including lactate, pyruvate and aspartate, which have been shown to support the growth of different cancer cell types (3). Moreover, through the generation of a stiffer ECM, CAFs drive changes in cancer cell metabolism, leading to an increase in one-carbon metabolism and oxidative glutamine metabolism in the TCA cycle. Furthermore, ECM components, such as fibronectin and laminin, have been shown to be internalized in an integrin-dependent manner and to control nutrient signaling. On the other hand, it is now clear that nutrient-dependent signaling pathways can control integrin trafficking and activation in diverse ways. However, more work needs to be done to elucidate how nutrient starvation impinges on vesicular trafficking and receptor internalization in complex 3D environments.

Elevated rates of tumor growth and limited blood supply result in a nutrient-deprived TME. These pushes cancer cells to adopt different strategies to obtain metabolic fuels. These include the internalization of ECP, ECM and cell debris. Despite all these processes have been observed, the cross-talk between different nutrient scavenging strategies and how cells can switch between them have not been fully addressed yet. Importantly, it is not known how changes in the TME, including immune cell infiltration and ECM composition and structure, affect the ability of cancer cells to grow under nutrient deprived conditions.

Altogether, understanding how cancer cells can thrive within the challenging TME will lead to the identification of cancer vulnerabilities. These could then be exploited for the development of novel strategies for targeting unresponsive tumors.

### AUTHOR CONTRIBUTIONS

MN and ER wrote this manuscript and generated the figures.

### FUNDING

MN was supported by the Faculty of Science, The University of Sheffield. ER was supported by the Academy of Medical Sciences, Wellcome Trust and British Heart Foundation Springboard Award [SBF003/1045].

tumor cells promote MCL-1 dependency in estrogen receptor-positive breast cancers. Oncogene. (2019) 38:3261–73. doi: 10.1038/s41388-018- 0635-z


derived fatty acids drive breast cancer cell proliferation and migration. Cancer Metab. (2017) 5:1. doi: 10.1186/s40170-016-0163-7

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

Copyright © 2020 Nazemi and Rainero. 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 in Chemotherapy Resistance

Maria Andrea Desbats 1,2, Isabella Giacomini <sup>3</sup> , Tommaso Prayer-Galetti <sup>4</sup> and Monica Montopoli 2,3 \*

*<sup>1</sup> Department of Medicine, University of Padova, Padova, Italy, <sup>2</sup> Veneto Institute of Molecular Medicine, Padova, Italy, <sup>3</sup> Department of Pharmaceutical and Pharmacological Sciences, University of Padova, Padova, Italy, <sup>4</sup> U.O.C. Urology, Azienda Ospedaliera di Padova, Padova, Italy*

Resistance of cancer cells to chemotherapy is the first cause of cancer-associated death. Thus, new strategies to deal with the evasion of drug response and to improve clinical outcomes are needed. Genetic and epigenetic mechanisms associated with uncontrolled cell growth result in metabolism reprogramming. Cancer cells enhance anabolic pathways and acquire the ability to use different carbon sources besides glucose. An oxygen and nutrient-poor tumor microenvironment determines metabolic interactions among normal cells, cancer cells and the immune system giving rise to metabolically heterogeneous tumors which will partially respond to metabolic therapy. Here we go into the best-known cancer metabolic profiles and discuss several studies that reported tumors sensitization to chemotherapy by modulating metabolic pathways. Uncovering metabolic dependencies across different chemotherapy treatments could help to rationalize the use of metabolic modulators to overcome therapy resistance.

#### Edited by:

*Paolo E. Porporato, University of Turin, Italy*

#### Reviewed by:

*Elisa Giannoni, University of Florence, Italy Flora Guerra, University of Salento, Italy*

#### \*Correspondence:

*Monica Montopoli monica.montopoli@unipd.it*

#### Specialty section:

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

Received: *31 October 2019* Accepted: *18 February 2020* Published: *06 March 2020*

#### Citation:

*Desbats MA, Giacomini I, Prayer-Galetti T and Montopoli M (2020) Metabolic Plasticity in Chemotherapy Resistance. Front. Oncol. 10:281. doi: 10.3389/fonc.2020.00281*

Keywords: cancer, metabolic reprogramming, TCA cycle, Warburg effect, metabolic vulnerabilities, chemoresistance

### METABOLIC REPROGRAMMING

The metabolic program between non-proliferating and proliferating cells is different. Nonproliferating cells rely mostly on catabolic reactions while proliferating cells must balance catabolic and anabolic reactions required to sustain enhanced cellular growth (1–3). In normally proliferating cells most **ATP** from glucose is obtained by glycolisis, tricarboxylic acid cycle (**TCA**) and oxidative phosphorylation (**OxPhos)**, while nucleotides, aminoacids, and lipids are provided by intermediate metabolites of these pathways; such as acetyl-CoA for fatty acids, glycolytic intermediates for non-essential aminoacids, and ribose for nucleotides. Tumor cells are characterized by metabolic hallmarks similar to highly proliferating normal cells but, in addition, they develop a high plasticity to metabolic rewiring to sustain enhanced cellular growth in changing microenvironmental conditions (4).

Back in the 1920's, Otto Warburg observed that many tumors depended on glycolysis as the sole source of ATP; even in the presence of oxygen (aerobic glycolysis) (5). Accordingly, the rate of glucose entry to cancer cells was found 20-to-30- fold higher than in normal cells (6), and glucose transporters and key glycolytic enzymes were heavily upregulated (7). Cancer cells under hypoxia induce pyruvate dehydrogenase kinase (**PDK**) that inactivates pyruvate dehydrogenase (**PDH**) (8). Thus, most glucose-derived pyruvate does not enter the TCA cycle and is converted in lactate by the action of lactate dehydrogenase (**LDH**) (9). This is because most tumors produce great quantities of lactate, which is very striking, since glycolysis produces only 2 ATP molecules for each glucose, while oxidative phosphorylation between 30 and 32 ATPs.

**70**

Later it became clear that in cancer cells glucose is consumed mainly to supply glycolitic intermediates for anabolic pathways. Glucose-6-phosphate can be oxidized by glucose-6-phosphate dehydrogenase (**G6PD**) to produce reduced nicotinamide adenine dinucleotide phosphate (**NADPH**) and ribose-5 phosphate (**R5P**) through the pentose phosphate pathway (**PPP**). **NADPH** and **R5P** are required for nucleotide synthesis, but also to sustain biosynthetic reactions and to maintain the redox capacity of the cell (1). Moreover, 3-phosphoglycerate could serve as a precursor for serine and glycine metabolism through the one-carbon cycle (10). Pyruvate instead that can be converted into alanine by alanine aminotransferase (**ALT**) (11). In turn, these aminoacids can be metabolized for nucleotide synthesis, DNA methylation, glutathione production and **NADPH** generation (12). Interestingly, several **PPP** enzymes and 3-phosphoglycerate dehydrogenase (**PHGDH**) were found upregulated in some cancer (13–16).

Unlike originally thought, aerobic glycolysis in cancer cells is not a sign of defective oxidative phosphorylation. Instead, high rates of glycolysis inhibit mitochondrial respiration, a phenomenon termed the "Crabtree effect" (17). Indeed, mitochondria function is essential for cancer cell proliferation (18). Mitochondrial redox homeostasis is crucial for maintaining cellular aspartate levels critical for nucleotide synthesis (19). Indeed, aspartate was shown essential for in vivo tumor growth (20).

Of great significance, cancer cells require **TCA** cycle intermediates for biosynthetic pathways and **NADPH** production (21). The **TCA** cycle generates citrate that can be exported to the cytosol through the mitochondrial tricarboxylate carrier (**SLC25A1**) to be converted into acetyl-CoA and oxaloacetate by ATP citrate lyase (**ACLY**). (22). Acetyl-CoA can either be employed for fatty acid and cholesterol synthesis (to support membrane biogenesis) or used for protein acetylation reactions, which regulate nuclear transcription as well as cytoplasmic processes like autophagy (23). The **TCA** cycle also provides metabolic precursors for the synthesis of non-essential amino acids, such as aspartate and asparagine from oxaloacetate, or proline, arginine and glutamate from α-ketoglutarate. To cope with the continuous efflux of intermediates cancer cells replenish the **TCA** cycle by increasing or developing the ability to use various carbon sources; including glutamine, acetate, lactate, serine, and glycine (24–27). In particular, tumor cells consume great quantities of aminoacids.

Glutamine is the major contributor of **TCA** intermediates in many cancer cell lines (28). Glutamine is transported into the cell through plasma membrane transporters, like **SLC1A5** (**ASCT2**) and **SLC7A5** (29) and converted into glutamate by glutaminase (**GLS**). Then glutamate is transformed into α-ketoglutarate, by either glutamate dehydrogenase (**GDH**) or transaminases; and αketoglutarate enters the **TCA** cycle to maintain the production of citrate. Glutamine can also be directly converted into citrate by reductive carboxylation. The reductive carboxylation of αketoglutarate by the inverse reaction of isocitrate dehydrogenase (**IDH**) generates citrate (30). Glutamine reductive carboxylation is particularly important in tumor cells under hypoxic conditions or when mitochondrial respiration is impaired (31). Moreover, **GLS** and **GDH** are upregulated in a wide variety of tumors and its inhibition has been shown to diminish tumorigenesis (32, 33).

Another contributor of **TCA** intermediates is lactate. Some cancer cells can use lactate produced by aerobic glycolysis as a source of energy. More than 50% of the total **TCA** cycle intermediates in breast cancer cells under glucose deprivation derived from lactate (34). Moreover, overexpression of lactate transporters (**MCTs**) is a common finding in some cancers (35).

Enhanced glycolisis and glutamine metabolism in cancer cells support the increase of de novo fatty acids synthesis (36). Fastproliferating cancer cells use fatty acids and cholesterol for biosynthesis of cell membranes, cell signaling and secondary messengers (37), as well as for lipid catabolism through fatty acid β-oxidation (**FAO**) during nutrient deprivation (38). In some cancers such us prostate cancer and lymphoma, lipiddependent metabolism becomes essential for energy production (39). In physiological conditions, lipid synthesis is restricted to specialized tissues, such as the liver and adipose tissues. Normal cells uptake lipids from the bloodstream, while cancer cells could obtain lipids and lipoproteins exogenously or by de novo synthesis (38). A wide variety of tumors have increased expression of crucial lipogenic enzymes such us **ACLY**, acetyl-CoA-carboxylase (**ACC**), fatty acid synthase (**FASN**) (38, 40, 41); as well as present an increase in the transcriptional activities of the sterol regulatory element-binding proteins (**SREBPs**) (42, 43). The upregulation of lipogenic enzymes seems required for tumor progression (40). Interesstingly, some cancer cells harbor adipocyte characteristics like storing excess lipids in lipid droplets (**LD**) (44). **LD** are intracellular storage organelles of neutral lipids mainly found in adipose tissue, but observed in several cell types and tissues (45, 46). **LDs** are dynamic, and their accumulation seem to confer survival advantages to cancer cells (47). Drugs that specifically target **LD** formation are thought to hold greater therapeutic potential compared with general lipid biosynthesis inhibitors (48, 49).

Enhanced glycolisis, glutamine metabolism and fatty acids synthesis are features shared by many cancer cell lines. However, the metabolic phenotype of the tumor in vivo is highly heterogeneous, resulting from the combination of intrinsic (genetic and epigenetic changes, tissue of origin, state of differentiation) and extrinsic (oxygen and nutrient availability, metabolic interactions within the tumor microenvironment) factors (50).

### ROLE OF ONCOGENES AND TUMOR SUPPRESSOR GENES IN METABOLISM REPROGRAMMING

One of the intrinsic factors that determine the tumor metabolic phenotype is the activation of oncogenes or deactivation of tumor suppressor genes which result in a metabolic rewiring (51). Tumor metabolism is distinct in tumors harboring different oncogenic alterations. Oncogenes such as RAS, MYC, or PI3K, favor glycolysis over oxidative phosphorylation; whereas tumor suppressors such as p53, PTEN, Von Hippel–Lindau (VHL), or liver kinase B1 (LKB1) have the opposite effect (52).

In particular, MYC expression could activate the pentose phosphate pathway, purine/pyrimidine synthesis and fatty acid oxidation under chemotherapy and radiotherapy (53). MYC directly regulates several glycolytic enzymes such as glucose transporter 1 (**GLUT1**), hexokinase 2 (**HK2**), phosphofructokinase muscle type (**PFKM**), and enolase 1 (**ENO1**); as well as lactate dehydrogenase A (**LDHA**) (54). Moreover, MYC expression increases glutamine uptake and glutaminolysis (55, 56) by inducing the expression of glutamine transporters **SLC1A5** and **SLC7A5** and by repressing the transcription of microRNA-23a/b which targets glutaminase 1 (**GLS1**) (56, 57).

p53 can directly or indirectly influence the expression of genes involved in glucose, **OxPhos** and lipid metabolism, among other pathways (58). p53 inhibits glycolysis dowregulating **GLUTs** and the glycolytic enzyme phosphoglycerate mutase (**PGAM**) (59, 60). p53 also induce the expression of TIGAR (TP53 induced glycolysis and apoptosis regulator), which indirectly inhibits phosphofructose kinase 1 (**PFK1**) diverting glycolytic intermediates into the **PPP** (61). p53 decreases fatty acid synthesis by also inhibiting the **PPP**. p53 directly binds and inhibits **G6PD** leading to decreased production of **NADPH** (14). Moreover, p53 directly repress the expression of SREBP-1 which regulates the expression of fatty acid synthesis enzymes (62). On the other hand p53 enhances fatty acid oxidation. p53 induces two important enzymes involved in fatty acid oxidation, Lipin 1 and carnitine palmitoyltransferase (**CPT1C**) (63, 64). p53 was also reported to transcriptionally induce malonyl-CoA decarboxylase (**MCD**), which catalyzes the conversion of malonyl-CoA to acetyl-CoA, to promote fatty acid oxidation and prevent lipid accumulation (65). p53 enhances mitochondrial **OxPhos** by inducing the expression of the cytochrome c oxidase (COX, complex IV) assembly factor SCO2 (66) or the expression of AIF; which maintains the integrity of mitochondrial NADH:ubiquinone oxidoreductase (complex I) (67). In addition, p53 promotes **OxPhos** by repressing the transcription of pyruvate dehydrogenase kinase 2 (**PDK2**), which inhibits **PDH** (68). **PDK2** repression activates **PDH** reducing the conversion of pyruvate to lactate and increasing the conversion of pyruvate to acetyl-CoA (68).

RAS can influence the glycolytic metabolism through the PI3K-mTOR pathway, or by upregulating glucose flux through hexosamine biosynthesis pathway (**HBP**) or the **PPP** (53). In addition, mutant KRAS activated lipogenesis through induction of **FAS** (69). In a Kras-driven mutant model of spontaneous lung tumorigenesis the uptake and utilization of branched-chain amino acids (**BCAAs**) such as leucine and valine, were increased, as well as the expression of the enzymes responsible for their catabolism (70).

PTEN decreases glycolysis and promotes oxidative phosphorylation. MEFs from PTEN tg mice present high levels of peroxisome proliferator-activated receptor g coactivator-1α (**PGC-1**α), increase the number of mitochondria, increment the levels of oxygen consumption and ATP production, and diminish lactate secretion. Moreover, PTEN decreases the levels of pyruvate kinase isozyme M2 (**PKM2**) and 6- phosphofructo-1-kinase/fructose-2,6-biphosphatase isoform 3 (**PFKFB3**); and elicits the inhibition of the pro-tumorigenic glutaminase **GLS1** thus contributing to the cancer-protection (71).

Of note, most studies on the role of oncogenes/tumor suppressor genes in metabolic reprogramming were performed using cell models with single genetic modifications. It's hard to translate this findings to the tumor in vivo which harbors many genetic defects, and whose metabolic profile will depend on their combination.

### METABOLIC HETEROGENEITY IN TUMORS

Based on the metabolic strategies prioritized by several solid cancers Lehuede et al. (72) proposed a classification of cancerspecific metabolic phenotypes in glycolytic and oxidative tumors. While lung, liver, colorectal cancers, and leukemias rely mostly on glycolysis; lymphomas, melanomas, and glioblastomas behave as oxidative tumors (72). However, there is not a uniform metabolic phenotype across tumors with a similar genetic defect in different organs or genetically different tumors in the same organ (73). A large study of metabolic features in 180 patientderived melanoma xenografts excluded a general metabolomic signature (74).

Cancer cells reside in poor oxygen and nutrition environments and hence attempt to reprogram the preexisting tissue metabolism for survival (75). The fact that some regions of the tumor could have more access to oxygen or various carbon sources than others (73) explains why tumor cells are metabolically heterogeneous. Intra-tumoral metabolic heterogeneity is maintained through coupled metabolic interactions between distinct cell populations coexisting in the tumor microenvironment. Stromal and tumor cells can compete or alternatively form symbiotic relationships where the metabolic products of a population become a nutrient of another (76).

Tumor cells can promote a "Reverse Warburg effect" in neighboring Cancer-associated fibroblasts (**CAFs**) (77). **CAFs** are stromal cells which often dominate the tumor microenvironment. Reactive oxygen species (**ROS**) produced by cancer cells activates HIF-1α and NFkB in **CAFs** inducing the production and release of energy-rich metabolites as lactate. Cancer cells could in turn take up lactate to fuel mitochondria respiration for energy production and anabolic metabolism (78, 79). This metabolic symbiosis may be controlled by the differential expression of lactate monocarboxylate transporters **MCT1** and **MCT4**. Lactate is released from **CAFs** by **MCT4** and taken up by **MCT1** in cancer cells. Indeed, lactate transporters inhibition reduces lactate uptake, induces a switch to glycolysis, and blocks metabolic symbiosis and tumor progression (80). Interestingly, higher levels of **MCT1** confer a higher metastatic potential to melanoma cells as metastasizing cells depend on **MCT1** to manage oxidative stress (81).

Cancer-associated adipocytes (**CAAs**) are adipocytes infiltrated into the tumor tissue (82). They provide carbon sources, growth factors, and cytokines affecting tumor growth, metastasis, and drug responses (83). **CAAs** frequently release fibroblast growth factors like (FGFs), leptin, adiponectin, IL-1β,

IL-6, TNF-α, CCL2, and CCL5; while cancer cells produce signaling molecules to trigger adipocyte lipolysis (84). In the presence of **CAAs** some cancer cells can acquire exogenous free fatty acids (**FFAs**) released by **CAAs** through the cell surface fatty acid translocase **CD36** and switch their metabolic program from glycolysis to **FAO** (38). In vitro, ovarian cancer cells induce white adipocytes lipolysis, fatty acids uptake and **FAO** (85).

Tumor metabolism also modulates the activity of tumorassociated immune populations. Activated T cells and cancer cells share some metabolic similarities (86) giving rise to a competition for nutrients which could impair the immune function (87). Naive CD4 T cells use mitochondrial **OxPhos** as a primary energy source, but upon activation they increase the expression of **GLUT1** and shift to aerobic glycolysis (88). Also TAMs Tumor Associated Macrophages (**TAMs**) **M1** rely on glycolysis to meet increased energetic demands (89). Moreover, increased lactate levels due to enhanced tumoral glycolisis can lead to diminished antitumour immunity. Lactate inhibits FIP200, leading to defective autophagy and increased apoptosis of naive T cells (90). Lactate can also suppress NK and dendritic cell function but does not affect regulatory T (**Treg**) cell function (91). Lactate could also induce the conversion of **M1** to **M2** pro-tumoral macrophages (92). CD8+ T cells and NK cells, are also sensitive to glutamine, serine, glycine, leucine, isoleucine and valine aminoacid restriction (93, 94). Moreover, limited availability of extracellular glutamine shifted the balance from Th1 to Treg cells (95).

Some cancer cells harbor an "hybrid glycolysis/ **OxPhos** phenotype" which allows them to use both glycolysis and the byproducts from glycolysis by **OxPhos**for energy production and biomass synthesis (96). Metabolic plasticity may be specifically associated with metastasis and therapy-resistance because a hybrid metabolism could maintain low **ROS** levels which induce a moderate stress response and the appearance of mutations that further stimulate tumorigenesis and metastasis (97). Dual inhibition of glycolysis (by 2-Deoxy-d-glucose, 2-DG) and **OxPhos** (by metformin) has been shown to effectively repress tumor growth and metastasis across multiple preclinical cancer models (98). Thus, a combination of glycolytic and **OxPhos** inhibitors could effectively eliminate the tumor survival potential of hybrid cells (99).

Understanding the factors that influence tumor heterogeneity is fundamental for the development of therapies that could act modulating tumor metabolism. Up to now tumor heterogeneity and toxicity issues has limited the success of most clinical trials targeting metabolic pathways (**Figure 1**).

### TARGETING CANCER METABOLISM TO OVERCOME DRUG RESISTANCE

Recently, metabolic reprogramming has been shown to play a role in the response of cancer cells to widely-used first-line chemotherapeutics (100).

Chemotherapeutic drugs target a differential feature of cancer cells that help them to actively proliferate. The main types of chemotherapy agents used in the clinics are: alkylating agents and platinants (damage DNA), such as cisplatin (101); cytotoxic antibiotics (bind DNA to prevent DNA and/or RNA synthesis); inhibitors of topoisomerase (damage DNA), such as daunorubicin, doxorubicin, irinotecan and etoposide; antimetabolites (interfere with intermediary metabolism of proliferating cells), such as gemcitabine; anti-microtubule agents (target microtubules and associated proteins required in cell division), such as paclitaxel and docetaxel (102); hormonal agents (inhibit hormone synthesis or function as hormone receptor agonist/antagonist) (103), such as tamoxifen or enzalutamide; and immunotherapy (target cancer cells that express a specific antigen or boost the natural ability of T cells to fight cancer), such as trastuzumab.

Tumor recurrence results from the ability of specific tumor subpopulations to resist treatment and expand. Resistance constitutes a lack of response to drug-induced tumor growth inhibition and it may be inherent to a subpopulation of cancer cells or can be acquired as a consequence of drug exposure. Chemoresistance is caused through genetic mutations in various proteins involved in cellular mechanisms such as cell cycle, apoptosis and cell adhesion (104). Reported chemoresistance mechanisms include: altered drug membrane transport, mutation, increased expression of drug targets, decreased drug activation, increased drug degradation due to altered expression of drug-metabolizing enzymes, drug inactivation due to conjugation with glutathione, altered drug subcellular redistribution, drug interactions, enhanced DNA repair, overexpression of anti-apoptotic genes, inactivation of apoptotic gene products, among others (105).

In the last decades, several studies have demonstrated that metabolic reprogramming plays an important role in the onset of chemotherapy resistance (106). Mostly due to the fact that chemotherapy agents used in the clinics cause a compensatory metabolic reprogramming in cancer cells (**Figure 2**). Thus, implementation of combinatorial therapies with chemotherapeutic drugs and metabolic modulators (**Table 1**) might provide a way to overcome therapy resistance.

### TARGETING GLUCOSE METABOLISM

As we already mentioned above, intensive aerobic glycolysis generates the accumulation of lactate that results in acidosis and promotes tumor progression and metastasis by inducing


immunosuppression, vascularization, aggressive proliferation, migration, invasion and therapy resistance (123, 124). It has been demonstrated that enhanced glucose uptake and improved aerobic glycolysis are capable to induce the intrinsic or acquired resistance to chemotherapy in several tumor cells such as breast (125), or ovarian (107). Several key glycolytic enzymes and glucose transporters inhibitors are currently in preclinical or clinical development to counteract resistance to chemotherapeutic drugs (100, 107, 126–129).

Some reports proposed that aerobic glycolysis is an important pathway for colorectal cancer (CRC) development. In fact, the overexpression of the immune checkpoint protein B7- H3 in CRC cells enhanced glucose consumption and lactate release by **HK2** expression, while B7-H3 knockdown had the opposite effect. Moreover, it is known that the depletion of **HK2** expression or **HK2** inhibition blocked aerobic glycolysis and CRC chemo-resistance (130). Recent studies reported that human colorectal adenocarcinoma doxorubicin-resistant cells (LoVo DOX) presents over expression of **GLUT1**. Thus, the treatment with silybin (a modulator of **GLUTs**) resulted synergic with the chemotherapeutic agents and it was able to overcome doxorubicin resistance (131).

Several key glycolytic enzymes, comprising **HK2**, **PFK**, and **PKM2**, are highly expressed in ovarian cancer cells and were implicated in anti-apoptotic and cell survival processes which correlate with chemo-resistance. These enzymes are controlled by oncogenes (e.g., Akt, mTOR) and tumor suppressors (e.g., p53) that may drive deregulated metabolism and ovarian cancer development (132). A few publications reported that ovarian cell lines with high glycolysis rate also presented high **OxPhos** activity, showing that most ovarian tumor cell lines prefer a highly glycolytic metabolic phenotype (133). Several inhibitors of glycolysis, such as **2-DG**, 3-bromopyruvate (**3-BrPA**) or lonidamide (**LND**), have been studied in recent years. The combined treatment between metformin and **2-DG** decreased the cellular growth of ovarian cancer cells (134). Moreover, **2- DG** was able to sensitize cisplatin (**CDDP**)-resistant and radioresistant cervical CaSki cell lines (135). **3-BrPA** (pyruvate analog) is an inhibitor of **HK2** and an alkylating agent. **LND** is another inhibitor of **HK2**, that enhanced **CDDP** activity in ovarian cancer cells (136). **LND** in combination with the chemotherapeutic agents, **CDDP**, and paclitaxel, presented a good activity and tolerability (137). The **PPP** is another pathway involved in ovarian drug resistance. It was demonstrated that ovarian **CDDP**-resistant cells (C13) showed increased glucose uptake, the up-regulation of the glucose transporter **GLUT1** and increased expression and activity of **G6PD**, in comparison to **CDDP**sensitive clones (2008). A combination of 6-nicotinamide (**6- AN**) (the **G6PD** inhibitor) and **CDDP** leads to a resensitization of **CDDP**-resistant cells (107). Moreover, since another ovarian cisplatin-resistant cancer cell line, IGROV PT, presented a higher expression of **G6PD**, the same combination has been loaded in liposomes and tested. The results showed a resensitization of resistant cells to cisplatin (138).

The increment in glycolysis is a common characteristic of drug-resistant breast cancer cells independent of the chemotherapeutic treatment, but this augmented activity is regulated in different ways in several resistant breast tumors. It has been demonstrated that triple-negative breast cancer (TNBC) and HER2 positive breast cancer possess higher rate of glycolytic activity than estrogen receptor-positive (ER+) breast cancer cells (139–141). In TNBC, it was shown that EGF pathways are activator of the first step in glycolysis (142) and that MYC modulates this metabolic phenotype by inhibiting the expression of the thioredoxin-interacting protein (143). The different expression of **GLUT** isoforms in breast cancer correlates with tumor cell differentiation, pathological grade, and prognosis. Invasive breast cancer, HER-2 positive, and TNBC, mostly present with a higher glycolysis rate due to the highest expression of **GLUT** (139). The most invasive breast cancer type, TNBC, has the highest **GLUT-1** expression (139). Moreover, overexpression of ErbB2 increased the expression of **LDHA**; promoting glycolysis and breast tumor growth (144). Increased glycolytic rate and a higher sensitivity toward inhibition of glycolysis were demonstrated in lapatinib-resistant BT474 breast cancer cells by a multi-omics approach (125). Curiously, the increase glycolytic activity in BT474 cells was not resulting from an overexpression of glycolytic enzymes, but merely from modifications in the phosphorylation state of glycolytic enzymes, demonstrating that post-translational changes alone can modulate glycolysis. In trastuzumab-resistant ErbB2-positive breast cancer cells, the improved glycolytic rate is regulated by heat shock factor 1 and **LDHA** and inhibition of glycolysis with **2-DG** and the **LDH** inhibitor oxamate by-pass trastuzumab resistance (145). Finally, in paclitaxel-resistant breast cancer cells, synergistic effects on inducing apoptosis were shown in **LDHA** downregulated cells or with oxamate (a pyruvate analog that inhibits the conversion of pyruvate to lactate) association (146). **LDHA** expression and activity are higher in taxol-resistant breast cancer cells. Downregulation of **LDHA** or oxamate treatment resensitizes taxol-resistant cells to taxol (146).

Differently to other tumor cell types that showed a higher rate of glucose consumption early in the modification process, prostate cancer (PCa) cells shift to the Warburg effect only in the metastatic stage, excluding the possibility to use advanced diagnostic procedures like standard FDG-PET scan for the detection of cancer in the early stages (147, 148). Glucose transporters have not been contemplated in PCa evolution because glucose metabolism in the prostate gland is different than in other organs. However, the relevance of **GLUTs** transporters has been lately proposed since the importance of increasing nutrients uptake, comprising glucose is clearly confirmed in PCa. In PCa androgens induce glucose uptake, upregulation of **GLUT** transporters and increased the AMPK pathway (149, 150). Glycolysis varies between androgen-sensitive and insensitive cells and it has been demonstrated that more aggressive tumors showed a higher glucose dependence (151). Indeed, prostate cancer cells switch to aerobic glycolysis only in the metastatic stage (147, 148). Even if the metabolic mechanism that supports prostate cancer metastasis has not been elucidated, in androgensensitive cells LNCaP and VCaP, androgen signaling induces both glycolysis and **OxPhos** (152). An augmented activity of key glycolytic enzymes by androgens has been established. **HK2** phosphorylation is prompt by androgens via PKA signaling, while **PFKFB2** is induced by direct binding of androgen receptor (AR) to **PFKFB2** promoter. Activation of **PFKFB2** produces a constitutive activation of 6-phosphofructo-2-kinase 2 (**PFK2**), which is entailed in the second irreversible reaction of the glycolytic pathway (150).

Combining different glycolytic inhibitors with chemotherapeutic agents could be a strategy to overcome drug resistance. To increase the anti-tumor activity **2-DG** was used in vitro and in vivo in combination with inhibitors of lysosomal permeabilization (153). Moreover, the combined treatment between **2-DG** and fenofibrate (PPARα agonist) caused a synergic effect in cancer cell growth (154). The combination of **2-DG** and paclitaxel in mouse xenografts models of human osteosarcoma and non-small cell lung cancer resulted in a significant reduction in tumor growth (155). **3-BrPa** use in vivo conditions resulted in anti-tumor activity after a single injection in a rabbit VX2 hepatoma model (156). Moreover, cells treated with **3-BrPa** enhanced doxorubicin-resistant cells response to the drug (112). Dichloroacetate (**DCA**), a **PDK1** inhibitor, was frequently used in combination with different chemotherapeutics agents (157). The combined treatment of **DCA** with paclitaxel was able to sensitize NSCLC resistant cells (158). Moreover, the combined treatment of **DCA** and **CDDP** was able to decrease tumor growth in advanced bladder cancer (159). The silencing of **PKM2** in lung cancer cells enhanced the efficacy of docetaxel (160). Another group reported that **PDK3** knockdown inhibited hypoxia-induced glycolysis and increased the susceptibility of cancer cells to paclitaxel (161). Cao et al. demonstrated that leukemia daunorubicinresistant cells show increased expression of GLUT1 and that the combined treatment between daunorubicin and phloretin, an inhibitor of glucose transporter sensitizes K562/Dox cells (113). Doxorubicin-resistant cell lines from anaplastic thyroid cancer presented a high expression of **G6GD** (an enzyme of **PPP**). **G6GD** knockdown or the anthraquinone physcion decreased **G6GD** activity and resensitized doxorubicin-resistant cells (111).

### TARGETING GLUTAMINE METABOLISM

Tumor cells increase glutamine metabolism to preserve the citric acid cycle, especially given the loss of the entry from pyruvate, in order to adapt to the modifications in the glycolytic pathway (162). Also, glutamine metabolism contributes to cancer cell chemoresistance. Recent studies demonstrated that the use of small molecules, of which several are in clinical trials, to inhibit key enzymes in glutaminolysis pathways is effective in slowing the proliferation of cancer cells (163–168).

Bis-2-(5-phenylacetamido-1,3,4-thiadiazol-2-yl)ethyl sulfide (**BPTES**) was recognized to be the first allosteric inhibitor of **GLS1** (169). It has demonstrated to reduce the proliferation in several cancer cell types in vitro and in xenograft models. Unfortunately, its poor metabolic stability and low solubility have discouraged its clinical development (169). For this reason, new inhibitors have been developed, such as **CB-839** that is a more potent inhibitor of **GLS1** than BPTES (170). **CB-839** reduced proliferation of mouse HCC cells at very low concentration, while **BPTES**, at the same concentration, had no activity (171). **CB-839** is ongoing in phase 1 clinical trial for the treatment of various cancer types [Study of the Glutaminase Inhibitor CB-839 in Solid Tumors https://clinicaltrials.gov/ct2/show/NCT02071862]. Glutamine analogs, such as azaserine, acivicin, and 6-diazo-5 oxo-L-norleucine (**DON**), are one strategy to disrupt glutamine metabolic pathways They form covalent bonds with Ser286 in the active site of GLS1 (166). These compounds have demonstrated to block the proliferation of a variety of cancers and have shown their efficacy in some clinical trials (163). Unfortunately, the main problem related to the clinical use of azaserine, acivicin, and **DON** is their lower selectivities toward **GLS1**, since they inhibit other glutamine-dependent enzymes. Then, **compound 968** was identified as an allosteric inhibitor of **GLS1**; and was shown to have cytotoxic effects in lymphoma, breast cancer, glioblastoma, and lung cancer (172–176).

It has been demonstrated that high expression of **GLS** can promote glutamine-independent growth and resistance to therapies that limit glutamine metabolism (177, 178). Thus, other pharmacological approaches to target glutamine metabolism include the use of glutamine synthetase or **GLUD** (Glutamate dehydrogenase) inhibitors (179).

Analysis in vitro demonstrated that a high glutamine flux protected MCF7 cells from tamoxifen-induced apoptosis (180). Indeed, a higher content of glutamate was correlated with breast cancer outcomes in patients (181). Metabolomic analysis of 270 breast tumor samples and 97 normal breast samples showed that breast tumor cells had a higher glutamate-to-glutamine ratio than normal cells (182). Another study showed that highly invasive and drug-resistant breast cancer cells presented increased glutamine metabolism, increased glutamate-to-glutamine ratio, and a higher glutaminase expression compared to non-invasive breast cancer cells (172). Moreover, deprivation of glutamine or **BPTES** treatment in combination with **CDDP** or etoposide enhanced chemotherapy cytotoxicity on breast cancer HCC1937 cells (183). Anti-proliferative effects of 1,4-di(5-amino-1,3,4 thiadiazol-2-yl)butane compound, **GLS1** inhibitor, on human breast tumor lines are similar to **BPTES** or **CB-839** (184). Cotreatment of **CB-839** and everolimus interrupts the growth of these endocrine-resistant xenografts (185).

**GLS1** and **GLS2** inhibitors or **BPTES** co-administered with doxorubicin demonstrated a synergistic activity decreasing proliferation of the human pancreas adenocarcinoma ascites metastasis (AsPC-1) cells (186). Disruption of glutamine metabolic pathways improved the efficacy of gemcitabine treatment (nucleoside analog that works by blocking DNA replication) in pancreatic cancer (187).

Some studies have revealed that some invasive ovarian tumor cells improve the use of glutamine to fuel **TCA** cycle (188). Yuan et al. demonstrated that **compound 968** is able to block cell proliferation and sensitize paclitaxel in ovarian cancer (189). Moreover, it has been demonstrated that ovarian cancer **CDDP**resistant cells present increased glutamine consumption and increased expression of the glutamine transporter **ASCT2** and **GLS**. Thus, the combined treatment of **BPTES** and **CDDP** resensitized **CDDP**-resistant cells (108). Another molecule is epigallocatechin gallate (**EGCG**), which is a **GLUD** inhibitor **GLUD**. This compound combined with **CDDP** had a synergic effect on A2780(cisR) ovarian cancer cells becoming a strategy to overcome cisplatin resistance (190).

mTOR inhibitors-resistant glioblastoma cells have a compensatory increase in glutamine metabolism, suggesting that combined inhibition of **GLS1** and mTOR could potentially overcome resistance (191).

### TARGETING LIPID METABOLISM

Lipid metabolism is another important player in the development of chemoresistance. The interest in therapeutic strategies directed to block lipid synthesis, lipid uptake, intracellular lipolysis/lipid utilization, and lipid storage is growing (192).

Among the enzyme that regulates lipid metabolism, **FASN** is an important one and it correlates with poor prognosis in various types of cancer and also interferes with drug efficacy (193). Moreover, **FASN** overexpression induces resistance to antitumoral drugs such as adriamycin and mitoxantrone in breast cancer cells (194), gemcitabine-resistant pancreatic cells (195), cisplatin-resistant ovarian cancer cells (110), and radiotherapy resistant head and neck squamous cell carcinomas (196).

Inhibitor compounds targeting lipogenic enzymes (such as **FASN**, **ACLY**, and **ACC**) have been studied and their anticancer activity has been demonstrated in several preclinical models (197–199). Besides the promising data, serious side effects of these compounds have led to their clinical development exclusion. Various **FASN** inhibitors, such as Cerulenin, C75, orlistat, C93, C247, and GSK837149A, have been identified (200). The inhibition of **FASN** demonstrated to represent an excellent target, when used in in vitro, in xenograft and genetically induced mouse model analysis (200). Inhibitors of **FASN** induced cancer cells death directly or sensitized them to chemotherapic drugs, such as 5-fluorouracil and trastuzumab (201–204).

It has been reported, by genomic profiling, that **CPT1** and fatty acid input into an oxidative pathway are decreased in four aggressive cancer cells, including melanoma, breast, ovarian, and prostate malignancies, respect to their non-aggressive counterparts (205). Recent studies reported a negative correlation between **FASN** inhibition and the consequent effect on body weight and food intake. In fact, a worsen eating that leads to weight loss was observed in mice treated with cerulenin and C75 and the cause seemed to be related to the **CPT**-1 inhibition in the hypothalamus (206–208).

The **SPHK1** sphingosine Kinase 1 isozyme has been largely studied and its several functions in tumor development have been demonstrated, while the **SPHK2** has not been as well-studied (209–213). Several studies in vitro and in vivo (only preclinical) demonstrated that ABC294640, the **SPHK2**-specific inhibitor, is able to inhibit proliferation of cancer cells or tumors more effectively or similarly than agents targeting **SPHK1**, in several tumor models, including ovarian (214), multiple myeloma (215), lung (216), kidney (217), breast (218), prostate (219), and pancreatic cancers (220).

Liver X receptor (LXR) is a crucial transcriptional regulator of cholesterol homeostasis and SR9243 is an LXR inverse agonist. SR9243 is able to kill selectively cancer cells because it inhibits the Warburg effect and lipogenesis and so the inhibition leads to the formation of an environment not favorable to cancer cells (221).

Recently, several studies have shown that dysregulated sphingolipid metabolism is a key contributor to the progression and resistance of ovarian cancer. By RNA-seq, Dobbin and colleagues revealed transcriptional variants between matched pairs of carboplatin and paclitaxel-treated vs. control patientderived xenograft (PDX) models of ovarian cancer. In particular, they identified that S1P signaling is modified pathways following chemotherapy treatment (222). Sphingolipid metabolizing enzymes strictly related in modulating the ceramidesphingosine-S1P rheostat play a key role in cell proliferation and have been directly correlated with drug resistance in ovarian cancer (223, 224). Specifically, increased expression of ceramide transport protein (**CERT**), **SPHK1**, **SPHK2**, and glucosylceramide synthase (**GCS**) have been correlated with resistance to paclitaxel, doxorubicin, and N-(4-hydroxylphenyl) retinamide (fenretinide) chemotherapies and cytotoxicity (225–230). The sphingolipid-mediated sphingosine-1-phosphate (**S1P**) pathway may represent a promising new pharmacological target to counteract the chemoresistance in ovarian cancer cells. Few therapeutic compounds specifically target **S1P** pathway proteins, but this pathway can modify the response of several chemotherapeutic treatments, including docetaxel, doxorubicin, and cyclophosphamide (231–234). Several approaches have been studied for modulating sphingolipid metabolism, and some of them consist in the use of combined treatment between ceramide analogs and chemotherapeutic agents (235–237). Treatments that showed activity in resistance ovarian cancer models include the use of synthetic ceramide analogs, inhibitors of **SPHK**, neutralization of secreted **S1P**, and **S1PR** antagonists. For example, the combined treatment of C6-ceramide with paclitaxel showed a synergic effect in suppressing cell proliferation and migration of CAOV3 ovarian cancer cells (238, 239). Moreover, drug delivery systems seem to be useful. In fact, a resensitization to paclitaxel of taxane-resistant SKOV3.TR ovarian cancer cells have been shown with the combination of paclitaxel with C6-ceramide-encapsulated in poly(ethylene oxide)-modified poly(epsilon-caprolactone) (PEO-PCL) nanoparticles (235). Kelly M. and colleagues demonstrated that the combined treatment of tamoxifen with the Sphingosine kinase 1 (**SK1**) inhibitor FTY720 blocks proliferation of both ERα-positive and ERα-negative drug-resistant cell lines and an ERα-positive PDX model of ovarian tumor (240). The multiple mechanisms of action of tamoxifen and its relatively high therapeutic index provide a strong rationale for combining tamoxifen with FTY720, as a strategy for treating ovarian tumors and circumventing drug resistance (226, 241–243).

It has been demonstrated that tamoxifen-resistant breast cells, T-47D, present an increased level of neutral lipids, in particular, cholesterol esters and triglycerides, and increased expression of Peroxisome Proliferator-Activated Receptor Gamma (PPARγ) (121). **CDDP**-resistant human ovarian cancer cell lines shift their metabolism toward a lipogenic phenotype and accumulate lipid droplets (244). Moreover, **CDDP**-resistant lung cells have an increased expression of **FASN** and that inhibiting **FASN** could decrease the metastatic potential of **CDDP**-resistant cells (245). It has been demonstrated that a combination of orlistat, an inhibitor of **FASN** and cisplatin, in vivo, causes a delay in tumor growth (110). A high fatty acid synthase (**FASN**) activity is also involved in ErbB2-induced breast cancer chemoresistance to docetaxel (116). It has been demonstrated that prostateresistant cells, C4-2R and MR49F (enzalutamide-resistant cells) respect to C4-2 and LNCaP have increased expression of 3 hydroxy-3-methyl-glutaryl-coenzyme A reductase (**HMGCR**). They showed that the combined treatment between simvastatin and enzalutamide sensitizes resistant cells in vitro. Moreover, tests in vivo in xenografts mice demonstrate a decrease in tumor cell proliferation (122).

Inhibitors of **CPT1**, such as etomoxir or ranolazine, have demonstrated promising results in different types of tumors. In fact, the combined treatment of etomoxir and orlistat is able to inhibit cell proliferation in LnCaP and VCaP prostate cancer cells (246). Moreover, the treatment of human leukemia cells with etomoxir or ranolazine can induce apoptosis cell death (115).

### TARGETING MITOCHONDRIA METABOLISM

During cancer cells' adaptation to an hypoxic microenvironment, mitochondria have been demonstrated to be fundamental during solid tumor metastasis and in chemoresistance (247– 249). Targeting mitochondrial-dependent metabolism to overcome drug resistance is an area of intense research. The increase of antioxidant pathways that help cancer cells to neutralize mitochondrial **ROS** is a common strategy adopted by some tumors to become drug-resistant (250). Moreover, mitochondria could promote therapy resistance by reducing the mitochondrial permeability transition (**MPT**) and inducing apoptosis resistance (251).

Mitochondria also appear responsible for the accumulation of oncometabolites such as fumarate, succinate, and 2 hydroxyglutarate (**2-HG**). Indeed, Succinate dehydrogenase complex iron-sulfur subunit B (**SDHB**), fumarate hydratase (**FH**), isocitrate dehydrogenase [NADP(+)] 1, cytosolic (**IDH1**), isocitrate dehydrogenase [NADP(+)] 2 and mitochondrial (**IDH2**) may be affected by germline or somatic mutations in a variety of human tumors (252). Fumarate, succinate and **2- HG** accumulation is sufficient to drive malignant transformation and thus behave like bona fide oncometabolite (253). These oncometabolites share the capacity to inhibit α-ketoglutaratedependent enzymes that control gene expression at the epigenetic level, such as Jumonji domain (JMJ) histone lysine demethylases (254) as well as ten-eleven translocation (TET) dioxygenases (255), resulting in the expression of a potentially oncogenic transcriptional program associated with a block in terminal differentiation (256).

Dysregulation of mitophagy (removing of abnormal mitochondria) contributes to neoplastic progression and drug resistance in various tumors (257). Enhanced mitophagy can protect cancer cells during chemotherapy and radiotherapy preventing apoptosis (258). On the other hand, excessive mitochondrial clearance may induce metabolic disorders and cell death (259). Therefore, mitophagy likely plays a dual role in cancer drug resistance (260). Mitophagy inhibition enhances the sensitivity of breast cancer cells to classical paclitaxel (114).

mtDNA has an essential role on tumorigenesis and chemoresistance. mtDNA pathogenic point mutations and changes in copy number, were shown to induce cancer progression (261) and have been associated with cancer development to a more malignant phenotype with poor prognosis in vivo (262–264). However, most mtDNA mutations are neutral missense mutations present in homoplasmy (265), suggesting that severe mutations are negatively selected. Indeed, mtDNA mutations per se are not able to induce carcinogenesis (266) but some mtDNA polymorphisms correlated with tumor development (267–270).

In particular, mutations in mitochondria encoded Complex I (CI) subunits could affect tumor progression depending on their mutational load and its detrimental activity (271). Based on these observations, Gasparre et al. introduced the concept of oncojanus: severe CI assembly mutations can promote tumorigenesis below a threshold level; but above that level they behave as antitumorigenic due to CI assembly defects. On the other side, non-disassembling mild mtDNA CI mutations could stimulate tumor proliferation and metastases. The oncojanus function of CI subunits was described for both mitochondrial and nuclear encoded CI subunits (272, 273). CI disruption inhibits OxPhos, promote NADH accumulation, inhibition of α-ketoglutarate dehydrogenase and increase the α-Ketoglutarate (KG)/succinate ratio. The α-Ketoglutarate (KG)/succinate imbalance activates prolylhydroxylases (PDH) enzymes responsible for the hydroxylation and degradation of HIF-1α even in hypoxic conditions (271, 274). Of note, genetical and pharmacological targeting of CI activity in osteosarcoma and colorectal cancer cell models successfully converted a carcinoma into a benign low-proliferating and noninvasive oncocytic tumor (273).

Moreover, the oncojanus effect was also observed in ovarian cancer after chemotherapy (275). A missense mtDNA point mutation in the MTND4 subunit of CI appeared after carboplatin treatment and generated a mild energetic defect allowing paclitaxel chemoresistance. When mutated MTND4 arrived to a certain threshold CI activity was impaired turning cancer cells into an oncocytic phenotype. Later it was demonstrated that the accumulation of deleterious mtDNA mutations induced by carboplatin in ovarian cancer cell lines conferred paclitaxel resistance through the reduction of filamentous tubulin (276). The treatment of A549 non-small cell lung cancer cells with CDDP induced an homoplasmic shift of a non-synonymous mutation in the CI protein MTND2 resulting in chemoresistance to cisplatin; which was correlated with the upregulation of the nuclear PGC-1α and PGC-1β and increased mitochondrial biogenesis (277).

mtDNA depletion in cancer cells under drug treatment promotes invasion and metastasis, induces expression of epithelial-to-mesenchymal (EMT) proteins (278) and activates pro-survival and antiapoptotic pathways (279, 280). Although the detailed molecular mechanism remains to be determined, several studies have demonstrated that reduced mtDNA content promotes activation of a mitochondria-to-nucleus signaling leading to increased expression of anti-apoptotic genes, including Bcl-2, and activation of pro-survival enzymes, such as Akt (280), that likely play a role in conferring resistance to apoptosis induced by drug treatment. mtDNA depletion in androgen-dependent LNCaP prostate cancer cells resulted in the loss of androgen dependence and increased resistance to paclitaxel (118, 119).

Horizontal transfer of mtDNA to cancer cells with a low respiratory function was correlated with recovery of respiration and increased tumor-initiating efficacy (281). mtDNA exchange through intercellular bridges or exosomes (extracellular vesicles implicated in cell-cell communication and transmission of disease states) could induce drug resistance by promoting a cancer stem cell (CSC) phenotype (282) Interestingly, exosomes containing mtDNA and mitochondrial proteins involved in mitochondrial fusion and biogenesis were found in the serum of prostate cancer patients as well as in the tumor itself (283). MSCs also protected AML leukemia cells from chemotherapeutic cytotoxicity by transferring them functional mitochondria. These effects occur together with mitochondrial fragmentation controlled by ERK-mediated Drp1 phosphorylation. Thus, disruption of leukemia cells/stromal interactions and targeting mitochondrial dynamics may provide a novel strategy that could be combined with conventional chemotherapeutic agents for leukemia treatment (284). It was demonstrated that the coculture of leukemia cells with mesenchymal stem cells (MSCs) increased the expression of uncoupling protein 2 (**UCP2**) in leukemia cells; uncoupling oxidative phosphorylation and decreasing **ROS** production (285, 286). Recent studies have demonstrated the protective role of mitochondrial metabolism on cancer cells exposed to chemotherapy cytotoxicity. In acute myeloid leukemia (AML) adenosine 5′ -monophosphate (AMP)-activated protein kinase (**AMPK**) signaling promoted glucose uptake and a shift to glycolysis decreasing intracellular ROS (287).

Phenformin is a biguanide similar to metformin that targets complex I of mitochondria. It was identified to be more potent in decreasing cell growth in non-small cell lung cancer, but unfortunately drug-resistance emerged (288). It has been hypothezed that the development of resistance is dependent on functional LKB1-**AMPK** signaling, which improves a switch in their metabolism to bypass inhibitory effects of phenformin. Since the serious side effects of the drug, it was withdrawned from the market. Besides this, new studies on phenformin have been conducted (288).

**PGC-1**α is a transcription co-activator that regulates mitochondria biogenesis and it is involved in energy metabolism. It has been reported that **CDDP** treatment increases **PGC-1**α levels in a small cell lung carcinoma cell line and **PGC-1**α silencing sensitizes cells to this drug (109). Increased mitochondrial mass confers stem-like properties to breast cancer cell lines MDA-MB-231 and MCF7 and enables their resistance to paclitaxel (117). **PGC-1**α induction may also cause chemoresistance by activating a metabolic shift to bypass **ATP** request, as shown for cells treated with inhibitors of BRAF (289). This is also confirmed for 5-FU resistance, which increased **PGC-1**α expression and so modified cellular cancer metabolism to modulate the energetic stress induced by treatment (290–292).

Numerous emerging studies demonstrated a correlation between mitochondrial dynamics and cell survival (293–298). The co-culture of leukemia cells with MSCs also altered the mitochondrial dynamics of leukemia cells due to an increase of the activating phosphorylation of Dynamin-1-like protein (Drp1) at S616. Drp1 is a GTPase that regulates mitochondrial fission. Leukemia cells overexpressing wild-type Drp1 or Drp1 S616E presented fragmented mitochondria, reduced mitochondrial **ROS** levels, increased glycolysis, and improved drug resistance (299–301). Drp1 S616 phosphorylation through the stimulation of mitochondria fission and glycolysis seemed required to RAS-induced transformation (302). Moreover, inhibition of Drp1 activity caused mitochondrial fusion and impeded tumor growth (303). The cell cycle inhibitor Cytarabine is a normally chemotherapeutic therapy for leukemias and lymphomas, with reduced clinical implications due to the development of resistance (304, 305). Among the probable mechanisms, Cytarabine treatment improved the increase of a chemoresistant leukemic stem cell population with high **FAO**/**OxPhos** activity. In light of this, the **FAO** inhibitor etomoxir blocks oxygen consumption in acute myeloid leukemia cells and sensitized cells to cytarabine (306).

Recent evidence suggests that chemoresistant ovarian cancer has an increase in **OxPhos** dependence. Improved **OxPhos** in ovarian cancer cells increase IL-6 production (307) which facilitates tumor cell survival and proliferation (308), changing efficacy to chemotherapy, and reduce progression-free survival of ovarian cancer patients (309). Ovarian cancer cell migration was shown to be sustained by pyruvate, involving the mitochondrial activity during metastasis (310). Other studies have revealed that some invasive ovarian tumor cells improve the use of glutamine to fuel the **TCA** cycle (188). CD44+CD117+ ovarian tumor stem cells derived from epithelial ovarian cancer patients exhibited both high glucose uptake and a high **OxPhos** phenotype, which was correlated with their amplified capacity to live under a glucose-free context (311). On the other hand, CD44+MyD88+ cancer stem cells had a mainly glycolytic phenotype and suggested that therapy with glycolytic inhibitors could be favorable to increase patient's survival (312). Mouse ovarian cancer-initiating cells (putative cancer stem cells) harbor a highly flexible metabolic phenotype, whereby they could use either glycolysis or **OxPhos** under stress (313). It was proposed that most ovarian tumor cells may use either glycolysis or OXPHOS and that such plasticity increased their "cellular fitness" (310, 314, 315). The shift from glycolysis to **OxPhos** has also been showed upon MYC/KRAS or MYC/ERBB2 removal in breast cancer cells (316, 317), and also in glioma cells because of the acquired resistance to phosphoinositide-3-kinase (PI3K) (318). Moreover, PI3K resistance in breast cancer cells is related to the shift from glucose to lactate (34). Inhibitors of mitochondrial respiration become therapeutic strategies in ovarian cancer cells because of their dependence on **OxPhos**. In fact, cancer-selective inhibition of the electron transport chain (**ETC**) could kill ovarian cancer cells directly without affecting normal cells. The complex I inhibitor BAY 87-2243 has been studied in a Phase 1 study (NCT01297530) but results were not showed (Clinicaltrials.gov). Several strategies targeting mitochondria CIII complex, such as the thiol modifier β-phenylethylisothiocyanates (**PEITC**), have been developed. In particular, **PEITC** can enhance the **ROS** level, decreasing **OxPhos** and, consequently, causing prostate cancer cell death by apoptosis (319). Moreover, the same compound combined with metformin demonstrated cytotoxicity in human ovarian cancer cells (320). Also, treatment with ABT-737, the inhibitor of Bcl-2 proteins (321) and the FAO inhibitor perhexiline also was capable of sensitized **CDDP**-resistant ovarian cancer cells (322) by modulating mitochondrial metabolism. The main drugs to target mitochondria or disrupt **OxPhos** are antibiotic or antiparasitic agents. Among these, azithromycin and doxycycline target mitochondrial protein synthesis, while salinomycin targets mitochondrial K+/H+ exchange (323). Azithromycin, combined with **CDDP** and paclitaxel, is able to reduce side effects and to enhance patients' survival (324). Doxycycline is able to inhibit cellular growth of ovarian cells SKOV3 and SKOV3/DPP and to sensitize resistant cells to **CDDP** (325). Salinomycin is another antibiotic able to inhibit cell growth, especially on cancer cells compared to normal epithelial cells (326).

Resistance to 5-FU has been associated to aumgmented mitochondrial mass and activity, increase **ETC** enzymes expression and higher level of oxygen consumption (290, 327). So, due to their **OxPhos**-dependency, resistant cells were showed to be sensitive to Complex I inhibition by metformin (327). **OxPhos** involvement to 5-FU resistance was correlated to the development of stemness-related phenotype, stricly linking CSCs to mitochondrial metabolism as described previously (328, 329). According, the activation of mitochondrial **FAO** is able to promote stemness in gastric cancer cells and consequently there is chemoresistance to 5-FU induced by tumor-associated mesenchymal stem cells. In fact, treatment with the **FAO** inhibitor etomoxir was capable to partially reduce FU resistance (330).

BRAFV600E is a mutation found in stage IIIc or stage IV melanoma. Chapman and co-workers demonstrated that its inhibition with vemurafenib leads to the shift to **OxPhos** and the switch is useful to treat resistant melanoma cells (331). Elesclomol sodium (STA-4783) is a compound targeting **ROS** in tumor cells. Its mechanism is strictly related to inhibition of electron transport flux and so increase **ROS** generation causing oxidative stress in both malignant and healthy cells. However, as cancer cells have already higher **ROS** production, this drug will be capable to cause cytotoxicity selectively in malignant cells, resulting in activation of apoptosis in melanoma cancer cells (332). STA-4783 alone and in association with paclitaxel revealed promising results in phase I/II clinical studies in patients with refractory solid tumors (333, 334), but unfortunately serious side effects lead to stop phase III study in melanoma patients (335).

Lee and coworkers demonstrated that an enhanced mitochondrial oxidative phosphorylation characterizes irinotecan-resistant NSCLC cells. They tested a combined treatment between gossypol (a molecule that inhibits aldehyde dehydrogenase) and phenformin (a molecule that inhibits mitochondrial complex I), concluding that the combination leads to sensitization of irinotecan-resistant NSCLC cells (336). Another drug targeting mitochondria is apogossypol, which is semisynthetic. It has been demonstrated to be cytotoxic in murine B cells (337). Metformin, a drug usually used for the treatment of type 2 diabetes, has demonstrated anti-cancer properties. In fact, the combined treatment of metformin and paclitaxel showed anticancer activity in vivo and was able to arrest the cell cycle in vitro in human breast MCF-7 and human lung A459 cancer cells (338). PC3 prostate cancer cells docetaxel-resistant shift their metabolism from glycolysis toward OXPHOS and this is linked to EMT phenotype. The combination of chemotherapy and OXPHOS inhibitors limited docetaxel-associated drug resistance and progression toward metastasis (120).

### REFERENCES


In short, targeting glycolysis, **PPP**, **OxPhos**, and fatty acid synthesis and oxidation could enhance chemotherapy and radiation responsiveness and overcome therapy resistance. Importantly, therapy-resistant tumors present different metabolic phenotypes related to non-treated tumors, thus it's needed a better understanding of the new dependencies and vulnerabilities across different chemotherapy treatments in different tumors to reduce toxicity and to increase the efficacy of chemotherapeutic drugs.

### CONCLUDING REMARKS

Metabolic deregulation is an established hallmark of cancer, thus the elucidation of novel therapy combinations based on new tumor-specific metabolic liabilities after chemotherapy will be essential to the development rational metabolic therapeutic strategies to overcome drug resistance.

### AUTHOR CONTRIBUTIONS

MD, IG, TP-G, and MM wrote the manuscript.

### FUNDING

This work was supported by AIRC Inverstigator Grant 2018\_22030 and MIUR PRIN 2017\_2017237P5X.


sensitizes glioblastomas to antilipogenic therapy. Sci Signal. (2009) 2:ra82. doi: 10.1126/scisignal.2000446


delays tumor growth in the neurofibromatosis type I tumor model. Cell Death Differ. (2013) 20:659–68. doi: 10.1038/cdd.2012.168


energy for rapid tumor growth. Nat Med. (2011) 17:1498–503. doi: 10.1038/ nm.2492


breast cancer. Int J Cancer. (2015) 136:1619–28. doi: 10.1002/ijc. 29152


mitochondrial DNA mutations in human cancer. eLife. (2014) 3:e02935. doi: 10.7554/eLife.02935.028


fission and MAPK-driven tumor growth. Mol Cell. (2015) 57:537–51. doi: 10.1016/j.molcel.2015.01.002


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

Copyright © 2020 Desbats, Giacomini, Prayer-Galetti and Montopoli. 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.

# Lactate Beyond a Waste Metabolite: Metabolic Affairs and Signaling in Malignancy

Fátima Baltazar 1,2 \* † , Julieta Afonso1,2†, Marta Costa1,2† and Sara Granja1,2†

<sup>1</sup> School of Medicine, Life and Health Sciences Research Institute (ICVS), University of Minho, Braga, Portugal, 2 ICVS/3B's—PT Government Associate Laboratory, Guimarães, Portugal

To sustain their high proliferation rates, most cancer cells rely on glycolytic metabolism, with production of lactic acid. For many years, lactate was seen as a metabolic waste of glycolytic metabolism; however, recent evidence has revealed new roles of lactate in the tumor microenvironment, either as metabolic fuel or as a signaling molecule. Lactate plays a key role in the different models of metabolic crosstalk proposed in malignant tumors: among cancer cells displaying complementary metabolic phenotypes and between cancer cells and other tumor microenvironment associated cells, including endothelial cells, fibroblasts, and diverse immune cells. This cell metabolic symbiosis/slavery supports several cancer aggressiveness features, including increased angiogenesis, immunological escape, invasion, metastasis, and resistance to therapy. Lactate transport is mediated by the monocarboxylate transporter (MCT) family, while another large family of G protein-coupled receptors (GPCRs), not yet fully characterized in the cancer context, is involved in lactate/acidosis signaling. In this mini-review, we will focus on the role of lactate in the tumor microenvironment, from metabolic affairs to signaling, including the function of lactate in the cancer–cancer and cancer–stromal shuttles, as well as a signaling oncometabolite. We will also review the prognostic value of lactate metabolism and therapeutic approaches designed to target lactate production and transport.

Keywords: lactate, warburg effect, monocarboxylate transporters, GPR81, metabolic fuel, lactate shuttles, signaling molecule

### INTRODUCTION

The first discoveries involving lactate were reported in 1808, when it was described in the muscle of animals; only many years later was lactate associated with energy metabolism in muscle contraction (1, 2). The glycolysis pathway, transformation of glucose into pyruvate and ATP, was described in the 1940s by the joined efforts of numerous scientists, in a cascade-like chronology. It started with the discovery of fermentation in microorganisms by Louis Pasteur; then, Meyerhof et al. described the lactate cycle, providing essential knowledge on the transformation of energy in cells (3–5).

Lactate formation and functions were incorrectly described for a long time, since lactate was considered as a waste product of cellular metabolism (6). Although history did not give lactate its due importance, it is believed at presentthat lactate has a crucial role, especially as a shuttle molecule. The concept was introduced by Brooks more than 30 years ago (7–9), and despite some initial disbelief (10, 11), several reports have finally acknowledged the role of lactate in

#### Edited by:

Paolo E. Porporato, University of Turin, Italy

#### Reviewed by:

Jacques Pouyssegur, Université Côte d'Azur, France Franklin David Rumjanek, Federal University of Rio de Janeiro, Brazil

> \*Correspondence: Fátima Baltazar fbaltazar@med.uminho.pt

†These authors have contributed equally to this work

#### Specialty section:

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

Received: 15 December 2019 Accepted: 11 February 2020 Published: 18 March 2020

#### Citation:

Baltazar F, Afonso J, Costa M and Granja S (2020) Lactate Beyond a Waste Metabolite: Metabolic Affairs and Signaling in Malignancy. Front. Oncol. 10:231. doi: 10.3389/fonc.2020.00231

**90**

shuttles between glycolytic and oxidative cells, being the product of one and used by another (12). It is well-established that lactate is formed from the reduction of pyruvate via lactate dehydrogenase (LDHA), under aerobic or anaerobic conditions, produced, and transformed continuously by resting/exercising muscle, brain, heart, and gut tissues (13). Lactate is a major source of energy, the major gluconeogenic precursor and, as a signaling molecule, is capable of inducing autocrine, paracrine, and endocrine-like effects. This molecule is responsible for several homeostatic functions: For instance, in hepatocytes, it feeds gluconeogenesis; in the brain, it is used by astrocytes and neurons for oxidative metabolism (12, 14).

Lactate homeostasis in a healthy environment requires adequate transporters. The lactate transporters, monocarboxylate transporters (MCTs), are members of the SLC16 gene family and several have been identified by gene homology, as it will be further explained (15). Physiological levels of lactate are considered to be in the range of 1.5–3 mM in blood and tissue from healthy individuals (14, 16); higher values are usually an indication of a health problem. Lactate shuttles are key players in many conditions involving pregnancy and reproduction (17, 18), the human heart (19), brain (12), and cancer (13). The use of lactate levels as a marker of clinical outcome was first suggested in 1964 by Broder and Weil, when studying patients with undifferentiated shock (20). Since then, high lactate levels have been associated with several diseases such as shock, cardiac arrest, trauma, ischemia, diabetic ketoacidosis, liver dysfunction, and sepsis (21). Lactate also modulates the immune system and promotes immune-inflammatory responses (22). The levels of lactate are increased in several inflammatory and autoimmune disorders, and lactate transporters were overexpressed at the surface of immune cells (14, 23). Lactate accumulation and transport has become particularly relevant in rheumatoid arthritis, where MCT4 inhibition was pointed as a possible therapeutic strategy (24). In the cancer setting, Otto Warburg was the first to observe that tumor cells share a common metabolic feature: high glucose consumption and increased glycolysis leading to lactate production, regardless of oxygen availability (25, 26).

### ROLE OF LACTATE IN CANCER METABOLIC REWIRING

Cancer metabolism emerged as an area of research that has increasingly gained attention in the last decades. In order to sustain the proliferative phenotype, cancer cells enroll metabolic changes, such as the "Warburg effect" disclosed by Otto Warburg in 1926 (27). These changes consist on upregulation of glucose metabolism (glycolysis) even in the presence of oxygen, thereby producing high levels of lactate and reducing the use of the tricarboxylic acid (TCA) cycle. This addictive glycolytic phenotype arises as a distinctive metabolic characteristic of many types of cancer, being introduced as a new hallmark of cancer in 2011 (28).

## Oncogenic Triggers of Glycolytic Metabolism

Tumorigenesis is characterized by genetic alterations, and several findings demonstrate that high expression of specific transcription factors or oncogenic tumor pathways, principally MYC, hypoxia-inducible factor-1 alpha (HIF-1α), nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB), and phosphatidylinositol-3-OH kinase (PI3K), can sustain the Warburg effect. As the tumor starts to grow, oxygen diffusion becomes limited and cancer cells respond to these environmental changes by upregulating HIF-1α (29). HIF-1α, in turn, induces the overexpression of key players in the conversion of glucose into lactate, such as glucose transporters (GLUTs) and hexokinase (HK) 1 and 2, which are responsible for the initial steps of glycolysis; lactate dehydrogenase A (LDHA), which transforms pyruvate into lactate (30, 31); and the lactate-extruding monocarboxylate transporter 4 (MCT4) (32). Conversely, HIF-1α blocks the entry of pyruvate into the TCA cycle by upregulating pyruvate dehydrogenase kinase 1 (PDK1), driving tumor cell energy to glycolysis (33). The major oncogene Ras, when mutated, can also induce glycolysis through the activation of the mammalian target of rapamycin complex I (mTORC1). Akt kinase activation by PI3K results in increased glucose uptake, HK2 targeting to the mitochondria, and increase in glycolytic flux (34), while the transcription factor MYC increases glutaminolysis and upregulates MCT1 expression (35). Cancer cell metabolism is also influenced by the activity of tumor suppressor genes. Loss of the p53 protein prevents expression of the synthesis of cytochrome c oxidase (SCO2) gene, decreasing mitochondrial respiration (36). Lactate can also function as a paracrine tumor molecule (37). Acidosis often precedes angiogenesis and lactate can stimulate HIF expression independently of hypoxia (38). Thus, instead of one event promoting the Warburg effect, numerous factors play a role in determining the fate of glucose in cancer cells. Also true is that somatic mutations in genes involved in metabolism either cause/predispose cells to become malignant. For instance, mutations in succinate dehydrogenase are related to paraganglioma, and mutations in fumarase can induce leiomyoma and leiomyosarcoma formation (39); isocitrate dehydrogenase mutations are related to glioma development (39).

### Lactate Transport in Cancer

The major oncometabolite resulting from tumor metabolic rewiring is lactate, which is abundant in the tumor microenvironment (TME). Because lactic acid is hydrophilic and a weak acid, its transport across membranes requires transporters that belong to the monocarboxylate transporter family. MCTs 1–4 facilitate the transmembrane H+-linked transport of monocarboxylates, including lactate, pyruvate, acetoacetate, and β-hydroxybutyrate (15), having the cell surface glycoprotein CD147 as an obligatory chaperone (40). MCT1 and MCT4 isoforms are strongly associated with the hyperglycolytic phenotype of cancer cells. These transporters display distinct affinities for monocarboxylic acids that are associated with their expression patterns within tissues (41). MCT1 expression Baltazar et al. Lactate Beyond a Waste Metabolite

was identified in most tissues, being associated with the uptake/extrusion of lactate, while MCT4 has an important role in the export of lactate in highly glycolytic tissues. MCTs have been widely studied by our group and others, and found to be robustly expressed in a variety of solid human tumors such as colon, glioblastoma, breast, prostate, stomach and others, as depicted in **Table 1** [for a comprehensive review see (41, 72)]. MCT expression/cell localization can differ from cancer to cancer. Importantly, the prognostic potential of MCTs was found in various tumor types (**Table 1**). Given the key role of MCTs in cancer, these transporters are promising therapeutic targets in cancer. A less studied family of lactate transporters, also known to facilitate the transport of monocarboxylates, are the sodium-coupled monocarboxylate transporters (SMCTs), containing two members, SLC5A8 and SLC5A12 (73).

## LACTATE ROLES IN THE TUMOR MICROENVIRONMENT: FROM METABOLIC AFFAIRS TO SIGNALING

Tumor growth occurs under a nutrient/oxygen-restrictive microenvironment where cancer cells are enrolled in a reprogrammed metabolism that allows them to surpass those limitations while facilitating malignant dissemination. Not only cancer cells, but also cancer-associated stromal cells, take part in such scavenging program, being cytosolic lactate the main driver of those metabolic alterations. Lactate is formed exclusively from pyruvate regardless of oxygen availability, and robustly exported to the TME, reaching concentrations that can be 20-fold higher (about 40 mM) (74) than in non-tumoral tissues (about 1.5–3 mM) (13, 15). Lactate is a major fuel source, providing energetic and anabolic support to cancer cells, and an important oncometabolite with both extracellular and intracellular signaling functions that equally contribute to cancer progression (**Figure 1**) (75, 76).

### Lactate as a Metabolic Substrate

Lactate acts as a powerful regulator of multiple hallmarks of cancer, supporting cell proliferation and promoting immune suppression, angiogenesis, migration, metastasis, and resistance to therapy (20), namely chemotherapy (77), radiotherapy (78), and targeted therapy (79, 80). Cancer cells exploiting aerobic glycolysis upregulate GLUTs and MCTs, secreting large amounts of lactate (**Figure 1**), while deviating glycolytic intermediates to fuel alternative anabolic pathways (e.g., pentose phosphate pathway), thus sustaining rapid cell proliferation (81). Due to the metabolic heterogeneity of the TME, cancer cells are also able to engage into context-dependent metabolic affairs, similarly to what occurs in muscle (82) and brain (83). One such example is the symbiosis between well and poorly-oxygenated cancer cell populations (**Figure 1**): at the hypoxic, nutrient-poor/normoxic, nutrient-rich interface, lactate is released by glycolytic cancer cells through MCT4, and taken up by oxidative cancer cells through MCT1, where it fuels oxidative phosphorylation, thus sparing glucose for glycolytic cancer cells (84). Occurrence of this "two compartment model" was additionally described between



↑, increased; ↓, decreased; +, positive expression; CAFs, cancer-associated fibroblasts; DFS, disease-free survival; MFS, metastasis-free survival; OS, overall survival; PFS, progression-free survival; RFS, recurrence-free survival; TNM, tumor, node, metastasis.

lactate-avid breast cancer cells and "corruptible" glycolytic cancer-associated fibroblasts (CAF) by Lisantis's group (85), and further amplified to a "three compartment model" in the study by Curry et al. (86); in head and neck cancer samples, the authors showed that catabolic compartments composed of Warburg-adapted MCT4-expressing cancer cells and CAFs provided anabolic MCT1-expressing cancer cells with glycolysis-originating lactate (86). Since those original observations, numerous studies reported similar associations in several cancer models, such as non-Hodgkin lymphoma (87), pancreas (88), lung (89), prostate (90), and bladder (91); clinically, this metabolic phenotype has been associated with cancer aggressiveness, resistance to therapy and poor survival (89–91). Several mechanisms have been pointed out as mediators of those metabolic affairs, such as secretion of growth factors [e.g., cancer cell-secreted basic fibroblast growth factor (bFGF) in response to CAF-secreted hepatocyte growth factor (HGF) (92)], interleukins [e.g., IL-1β secretion by cancer cells (93)], and exosomal microRNAs (94). Interestingly, microRNA-containing exosomes secreted by CAFs were able to inhibit oxidative metabolism in cancer cells, while providing them with intact metabolites, namely glucose, to sustain their growth (95). In such an inverted scenario, CAFs oxidize cancer cell-derived lactate to support tumor proliferation (96); this has been correlated with resistance to targeted therapy (80).

In addition to CAFs, the metabolic promiscuity described above involves other cells of the TME, namely immune and endothelial cells (**Figure 1**). Cytotoxic T cells' transition from an anergic to a fully activated state relies on an accelerated glucose metabolism (97) and, in a glucose-restricted TME, cancer cells easily succeed in such metabolic competition (98). Dampening of lymphocyte proliferation and motility, cytokine production and cytotoxic activity ultimately leads to immunosuppression, as a result of the excess cancer cell-derived lactate that blocks lactate export by immune cells (99) and might be inclusively taken up by those cells, thus impairing their glycolysis-dependent activation (100). Lactate also mediates polarization of macrophages from an M1- (anti-tumoral type) to an M2-like phenotype (pro-tumoral type) (101, 102); induction of VEGF (vascular endothelial growth factor) expression has been linked to this pro-tumoral state (103). Moreover, it was proposed that endothelial cells rely on extracellular lactate uptake, via MCT1, as a fuel source for their oxidative metabolism, promoting VEGF/VEGFR-2 production through HIF-1α stabilization, endothelial cell migration and tube formation (104–106).

Lactate stimulates motility, migration and invasion of cancer cells (38, 107), as a probable result of CD44 expression and hyaluronan production (108), as well as activation of matrix metalloproteinases (MMPs) (109), both promoted by extracellular acidosis.

### Lactate as a Signaling Metabolite

As stated above, lactate can serve additional purposes beyond acting as a metabolic substrate, functioning as an intracellular signaling mediator and as an extracellular ligand. At the intracellular level (**Figure 1**), hypoxia adaptation is assured by lactate in HIF-1α-dependent [direct HIF-1α stabilization by prolyl hydroxylase (PHD) 2 inhibition (110)] and independent [binding to N-Myc downstream-regulated (NDRG3) protein, preventing association with PHD2 (111)] fashions. HIF-2α stabilization is also induced by lactate, which ultimately potentiates glutaminolysis in cancer cells (112). Lactate promotes HIF-1α-mediated VEGF expression in the cancer cell, and expression of bFGF and VEGFR-2 by neighboring endothelial cells (105). Apart from this HIF-1α-dependent angiogenic signals, endothelial cells are also activated by NF-κB stabilization (104) in a lactate-dependent manner. NF-κB activation in cancer cells' instructed CAFs additionally drives resistance to targeted therapies, being lactate secreted by cancer cells the instructor in such phenotype (80). Pyruvate kinase M2/HIF-1α-driven gene expression in prostate cancer cells promoted epithelial-to-mesenchymal in response to CAF-secreted lactate (113). Immune suppression is also mediated by lactate signaling, as different studies reported that this oncometabolite interferes with key tumor pathways that lead to IFN-γ production by cytotoxic T cells (114), activates the IL-23/IL-17 proinflammatory pathway (115) and promotes polarization of macrophages toward an M2-like phenotype (103). Recently, functions in histone posttranslational modification, termed histone lysine lactylation, were attributed to lactate and shown to regulate gene expression in macrophages; increased lactate production led to this epigenetic modification, inducing an M2-like phenotype during wound healing (116).

The signaling functions of lactate at the extracellular space (**Figure 1**) are mediated by the lactate-activated G-proteincoupled receptor GPR81 (5, 117). Its expression is not limited to the plasma membrane but also to other intracellular organelles (118). GPR81 activation occurs at a lactate concentration of 0.2–1.0 mM (119), followed by cyclic AMP downregulation and inhibition of protein kinase A (PKA)-mediated signaling (120). In the physiological context, lactate binds to GPR81, which inhibits lipolysis in fat cells (121). In the cancer context, modulation of lactate-sensing proteins, such as MCTs, ultimately leads to tumor proliferation and dissemination (122), escape from the immune system (123) and therapy resistance (124). GPR81 expression is upregulated in cervical (125), breast (126) and liver cancer (122), and associated with the progression of cervical squamous carcinoma (125). GPR81 is highly expressed in different cancer cell lines including colon, breast, lung, hepatocellular, cervical, and pancreatic (122, 126). In vitro, GPR81 expression associated with cancer cell survival, proliferation, migration, invasion and resistance to chemotherapy, and is involved in the suppression of antitumor immunity by promoting the overexpression of PD-L1 in lung cancer cells lines (123, 124, 126). Knockdown of GPR81 in a xenograft cancer model resulted in reduction of tumor growth and metastasis (122, 126).

The putative lactate sensors GPR4, GPR65, GPR68, and GPR132 have been described as proton-sensitive, and are activated at the acidic TME due to the low pH levels obtained from lactic and carbonic acids (5). GPR132 and GPR65 were additionally described in macrophages and linked to their polarization toward a pro-tumoral phenotype (127, 128). It remains to be clarified whether the modulation of lactate-sensing signaling pathways occurs through a direct GPR-lactate interaction ou through a conformational modification in the receptor induced by lactic acidosis.

### LACTATE METABOLISM AS A PROGNOSTIC AND THERAPEUTIC TOOL

As mentioned above, lactate levels in tissues can mirror their metabolic status. Lactate concentrations vary either within healthy or diseased tissues, reflecting the distribution of the metabolic activity in the tissue, phenomenon known as "metabolic zonation" (129). This term was first described in the liver, in which there is hepatocyte metabolic heterogeneity along the porto-central axis, resulting from the physiological occurring oxygen gradient (130). In solid malignant tumors, which are characterized by high heterogeneity, "metabolic zonation" results from different intrinsic properties of cancer cells, co-existence of different cell populations within the tumor and different distribution of the vascular supply (129). The clinical significance of the variable levels of lactate in human solid tumors was first described in 2000 by the group of Walenta et al. (131), where significantly higher lactate levels in cervical metastatic tumors were found, compared with nonmetastatic malignancies, suggesting that tumor lactate content could be used as a prognostic biomarker. Interestingly, the levels of lactate were inversely correlated with the levels of glucose, and directly correlated with the expression of MCT4. Later studies have also linked intratumoral lactate levels with higher incidence of distant metastasis and poor patient survival (76).

Prognostic value has also been attributed to important players in lactate metabolism, namely involved in lactate production (LDHA) and lactate transport (MCTs). There are several studies on the prognostic value of LDH levels, supported by systematic reviews/meta-analyses (132). As examples, higher pretreatment LDH concentration is associated with increased risk of overall mortality in lung cancer patients (133), high LDH serum levels are associated with lower event-free survival (EFS) in osteosarcoma patients (134) and with overall survival (OS)/progression-free survival (PFS) in urinary system cancer patients (135). Besides serum LDH, LDHA levels in cancer tissues have been reported as a biomarker of malignancy and prognosis (136). As examples, upregulation of LDHA levels in pancreatic and esophageal cancer have been associated with metastasis, tumor stage, tumor recurrence, and patient survival (137). However, LDHA expression in malignant tumor tissues does not correlate consistently with serum LDH levels, which may indicate that these are independent prognostic factors in cancer (138, 139). Studies on lactate transporters (MCTs/CD147) are not as solid as for LDH, but also reveal prognostic value (**Table 1**) (41). In a recent meta-analysis Bovenzi et al. (140) identified association between increased MCT4/CD147 expression with decreased OS and disease-free survival (DFS) across many cancer types, while there was no clear association for MCT1 expression with these parameters.

Besides prognostic biomarkers, both LDHA and MCTs have been recognized as attractive targets for cancer therapy. LDHA overexpression has been associated with increased cancer aggressiveness and targeting has been tackled both genetically and pharmacologically. There are different types of LDHA inhibitors, including the pyruvate-competitive (e.g., oxamate), the NADH-competitive (e.g., gossypol), the pyruvate and NADH-competitive (N-hydroxyindoles), and the free enzymebinding inhibitors (galloflavin) (138). LDH pharmacological inhibition reduces lactate production, impairs cell proliferation in vitro, and reduces tumor size in vivo, either with LDH inhibitors alone or in combination with other agents (138, 141). Additionally, gossypol has also demonstrated promising results in different clinical trials, being relatively safe and effective in reducing tumor markers (138, 142). These results are supported by LDHA silencing studies in tumor models, where cell proliferation, migration and tumor growth were prevented (143, 144). However, genetic studies deleting LDHA/LDHB or glucose phospho-isomerase (GPI) have demonstrated that Warburg effect is dispensable as agressive tumors, relying on OXPHOS, are able to survive and to develop tumors in nude mice (145).

As stated above, upregulation of MCT1 and MCT4 has been described in a variety of human cancers, and inhibition of MCT activity has been showing promising results in preclinical models (41). In vitro, MCT inhibition impairs lactate transport, cell proliferation, invasion and migration, and induces cell death, while it delays tumor growth, induces necrosis and decreases invasion in vivo. MCT activity has been inhibited either genetically (gene downregulation or knockout) or using pharmacological inhibitors. MCT classical inhibitors include the α-cyano-4-hydroxycinnamate CHC, 4,4′ -di-isothiocyanostilbene-2,2′ -disulfonates (e.g., DIDS) and flavonoids (e.g., quercetin) (15). However, these inhibitors are not MCT/MCT isoform specific, which prompted the search for new inhibitors. AstraZeneca developed MCT1 specific inhibitors, and one of them (AZD3965) already reached clinical trials in the cancer setting. After a first evaluation of compound tolerability, the trial is now set to evaluate the effect of AZD3965 in MCT1 positive tumors with pre-clinical positive results (146).

Information on the use of lactate by tumors could also be of value in cancer therapy. Van Hée et al. developed a PET tracer

### REFERENCES


of lactate [(±)- [18F]-3-fluoro-2-hydroxypropionate, [18F]-FLac] to monitor MCT1-dependent lactate uptake in tumors (147). The authors propose that this tracer can be used to predict response to treatments that disrupt lactate consumption, with potential to allow personalized patient treatment.

### DISCUSSION

Along the previous decades, the role of lactate has been overlooked, as it was seen as a mere metabolic waste of cell glycolytic metabolism. However, recent evidence has been revealing new and important oncogenic roles of lactate in malignant tumors. Lactate can function either as metabolic fuel for oxidative cells or as signaling molecule in the TME, being responsible for several aggressiveness cancer cell features, namely proliferation, migration and invasion, angiogenesis, escape to the immune system and resistance to therapy (**Figure 1**). Besides, upregulation of key proteins involved in lactate metabolism, namely LDHA and MCTs, have demonstrated clinical prognostic value and are seen as rational targets for cancer therapy. Thus, given the important role of lactate metabolism in cancer aggressiveness and response to therapy, lactate metabolism inhibitors should to be further explored in the clinical setting, especially in combination with classical therapy, molecular targeted drugs and immunotherapy.

### AUTHOR CONTRIBUTIONS

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

### FUNDING

This article has been developed under the scope of the project NORTE-01-0145-FEDER- 000013, supported by the Northern Portugal Regional Operational Programme (NORTE 2020) under the Portugal Partnership Agreement, through the European Regional Development Fund (FEDER), and through the Competitiveness Factors Operational Programme (COMPETE) and by National funds, through the Foundation for Science and Technology (FCT), under the scope of the project POCI-01-0145-FEDER-007038. JA and SG received fellowships from FCT, ref. SFRH/BPD/116784/2016 and SFRH/BPD/117858/2016, respectively.


Circulation, Respiration and Metabolism: Current Comparative Approaches. Champaign, IL: Human Kinetics (2014). p. 3-65.


associated with CD147 in cervical carcinoma. Disease Mark. (2009) 26:97–103. doi: 10.1155/2009/169678


mitochondrial dysfunction in cancer. Biochim Biophys Acta. (2017) 1868:246–57. doi: 10.1016/j.bbcan.2017.05.003


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

The handling editor declared a shared affiliation, though no other collaboration, with one of the authors FB.

Copyright © 2020 Baltazar, Afonso, Costa and Granja. 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.

# NAMPT and NAPRT: Two Metabolic Enzymes With Key Roles in Inflammation

Valentina Audrito\*, Vincenzo Gianluca Messana and Silvia Deaglio\*

*Laboratory of Tumor Immunogenetics, Department of Medical Sciences, University of Turin, Turin, Italy*

Nicotinamide phosphoribosyltransferase (NAMPT) and nicotinate phosphoribosyltransferase (NAPRT) are two intracellular enzymes that catalyze the first step in the biosynthesis of NAD from nicotinamide and nicotinic acid, respectively. By fine tuning intracellular NAD levels, they are involved in the regulation/reprogramming of cellular metabolism and in the control of the activity of NAD-dependent enzymes, including sirtuins, PARPs, and NADases. However, during evolution they both acquired novel functions as extracellular endogenous mediators of inflammation. It is well-known that cellular stress and/or damage induce release in the extracellular milieu of endogenous molecules, called alarmins or damage-associated molecular patterns (DAMPs), which modulate immune functions through binding pattern recognition receptors (PRRs), such as Toll-like receptors (TLRs), and activate inflammatory responses. Increasing evidence suggests that extracellular (e)NAMPT and eNAPRT are novel soluble factors with cytokine/adipokine/DAMP-like actions. Elevated eNAMPT were reported in several metabolic and inflammatory disorders, including obesity, diabetes, and cancer, while eNAPRT is emerging as a biomarker of sepsis and septic shock. This review will discuss available data concerning the dual role of this unique family of enzymes.

### Edited by:

*Alessandra Castegna, University of Bari Aldo Moro, Italy*

### Reviewed by:

*Georg F. Weber, University of Cincinnati, United States Lorena Pochini, University of Calabria, Italy*

#### \*Correspondence:

*Silvia Deaglio silvia.deaglio@unito.it Valentina Audrito valentina.audrito@unito.it*

#### Specialty section:

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

Received: *27 November 2019* Accepted: *02 March 2020* Published: *19 March 2020*

#### Citation:

*Audrito V, Messana VG and Deaglio S (2020) NAMPT and NAPRT: Two Metabolic Enzymes With Key Roles in Inflammation. Front. Oncol. 10:358. doi: 10.3389/fonc.2020.00358* Keywords: inflammation, cancer, signaling, metabolism, DAMPs, NAMPT, NAPRT, TLRs

## INTRODUCTION

One of the key roles of the innate immune system is to initiate immune responses against invasive pathogens. Pathogen-associated molecular patterns (PAMPs) include sugars/lipoproteins or nucleic acids [i.e., bacterial DNA as unmethylated repeats of dinucleotide CpG, double-stranded (ds) or single-stranded (ss) RNA] (1, 2). PAMPs can initiate immune responses through the activation of classical pattern recognition receptors (PRRs), among which there are toll-like receptors (TLRs), NOD-like receptors (NLRs), retinoic acid inducible gene I (RIG- I)-like receptors (RLRs), C-type lectin receptors (CLRs), multiple intracellular DNA sensors, and other non-PRRs DAMPs receptors (2–4). However, these receptors can be engaged also by endogenous ligands. It is now largely accepted that cells in conditions of hypoxia, acidosis, redox imbalance, hypertonic/hypotonic stress, and intracellular ion or cytoskeleton perturbations, can release small endogenous molecules called damage-associated molecular patterns (DAMPs) or sometimes "danger signals" or "alarmins," triggering immune responses through the activation of PRRs (4–6). Intriguingly, many of these DAMPs have a well-characterized intracellular function and have been serendipitously identified in the extracellular space where they initiate inflammatory responses, independently of pathogen infection, a phenomenon referred to as sterile inflammation (4, 7, 8). Similar to pathogen-induced inflammation, DAMPs can prime neutrophils, macrophages, and dendritic cells (DCs), but also non-immune cells, including endothelial and epithelial cells and fibroblasts (7). Activation of these cells leads to the production of several cytokines and chemokines, which in turn recruit inflammatory elements and trigger adaptive immune responses. Although sterile inflammation plays an essential role in tissue repair and regeneration, unresolved chronic inflammation is deleterious to the host leading to the development of metabolic, neurodegenerative, autoimmune disorders, and cancer (4).

Since their original definition as DAMPs in 2003, the list of endogenous molecules are increased considerably (4) and now includes high-mobility group box 1 protein (HMGB-1), heat shock proteins (HSPs), histone and extracellular matrix components (for example, hyaluronic acid and biglycan). All these molecules exert pro-inflammatory functions through binding to TLRs. HMGB-1 is among the most studied DAMPs. It is a nuclear DNA binding protein that can be found in the extracellular space not only as a consequence of necrosis, but also through dedicated secretion mechanisms (9, 10). Extracellularly, HMGB-1 elicits pro-inflammatory effects linked to consequent TLR4 binding and activation of the nuclear-factor kappa B (NFkB) signaling pathway (11, 12). In animal models, HMGB-1 is as a late mediator of lethal systemic inflammation, involved in delayed endotoxin lethality (13). Others DAMPs include F-actin, Sin3A associated protein 130 (SAP130), β-glucosylceramide and N-glycans binding to CLRs; monosodium urate (MSU) crystals, cholesterol crystals, β-amyloid (Aβ), and adenosine 5′ triphosphate (ATP) that activate NLRP3 inflammasome (4). In addition, numerous cytokines [i.e., interleukin (IL)-1β, tumor necrosis factor (TNF), and type I interferon (IFN-I)], proinflammatory proteins, such as interferon-induced protein 35, and bioactive lipids like lysophospholipids, are referred as "inducible DAMPs" or "conditional DAMPs" (14).

Nucleotides and nucleosides, for long time considered simply electron-shuttling agents involved in supporting energy metabolism, are gaining interest together with the network of enzymes that control their synthesis and degradation. Interestingly, while all these factors have a well-characterized intracellular function, they can be released in the extracellular space, where they bind and activate different sets of cellular receptors, including purinergic and PRRs. For example, ATP a key intracellular energy molecule, can be massively released by passive leakage when cells become injured, stressed, or even necrotic, acting as a DAMP (15). Extracellular ATP and its derivative nucleotides (adenosine, AMP, ADP) synthesized by endonucleotidases achieve many of their effects through purinergic receptors, via inflammatory cascades and the production of proinflammatory cytokines (16, 17). Among the enzymes involved in nicotinamide adenine dinucleotide (NAD) synthesis, nicotinamide phosphoribosyltransferase (NAMPT)—the focus of this review—emerges as new mediator of inflammation. Intracellularly, it catalyzes the first and ratelimiting step in the biosynthesis of NAD from nicotinamide (Nam) (18, 19). Increased eNAMPT levels are reported in conditions of acute or chronic inflammation (18, 20– 25). eNAMPT effects are mostly linked to the activation of an inflammatory signature mainly in macrophages, with recent data suggesting that it binds TLR4, therefore adding the enzyme to the number of "danger" signals activating this receptor (26). NAMPT is structurally and functionally related to a second NAD-biosynthetic enzyme (NBE), i.e., nicotinate phosphoribosyltransferase (NAPRT), which is rate-limiting in the NAD salvage pathway that starts form nicotinic acid (Na) (27–29). Our group recently discovered the presence of NAPRT in extracellular fluids (eNAPRT), highlighting a role also for this enzyme as a ligand for TLR4.

This review summarizes the current knowledge on NAMPT and NAPRT, as intracellular NBEs involved in the regulation/reprogramming of cellular metabolism, and as cytokines/DAMPs in the extracellular environment. Lastly, we will discuss the role of these enzymes especially in relation to the development of inflammatory conditions, including cancer, and their potential therapeutic values.

### NAD LEVELS MODULATE CELLULAR TRANSCRIPTIONAL RESPONSES AND METABOLIC ADAPTATION

Our knowledge on NAD biology has grown exponentially over the past few years, including biosynthetic and degrading pathways. A general decrease in cellular NAD is described in many age-related diseases, whereas increased NAD levels are associated to inflammatory conditions, including cancer. **Figure 1** illustrates the main NAD-biosynthetic and -consuming pathways, as well as the crosstalk between intracellular (i)NAD and eNAD.

### NAD: Energy Cofactor

As energetic co-enzyme, NAD is essential as electron acceptor donor in various metabolic pathways including cytosolic glycolysis, serine biosynthesis, tricarboxylic acid cycle (TCA), oxidative phosphorylation, as well as cell redox state homeostasis redox reactions (30, 31). Cofactor of almost 300 dehydrogenase, NAD is primarily used during glycolysis in the sixth step of the enzymatic chain by glyceraldehyde phosphate dehydrogenase (GAPDH) and at the end of the process by lactate dehydrogenase (LDH), catalyzing the interconversion of pyruvate and lactate and simultaneously of NADH and NAD. The final glycolytic product pyruvate, can be metabolized to produce acetylCoA by the pyruvate dehydrogenase complex (PDC), a reaction accompanied by NAD reduction to NADH (32). During the TCA cycle, NAD is reduced to NADH moieties in several key steps by isocitrate dehydrogenase (IDH), oxoglutarate dehydrogenase (OGD), and malate dehydrogenase (MDH). NADH produced in all these reactions, working as electron equivalent redistributor, is used by the electron transport chain (ETC) to generate ATP (33).

The ratio between NAD/NADH and their relative phosphorylated form (NADP/NADPH), are also critical for enzymatic defense systems against oxidative stress, regulating

FIGURE 1 | Intra/extra NAD interconnections and NAD-metabolizing enzymes. Schematic representation of the network of NAD-metabolizing cell surface and intracellular enzymes and their products. Several NAD precursors derived from diet can be internalized to generate iNAD, via NBE activities, to support energy metabolism, signaling, and other biological processes through the activities of intracellular NAD-consuming enzymes (PARPs and Sirtuins). These enzymes release Nam that, in turn, via NAMPT-dependent salvage pathway, support NAD production. Once in the extracellular space because of secretion/leakage, via Cx43, or because of direct extracellular synthesis from precursors (not confirmed), eNAD can function by binding purinergic receptors (P2Y, P2X), an event that leads to intracellular signaling and inflammatory conditions. Alternatively, eNAD can also be metabolized by a series of ecto-enzymes of the cell surface (CD38/CD157, ARTs, CD73, ENPP1) generating different metabolites (cADPR, ADPR, and NAADP) involved mainly in Ca2+-signaling. The end product of the reaction, adenosine, can modify signal transduction by acting on P1 purinergic receptors, generally leading to immunosuppression. In the square brackets are indicated the range of iNAD [200–500µM] or eNAD [100–500 nM]. Trp, tryptophan; Nam, nicotinamide; NR, nicotinamide riboside; Na, nicotinic acid; NBEs, NAD-biosynthetic enzymes; NAMPT, nicotinamide phosphoribosyltransferase; ARTs, mono adenosine diphosphate (ADP)-ribose transferases; PARPs, poly ADP-ribose polymerases; Cx43, connexin 43; ADPR, ADP ribose; cADPR, cyclic ADP ribose; NAADP, nicotinic acid adenine dinucleotide phosphate; NMN, nicotinamide mononucleotide; ADO, adenosine; ENPP1, ectonucleotide pyrophosphatase/phosphodiesterases.

redox homeostasis through the main cellular scavenging systems which are the glutathione (GSH/GSSG) and the thioredoxinmediated (Trx-SH/Trx-SS) mechanisms (34–37). In this context, NADPH is the indispensable reducing agent for ROS elimination and redox homeostasis, primarily produced by glucose-6 phosphate dehydrogenase (G6PD) and -phosphogluconate dehydrogenase (6PGD) in the pentose phosphate pathway (36).

## NAD: A Pleiotropic Signaling Molecule

Independently of its redox properties, NAD is also the substrate of enzymes with fundamental roles in gene expression and cell signaling (38). In these reactions, NAD is cleaved at the glycosidic bond between Nam and ADP-ribose acquiring the characteristic of signaling molecule (27).

The large family of NAD consuming enzymes includes mono adenosine diphosphate (ADP)-ribose transferases (ARTs) and poly ADP-ribose polymerases (PARPs), the NADdependent deacetylases, sirtuins (SIRT1-7), and the cyclic ADP-ribose hydrolases, NAD glycohydrolases, ectonucleotide pyrophosphatase/phosphodiesterases and ecto-5'-nucleotidase (CD38/CD157 and ENPP1/CD73) (19, 39, 40) (**Figure 1**).

Through their functional activities of post-translational modifications (ADP-ribosylation and deacetylation), or through the modulation of Ca2<sup>+</sup> signaling, these enzymes regulate gene transcription, cell differentiation, cell cycle progression, circadian rhythm, DNA repair, chromatin stability, cell adaptation to stress signals, and immune responses (41, 42). Therefore, PARPs and sirtuins represent connecting elements between the metabolic state of a cell and its signaling and transcriptional activities (43).

### Extracellular NAD and Its Biological Role

The eNAD concentration is in the range of 100–500 nM, considerably lower than its intracellular levels (200–500µM) (39, 44–46). eNAD and iNAD levels are highly linked, due to intraextra membrane transport of NAD precursors, intermediates of reaction and NAD itself (47). The canonical view is that NAD is unable to cross lipid bilayers, but it enters the cell using dedicated NAD transporters, such as connexin 43 (Cx43) channels, or exits through exocytosis (45, 48–50). In addition, conditions of environmental stress can induce NAD release (51–53). On the other hand, whether there is direct eNAD synthesis remains controversial (39) (**Figure 1**), despite the presence of extracellular precursors and biosynthetic enzymes. Specifically, it is known that among the different forms of vitamin B3 (NAD precursor), transport of Na is mediated by membrane carrier systems potentially including either a pH-dependent anion antiporter or a proton cotransporter (54, 55). Nam is present extracellularly and its uptake is possible either as direct transport in intact form or converted to salvage pathway metabolites. However, NAMPT's substrates ATP and 5-phosphoribosyl-1-pyrophosphate (PRPP) were shown to be unavailable in sufficient quantities in the extracellular space (56) to support direct eNAD generation.

eNAD can bind different subtypes of purinergic P2 receptors, including P2Y11, leading to the opening of L-type Ca2<sup>+</sup> channels and activation of a cAMP/cADPR/[Ca2+]i signaling cascade, ultimately causing increased proliferation and migration (57). In T cells and monocytes, P2X7 receptor activation generally results in Ca2<sup>+</sup> internalization, opening a non-selective, large membrane pore, causing cell death (58, 59). eNAD also acts as a neurotransmitter, released by stimulated terminals of mammalian central nervous system and peripheral nervous system neurons, binding to post-synaptic P2Y1 receptors, similar to ATP (60).

The very low levels of eNAD are due to its rapid metabolism/degradation by NAD-catabolic enzymes present on the surface of the cell (61), suggesting that also NAD metabolites may mediate cellular responses in the extracellular environment.

eNAD is degraded by three main classes of specific ectoenzymes: CD38 and CD157 (62, 63), ARTs (64), ENPP1 and CD73 (61, 65, 66). NADase, ENPP1 and CD73 can lead to the formation of adenosine (ADO), a potent immunosuppressant factor, independently of the activity of CD39 (61, 67, 68). Beside generating ADO, eNAD can be degraded to nicotinamide mononucleotide (NMN) by CD38, generating Nam which can cross plasma membranes and be re-converted to NAD through NAMPT and NMN adenylyltransferase (NMNAT) (69). On the other side, NMN can be also used by CD73, which generates nicotinamide riboside (NR) (66, 70), that, likely through equilibrative nucleoside transporters (ENTs), can be imported as NAD precursor (71, 72) (**Figure 1**). Recently, Slc12a8 was identified as specific NMN transporter (73), suggesting that NMN can be internalized without conversion to NR. Studies on cell type expression pattern of this transporter will clarify this possibility.

### NAD BIOSYNTHESIS: THE ENZYMATIC FUNCTIONS OF NAMPT AND NAPRT

NAD turnover within the cell is dynamic, displaying circadian oscillations that are regulated by the core clock machinery CLOCK:BMAL1 (74, 75). Total intracellular levels are maintained between 200 and 500µM, depending on the cell type or tissue, increasing in response to different stimuli (43). NAD homeostasis is the result of the balance between a number of NAD-consuming reactions and NAD-biosynthetic routes, via three distinct pathways: the de novo biosynthetic pathway, the Preiss–Handler pathway, and the salvage pathway, as reviewed in Houtkooper et al. (27), Ruggieri et al. (29), and Audrito et al. (42) and illustrated in **Figure 2**.

Specifically, de novo NAD biosynthesis starts with the catabolism of the amino acid tryptophan to kynurenine by indoleamine-2,3-dioxygenase. Kynurenine is then metabolized through the kynurenine pathway to quinolinic acid (QA), which is converted by quinolate phosphoribosyltransferase (QPRT), rate-limiting enzyme, to Na mononucleotide (NaMN). The Preiss–Handler pathway metabolizes kynurenine pathway– derived NaMN or diet-derived Na, or Na as a product of Nam deamidation by intestinal flora (76) to NAD, via NAPRT ratelimiting activity. In the salvage pathway, NAMPT metabolizes Nam and PRPP to NMN in a rate limiting step, which is then converted into NAD. In a further salvage route, NR, derived from diet, can be used by nicotinamide riboside kinase (NRK), to generate NAD (**Figure 2**).

Quantitatively, the Nam salvage pathway is the most relevant in mammalian cells. Several lines of evidence support this observation. First, Nam is the most abundant NAD precursor in the bloodstream (39), and can be easily introduced by diet (vitamin B3). Second, Nam is a by-product of all NADmetabolizing enzymes activity, increasing its availability (77). Third, the rate limiting enzyme NAMPT (EC 2.4.2.12) is expressed in all mammalian tissues (78), as detailed below. Linked to this, NAMPT gene deletion in mice is embryonically lethal (79), suggesting the importance of this route to regenerate NAD. In this pathway, Nam N-methyltransefase (NNMT) recently emerged as an evolutionarily conserved regulator of Nam availability. In fact, NNMT N-methylates Nam preventing its accumulation and inhibition of NAD-consuming enzymes, while on the other side, limiting its availability to NAMPT (80, 81).

The functional NAMPT forms a homodimer to catalyze the conversion of Nam and PRPP to NMN. Structural and site-directed mutagenesis studies by Khan et al. demonstrated that Asp219 is fundamental in defining the substrate specificity of NAMPT (82). Wang et al. showed that NAMPT has an autophosphorylation activity and hydrolyzes ATP. Autophosphorylation can increase its enzymatic activity (83). Recently, NAMPT was found to be a direct substrate of SIRT6 deacetylation, a post-translational mechanism that upregulates its enzymatic activity (84). On the contrary, mutations of His247, a central conserved residue in the active site of the enzyme, significantly decreases or abolishes NAMPT enzymatic activity (83).

NAPRT (EC 2.4.2.11) catalyzes the conversion of Na and PRPP to NaMN and pyrophosphate (PPi). The enzyme, originally named NaMN pyrophosphorylase, was first described by Handler in human erythrocytes, where it increases NAD levels (85).

NAPRT activity is more tissue-specific. Although enzyme activity can be detected in most mouse tissues (86), Na acts as a more efficient precursor than Nam in mice liver, intestine, heart and kidney (87). Furthermore, Na is more efficient than Nam in raising NAD levels in cells exposed to oxidative stress (56, 85, 88).

Contrary to NAMPT, NAPRT is not inhibited by NAD, which explains its significantly higher efficiency in raising NAD levels in vivo (56, 89). Moreover, NAPRT is strongly activated by phosphate (85), while ATP behaves as an allosteric modulator of the enzyme (29, 85, 90).

In 2015 Marletta et al. resolved the structure of human (h)NAPRT, highlighting a high degree of structural homology between the human and the bacterial NaPRTases due to evolutionary adaptation (91). As with NAMPT, the functional NAPRT enzyme works as dimers, and despite sharing very limited sequence similarity, hNAPRT shows a molecular fold that closely resembles that firstly described for hNAMPT (83). This opened new hypotheses of shared motifs in NAMPT and NAPRT involved in the binding of extracellular proteins to the receptor, as described in section eNAMPT Functions.

The presence of these multiple NAD biosynthetic routes most likely reflects differences in tissue distribution and/or intracellular compartmentalization of NBEs (39, 46, 76, 92, 93). Our group recently showed that NAMPT and NAPRT are mainly located in cytoplasm and nucleus, while NRK is more expressed in mitochondria, impacting on iNAD levels and response to NAMPT inhibitors (39, 46, 76, 92, 93).

### Identification, Characterization, and Expression of NAMPT and NAPRT NAMPT

The enzyme NAMPT is highly conserved with orthologs in bacteria (94), invertebrate sponges (95), amphibians (96), birds and mammals (97). Not long after its discovery in 1994 by Samal et al. as a pre-B-cell colony enhancing factor (PBEF) secreted by activated lymphocytes and bone marrow stromal cells, Rongvaux et al. (98) showed that murine PBEF could catalyze the conversion of Nam to NMN, a ratelimiting step in NAD biosynthesis. These authors also showed that Actinobacillus pleuropneumonia, a bacterium lacking the NadV gene, transformed with murine PBEF acquires NAD independence, confirming that the enzymatic activity is evolutionarily conserved from bacteria to mammals (98).

In recent years, NAMPT has received increasing attention due to new evidence indicating that it is a pleiotropic protein that may function as NBE, as well as growth factor, cytokine and adipokine [reviewed in (18, 25)]. Although NAMPT lacks the typical signal peptide needed for extracellular secretion, the mature protein can be found in the medium of many cellular cultures due to an active secretion mechanism (99, 100). However, in conditions of cell damage eNAMPT can be released as passive diffusion across cell membranes, as usual for other DAMPS. In addition, the 3' untranslated region (UTR) contains multiple TATT motifs that are characteristic of cytokines (99).

The human NAMPT gene spans over 34.7 kb on the long arm of chromosome 7 (7q22) and contains 11 exons and 10 introns (101, 102). Two distinct promoter sites are present in the 5'-flanking region, suggesting the possibility of tissue specific differential expression (101). The region proximal to the promoter is GC-rich and contains 12 binding sites for specificity protein 1 (SP-1), multiple activating protein 2 (AP-2), lymphoid enhancer-binding factor 1 (LF-1), cAMP response elementbinding protein (CREB), and signal transducer and activator of transcription (STAT) binding sites (101, 103). Furthermore, the presence of two hypoxia inducible factor (HIF) response elements (HREs) suggest that the gene is upregulated under hypoxic conditions (104). The distal promoter region contains several CAAT boxes and TATA-like sequences, as well as binding sites for nuclear factor 1 (NF-1), nuclear factor kappa-lightchain-enhancer of activated B cells (NF-κB), CCAAT/enhancer binding protein (C/EBPβ), the glucocorticoid receptor (GR), and activating protein 1 (AP-1) (101). The majority of these transcription factors, including NF-1, AP-1, AP-2, NF-κB, and STAT, regulate cytokine expression and their presence in the promoter region of NAMPT suggests a role for this enzyme in immunity (42, 105).

Recently, 65 kb downstream of the NAMPT transcription start site on chromosome 7 (hg19: 105,856,018–105,860,658), a putative NAMPT enhancer was identified as specifically marked by H3K27ac and/or an accessible DNase I hypersensitive (DHS) signal (106). Fine-mapping of the 4.6-kb putative enhancer by stepwise 1-kb deletions or insertions identified the 1-kb enhancer "B" region as responsible for (i) the control of expression of the NAMPT gene through c-MYC and MAX activities. (ii) In addition, it is the target of H3K27 acetylation; (iii) it regulates iNAD levels; and (iv) it is required for cell survival in NAMPTdependent tumors (106). Some genetic polymorphisms were identified in the human NAMPT gene, potentially responsible for NAMPT expression. Different representation of these Single Nucleotide Polymorphisms (SNPs) were described in patients with acute respiratory distress syndrome, type 2 diabetes, glucose and lipid metabolism alterations, diastolic blood pressure and hypertensive disorders compared to controls (107).

In the cell, NAMPT is abundant in the cytosol and present in the nucleus (108–110). Recently, Svoboda et al. showed that nuclear NAMPT translocation is a regulated process induced by genotoxic, oxidative, or dicarbonyl stress, mainly to finance NAD production for increased PARP and sirtuin activity (111). Moreover, NAMPT cytosol/nucleus localization changes according to cell cycle phases: it is excluded from the nucleus immediately after mitosis and it migrates back into it as the cell cycle progresses (111). These data were confirmed also by Grolla et al. that demonstrated a transport of NAMPT into the nucleus, GAPDH-mediated, in response to DNA damage (112). On the contrary, the presence of NAMPT in mitochondria remains controversial (30, 46, 109).

Furthermore, an increasing number of cell types have been shown to release eNAMPT, including adipocytes, hepatocytes, cardiomyocytes, activated immune cells and several tumor cells (100, 113–117). While it was shown that a regulated positive secretory process exists (79), the exact mechanisms of release are presently under investigation. The most accredited hypothesis, yet to be confirmed in most cell types, is that eNAMPT is secreted through a "non-classical" secretory pathway, which is not blocked by monensin and brefeldin A, two inhibitors of the classical endoplasmatic reticulum (ER)–Golgi secretory pathway (79, 113, 118, 119). A recent paper showed that eNAMPT is carried in extracellular vesicles (EVs) through systemic circulation in mice and humans. EV-contained-eNAMPT is internalized into cells, enhancing NAD synthesis (120). The same conclusion was obtained by another group identifying that eNAMPT is actively secreted via exosomes from microglia during neuroinflammation due to ischemic injury (121). However, this mechanism of secretion could be context dependent: in fact in 3T3-L1 adipocytes eNAMPT release and secretion do not appear to occur through microvesicles (113).

Whether the extracellular form possesses specific differences in terms of truncations or post-translational modifications is presently unclear. Different groups suggested that deacetylation by sirtuins can impact eNAMPT secretion (84, 122), adding a new layer of complexity.

### NAPRT

Highly conserved across species, the human NAPRT gene is located at chromosome 8q24.3, containing 12 exons. Similar to NAMPT, intracellular NAPRT is located in both the nucleus and the cytoplasm, but not detected in mitochondria (46, 71, 123). Our group firstly reported the presence of an extracellular form of NAPRT in biological fluids in physiological (healthy donor's blood) and inflammatory conditions opening a new field of investigations (124).

Several information about NAPRT expression and regulation emerged in tumors, in relation to the efficacy of NAMPT inhibitors (NAMPTi) as potential anti-cancer agents (125, 126), as described in the dedicated section NAMPT and NAPRT in Tumors.

### EXTRACELLULAR NAMPT AND NAPRT: ADIPOCYTOKINES AND DAMPS

In addition to a direct effect on NAD and its metabolites, the enzymes involved in synthesis of NAD also have important extracellular functions, as summarized in **Figure 3**.

### eNAMPT Functions

eNAMPT/PBEF was first identified as an immunomodulatory cytokine able to synergize with interleukin 7 (IL-7) and stem cell factor (SCF) to promote pre-B cell colony formation (99). It is now well-established that eNAMPT is a soluble factor that is up-regulated upon activation in innate and adaptive immune cells, including neutrophils, monocytes, and macrophages, and in epithelial and endothelial cells (18, 127). NAMPT expression can be rapidly induced by inflammatory signals, in particular both pathogen-derived lipopolysaccharide (LPS) and host-derived inflammatory stimuli (TNF-α, IL-1β, IL-6, leptin) in amniotic epithelial cells, macrophages, human osteoarthritic chondrocytes and a synovial fibroblast cell line (101, 103, 128, 129).

eNAMPT has a variety of biological functions (**Figure 3**): (i) it is an important mediator of inflammatory programs (18, 130) and (ii) it acts as a cytokine that modulates the

FIGURE 3 | Extracellular functions of NAMPT and NAPRT. iNAMPT and iNAPRT are involved in NAD generation inside of cells, but can be also secreted, through unknown mechanisms, in the extracellular space due to cellular stresses (damage/inflammation/pathological conditions). Extracellularly, they can act as adipocytokine/DAMP binding to TLR4 and triggering intracellular signaling promoting differentiation/polarization of myeloid cells, activation of inflammosome, secretion of pro or anti-inflammatory cytokines. The final outcome depends on the cellular context, for example in tumors eNAMPT creates an immunosuppressive microenvironment, favoring cancer progression, while eNAPRT in sepsis amplifies the inflammatory response. TLR4, Toll-like receptor 4; MD2, myeloid differentiation 2; MYD88, myeloid differentiation primary response gene 88; NAMPT, nicotinamide phosphoribosyltransferase; NAPRT, nicotinate phosphoribosyltransferase; NBEs, NAD-biosynthetic enzymes; Nam, nicotinamide; Na, nicotinic acid; DAMP, damage-associated molecular pattern; PARPs, poly ADP-ribose polymerases.

immune response (42). Notably, the cytokine-like functions appear, at least in part, independent of the protein catalytic activity, as inferred by the use of an enzymatically inactive NAMPT H247E mutant that retains the ability to activate signaling pathways (26, 83, 131, 132). In keeping with this view, NAMPT's substrates PRPP and ATP are apparently unavailable extracellularly to sustain its enzymatic activity (56). Following NAMPT treatment, interleukins IL-1β, IL-6, IL-10, and tumor necrosis factor- α (TNF-α) are up-regulated and secreted by peripheral blood mononuclear cells (PBMCs) and CD14<sup>+</sup> monocytes (133). Co-stimulatory molecules such as CD40, CD54, and CD80 are also up-regulated in response to eNAMPT exposure, an effect mediated through the activation of PI3-kinase and MAPKs pathways (133). Furthermore, in macrophages NAMPT increases matrix metalloproteinases (MMPs) expression and activity (134). In addition, (iii) eNAMPT has anti-apoptotic effects on immune cells, including neutrophils and macrophages, for example it promotes macrophage survival after induction of endoplasmic reticulum (ER) stress triggering IL-6 secretion and phosphorylation of STAT3 (103). (iv) eNAMPT is also reported as an adipokine, also known as visfatin, playing a critical role in the regulation of glucose-stimulated insulin secretion in pancreatic β cells (21, 135). While a direct role for insulin receptor in eNAMPT-mediated cytokine release was discarded (18, 133), eNAMPT is up-regulated in obese and diabetic patients: it is enriched in visceral fat and secreted by adipocytes (113, 136, 137). The role as adipokine seems more related to the extracellular generation of NMN: in fact systemic administration of NMN to aged mice or mice subjected to a high-fat diet restores normal NAD levels in white adipose tissue and liver, and ameliorates glucose intolerance and type II diabetic syndrome (138). (v) eNAMPT can also act as a pro-angiogenic factor, promoting endothelial cell proliferation, migration, and capillary tube formation in a concentrationdependent manner in human umbilical vein endothelial cells (HUVEC) (139–142). These proliferative effects of eNAMPT seem to be mediated, or at least partially mediated by vascular endothelial cell growth factor (VEGF), the master regulator of endothelial cell program (139). Thus, eNAMPT upregulates VEGF synthesis and secretion, as well as the expression of the VEGF receptor 2, which has been proposed to mediate the angiogenic actions of VEGF (139). Beside VEGF, in endothelial cells eNAMPT upregulates production of other pro-angiogenic soluble factors, such as fibroblast growth factor-2 (FGF-2), monocyte chemoattractant protein-1 (MCP-1) and IL-6 (143, 144). Indeed, both MCP-1 and FGF-2 have also been identified as mediators of eNAMPT-induced angiogenesis (143). Beyond in vitro studies, the angiogenic activities of eNAMPT were demonstrated in ex vivo and in vivo approaches (139, 140).

### eNAPRT Functions

Starting from the structural and functional similarity between human NAMPT and NAPRT (91), our group for the first time investigated whether NAPRT exists in an extracellular form, thus sharing with NAMPT its moonlighting abilities (124). By setting up a new luminex/ELISA assay, we dosed eNAPRT in a cohort of > 100 plasma from normal blood donors (HD), highlighting a mean concentration similar to that recorded for eNAMPT (in the range of 1.5–2.0 ng/ml), with no differences according to gender or age. We used mass spectrometry to confirm the presence of NAPRT peptides in human plasma. Moreover, we demonstrated that endogenous eNAPRT is enzymatically active (86, 124). Analyzing eNAPRT in sera from patients with acute or chronic inflammatory conditions, we demonstrated that this enzyme strongly increased in acute inflammatory diseases such as sepsis and septic shock, driving inflammatory responses related to the activation of macrophages (**Figure 3**). We also observed that cellular stress [i.e., treatment with TNF-α and cycloheximide to trigger apoptosis, or with ionomycin and carbonyl cyanide 3-chlorophenylhydrazone (CCCP) to trigger necrosis] is accompanied by marked increase of eNAPRT in macrophage culture media (124), as previously described also for HMGB-1 (9) and others DAMPs (145).

### eNAMPT/eNAPRT in Myeloid Cells Function: the Role of TLR4

NAD synthesis has a driving role in myeloid differentiation and in supporting macrophage inflammatory responses (42, 146, 147), prompting investigation on the function of NAMPT and NAPRT in these cells.

Increasing evidence demonstrated a direct role of NAMPT in regulating the differentiation program and the metabolic phenotypes of myeloid cells. Both iNAMPT and eNAMPT influence monocyte/macrophages differentiation, polarization and migration (132, 146, 148). We described a role for eNAMPT in creating an immunosuppressive and tumorpromoting microenvironment in chronic lymphocytic leukemia, where eNAMPT is important for the differentiation of monocytes toward tumor-supporting M2 macrophages (132). Recently, it was demonstrated that iNAMPT acts also on myeloid-derived suppressor cells (MDSCs), where NAMPT blocks CXCR4 transcription, via a SIRT1/HIF-1α axis. The activation of this circuit, in turn, leads to MDSCs mobilization and enhances the production of nitric oxide, promoting immunosuppression (149).

The NAMPT/NAD/SIRT1 axis seems to play a relevant role in myeloid cell activation. NAMPT-dependent NAD generation is crucial in the metabolic switch characterizing the transition from the early stage of acute inflammation, primarily relies on glycolysis, to the later adaptation phase more dependent on fatty acid oxidation (FAO) for energy production (150–152). Moreover, NAMPT/NAD levels significantly increased during activation of pro-inflammatory M1 macrophages (153). In a further feedback loop some cytokines, including IL-6 and TNF-α, induced during monocyte activation, are able to promote NAMPT expression via HIF-1α. In turn, NAMPT, triggering NF-kB signaling pathway, sustains IL6 and TNFA transcription forcing myeloid cell activation (131). It has been also shown that NAMPT/NAD/SIRT1 axis can regulate neutrophilic granulocyte differentiation via CCAAT/enhancer-binding protein α/β (C/EBPα/β) induction, ultimately, up-regulating granulocyte colony-stimulating factor (G-CSF). In turn, G-CSF further increases NAMPT levels (148). NAMPT inhibition, reducing NAD levels, thereby decreasing SIRT1 activity, leads to the dramatic elevation of acetylated C/EBPα levels and reduces amounts of total C/EBPα protein, accompanied by diminished mRNA expression of C/EBPα target genes (G-CSF, G-CSFR, and ELANE) (148, 154). Moreover, exposure of the acute myeloid leukemia cell line HL-60 to recombinant NAMPT or NAMPT overexpression induced myeloid differentiation of these cells per se (154).

A controversial issue in NAMPT biology is whether its cytokine-like properties are all linked to its enzymatic activity or are mediated by the binding to a cell surface receptor. In 2015, Camp et al. showed that eNAMPT produces robust TLR4 mediated NF-kB signaling activation, by directly binding TLR4- MD2 (26) (**Figure 3**). However, due to possible contamination of LPS, the natural ligand of TLR4, in the recombinant NAMPT preparations used to treat cells in vitro, the interpretation of these results remains controversial. Our group recently confirmed the binding of eNAMPT to TLR4 in macrophage cellular model (124), performing surface plasmon resonance (SPR) under the same conditions previously established for the NAMPT-TLR4 interaction (26). More recently, the same group that firstly identified TLR4 as NAMPT receptor published new details about this interaction (155). At the same time, a direct role of NAD in activating the inflammasome was recently reported by Yang et al. The authors demonstrated that NAD manipulation, using NAMPT inhibitors or the treatment with NAD precursors, affects TLR4-mediated NF-κB activation and PYD-domain 3 (NLRP3) inflammasome activity connecting intracellular NAD levels and inflammation (156).

Similar properties were attributed to eNAPRT. By using a surface coated with an anti-NAPRT antibody, we showed that a pre-mixed solution of recombinant (r)NAPRT and rTLR4 resulted in increased binding when compared to rNAPRT alone, indicating that a direct molecular interaction was occurring between the proteins (124). TLR4 triggering by rNAPRT activates an inflammatory signature in human macrophages differentiated from PBMC of healthy donors, as observed also using rNAMPT, promoting robust activation of NF-κB signaling, transcription and secretion of proinflammatory cytokines, including IL-1β, IL-8, TNF-α, CCL3, and inflammatory mediators such as caspase-1 (CASP1) and P2X purinoreceptor (124) (**Figure 3**). These effects are lost in TLR4-silenced macrophages. Accordingly, in macrophages, derived from TLR4−/<sup>−</sup> mice, rNAPRT exposure was not able to activate NF-κB signaling and cytokine production. Lastly, we demonstrated that the rNAPRT enzymatic deficient mutant is still able to trigger inflammosome in macrophages, indicating that the enzymatic activity is irrelevant to the pro-inflammatory functions of eNAPRT.

rNAPRT, as previously observed for eNAMPT (132, 146, 148), is also able to force monocyte differentiation into macrophages, up-regulating macrophage colony-stimulating factor (M-CFS). This function in triggering macrophage differentiation is a unique feature of eNAMPT/eNAPRT and not shared by LPS, suggesting that even though TLR4 is a receptor for multiple soluble factors and proteins, each specific ligand has a peculiar role. Notably, eNAPRT could be detected in macrophage culture supernatants, suggesting that macrophages are a source of eNAPRT in vivo.

Lastly, in this paper, we demonstrated that the signaling functions of hNAMPT and hNAPRT are not an evolutionarily conserved trait. In fact, the bacterial rNAPRT (PncB) or the bacterial rNAMPT (NadV) invariably failed to activate NFκB signaling in macrophages. Furthermore, a comparison of the surface properties of the bacterial and hNAPRT proteins revealed the presence in hNAPRT of an arginine-rich stretch (65**R**FL**R**AF**R**L**R**) forming a large mouth-like positively charged area on the top of the dimer, which is absent in the bacterial ortholog, but is present in a similar form in NAMPT, and could be involved in the binding to TLR4 (124). Even if several issues remain to be investigated, these data support the notion of another NBE acting as extracellular mediator with a direct role in macrophage functions, binding TLR4.

### NAMPT AND NAPRT AS BIOMARKERS OF CHRONIC AND ACUTE INFLAMMATORY DISEASES

iNAMPT over-expression as well as increased circulating levels of eNAMPT were documented in metabolic/inflammatory conditions including obesity, type 2 diabetes, metabolic syndromes, atherogenic inflammatory diseases, therefore supporting a role for eNAMPT as a potential biomarker of cardio- cerebro-vascular disorders (157–160). Enhanced eNAMPT levels are also described in kidney transplantation recipients (161), polycystic ovary syndrome (162), preeclampsia (163), and acute coronary syndrome (158, 164). Increased eNAMPT levels were additionally reported in non-metabolic chronic inflammatory diseases [i.e., osteoarthritis (103) and acute lung injury (ALI) (165, 166)], characterized by systemic inflammation. eNAMPT also seems to play a role in several types of infections like sepsis (167, 168) or intrauterine infection (chorioamnionitis) (169), and in autoimmune inflammatory diseases including psoriasis (170), rheumatoid arthritis (RA) (171) Crohn's disease (CD) and ulcerative colitis (UC) (172). **Table 1** summarized main activities of i/eNAMPT in these pathological conditions.

The first indication that NAPRT can be present in the extracellular space was published by our group in 2019. We dosed eNAPRT in sera from patients with sepsis or septic shock due to bacterial infections. Our results indicated that median eNAPRT levels picked-up to ∼25 ng/ml in septic individuals (compared to a median of about 2 ng/ml in HD), underlying high levels of this enzyme in this acute inflammatory condition (124). Circulating NAPRT has a role in mediating endotoxin tolerance at low/physiological doses, in fact the highest plasmatic eNAPRT levels were dosed in patients who died because of septic shock, while those with low concentrations survived. We confirmed a significant association between high levels of eNAPRT and mortality, suggesting that eNAPRT is a novel risk factor in sepsis (124). Even if the biological explanation behind this observation is still partly missing, findings in our work are significant starting points to evaluate the functional

#### TABLE 1 | NAMPT and NAPRT functions in chronic and acute inflammatory diseases.


*HD, healthy donors; Ab, antibody.*

role of eNAPRT as DAMP in sepsis, but also in others acute inflammatory conditions (**Table 1**).

### NAMPT and NAPRT in Tumors

In tumors increased i/eNAMPT have been reported, not only as biomarkers, but also as drivers of tumor progression (18, 25, 178), detailed in **Table 1**. Cancer cells require high energetic needs to support their proliferation. Increased demand of NAD, obtained through NAMPT overexpression, is needed to finance cellular metabolism and NAD-consuming reactions, including DNA repair activity (41). NAMPT is overexpressed in a broad range of solid tumors including colorectal, ovarian, breast, gastric, prostate, thyroid, pancreatic cancers, melanoma, gliomas, sarcoma, endometrial carcinomas, and hematological malignancies, as reviewed in Dalamaga et al. (178), Yaku et al. (179), Audrito et al. (25), and Chowdhry et al. (106). NAMPT, as intracellular and extracellular factor, exerts a direct role on tumor cells increasing tumor aggressiveness, correlating with worse prognosis and regulating different processes including metabolic adaptation, DNA repair, gene expression, signaling pathways, cell growth, invasion, stemness, epithelial to mesenchymal transition program, metastatization, angiogenesis, secretion of both proinflammatory and immunosuppressive cytokines, resistance to genotoxic stress, as reviewed in Dalamaga et al. (178) and Audrito et al. (25). Very recently, Nacarelli et al. described also a role of NAMPT in governing the strength of the proinflammatory senescence-associated secretory (SASP) phenotype observed during senescence, a process implicated in tissue aging and cancer (181).

Recently, amplification of NAPRT gene was detected in prostate, ovarian, and pancreatic cancers (106, 123). NAPRT gene amplification in tumors correlated with NAPRT expression in matched normal tissues, suggesting a role for tissue context in determining which cancers amplify NAPRT (106). Duarte-Pereira et al. in 2016 extensively studied expression of NAMPT and NAPRT in different tumor types and normal tissues (88). The initial step in that study was to evaluate NAPRT and NAMPT expression in a set of normal human tissues, highlighting a widespread expression for both genes. In tumors, while NAMPT was expressed at mRNA and protein levels in all samples analyzed, NAPRT protein levels were highly diverse, being undetected in several cases. Likewise, NAPRT protein is differentially expressed between cell lines, with markedly decreased expression in carcinoma cell lines MKN28 (gastric), 786-O (renal), HCT116 (colorectal), and in all leukemia cell lines tested (HL-60, NB4, and ML2) (88). Another paper highlighted a role for NAPRT, together with NAMPT, as negative prognostic marker in colorectal cancer, based on TCGA RNAsequencing data and protein tissue array (180). In a recent work the overexpression of NAPRT in ovarian cancer, correlated with a BRCAness gene expression signature. In this context, NAPRT silencing reduced energy status, protein synthesis, and cell size (123). These results suggest that both transcriptional and post-transcriptional mechanisms regulate the expression of the NAPRT gene in cancer types, including mutations in transcription factor binding sites of CREB and Sp1, to promoter methylation and alternative splicing (88). Epigenetic silencing of NAPRT, driven by the hypermethylation of CpG islands activity of mutant Protein Phosphatase Mg2+/Mn2<sup>+</sup> Dependent 1D (PPM1D), also known as Wip1, is a recently defined mechanism. As a consequence, PPM1D mutated tumors are particularly sensitive to NAMPTi (182). It was shown that the lack of NAPRT expression in some tumors, such as neuroblastoma, glioblastoma (183) or lymphomas (184), puts NAPRT as a biomarker for the use of Na as a chemoprotectant agent during treatment with NAMPT inhibitors (126). In NAPRT-negative tumors, NAMPT inhibition provides a novel synthetic lethal therapeutic strategy by inducing metabolic stress, while normal cells are rescued by Na via activation of the NAPRT pathway (123, 183–185).

We demonstrated the presence of eNAPRT in sera from patients with a diagnosis of cancer, including solid tumors (prostate, lung and bladder cancer, mesothelioma and metastatic melanoma) and hematological malignancies [myeloma, chronic lymphocytic leukemia (CLL), and diffuse large cell lymphoma (DLCL)] (124) as summarized in **Table 1**. The median value of circulating eNAPRT is double compared to HD, suggesting a possible role of this enzyme in tumor microenvironment.

Several issues remain to be addressed. First and foremost, it will be important to understand the relationship between eNAPRT and eNAMPT: our findings suggest that they have multiple roles in acute vs. chronic inflammation, engaging TLR4 in different pathological conditions (**Table 1**) and alerting the immune system to distinct sets of "dangers."

### IMPLICATIONS FOR THERAPY AND CONCLUDING REMARKS

NAMPT inhibitors were primarily developed as anticancer agents, depleting NAD and causing metabolic crash and tumor cell death (**Table 1**).

For iNAMPT selective pharmacological inhibitors exist, the best studied being FK866 (also known as APO866) and GMX1778 (also known as CHS-828), among others (25, 178, 186–189). These inhibitors have been studied in cancer cell lines and animal models showing cytotoxicity and tumor regression (178, 190). Despite these important results in vitro and in vivo, phase I clinical trials in advanced solid tumors and leukemia showed no objective tumor remission and toxicity (191, 192). One of the known mechanism leading to the partial failure of NAMPTi treatment is due to the concomitant expression of NAPRT (88, 106, 123, 125, 126, 184), that can overcome NAMPT inhibition. A complete analysis of expression of these two NBE in tumors should be made to design better therapeutic strategies that deplete NAD improving efficacy. Development of novel NAPRTi, to obtain complete depletion of NAD in tumor insensitive to NAMPTi due to the overexpression of NAPRT should also be considered. Previous studies indicated the ability of structural analogs of Na to inhibit NAPRT enzymatic activity (85, 89). Among this class of compound, 2-hydroxinicotinic acid (2-HNA) is the most promising, showing significant inhibition of NAPRT enzymatic activity and function in ovarian cancer in vitro and in xenograft models (123). The use of NAPRT inhibitors appears as a promising strategy to overcome NAPRT-mediated resistance to NAMPT inhibitors in patients (**Table 1**).

Blocking the extracellular cytokine-like function of eNAMPT and eNAPRT would be very useful to restore immune competence in cancer, as well as, infection setting (**Table 1**). In the tumor microenvironment, neutralization of eNAMPT using blocking antibodies could be effective to repolarize myeloid cells (TAMs/MDSCs) against tumor. Some groups/companies are working on the production of these antibodies (193), hypothesizing a combination strategy with immunotherapy, or a double inhibition of i/eNAMPT. Blocking eNAPRT in acute inflammatory conditions, such as in septic patients, could be an important strategy to prevent the damaging action of a massive secretion of eNAPRT leading to decreased survival of patients, but this remains, at this moment, only a speculative hypothesis.

In conclusion, in this review we summarized current knowledge on these two old enzymes involved in NAD biosynthesis that can powerfully modulate immune responses. If NAMPT has now an acknowledged role in regulating several cellular processes in physiological and pathological conditions,

### REFERENCES


and as biomarker in several diseases, the biology of NAPRT, especially as new soluble factor, acting as DAMP in acute inflammation, needs to be extensively studied to determine potential pharmacological settings.

### AUTHOR CONTRIBUTIONS

VA and SD have made a substantial, direct and intellectual contribution to the work, contributed equally to writing the manuscript, and approved it for publication. VM contributed to writing the final version of the manuscript and approved it for publication.

### FUNDING

This work was supported by Gilead Fellowship program 2018, by the Ministry of Education University and Research-MIUR, PRIN Project 2017CBNCYT and Progetto strategico di Eccellenza Dipartimentale #D15D18000410001 (the latter awarded to the Dept. of Medical Sciences, University of Turin) and ITN INTEGRATA program (grant agreement 813284).


of CD39 drives a novel adenosinergic loop in human T lymphocytes. Oncoimmunology. (2013) 2:e26246. doi: 10.4161/onci.26246


like receptor 4 and mediates inflammation. Nat Commun. (2019) 10:4116. doi: 10.1038/s41467-019-12055-2


diagnostic and therapeutic perspectives and challenges. Metabolism. (2018) 82:72–87. doi: 10.1016/j.metabol.2018.01.001


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

Copyright © 2020 Audrito, Messana and Deaglio. 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.

# Nutritional Exchanges Within Tumor Microenvironment: Impact for Cancer Aggressiveness

#### Giuseppina Comito<sup>1</sup> , Luigi Ippolito<sup>1</sup> , Paola Chiarugi 1,2 \* and Paolo Cirri 1,2 \*

*<sup>1</sup> Department of Experimental and Clinical Biomedical Sciences, University of Florence, Florence, Italy, <sup>2</sup> Excellence Center for Research, Transfer and High Education DenoTHE, University of Florence, Florence, Italy*

Neoplastic tissues are composed not only by tumor cells but also by several non-transformed stromal cells, such as cancer-associated fibroblasts, endothelial and immune cells, that actively participate to tumor progression. Starting from the very beginning of carcinogenesis, tumor cells, through the release of paracrine soluble factors and vesicles, i.e., exosomes, modify the behavior of the neighboring cells, so that they can give efficient support for cancer cell proliferation and spreading. A mandatory role in tumor progression has been recently acknowledged to metabolic deregulation. Beside undergoing a metabolic reprogramming coherent to their high proliferation rate, tumor cells also rewire the metabolic assets of their stromal cells, educating them to serve as nutrient donors. Hence, an alteration in the composition and in the flow rate of many nutrients within tumor microenvironment has been associated with malignancy progression. This review is focused on metabolic remodeling of the different cell populations within tumor microenvironment, dealing with reciprocal re-education through the symbiotic sharing of metabolites, behaving both as nutrients and as transcriptional regulators, describing their impact on tumor growth and metastasis.

Keywords: tumor microenvironment, OXPHOS, EVs, EMT, CAFs

## INTRODUCTION

### Tumor Microenvironment

A solid tumor is a dysfunctional neoplastic tissue characterized by uncontrolled growth and chaotic histological organization and it is composed, in addition to cancer cells, by heterogeneous subsets of non-transformed cells, such as mesenchymal stem cells, fibroblasts, endothelial cells, adipocytes and immune cells, establishing a complex tumor microenvironment (TME) with peculiar structural and biophysical characteristics [i.e., altered extracellular matrix (ECM) composition, acidity and hypoxia]. The features of the neoplastic parenchyma are well instructed through a complex interplay between cancer and stromal cells, orchestrated by soluble factors, metabolites, extracellular vesicles (EVs), as well as cell-to-cell interaction.

In physiologic conditions fibroblasts are the main cellular component of connective tissue and they are involved in providing structural scaffolding and trophic ancillary function for the epithelial cells of the tissues (1). In tumors, cytokines released by cancer cells convert fibroblasts into a permanently activated, myofibroblast-like, form called cancer associated fibroblasts (CAFs) (2). This chronic activation of fibroblasts within TME is crucial for cancer progression. Indeed, CAFs

### Edited by:

*Paolo E. Porporato, University of Turin, Italy*

#### Reviewed by:

*Cinzia Domenicotti, University of Genoa, Italy Sofia Avnet, Rizzoli Orthopedic Institute (IRCCS), Italy*

### \*Correspondence:

*Paola Chiarugi paola.chiarugi@unifi.it Paolo Cirri paolo.cirri@unifi.it*

#### Specialty section:

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

Received: *27 November 2019* Accepted: *05 March 2020* Published: *24 March 2020*

#### Citation:

*Comito G, Ippolito L, Chiarugi P and Cirri P (2020) Nutritional Exchanges Within Tumor Microenvironment: Impact for Cancer Aggressiveness. Front. Oncol. 10:396. doi: 10.3389/fonc.2020.00396*

**117**

are responsible for an abnormal ECM deposition and remodeling, for a persistent inflammation mediated by soluble factors (i.e., cytokines) (SDF-1, CXCL14, etc.) leading to new vessels formation and recruitment of immune cells within the TME, events particularly important for the nutrients supply and metastatic dissemination, respectively (3). CAFs also exert an immunomodulating role, mainly by enhancing the M2/M1 macrophage and the Th2/Th1 ratio (4, 5). Besides their immunomodulating and pro-angiogenic activity, CAFs are able to promote epithelial–mesenchymal transition (EMT) in cancer cells, conferring them proinvasive and stem-like features (6). Finally, CAFs play a mandatory role in cancer cells dissemination, since they can escort metastatic cancer cells in the bloodstream, favoring their implantation at distal sites (7).

Mesenchymal stem cells (MSCs) are multipotent stromal cells recruited into TME mainly from adipose tissue and bone marrow in response to several growth factors, i.e., platelet derived growth factor (PDGF), vascular endothelial growth factor (VEGF), transforming growth factor-β (TGF-β) as well as EVs released by cancer cells (8). MSCs possess self-renewal ability and are able to differentiate into several cell types within TME, such as CAFs. For example, in neuroblastoma tumor CAFs share phenotypic and functional characteristics with bone marrow-derived MSCs (9), while in vitro conditioning of bone marrow-derived MSCs cells with tumor-derived medium lead to the acquiring of a CAF-like phenotype sustaining tumor growth both in vitro and in vivo (10). In addition to the protumorigenic functions, broadly shared with CAFs, MSCs show an immunosuppressive role in both prostate and melanoma cancer models (11, 12).

Neo-angiogenesis, the growth of new blood vessels from the existing vasculature, is a key step in tumor progression. Under the stimulation of pro-angiogenetic cytokines, endothelial cells within TME provokes a wide but disorganized rearrangement of vessel architecture characterized by altered permeability which is crucial for tumor cells metastatic spreading. In addition, tumor endothelial cells can secrete angiocrine factors, such as CSF-1 or interleukin (IL)-8 (13) promoting cancer cells migration along with neutrophils infiltration, hence widening their functions in tumor progression (14).

Tumor- or CAF-derived cytokines also are able to induce monocytes recruitment within the tumor mass where they were activated to M1-like macrophages by CSF-1 and IFN-γ (15). Conversely, macrophages stimulation with type 2 T helper cell cytokines, such as IL-4 and IL-10, leads to phenotype called M2-like endowed with pro-tumor characteristics, likely taking part in all steps of the metastatic route (16).

Tumor-associated neutrophils (TANs) are divided into two sub-populations, showing antitumoral activity (N1-like phenotype) or protumoral activity (N2 phenotype). Neutrophils, recruited in TME by CXCL2 and CXCL5 cytokines, actively participate in the metastatic process by enhancing tumor cell expression of pro-metastatic genes (15), as well as associating with circulating breast tumor cells, helping them to proliferate once they reach the secondary site (11).

Many lymphocytes subtypes are present in TME as CD4<sup>+</sup> helper cells, immunosuppressive regulatory T-cells (Tregs) and CD8<sup>+</sup> cytotoxic T-cells, recruited by several chemokines produced by cancer and stromal cells. The histological origin, the composition and the density of the cells that constitute tumor-infiltrating lymphocytes, together with hormonal context within TME can determine tumor progression and clinical outcome. For example, a lot of evidence have addressed the role of cytotoxic CD8<sup>+</sup> T cells, whose presence and activity is associated with a good prognosis, while the infiltration of Tregs, an immunosuppressive T cell subpopulation, has been shown to be associated with poor prognosis in several tumors (17).

Adipocytes are recently emerging as important contributors to cancer progression. Cancer cells through chemokines secretion can convert adipocytes into their activated form Cancer-Associated Adipocytes (CAAs), that has been reported to promote IL-6-mediated EMT in cancer cells (18). CAAs-secreted leptin has a proliferative effect on cancer cells (19) as well as a pro-angiogenic role (20). On the contrary, adiponectin secretion decreases in CAA with respect to normal adipocytes, suggesting an anti-proliferative effect of this adipokine on cancer cells (19).

Recent advances in tumor biology showed the importance of a highly tuned exchange of nutrients within TME, impacting on tumor progression (21). A consequence of the cytokinesmediated cross-talk between cancer and stromal cells is the metabolic reprogramming of all cellular components of the tumor, aimed at maximizing the proliferative capacity of tumor cells. In this view CAFs, which are the major component of tumor stroma, together with adipocytes, give a feed support to tumor cells, increasing their growth rate. In addition, some nutrients exchanged in the TME, also play an essential signaling function acting as epigenetic switches, leading to activation of EMT and inhibition of immune cell response. The multifaceted significance of nutrients exchange is discussed in the chapters below.

### Metabolic Deregulation in Cancer

A tumor, consisting of a heterogeneous and complex network of cancer and stromal cell populations, needs to adapt all the metabolic functions to support the demands of uncontrolled growth and to support disease progression. The metabolic alterations of a tumor come from both the oncogenic signaling that orchestrate distinct metabolic pathways and the environmental context that promotes nutrient-based intercellular cross-talk and/or competition.

Actually, it is widely recognized that cancer cells need to meet their bioenergetic and biosynthetic demands to maintain a high tumor cell growth rate. Tumor cells require a high rate

**Abbreviations:** α-KG, α-ketoglutarate; CAAs, Cancer Associate Adipocytes; CAFs, cancer associated fibroblasts; ECM, extracellular matrix; CSF, colony stimulating factor; CXCL, chemokine (C-X-C motif) ligand; EMT, epithelial-tomesenchymal transition; EVs, extracellular vesicles; FAs, fatty acids; FAO, fatty acid oxidation; HIF-1-α, hypoxia-inducible factor 1-α; IFN, interferon; IL, interleukin; LDs, lipid droplets; MCT, monocarboxylate transporter; MSCs, mesenchymal stem cells; OXPHOS, oxidative phosphorylation; PDGF, platelet derived growth factor; SDF-1, Stromal cell-Derived Factor-1; SLCs, solute carriers; TAMs, tumor associated macrophages; TGF-β, transforming growth factor-β; TCA, tricarboxylic acid cycle; TME, tumor microenvironment; VEGF, vascular endothelial growth factor; VEGF, vascular endothelial growth factor receptor.

of biosynthesis of macromolecules (lipids, amino acids, nucleic acids) in order to maintain the cellular redox balance and, at the same time, to compensate their energy-consuming processes, ultimately culminating in fueling tumor growth and progression. However, the metabolic reprogramming of cancer cells is crucial also for the signaling role exerted by the metabolites.

Intriguingly, the metabolic flux is mainly derived from the glucose in cancer cells, known as the Warburg effect, that is the ability of cancer cell to massively upload glucose, thanks to the upregulation of glucose transporters GLUT1-3, in order to (i) provide precursors and intermediary metabolites, useful for the tumor-associated biosynthetic machinery, and to (ii) produce high amounts of lactate, even in the presence of oxygen. Warburg metabolism is one of the most commonly observed examples of metabolic reprogramming in highly proliferating cells, such as cancer cells and non-transformed cells (i.e., T lymphocytes), taking advantage from the rapid production of ATP and the synthesis of glucose-derived macromolecules (22).

The collateral metabolic fluxes arising from aerobic glycolysis lead to the activation of specific pathways such as the pentosephosphate-pathway (PPP) and the one-carbon metabolism. PPP is important for tumor cells as it generates pentoses useful for DNA/RNA synthesis and feeds the nicotinamide-adenine dinucleotide phosphate (NADPH) pool, which is needed for fatty acid synthesis and cell survival under oxidative stress conditions. The harsh TME, as well the oncogenic background, are responsible of the increase of reactive oxygen species (ROS) in tumor cells. These highly reactive molecules can detrimentally modify the intracellular environment as well as activate certain pro-tumoral signaling pathways, under certain sub-toxic levels. Accordingly, to challenge the toxic levels of ROS, tumor cells increase their antioxidant capacity to allow cancer progression and PPP activation is oriented in such way. Oxidative stress can be counteracted by the production of NADPH by the oxidative branch of the pentose phosphate pathway, as it is used by the glutathione reductase enzyme in the reduction reaction of oxidized glutathione (GSSG). To note, glutathione (GSH) is the one of most important antioxidant molecule within the cell, it is synthesized from glutamine carbons and conditions of oxidative stress increase the conversion of GSH (reduced, physiological form) to GSSG (oxidized), which is potentially toxic for the cell, as it acts as a pro-oxidant (23). The deregulation of glutathione metabolism is broadly identifiable in the majority of cancers as the genes involved in GSH turnover or utilization are under the transcriptional control of classical tumorigenic pathways, primarily the nuclear factor erythroid 2-related factor 2 (NRF2) signaling which drives the antioxidant response and control the transcription of glutamate-cysteine ligase, the first enzyme of the cellular GSH biosynthetic pathway. In addition, the hypoxic signaling is a driving force for the activation of GSH production and it has been associated with the enrichment of breast cancer stem cell niche following chemotherapy treatments (24). GSH alterations have been identified in metabolically deregulated tumors, such as tumors deficient for fumarate hydratase enzyme. Strikingly, the accumulation of fumarate in FH-deficient cancer cell lines leads to the formation of a peculiar molecule between fumarate and glutathione (GSH), which depletes intracellular NADPH and enhances oxidative stress (25). Also, in MYC-driven liver tumors, particular for a decreased incorporation of glutamine, the attenuation of expression of glutamate-cysteine ligase contributes to GSH depletion (26). A key role of glutathione is also emerging in the context of tumor microenvironment. In particular, CAFs were shown to diminish the accumulation of genotoxic agents in cancer cells in a glutathione-dependent manner. In fact, CAFs release high levels of thiols, including glutathione and cysteine, which increase intracellular GSH levels in tumors counteracting drug-dependent oxidative stress and apoptotic response (27, 28). Furthermore, glycolysis can divert glucose-derived intermediates to one-carbon pathway that is important for serine synthesis (29). It supplies methyl groups to the one-carbon and folate pools, contributing to amino acid and nucleotides synthesis, methylation reactions, and NADPH production. Finally, the Warburg-associated fermentation of pyruvate into lactate, catalyzed by lactate dehydrogenase A enzyme, culminates in its extrusion in the extracellular milieu via the monocarboxylic acids transporter MCT4. Lactate release, coupled with H+, increases external acidity and deliver to TME a peculiar molecule losing its classification as waste product, as it plays both a signaling and a metabolic role, thereby altering the immune cell landscape, increasing tumor invasive capacity and supplying an appealing carbon source for other cell populations (30) (see below).

Warburg metabolism is an aspect of a highly multilayered cancer metabolism, as cancer cells have adapted multiple mechanisms to exploit metabolic substrates through mitochondria. To note, many reactions of the tricarboxylic acid (TCA) cycle are reversible and multiple mitochondriaassociated anaplerotic circuitries ensure such a metabolic adaptation of cancer cells.

In addition to glucose-derived pyruvate, fatty acids (FAs) and amino acids can feed the TCA cycle to sustain mitochondrial activity in malignant cells and produce ATP via oxidative phosphorylation. A key role of the TCA cycle in proliferating cells is to act as a biosynthesis hub and this function differs from that occurring in non-proliferating cells, where TCA cycle serves to provide the maximal ATP production. During tumor cell proliferation, however, much of the carbon that enters the TCA cycle is used in biosynthetic pathways. In this scenario, tumor mitochondrial metabolism represents a cataplerotic center by providing building blocks for anabolic processes. Synthesis of lipids (fatty acids, cholesterol, and isoprenoids) is a crucial example of cataplerosis in tumor cells and the activation of lipid biogenesis could play an active role in cell transformation and cancer development, as lipids have important roles in membrane structure, cellular signaling and protein regulation, beyond energetics. Glucose is a major lipogenic substrate as it can be oxidized and mediates the transfer of mitochondrial citrate out to the cytosol to be converted to oxaloacetate (OAA) and the lipogenic precursor acetyl-CoA, which can either be used for fatty acid and cholesterol synthesis or for epigenetic purposes (i.e., acetylation reactions), by providing a pool for chromatinmodifying enzymes such as the acetyltransferases (31). However, the biosynthesis of fatty acid chains, upon conversion of citrate to acetyl-CoA via ATP-citrate lyase (ACLY), is sustained by the carboxylation of cytosolic acetyl-CoA by acetyl-CoA carboxylase (ACC) to produce malonyl-CoA, which is further assembled into long fatty acid chains by fatty acid synthase (FASN). As ACLY, ACC, and FASN are frequently upregulated in tumor cells and their inhibition reduces tumor growth, it is widely recognized that the increased capacity for producing lipids de novo is a crucial determinant for the tumor progression. In addition, cholesterol synthesis plays a role in the tumor malignancy, as the interference with such pathway through statins treatment provokes a detrimental effect on tumor growth in vitro and in vivo (32).

The major anaplerotic substrate in growing cells is glutamine, the most rapidly consumed nutrient by many human cancer cells. Indeed, most of them display addiction to glutamine, thereby boosting its uptake mainly through SLC1A5/ASCT2 transporter, and its catabolism (glutaminolysis) via the activity of mitochondrial glutaminolytic enzymes, glutaminase and glutamate dehydrogenase. Glutamine entry and metabolism is mainly supported by c-Myc, a transcription factor upregulated in several cancer cells. To this end, c-Myc induces the transcription of glutamine transporters, and of glutamine-utilizing enzymes, such as glutaminase, phosphoribosyl pyrophosphate synthetase and carbamoyl-phosphate synthetase 2.

Glutamine is important for energetic demands of cancer cells as it provides carbons to replenish the TCA cycle. However, it also provides nitrogen for biosynthesis of purine and pyrimidine nucleotides, as well as of nonessential amino acids. Glutamine metabolism also contributes to the production of glutathione, thus playing a role in the cellular anti-oxidant defense, and serves as a precursor to lipid synthesis via α-ketoglutarate (KG)-tocitrate conversion namely reductive carboxylation (33, 34). In addition, many epigenetic modifications and cellular processes are regulated by glutamine-derived α-KG, which is a cofactor of dioxygenase enzymes, including the ten eleven translocases (TET) family and the Jumonji (JMJ) family, thereby affecting, respectively, DNA and histone demethylation (35).

In keeping, in addition to catabolic, energetic and anabolic requirements for cancer growth by exploiting TCA cycle, an intracellular signal transduction cascade is mediated by other TCA cycle metabolites (36, 37). In tumors harboring the loss of the mitochondrial enzymes succinate dehydrogenase or fumarate hydratase, the respective accumulation of succinate or fumarate has been shown to inhibit the enzymatic activity of α-KGdependent dioxygenases. Hence, these enzymes are important for different purposes such as hypoxia inducible factor (HIF)-1 stability, as well as epigenome rewiring. In keeping, in tumors that have lost succinate dehydrogenase or fumarate hydratase, HIF-1 is activated under normoxic conditions, resulting in the activation of pseudohypoxic pathways and in the enhancement of tumor malignancy (38).

Thus, beyond the genetic alterations, the ability of tumor cells to engage different metabolic behaviors according to the metabolic scenario provided by the microenvironment (oxygen levels, austere availability of nutrients, stromal cues) greatly contributes to a high metabolic plasticity which consequently increases tumor heterogeneity. Indeed, a tumor cell needs to face the environmental scenario, by displaying a metabolic plasticity useful to orchestrate the selection, the upload and the consequent exploitation of the available nutrients in the TME. Thus, the metabolic reprogramming occurring in a cancer cell encompasses multiple strategies, among which is a non-cell autonomous one, mainly involving the tumor-associated stromal components that supply nutrients and establish metabolic networks with tumor cell compartment, thereby shaping their malignant phenotype.

### Nutrients Exchanged in TME

Besides the metabolic reprogramming of a tumor cell harboring high mutational landscape, recent discoveries have highlighted nutrients available in the TME as crucial molecules acting on the acquisition of a peculiar metabolic and phenotypic plasticity in tumor cells allowing them to adapt to the peculiar features of the TME they face with (e.g., cytokine delivery, oxidative, acidic, and nutritional stress). Tumor-associated stromal and tumor populations dynamically communicate each other through metabolic connections, causing a reciprocal tumor-stroma metabolic interplay. Such metabolic symbiosis reasonably have the advantage to supply each other different metabolites able to reprogram anabolic and catabolic processes in the recipient subpopulations (**Figure 1**). Of note, nutrients arisen by stromal populations can overcome metabolic constrains within the tumor, circumventing oncogenes or tumor suppressors regulation of several metabolic enzymes, thus rewriting cancer mass evolution.

Several nutrients will be exchanged within TME, as explained below.

### Lactate

Mitochondrial exploitation of lactate over glucose has been reported in human lung tumors, highlighting the contribution of stroma for anabolic purposes and TCA replenishment driven by such metabolite (39). A clear lactate-based tumorstroma cross-talk has been reported in several tumor models, including the prostate carcinoma. CAFs predominantly exhibit aerobic glycolysis and secrete lactate through monocarboxylate transporter (MCT)-4, whose expression is under the redox or succinate-dependent HIF-1 control (40, 41). Cancer cells educate CAFs to secrete lactate, exploiting directly the environmental lactate, uploaded through MCT-1. Once imported, lactate is able to rewire cancer cell metabolism, causing a shift from glycolysis to oxidative phosphorylation (OXPHOS) (42). The inward of stromal lactate provokes the unbalance of NAD+/NADH ratio (see the lactate-to-pyruvate conversion and its oxidation), causing ad hoc adaptive changes in cancer cells, such as sirtuin1-mediated de-acetylation/activation of the transcriptional co-activator peroxisome proliferator-activated receptor-gamma coactivator-1 (PGC-1α). This molecular signature has been found as crucial for the enhancement of the tumor mitochondrial mass and function of stroma-reprogrammed prostate cancer cells, as reported in other models of tumor progression (43, 44). Moreover, the simultaneous increase of the GLUT-1 carrier in CAFs, as well as activation of the mitochondrial pyruvate dehydrogenase complex, concur to significantly reprogram the metabolism of both tumor and stromal compartments

establishing a metabolic symbiosis (45, 46). The in vitro definition of such metabolic symbiosis has been confirmed in vivo by recent isotope tracer measurements, showing a rapid exchange of lactate between the tumor and circulation in several cancer models (39, 47). Interestingly, the MCT-mediating lactate influx and efflux activity involves protons (H+), thereby leading to an apparent paradox. As several accessory cells in TME concur to decrease extracellular pH [due to the overexpression of carbonic anhydrases or proton pumps, commonly occurring in both cancer cells and CAFs (48, 49)], lactate/H+-coupled transport by MCTs tends to drive lactate from the interstitium into tumor cells (50), allowing them to anabolize this nutrient (51). Furthermore, the physical association between carbonic anhydrases (CA) and lactate transporters MCT-1/4 plays a key role in the regulation of the directional flux of lactate, as well as in the protonation/deprotonation of proteins, thereby affecting their functions. CAII is associated to the cytosolic part of the MCT transporters, while CAIX is associated to the extracellular face and MCT-assembled CD147 chaperone. MCT-associated CAs non-enzymatically cooperate to drive the lactate/H<sup>+</sup> symport/export in/by cancer cells. Indeed, protons diffuse very slowly within the cell (52); for this reason, in order to allow a more efficient extrusion of H<sup>+</sup> and lactate from the cell, the MCT does not extract protons directly from cytosol, but rather from protonated residues located in a peculiar antenna of CAII. Similarly to the cytoplasm, the diffusion of H<sup>+</sup> in the extracellular space is restricted and protons have to be removed from the extracellular side by CAIX, shuttled to protonated residues and then released to the extracellular space. Hence, lactate bi-directional flux is strictly linked to acidity, both intracellular and extracellular, and mostly to CAs activity. Fascinatingly, as these enzymes links acidity to 1-C metabolism, catalyzing the formation of both H<sup>+</sup> and HCO<sup>−</sup> 3 , it is conceivable that lactate flux leads to regulation of one carbon metabolism as well. Importantly, besides its role in the modulation of MCT-mediated lactate transport into cancer cells, the high amount of H<sup>+</sup> within the TME can modify other aspects of tumor metabolism, as it promotes a preferential exploitation of glutamine and lipids—as sources of energy and biosynthesis—in cancer cells, over the canonical glycolytic metabolism. Indeed, a decrease in HIF1α activation (resulting from direct acetylation) concomitantly with a reduction in the expression of glycolytic enzymes as well as the glucose transporter GLUT1 and the lactate transporter MCT4 has been reported (53, 54). Also, acidosis drives the reprogramming of lipid metabolism by triggering an increase in HIF2α activity which stimulates the reductive and oxidative glutamine metabolism, ultimately sustaining the co-existence of synthesis and oxidation of Fas (55). Notably, the newly synthetized lipids, stored in lipid droplets, represent a readily available energy to support anoikis resistance and invasiveness in cancer cells adapted to the acidic conditions (56).

Finally, immune cells may be forced to experience environmental lactate with beneficial or detrimental consequences on their differentiation. Indeed, in a prostate cancer model, lactate released by glycolytic CAFs causes a clear shaping of T-cell polarization, by reducing the percentage of the anti-tumoral Th1 subset cells and increasing protumorigenic Treg cells subpopulation. Both Th1 and Treg cells are reprogrammed by an activation of a lactate-driven epigenetic pathway, causing activation of T-bet or NF-kB/FoxP3 transcription factors. This lactate-based reprogramming of T-cell response leads to enhance malignancy of prostate tumors, thereby confirming the immunomodulatory role of lactate (51). Furthermore, cancer cell-derived lactate is able to polarize M1 macrophages into M2 ones, activating different signaling cascades (i.e., VEGF, arginase-1) in macrophages undergoing pro-tumor differentiation (57). Notably, another signaling role of the environmental lactate has been investigated in endothelium as lactate could be up-taken by endothelial cells through the MCT-1, thus stimulating the autocrine NF-κB/IL-8 (CXCL8) pathway which affects tumor angiogenesis in terms of endothelial migration, vessels permeability and morphology (58). Importantly, besides the canonical role as nutrient, lactate has been shown to act as a hormone, as it is able to activate signaling pathways downstream the hydroxycarboxylic acid receptor 1, formerly known as G protein-coupled receptor 81 (GPR81). This receptor, sensitive to low concentrations of lactate (1–5 mM), is coupled to Gi/q, and activation of the receptor results in decreased cellular levels of cAMP and increased cellular levels of Ca2+, leading a fascinating hypothesis of an autocrine and paracrine role of lactate in cancerous and stromal cell types via surface receptor (59, 60). Very recently, another non-nutritional role of lactate has been reported by Zhang et al. (61). These authors reported that lactate produced in hypoxic environment is involved in a peculiar post-transcriptional modification of histones (i.e., lactylation), a process occurring with different temporal dynamics from acetylation and, so far, involved in wound healing and pro-inflammatory signals (61). Although the role of lactylation in TME deserves future considerations, it is likely that this hypothesis deserves consideration in future studies. However, the lack of definitive information regarding the microenvironmental fluxes of lactate in the different cellular compartments highlights the general need for enlarging and setting new tracing studies in the tumor extracellular milieu.

### TCA Intermediates

Oncometabolites are a group of metabolites, including succinate, fumarate, and 2-hydroxyglutarate, accumulated in cancer cells generally as a consequence of mutations in genes coding for the related metabolic enzymes, that are succinate dehydrogenase (SDH), fumarate hydratase, or isocitrate dehydrogenase, or of alterations in their enzymatic activity (62). These metabolites are involved in the dysregulation of several cellular processes, mainly through the competitive inhibition of α-KG-dependent dioxygenases, causing pseudohypoxia via HIF-1 stabilization, protein post-transcriptional modifications, as well as epigenetic alterations in cancer cells. Mutations of SDH subunit genes are recurrent in some cancer types including hereditary pheochromocytoma syndrome and paraganglioma. In any case, there is a loss of function of the SDH enzyme, causing succinate accumulation (63, 64). Germline mutations in the SDH subunits have also been shown to cause gastrointestinal renal, pancreatic neuroendocrine, thyroid, and neuroblastoma tumors (65), although SDH activity can also be epigenetically inhibited via the binding of the chaperone tumor necrosis factor receptor-associated protein 1 (49), or through the competitive inhibition of the metabolite itaconate highly enriched in reactive macrophages (66). Increased concentrations of succinate may induce metabolic reprogramming within TME and concur to promote cancer progression. Accumulation of succinate is correlated with a state of pseudohypoxia, due to its ability to inhibit prolyl hydroxylation of HIF-1α, leading to stabilization of the transcription factor and activation of HIF-controlled genes involved in glycolysis, angiogenesis, and EMT (67). Moreover, accumulation of fumarate can reduce the expression of the anti-metastatic miRNA cluster mir-200ba429, by inhibiting demethylation of the CpG islands in the regulatory region via regulation of DNA demethylases TETs. This epigenetic rewiring promotes activation of EMT programme and the increase of metastatic potential in a model of renal cancer (38).

Moreover, succinate post-translationally modifies lysine residues of proteins through succinylation, including L-lactate dehydrogenase A, glyceraldehyde 3-phosphate dehydrogenase, glutamate carrier-1, uncoupling protein-1 and malate dehydrogenase, all enzymes involved in the reprogramming of cancer cell metabolism (68, 69).

Beside the direct role of oncometabolites in those cancer cells undergoing their accumulation, a new original view also supports the oncometabolites as signaling molecules, secreted by dying cells or by neighboring stromal populations. In this line, we have recently reported that exposure of prostate cancer cells to CAFs, while undergoing mitochondrial deregulation and OXPHOS addiction, leads to the accumulation of TCA cycle intermediates, consistently with lactate oxidative exploitation (51). Cancer cells, uploading lactate secreted by CAFs, fuel TCA cycle and accumulate succinate and fumarate, likely linked to their ability to drive a pseudohypoxic HIF-1-mediated EMT motility (51). Upon deregulation of TCA, succinate and fumarate are also secreted in TME, although indications on their specific destiny are lacking. In keeping with a role as extracellular signal, succinate can bind to its cognate receptor namely SUCNR1 (68). SUCNR1, belonging to the family of G protein-coupled receptors, is expressed in kidney, liver, brain, bone marrow, as well as in several cancers (70), and is reported to control cell proliferation, migration, capillary formation and development of new vessels formation, VEGF secretion, as well as stem cell functions (71, 72).

Cancer cells can either accumulate succinate, or eventually upload it from the TME. Indeed, a plasma membrane Na(+) dependent dicarboxylic acid transporter NaDC3 (also called SLC13A3), able to specifically upload succinate, has been reported in prostate cancer cells. The real contribution of extracellular succinate is not clear, as the block of the succinate plasma membrane carrier is not sufficient to inhibit cancer growth in a PTEN-loss model of prostate cancer, while succinate-supported respiration is mandatory for prostate cancer malignancy (73). During inflammation, succinate may be secreted by inflammatory macrophages and accumulate into TME (74), as reported in murine ischemic tissues (75), central nervous system inflammation and in rheumatoid arthritis inflammation. Interestingly, macrophages express GPR91 and, in response to inflammatory signals like lipopolysaccaride, activate a GPR91-mediated signal transduction that sustains the proinflammatory phenotype and leads to IL-1β production (76). This represents a novel mechanism by which succinate fuels inflammation in an autocrine manner to sustain and amplify the inflammatory response (77).

However, fascinating evidence report that cancer cell-secreted succinate elicits M2 macrophage polarization and positively regulates cancer metastasis via SUCNR1 (78), thus enlarging the class of tumor metabolic factors affecting TME and tumor phenotypic rearrangement.

### Citrate

Citrate is the primary substrate for fatty acid synthesis and is metabolized in the cytoplasm by ATP-citrate lyase to serve acetyl-CoA moieties for lipid synthesis. Citrate-derived acetyl-CoA also contributes to amino acid synthesis, as well as to protein acetylation (79, 80), both processes critical for proliferating cells. Sources of citrate for cancer cells are their own Krebs cycle, reductive carboxylation of α-KG originating from glutaminolysis (81), as well as the direct importation from TME through a plasma membrane-specific variant of the mitochondrial citrate transporter (82). Consistent with the hypothesis of extracellular citrate as a key nutrient able to affect cancer aggressiveness the blocking of the plasma membrane citrate carrier (variant of the SLC25A1), expressed in several malignant cancers, results in decreased tumor growth in immunodeficient mice and altered tumor metabolism. Moreover, decreased blood citrate levels have been associated with some tumors including those in the lung, bladder, and pancreas (83).

Finally, citrate, upon conversion into isocitrate, can also fuel itaconate biosynthesis as a TCA cycle by-product from the decarboxylation of cis-aconitate. Itaconate production is active in macrophages upon exposure to inflammatory stimuli, playing a direct antimicrobial effect, markedly affecting immunomodulation, suppression of inflammation and tolerance (66). Itaconate acts mainly by inhibiting SDH, causing accumulation of succinate in LPS activated macrophages, and this was associated to reduced mitochondrial respiration, ROS production, HIF-1 pseudohypoxic activation, proinflammatory cytokine release, and inflammasome activation (84). Although itaconate plays clearly a key role within TME by regulating macrophage activation, its release in TME has not been yet reported.

### Glutamine and Other Aminoacids

Intriguingly, although lactate is the most abundant nutrient provided in the TME, CAFs are also able to supply amino acids like glutamine to cancer cells. Epithelial cancer cells incorporate fibroblasts-derived glutamine replenishing their TCA cycle, as well as promoting an increase in aspartate-mediated nucleotide anabolism, the accumulation of oxidized glutathione and the activation of protein synthesis (85). Glutamine dependency as it is exploited as a carbon source for the energetic purposes and as a nitrogen source for nucleotide biosynthesis reflects the fact such amino acid is the most commonly depleted amino acid in TME (33). In agreement, glutamine-restricted TME are truly dependent on tumor-stroma glutamine cross-feeding. In ovarian carcinoma, CAFs metabolism diverge from classical glucose exploitation, but activate glutamine synthesis, thereby serving this amino acid to cancer cells. Hence, due to the metabolic pressure applied by cancer cells, CAFs increase their incorporation of glucose-derived carbons into TCA metabolites and branched-chain amino acids-derived nitrogen to glutamine synthesis. Cancer cells educate CAFs to enhance their capability to use different nutrient sources to synthesize glutamine, in order to support cancer cell mitochondrial activity through glutaminolysis in stressed TME (86). A similar nutrient crosstalk mediated by exchanged glutamine has also been reported in models of astrocytes:glioblastoma and adipocytes:pancreatic cancer cells, as glutamine fuels the de novo purine biosynthesis (87, 88). Interestingly, glutamine within TME can also be active in rescheduling macrophages polarization toward the malignant M2 phenotype and enhancing cancer aggressiveness. Indeed, pharmacologic impairment of glutamine synthetase skews M2-polarized macrophages toward the M1-like phenotype. As a result of these metabolic changes M2 macrophages display a decreased ability to recruit immune and endothelial cells (89).

Beside glutamine, upon stromal autophagy activation, also alanine is largely secreted by pancreatic stellate cells, a stromal population very similar to activated CAFs. Alanine is uploaded by pancreatic cancer cells and fuels their TCA cycle over the glucose/glutamine-derived carbons, and this mitochondrial exploitation leads to an increased biosynthesis of lipids and non-essential amino acids (90). The metabolic rescheduling of pancreatic stroma has profound effects on cancer cells, as the contact with reactive stroma induces widespread histone acetylation in cancer cells, thereby serving to epigenetic purposes (91).

Moreover, mechanical signals sent by ECM composition and stiffness are able to reprogram CAFs and cancer cells toward a peculiar metabolic cross-talk mediated by exchange of aspartate and glutamate via the SLC1A3 transporter (92). The cross-talk

is directional, as CAFs-derived aspartate feeds TCA cycle by sustaining the pyrimidine biosynthesis in cancer cells exposed to enhanced ECM stiffness, while glutamate provided by cancer cells is used by CAFs to maintain redox homeostasis through glutathione biosynthesis.

Finally, the tumor stroma cross-talk may affect kynurenine synthesis, a metabolite of tryptophan catabolism, through activation of the tryptophan 2,3-dioxygenase in CAFs. The shuttled kynurenine, uploaded by cancer cells, engages the EMT pathway, enhancing malignancy and immune suppression through regulation of dendritic and Th1 and Th2 subset of T cells (93).

### Lipids

Lipids are surely key components of TME rescheduling of cancer cell metabolism. Indeed, lipids can be accumulated in cells, segregated to lipid droplets (LDs) due to physicochemical reasons, mainly as triglycerides and cholesterol derivatives. Their mobilization upon energetic request is under the control of specific lipases, tightly regulated by TME stimuli. Catabolism of triglycerides by adipose triglyceride lipase (ATGL) releases fatty acids (FAs), mainly used for energetic purposes via TCA cycle fueling, or for serving acetyl-CoA moieties for acetylation of proteins, either belonging to nuclear or cytosolic compartments. Cholesterol can be converted to 22- or 27-hydroxycholesterol, which activate liver X receptor signaling to up-regulate cholesterol efflux via regulation of the ATP-binding cassette transporters (94).

FAs, once released from LDs due to activation of ATGL, can be delivered to TME for fueling energetic needs of neighboring cells. To this end lipids can be loaded on secreted vesicles or translocated across the phospholipid bilayers of the plasma membrane through either passive diffusion or a proteinmediated transport system. Several membrane-associated FA binding proteins and transporters reportedly facilitate the transport process, including FA translocase (FAT, also named CD36), Fatty Acid Transport Protein and Plasma Membrane Fatty Acid Binding Protein. Highly aggressive prostate cancers show high expression of CD36 which facilitates the intake of exogenous FAs, and the subsequent LDs mobilization provokes a significant alteration in intracellular lipid content in terms of acyl-carnitines, monoacylglycerols and other lysophospholipids (95). Other findings have found the breast cancer cells resistant to HER2 therapy upregulate CD36, and thus acquiring an increased lipid metabolism and metabolic plasticity, both crucial for promoting resistant cells the adaptation and survival under nutrient deprivation and drug toxicity (96). Also, hypoxic breast and glioblastoma cells cancer cells upload FAs from the TME. The exploitation of triglycerides derived from accumulated LDs provides them ATP to face conditions of reoxygenation frequently occurring in a harsh TME (97). To note, stromal adipocytes are the main lipids donors in TME of several cancers. During melanoma progression adipocyte-derived lipids are taken up by FAT proteins, aberrantly expressed in melanoma, causing lipid upload and enhanced invasion and melanoma cell growth (98). The translocation of FAs in melanoma cells is also mediated by vesicles, as indicated by proteomic analysis of peritumoral adipocyte exosomes, rich in either lipids and enzymes involved in their catabolism (99). Moreover, in ovarian cancers, adipocytes promote tumor progression again through the provision of FAs. Although the exact mechanism through which adipocyte-derived FAs are transported into ovarian carcinoma cells remains uncertain, a role has been proposed for FAT/CD36 carrier (100). Besides adipocyte predominance in TME lipid supply, interestingly, levels of n-3 and n-6 polyunsaturated fatty acids (PUFA) and glycerophospholipids (e.g., phosphatidylcholine) have been highly detected in tumor cells cultured with endothelial cells (101). Collectively, these findings demonstrate that FATBPs and CD36 play a key role in tumor microenvironment metabolic cross-talk, driving the dependency of tumor cells toward exogenous lipid rewiring cancer cell metabolism and behavior. Furthermore, adiposederived lipids have been also shown to mediate ovarian cancer chemoresistance (102). Indeed, a lipidomic analysis revealed that arachidonic acid AA is the key chemo-protective lipid mediator, although it is not known if arachidonate activity is due to its uploading or if it acts as a signaling molecule, as its sister companions prostaglandins. Finally, breast cancer cells promote lipolysis in peritumoral adipocytes leading to the release of FAs in the TME (103). Particularly, cancer cell-derived inflammatory signals induce an adipose triglyceride lipase-dependent catabolic pathway. The mobilized FAs, upon secretion, are transferred to cancer cells where they are stored in LDs or used through the carnitine palmitoyltransferase I-dependent fatty acid β-oxidation pathway, fueling a high mitochondrial activity.

### Mitochondria

Nutrients are not only the unique metabolic molecules to be exchanges. Strikingly, horizontal transfer of intact and functional organelles (e.g., mitochondria) from stromal to cancer cells has been observed in TME. Cancer cells may exploit traveled mitochondria either to start or boost OXPHOS metabolism. Indeed, mitochondria-defective cancer cells de novo acquire mitochondria from TME to rescue a respiration they cannot carry out (104, 105). Oxidative stress is the driver for this mitochondrial transfer via cytoplasmic bridges (tunneling nanotubes) formed between bone marrow-derived MSCs and recipient leukemic blasts. The final outcome is an increase in mitochondrial mass, OXPHOS and ATP production as well as the drug resistance of cancer cells (106–108). Remarkably, CAFs channel their own mitochondria through intercellular interactions to further boost metabolism of OXPHOS-addicted prostate cancer cells. The molecular driver of such behavior seems to be again the lactate as its presence putatively enhances the formation of such mitochondria roads. These de novo achieved mitochondria are finely active for OXPHOS metabolism, ROS production and EMT promotion in cancer cells (51). Of note, these exchanges of intact mitochondria in prostate cancer also occur in xenografts of mice models and are not restricted to mitochondria-defective cancer cells.

### Microvesicles

Extracellular vesicles (EVs) trafficking has been recently described as a new form of intercellular communication (87), with a high impact on the nutritional exchanges within tumor microenvironment either directly or indirectly through the cell-cell exchange of metabolic enzymes. EVs are approximately spherical structures limited by a lipid bilayer and containing bioactive components, such as proteins, lipids and nucleic acids. EVs are secreted by many cell populations, including fibroblasts (88), hematopoietic-derived cells, epithelial cells, neurons and tumor cells (109–111). EVs are classified into two main distinct subtypes, depending on their biogenesis, size, morphology and protein composition: exosomes and ectosomes/microvesicles (112). Exosomes are vesicles with a diameter of 50–150 nm, which are formed via inward budding of late endosomes membrane, the so called multivesicular bodies, which can fuse with the plasma membrane, releasing exosomes into the extracellular environment. On the other hand, microvesicles are directly produced by plasma membrane blebbing, are larger than exosomes, ranging from 100 nm to 1µm in diameter (113). Once released into the extracellular environment they can interact to recipient cells receptors thereby triggering signal transduction events or they can fuse with the plasmamembrane of the acceptor cell releasing their content in the cytoplasm. EVs trafficking is involved in both physiological and pathological contexts such as: immunity (114), tissue regeneration (115), stem cell biology (116), angiogenesis (117), and tumor progression (118). Here we will focus our attention on EVs mediated-cross-talk in the context of tumor microenvironment. Tumor derived EVs are classically viewed as a way to alter tumor microenvironment to facilitate cancer progression via the transfer of proteins such as: (i) epidermal growth factor receptor-vIII, an oncogenic receptor (119); (ii) multidrug resistance-associated protein 1 a membrane protein mediating export of organic anions and drugs from the cytoplasm (120); (iii) pro-angiogenic proteins, i.e., TGF-β and VEGF, etc. (121). In addition, miRNA transferred by cancer EVs induce, by means of a still unknown mechanism, the secretion of CAFs chemokines such as CXCL1 and CXCL8 that correlate with poorer survival in gastric cancer patients (122).

More recently, it has become evident that CAFs are also able to produce and secrete EVs, thereby underlining the bidirectional importance of EVs trafficking within TME. Proteins and miRNA produced by CAFs and conveyed through EVs to tumor cells influence their behavior, supporting cancer cells growth rate and survival (123, 124), aggressiveness (125, 126) and favoring chemoresistance (127). In addition, CAFderived EVs, uploaded by tumor cells, induce metabolic changes in acceptor cells such as enhanced glycolysis and glutamine metabolism rate, decreased oxygen consumption rate and down-regulation of mitochondrial function (128). The growing evidences about the involvement of EVs trafficking in regulating tumor cells metabolism has pushing the focus on the study of EV-transferred metabolites between different subsets of cells within tumor microenvironment. EVs-mediated trafficking of metabolites may be of particular importance for cancer cells that need a very high rate of metabolites influx to sustain their rapid cell growth. Currently, metabolomic studies on EVs has addressed their metabolic content in terms of lipids. It is reported that EVs transport plasma-membrane derived lipids i.e., sphingolipids, sterols, glycerophospholipids, fatty acids, and sphingolipids (129–131), with different relative proportion and composition reflecting those of their parental cells. Fewer studies are currently available describing the complete metabolome of the EVs, but what it is clearly emerging that, besides lipids, EVs contain many other organic molecules such as vitamins, amino acids, sugars, nucleotides, carnitines and aromatic compounds (128, 132, 133).

However, a very efficient way to induce a change in recipient cells phenotype, with respect to the simple transport of metabolites, is the transfer of enzymes involved in cellular metabolism. The analysis of the "vesiclepedia" database using informatics tools that clusterize the proteins contained into the EVs using a functional criterion, reveals that over 25% of them are directly involved in cell metabolism (113). In this view, EVs trafficking can be seen also as a metabolic coordination platform between the different cell populations within the solid tumor. This "metabolic synchronization" allows the optimization of the overall request for metabolites between the various cellular components of the tumor in order to support the survival and the neoplastic expansion of the tissue.

### SUMMARY

Tumor-stroma metabolic cross-talk mainly portrays the setting where tumor confiscates metabolic nutrients, including lactate, amino acids and fatty acids, from local and/or stromal sources. This event provokes the catabolic pathways, such as autophagy, glycolysis and lipolysis in the tumorassociated cellular compartment. This interplay is absolutely reciprocal, as the interactions between stromal and tumor cells mutually reprogram the metabolism of each cell population. Highly aggressive cancer cells experience specific metabolic reprogramming, aimed at optimizing and functionally exploiting stromal cues (i.e., metabolites, vesicles, organelles), likely representing the critical transducers of the rewiring of the cancer metabolism within the TME. Stromal-induced mitochondrial dysregulation, in terms of oncometabolites production, ROS production and organelle biogenesis or transfer, contributes to the proliferative and metastatic potential of neoplastic cells. The flexibility that metabolic deregulation of cancer cells upon education by TME, often referred as metabolic plasticity, provides tumor cells the correct tools to face environmental hostile conditions.

Few therapeutic approaches have been developed to target tumor:stroma:metabolic interplay, among these we can cite glycolytic inhibitors to target the stromal component or pseudohypoxic cancer cells, or mitochondrial inhibitors for targeting mitochondrial metabolism in OXPHOS-addicted populations (134). The main obstacle in such targeting is the metabolic plasticity arising in this stroma-cancer communication: any strategy targeting one side of the tandem rapidly results in rescue of the other part of the tandem, to shift metabolism toward adaptation to dynamic environment. Although repurposing efforts to develop new metabolic inhibitors for cancer therapy to implement treatments is highly warranted, the preliminary strongest effort needed right now is the identification of the molecular player of metabolic plasticity, in order to efficiently target the adaptive symbiosis of tumor: stroma tandem.

### REFERENCES


### AUTHOR CONTRIBUTIONS

PCh, PCi, LI, and GC contributed to the writing of this manuscript. Figure was rendered by LI. Editing was performed by all authors of this paper.

### FUNDING

This work was supported by PRIN 2017 (grant to PCh) and Fondazione CR Firenze, AIRC (grant 19515 to PCh).


proliferation of pancreatic cancer cells. Oncogene. (2017) 36:1770–8. doi: 10.1038/onc.2016.353


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

Copyright © 2020 Comito, Ippolito, Chiarugi and Cirri. 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.

# Causes and Consequences of Variable Tumor Cell Metabolism on Heritable Modifications and Tumor Evolution

#### Bryce Ordway <sup>1</sup> , Pawel Swietach<sup>2</sup> , Robert J. Gillies <sup>1</sup> and Mehdi Damaghi 1,3 \*

<sup>1</sup> Department of Cancer Physiology, H Lee Moffitt Cancer Center and Research Institute, Tampa, FL, United States, <sup>2</sup> Department of Physiology, Anatomy and Genetics, University of Oxford, Oxford, United Kingdom, <sup>3</sup> Department of Oncologic Sciences, Morsani College of Medicine, University of South Florida, Tampa, FL, United States

When cancer research advanced into the post-genomic era, it was widely anticipated that the sought-after cure will be delivered promptly. Instead, it became apparent that an understanding of cancer genomics, alone, is unable to translate the wealth of information into successful cures. While gene sequencing has significantly improved our understanding of the natural history of cancer and identified candidates for therapeutic targets, it cannot predict the impact of the biological response to therapies. Hence, patients with a common mutational profile may respond differently to the same therapy, due in part to different microenvironments impacting on gene regulation. This complexity arises from a feedback circuit involving epigenetic modifications made to genes by the metabolic byproducts of cancer cells. New insights into epigenetic mechanisms, activated early in the process of carcinogenesis, have been able to describe phenotypes which cannot be inferred from mutational analyses per se. Epigenetic changes can propagate throughout a tumor via heritable modifications that have long-lasting consequences on ensuing phenotypes. Such heritable epigenetic changes can be evoked profoundly by cancer cell metabolites, which then exercise a broad remit of actions across all stages of carcinogenesis, culminating with a meaningful impact on the tumor's response to therapy. This review outlines some of the cross-talk between heritable epigenetic changes and tumor cell metabolism, and the consequences of such changes on tumor progression.

Keywords: tumor evolution, acidosis and oxidative stress, nutrient sensing and signaling, tumor micoenvironment, epigenetic regulation

### MOLECULAR MECHANISMS OF MICROENVIRONMENTAL SENSING

Cancer evolution operates through selection, which requires a degree of phenotypic diversity to present a range of possible responses to microenvironmental selection forces, some of which confer selective advantage (1, 2). Tumors can be described as complete ecosystems, containing cancer cells, stromal cells, vasculature, extracellular matrix, and the chemical milieu consisting of variables such as pH and oxygen tension (3–5). During tumorigenesis—and similarly in response to therapy—the tumor ecosystem shows considerable plasticity because cancer cells shape their microenvironment,

#### Edited by:

Paolo E. Porporato, University of Turin, Italy

#### Reviewed by:

Valéry L. Payen, Fonds National de la Recherche Scientifique (FNRS), Belgium Liwei Lang, Augusta University, United States

> \*Correspondence: Mehdi Damaghi mehdi.damaghi@moffitt.org

#### Specialty section:

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

Received: 14 December 2019 Accepted: 03 March 2020 Published: 27 March 2020

#### Citation:

Ordway B, Swietach P, Gillies RJ and Damaghi M (2020) Causes and Consequences of Variable Tumor Cell Metabolism on Heritable Modifications and Tumor Evolution. Front. Oncol. 10:373. doi: 10.3389/fonc.2020.00373

**130**

to which subsequent generations of them must adapt to thrive, and these adaptations, in turn, fine-tune the microenvironment (6). During the various stages of tumor progression, cells can be exposed to highly variable chemical stimulations, largely attributed to variable blood perfusion; for example, oxygen deprivation (hypoxia), nutrient deprivation, metabolic endproduct build-up, and increased acidity. Overall, these stimuli would be considered, by normal standards, to be survivable to most cells but exerts some cost on cells fitness, and it is therefore axiomatic that cancer cells must adapt to these conditions if they are to thrive. Although the stimuli are survivable, they still impose stress on the cells which changes their fitness, requiring acquisition of a novel homeostatic balance that costs more energy for cells and can be lethal for cells in competition with other cells.

Hypoxia is one of the main environmental factors a cancer cell must face in order to survive, thrive, and progress. Hypoxia imposes a metabolic stress on the cell, hindering its ability to carry out aerobic respiration. Therefore, the cell must be able to adapt to a hypoxic environment in order to survive. The cellular response to hypoxia is robust, and exerts most of its force via the transcription factors HIF-1α and HIF-2α. Another HIF family protein, HIF-3α, functions to repress the responses directed by HIF-1α and HIF-2α. All three of these proteins carry out their function via dimerization with constitutively expressed HIFβ proteins in the nucleus which allows them to directly modulate transcription of proteins involved in the hypoxia response (7, 8), or in the case of HIF-3α repress the transcriptional response. HIF-1α is a constitutively expressed protein, whose activity is regulated by the hydroxylation of conserved proline residues. In microenvironments of high oxygen tension (>5%), the proline residues are hydroxylated, tagging the protein for degradation by E3 ubiquitin ligases (9). When oxygen concentrations are below the tolerable threshold for a given cell type, HIF-1α is not degraded, and increases in concentration to allow HIF-1α to induce transcription of its client genes (10).

Nutrient deprivation is another major stressor within the tumor microenvironment. When the cell experiences a critical reduction in a particular nutrient, it must swiftly adjust in order to maintain productivity in the terms of metabolism, proliferation, migration, or other processes essential to evolutionary success and survival. Two main nutrient sensing proteins implicated in cancer are AMPactivated protein kinase (AMPK) and the mammalian target of rapamycin (mTOR) (11, 12). These proteins are capable of sensing the current energy status of the cell and nutrient availability, respectively.

The metabolic pathway directed by AMPK is highly contextspecific; depending on the nutrient status of the cell. AMPK is responsible for the cellular response to glucose deprivation and acts as a metabolic switch from a highly glycolytic state to an oxidative state depending on the availability of glucose. This is particularly important in the context of cancer and the highly plastic nature of cancer metabolism. AMPK is activated by 5′ - AMP, which indicates that the cell is not regenerating ATP at a fast-enough rate to meet demand. This induces the uptake of glucose and the induction of glycolysis to replenish the cellular ATP (11). The induction of glycolysis via-a-vis respiration is likely due to the promptness with which glycolytic activation can occur (13).

mTOR is present in the form of two different complexes, mTORC1 and mTORC2. These two complexes participate in associated, but distinct signaling pathways in nutrient sensing. mTORC1 becomes activated in response to various growth factors and amino acids that promote cell growth and proliferation. When inactive, mTORC1 represses growth and induces an autophagic response. mTORC2 is a sensor of glucose but also plays a role in amino acid signaling. mTORC2 is activated by acetyl-coenzyme A (Ac-CoA), which is produced in the cytoplasm via citrate lyase, when glycolytic flux is abundant. The result of mTORC2 activation is increased cell proliferation, in response to the increased glucose metabolism. mTORC2 has also been implicated in amino acid sensing by having the ability to suppress the function of the glutamine-cysteine transporter, system Xc transporter-related protein (12).

Aberrant perfusion in the tumor microenvironment allows a significant build-up of metabolites in the tumor interstitial fluid. The main metabolite that is commonly accumulated in the tumor interstitial fluid is lactic acid, which is associated with a decrease in pH. A decrease in extracellular and intracellular pH can dramatically modulate the activities of enzymes, some of which are more sensitive than others, depending on how significant the change in pH is and the isoelectric point of the enzymes optimal activity (14). These alterations in enzyme activity are pleiotropic and leads to metabolic reprogramming. Lactate in the tumor microenvironment is a by-product of increased glucose fermentation, which occurs even in the presence of oxygen, known as aerobic glycolysis or the Warburg Effect. Once produced, lactate is shuttled out of the cell, stoichiometrically with a proton, by monocarboxylate transporters (MCTs) 1–4. Sensing of extracellular pH is accomplished through a variety of plasma membrane associated proteins including two major classes of acid-sensing receptors: (i) G-protein coupled receptors (GPCR) such as Ovarian cancer G protein-coupled receptor 1 (OGR1), G-protein coupled receptor 4 (GPR4), T-cell death-associated gene 8 (TDAG8), and ii) Acid-sensitive ion channels (ASICs) which include 7 proteins from 4 genes: 1a/b,2a/b,3,4,5, and Ca2<sup>+</sup> channel that includes transient receptor protein channel vanilloid subfamily 1 and 2 (TRPV1 and TRPV2) (15). Lactate can also be sensed and regulate cellular functions by activating the G protein-coupled receptors HCA1/GPR81, HCA2/GPR109A, and HCA3/GPR109B. These hydroxy-carboxylic acids (HCA) receptors control physiological homeostasis under changing metabolic and dietary conditions (16).

Cancer cells commonly overexpress many of the aforementioned acid sensors, and this can be correlated to tumor progression and poor outcome (17, 18). Therefore, investigating these sensors as a factor in malignancy may identify relevant prognostic biomarkers or may reveal new therapeutic vulnerabilities. The sensors can be connected to pathways to activate transcription factors or overexpress other genes and proteins (**Figure 1**). However, considering the ever-changing state of the microenvironment, we propose that epigenetic regulation may be a more effective factor in stabilization of emerging phenotypes in cancer cells. Adaptation

membrane of the cell are able to sense the extracellular pH, but have no clear mechanism for altering the activity of epigenetic modifiers.

to an acid-microenvironment has been shown to alter cell state by pushing cells into a partial EMT phenotype (19); this may be a manifestiation of these acid sensors inducing a stable epigenetic change.

### HERITABLE EPIGENETIC MODIFICATIONS ACQUIRED THROUGH MICROENVIRONMENTAL SENSING

One mechanism enabling cancer cells to adapt is through changes to gene expression via epigenetic regulation. Some modalities of epigenetic regulation are transient (e.g., histone acetylation) and are imposed to help cancer cells survive acute disruptions in their microenvironmental homeostasis. In contrast, other epigenetic mechanisms are more persistent (e.g., DNA methylation) and have the ability to be passed down through generations to endow further generations with memory on how to survive in the tumoral microenvironment.

The term epigenetics was first created by CH Waddington who described it as "the causal interactions between genes and their products, which bring the phenotype into being." While a commonly agreed upon definition is hard to find today, the term epigenetics in the modern era is commonly described as a permanent change in the way genes are expressed. Types of epigenetic regulation include histone modifications, DNA Methylation, and non-coding RNA (20), which can all impact one another to create a complex regulatory dynamic. A major question is: "How do the external factors of the tumoral microenvironment play into altering this complex dynamic"?

It is commonly known that the epigenetic signatures of cancer cells are different compared to their untransformed counterparts (21–25). Many of these epigenetic alterations exert their function by altering the metabolism of cancer cells (26), and are acquired by signaling cascades initiated by sensing of the extracellular environment. Some of the signaling cascades that can lead to changes of epigenetic signatures were implied in the previous section and the specific alterations they are involved in will be discussed herein.

In many solid tumors, intra-ductal hyperplasia leads to significant alterations in the physical microenvironment, especially in peri-luminal cells that are far (>160 microns) from their blood supplies. Importantly, the diffusion distance of oxygen in tissues is 100–200 microns (27), meaning that the periluminal cells of DCIS can be profoundly hypoxic. The depth and duration of hypoxia is dependent on the blood flow of the surrounding stroma. Hypoxia eventually selects for metabolic reprogramming, leading to acidosis, as well as exacerbating nutrient deprivation. Over many years in this environment, these forces (hypoxia, acidity, nutrient deprivation) select for cells with more highly adaptable, aggressive, de-differentiated phenotypes (6, 28). The resulting acidosis leads to genome instability, which could increase the rate of cancer evolution (29). The source of cytoplasmic (and nuclear) acidosis is lactate accumulation as a byproduct of glycolysis. Lactate has been shown to have a variety of effects on the epigenetic mechanisms of the cell, some of which are confounding. Direct inhibition of histone deacetylases (HDACs) by lactate has been shown in separate studies (30, 31); while others have reported multiple times on the increase in activity of the Sirtuin family of histone deacetylases upon exposure of cells to a chronically acidic extracellular environment (32, 33). These examples represent combined sensing and epigenetic effector mechanisms that act directly on altering the epigenetic status of the cell.

The oxygen sensing protein HIF-1α can regulate epigenetics in a variety of ways. Upon activation, HIF-1α leads to downstream signaling cascades important for the survival of cells in low oxygen environments. The ultimate effect of some of these signaling cascades is the epigenetic alteration of gene regulation (**Figure 1**). Two epigenetic mechanisms influenced by HIF-1α are histone methylation and DNA methylation (**Figure 1**). Unlike other epigenetic mechanisms, histone methylation can act in both activating and repressing fashions depending on the specific location of the covalent modification. The histone demethylase JMJD2B is activated by HIF-1α (**Figure 1**), and is specifically targeted to H3K9me2/3 to demethylate the mark to a monomethylated state (34, 35). The expression of ten-eleven translocation proteins 1/3 (TET1/3) is also upregulated upon stabilization of HIF-1α. TET1/3 are 5-methylcytosine (5mC) oxidases, which convert 5mC into either 5-hydroxymethylcytosine (5hmC), 5-formylcytosine (5fC), or 5-carboxylcytosine (5caC) via sequential reactions (36). The result of these reactions is the deactivation of the methylation mark and the subsequent reactivation of the sequence being repressed by the methylation. This has been validated in neuroblastoma where it was shown that HIF-1α can induce HIF-1α/hypoxia specific DNA methylation signatures (37). The fact that HIF-1α activates DNA methylation supports our hypothesis of inheritable epigenetic changes to next generation that can be tracked in cancer cells. Contrary to the upregulation of TET by hypoxia-induced transcriptional programs, TET proteins have been shown to have their activity reduced directly by low oxygen availability in a tumor setting. TET activity is lost in vitro when exposed to hypoxic conditions, possibly via hypermethylation of tumor suppressor promoters in hypoxic regions of tumor samples (38). The opposing forces of the transcriptional and functional regulation of TET proteins may demonstrate a physiological feedback system for regulating the epigenetic response to oxygen deprivation in order to attenuate the response (**Figure 1**).

The sensing of nutrients by a cell is vitally important to its survival and can have long term effects on the downstream lineage of that cell via epigenetic modifications. Having this feed forward system of epigenetic regulation directed by nutrient signaling allows for increased fitness for subsequent generations. As mentioned previously, cellular nutrient sensing is mainly achieved through 3 essential proteins and protein complexes: AMPK, mTORC1, and mTORC2 (**Figure 1**). AMPK is activated in response to cellular metabolic stress, and modulates gene transcription epigenetically in order to respond to this stress. Unlike the mTORC1/2 complexes, AMPK is able to modulate transcription directly by phosphorylation of Ser<sup>36</sup> on histone H2B (39). This phosphorylation mark directly promotes the transcription of response genes needed to handle metabolic stress. Other papers report the direct phosphorylation of Ser<sup>36</sup> on H2B by S6K1 (40), which is also a player in the LKB1- AMPK-mTORC1 signaling axis, with S6K1 being phosphorylated by mTORC1. mTORC1 is another player in the epigenetic response to nutrient sensing. As mentioned previously, mTORC1 is capable of sensing various growth factors and amino acids. A downstream target of mTORC1 nutrient sensing is SIRT4, which is repressed in response to mTORC1 activation (41). SIRT4 is a lysine deacylase (**Figure 2**) (42), that has the ability to inhibit glutamine metabolism by inhibiting glutamate dehydrogenase (GDH). This inhibition of SIRT4 is achieved at the transcriptional level by mTORC1 stabilizing the CREB2 βTrCp complex, preventing CREB2 from activating transcription of SIRT4 (41). mTORC2 exerts its epigenetic function by activation of the AKT and SGK1 proteins. The effect of these proteins on epigenetic regulation is the inhibition of KMT2D, which is a histone methyltransferase specifically targeting H3K4 (**Figure 2**) (43). Inhibition of KMT2D has been shown to have anti-tumor effects in some cancers by not allowing the FOXA1- PBX-ER complex to access the DNA for transcription (44).

While the aforementioned mechanisms involve the sensing of nutrients to transduce downstream epigenetic changes, alterations in acetate level can directly influence the epigenetic status of the cell. Free Acetyl-CoA in the cell nucleus is produced by Acetyl-CoA synthetase (ACSS2), which catalyzes the conversion from Acetate, and by ATP-citrate lyase (ACLY) which catalyzes the conversion from citrate. The levels of free Acetyl-CoA directly influence the global acetylation status of histones (45), and henceforth have the ability to regulate epigenetics without the direct manipulation of an enzyme intermediate. While there is no direct sensing mechanism, this level of regulation could be seen as a sensor of the glycolytic state of the cell considering it has been shown that decreasing the amount of glucose available to a cell reduces the Acetyl-CoA abundance and lowers global histone acetylation (46).

A recently described mechanism of both environmental sensing and epigenetic modification is that of histone lactylation. In 2019, Zhang et al. described for the first time the modification of histones by lactate (47). As is commonly known, lactate is built-up as a byproduct of glycolysis. This epigenetic modification may provide a direct mechanism for the regulation

of gene expression in response to fermentative glycolytic activity of the cell.

### EFFECT OF HERITABLE EPIGENETIC MODIFICATIONS ON TUMOR METABOLISM

Many epigenetic alterations that are acquired throughout tumor progression alter the metabolism of the tumor's cellular population. While the previous section covered specific epigenetic alterations that occur in response to the metabolic microenvironment, this section will describe the role of epigenetic modifications in altering cancer cell metabolism.

Lactate itself has the ability to alter the activity of epigenetic modifier proteins. The inhibition of HDAC's by lactate demonstrated in previous studies (30, 31) has yet to be phenotypically implicated in the alteration of metabolic processes; yet it is likely this lactate-mediated mechanism will play a role in altering metabolism. The activation of SIRT1 by extracellular acidosis, which is a consequence of acid-inhibition of glycolysis and the subsequent build-up of NAD+, has been shown to alter cellular metabolism through histone deacetylation leading to increased transcription of HIF-2α. This SIRT1/HIF-2α axis promotes the oxidative metabolism of glutamine, and suppresses the effects of HIF-1α, inhibiting hypoxia mediated induction of glycolysis (32). Corbet et al. in a later study showed that SIRT1 as well as SIRT6 are essential for histone deacetylation and the induction of fatty acid metabolism when cells are chronically exposed to an acidic extracellular environment (33). From a cell survival standpoint, this switch to other methods of energy metabolism when lactate has accumulated is intuitive, and has relevancy in the context of cancer progression that is discussed later.

In response to hypoxia, HIF-1α activation leads to the induction of JMJD2B activity. JMJD2B has been shown to be upregulated in ER-positive breast cancer (48) and bladder cancer (49), and its upregulation has been directly linked to induction by HIF-1α in colorectal cancer (50) and gastric cancer (51). This activation of JMJD2B directly drives the demethylation of H3K9me2/3 to its monomethylated state (**Figure 2**). JMJD2B has been shown to play a role in altering the expression of many cancer associated genes including cyclin-dependent kinase 6 (CDK6) (49), and carbonic anhydrase 9 (CA9) (50), which can directly affect the transmembrane pH gradient. Also induced by hypoxia and HIF-1α activation are the expression of TET proteins 1 and 3 (36). As mentioned previously, TET proteins are 5mC oxidases that allow for the expression of genes repressed by DNA methylation. Induction of TET in neuroblastoma has been shown to increase transcription of hypoxia response gene (52), and TET1 has been shown to be overexpressed in triple negative breast cancer (TNBC) where it is associated with hypomethylation (**Figure 2**) (53). Hypomethylation increases the expression of associated genes such as Hexokinase II (HK2) in liver cancer (54) and glioblastoma multiforme (55). Hexokinase II catalyzes the conversion of glucose to glucose-6-phosphate, an essential step in glycolysis. While CA9 and HK2 are both direct transcriptional targets of the HIF-1α mediated hypoxic response,


TABLE 1 | Represenation of the types of epigenetic modifications that can be induced by environmental sensing, and the specific modifications made.

Informtion including the half-life of the modification, and how the modifications are written, erased, and read is included. This demonstrates the wide variety and complexity of epigenetic modifications that can be made in response to sensing, and the time-scales at which their effects are applicable.

this epigenetic regulation is important because it imposes the upregulation of these proteins over longer time scales (**Table 1**) and allows for the maintenance of the metabolic phenotype independently of oxygen status.

Epigenetic marks induced by nutrient sensing proteins and complexes have the ability to greatly alter cellular metabolism, making a useful feed-forward mechanism for acclimation and adaptation to the current metabolic microenvironment. The phosphorylation of Ser<sup>36</sup> on Histone 2B is a significant epigenetic mark made by two proteins involved in nutrient sensing: AMPK and S6K1 (39, 40). It has been shown that phosphorylation of Ser<sup>36</sup> on Histone 2B is significantly increased upon treatment of cells with 2-Deoxy Glucose (39), which mimics a glucose deprived environment. The resulting effect on the cellular transcription from phosphorylation of Ser<sup>36</sup> on Histone 2B by AMPK is the recruitment of EZH2 (40). EZH2 is a histone methyltransferase that trimethylates Lysine 27 on histone 3 when recruited. It has been shown in Drosophila that trimethylation of Lysine 27 on histone 3 reduces the glycolytic tendencies of the cell (56). Considering the presence of this mark in glucosedeprived cellular states, it would intuitively make sense that the presence of this mark would decrease the glycolytic capacity of the cell.

The suppression of SIRT4 by mTORC1 has profound effects on the metabolism of cancer cells, specifically inhibiting glutamine metabolism through inhibition of GDH. In colorectal cancer, decreased SIRT4 expression has been correlated with progression and increased invasive potential of cancer cells (57), and in both colorectal and gastric cancers lower SIRT4 expression is associated with poor prognosis (57, 58). All in all, this leads to the conclusion that when in the presence of sufficient amino acids and growth factors, the activation of mTORC1 will lead to the inhibition of SIRT4 and the subsequent reactivation of glutamine metabolism which can promote tumor growth (**Figure 2**).

mTORC2 has the ability to modulate activity of KMT2D. KMT2D is inhibited downstream during mTORC2 activation, which in response, inhibits the access of the FOXA1-PBX1-ER complex from binding the DNA. Prevention of this complex from binding the DNA has been shown to reduce the expression of key proteins including: GREB1, SERPINA1, cFOS, and MYC (59). Of these proteins, MYC has been shown to have the most substantial effects on reprogramming cancer metabolism in a type-specific manner. A comprehensive list of metabolic alterations in specific cancer types driven by MYC has recently been reviewed (60). The effect MYC has on glycolysis is highly variable depending on the cancer type, with a MYC-associated increase in non-small cell lung cancer and hepatocellular carcinoma, and a MYC associated decrease in renal cell carcinoma and prostatic intraepithelial neoplasia. MYC's effect on glutaminolysis was cohesive in the various cancer types, with an associated increase demonstrated in hepatocellular carcinoma, pancreatic ductal adenocarcinoma, and renal cell carcinoma.

In addition to the previously mentioned alterations in gene expression caused by stimulus-induced epigenetic modifications, epigenetic upregulation of MCT4 via hypomethylation of the SLC16A3 promoter has been shown in renal cancers (61). Although no specific mechanism can be tied to this alteration, this increase in MCT4 expression will have profound consequences on the long-term progression and evolution of the tumor.

### CONSEQUENCES OF EPIGENETICALLY ALTERED METABOLISM ON TUMOR PROGRESSION

As described previously, a resulting consequence of oxygen deprivation and stabilization of HIF-1α is the induction of TET1/3 and hypomethylation of the genome. Hypomethylation

has been shown to upregulate the expression of Hexokinase 2 (54, 55), which is associated with an increase in glycolysis (**Figure 3**). Increased glycolysis will increase acidosis in the tumor microenvironment that can induce extracellular matrix remodeling (15). Thus, increased glycolysis and its sequelae are barriers that cancer cells must overcome in order to meet the energy demands of rapid proliferation and to survive and thrive in a more hostile environment. Altered glycolysis can also lead to Warburg phenotype leading to even more acidic microenvironment and more altered genome and epigenome alteration (28, 62).

Despite the differing sensing mechanisms of the mTORC1/mTORC2 complexes and SIRT1, all mechanisms converge on a single metabolic alteration caused by epigenetic modification: the increase of glutamine metabolism. This may pose the opportunity to target glutamine metabolism as a cancer therapeutic, an idea that has drawn enough attention to warrant various reviews on that subject alone (63, 64). Although this may seem like a rational therapeutic target, cautionary narratives have been proposed as to the possible outcome of creating a resistant cellular population with a heightened metabolic capacity (65). While glutamine metabolism is non-essential in a normally proliferating cell, under periods of rapid proliferation, like tumor growth, glutaminolysis is an essential process (66). This phenotype is selected for due to the high demand for metabolic building blocks produced from the TCA cycle. In the TCA cycle alpha-ketoglutarate can be carboxylated to citrate, which, if in abundance, is translocated to the cytoplasm where it is used for fatty acid synthesis. The end product of glutaminolysis is alpha-ketoglutarate, which is shunted into the TCA cycle to accelerate the process (67). Supporting the TCA cycle with the necessary building blocks will give more chance to glucose to be turned into lactate in glycolysis and augment a Warburg phenotype. Glutamine-fed TCA cycle can also give more freedom to cancer cells to use glycolysis for their fluctuating ATP demand as a quick local source of energy for cancer cells (13).

### DISCUSSION

Genomic data describing tumors and cancer cell populations is valuable information for studying and classifying a cancers phenotypic characteristics. Although this information is valuable, it does not tell the entire story when it comes to cancer initiation and progression. Recent studies have demonstrated the presence of single and multiple driver mutations in significant proportions of cells comprising healthy tissue (68, 69). This poses the question as to how cancer arises, as the once thought sufficient accumulation of driver mutations has been discounted. While it is no doubt these mutations are necessary and play a significant role in cancer progression, it is clear there is more to the story.

Metabolic cross-talk and feedback is essential for the survival of any cell in a highly variable environment. There are many ways cells can sense perturbations in their microenvironment, including nutrient, and energy demands. These sensing mechanisms can transduce signals that lead to the alteration of the cell's epigenome. The epigenetic alterations driven by the alterations to the cellular environment and metabolic state can go on to influence the metabolism of the cell in a feed-forward mechanism. These changes made can have long lasting and heritable effects and go on to influence the progression of the tumor.

Nutrient sensing by cells is essential for the survival and acclimation to an unstable environment (**Figure 3**). Permanent changes induced by a given environment may be beneficial to a cells lineage, in that they are pre-programmed to deal with the environments endured by their predecessors. In cancer, cells are dynamically exposed to a variety of environmental conditions that would be extremely difficult to survive without epigenetic acclimation handed down from predecessors. Founding cancer populations may have a difficult time surviving the variety of harsh environmental factors, but genetic reprogramming facilitated by epigenetic change would allow the daughter populations to have an increased fitness.

The hyper-glycolytic state of cancer cells is a hallmark of their progression and aggressive state (28). At present, a conclusive mechanism as to the induction of this glycolytic state has yet to be achieved. Described in this paper are various mechanisms in which a cell is able to sense its' current environmental and energetic state of being, some of which can lead to long-lasting changes in the metabolic programming of a cell. Many of these semi-permanent changes converge on the regulation of cellular

### REFERENCES


energetics, in particular, glycolysis. It is therefore conceivable that the mechanism for the induction of a glycolytic state in cancer cells may not be a proposed "switch" or genetic mutation, but instead the accumulation of various epigenetic alterations that permanently reprogram the cellular population (**Figure 3**). It is possible that this reprogramming would occur early on in tumorigenesis, caused by a lack of perfusion in the core of the tumor which would cause many of the environmental perturbations that induce the epigenetic alterations described in this paper. What is certain, is that the spatial and temporal aspects of these epigenetic modifications would be vitally important for directing tumor progression. More studies need to be completed in order to elucidate how these epigenetic changes occur directly in relation to tumor growth in models and in the patient.

### AUTHOR CONTRIBUTIONS

MD and RG developed the idea. All authors wrote the manuscript.

### FUNDING

This work was supported by grants: U54CA193489, Cancer as a Complex Adaptive System, and P30CA076292, Moffitt Cancer Center Support Grant.


with consequences for clinical outcome. Clin Cancer Res. (2013) 19:5170–81. doi: 10.1158/1078-0432.CCR-13-1180


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

Copyright © 2020 Ordway, Swietach, Gillies and Damaghi. 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.

# PI(3,4)P2 Signaling in Cancer and Metabolism

Luca Gozzelino† , Maria Chiara De Santis † , Federico Gulluni † , Emilio Hirsch and Miriam Martini\*

*Department of Molecular Biotechnology and Health Sciences, University of Torino, Turin, Italy*

The phosphatidylinositide 3 kinases (PI3Ks) and their downstream mediators AKT and mammalian target of rapamycin (mTOR) are central regulators of glycolysis, cancer metabolism, and cancer cell proliferation. At the molecular level, PI3K signaling involves the generation of the second messenger lipids phosphatidylinositol 3,4,5-trisphosphate [PI(3,4,5)P3] and phosphatidylinositol 3,4-bisphosphate [PI(3,4)P2]. There is increasing evidence that PI(3,4)P2 is not only the waste product for the removal of PI(3,4,5)P3 but can also act as a signaling molecule. The selective cellular functions for PI(3,4)P2 independent of PI(3,4,5)P3 have been recently described, including clathrin-mediated endocytosis and mTOR regulation. However, the specific spatiotemporal dynamics and signaling role of PI3K minor lipid messenger PI(3,4)P2 are not well-understood. This review aims at highlighting the biological functions of this lipid downstream of phosphoinositide kinases and phosphatases and its implication in cancer metabolism.

#### Edited by:

*Daniel McVicar, National Cancer Institute (NCI), United States*

#### Reviewed by:

*Daniel Pereira Bezerra, Oswaldo Cruz Foundation (Fiocruz), Brazil Krishna Beer Singh, University of Pittsburgh, United States*

> \*Correspondence: *Miriam Martini miriam.martini@unito.it*

*†These authors have contributed equally to this work*

#### Specialty section:

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

Received: *30 October 2019* Accepted: *02 March 2020* Published: *31 March 2020*

#### Citation:

*Gozzelino L, De Santis MC, Gulluni F, Hirsch E and Martini M (2020) PI(3,4)P2 Signaling in Cancer and Metabolism. Front. Oncol. 10:360. doi: 10.3389/fonc.2020.00360* Keywords: AKT, INPP4, PTEN, cancer biology, phosphatases, phosphoinositide, PI3K, cancer metabolism

### INTRODUCTION

Phosphoinositides are synthetized in the endoplasmic reticulum by phosphatidylinositol synthase (PIS) and consist of a glycerol backbone, an inositol ring, and two fatty acid chains; in humans, these are usually enriched with stearic acid and arachidonic acid at the sn-1 and sn-2 position of their glycerol backbone, respectively (1, 2). Different phosphatidylinositol kinases can catalyze the binding of phosphate groups at the position 3, 4, or 5 of the inositol ring, while several phosphatases can specifically remove the phosphate groups (**Figure 1**). The different combinations of phosphorylation give rise to seven different PIs (phosphoinositides) (3). This variety of membrane signals leads to the finely tuned recruitment of different effectors at different time points, contributing to the maintenance of membrane identity and the control of signaling pathways and cytoskeletal and membrane dynamics (**Figure 2**). There is growing evidence that phosphatidylinositol 3,4-bisphosphate [PI(3,4)P2] is a critical second messenger in cancer, regulating vesicular trafficking, clathrin-mediated endocytosis, cytoskeletal rearrangements (lamellipodia and invadopodia), and cell metabolism [micropinocytosis and mammalian target of rapamycin (mTOR) signaling] (4).

In this review, we will discuss the specific contributions of kinases and phosphatases to PI(3,4)P2 synthesis and how they regulate PI(3,4)P2-dependent cellular functions.

### PI(3,4)P2 GENERATION BY KINASES

The synthesis of PI(3,4)P2 can proceed via class I and class II phosphoinositide 3-kinases (PI3Ks) that can directly phosphorylate the 3-OH of the plasma membrane (PM) phosphoinositide PI(4)P (5).

PI3Ks are a large family of lipid enzymes that phosphorylate the 3′ -OH group of PI at the plasma membrane (**Figure 1**). PI3K signaling encompasses the generation of phosphatidylinositol 3,4,5-trisphosphate [PI(3,4,5)P3] and PI(3,4)P2 that activate downstream effector proteins like serine/threonine kinase AKT (5, 6).

PI3Ks have been divided into three classes according to their structural characteristics and substrate specificity. Class I PI3Ks are the most commonly studied enzymes that are activated directly by cell surface receptors like receptor tyrosine kinase (RTK) and G-protein-coupled receptors (GPCRs). Besides the class I enzymes, recent studies revealed the importance of class II PI3Ks in cell proliferation, migration, and metabolism (6). Class III PI3K consists of a single catalytic vacuolar proteinsorting defective 34 (Vps34) subunit that generates only PI(3)P, an important regulator of membrane trafficking and mTOR signaling mediator (5).

There is increasing evidence that the three class II PI3K isoforms (PI3K-C2α, PI3K-C2β, and PI3K-C2γ) have distinct and non-overlapping cellular roles. Class II PI3Ks generate PI(3)P and PI(3,4)P2 from PI and PI(4)P, respectively, on spatially defined membrane sections regulating clathrin-mediated endocytosis (7), primary cilium function (8), and insulin signaling and sensitivity (9). The three isoforms of class II PI3Ks are homologous in sequence but differ in catalytic activities and biological functions. In particular, PI3K-C2α and PI3K-C2β are expressed in a wide range of tissues where they are catalytically active in several subcellular compartments, differently from PI3K-C2γ isoform that is present in a restricted number of tissues (6).

Several papers showed a dose-dependent effect of PI3K-C2α on proliferation (10, 11), emerging as the first tumor suppressor of the PI3Ks in breast cancer (11). Increased levels of PI3K-C2β expression promote tumorigenesis in breast, ovarian, prostate neuroblastoma, and esophageal cancers, possibly by an AKTdependent mechanism (12–15). Moreover, PI3K-C2β expression is associated with proliferation and invasion in breast cancer, being highly expressed in lymph nodes metastases compared to matching primary tumors, suggesting a pivotal role in the metastatic process.

A recent study reported that PI3K-C2β represses mTORC1 activity at lysosomes by the synthesis of a specific subcellular pool of PI(3,4)P2 (**Figure 3**) (16). Upon depletion of growth factors, PI3K-C2β represses mTORC1 activity through the

association of 14-3-3 proteins with the Raptor subunit of mTORC1 (16). In addition, the same group reported that protein kinase N (PKN) directly phosphorylates PI3K-C2β to induce its association with inhibitory 14-3-3 proteins, thus facilitating mTORC1 signaling (17).

### PI(3,4)P2 GENERATION BY PHOSPHATASES

### 5-Phosphatases: SHIP and INPP5

In addition to kinases, the PI(3,4)P2 levels are also controlled by specific phosphatase enzymes that hydrolyze PI(3,4,5)P3 (**Figure 1**). In cancer, the activity of SHIP1 and SH2-containing inositol 5′ -polyphosphatases (SHIP2) and 3-phosphatase tensin homolog (PTEN) is generally considered as a negative regulator of the PI3K axis by reducing the PI(3,4,5)P3 levels at the plasma membrane. The SHIP family includes two gene products, SHIP1 (encoded by INPP5D gene) and SHIP2 (INPPL1 gene), that share 43% of sequence identity at the amino acid level. They are composed of a SH2 domain at their N-terminal region, a 5-phosphatase domain and a prolin-rich C-terminal region that can be cleaved, generating different SHIP isoforms (18). SHIP acts as a specific phosphoinositide 5-phosphatases that dephosphorylates the PI(3,4,5)P3 to produce PI(3,4)P2 (19). SHIP1 expression is mainly restricted at the hematopoietic compartment, where it plays a key role in cytokine signaling, whereas SHIP2 has a wider expression pattern, including in the skeletal muscle, the heart, and the pancreas (20). SHIP1/2 are known to act as tumor suppressors, being frequently downregulated in several types of cancer, and they counteract the PI3K pathway by producing PI(3,4)P2 (21, 22). Nevertheless, there is growing evidence indicating that PI(3,4)P2 is critical for AKT activation, thus suggesting that SHIP1/2 have a

proto-oncogenic activity. Recently, pan and selective SHIP chemical inhibitors have been identified for the treatment of hematological malignancies (**Table 1**). In particular, SHIP1 selective chemical inhibitor 3AC (3α-aminocholestane) is able to block multiple myeloma (MM) cell lines, inducing apoptosis, and cell cycle arrest. Interestingly, SHIP inhibition can be rescued by the addition of exogenous PI(3,4)P2, suggesting that these inhibitors may have a broader application in multiple tumor types other than MM (23).

INPP5E has a substrate specificity similar to that of SHIP, being a 5-phosphatase which dephosphorylates PI(3,4,5)P3 and PI(4,5)P2 (24). INPP5E has a broad pattern of expression and it was reported to negatively regulate insulin-like growth factor 1-mediated AKT phosphorylation, counteracting the PI3K axis. Similarly, INPP5K, also known as skeletal muscle and kidney-enriched inositol 5-phosphatase (SKIP), suppresses insulin signaling in the skeletal muscle in an AKT-dependent manner (25, 26).

Differently from SHIP which mainly dephosphorylates PI(3,4,5)P3 and PI(4,5)P2, inositol polyphosphate 5-phosphatase (PIPP) is mainly involved in the hydrolysis of PI(3,4,5)P3 to PI(3,4)P2 (**Figure 1**). PIPP, also named as INNP5J, has two prolin-rich domains at the N- and C-terminal, respectively, and a SKIP C-terminal homology domain. In cancer, loss of enzymatic activity of the PIPP induces PI(3,4,5)P3 accumulation, promoting AKT1-dependent tumor growth and progression (27). In contrast, PIPP overexpression in esophageal squamous cell carcinoma decreases PI(3,4,5)P3 levels and AKT phosphorylation, concomitantly suppressing cell proliferation and anchorage-independent growth (28).



*PM, plasma membrane; EE, early endosome; LE, late endosome; GBM, glioblastoma; BC, breast cancer; CRC, colorectal cancer; AML, acute myeloid leukemia; NSCL, non-small cell lung.*

### 3-Phosphatase

Among the phosphatases, PTEN is one of the major regulators of PIs metabolism in the cell. PTEN is responsible for the hydrolysis of PI(3,4,5)P3 to PI(4,5)P2. From a structural point of view, PTEN consists of a phosphatase domain that contains the active site and a C2 domain that binds to the phospholipid membrane (29). The tumor suppressor activity of PTEN is based on the reduction of PI(3,4,5)P3 levels, repressing AKT activation and regulating a variety of cellular processes including proliferation, survival energy metabolism, and cellular architecture (30). Inactivating mutations of the PTEN gene are frequently reported in human tumors, including genomic locus deletions, missense/nonsense mutations, promoter methylation, and regulation of oncogenic microRNAs (31, 32). In addition, germline mutations in PTEN gene are associated with hereditary tumor syndromes such as PTEN hamartoma tumor syndromes and Cowden and Bannayan–Riley–Ruvalcaba syndrome (33).

Although it is widely accepted that PTEN is the major negative regulator of the PI3K axis through PI(3,4,5)P3 hydrolysis, there is growing evidence showing that PTEN can also function as a PI(3,4)P2 3-phosphatase in normal and pathological conditions (34, 35). Malek et al. recently reported that PTEN loss alone has no detectable effect on the PI(3,4)P2 levels, while the concomitant loss of PTEN and INPP4B, a PI(3,4)P2 4 phosphatase, leads to PI(3,4)P2 accumulation and increased cell growth and migration within epidermal growth factor (EGF) stimulation (34). Similarly, PTEN deletion inversely correlates with PI(3,4)P2 levels in the EGF-stimulated mouse model of prostate cancer and human prostate and breast cancer cell lines. Interestingly, PTEN/INPP4B loss in non-transformed cells (Mcf10a) potentiates EGF-dependent AKT phosphorylation and invadopodia formation, although it has opposite effects in breast and prostate cancer cells, reducing AKT activation possibly by inhibiting class I PI3K signaling (35). Finally, PTEN loss leads to PI3K/AKT-mediated mTOR activation, regulating cell metabolism and promoting glycolysis and pentose phosphate pathway (PPP) (36, 37).

These results suggest a novel functional role for PTEN together with INPP4B, regulating PI(3,4)P2 levels upon EGF stimulation and compensating each other in cancer (34, 35). It is possible that technical difficulties related to PI(3,4)P2 measurement in vitro, using recombinant proteins, and in vivo (cellular extracts) masked this new PTEN activity. However, the functional role of PI(3,4)P2 in promoting and sustaining cancer growth in PTEN-dependent tumors requires further investigation.

### 4-Phosphatases: INPP4A and INPP4B

Inositol polyphosphate 4-phosphatase (INPP4A) and its isoenzyme INPP4B are magnesium-independent phosphatases that hydrolyze the 4-phosphate from PI(3,4)P2 to form phosphatidylinositol-3-phosphate [PI(3)P] (38–40). Recent findings report that INPP4 can also hydrolyze PI(4,5)P2 and PIP3 in vitro (41).

INPP4A and INPP4B consist of an N-terminal C2 domain, a PEST domain, and a C-terminal lipid phosphatase "CX5R" motif and they share about 40% of sequence homology (38, 39). In cancer, INPP4A and INPP4B are considered to act as tumor suppressors by inhibiting the PI3K/AKT signaling pathway in several tumor types including breast, ovary, lung, pancreatic, melanoma, and esophageal cancer (40, 42, 43). In basal-like breast cancer patients, reduced INPP4B expression is associated with poor survival rate. Similarly, in prostate cancer, loss of INPP4B correlates with poor prognosis and reduced time to biochemical recurrence by promoting AKT activation (44, 45). In follicular variant of papillary thyroid carcinoma, loss of INPP4B expression or catalytic activity in mice is not oncogenic per se (41, 46). However, depletion of INPP4B phosphatase domain in a Pten+/<sup>−</sup> mice background results in malignant cancer and lung metastases by activating PI(3,4,5)P3-mediated AKT2 signaling. In normal tissue (41), INPP4B dephosphorylates both PI(3,4)P2 and PI(3,4,5)P3, but a concomitant loss of PTEN and INPP4B results in PI(3,4,5)P3 accumulation and enhanced AKT signaling, favoring tumor growth and progression (47).

Conversely, in acute myeloid leukemia, two recent reports showed that INPP4B over-expression correlates with poor prognosis (48, 49), suggesting that this effect is not due to the phosphatase activity of the enzyme (49). Therefore, in tumors with high INPP4B expression, the hydrolysis of PI(3,4)P2 can promote the activation of an alternative effector, such as serum and glucocorticoid-regulated kinase 3 (SGK3). In breast cancer, SGK3 is amplified and acts as PIK3CA oncogenic effector in a INPP4B phosphatase-dependent manner (50). In fact, PI(3)P binds to the SGK3 PX domain, leading to enhanced SGK3 activation and inhibition of AKT phosphorylation. INPP4B directly activates SGK3 through the hydrolysis of PI(3,4)P2, driving cell migration, anchorage-independent growth, and in vivo tumor development (50). Hence, the predominant mechanism by which INPP4B functions in cancer cells mostly relies on its ability to reduce PI(3,4)P2 levels. Further studies are required to better understand its oncogenic role in either promoting or inhibiting AKT activation in a specific genetic background or cellular context.

It is therefore possible that the level of activity of 4 posphatase differently affects oncogenic transformation on the basis of 5-phosphatase activity. As a result, one can hypothesize two scenarios:


Future work will lead to a further understanding of the nature of PI(3,4)P2 phosphatases in a specific cellular context and in cancer.

### PI(3,4)P2 IN CYTOSKELETON REARRANGEMENTS

Recent studies indicate that PI(3,4)P2 can promote cytoskeletal rearrangements at the plasma membrane by regulating lamellipodia maturation, podosomes, and invadopodia formation (4). Lamellipodia are actin-formed structures that protrude at the leading edge of the cell, which are essential for cell motility. Several works showed that the formation of these structures is highly dependent on the recruitment of lamellopodin to the plasma membrane (51–53). Lamellopodin promotes the growth of actin filaments by regulating the proteins of the enabled/vasodilator-stimulated phosphoprotein (Ena/Vasp) family, which have also been implicated in the interactions with talin and integrins (54). Of interest is the fact that lamellipodin contains a PH domain that mediates the binding to PI(3,4)P2, which possibly is produced by the 5-dephosphorylation of class I PI3K-derived PI(3,4,5)P3 at the leading edge (55).

SHIP2-derived PI(3,4)P2 also participates in the formation of podosomes and invadopodia (56). The primary function of these actin-rich structures is to degrade the extracellular matrix at the base of the cell to allow cell migration or to promote metastasis, respectively. In particular, mature invadopodia interact with the microtubule cytoskeleton, which delivers vesicular cargoes containing both membrane bound membrane type I-matrix metalloproteinase and soluble matrix metallopeptidases (57). Similar to the lamellipodia or the dorsal ruffles, the regulation of actin polymerization is strictly controlled by PIs. Indeed one of the main podosomal/invadopodial core proteins, tyrosine kinase substrate with 5 SH3 domains (Tks5), possesses a PH domain which binds PI(3)P and PI(3,4)P2 (58) and recruits N-Wiskott–Aldrich syndrome gene-like protein, actin-related proteins, dynamin-2, growth factor receptor-bound protein 2, and other proteins involved in the formation of podosomes. PI(3,4)P2 regulation in invadopodia is crucial in cancer invasion and metastasis. It has been demonstrated that, in breast cancer, the activation of PI3Kα or PI3Kβ is implicated in invadopodia formation and matrix degradation, following integrin signaling, by increasing the levels of PI(3,4,5)P3 used by SHIP2 as substrate to produce PI(3,4)P2 (59, 60). Moreover, decreased Tks5 expression correlates with reduction in tumor growth, metastasis, and angiogenesis in vivo (61). Reduced SHIP2 expression in PTEN-deficient glioblastoma cell lines is also linked to increased cell migration ability. In addition, SHIP2 regulates focal adhesion (FA) dynamics through the hydrolysis of PI(4,5)P2 to PI(4)P, inhibiting cell migration (62–64). In addition, SHIP2 generates PI(3,4)P2 at FA and invadopodia, inducing the development of mature FA, and lamellipodia extrusion (65). In contrast, increased SHIP2 protein levels correlate with lymph node metastasis, TNM stage, and reduced 5 year survival rate (66) in non-small-cell lung carcinoma and with increased cell migration and metastasis in breast cancer cells (67, 68). These findings indicate a novel mechanistic role of PI(3,4)P2 in the regulation of cell migration and invasion in cancer cells.

## PI(3,4)P2 IN NUTRIENT SCAVENGING

Advances in the study of PIs led to a deeper understanding of the processes regulating insulin signaling, energy homeostasis, signal transduction, and intracellular trafficking.

### Glucose and Cholesterol Metabolism

PI3K-C2α is known to promote glucose uptake by regulating the insulin-dependent translocation of glucose transporter 4 (GLUT4) to the plasma membrane (69). In addition, the knockdown of PI3K-C2α levels results in the rerouting of the insulin signal, promoting a beta-cell switch from a glucoseresponsive to a proliferative state (70). Recent work showed that reduced levels of liver-specific class II PI3K, PI3K-C2γ, induces the development of type II pre-diabetic condition by modulating an endosomal pool of PI(3,4)P2 required to sustain AKT2 activation (71). PI3K-C2γ null mice display reduced levels of liver glycogen and develop hyperlipidemia, adiposity as well as insulin resistance with age or after a high-fat diet. Conversely, SHIP2 null mice have normal levels of glucose and insulin but are resistant to weight gain in high-fat diet conditions (72). These results indicate that PI(3,4)P2 in vivo modulates glucose homeostasis in physiological conditions, regulating insulin secretion and glucose uptake.

Cholesterol metabolism involves multiple organelles, including the endoplasmic reticulum (ER), the Golgi apparatus, and the endosomal/lysosomal compartment, where exogenous cholesterol is catabolized. In particular, exogenous cholesterol is transported as cholesteryl ester by low-density lipoproteins, which can be internalized via receptor-mediated endocytosis. This endocytic pathway results in the fusion of endosomes with lysosomes, where cholesteryl ester is hydrolyzed and cholesterol is sorted to the ER or membranes. It is becoming more and more evident that these sorting routes are tightly controlled by PIs and, vice versa, the amount of cholesterol can regulate PIs turnover and lysosomal activity. Sterols and lipids can be moved between organelles and membranes by lipid transfer proteins in a non-vesicular way. Proteins of the oxysterol-binding protein 1 or oxysterol-related proteins (ORP) families can associate with phospholipids, uptake cholesterol, and mediate its transport toward acceptor membranes (73, 74). Moreover, ORP1L has also been demonstrated to be a cholesterol sensor that regulates the microtubule-dependent movement of lysosomes and late endosomes. Recent work demonstrated that ORP1L is able to strongly bind to PI(3,4)P2 on the surface of lysosomes and functions as a shuttle to transport cholesterol toward the ER (74). Through a structural analysis of the protein, the authors demonstrated that the binding of ORP1L to PI(3,4)P2 allows the cholesterol uptake from lysosomes, which in turn loosened the binding of ORP1L to the lysosomal membranes, promoting the detachment of the protein and the delivery of lipids to the ER. Interestingly, sorting of cholesterol from lysosomes dramatically decreased in ORP1L knockout cells or when PI3K-C2β was silenced, indicating that the two proteins cooperate in regulating the endosomal cholesterol efflux. Given that lysosomal cholesterol activates mTORC1 through solute carrier family 38 member 9 (75), it is evident that both the PI(3,4)P2 production by PI3K-C2β and the subsequent PI(3,4)P2-regulated cholesterol export from lysosomes contribute to the silencing of mTORC1.

### Macropinocytosis and Lysosomal Catabolism

In order to support tumor growth and fuel biosynthetic pathways, cancer cells are able to modulate vesicle generation from the plasma membrane processes (namely, endocytosis) to scavenge proteins and lipids from the extracellular space. Macropinocytosis is a regulated form of endocytosis in which cells can uptake nutrients and small particles from the extracellular environment fluid phase. Macropinosomes are formed by the folding back of dorsal membrane ruffles that fuse with the plasma membrane and create large endocytic structures (76). Oncogenic Ras can stimulate nutrient uptake by the upregulation of nutrient transporters to the plasma membrane and by inducing the membrane ruffles in a PI3K-dependent manner. Recently, it has been reported that Ras-transformed cells depend on macropinocytosis to internalize extracellular protein into the cell and sustain their metabolic needs (77). In particular, oncogenic Ras regulates the early events of macropinosome formation through the activation of class I PI3K and PI(3,4,5)P3 production at the plasma membrane (78). The actin-driven membrane ruffling is under control of class I PI3Ks. The local production of PI(3,4,5)P3 coordinates the activity of Rho and Arf GTPases, which organize the cortical cytoskeleton to form dorsal ruffles (**Figure 2**), and recruits PH-domain-containing effector molecules that shape the macropinosomes, such as AKT and a myosin-I motor protein (79). However, the closure of the dorsal ruffles to form the macropinocytic vesicle needs a sequential breakdown of PIs from PI(3,4,5)P3 (80). It has been observed that SHIP2 catalyzes the conversion from PI(3,4,5)P3 to PI(3,4)P2, leading to the recruitment of TAPP1 to the dorsal ruffles. TAPP1 is then able to recruit the actin-binding protein syntrophin (81), contributing to cytoskeletal rearrangement. However, it is also conceivable that PI(3,4)P2 could promote the recruitment of other effectors, like the SNX (sortin nexin) family proteins, to detach the nascent macropinosome from the plasma membrane, with a similar mechanism to CME. Next, INPP4B activity is required during this process to sequentially dephosphorylate PI(3,4)P2, indicating that this lipid is not only a degradation product of PI(3,4,5)P3 but has its own functions during micropynocitosis (80). The final hydrolysis of PI(3)P to PIs by the phosphatases MTM6 and MTM9 (myotubularinrelated protein) is then necessary for the closure of the membrane ruffles and the completion of the pinocytic process (80).

The requirement of PI(3,4,5)P3 in macropinocytosis is demonstrated using PI3K inhibitors (82). Reducing PI(3,4,5)P3 levels, by the use of wortmannin or LY294002, can block the closure of macropinosomes before the circular dorsal ruffle formation (80). In addition, in tumors harboring PTEN loss, the expression of class I PI3Kβ is required to maintain elevated levels of micropinocytosis (83).

The fate of the extracellular macro-nutrients obtained by micropinocytosis is then regulated by mTORC1, which inhibits the lysosomal catabolism of proteins. Consistently, in amino-acid-starved conditions, the inhibition of mTORC1 sustains tumor growth by upregulating macropinocytosis and the catabolism of engulfed proteins (84). Aside from regulating the scavenging of extracellular nutrients, mTORC1 also negatively regulates autophagy in nutrient-rich conditions through the regulation of a protein complex composed of unc-51-like kinase 1, autophagy-related gene 13, and focal adhesion kinase family-interacting protein of 200 kDa (85). Therefore, mTORC1 suppresses the use of proteins as nutrients, limiting the cellular metabolic flexibility during times of nutrient abundance.

Recent studies revealed a link between lysosomal position and nutrient signaling via mTORC1 (86). Lysosome distribution is tightly controlled by a complex interplay between small GTPases, such as Rab7, and kinesins, that are in turn regulated by several factors including PI. In particular, under conditions of growth factor deprivation, PI3K-C2β is responsible for the production of a lysosomal pool of PI(3,4)P2 that inhibits mTORC1 and facilitates the perinuclear clustering of lysosomes (16). These findings suggest that the lysosomal generation of PI, including PI(3)P and PI(3,4)P2, depends on nutrient levels, directly regulating mTORC1 signaling.

Given the key role of lysosomes in nutrient sensing, future studies will need to address how PIs couple mTORC1 regulation via the autophagy/lysosome pathway.

### Clathrin-Mediated Endocytosis

Besides micropinocytosis, which is a clathrin- and caveolinindependent endocytotic process, CME is one of the main processes through which cells internalize surface molecules and proteins, including receptors, nutrients, and growth factors. The initialization and the timing of CME are highly regulated by a close interplay between the different PI3Ks and phosphatases, which modify the composition of the plasma membrane in the endocytic pits and recruit several effectors, such as adaptors, membrane-deforming scaffolds, and actin modulatory factors (87). For many years, PI(4,5)P2 was thought to be the main regulator of the process, given the presence of a multitude of PI(4,5)P2 binding proteins in the endocytic machinery. Early-acting clathrin adaptors AP-2 (88), membrane remodeling proteins containing BAR domains (89), and dynamin (90) are all able to bind PI(4,5)P2, giving a functional identity to the clathrin-coated pits (CCPs) and promoting the invagination and the fission of the clathrin-coated vesicles. However, it has been recently demonstrated that PI(3,4)P2 also exerts a crucial role in the maturation of the vesicles (7) (**Figure 2**). During the early stages of CME, different 5-phosphatases such as SHIP2 or synaptojanin are recruited to the CCP, suggesting that the levels of PI(4,5)P2 decline as the vesicle matures (91). The resulting accumulation of PI(4)P is used as a substrate for the PI3K-C2αdependent synthesis of PI(3,4)P2. The pool of PI(3,4)P2 is able to recruit the PX-BAR domain proteins SNX9 and SNX18, which contribute to the formation of the narrow neck in the nascent vesicles that will be finally cut by dynamin, leading to the release of the vesicles (7). The production of PI(3,4)P2 also contributes to the recruitment and activation of the adaptor protein FCHSD2 (FCH and double SH3 domains protein 2) (92). This F-BAR containing protein promotes the formation of actin structures around the CCPs, leading to the efficient invagination of the plasma membrane.

Recent findings showed that PI(3,4)P2 can also be directly synthetized during the process of clathrin-mediated pinocytosis by PI3K-C2β (93). In this work, PI3K-C2α is recruited to the clathrin-coated structures through its binding to PI(4,5)P2, in a similar way to CME, while PI3K-C2β is recruited to the plasma membrane by the interaction with the scaffold protein ITSN1 (intersectin 1). The production of PI(3,4)P2 by PI3K-C2β recruits FCHSD2, which contributes to actin polymerization and drives the mechanical force needed for the abscission of the vesicle.

CME is a complex process that requires several adaptors, membrane curvature effectors, and a membrane scission machinery. Therefore, the formation of the clathrin-coated pits and the release of clathrin-coated vesicles take a relatively long time. However, in specific conditions, cells can internalize some receptors and ligand–receptor complexes in a faster way, through the so-called fast endophilin-mediated endocytosis (FEME), which does not require clathrin. This endocytic route relies on endophilin, a protein containing a SH3 domain, a BAR domain, and possesses multiple amphipathic helices (94). The coexistence of these domains allows endophilin to recognize cargo receptors by itself, to promote membrane curvature, and to support membrane scission in collaboration with dynamin (95, 96). The process of FEME starts with the activation of the plasma membrane receptors and the production of PI(3,4,5)P3. Subsequently, the production of PI(3,4)P2 by SHIP1/2 leads to the recruitment of lamellipodin and its binding partner endophilin, which induces actin rearrangement together with WASP or Ena/Vasp proteins (96) (**Table 1**). Thus, PI(3,4)P2 seems to act in a very similar way both in CME and in FEME.

### ROLE OF PI(3,4)P2 IN CANCER METABOLISM

The formation of PI(3,4)P2 has different effects in the biology of the cell: it is an alternative pathway for the removal of PI(3,4,5)P3, a source for the production of PI(3)P, and the origin of a second messenger. The inappropriate accumulation of PI(3,4)P2 due to excess in production or defects in degradation contributes to several disorders caused by mutations in kinases and phosphatases, respectively. The main source of PI(3,4)P2 derives from the dephosphorylation of PI(3,4,5)P3 mediated by SHIP. While it is clear that the tumor suppressor function of PTEN is due to the removal of PI(3,4,5)P3 and PI(3,4)P2, the effect of SHIP on tumor transformation was debated. The tumor suppressor role of the 5-phosphatases can be associated to the removal of the PI(3,4,5)P3-dependent signaling and consequent downregulation of class I PI3K signaling or to a negative feedback loop of PI(3,4)P2 (27). On the contrary, it is now clear that, in a certain context, SHIP may facilitate tumor cell survival, contrarily to PTEN, due to a different effect on AKT. In particular, the removal of PI(3,4,5)P3 is anti-tumoral if followed by PI(4,5)P2 production mediated by PTEN, while it is protumoral if it results in SHIP-mediated accumulation of PI(3,4)P2. Accordingly, the increase in PI(3,4)P2 levels, due to the loss of the tumor suppressor function of 4-phosphates, is usually considered as pro-growth (40), and mutant mice for INPP4B show mammary epithelial transformation. The SHIP-mediated increase of PI(3,4)P2 enhances the number of docking sites at the plasma membrane in order to recruit and activate PH-containing kinases such as AKT. Thus, SHIP, which is mainly expressed in blood cells, promotes MM and leukemia cell transformation. The requirement of both PI(3,4,5)P3 and PI(3,4)P2 to sustain malignant transformation is known as "the two PIP hypothesis" (18). In line with this concept, both the agonistic and the antagonistic compounds of SHIP are efficient in killing multiple myeloma cells. SHIP inhibition using 3AC abrogates MM growth (23); however, another group demonstrated that the reduction of SHIP expression levels did not affect MM cell proliferation.

While the loss of INPP4 [with PI(3,4)P2 accumulation] or SHIP [with PI(3,4,5)P3 increase] is not sufficient to promote tumor formation, PTEN deficiency alone (impacting on both lipids) is associated with a high incidence of cancer (**Table 1**). These data suggest that the uncontrolled production of both lipids is necessary to induce cell transformation mediated by PI3K dysregulation. Alternatively, some phosphatases may have partially redundant roles; thus, single mutants are not enough to impact on tumor onset.

Another way of PI(3,4)P2 derivation comes from class II PI3K. However, by now, there is no direct evidence about class II PI3Kderived PI(3,4)P2 in tumor. In fact, the only study based on a cancer mouse model and a patient cohort demonstrates the implication of PI3K-C2α in breast cancer progression due to its scaffold rather than to its catalytic function (11).

It is becoming clear that PI(3,4)P2 is not only a lipid component that characterizes membrane identity but that it also acts as a signaling molecule. The pathologies associated with PI(3,4)P2 accumulation are characterized not only by mutations in kinases and/or phosphatases but also by alterations in the PI(3,4)P2-dependent signaling pathway. The PI(3,4)P2 effectors contain 3-phoshoinositide binding domains, including PH, PX, FYVE, ANTH, ENTH, and FERM. Although some proteins selectively bind to PI(3,4,5)P3, most proteins, such as AKT and PDK, have dual specificities for both PI(3,4,5)P3 and PI(3,4)P2, making it difficult to discriminate between the functional contributions of the two lipids in different signaling contexts. However, it is becoming clear that PI(3,4,5)P3 mainly regulates AKT1 at the plasma membrane, while PI(3,4)P2 regulates a spatially restricted pool of AKT2 in endosomes. The AKT-mediated phosphorylation of mouse double minute 2 homolog, a negative regulator of p53, causes the activation of pyruvate kinase isozymes, which are involved in the last step of glycolysis, preventing serine synthesis and promoting glycolysis (97). By now, the only convincing specific effect mediated by PI(3,4)P2 is the Tapp1/2-induced feedback inhibition of class I PI3K. In response to insulin stimulation, the ko for Tapp1 results in increased AKT stimulation in the heart and the muscles (98). Interestingly, the same phospholipid mediates opposite effects according to its localization: at the plasma membrane, PI(3,4)P2 downstream class I PI3K activates mTOR and pro-survival signals, while on the lysosome it inhibits mTOR in serum-starved conditions. At the plasma membrane, PI3K-C2α promotes

### REFERENCES

1. Agranoff BW, Bradley R M, Brady RO. The enzymatic synthesis of inositol phosphatide. J Biol Chem. (1958) 233:10 77–83.

glucose uptake, impacting on GLUT4 translocation upon insulin stimulation (69).

On the contrary, another class II PI3K, PI3K-C2γ, controls the endosomal pool of PI(3,4)P2 on endosomes, where it regulates glycogen deposition in the liver. Ko mice display insulin resistance and dyslipidemia with age (71). Given its implication in metabolism and its specific pattern of expression, PI3K-C2γ could be involved in tumors highly dependent on metabolism, such as pancreatic cancer.

PI3K-C2β is also involved in the production of PI(3,4)P2 at endomembranes, in particular, on the lysosome. After nutrient deprivation, the PI3K-C2β-derived phospholipid inhibits mTOR, suggesting a metabolic impact of this kinase (16). In particular, mTORC1 activation sustains glycolysis and the PPP oxidative arm, promoting the expression of glucose-6-phosphate dehydrogenase. In addition, mTOR is responsible of the de novo purine synthesis for nucleic acid production and of mitochondrial biogenesis in order to boost ATP production (37, 99). It was recently demonstrated that both PI3K-C2α and PI3K-C2β regulate clathrin-mediated pinocytosis, a fluid endocytosis involved in the acquisition of nutrients from the extracellular environment (93). This study further supports the involvement of both class II PI3K in cell metabolism control.

Understanding how PI(3,4)P2 contributes to several diseases, mainly linked to cancer and metabolism, would be therapeutically beneficial. In particular, the alteration of the PI3K signaling as well as the defects in phosphoinositide-metabolizing enzymes are the main causes of PI(3,4)P2-related diseases. Given that the production of PI(3,4)P2 comes from different pathways, the manipulation of only one enzyme is not sufficient to interfere with its physiological function. By now, the main inhibitors in clinical trials target class I PI3K, while drugs against phosphatases and specific for class II PI3Ks are still at the preclinical level and need further studies.

### AUTHOR CONTRIBUTIONS

LG, MD, FG, EH, and MM wrote the manuscript. FG did the artwork. MD made the table. All authors contributed to manuscript revision, read, and approved the submitted version.

### FUNDING

This work was supported by the Associazione Italiana per la Ricerca sul Cancro (AIRC21875). MD was supported by FIRC/AIRC fellowships (Grant no. 22248). FG was supported by Fondazione Pezcoller/SIC -Patrizia Coser. MM was supported by Ricerca Sanitaria Finalizzata Giovani Ricercatori (GR-2018- 12365776) and Worldwide Cancer Research grant (20-0033).

2. Epand RM. Recognition of polyunsaturated acyl chains by enzymes acting on membrane lipids. Biochim Biophys Acta. (2012) 1818:957–62. doi: 10.1016/j.bbamem.2011.07.018

3. Balla T. Phosphoinositides: tiny lipids with giant impact on cell regulation. Physiol Rev. (2013) 93:1019–137. doi: 10.1152/physrev.00028.2012


expression in prostate carcinoma predicts for recurrence and poor long term survival. Prostate. (2015) 75:92–102. doi: 10.1002/pros.22895


domain-containing proteins leads to GTP-dependent membrane scission. J Biol Chem. (2013) 288:6651–61. doi: 10.1074/jbc.M112.444869


**Conflict of Interest:** EH is co-founder of Kither Biotech, a company involved in the development of PI3K inhibitors.

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

Copyright © 2020 Gozzelino, De Santis, Gulluni, Hirsch and Martini. 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.

# In-vitro NMR Studies of Prostate Tumor Cell Metabolism by Means of Hyperpolarized [1-13C]Pyruvate Obtained Using the PHIP-SAH Method

Eleonora Cavallari <sup>1</sup> , Carla Carrera<sup>2</sup> , Ginevra Di Matteo<sup>1</sup> , Oksana Bondar <sup>1</sup> , Silvio Aime<sup>1</sup> and Francesca Reineri <sup>1</sup> \*

*<sup>1</sup> Department of Molecular Biotechnology and Health Sciences, Center of Molecular Imaging, University of Turin, Turin, Italy, 2 Institute of Biostructures and Bioimaging, National Research Council, Turin, Italy*

Nuclear Magnetic Resonance allows the non-invasive detection and quantitation of metabolites to be carried out in cells and tissues. This means that that metabolic changes can be revealed without the need for sample processing and the destruction of the biological matrix. The main limitation to the application of this method to biological studies is its intrinsic low sensitivity. The introduction of hyperpolarization techniques and, in particular, of dissolution-Dynamic Nuclear Polarization (d-DNP) and ParaHydrogen Induced Polarization (PHIP) is a significant breakthrough for the field as the MR signals of molecules and, most importantly, metabolites, can be increased by some orders of magnitude. Hyperpolarized pyruvate is the metabolite that has been most widely used for the investigation of metabolic alterations in cancer and other diseases. Although d-DNP is currently the gold-standard hyperpolarization method, its high costs and intrinsically slow hyperpolarization procedure are a hurdle to the application of this tool. However, PHIP is cost effective and fast and hyperpolarized pyruvate can be obtained using the so-called Side Arm Hydrogenation approach (PHIP-SAH). The potential toxicity of a solution of the hyperpolarized metabolite that is obtained in this way is presented herein. HP pyruvate has then been used for metabolic studies on different prostate cancer cells lines (DU145, PC3, and LnCap). The results obtained using the HP metabolite have been compared with those from conventional biochemical assays.

Keywords: nuclear magnetic resonance, pyruvate, hyperpolarization, para-hydrogen, metabolism

### INTRODUCTION

Citation:

*Cavallari E, Carrera C, Di Matteo G, Bondar O, Aime S and Reineri F (2020) In-vitro NMR Studies of Prostate Tumor Cell Metabolism by Means of Hyperpolarized [1-*13*C]Pyruvate Obtained Using the PHIP-SAH Method. Front. Oncol. 10:497. doi: 10.3389/fonc.2020.00497*

NMR is a powerful and non-invasive tool for the investigation of cellular metabolism as it allows a wide range of metabolites to be detected either in cell cultures or in samples obtained from tissues resected from living systems. The main drawback of magnetic resonance spectroscopy in biological specimens is the low

intensity of the MR signals. Thermal nuclear spin polarization P is typically in the order of 10−<sup>5</sup> -10−<sup>6</sup> for conventional high-field NMR spectrometers. The signal from water protons is predominant by far, while cell metabolites, whose concentration is about 10,000 times lower, can only be observed in vivo with low spatial resolution and after long acquisition times. Moreover,

#### Edited by:

*Alessandra Castegna, University of Bari Aldo Moro, Italy*

#### Reviewed by:

*Krishna Beer Singh, University of Pittsburgh, United States Stephen John Ralph, Griffith University, Australia*

> \*Correspondence: *Francesca Reineri francesca.reineri@unito.it*

#### Specialty section:

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

Received: *13 December 2019* Accepted: *19 March 2020* Published: *17 April 2020* the <sup>1</sup>H-NMR spectra acquired in vivo or in tissues can only provide information about the concentration of metabolites in a steady-state.

The breakthrough in the field came with the introduction of the d-DNP hyperpolarization method (1), which increased the sensitivity of MR signals by some orders of magnitude, thus allowing the visualization of metabolites to be performed in cells and in vivo. More interestingly, the acquisition of timeresolved spectra becomes possible, thus providing information about the kinetics of metabolic transformations, in cell cultures and in vivo (2, 3).

[1-13C]pyruvate is the substrate that has been used most widely for the study of cellular metabolism and its alterations in diseases such as cancer, heart failure and stroke (4–8). Pyruvate plays a central role in cellular metabolism as it can be converted into a number of metabolites, depending on cellular conditions. In most normal tissues, pyruvate dehydrogenase catalyzes the decarboxylation of the pyruvate that enters the TCA cycle. Pyruvate can also be transaminated by alanine aminotransferase or reduced to lactate by lactate dehydrogenase. The presence of disease can alter metabolic fluxes through the different enzymes and an increased glycolytic flux is often observed in tumors. The use of hyperpolarized pyruvate has shown a marked upregulation in the exchange of the <sup>13</sup>C hyperpolarized label, mediated by the LDH enzyme, between pyruvate and lactate in tumor cells (9). Hyperpolarized [1-13C]pyruvate is currently under intense scrutiny as a potential probe with which to assess the presence and grading of prostate cancer in humans (10).

Unfortunately, d-DNP is an inherently slow, technologically demanding and extremely expensive technique, limiting the application of this powerful tool to a few laboratories worldwide.

ParaHydrogen Induced Polarization (PHIP) allows hyperpolarized substrates to be obtained in a few seconds and with much lower costs than d-DNP (11). This hyperpolarization method is based on the hydrogenation, catalyzed by metal complexes, of an unsaturated precursor of the target molecule (12–14). On this basis the number of PHIP-polarizable substrates appears to be limited by the availability of the proper unsaturated precursor and the concerns about the biological applications of the obtained hyperpolarized products are related to the presence of residual metal catalyst and other impurities that are associated to the hydrogenation reaction. The PHIP- side-arm hydrogenation (PHIP-SAH) approach (15) is a good step ahead as it circumvented both issues and can provide aqueous solutions of hyperpolarized pyruvate, or other metabolites (16), that can be safely used for in-vivo and in-cells studies (17, 18). Nevertheless, although the PHIP-SAH procedure allows most of the toxic compounds to be removed (catalyst and solvent), traces that may affect the viability and metabolism of cells may still be present in the aqueous phase of the final product.

This work thoroughly investigate the toxicity of aqueous solutions of the HP metabolites by performing in-vitro cytotoxicity analyses on prostate cancer cell lines.

HP [1-13C]pyruvate was found to be a sensitive tool for investigation into the difference in the metabolic phenotype of two highly aggressive and metastasizing human prostate carcinoma cell lines, PC3 and DU145 and another, less aggressive one, LNCaP.

### MATERIALS AND METHODS

### [1-13C]pyruvate Hyperpolarization

<sup>13</sup>C-labeled HP pyruvate was obtained using the PHIP-SAH method (**Figure 1**). The hydrogenation catalyst ([1,4 bis(diphenylphosphino)butane](1,5-cyclooctadiene) rhodium(I) tetrafluoroborate, Sigma Aldrich, 1.38 µmol) was dissolved in deuterated chloroform (CDCl3, 100 µl). The propargylic ester of [1-13C]pyruvate (3 µl, 26.6 µmol) was added to this solution. A 5 mm NMR tube (equipped with a gas valve) was used as the hydrogenation reactor and was pressurized with 2.1 bar of paraenriched hydrogen (∼86% enriched), while the NMR tube was kept in a liquid nitrogen bath.

The sample was kept frozen in liquid nitrogen (time intervals from 0.5 to 7 h) until the start of the hyperpolarization experiment. The hyperpolarization procedure consisted of the following steps:


to the aqueous phase to reach physiological pH (7.2 ± 0.2 was measured).

d) Phase separation: the aqueous solution of the HP product (250 ul) was taken up into a syringe and diluted with 600 µl of distilled H2O before being transported to the NMR spectrometer, where 230 ± 11 ul of this solution was quickly added to the NMR tube containing the cell suspension.

The concentration and hyperpolarization of [1-13C]pyruvate was measured at the end of each experiment by acquiring a <sup>13</sup>C-NMR spectrum of the thermally polarized product with the addition of <sup>13</sup>C-urea as an internal reference. The pyruvate concentration in the cell suspensions was found to be 5.0 ± 0.4 mM.

## <sup>13</sup>C-NMR Experiment Set-Up

The <sup>13</sup>C-NMR spectra were carried out using 5 mm NMR tubes, instead of the previously reported 10 mm sample tubes (18). A standard 5 mm-OD Wilmad NMR tube was equipped with a cap modified with a central hole (inner diameter 2 mm) and a glass tube (2 mm OD, 1.75 mm ID, 8 mm L) was inserted into the cap. A PTFE tube (1 mm OD, 0.75 mm ID, 750 mm L) was placed inside the glass capillary tube and kept in the axial position, a few mm above the cells suspension. The quick injection of the aqueous solution of the hyperpolarized substrate through the PTFE injection line allowed the cell suspension to mix with the HP pyruvate. The PTFE tube was removed immediately after the addition of the HP-pyruvate in order to avoid B<sup>0</sup> inhomogeneities during the acquisition of the <sup>13</sup>C-NMR spectra.

### <sup>13</sup>C-NMR Experiments: Cells Preparations

The metabolic transformation of [1-13C]pyruvate, hyperpolarized through the PHIP-SAH procedure, into [1- <sup>13</sup>C]lactate, was assessed on the following specimens: (I) prostate cancer cells suspended in their growth medium; (II) intact cells suspended in lactate-enriched medium; (III) lysed cells. The samples (volume of 300 ul) were placed in the 5 mm NMR tube equipped with the Teflon transfer line ready to receive the addition of the HP substrate. The NMR tube was positioned into the NMR spectrometer (600 MHz Bruker Avance) where it was kept at 310 K.

The contents of the specimens were as follows:


cell suspension in liquid nitrogen. The sample-containing cells were prepared as in II. In this set of experiments L-lactate (10 mM) was also added to the growth medium and left to equilibrate for 10 min in the NMR spectrometer before the HP experiment.

The acquisitions of the <sup>13</sup>C-NMR spectra were carried out using a small flip angle (18◦ ). A 2 s delay was applied between successive acquisitions. The first acquisition started a few seconds before the injection of the HP substrate.

The vitality of the cells at the end of the hyperpolarized experiment was checked using the trypan blue exclusion test. The viability in all of the experiments was 95%.

At the end of the experiment the number of cells was checked via the quantification of Bradford proteins, using the specific calibration line for each cell line.

The <sup>13</sup>C NMR spectra were acquired on a Bruker Avance 14.1 T NMR spectrometer using a 5 mm BBO probe equipped with <sup>1</sup>H and <sup>13</sup>C coils.

### Cells Cultures

The PC3, DU145, and LNCaP (prostate carcinoma) cell lines were purchased from American Type Culture (ATCC <sup>R</sup> ). Cells were kept for 72 h in 175 cm<sup>2</sup> flasks at 310 K in a humidified atmosphere with 5% CO<sup>2</sup> in the recommended culture media as suggested by ATCC, in particular: LNCaP: RPMI1640 (D-glucose 25.0 mM); PC3: F-12K (D-glucose 7.0 mM); DU145: EMEM (Dglucose 5.6 mM). All the cells were used within the first 10 passages from unfreezing.

### In-vitro Cytotoxicity

To assess the adverse effects of the aqueous PHIP-SAH hyperpolarization derived solutions, the MTT test (based on the enzymatic reduction of the tetrazolium salt MTT [3-(43 dimethylthiazol-2-yl)-2,5-diphenyl-tetrazoliumbromidein]) in living, metabolically active cells was carried out on two prostate cancer cell lines (DU145 and PC3).

For these tests, cells were harvested via trypsinization, resuspended in fresh medium, and plated in the wells of 96 well-microtiter plates at a volume of 0.1 ml. Routinely 7,000– 16,000 cells, depending on the cell-lines growth curves, were plated in each well. After 24 h, to ensure the cells adhesion, the culture medium was removed, and replaced with a fresh medium of the same composition as the one used in the hyperpolarization procedure, i.e., 15% of the aqueous solution in the final volume (0.1 ml).

At the end of the incubation period (1, 6, and 24 h), the medium was replaced with 0.1 ml of a 5 mg/ml solution of MTT (purchased from Sigma, St. Louis, MO) in phosphate-buffered saline (PBS 1X). After 4 h of incubation with MTT at 310 K, the supernatant was carefully sucked off and a solubilization solution, 0.15 ml of dimethyl sulfoxide, was added to dissolve the insoluble purple formazan product into a colored solution and the absorbance at 570 nm was read by a spectrophotometer.

Cells were plated in triplicate to minimize the variability of the results. In each plate, 3 control wells for each cell line were included.

The PC3 and DU145 cells were treated with the following solutions:


### Lactate Dehydrogenase Assay

A commercial kit (Sigma-Aldrich MAK066) was used to measure the LDH activity in the cell lines. The kit was used according to the manufacturer's instructions. Briefly, 1 × 10<sup>6</sup> cells per sample were rapidly homogenized by sonication (30% power, 21 W, for 30 s) over ice in 500 µl of cold LDH Assay buffer, and was then centrifuged at 10,000 g for 15 min at 277 K to remove the insoluble material.

To ensure the readings were within the linear range of the standard curve, 4–10 µl samples were added into duplicate wells of a 96-well-plate, bringing the sample to the final volume of 50 µl with LDH Assay Buffer. After the addition of 50 µl of the Master Reaction Mix the NADH production kinetics were measured via the absorbance of the specific probe at 450 nm.

### Extracellular Lactate Assessment

To determine the concentration of lactate in the supernatant, cells were seeded on 75 mm<sup>2</sup> plates. After overnight adhesion in standard conditions (310 K and 5% CO2), the culture medium was replaced with a new one. After 72 h, the medium was collected, and the cells were harvested by trypsinization and then counted. The applied procedure allowed the amount of lactate measured in the culture medium to be normalized to the number of cells.

In order to immediately quench any possible residual metabolism, one volume of culture medium was mixed with two volumes of cold methanol in a vial, snap-frozen and left 2 h in liquid nitrogen. Proteins were then allowed to precipitate at 250 K for 30 min and the sample was centrifuged at 16,000 g at 277 K for 20 min. The supernatant was then collected and immediately lyophilized to remove the methanol for subsequent measurements, and was then reconstituted with 0.6 ml of phosphate buffer (0.15 M K2HPO4, pH 7.0) in deuterated water (D2O) for <sup>1</sup>H NMR quantification.

The amount of extracellular lactate was determined using NMR <sup>1</sup>H spectrometry (see **Supplementary Material**). Experiments were carried out in triplicate and the lactate concentration in the samples (µmol/cell) was calculated based on the internal standard reference. A known amount (0.35 mM) of 3-(trimethylsilyl)-propionic-d<sup>4</sup> acid sodium salt (TSP-d4) was added to act as a chemical shift reference for the calibration of the NMR data (at 0.0 ppm) as well as an internal standard for quantitation.

### Intracellular Lactate Concentration

Methanol-chloroform-water extraction (M/C) was used to extract metabolites from cells (19).

Briefly, cell pellets were promptly quenched in liquid nitrogen followed by the addition of 0.5 ml of cold methanol-chloroform solution at a ratio of 2:1. After thawing over ice, samples were vortexed for 60 s and sonicated. After 15 min of contact with the M/C solution, 0.25 ml of chloroform and 0.25 ml of distilled water were added to the mixture to yield an emulsion, which was vortexed and centrifuged at 13,000 g for 20 min at 277 K. The upper layer which contained the water-soluble metabolites was collected, lyophilized and reconstituted with 0.6 ml of phosphate buffer (0.15 M K2HPO4, pH 7.0) in deuterated water (D2O) for <sup>1</sup>H NMR quantification (see **Supplementary Material**).

An internal standard (TSP-d4) concentration of 0.03 mM was used as the reference.

### RESULTS

### Biochemical Assays Cytotoxicity

The cytotoxicity of the aqueous HP-pyruvate-containing solutions obtained by PHIP-SAH (solution I) and of single components was assessed. Cell viability was not affected by solution I after 1 and 6 h of treatment while a toxicity effect was observed after 24 h (**Figure 2A**). The solutions that contained traces of chloroform plus catalyst (solution II) and chloroform alone (solution III), showed a significant effect on the cell viability after 6 h of treatment (**Figures 2B,C**). As their toxicity appeared to be the same, one can draw the conclusion that it is mainly associated with the presence of the chloroform traces in the water phase. The hydrolysis side product allyl alcohol (solution IV) did not show any significant effect on the cell viability (**Figure 2D**). The presence of ethanol, the hydrogenation co-solvent, in the water solution, had a toxicity effect after 24 h treatment (**Figure 2E**).

### LDH Activity

The conventional biochemical assay used for the measurement of LDH activity showed that this enzyme was significantly more active in PC3 than in DU145 cells (**Figure 3A**), while in the LNCaP cells the activity of LDH was significantly lower than in both the other. This latter finding is in agreement with the fact that LNCaP cells are less glycolytic than the other two, more aggressive prostate tumor models. It has already been

FIGURE 2 | Results on the biocompatibility of the aqueous solutions obtained from the PHIP-SAH procedure. All cell lines were treated with: (A) an aqueous solution of pyruvate obtained from the PHIP-SAH procedure; (B) an aqueous solution containing traces of chloroform and catalyst; (C) an aqueous solution containing chloroform traces; (D) an aqueous solution of allyl alcohol and (E) the aqueous solution of the product of the PHIP-SAH procedure, to which ethanol was added. The cells viability was assessed by MTT assay. Data are the mean +SD from three independent experiments (*n* = 3). \**P* ≤ 0.05, \*\*\**P* ≤ 0.001, ns: not significant vs. untreated control (100% viable); one-way ANOVA.

shown that the metabolic phenotype of the LNCaP cells is dramatically different from that of PC3 and DU145 cells (20). The difference of LDH activity between PC3 and DU145 cells is also significant, being PC3 cells much more glycolytic than DU145 cells. This difference has been further investigated by means of the metabolomic measurements carried out in growth media and cells extracts.

### <sup>1</sup>H-NMR Metabolomic Measurements

From the <sup>1</sup>H-NMR spectra of the culture medium it was found that the production and accumulation of lactate in the extracellular space is slower in the less aggressive cells (LNCaP). If only the other two, more aggressive cell lines are considered, the lactate production is faster in DU145 than in PC3 cells (**Figure 3B**). Conversely, the intracellular lactate pool was larger in PC3 than in DU145 cells (**Figure 3C**).

Measurements of the glucose content in the extracellular medium, at different timepoints, showed that the rate of glucose consumption was faster in DU145 than in PC3 cells (**Figure 3D**). These results would appear to imply that glycolytic efficiency is higher in PC3 than in DU145 cells.

### <sup>13</sup>C-Hyperpolarization Experiments

When HP-[1-13C]pyruvate was added to the cells suspended in their growth medium and then kept in the NMR spectrometer at physiological temperature, lactate signal build-up was observed immediately in the <sup>13</sup>C-NMR spectra series. This was due to the rapid exchange of the <sup>13</sup>C hyperpolarized label between

pyruvate and lactate that occurs in the intracellular compartment (**Figure 4**). The lactate signal reached a maximum at about 20 s, and then decayed due to the T<sup>1</sup> relaxation processes. The <sup>13</sup>C signals that correspond to pyruvate and lactate were integrated and the time-dependent signal intensities of lactate and pyruvate were interpolated using a set of functions in order to obtain information about the kinetics of the metabolic process. The experimental results can be interpolated using a four-pool model that takes into account intra- and extra-cellular pyruvate and lactate, all mutually exchanging (21, 22). However, in the present case, the spectral resolution did not allow us to discriminate between the intra and extracellular metabolite signals, meaning that the two-compartments model was used in order to avoid errors that might be caused by over-parametrization (23).

The observed pyruvate and lactate peak intensities can be fitted to a simple two-sites exchange model

$$\text{Pyr} = \text{Lac}$$

according to which, the <sup>13</sup>C-NMR signals of [1-13C]pyruvate and [1-13C]lactate are described by the coupled differential equations (24)

$$\frac{d\mathcal{P}\mathcal{Y}r}{dt} = -k\_{\rm PL}\mathcal{P}\mathcal{Y}r\left(t\right) + k\_{\rm LP}La\left(t\right) - \frac{1}{T\_{1P}}\mathcal{P}\mathcal{Y}r\left(t\right) \tag{1}$$

$$\frac{dLac}{dt} = \ +k\_{PL}Pyr(t) - k\_{LP}Lac(t) - \frac{1}{T\_{1L}}Lac(t) \tag{2}$$

where kPL and kLP are the kinetic constants of the pyruvatetolactate conversion and the T<sup>1</sup> values are the decay time constants of the <sup>13</sup>C carbonyl sites. It has been shown that the model can be further simplified by setting the back-conversion rate (kLP) to zero (25). It follows that one deals with a two compartment model characterized by an unidirectional flow with the kinetic constant kPL being an apparent overall pyruvate to lactate conversion rate. The solutions of equations 1 and 2, given that Pyr(t = 0) = [Pyr]∗Z, where [Pyr] is the concentration of hyperpolarized pyruvate added to the test tube, Z is the enhancement factor, and Lac(t = 0) = 0, are

$$\mathcal{P}pr = \mathcal{P}pr\left(t\_0\right) \cdot \exp\left(-\left(k\_{PL} + \frac{1}{T\_{1P}}\right) \cdot t\right) \tag{3}$$

$$\begin{split} \mathcal{L}ac &= \frac{T\_1^{\mathcal{L}ac} \, k\_{PL} \, \mathcal{P}pr\left(t\_0\right)}{T\_1^{\mathcal{L}ac} \left(k\_{PL} + \frac{1}{T\_1^{\mathcal{V}}}\right) - 1} \left(\exp\left(-\frac{t}{T\_1^{\mathcal{L}ac}}\right) \\ &- \exp\left(-\left(k\_{PL} + \frac{1}{T\_1^{\mathcal{P}\mathcal{V}}}\right) t\right) \end{split} \tag{4}$$

We must also be consider that the small flip angle pulses (ϕ) that were used lead to further loss of polarization. In order to take into account this factor, an envelope function exp(−λt) was added to the observed signal intensity, where λ = l n(cosϕ) 1t and 1t is the time interval between successive pulses (26)

$$Pyr' = Pyr \, . \, \exp\left(-\lambda \, t\right) \tag{5}$$

$$Lac' = Lac \, . \, \exp\left(-\lambda t\right) \tag{6}$$

The kinetic constants obtained from the fittings were normalized to the number of cells and the pyruvate-to-lactate conversion rate

FIGURE 5 | Rate of pyruvate to lactate conversion obtained from the experiments carried out using HP [1-13C]pyruvate on cells, in different conditions: (A) intact cells (DU145, PC3 and LNCaP cells) cultured in their proper culture medium, (B) intact cells suspended in the medium with added lactate; (C) lysed cells and (D) cells cultured in the same medium (DMEM). \**P* ≤ 0.05, \*\**P* ≤ 0.01, \*\*\**P* ≤ 0.001; unpaired t-test.

was calculated as follows

$$\nu\_{PL} = \frac{k\_{PL} \left[ P \circ r \right]\_{t0}.\nu}{n \text{ cells}}$$

where v is the volume of the sample.

From the interpolation of the <sup>13</sup>C-NMR time-dependent signals LNCaP cells resulted to be less glycolytic than the other two (PC3 and DU145), in agreement with the fact that LNCaP cells have a more oxidative metabolic phenotype than the other two (20), while DU145 were significantly more glycolytic than PC3 cells (**Figure 5A**).

In the second set of experiments, lactate was added to the medium in which cells were suspended, during the 13C-NMR experiment, as reported by Day et al. (27). In this case, the pyruvate-to-lactate conversion rate become significantly faster in PC3 than in DU145 cells, while the apparent glycolytic efficiency of the LNCaP cells was still significantly lower than in the other two cell lines (**Figure 5B**).

The observed pyruvate-to-lactate exchange rate in intact cells is the result of concomitant processes, namely the rate of metabolite transport through the cellular membrane, and the efficiency of the LDH enzyme. In order to get more information about the differences in the metabolic phenotype of the two, more aggressive cell lines (PC3 and DU145), another series of experiments has been carried out on lysed cells. In these experiments, the contribution of MCT1 mediated transport of pyruvate to the extra to the intracellular compartment was removed, while lactate was added to the medium in order to mimic its intracellular concentration. In these experiments, the exchange kinetics became higher in both cell lines (PC3 and DU145) and were still slightly, even if not significantly, faster in DU145 than in PC3 cells (**Figure 5C**).

It must also be noticed that the difference between PC3 and DU145 cells was not significant when the cells were maintained in the same growth medium (DMEM) (20) (**Figure 5D**), while the growth rate of LNCap cells was exceedingly low.

### DISCUSSION

One of the main concerns with the application of parahydrogen hyperpolarized substrates in metabolic studies, both in cells and in vivo, is the presence of toxic impurities in the aqueous solution of the HP metabolite, traces of hydrogenation catalyst, organic solvent, and reaction side-products. The cytotoxicity experiments carried out in this work indicate that the aqueous solution of the HP substrate has moderate toxicity when cells are incubated with the product solution for 24 h. Interestingly, the toxicity appeared to be mainly caused by traces of organic solvent (chloroform) while the hydrolysis side-product allyl-alcohol did not affect cell viability after 24 h of incubation. Most importantly, experiments carried out using hyperpolarized substrates usually take place in a timeframe of a few minutes after the perfusion of the HP substrate. Therefore, on the basis of the herein reported results, it can be reasonably assumed that cellular metabolism is not affected by these chemicals during the time course of the experiment.

In this work, the experimental set-up and procedure for in-cell studies has been modified, compared to the previously reported protocol (18). The improved system for the addition of the hyperpolarized product through the cell suspension allowed us to use small size NMR tubes (5 mm NMR tubes) instead of the previously reported 10 mm NMR tubes. The use of a smaller volume is a significant improvement as the amount of cells necessary for each experiment was drastically reduced, from 20 to 8–10 M, and the spectral resolution was also considerably increased.

When cellular metabolism was interrogated using HP-[1- <sup>13</sup>C]pyruvate in intact cells, the pyruvate to lactate conversion rate was slower in LNCaP than in the other two cell lines. This is in agreement with the fact that the metabolic phenotype of LNCaP is significantly different from that of the other two cell lines and is also coherent with the lower LDH activity measured using the conventional biochemical assay. Differently, the pyruvate to lactate conversion rate of PC3 and DU145 cells, measured using hyperpolarized pyruvate, was contradictory with the LDH activity obtained from the biochemical assay applied to both cell lines. In fact, while in the first set of experiments, DU145 appeared more glycolytic than PC3 cells (**Figure 5A**), the second showed a higher LDH activity in the PC3 cells (**Figure 3A**).

In order to clarify this apparent contradiction, other series of experiments using HP-[1-13C]pyruvate have been carried out on intact cells suspended in their growth medium with added lactate and on lysed PC3 and DU145 cells.

When lactate was added to the extracellular medium of the intact cells, the apparent pyruvate-to-lactate exchange rate was significantly increased in all the cell lines and in particular in the PC3 cells. (**Figure 5B**).

Conversely, in the experiments carried out on lysed cells, there was not a significant difference between the <sup>13</sup>C label exchange kinetics of these two cell lines (**Figure 5D**).

In order to account for these experimental observations, two different hypotheses can be put forward, one relying on the oxidative lactate metabolism and the other based on the MCT4 mediated transport of lactate through the cellular membrane.

The first is based on the well-known fact that different isoforms of LDH exist, that are tetramers of two kinds of subunits (the M type and the H type), and have different affinity either for pyruvate or for lactate. These subunits are encoded by two similar genes (LDH-A and LDH-B, respectively). While LDH-A supports the ability of malignant cells to convert pyruvate into lactate, LDH-B is related to the oxidative use of lactate (28). Therefore, the faster <sup>13</sup>C label exchange between pyruvate and lactate observed in the PC3 cells, after the addition of lactate to the extracellular medium, might be due to higher LDH-B expression, that leads to the more efficient oxidation of lactate to pyruvate.

The biochemical assays seemed also to support this hypothesis. In fact the biochemical LDH assay measured the rate of lactate-to-pyruvate conversion, through the spectrophotometric observation of NADH formation, and the LDH activity was higher in PC3 than in DU145 cells. From the glucose measurements carried out on the extracellular medium it was also evident that glucose consumption and lactate production was more efficient in DU145 cells, which accumulate lactate in the extracellular compartment. On the other hand, PC3 cells showed smaller glycolytic efficiency and a larger amount of lactate in the intracellular space, which may imply a more efficient use of lactate rather than glucose as an oxidative substrate.

All of these observations seemed to point toward the view that the oxidation of lactate into pyruvate is more efficient in PC3 cells, while the LDH enzyme works preferentially as a pyruvate reductase in the DU145 cells. The oxidative lactate metabolism, associated with MCT-1 facilitated lactate uptake, is at the core of a metabolic adaptation of cancer cells called metabolic symbiosis (28).

The other possible explanation takes into account the effect of MCT-mediated transport on the pyruvate-to-lactate exchange rate, that has previously been investigated using hyperpolarized pyruvate in cells (25, 29, 30) and in vivo (31). MCT-1 has a broader distribution and has been associated with the uptake and efflux of pyruvate, L-lactate and others through the plasma membrane, while MCT-4 are mostly associated with the export of lactate in cells with high glycolytic rates (32, 33). The effect of the overexpression of MCT-4 in cancer cells on the pyruvateto-lactate exchange rate observed using hyperpolarized pyruvate had been investigated as well (23). A recent study on MCT4 carried out by Contreras-Baeza et al. (34) showed that MCT4 is a high-affinity lactate transporter with a somewhat lower affinity for pyruvate. This property confers MCT-4 expressing cells the ability to export lactate against high ambient lactate level, thus maintaining their lactate producing role, while MCT1 cells revert from lactate producers to consumers. A higher expression of MCT4 in DU145 cells than in PC3 may also explain the observation that the pyruvate-to-lactate exchange rate becomes more marked in the former cell line when the concentration of lactate in the extracellular medium is increased. It must also be noticed that the addition of lactate to the medium, in the experiments carried out on lysed cells (**Figure 5D**), did not lead to any significant difference in the exchange rate.

Although a thorough investigation of the expression of the different isoenzymes (LDHA and LDHB) and MCTs (MC1 ad MCT4) in the two, more aggressive prostate cancer cell lines (PC3 and DU145) would be needed, it can be concluded that the experiments reported herein have shown that the polarization obtained on [1-13C]pyruvate, as obtained using the PHIP-SAH methodology is more than sufficient for investigation of the differences in the metabolic phenotype of prostate cancer cells characterized by different aggressiveness (LNCaP, PC3, and DU145). These findings pave the way for a number of possible NMR investigations of cellular metabolism that go well-beyond the pyruvate/lactate transformation investigated in this work.

### DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

## AUTHOR CONTRIBUTIONS

The study was conceived and designed by EC and FR, implemented with the help of CC and OB. CC synthesized the substrates and optimized the preparation of the aqueous solution of the HP product. EC and FR carried out the analysis of the data from hyperpolarized experiments. GD and EC performed the biochemical assays and metabolomic measurements. SA provided conceptual advices on the study. FR, SA, and EC wrote the manuscript. All the authors contributed to the discussion of the results, revised, and approved the final version of the manuscript.

### FUNDING

EC was a recipient of a fellowship from the Fondazione Umberto Veronesi FUV (Post-doctoral Fellowships 2018). This project has received funding from the Compagnia di San Paolo (Athenaeum Research 2016, n. CSTO164550) and from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie Grant Agreement No. 766402.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2020.00497/full#supplementary-material

### REFERENCES


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

Copyright © 2020 Cavallari, Carrera, Di Matteo, Bondar, Aime and Reineri. 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.

# Understanding Metal Dynamics Between Cancer Cells and Macrophages: Competition or Synergism?

Marina Serra1,2, Amedeo Columbano<sup>1</sup> , Ummi Ammarah2,3, Massimiliano Mazzone2,3 \* and Alessio Menga2,3 \*

*<sup>1</sup> Department of Biomedical Sciences, University of Cagliari, Cagliari, Italy, <sup>2</sup> Laboratory of Tumor Inflammation and Angiogenesis, Center for Cancer Biology (CCB), VIB, Leuven, Belgium, <sup>3</sup> Department of Molecular Biotechnology and Health Sciences, Molecular Biotechnology Center – MBC, University of Torino, Turin, Italy*

#### Edited by:

*Daniel McVicar, National Cancer Institute (NCI), United States*

#### Reviewed by:

*Ana Patricia Da Silva Gomes, Cornell University, United States Cesare Indiveri, University of Calabria, Italy*

#### \*Correspondence:

*Massimiliano Mazzone massimiliano.mazzone@ kuleuven.vib.be; massimiliano.mazzone@unito.it Alessio Menga mengaalessio@gmail.com*

#### Specialty section:

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

Received: *13 December 2019* Accepted: *07 April 2020* Published: *30 April 2020*

#### Citation:

*Serra M, Columbano A, Ammarah U, Mazzone M and Menga A (2020) Understanding Metal Dynamics Between Cancer Cells and Macrophages: Competition or Synergism? Front. Oncol. 10:646. doi: 10.3389/fonc.2020.00646* Metal ions, such as selenium, copper, zinc, and iron are naturally present in the environment (air, drinking water, and food) and are vital for cellular functions at chemical, molecular, and biological levels. These trace elements are involved in various biochemical reactions by acting as cofactors for many enzymes and control important biological processes by binding to the receptors and transcription factors. Moreover, they are essential for the stabilization of the cellular structures and for the maintenance of genome stability. A body of preclinical and clinical evidence indicates that dysregulation of metal homeostasis, both at intracellular and tissue level, contributes to the pathogenesis of many different types of cancer. These trace minerals play a crucial role in preventing or accelerating neoplastic cell transformation and in modulating the inflammatory and pro-tumorigenic response in immune cells, such as macrophages, by controlling a plethora of metabolic reactions. In this context, macrophages and cancer cells interact in different manners and some of these interactions are modulated by availability of metals. The current review discusses the new findings and focuses on the involvement of these micronutrients in metabolic and cellular signaling mechanisms that influence macrophage functions, onset of cancer and its progression. An improved understanding of "metallic" cross-talk between macrophages and cancer cells may pave the way for innovative pharmaceutical or dietary interventions in order to restore the balance of these trace elements and also strengthen the chemotherapeutic treatment.

Keywords: cancer, iron, selenium, copper, zinc, metabolism, crosstalk, TAMs

### INTRODUCTION

Tumors occur as a result of the complex interaction between malignant, stromal, immune cells, and vascular system, as these different components communicate with each other via cell–cell contact-dependent mechanisms, soluble messengers and metabolites (1, 2). It is firmly established that the immune system can be reprogrammed by tumor cells to become ineffective, inactivated, or even acquire a tumor promoting phenotype (3). In this special tumor microenvironment the macrophages are particularly abundant and play an important role in tumor development by modulating inflammation, immune suppression, and angiogenesis (2). Many kinds of molecules including growth factors, inflammatory cytokines, chemokines, reactive oxygen, and nitrogen species (ROS and RNS, respectively) from tumorassociated macrophages (TAMs) are involved in the maintenance of a pro-tumorigenic microenvironment and in facilitating metastatic dissemination (3). Recent evidences have highlighted the metabolic signals as important mediators of macrophage function in the crosstalk between cancer and the immune system (4–6). In this metabolic context, cancer patients are characterized by a variety of perturbations in homeostasis of metal ions such as zinc, iron, selenium, and copper both at intratumoral and systemic level (7, 8). A large body of preclinical and clinical studies related to dietary deficiencies, indicates that this metal dysregulation triggers neoplastic transformation of cells and/or reduces anti-tumorigenic functions of immune cells by controlling a plethora of chemical and biological reactions (9). Selenium, copper, zinc, and iron are chemical elements of particular interest given their natural presence in the environment (air, drinking water, and food) and their capacity to stabilize cellular structures, to protect the genome stability, to control metabolic enzymes, receptors, transcription factors at very small concentrations (8, 10). The purpose of this review is to consider the contribution of these trace elements during neoplastic transformation and their involvement in tumorinduced immune evasion (7). Here, we will focus on how metal ions modulate TAMs functions in sustaining immunesuppressive environment that protects tumor cell growth or conversely, how the activity of cancer cells influences TAMs via metallic interplay. New pharmaceutical or dietary intervention strategies with the aim of restoring metal homeostasis, may in the future arise from an improved understanding of "metallic" crosstalk between macrophages and cancer cells.

### MAIN

### Zinc

Zinc is a vital mineral in many homeostatic mechanisms of the body (11). It activates metabolic enzymes, it is involved in carbonic acid and alcohol formation, it acts as a cofactor for some antioxidant enzymes, such as superoxide dismutase (SOD) and it is essential for the activity of transcription factors and/or proteins regulating gene transcription (9, 10, 12). It is also involved in the signaling pathways of proliferation, differentiation, apoptosis, and cell cycle regulation. Zinc is also crucial for the immune system, since its dyshomeostasis has an effect on proliferation, activation, and apoptosis of immune cells such as monocytes, natural killer-, T-, and B-cells (12, 13). Due to its ubiquitous presence, the immune-modulating properties and the potential ability to alter the function of various important proteins, zinc plays both a direct and an indirect role in the initiation and progression of cancer (14, 15). Moreover, zinc might enhance or decrease the signaling between immune cells and neoplastic cells, by altering membrane structure and receptor expressions (9). The role of zinc homeostasis in regulation of immune system and tumor progression is very complex, depending on its concentration, distribution as well as its temporal pattern (16, 17). Indeed, Zn appears to be protective in some conditions, whereas it is harmful in cases of environmental overexposure (8). Intake of dietary zinc is associated with a reduced risk of gastric, breast, esophageal, prostatic, and colorectal cancer (16), but at plasma concentrations not exceeding 30µM, in order to avoid immune-suppressive effects (9).

### Role in Cancer Cells

Many studies support the involvement of two families of metal transporters, namely ZnTs and ZIPs, in different types of cancers (17, 18). The ZnT (SLC30) family reduces cytoplasmic zinc concentrations whereas the ZIP (SLC39) family does the opposite function (19–21). Zn transporters are regulated by the status of zinc itself, hormones, growth factors, as well as cellular redox status (22). Their altered levels of expression or abnormal activities contribute to Zn dyshomeostasis in prostate, pancreatic, breast, and esophageal cancers (16, 17). Ambiguous changes in the expression levels of the zinc efflux transporters (ZnTs) have also been observed during tumorigenesis (21). On one hand, ZnT1 and ZnT2 expression increases in highly aggressive and metastatic basal breast cancer compared to low-invasive luminal, making the cells resistant to Zn toxicity (19, 23, 24). On the other hand, in different cases of more advanced prostatic cancer, ZnT1 and ZnT4 expression (in cytoplasmic vesicles, Golgi apparatus, and plasma membrane) decreases (23–25). Notably, ZnT transporters levels are very low also in pancreatic cancer compared to normal tissues (16, 25), while ZnT7 gene expression is up-regulated in esophageal cancer (17).

Compared to ZnT transporters, many more data are available regarding the association between zinc influx transporters (ZIP) and cancers (11, 19, 24, 25). ZIP1–4 is down-regulated in prostate cancer tissues resulting in low Zn levels in prostate gland (18). Zinc deficiency in prostatic cancer cells is responsible for an increased activity of mitochondrial aconitase and cytochrome c reductase, with consequent high citrate oxidation and respiration, as well as high rate of proliferation and invasiveness (26). In pancreatic cancer tissues all ZIP proteins with the exception of ZIP4 are downregulated, leading to low intracellular Zn concentrations, and increased resistance of the malignant cells to Zn cytotoxic effects (11, 13). In different breast cancer subtypes, zinc distribution, and zinc influx transporter levels show distinct profiles (16, 25). The luminal breast cancer, compared to the basal one, displays the upregulation of several ZIP proteins (ZIP 3, 5, 6, 10, 14) suggesting an increased need of cellular Zn uptake to meet the metabolic demand (25). Intracellular Zn homeostasis is tightly controlled not only by the regulation of the flux across the membranes, but also by buffering of free Zinc by metallothionein and its storage in subcellular organelles, such as vesicles (17, 24). Metallothioneins are small, cysteine-rich, metal-binding proteins which are responsible for maintaining metal homeostasis by acting as metallochaperones, metal donors, and acceptors for enzymes and transcription factors (22). In advanced prostate cancer the expression of metallothioneins, particularly MT1 and MT2, is lower compared to normal tissue and this is associated with increased risk of cancer relapse (21, 24). Conversely, the aggressive basal-like breast cancer exhibits higher levels of metallothioneins than luminal (ER+) and HER2 overexpressing tumors, in order to buffer cytoplasmic Zn and protect the malignant cells from Zn toxicity (21, 25). The behaviors of malignant cancer cells that might appear contradictory in terms of Zn management, within the same type of tumor, must be contextualized to the molecular phenotype of cancer, degree of invasiveness, metastatic potential, and response to therapy. For example, the high invasive basallike breast cancer tends to throw out and chelate zinc (21, 25), probably with the aim to preserve mitochondrial aconitase and cytochrome c reductase activities and to sustain high oxidative metabolism. Whereas the luminal-like is more inclined to zinc uptake (17), probably in order to avoid an uncontrolled oxidative damage through superoxide dismutase (SOD) activity. The complex interplay between zinc transporters/metallochaperones and zinc signaling is just beginning to be deciphered and requires further investigation. Despite accumulating evidences, whether the accumulation of intracellular zinc pools or the Zn secretion is a "driver" for carcinogenesis is still unclear.

### Role in Macrophages

The regulation of zinc homeostasis is particularly complex also in immune cells, in particular macrophages. Multiple ZnT/ZIP members are expressed in macrophages, indicating that these transporters have a very important role in physiological conditions (13, 27). Various functions, such as phagocytosis or the secretion of immune-mediating factors can be impaired by deregulation of zinc homeostasis, which ultimately leads to induction or exacerbation of various inflammatory and/or disease processes (22, 28, 29). The relationship between zinc and macrophage functions is very controversial and difficult to figure out. For example, while intracellular zinc levels are induced during early stage of macrophage differentiation whereby they enhance the adhesion process, zinc deficiency inhibits many functions including intracellular killing, cytokine production, and phagocytosis (30, 31). On the one hand, Zn depletion increases monocytes maturation into macrophages (12, 29, 32), on the other hand, it induces apoptosis in macrophages by p53-dependent mechanisms (31, 33). The relationship between zinc and oxidative burst after bacterial infection is also a matter of debate (22, 34). Indeed macrophages utilize two opposite strategies to kill phagocytosed pathogens, (i) by reducing the phagosome zinc content or (ii) by intoxicating them with excess amounts of this metal (12). The relationship between zinc and inflammatory signaling in monocytes/macrophages is still unclear. Chronic zinc deficiency activates the NLRP3 inflammasome and induces the secretion of IL-1β in macrophages, while a short term deficiency inhibits inflammatory activation (29). In addition, LPS treatment of human macrophages in zinc supplemented media increases ZIP8 expression and zinc uptake with consequent C/EBPβ inhibition and the subsequent increase in the pro-inflammatory cytokine response (35).

Zn homeostasis in pro- and anti-inflammatory conditions is also controlled by metallothioneins (MTs). These metal-binding proteins play a fundamental role in macrophage function and in cytokine signaling modulation (12, 22). In response to the pro-inflammatory or M1 cytokines, the macrophages increase Zn uptake by ZIP2 and Zn sequestration by MT1 and MT2, in order to yield a Zn-deficient environment and "steal" this essential element to the pathogen (34, 36). In M2 macrophages polarized with IL-4 or IL-13, MT3 is elevated and suppresses macrophage defenses (22, 36). MT3 renders Zn-pool labile and readily accessible to the pathogen, instead of sequestering it (34, 36). Overall, a lot is yet to be unveiled about the involvement of the metallothionein-Zn axis in immune processes. Indeed, the literature concerning the role of TAMs in maintaining of zinc homeostasis into tumor microenvironment is presently very limited. Ge et al. have highlighted that ZIP8 mediates Zn uptake and that different metallothioneins are induced in TAMs obtained from monocytes treated with melanoma-conditioned medium (30). They concluded that metallothioneins in TAMs sustain high levels of intracellular zinc protecting the cells from stress-induced apoptosis. Overall, the mechanism of how MTs and Zn transporters control TAMs functions in the tumor is limited and further investigation is required.

### The "Metallic" Cross-Talk Between Macrophages and Cancer Cells

Our understanding of the significance of ZnT, ZIP, and MTs expression within cancer cells and macrophages is still primitive. ZnT, ZIP, and MTs gene expression varies not only in different tumors but also within the tumor. Elevated zinc levels in tumor are characteristic of patients displaying breast, esophageal, lung, and gastrointestinal cancer (16, 17). Zn accumulation in these tumors is in agreement with increased expression of cellular Zn importing proteins compared to normal tissues, suggesting that this mechanism allows them to survive (17). Additionally, liver, kidney, and lung metastasis display higher zinc content in peritumoral tissue than the corresponding normal one or the tumor itself (13).

Zinc levels can be directly affected also by the tumor microenvironment. For example, pro-inflammatory mast cells release into cancer microenvironment granules of zinc affecting the cellular response and worsening the prognosis of most cancer patients (13). In this context, one could speculate that M2-like macrophages in the tumor microenvironment render Zn-pool labile and readily accessible to the cancer cells by MT3 and ZnT efflux transporters **(Figure 1)**. Unlike other cancer types, prostate, and skin tumors display lower zinc levels compared to normal tissues (13). Malignant prostate cells are deprived of the ability to accumulate zinc, due to the loss of ZIP1 expression and this is correlated with a metabolic transformation (26). In agreement with Zn "phobic" phenotype of skin tumor, TAMs obtained from monocytes treated with melanoma-conditioned medium, import zinc, and sustain high intracellular levels by upregulating ZIP8 and metallothioneins (30), thus contributing to protection of cancer cells **(Figure 1)**.

Although it is not applicable to any tumor type, it is possible to hypothesize that higher or lower Zn addiction might represent one of the mechanisms by which cancer cells apply a metabolic pressure on the TAMs, leading to immunosuppression, or conversely confer metabolic support to cancer cells.

Evidence of zinc crosstalk between cancer cells and macrophages could unveil a totally new scenario in which novel cellular targets for therapeutic intervention may emerge.

context, the M2-like macrophages in the tumor microenvironment could render Zn-pool labile and readily accessible to cancer cells by metallothionein MT3 and ZnT efflux transporters. Unlike other cancer types, prostate and skin tumors have zinc levels lower than normal tissues. Malignant prostate cells lose zinc importer protein ZIP1 and the ability to accumulate zinc, this in turn is associated with a metabolic rewiring, an increased activity of mitochondrial aconitase (ACO) and consequent high citrate oxidation and respiration. In skin tumor, TAMs import zinc and sustain high intracellular levels by upregulation of ZIP8 and metallothioneins, so contributing to protection of cancer cells from Zn toxicity. Higher Zn abduction might be inferred as one of the mechanisms through which TAMs sustain high oxidative metabolism of cancer cells. In parts the figures are based on speculations and have been prepared by assembling in-house built cellular metabolic pathway outlines with a modified and adapted version of BioRender images.

### Opportunities for Improvement of Cancer Therapy

Several studies suggest the correlation between zinc deficiency and cancer, and some of them support the necessity of zinc supplementation in preventing or treating tumors (9, 37). Yu and co-workers have demonstrated in a murine model of pancreatic cancer that its supplementation via zinc metallochaperones (ZMCs) is able to reactivate quickly and effectively zinc deficient mutants p53 and to recover their wild type transcriptional activities and pro-apoptotic mechanisms (38, 39). These pre-clinical studies might be translated to patients once p53 status of their tumors and zinc-deficient mutations are determined (38). Another way to replenish zinc is by the administration of zinc oxide (ZnO) nanoparticles or sulfate/gluconate formulations. Zn gluconate, used as an adjuvant therapy, has demonstrated its efficacy in stimulating the immune system and in improving the effects of chemotherapy against acute lymphocytic leukemia **(Table 1)** (44). Zinc sulfate, although at concentrations which exceed those observed in plasma, has revealed cytotoxic effects in colon cancer cells and tumorigenic esophageal epithelial cells **(Table 1)** (40, 41). Moreover, as well as zinc oxide (ZnO) nanoparticles, zinc sulfate induces a proinflammatory phenotype in a macrophage cell line and in peritoneal macrophages **(Table 1)** (31, 35), and this may pave the way for innovative TAM-specific agents able to switch the M2-like phenotype toward a tumorinhibiting M1-like phenotype. On the other hand, excessive zinc supplementation can generate side effects, such as high blood pressure (45). Before starting zinc-based therapy, it would be essential to profile zinc levels in patients and to contextualize them to the molecular phenotype of cancer, histological grading, metastatic potential etc. In luminal-like breast cancer context characterized by zinc requirement, a zinc-based therapy would be counterproductive since it would increase the aggressiveness of the tumor, whereas it would be useful a therapy with strong zinc-chelators. Hashemi and co-workers have demonstrated the cytotoxic power of the cell membrane permeable zinc chelator, N,N,N',N'-tetrakis(2 pyridylmethyl)ethylenediamine (TPEN) and the membrane



impermeable zinc chelator, diethylenetriaminepentacetic acid, (DTPA) against breast cancer cells (46).

### Iron

Iron (Fe) is an essential metal for mammalian cells, since Fe-S clusters are the basis of the catalytic activity of many enzymes necessary for heterochromatin stabilization, epigenetic modulations, mitochondrial respiration, TCA cycle etc (47). Iron exhibits a dual effect: on one hand it promotes cell proliferation and growth, on the other hand can induce oxidative damage to DNA, proteins, lipid membranes (i.e., ferroptosis) by production of free oxygen species (ROS) through Fe2+-O<sup>2</sup> reactions and Fenton chemistry (48). Due to iron ability to cause severe DNA strand breaks and modulate epigenome, its dyshomeostasis could be responsible for neoplastic transformation and aggressive tumor cell behavior (48, 49).

### Role in Cancer Cells

Iron homeostasis and cancer biology are tightly inter-connected, indeed the iron pool is necessary not only for early steps of tumor development, enhanced survival, and proliferation of neoplastic cells, but also for the promotion of metastatic cascade (47, 50). Here, iron is involved in remodeling the extracellular matrix and in the motility of cancer cells (50). Therefore, not surprisingly, elevated levels of Fe have been identified as a risk factor in cancer development and progression (47, 51). In this regard, the role of iron in cancer has been also highlighted by several in vivo models (47, 52). In particular, a low-iron diet has been shown to be effective in delaying tumor growth and increasing the survival of mice (53). Malignant tumors display the overexpression of many iron-related genes, and for this reason they compete with liver and spleen for Fe storage, leading to inadequate erythropoiesis and eventually anemia (54). The expression in cancer cells of genes, such as the transferrin receptor (TfR1), ferritin light chain (FTL), and the iron regulatory protein (IRP)-2, is associated with poor prognosis, a higher grade of tumor, and increased resistance to chemotherapy (55, 56). Tumor cells increase iron uptake through the upregulation of divalent metal transporter-1 (DMT1), transferrin/transferrin-receptor (Tf/TfR), and lipocalin-2/lipocalin-2receptor (Lcn-2/Lcn-2R) systems, and its storage by ferritin (FT) heavy chain (FTH) and FTL overexpression (48, 57). The increased iron level in cytosolic compartment supports cellular proliferation and survival functions via cyclinD1/CDK4 overexpression—p21 down regulation and via perturbations in the global histone and DNA methylation (49, 58). At the same time, cancer cells increase mitochondrial uptake of iron via mitoferrin-2 (Mfrn-2) and upregulate frataxin in order to sequester excess iron (that could lead to increased oxidative stress) and deliver it to Fe-S cluster assembly enzyme (ISCU), to allow for Fe-S cluster formation (58–60). To reduce the risk of iron overloading-dependent lipid peroxidation (that leads to non-apoptotic form of cell death known as ferroptosis) cancer cells rely on the selenoprotein glutathione dependent peroxidase 4 (GPX4) activity, which decreases intracellular radicals and protects mitochondrial metabolism from ROSinduced membrane damage (61, 62). Iron accumulation in cancer cells is also exacerbated by deregulation of iron exporter ferroportin1 (FPN1). In invasive tumor areas, FPN expression is lower compared to normal tissue and inversely correlated with patient survival and disease outcome (48, 63, 64). The expression of FPN is regulated by hepcidin, a protein linked to cancer driven inflammation which induces internalization and degradation of FPN upon its binding (48, 65). In cancer patients, elevated levels of hepcidin allow to control local tumor iron efflux by an autocrine/paracrine regulatory mechanism (48, 66). Given the complex network of iron regulatory genes in cancer cells a better understanding of their regulation and interplay is necessary.

### Role in Macrophages

Immune cells such as macrophages and T cells require iron to shape their phenotype and determine their responses (67, 68). Macrophages have a very important role to play in iron recycling from the RBCs. In spleen and liver, macrophages swallow up senescent RBCs and heme oxygenases (HO-1 and HO-2) catabolize the heme. The iron resulting from this process is then stored either in ferritin (FT) or exported via ferroportin (FPN) (69, 70). The FThigh and FPN1low pro-inflammatory macrophages display an iron sequestering phenotype characterized by iron withdrawal, restriction and storage (71, 72). Furthermore, these kind of macrophages enhance the uptake of iron-containing heme clusters and the expression of heme oxygenase 1 (HO-1) in order to recycle hemeiron and increase labile iron pool (LIP) (71, 72). It is worthy of interest that excess amounts of heme or iron in hemorrhagic tumor areas, caused by hemolytic red blood cells (RBCs), shift the pro-tumoral macrophage phenotype toward a pro-inflammatory and anti-tumoral one, which in turn exacerbates tissue damage (67). On the other hand, the FTlow FPN1high anti-inflammatory macrophages are predisposed to iron export and redistribution to the extracellular space, supporting the demand of surrounding cells (47, 71, 72). It has been widely demonstrated, both in vitro and in vivo, that anti-inflammatory macrophages TAMs adopt a strong iron-release phenotype that contributes to tumor cell proliferation and growth (57). In some cases, the inability of their FPN to export iron, due to the high levels of local hepcidin, is bypassed thanks to the increased expression of highaffinity iron-binding protein lipocalin-2 (Lcn-2) (47, 57). Since tumors demand an excess of iron, further investigations on TAM heterogeneity and iron plasticity are urgently needed.

### The "Metallic" Cross-Talk Between Macrophages and Cancer Cells

In the tumor microenvironment both tumor and immune cells compete for nutrients and metal elements such as iron (47, 48, 73). As mentioned before, iron plays also an important role in cancer development (48). Several evidences firmly established the concept of iron crosstalk between cancer cells and macrophages (47, 74). During early stages of carcinogenesis, pro-inflammatory cytokines and the exposure to hemolytic red blood cells (RBCs) shift the macrophages toward an iron loaded phenotype (67). Consequently, it is not surprising that cancer and macrophage cells compete for iron uptake in the tumor microenvironment. Later, the pro-tumoral/anti-inflammatory macrophages adopt an iron-release phenotype and donate iron to the tumor microenvironment to support cancer progression (75) **(Figure 2)**. Iron can be released via FPN and loaded onto circulating Tf for its uptake by cancer cells via the TfR. Alternatively, TAMs can rely on lipocalin-2 or ferritin release to transfer iron (47, 74, 76). To date, it is not known if iron removal from tumor microenvironment by iron-demanding cancer cells could be responsible for a shift toward a pro-tumoral and anti-inflammatory M2-like phenotype, as it happens in a renal inflammatory context (77). A better understanding of how iron controls crosstalk between macrophages and cancer cells requires further investigation.

### Opportunities for Improvement of Cancer Therapy

Considering the role of iron in regulating immune and cancer cells functions, therapies targeting iron metabolism are urgently needed. Cancer cells are iron influx dependent, and in line with this concept the application of iron chelators, dietetic iron depletion, and interference with the hepcidin pathway represents a first intervention strategy in vivo and in vitro (47, 78, 79). Various iron chelators able to inhibit cancer cell growth and modulate global histone and DNA methylation have been employed for iron overload disorders (49, 51). But to date, none has obtained approval for the cancer treatment, due to unfavorable pharmacokinetics and lack of selectivity (48). At the same time, several drugs and antibodies that interfere with hepcidin expression or activation have been developed with promising effects, but unfortunately, the lack of long-term follow-up studies in patients does not allow to predict their efficacy and safety (80–82). Moreover, some FPN stabilizers are being developed, in order to reactivate iron efflux from tumor cells (48, 81). However, since the pathways that regulates the hepcidin-FPN axis are complex, further studies are needed. Another emerging possibility is to target excess iron in tumor cells through induction of ferroptosis (48, 83, 84). In this regard, GSH depletion by erastin and inactivation of GPX4 activity by FDA approved alkylating antineoplastic compound altretamine (hexamethylmelamine) have shown their efficacy as ferroptosisinducer (61, 62, 85, 86). It is worthy of interest that ferroptotic secretome released from dying cancer cells is able to promote the recruitment of immune cells and support an M1-type immune microenvironment (87). To date, there is an increasing reliance on the use of micro/nanoparticles in cancer therapy. The treatment of tumor-bearing mice with iron microparticles has resulted in M1-like iron-loaded macrophages and net tumor suppression (67, 88). Another type of iron nanoparticle, the FDA approved ferumoxytol, has been shown to reduce the tumor growth and polarizing the macrophages toward M1 like phenotype in mammary, liver, and lung cancers (89). Additional in vivo studies and clinical trials are required for many of these compounds to elucidate their specific anticancer properties and their efficacy. Moreover, it would be useful to correlate iron levels in serum and tumors with the molecular phenotype of cancer, in order to choose the best therapy.

### Copper

Copper is an essential transition metal required for fundamental metabolic processes, but it can be toxic if present in excess (90, 91). As catalytic cofactor of many enzymes, it is involved in the mitochondrial electron transport chain (cytochrome c oxidase), in the detoxification of reactive oxygen species (superoxide dismutase 1 and 3), in the conversion of hydroperoxides into hydroxides (glutathione peroxidase), in melanin formation (tyrosinase), and in "ferroxidation" (ceruloplasmin) (91). Copper ions are also fundamental for proteins involved in cell signaling pathways, cell differentiation and death, and for enzymes involved in nervous system physiology. This metal ion plays a crucial role in the development and maintenance of immune function (29, 92). Indeed copper-deficient patients display decreased numbers of myeloid precursors in the bone marrow and susceptibility to infections (29, 93).

The recommended daily intake of copper in healthy adults is 0.9 mg/day (94).

A reduced intake of copper causes neutropenia, anemia, hypotonia, deterioration of the nervous system,

been prepared by assembling in-house built cellular metabolic pathway outlines with a modified and adapted version of BioRender images.

neurodegenerative disorders, and severe intellectual disabilities. Whereas the overload of copper, mainly in the liver, brain, and kidney, results in redox copper toxicity (e.g., liver cirrhosis) (91, 95). Various studies suggest a strong involvement of altered copper and cupro-proteins levels in cancer (96, 97). Copper has the ability to catalyze redox reactions and during its dysregulation reactive oxygen species are generated so excessively that act as precursors for neoplastic transformation and metastasis formation (91, 98). Many types of cancer (brain, multiple myeloma, acute lymphoblastic leukemia, lung, reticulum cell sarcoma, cervical, breast, and stomach cancer) show increased intratumoral levels and/or altered overall distribution of copper (97).

### Role in Cancer Cells

An analysis of the human copper proteome in 18 different tumor types has revealed several copper genes like CTR1, ATOX1, ATP7B, COX17 to be up-regulated (91). The reduced copper (including the dietary pool) is transported inside the cells via CTR1, a high affinity membrane copper transporter. The increased copper flow via CTR1 is followed by loading onto copper chaperone ATOX1, which acts as a copper-dependent transcription factor promoting the transcription of cyclin D1 and prompting cell replication (91, 99). Furthermore, copper binds to copper chaperones like COX17 and SCO2, which deliver it to mitochondria and to target proteins involved in trans Golgi network, including ATP7A, and ATP7B (100). Since copper is essential for the activity of cytochrome c oxidase (Cox), mitochondria rely on the phosphate carrier SLC25A3 for its uptake (101), and on labile copper pool in endoplasmic reticulum as additional source (91, 102). The mitochondrial phosphate carrier SLC25A3 has been associated with chronic myeloid leukemia progression and might play a role in copper imbalance (103). MEK1 being a copper-binding protein has led to the hypothesis that this metal ion is involved in the RAS-RAF-MEK-ERK pathway, required for cell proliferation, and tumorigenesis (104). Copper not only binds to proteins directly involved in cancer progression, but also indirectly modulates their expression or activation. Copper inhibits prolyl hydroxylase thus stabilizing HIF-1α and increasing the transcription of various angiogenic genes (e.g., ceruloplasmin and VEGF) (105) and genes involved in the epithelial to mesenchymal transition (e.g., LOX) (91, 106). The copper-dependent enzyme LOX catalyzes the cross-linking of collagen and elastin in the extracellular matrix (ECM) and interacts with MEMO1 (Mediator of cell Motility 1), another copper-dependent redox enzyme (107). MEMO1 is involved in cell migration through modulation of the cytoskeleton and formation of adhesion sites. Furthermore, copper ions activate the endothelial Nitric Oxide Synthetase (eNOS), thus increasing the production of the vasodilator nitric oxide (NO) (108). Other studies are required to unveil the mechanisms by which these proteins within the cell are loaded with copper. The dysregulation of these protein functions could be the priming for processes such as, creation of premetastatic niches, escape of immune defense, and angiogenesis. Understanding the mechanism of these genes and protein may open up exciting avenues for developing them as potential cancer therapeutic targets.

loading onto the copper chaperone ATOX1, which acts as a copper-dependent transcription factor promoting cyclin D1 expression and cell replication. Since copper is essential for the activity of proteins, like cytochrome c oxidase (Cox), involved in the mitochondrial electron transport chain, mitochondria rely on the phosphate carrier SLC25A3 (PTP) for its uptake. Copper not only binds to proteins directly involved in cancer progression, such as MEK1, but also indirectly modulates their expression or activation. Copper inhibits prolyl hydroxylase thus stabilizing HIF-1α and increasing the transcription of several angiogenic genes (e.g., ceruloplasmin and VEGF) and genes involved in the epithelial to mesenchymal transition (e.g., LOX). Copper is essential for sustaining the pro-inflammatory phenotype of macrophages; indeed, as a component of the SOD enzyme which catalyzes the production of H2O<sup>2</sup> from superoxide, it contributes to the ROS-dependent killing capacity of macrophages. The removal of copper from microenvironment by cancer cells might drive the polarization of TAMs toward a pro-tumoral M2-like phenotype. In parts the figures are based on speculations and have been prepared by assembling in-house built cellular metabolic pathway outlines with a modified and adapted version of BioRender images.

### Role in Macrophages

Copper is an essential element for immunomodulatory functions (29). As a component of the SOD enzyme, which catalyzes the production of H2O<sup>2</sup> from superoxide, it sustains the activity of neutrophils and monocytes, and regulates macrophage antimicrobial functions by contributing to ROS-dependent killing capacity (29, 109). Indeed its deficiency leads to a defective respiratory burst, impaired phagocytosis, and killing ability, with consequent susceptibility to recurrent pulmonary and urinary infections as well as septicaemia (29, 110, 111). Macrophages activated with proinflammatory cytokines (IFNγ and TNFα) and LPS show increased copper uptake via CTR1, increased copper accumulation within the phagosomes due to bactericidal Fenton reactions, and finally increased ceruloplasmin activity (112). The copper-containing ferroxidase ceruloplasmin promotes iron export via FPN, thus starving intracellular bacteria of this essential element (29, 113). Furthermore, M1-like macrophages display also an increased copper transport to the mitochondria via COX17 for energy production, to SOD1 for antioxidant defense or to Atp7a for protein synthesis (29, 112). The literature on the role of copper in modulating M2-like macrophages and/or in sustaining TAMs function into tumor microenvironment is absent.

### The "Metallic" Cross-Talk Between Macrophages and Cancer Cells

There are not evidences on the copper crosstalk between cancer cells and macrophages, thus in this context we can only speculate. Several studies suggest a strong copper addiction of cancer cells (114, 115), that probably deprives TAMs of this essential element. Since copper is essential for sustaining the pro-inflammatory phenotype of macrophages (29, 113), its removal from tumor microenvironment could be responsible for a shift toward a pro-tumoral M2-like phenotype and for an immunosuppressive environment **(Figure 3)**. Overall, our understanding of how copper controls TAMs-cancer cells interplay requires further investigation, with the aim to plan in the future a better dietary intervention or to find novel targets and innovative therapeutic agents.

### Opportunities to Improve Cancer Therapy

The strong connection between copper and tumor development, as well as metastization has encouraged scientists to design and synthesize new copper-complexing agents to be used in chemotherapy with lower side effects (79, 91). The copperbinding compounds used as anticancer agents are divided in two groups: copper chelators, which sequester copper ions from cells, and copper ionophores, which vehicle copper inside cells increasing its intracellular levels and priming cytotoxic effects through multiple pathways (116, 117). The copper complexing species tetrathiomolybdate (TTM), disulfiram, and clioquinol have been employed in clinical trials, but only TTM has given the most promising results (117). In the latest years, the fact that copper is a limiting factor for multiple phases of tumor progression, has led the scientists to the identification of plant based natural molecules with chelating properties, able to exert antitumoral effects or improve the efficacy of already known drugs, with low side effects (91, 97). These compounds in the presence of copper act as pro-oxidants and produce reactive oxygen species so excessively to induce DNA degradation (91, 118). The effects of copper, copper oxide nanoparticles, and copper chelate have been evaluated not only on cancer cells but also on macrophages (88, 119). Chatterjee et al., discovered a novel copper chelate, copper N-(2-hydroxy acetophenone) glycinate (CuNG), able to reprogram TAMs in a proinflammatory type which in turn converts Treg and Th2 cells in anti-tumorigenic Th1 cells (120–122). This compound triggers in TAMs ROS-mediated activation of MAPKs and ERK1/2 pathways which lead to upregulation of IL-12 and simultaneous downregulation of TGF-β and IL-10 production (121). We may speculate on a bivalent role of these redox-active compounds like CuNG in a clinical approach. The sustained generation of ROS on the one hand would induce apoptosis of cancer cells, on the other hand would trigger proimmunogenic macrophages.

### Selenium

The metal ion selenium (Se) plays important role in different biological processes which are mediated by almost 25 selenoproteins (123). As a cofactor for antioxidant enzymes, it exhibits anti-inflammatory properties and inhibits oxidative damage as well as DNA alterations (9). Moreover, selenium homeostasis supports the innate and adaptive immune functions; indeed its deficiency is associated with T cells and NK cells dysfunction and with a reduced number of lymphocytes in both thymus and bursa (9). Selenium is generally transported by selenoprotein P (SEPP1) and its mutations and/or haplo insufficiency increases genomic instability and risk of cancer (124, 125). Indeed, populations with low Se intake are exposed to higher risk of cancer development and its supplementation in suboptimal doses enhances immune responses to prevent cancer growth, reduce relapse, and cancer-specific mortality (9, 14). However, supra-nutritional doses do not confer protection against cancer and are associated with toxicity (123).

### Role in Cancer Cells

A recent study on SELENOP (SEPP1) has led to the identification of several single nucleotide polymorphisms (SNPs) which decrease the expression or function of this metal in various tumor types, including hepatocellular carcinomas, gastric adenocarcinomas, colorectal cancer, and prostate cancer (126– 128). SEPP1 is one of the few selenoproteins (SePs) able to incorporate selenium, be secreted into the plasma, be absorbed by the other tissues, and degraded to free selenium for synthesis of other SePs (129). SEPP1 loss induces an oxidative stress which, on one hand, can increase DNA damage and favor tumor initiation, on the other, can promote cancer cell cytotoxicity (127, 128, 130). However, SEPP1 expression is not universally down regulated in all tumor types. Indeed, SEPP1 upregulation has been observed in metastatic melanoma and poorly differentiated prostate cancer (128, 131). In cancer cells having high basal levels of oxidative stress, increased expression of SEPP1 can protect from cytotoxic effects and also lead to increased tumor development, proliferation, and resistance to chemotherapy (128). Selenium by lowering ROS production/accumulation not only prevents DNA oxidation but also activates mechanisms that stimulate mitochondrial biogenesis, preserve mitochondrial membrane potential, and sustain metabolic performance (132).

The glutathione peroxidases GPxs (1,2,3,4, and 6) constitute some of the most thoroughly studied SePs, because of their role in oxidative stress and their contribution to tumorigenesis (133). These proteins have antioxidant properties and catalyze hydroperoxide reduction by using glutathione (GSH) as a reductant. GPx1 expression is decreased in many tumor types and its overexpression, both in vitro and in vivo, has been found to reduce the growth of cancer cells and carry out a protective role (128, 134, 135). On the other hand, GPx1 expression has been linked to higher tumor number and growth rate, as well as to chemo/radio resistance (136). Like GPx1, GPx2 appears to have a pro-tumorigenic role in esophagus and liver, whereas it exhibits an anti-inflammatory role in colon context. Indeed, its deficiency has been linked to colitis-associated tumorigenesis (137, 138). Among the glutathione peroxidases, GPx3 is the only one clearly acting as a tumor suppressor. In tumor cells, GPx3 is often a target of hypermethylation and its downregulation is associated with bad prognosis and chemoresistance in several types of tumor (128, 139). Other selenoproteins with a critical role in maintaining redox balance and in controlling the multiple stages of tumor progression are the thioredoxin reductases (TrxRs). They are selenium responsive elements able to trigger antioxidant defense mechanisms in response to selenium supplementation (128, 140). Several in vitro and in vivo studies agree that TrxRs can inhibit tumor growth by extinguishing oxidative damage and DNA alterations, especially in the context of inflammatory-driven cancers. However, in tumor cells with higher basal levels of oxidative stress, these TrxRs can increase the resistance to apoptosis and even to chemotherapy (128, 141). Much work still needs to be done to characterize SePs in tumorigenesis context and to identify and understand the mechanisms by which they influence neoplastic transformation. The contradictory behavior of malignant cancer cells in terms of selenium management, needs to be deepened and contextualized to type of tissue, molecular phenotype, and degree of invasiveness, in order to determine the benefits or not of selenium supplementation. Selenium as regulator of cell redox balance can have different effects, depending on whether or not the tumor is inflammatory-driven.

### Role in Macrophages

A great body of evidence has extensively highlighted the role of selenium in the modulation of immune processes, particularly in macrophages (124). Studies on macrophagespecific knockout of selenocysteine (Sec) tRNA gene (Trsp), have demonstrated that selenoproteins drive their polarization from a pro-inflammatory toward an anti-inflammatory phenotype, which aids in the resolution of inflammation and wound healing (124, 142, 143). In particular, loss of Trsp leads to a decrease in M2 macrophage markers, a corresponding increase in M1 macrophage markers, an altered regulation in extracellular matrix-related gene expression and a diminished migration of macrophages in a protein gel matrix (124, 144, 145).

This phenotypic switch is combined with changes in cellular metabolism, particularly of arachidonic acid (146). Selenium in macrophages, by differential regulation of expression of mPGES1, TXAS, and H-PGDS, plays an important role in bioactive oxylipids synthesis, such as cyclopentenone prostaglandins (CyPGs) (145, 146). In presence of selenium, the arachidonic acid is metabolized to 15d-PGJ2, which negatively

affects pro-inflammatory signal transduction pathways (146). Vunta et al., demonstrated that selenium deficiency in mice exacerbates the LPS-mediated infiltration of macrophages into the lungs and also that selenium reintegration in macrophages leads to a significant decrease in LPS-induced expression of cyclooxygenase-2 (COX-2) and tumor necrosis factor-a (TNF-a) (146). Furthermore, other studies have associated the ability of selenoproteins to downregulate the expression of pro-inflammatory genes and polarize the macrophages toward an M2 phenotype with the inhibition of histone and non-histone acetylation, the activation of PPARγ and the degradation of pro-inflammatory PGE2 (145, 147). Experiments of gene expression have revealed that SELENOP (SEPP1) is one of the most upregulated genes in the M2 macrophage phenotype (128, 148). Moreover, Solinas et al., have found SELENOP (SEPP1) upregulated 95-fold at the transcript level in macrophages polarized by cancer cells conditioned media (149). Despite the lack of experimental evidence, it is possible to hypothesize that the increased SELENOP in M2 macrophages may offset the loss of SELENOP in cancer cells and support metastasis by supplying it in a paracrine manner (150). On the other hand, Barrett et al., have highlighted a shift toward M2 phenotype stimulated by IFN-γ and LPS (M1) or IL-13 (M2) in bone marrow derived macrophages isolated from Sepp1+/<sup>−</sup> mice (124). In agreement with these results, other studies have associated the selenium deficiency to the loss of GPxs and phagocytic activities of macrophages (M1 feature) toward transformed cells (133, 145, 150).

Development of mouse models lacking selenoproteins in macrophages has paved the way for understanding immune modulatory properties of these proteins (143, 144). However, the role of individual selenoproteins in this process is yet to be investigated properly. Based on the in vivo studies, selenium supplementation is essential to effectively resolve inflammation in most instances (145). Thus, it remains to understand if also selenocompounds may play a protective role.

### The "Metallic" Cross-Talk Between Macrophages and Cancer Cells

The role of selenium in the cross-talk between macrophages and cancer cells has been demonstrated only in leukemia disease. In a Se-deficient microenvironment TAMs produce mostly PGE2 and TXA2 from arachidonic acid via the COX pathway, supporting the highly glycolytic cancer stem cells (CSC) (145) **(Figure 4)**. Following Se-supplementation, selenoproteins affect the production of 112-PGJ2 in the M2 macrophage. 112- PGJ2 released by macrophages activates in cancer stem cells the tumor suppressor protein p53, which in turn upregulates the TCA cycle, oxidative phosphorylation, and lowers glucose uptake (145, 151, 152). As a compensatory mechanism, the antioxidant machinery is increased, although it is not sufficient to control ROS production and to avoid apoptosis (145) **(Figure 4)**. Also in this case, a better understanding of how selenium controls TAMs-cancer cells interplay will require further investigation. To date, without adequate experimental evidences, we may only speculate that the absence of selenium transporter SEPP1 in tumors and its increased expression in M2-like macrophages, tip selenium balance toward an immunosuppressive and pro-tumorigenic microenvironment. One may suggest that lower Se uptake by cancer cells is one of the mechanisms by which they drive the macrophage shift toward a proangiogenic, immunosuppressive, and protumoral function.

### Opportunities for Improvement of Cancer Therapy

Selenium supplementation is an attractive and achievable way to decrease cancer incidence, since selenium compounds are generally cheap and, at appropriate doses, safe (153, 154). Various studies have identified many classes of natural as well as synthetic organoselenium compounds which act as cytotoxic agents, and the research is ongoing for identifying more such compounds (154–157). Keeping in mind the immunomodulatory function of selenium, selenium nanoparticles (SeNPs) have been synthesized (158, 159). SeNPs have potential to decrease tumor cell proliferation, drive the anti-tumor function of TAMs, and in virtue of their properties, be used for imaging diagnosis and cancer therapy with low costs and negligible side effects (154, 158, 159). An impressive number of in vitro and in vivo studies clearly confirms the scarce toxicity of selenium compounds as monotherapy and in combination with classical chemotherapy (154). Furthermore, they also seem to increase the therapeutic potential of other drugs and reduce their side effects. However, to date the antiproliferative and proapoptotic properties of selenite, selenium amino acids, and other selenium compounds have not been confirmed by clinical trials (155, 156). Since supra-nutritional doses do not confer protection against cancer, on the contrary are associated with toxicity, before choosing a selenium-based therapy, it would be essential to profile serum and tumoral levels of metal ion, and to contextualize them to type of tissue, molecular phenotype, histological grading, metastatic potential, and chemosensitivity. In tumors characterized by high basal levels of oxidative stress, resistance to ROS-, and chemotherapy-mediated apoptosis, a selenium-based therapy would be counterproductive since it would increase tumor development and proliferation. More focused in vivo studies and additional clinical trials are necessary.

### REFERENCES

1. Dehne N, Mora J, Namgaladze D, Weigert A, Brune B. Cancer cell and macrophage cross-talk in the tumor microenvironment. Curr Opin Pharmacol. (2017) 35:12–9. doi: 10.1016/j.coph.2017.04.007

## CONCLUDING REMARKS

The effects of zinc, iron, selenium, and copper on cancer cells and TAMs (in supplementation or deficiency context) vary with concentration and tumor type. To sum up: heme iron intake and high serum levels of iron are associated with increased risk of breast and liver cancer (160); copper overload causes liver, lung, urinary, stomach, and cervical cancer; zinc poisoning or deficiency are associated with breast, lung, gastric, colon, and prostatic cancer; lower selenium intake increases liver, gastric, colon, and prostatic cancer incidence. In most cases these ions have been studied individually and their combined contribution to cancer progression has been totally overlooked or not well understood. In cancer growth and immune escape context, it is very important to consider also the relationship and balance between these metal ions inside the tumoral tissue. For example, a lower Zn/Fe ratio in the malignant prostatic tissue is correlated with poor prognosis and increased resistance to chemotherapy (161). In this case, zinc deficiency and iron overload combine their metabolic effects to increase citrate oxidation and mitochondrial activity in cancer cells and support their energy status. Also the Se/Zn balance plays an important role in onset of cancer. When the selenium is in excess compared to zinc, the metallothionein system is dysregulated, thereby producing p53 loss of function and DNA integrity reduction (162). Moreover, the results of some studies suggest that there is a close relationship between Cu and Fe in macrophages. Indeed highly toxic ferrous iron, as result of decreased ceruloplasmin expression/activity and copper deficiency, accumulates in macrophages leading to severe dysfunction (163). How the different ions contribute collectively to all steps of carcinogenesis and immune suppression remains to be described. The few observations made in co-culture systems and small animal models need to be amplified, extended to ionion interactions and carefully translated to the human setting. Wisely designed clinical trials are necessary to establish how the neoplastic cells influence TAMs functions or conversely, by controlling metal ions flux. A better understanding of the metal dynamics by which cancer remodels its microenvironment, may aid the discovery of innovative therapies able to more effectively kill tumor cells, or at least limit tumor progression and metastatic dissemination.

## AUTHOR CONTRIBUTIONS

This manuscript was conceived jointly by all authors. MS wrote the first draft of the manuscript. AM designed the paper and figures. MS, AC, UA, MM, and AM have revised and approved the final manuscript.


or prevent tumour progression. Converg Sci Phys Oncol. (2017) 3:034002. doi: 10.1088/2057-1739/aa7e86


implications in prostate cancer. Arch Biochem Biophys. (2016) 611:100–12. doi: 10.1016/j.abb.2016.04.014


tumorigenesis. Nature. (2014) 509:492–6. doi: 10.1038/nature 13180


enzymes by Nrf2. J Biol Chem. (2008) 283:2021–30. doi: 10.1074/jbc.M708 352200


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

The handling Editor declared a past co-authorship with one of the authors MM.

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

# Radiation Induced Metabolic Alterations Associate With Tumor Aggressiveness and Poor Outcome in Glioblastoma

Kshama Gupta<sup>1</sup> \*, Ivan Vuckovic<sup>2</sup> , Song Zhang<sup>2</sup> , Yuning Xiong<sup>1</sup> , Brett L. Carlson<sup>3</sup> , Joshua Jacobs <sup>1</sup> , Ian Olson<sup>1</sup> , Xuan-Mai Petterson<sup>2</sup> , Slobodan I. Macura2,4, Jann Sarkaria<sup>3</sup> and Terry C. Burns <sup>1</sup> \*

*<sup>1</sup> Department of Neurologic Surgery, Mayo Clinic, Rochester, MN, United States, <sup>2</sup> Metabolomics Core Mayo Clinic, Rochester, MN, United States, <sup>3</sup> Department of Radiation Oncology, Mayo Clinic, Rochester, MN, United States, <sup>4</sup> Department of Biochemistry and Molecular Biology, Mayo Clinic, Rochester, MN, United States*

#### Edited by:

*Paolo E. Porporato, University of Turin, Italy*

#### Reviewed by: *Cesar Cardenas,*

*Universidad Mayor, Chile Yasumasa Kato, Ohu University, Japan*

#### \*Correspondence: *Kshama Gupta gupta.kshama@mayo.edu Terry C. Burns burns.terry@mayo.edu*

#### Specialty section:

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

Received: *16 December 2019* Accepted: *25 March 2020* Published: *05 May 2020*

#### Citation:

*Gupta K, Vuckovic I, Zhang S, Xiong Y, Carlson BL, Jacobs J, Olson I, Petterson X-M, Macura SI, Sarkaria J and Burns TC (2020) Radiation Induced Metabolic Alterations Associate With Tumor Aggressiveness and Poor Outcome in Glioblastoma. Front. Oncol. 10:535. doi: 10.3389/fonc.2020.00535* Glioblastoma (GBM) is uniformly fatal with a 1-year median survival, despite best available treatment, including radiotherapy (RT). Impacts of prior RT on tumor recurrence are poorly understood but may increase tumor aggressiveness. Metabolic changes have been investigated in radiation-induced brain injury; however, the tumor-promoting effect following prior radiation is lacking. Since RT is vital to GBM management, we quantified tumor-promoting effects of prior RT on patient-derived intracranial GBM xenografts and characterized metabolic alterations associated with the protumorigenic microenvironment. Human xenografts (GBM143) were implanted into nude mice 24 hrs following 20 Gy cranial radiation vs. sham animals. Tumors in pre-radiated mice were more proliferative and more infiltrative, yielding faster mortality (*p* < 0.0001). Histologic evaluation of tumor associated macrophage/microglia (TAMs) revealed cells with a more fully activated ameboid morphology in pre-radiated animals. Microdialyzates from radiated brain at the margin of tumor infiltration contralateral to the site of implantation were analyzed by unsupervised liquid chromatography-mass spectrometry (LC-MS). In pre-radiated animals, metabolites known to be associated with tumor progression (i.e., modified nucleotides and polyols) were identified. Whole-tissue metabolomic analysis of pre-radiated brain microenvironment for metabolic alterations in a separate cohort of nude mice using <sup>1</sup>H-NMR revealed a significant decrease in levels of antioxidants (glutathione (GSH) and ascorbate (ASC)), NAD+, Tricarboxylic acid cycle (TCA) intermediates, and rise in energy carriers (ATP, GTP). GSH and ASC showed highest Variable Importance on Projection prediction (VIPpred) (1.65) in Orthogonal Partial least square Discriminant Analysis (OPLS-DA); Ascorbate catabolism was identified by GC-MS. To assess longevity of radiation effects, we compared survival with implantation occurring 2 months vs. 24 hrs following radiation, finding worse survival in animals implanted at 2 months. These radiation-induced alterations are consistent with a chronic disease-like microenvironment characterized by reduced levels of antioxidants and NAD+, and elevated extracellular ATP and GTP serving as chemoattractants, promoting cell motility and vesicular secretion with decreased levels of GSH and ASC exacerbating oxidative stress. Taken together, these data suggest IR induces tumor-permissive changes in the microenvironment with metabolomic alterations that may facilitate tumor aggressiveness with important implications for recurrent glioblastoma. Harnessing these metabolomic insights may provide opportunities to attenuate RT-associated aggressiveness of recurrent GBM.

Keywords: radiation therapy (RT), glioblastoma (GBM), recurrence, tumor microenvironment (TME), metabolomics

### INTRODUCTION

Glioblastoma multiforme (GBM; World Health Organization grade IV) is the most common adult primary brain malignancy (1, 2), accounting for 50% of all gliomas across all age groups (2). Standard treatment includes surgical resection, radiation therapy (RT), and chemotherapy; however, the overall 5-years survival rate is <10% with mortality approaching 100% (3, 4) is unfavorable prognosis may be due to the high propensity of tumor recurrence, with many recurring within 1 year, and 90% of these tumors forming within the prior RT field (5–7).

Radiation-induced changes in the brain and tumor microenvironment (TME) injury results in molecular, cellular, and functional changes that can facilitate tumor aggressiveness upon recurrence (8). Such changes include decreased vascularity, innate immune activation, and altered pharmacokinetics, pharmacodynamics, and therapeutic efficacy of chemotherapy agents (9–12). Additionally, irradiation (IR) generated reactive oxygen and nitrogen species (ROS/RNS) play havoc with cellular proteins, DNA, and phospholipid membrane (13). Mitochondria exposed to radiation produce increased ROS that may contribute to RT-induced cell senescence (14–16).

Tumor cell metabolism is strikingly different from that of normal cells with a shift from energy-producing pathways to those generating macromolecules necessary for proliferation and tumor growth. Through a tricarboxylic acid cycle (TCA), healthy cells metabolize glucose and produce carbon dioxide within an oxygen-rich environment, which efficiently produces a large quantity of adenosine triphosphate (ATP) (17). In hypoxic environments, these cells produce large quantities of lactic acid by anaerobic glycolysis. Conversely, in aerobic conditions, tumor cells rely on glycolysis for energy production (18), resulting in elevated rates of glucose uptake and increase lactate production (19). Lactate production during active tumor growth alters the tumor microenvironment by promoting acidosis, serving as a metabolic cancer cell fuel source, and inducing immunosuppression. RT may also have immunosuppressive effects leading to increased tumor aggressiveness, with associated increases in proliferation and infiltration (20), which may be exacerbated by prior RT.

Metabolic alterations may be pro-tumorigenic, promoting glioma initiation and progression (21–25). RT-induced metabolic changes in GBM depend on tumor volume, location, and dose-time regime of RT-administration, all of which can vary treatment response (8, 26–31). While differential metabolism of glioma tumor cells can be targeted for regression of tumor growth, understanding the impact of radiation-induced metabolic alterations in GBM microenvironment can provide new avenues to maximize long term benefits of RT in GBM care. The major objective of this study is to investigate the interactions between irradiation, tumor aggressiveness, and the associated metabolic changes in the TME. We here evaluate the tumor-promoting effects of prior RT on patientderived intracranial GBM xenograft in mice and characterize the metabolic alterations associated with the pro-tumorigenic stromal microenvironment.

### MATERIALS AND METHODS

### Ethics Statement on Mice

Six to 8 weeks old female heterozygous Hsd: Athymic Nude-Foxn1nu/Foxn1<sup>+</sup> mice were purchased from Envigo (Indianapolis, IN). Six to 8-weeks-old male C57BL/6J mice were purchased from Jackson Laboratories (Bar Harbor, ME). Mice were housed at the Mayo Clinic animal care facility, which is accredited by the Association for Assessment and Accreditation of Laboratory and Animal Care International (AAALACI). Aging was induced in two separate cohorts of C57BL/6J mice [fed with regular diet or high-fat diet (D12492, Research diets)] by keeping them in-housed for 24 months (24 mo) at the Mayo Clinic animal care facility, i.e., a small cohort of 5 mice, 2 months old was maintained for 22 months fed throughout on regular diet to obtain an aged mice group (24 mo), and, another cohort of 5 mice (2 months old) was fed on high-fat-diet (HFD) to induce obesity and continued on HFD for 22 months to obtain an aged-obese mice group (24 mo). All animal procedures were performed with proper animal handling, adhering to the National Institutes of Health (NIH) guidelines and protocols approved by

**Abbreviations:** RT, Radiation therapy; IR, Irradiation; GBM, Glioblastoma; PDX, Patient derived xenograft; IC, Intracranial; TME, Tumor microenvironment; IH, Ipsilateral hemisphere; CH, Contralateral hemisphere; TAM, Tumor associated macrophages; CBCT, Cone beam computed tomography; IF, Immunofluorescence; H&E, Hematoxylin and Eosin; <sup>1</sup>H-NMR, Proton–Nuclear Magnetic Resonance spectroscopy; GC-MS, Gas chromatography–mass spectrometry; LC-MS, Liquid chromatography–mass spectrometry; PCA, Principle Component Analysis; PLS-DA, Partial least squares–Discriminant Analysis; OPLS-DA, Orthogonal Partial least squares–Discriminant Analysis; VIP, Variable Importance on Projection; VIP-pred, Value of the VIP variant for the predictive components; VIP-total, Total sum of VIP values for both predictive and orthogonal components; ROS, Reactive oxygen species; eROS, Extracellular reactive oxygen species; RNS, Reactive nitrogen species; TCA, Tricarboxylic Acid cycle; NAD, Nicotinamide adenine dinucleotide; ATP, Adenosine triphosphate; GTP, Guanosine triphosphate; eATP/eGTP, Extracellular ATP/Extracellular GTP; GSH, Glutathione (reduced); ASC, Ascorbate; ThrO, Threonic acid; OxA, Oxalic acid; NAA, N-acetyl aspartate; Crn, Creatinine; Cr, Creatine; PC, Phosphocholine; Ob (Aged Ob), Obese (Aged Obese); Gy, Gray; hrs/mo, Hours/Months.

the Institutional Animal Care and Use Committee (IACUC) at Mayo Clinic, Rochester.

### Cranial Irradiation of Mice

Cranial irradiation was administered using the X-RAD SmART irradiator (Precision X-ray, North Branford, CT), which uses a cone beam computed tomography (CBCT) for accurate target localization. The stereotactic coordinates were determined from the target-set on CBCT using the first scan for each mouse within all groups (values ranged between x = 0.25 to 0.35, y = −3.8 to −4.0, and z = −5.8 to −5.95, depending on mice and straintype). Whole brain RT was performed as described (32), using parallel opposed lateral beams with 10 mm square collimator. Radiation treatments included 10 Gy or 20 Gy single dose (20 Gy) administration, or 4 Gy × 10 dose-fractionation. Control group mice were handled similarly as the treated, but with no radiation dose administered (0 Gy).

### Intracranial Injections in Mice

Intracranial (IC) injections in athymic nude mice were performed as previously described (33). Briefly, pre-established human GBM xenograft line, GBM143 cells were obtained from flank tumors and cultured in vitro in Dulbecco's Modified Eagle Medium (DMEM, GibcoTM 41966029) media having 10% Fetal bovine serum (FBS) and antibiotics (penicillin-streptomycin), for 3 weeks (34, 35). Representative images for GBM143 cell growth were acquired in transmitted light, using EVOS <sup>R</sup> FL Cell Imaging System, Thermo Fisher Scientific (**Figure 1B**, **Figure S1A**). These cells were dissociated using TryplE (Cat# 12563011, Thermo Scientific) and resuspended in PBS at a concentration of 100,000 cells per µl (with injection volume 3 µl/mouse). Mice were anesthetized using Ketamine: Xylazine mixture (100 mg/kg Ketamine and 10 mg/kg Xylazine) injected intraperitoneally (IP) with a 0.5cc syringe. The surgical procedure involved the following steps: disinfecting mice head with Betadine, lubricating the eyes with artificial tears, making a 1 cm midline incision extending from just behind the eyes to the level of the ears using sterile scalpel while applying pressure to have the incision open. Using a cotton swab, the skull was cleared to have the bregma exposed, a point 1 mm anterior and 2 mm lateral from bregma was identified and drilled through the skull using an 8bit Dremel drill. For stereotactic injection, Hamilton syringe with a 26G needle assembly was cleaned thoroughly, fixed on the injection jig, and 3 µl of cell suspension drawn into it. Injection jig was sterilized by wiping with STERIS Spor-Klenz and draping it with a sterile towel. The mouse having its skull drilled was placed on the jig and fixed using a front teeth hook at mouthpiece and ear pins. Using the stereotactic controls, the needle was inserted 3 mm deep into the brain and, cell suspension having 300,000 cells/3 µl was injected at a rate of 1 µL/min for over 3 min using the syringe pump. The needle was maintained as inserted in place inside the skull for additional 1 min, and then drawn out gently using the stereotactic controls. The hole drilled in mouse-skull at site of tumor cell implantation was sealed using bone cement, and the wound sutured with 4-0 vicryl with rb-1 needle (Ethicon J304H). Triple antibiotic was applied to the incision and stitches to prevent infection, and the mouse was left in the warm cage to recover from anesthesia. Water was supplemented with children's ibuprofen starting 24 hrs prior to the procedure and continued for 48 hrs post-surgery. The scheme of experiments involving IC injections is illustrated in **Figures 1A,B** and **Figure S1F**.

### Histology and Immunofluorescence

Athymic nude mice injected with the established PDX line, GBM143, were euthanized using isoflurane overdose at day of moribund (i.e., after 58 days of tumor cell-implantation). PBS cardiac perfusion was performed prior to termination under fully anesthetized conditions to remove the circulating peripheral leukocytes from the brain. Brains were extracted, fixed in 10% buffered formalin for 24 hrs, paraffin embedded, and 5µm coronal sections were obtained (slicing strategy explained in **Figure S1B**). All processing after fixation was performed at Mayo Clinic Histology core, Scottsdale. For histologic analysis, slides were stained with hematoxylin and eosin (H&E) and visualized by bright field microscopy at 4X microscopic magnification using Leica DMI-6000B [software: Leica Application Suite X (Leica Microsystems, Wetzlar, Germany)]. Percent positive H&E stained area was assessed as illustrated in **Figure S1C**, to estimate relative tumor burden between the samples.

H&E stained sections were reviewed to identify appropriate tumor bearing regions and respective unstained slides processed for immunofluorescence (IF) staining with human-Lamin A+C and Ki67 antibodies using standard procedure. Briefly, slides were deparaffinized in xylene and rehydrated by washing (3 min each) in serially diluted ethanol from 100, 95, 75, 50%, and then distilled H2O. Antigen retrieval was performed using prewarmed 9.8 mM Sodium citrate buffer (pH 6.0, with 0.05%Tween 20) for 30 min in hot steamer. Slides were rinsed in distilled H2O and PBS, blocked in blocking solution (10% Normal goat serum and 1% BSA in PBS), and stained with primary antibody (diluted in blocking solution, 1:300) overnight in humidified chamber at 4◦C. The slides were washed in PBS (3 × 5min), stained with secondary antibody (diluted in blocking solution, 1:300) for 2 hrs at room temperature, washed, and mounted with ProLong Gold reagent having DAPI (P36935, Life Technologies). Images were acquired at 4X microscopic magnification and tiling was done using Leica DMI-6000B (software: Leica Application Suite X).

### Image Analysis

All IF stained slides were quantified and scored for single cell count in a defined region with x-y coordinates approximated at tumor center (for h-Lamin A+C, and Ki67) or at center of corpus callosum (for h-Lamin A+C), respectively, using Image J (36, 37) and Cell Profiler 2.2.0 (Broad Institute of Harvard and MIT) (38). Briefly, for single cell counting, an IF image obtained was imported into Image J, threshold was set, channels split, and image in relevant single channel was selected and converted to black and white (BW). An area template having fixed size was generated to define a contained region at tumor or at the center of corpus callosum, respectively, maintaining consistency between different sample slides. This defined area selectively masked was overlaid and appropriately positioned in the BW

FIGURE 1 | (A) Scheme of experiments. (B) Representative image (at 10X, transmitted light microscopy) of GBM143 xenograft line cultured for 3 weeks *in vitro* in media indicated; cells were collected and orthotopically implanted into cranially irradiated mice 24 hrs post-irradiation (IR). (C) Representative Immunofluorescence (IF) images (at 4X, tiling) for hLamin A+C staining from 0 Gy and 20 Gy-IR mice coronal sections to assess tumor growth and invasion. (D) Top: Representative images (at 20X) show IF staining at tumor; and the dot-plot of single cell count for hLamin A+C and Ki67 staining. Bottom: Representative images (at 20X) show IF staining at center of corpus callosum, and dot-plot of single cell count for hLamin A+C. IH, ipsilateral hemisphere; CH, contralateral hemisphere; IR, irradiation. Statistical significance is represented as \**p* < 0.05.

image, and all background cells out of the masked region were eliminated. The resultant image was transferred to Cell Profiler 2.2.0 software (Broad Institute of Harvard and MIT) (38), the masked region was cropped and used as input; the pipeline for single cell counting was run to detect nuclei and quantify cells within this defined region.

To evaluate microglial activation, slides were stained for Iba-1 using standard procedure for IF. Images represented with 20X magnification were acquired on Leica DMI-6000B (software: Leica Application Suite X) and 40X magnification on Zeiss Axio Observer Z.1 (Software: Zen 2.3 SP1, Jena, Germany). Microglial morphology was assessed using ImageJ (36, 37). Antibodies used: Rabbit monoclonal Anti- h-Lamin A+C [EPR4100] (Cat# Ab108595, Abcam, Cambridge, United Kingdom); Rat monoclonal Ki67 (SolA15) (Cat #14-5698-82, eBioscience Invitrogen, Waltham, MA); Rabbit monoclonal Anti-Iba-1 (Cat# 019-19741, Wako). Secondary antibodies from Jackson ImmunoResearch Laboratories, Inc. (West Grove, PA) included polyclonal affinity-pure whole IgG: Cy3-Goat Anti-Rabbit IgG (H+L) (code: 111-165-003) and Cy5-Goat Anti-Rat IgG (H+L) (code: 112-175-143).

### Microdialysis

To evaluate changes in the extracellular milieu of radiated brain, a small group of mice (3 mice per group) from 0 Gy and 20 Gy single-dose irradiated mice cohorts injected with GBM143 24 hrs post-IR, were microdialyzed on their contralateral hemisphere (non-tumor bearing side) at day 30 from tumor cell injection (scheme of experiment in **Figure 1A**). The microdialysis set-up and surgical procedure was followed as described from the facility of Dr. Doo-Sup Choi, at Mayo Clinic, Rochester, Minnesota (39). Briefly, the mice were housed singly for 2 hrs in the microdialysis room to acclimatize, and then anesthetized using Ketamine:Xylazine mixture. Survival surgery was performed on a rotating platform with stereotactic guidance under sterile conditions. A microdialysis probe with a 2.0 mm cellulose membrane (Brain Microdialysis, CX-I Series, Eicom, Kyoto, Japan; MW cut off: 50,000 Da) was inserted at a point 1 mm anterior and 2 mm lateral from bregma on the contralateral hemisphere and secured to the guide cannula. The probe was connected to a microsyringe pump (Eicom, Kyoto, Japan), which delivered Ringer's solution (145 mM NaCl, 2.7 mM KCl, 1.2 mM CaCl2, 1.0 mM MgCl2, pH 7.4) at a 1.0 µl/min flow rate. The samples were collected in 0.2 ml collection tubes maintained at 4 ◦C for 3.5 hrs, and then immediately frozen and stored at −80◦C until analyzed.

### Metabolomics

To assess the radiation induced metabolic alterations in the pre-radiated mice brain, Proton Nuclear magnetic resonance spectroscopy (1H-NMR) and Gas Chromatography- Mass Spectrometry (GC-MS) based metabolomics was performed on whole tissue extracts obtained from non-tumor bearing brain samples of two independent strains of mice: Athymic nudes and C57BL/6. The experimental design with mice groups included for each strain is illustrated in **Figure S2A**.

#### Proton Nuclear Magnetic Resonance Spectroscopy ( <sup>1</sup>H-NMR)

Athymic nude mice, 0 Gy-control, and 20 Gy single-dose irradiated (10 mice per group) were sacrificed and immediately frozen in liquid nitrogen. Brian tissues were collected on dry ice and pulverized in liquid nitrogen. The pulverized mouse brain tissue (∼55–60 mg) was homogenized and extracted with 300 µl of ice-cold 0.6 M perchloric acid (HClO4) solution. Samples tubes were vortexed, centrifuged at 10,000 g for 10 min at 4◦C, and supernatants collected (40). The extraction procedure was repeated on the pellets (with ∼150 µL HClO4) and supernatant obtained from two rounds of extraction were combined and neutralized with 140 µl of 2M potassium bicarbonate (KHCO3). In 400 µL aliquot of neutralized extract, 100 µL of 0.1M phosphate buffer, and 50 µL of 1 mM TSP-d<sup>4</sup> in D2O were added. Samples were vortexed for 20 s and transferred to 5 mm NMR tubes. The NMR signal was acquired on Bruker AVANCE III 600 MHz instrument (Bruker, Billerica, USA). <sup>1</sup>H-NMR spectra were recorded using 1D NOESY pulse sequence with presaturation (noesygppr1d) under the following conditions: 90 degree pulse for excitation, acquisition time 3.90 s, and relaxation delay 5 s. All spectra were acquired with 256 scans at room temperature (298 K) with 64k data points and 8,417 Hz (14 ppm) spectral width. The recorded <sup>1</sup>H-NMR spectra were phase and baseline corrected using TopSpin 3.5 software (Bruker, Billerica, MA). The spectra were then processed using Chenomx NMR Suite 8.3 software (Chenomx Inc., Edmonton, Canada). The compounds were identified by comparing spectra to database Chenomx 600 MHz Version 10 (Chenomx Inc., Edmonton, Canada) and literature data (40–46). Quantification was based on an internal standard (TSP-d4) peak integral. The metabolite concentrations were exported as µM in NMR sample and recalculated as µmol/g of wet tissue.

### Gas Chromatography–Mass Spectrometry (GC-MS)

For GC-MS analysis, 70 µl neutralized brain extracts (∼6.4 mg of tissue wet weight) from athymic nudes were obtained using perchloric acid extraction method with 2M KHCO<sup>3</sup> based neutralization as described for <sup>1</sup>H-NMR, centrifuged at 10,000 g for 10 min, and cleared supernatant collected in fresh 1.5 ml eppendorf tubes. These samples were completely dried in a SpeedVac concentrator run overnight. They were subsequently methoximated using 10 µL MOXTM reagent (Cat# TS-45950, ThermoScientific, Waltham, MA) at 30◦C for 90 min and then derivatized using 40 µL of N-methyl-N-trimethylsilyl trifluoroacetamide with 1% trimethylchlorosilane (MSTFA+1% TMCS: Cat# TS48915, ThermoScientific, Waltham, MA) at 37◦C for 30 min. Metabolite levels were determined using GC-MS (Hewlett-Packard, HP 5980B) with DB5-MS column. GC-MS spectra were deconvoluted using AMDIS software (NIST, Gaithersburg, MD) and SpectConnect software (Georgia Tech, Atlanta, GA, USA) was used to create metabolite peaks matrix. The Agilent Fiehn GC/MS Metabolomics RTL Library (Agilent, Santa Clara, CA) was used for metabolite identification. Ion count peak area was used for analysis of the relative abundance of the metabolites (47).

Similar to above, whole brain extracts using perchloric acid method were also prepared from a cohort of C57BL/6 mice and evaluated by <sup>1</sup>H-NMR and GC-MS. C57BL/6 mice included in the study were divided into five groups (with 4–5 mice per group) as follows: control (0 Gy), 20 Gy single-dose irradiated, 4 Gy × 10 fractionation-dose irradiated, aged (24 mo), and aged-obese (24 mo) (scheme included in **Figure S2A**).

### Data Analysis

Multivariate analysis of NMR data was performed using SIMCA 15 software (Sartorius Stedim Biotech, Göttingen, Germany). Principal component analysis (PCA) was used to detect any innate trends and potential outliers within the data. Supervised Partial Least Squares discriminant analysis (PLS-DA) and Orthogonal-Partial least square–discriminate analysis OPLS-DA were performed to obtain additional information including differences in the metabolite composition of groups, variable importance on projection (VIP) values, and regression coefficients. OPLS-DA models were calculated with unit variance scaling and the results were visualized in the form of score plots to show the group clusters. The VIP values and regression coefficients were calculated to identify the most important molecular variables for the clustering of specific groups. Nonparametric Wilcoxon rank sum test and Student T-test were performed to determine the statistically significant differences between the groups.

### Survival Curves

Athymic nudes, grouped as control (non-irradiated, 0 Gy) and irradiated with 20 Gy single dose, were divided into two study cohorts: (1) Short-term IR: where 5 mice from each group were injected with GBM143 cells after short-term prior IR-exposure of 24 hrs, and (2) Long-term IR: where 5 mice from each group were maintained for 2 months post-irradiation and then injected with GBM143 cells. Survival time (in days) for each mouse was recorded until 70 days post tumor cell injection. The overall survival was calculated by Kaplan-Meier method and log-rank test was used to compare the survival curves (48). Experimental design illustrated in **Figure S1F**.

### Statistical Representation

The difference between specific metabolites or a parameter measured across two groups was estimated for p-value, as indicated. Graphs were plotted using software(s): GraphPad Prism 8.2.0 (GraphPad, San Diego, CA), Heatmapper (Wishart Research Group, University of Alberta and Genome Canada) (49) and Microsoft Office Excel. Statistical significance is represented as p-values: <sup>∗</sup>p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

### RESULTS

### Effect of Radiation on Tumor Growth, Proliferation and Migration

Mice were cranially irradiated with either 20 Gy (single dose) or 0 Gy (control), and tissues were collected at moribund to be evaluated with histology for tumor growth. A small cohort of mice radiated with 10 Gy (single dose) and injected with GBM143 line was also compared with the 0 Gy and 20 Gy cohorts for relative tumor burden using haematoxylin and eosin (H&E) staining. No difference in tumor size was observed between 0 Gy and 10 Gy; however, 20 Gy irradiated samples had significantly higher percent of section area positive for tumor, indicated by H&E (∼15% positive H&E for 0 Gy and 10 Gy, and 27% for 20 Gy, with p-value of 0.033 between 0 Gy and 20 Gy), indicating an overall faster rate of tumor growth (**Figure S1D**). Thus, 10 Gy cohort was not pursued for further evaluation. Sections from 0 Gy and 20 Gy were analyzed for tumor growth and proliferation using human-Lamin A+C and Ki67 staining. There was observable difference between tumor size of 0 Gy and 20 Gy radiated mice based on h-Lamin A+C staining at the tumor. Also, more cells positively stained for h-Lamin A+C were present at the corpus callosum of 20 Gy mice (**Figure 1C**). Quantitative analysis performed by counting both h-Lamin A+C and Ki67 within the tumor to evaluate proliferation revealed higher trend of both the stains in 20 Gy, with h-Lamin A+C. The h-lamin A+C was however not significant, due to high-density tumor region evaluated for both 0 Gy and 20 Gy; but showed near to significant difference in Ki67 positive cells stained in that area (p = 0.05), indicating higher proliferative potential in tumors that were obtained from 20 Gy-pre-irradiated mice brain. Similarly, h-Lamin A+C was assessed in the midline corpus callosum to evaluate cell migration toward the contralateral hemisphere, as illustrated in **Figure S1E**. The h-Lamin A+C staining in 20 Gy was significantly higher with a p-value of 0.03, compared to 0 Gy mice in the midline corpus callosum, suggesting a higher number of cells migrating toward the contralateral hemisphere (**Figure 1D**).

### Metabolomics Microdialysis

To assess for radiation-induced changes in the extracellular milieu, a pilot experiment with intracranial microdialysis (in the contralateral hemisphere) was performed in a small cohort of athymic nude mice (n = 3), involving groups 0 Gy and 20 Gy, at day 30 after GBM143 injection and microdialysates were analyzed for untargeted liquid metabolic profiling using LC-MS (method described in **Supplementary Materials 1**). Principal component analysis could separate the groups 0 and 20 Gy, indicating metabolic changes in effect of irradiation. A trend toward elevated levels of metabolites relevant to cancer progression was observed in the 20 Gy mice, including modified nucleotides (N6-methyladenosine, pseudouridine), polyol (myo-inositol, quebrachitol) detected in the 20 Gy (**Supplementary Excel Sheet 3**). However, there were very limited identifiable metabolites with a total of <60 due to low sample volume obtained after a 3.5 hrs microdialysis run at a rate of 1 ul/min (**Supplementary Excel Sheet 3**). Moreover, due to technical challenges involved with keeping ≥4 mice per group in microdialysis and the limited volume of microdialysates collected for evaluation, significant conclusions could not be made. We therefore utilized a whole tissue metabolomics approach in non-tumor bearing mice to evaluate the metabolic changes post-irradiation.

### Proton Nuclear Magnetic Resonance (1H-NMR) Spectroscopic Analysis

We sought to identify the radiation induced metabolic alterations in the brain stroma associated with the observed outcome of higher tumor growth and proliferation in 20 Gy mice. To achieve this, whole brain metabolomics was performed in two separate mouse strains, athymic nude mice and C57BL/6 mice, as described in methods. Athymic nude mice were included since the tumor study described above was performed with human-PDX line in athymic nudes; C57BL/6 mice were included to eliminate strain dependence and to avoid potential confounding effects of immunodeficient mice. A cohort of aged-C57BL/6 mice (24 mo) with and without diet-induced-obesity was analyzed to assess whether or not the radiation-induced metabolic changes in the brain were similar to those induced by aging or obesity-induced senescence. A small group of C57BL/6 mice were administered a fractionated dose of 4 Gy × 10 for comparative analysis.

Data <sup>1</sup>H-NMR spectroscopic analysis revealed clear separation of 0 and 20 Gy mice cohorts from athymic nude mice, using PCA (**Figure 2Ai**). Supervised OPLS-DA further separated the two groups based on metabolite composition differences with predicted-variable importance in the projection (VIP) values shown. The most important molecular variables for clustering of specific groups include glutathione (GSH) and ascorbate (ASC) having VIPpred 1.65, along with differences in ATP and GTP levels as potentially distinguishing characteristics (**Figure 2Ai**). After IR, a significant reduction of GSH, ASC, and NAD<sup>+</sup> levels were observed, along with increases in ATP and GTP. Additionally, an overall reduced trend in TCA intermediates was observed in 20 Gy (**Figure 2Aii**). The multivariate analysis of NMR data performed using SIMCA 15 software for C57BL/6 mice demonstrated separation of groups: Aged 24 mo, Aged-Obese 24 mo, Control (0 Gy), 20 Gy singledose cranially irradiated, and 4 Gy × 10 cranial IR-fractionated. Supervised PLS-DA showed separation of 0 Gy from irradiated mice groups, 20 Gy and 4 Gy × 10 (**Figure 2B**) and all five groups (**Figure S2B**). Specifically, the aged-groups (aged: 24 mo and aged-obese: 24 mo) were separated into a different component compared to the 0, 20, and 4 Gy × 10 groups (**Figure S2B**). There was a better separation of groups shown in model: M4 (aged, 0 and 20 Gy) as compared to those shown in model:M5 (0, 20, and 4 Gy × 10) (**Figure S2B**, **Table S1** for model parameters). Comparing all irradiated mice (IR group: 20 and 4 Gy × 10 analyzed together) with 0 Gy using PLS-DA and OPLS-DA showed significant group separation. The VIP-total and VIPpred value estimation indicated the metabolites most relevant to this group separation, which included GTP, ATP, GSH, and ASC (**Figures S2Bi,ii**).

The relative abundance of metabolites identified post-IR for 20 Gy single dose from <sup>1</sup>H-NMR for C57BL/6 mice showed reduction in GSH and ASC levels and an increase in ATP and GTP. No significant difference was observed between 20 and 4 Gy × 10 (**Figure 2B**). To assess how metabolomic profile of the radiated brain (at doses 20 Gy, and, 4 Gy × 10) was different from age-related brain metabolomic profile, C57BL/6 aging-mice cohorts (24 mo) were evaluated for significantly altered metabolites in comparison to irradiated and control mice. Alterations specific to the aged-group involved increased levels of scyllo-inositol and sn-glycero-3-phosphocholine with concomitant reduction in O-phosphocholine. Other metabolites reduced significantly in aged-mice were NAA (N-acetyl aspartate), neurotransmitters, and intermediates of TCA cycle (**Figure S3**) (50–54). List of metabolites detected for athymic nude mice and C57BL/6 by <sup>1</sup>H-NMR are included in **Supplementary Excel sheet 1**.

### Gas Chromatography-Mass Spectrometry (GC-MS)

Lysates processed for <sup>1</sup>H-NMR were further evaluated using GC-MS. The heatmap for relative abundance of metabolites (i.e., normalized total peak area of a metabolite per mice), between athymic nude mice, 0 Gy and 20 Gy, is illustrated in **Figure S2Ci**. While there was internal variation observed within these groups, only a few significantly altered metabolites in 20 Gy were identified, which included an increased trend in urea and a reduction in levels of creatinine (Crn), N-acetyl aspartate (NAA), and NAA/Crn ratio post-irradiation. Importantly, ascorbic acid was significantly reduced in 20 Gy and threonic acid was increased, reflecting ascorbic acid catabolism (**Figure 3A**). The heatmap for relative abundance of metabolites averaged for each group of C57BL6 mice is included in **Figure S2Cii**. The significantly altered metabolites between control (0 Gy) and irradiated groups (20 Gy and 40 Gy × 10) involved increased levels in urea but no change in Crn, NAA, and NAA/Crn ratio. However, there was significant reduction in levels of ascorbic acid with concomitant rise in threonic acid observed post-irradiation, alike observed for the athymic nudes (**Figure 3B**). Collectively, the results of <sup>1</sup>H-NMR and GC-MS indicate involvement of ROS clearance with active utilization of GSH and ASC as antioxidants. Scheme for ASC and GSH cycle in clearance of ROS and the role of GSH in regeneration of ASC is illustrated, along with intermediates of ascorbic acid catabolism, in **Figure 3C**. Expected metabolic alterations upon irradiation involve an increase in levels of ROS, utilization and reduction in GSH and ASC, with concomitant increase in by-products of ASC catabolism, threonic acid (ThrO), and Oxalic acid (OxA) (**Figure 3C**).

Other metabolites contributing to the separation of the groups in C57BL6 mice, and their relative assessment with aged mice groups are shown in **Figure S4**. The heatmap showed distinct metabolomic signatures for aging from that of irradiation (**Figure S2Cii**). At individual metabolite levels, no significant difference was observed for cholesterol in aged-Obese mice, which could be due to high internal variation observed within the group or small cohort size (5 mice/group). However, there was a reduced trend in free fatty acids and overall higher cholesterol, as compared to others. Notable metabolites separating the aged groups from the irradiated involved: increased age-related markers, scyllo-inositol and sn-glycero-phosphocholine, and reduction in fumaric, succinic acids, and metabolic intermediates of glycolysis and TCA cycle (52–54). Metabolic variations common to both aged and radiated mice cohorts included a rise in threonic acid, oxalic acid, D-allose, and myo-inositol. Additionally, there was a slightly higher trend in urea and Crn; however, this was not significant for either aged or irradiated

the metabolite composition of groups with predicted-variable importance in the projection values shown in graph below. (ii) The graphs show, significantly altered

*(Continued)*

FIGURE 2 | metabolites between the 0 Gy vs. 20 Gy. (B) Multivariate analysis of NMR data performed using SIMCA 15 software (Sartorius Stedim Biotech, Göttingen, German) for cohort of C57BL/6 mice, having groups as indicated. Supervised Partial Least Squares discriminant analysis (PLS-DA) performed shows, separation of all three groups. Bar graphs show metabolites most significantly altered between groups. Additional graphs for metabolic variants observed in C57BL/6 mice are included in Figure S3. Statistical significance is represented as \**p* < 0.05; \*\**p* < 0.01.

mice groups (**Figure S4**). List of metabolites detected by GC-MS are included in **Supplementary Excel sheet 2**.

### Immunostaining for Microglia With Iba-1

To evaluate the status of inflammation in the radiated brain and tumor microenvironment in response to RT, immunostaining for Iba-1 was performed for microglia in coronal slices from mice cranially irradiated (0 Gy or 20 Gy), and injected 24 hrs post-IR with GBM143 PDX line (**Figure 4**). Microglial morphology was assessed in ipsilateral (IH) and contralateral (CH) hemispheres Microglia were observed to be enlarged, bushy, and branched for 0 Gy-GBM143, as opposed to amoeboid for 20 Gy-GBM143, indicating stages of higher activation and higher phagocytic activity for the 20 Gy-GBM143 injected mice (**Figures 4A,B**). Comparing the microglial staining in ipsilateral hemispheres of 0 Gy and 20 Gy-GBM143 with that of the ipsilateral hemispheres of two separate mice that were cranially irradiated with 20 Gy but not injected with any human-GBM PDX line (radiation controls): showed, negligible Iba1<sup>+</sup> microglia staining in the brain slices of 20 Gy-IR alone, indicating, the observed microglial activation to be an effect of crosstalk between irradiation and tumor pathogenesis. **Figure 4D** illustrates their relevance in our experimental setting with maximum microglial activation and phagocytic activity observed in 20 Gy mice.

### Effect of Radiation on GBM Outcome

Effect of radiation-associated metabolic alteration on GBM outcome was assessed by, survival analysis for irradiated mice cohorts, ST-IR and LT-IR, and their respective control groups, injected with GBM143. Significant reduction was seen in the survival of mice after irradiation ST-IR or LT-IR (**Figure 5**). The combined graph of ST-IR and LT-IR further showed a significant difference is survival of 20 Gy (ST-IR) and 20 Gy (LT-IR) with median survival of 58 and 51 days, respectively.

Collectively, our data demonstrate radiation-induced metabolic alterations, including a rise in energy carriers (ATP and GTP) and reduction in antioxidants (GSH and ASC) associated with tumor promoting cell processes (cell proliferation, migration, and inflammation) and poor GBM outcome. The proposed model is illustrated in **Figure 5B**, and its translational significance is illustrated in **Figure S5**.

### DISCUSSION

Radiation therapy (RT) is an indispensable treatment modality for management of majority of cancers, and the standard of care for GBM. While, RT exerts its therapeutic potential by killing the proliferative tumor cells, RT can severely impact the TME by altering the extracellular milieu at molecular and structural levels (8, 26, 55). Radiation induced brain injury is widely documented; however, pro-migratory effects of RT on GBM cells have recently gained attention. Independent groups have reported enhanced human glioma cell migration and invasion in response to radiation dose treatment (56–59). We here evaluated the tumor growth and migratory potential of human-PDX line (GBM143) in the pre-radiated brain microenvironment with purpose to recapitulate the tumor recurrence scenario observed in clinic. As a measure of tumor growth and invasion, we quantified the proliferative cells at tumor and migratory cells crossing the center of corpus callosum and, observed higher cell migration and proliferation of GBM143 PDX line implanted in mice brain pre-radiated with 20 Gy indicating a tumor permissive microenvironment of the brain post-RT (**Figure 1D**) (8).

Radiation treatment leads to production of ROS, which facilitates tumor cell cytotoxicity in effect of RT. Tumor cells adapt to this oxidative stress through several mechanisms, including metabolic shifts and elevated antioxidant peptide production and intratumoral hypoxia generation (60–62). However, the RT-induced redox state of the non-transformed cells in the tumor stroma and how it may cross-interact with transformed tumor cells to impact tumor growth is less studied. Increased ROS levels in response to IR can be pro-tumorigenic (21, 22).

Metabolomics has emerged as the state-of-the-art approach to identify cancer cell fate (63–69); and metabolic therapy for management of GBM has been discussed (70, 71). We evaluated the metabolic changes in the pre-radiated brain microenvironment in response to 20 Gy-IR and the association with observed tumor aggressivity and inflammatory microglial phenotype. Cell proliferation and migration are a direct function of the cell's energy state (21); therefore, utilizing <sup>1</sup>H-NMR we quantified energy carriers in the radiated brain stroma and, found elevated levels of ATP and GTP post 20 Gy-IR with reduced levels of antioxidants, glutathione, and ascorbate (**Figure 2**). Ascorbate and GSH serve as the prime cellular antioxidants. Glutathione can recycle itself and reduced ascorbate (72, 73). Active ASC catabolism with decreased levels of ASC and GSH were observed, which indicate active ROS scavenging. While studies have also shown reduced intracellular redox signaling pathway in response to radiation, which may contribute to radiation induced oxidative stress (74), the depletion of ROS scavengers due to their increased demand would cause further accumulation of intracellular ROS, exacerbating oxidative stress. Chronically high levels of ROS in the TME can facilitate tumor growth (62, 75). Similarly, while ATP and GTP are essential components of cellular homeostasis, a rise in these intracellular nucleotides can cause their export out of the cell through extracellular vesicles, thus elevating their levels in extracellular space (76, 77). Extracellular purinergic nucleotides can affect both stroma and tumor cell processes. Extracellular ATP (eATP) has been implicated in facilitating microglial chemotaxis, inflammation, and several neurological or neuropathological

FIGURE 3 | clearance of reactive oxygen species (ROS). Intermediates of ascorbic acid catabolism are represented in orange boxes. Reactions in green show ASC-dependent peroxide metabolism; reactions in the central gray box show GSH-dependent regeneration of ASC; and reactions in red show GSH-dependent peroxide metabolism. Box on the right illustrates the expected metabolic alterations upon irradiation, which include increases in levels of ROS, and utilization of GSH and ASC, with concomitant increase in by-products of ASC catabolism, Threonic acid (ThrO), and Oxalic acid (OxA). Key to metabolic cycle illustrated: ASC, Ascorbate; MDHA, Monodehydroascorbate; MDHAR, Monodehydroascorbate reductase; APX, ASC peroxidase; GR, GSH reductase; GRX, Glutaredoxin; PRX, Peroxiredoxin; ThrO, L-threonic acid; OxA, oxalic acid; DHA, Dehydroascorbic acid; GSH, Glutathione reduced; GSSG, Glutathione, oxidized; NADP+, Nicotinamide adenine dinucleotide phosphate. Statistical significance is represented as \**p* < 0.05; \*\**p* < 0.01.

processes (78). Additionally, it can be internalized by tumor cells, increasing their intracellular ATP levels conferring metabolic reprogramming, increased tumor aggressivity, and treatment resistance (79–83). A recent lung cancer study has shown eATP to be involved in epithelial-to-mesenchymal transition, cell migration, and metastasis (84). While the biological functions of extracellular guanosine or eGTP are less studied than adenosine or eATP, their relative concentrations can co-vary, and biological functions of these nucleotides can cross-interact (85, 86). GTP is an essential biomolecule that modulates cell signaling via G-proteins and small GTP-binding proteins to facilitate cell proliferation, cell migration, and vesicle trafficking, and, can modulate metabolism and tumor development (87–94). Exocytosis and vesicle secretion can further facilitate release of purinergic nucleotides, inflammatory molecules, enzymes, and ROS into the extracellular milieu, which collectively can alter the TME to become pro-tumorigenic (75, 83, 95–97). These cause-effect relation between metabolic alterations and their cell physiological processes in radiated brain stroma, are illustrated in **Figure 5B**.

Microglia are the prime cells of immune surveillance in normal brain and one of the main the cellular components of tumor associated macrophages (TAMs) in the immune microenvironment of GBM (98–101). A persistent activation of microglia is the hallmark of a chronic neuroinflamation. Microglial activation and its M1 polarization state is characteristically exhibited in traumatic brain injury; however, the extent to which M1 vs. M2 polarization states relate to radiation-induced changes in microglia remains unclear (8, 32, 102, 103). Upon inflammatory trigger, microglial activation matures with sequential changes in its morphology, from resting ramified state to hyper-ramified, bushy and highly phagocytic ameboid state (101, 104, 105). A prolonged activation of microglia leads to a vicious circle, where secretion of proinflammatory cytokines and other neurotoxic agents (ROS and RNS) leads to further neuronal damage and cell death, which maintains microglial cells in their activated status (103, 106, 107). Extracellular ATP (eATP) can act as a chemoattractant and facilitates microglial activation and, intracellular ATP and GTP are involved in microglial mobility and secretory processes of inflammatory cytokines (101, 108–111).

While microglial activation is reported after irradiation in both juvenile and adult rodent brain (74, 112) intriguingly, we observed trivial Iba1<sup>+</sup> cells in the radiation control, indicating a possible clearance of the activated microglia over time, as the brain samples were harvested 58 days post-IR, at a time-point close to moribund for tumor-bearing mice groups. The observed microglial morphology and tumor-stromal cross talk in tumor bearing mice implies that radiation induced metabolic alterations in brain stroma along with progressive pro-inflammatory damage caused by tumor growth could lead to continued feed-forward recruitment and activation of microglial cells at the tumor. Since, maximum deleterious alterations and tissue damage would be expected within 20 Gy radiated-GBM143 tumors, maximal phagocytic activity of microglia was observed in these, indicated by their all amoeboid phenotype.

The dose and time-dependence of radiation exposure can significantly alter the impact of RT on TME by affecting tumor or stromal cell behavior, migration, and treatment response (27, 29, 30, 113–120). The association between cancer, aging and therapyassociated aging is well documented (50, 51). High-dose IR effects include hemorrhage, cognitive decline, neurodegeneration, and premature senescence, which can progress over time (13, 15).

Multivariate analysis of <sup>1</sup>H-NMR data and heatmap of GC-MS data revealed a clear distinction between aged-mice groups from 20 Gy-single dose (**Figures S2B,Cii**). The metabolic changes observed in aged and irradiated-mice differed markedly in relative abundance of most of the metabolites assessed by <sup>1</sup>H-NMR and GC-MS (**Figures 3**, **4**). Increased levels of urea and decreases in NAA and creatine (Cr) or Crn levels have been observed in neuropathologies (116, 121). We observed a slight increase in urea with radiation in both mouse strains, but NAA and Crn levels were not consistent and demonstrated a decline only observed in athymic nude mice. These indicate a partial neurotoxic state induced by 20 Gy-IR; with no severe aginglike signatures in 20 Gy. This could in part be due to the timedependence of the experiment, where mice brain samples were harvested for metabolic analysis 24 hrs post-RT to mimic the time frame in which tumor implantations were performed post IR.

The association between the metabolic effects and time since radiation was investigated by performing a survival analysis. Shortest median survival in the LT-IR cohort indicates progressive IR-induced damage in tumor stroma, making it more permissive for tumor growth and recurrence. This corresponds to progressive radiation-induced brain injury and increased susceptibility to neuropathologies observed in patients treated with RT (122). Clinical correlation of the study is depicted in **Figure S5**.

### CONCLUSIONS

We identified an aggressive tumor behavior and microglial activation following 20 Gy single dose brain radiation, which could become more severe with time. Moreover, we found metabolic alterations with a rise in energy carriers (ATP and GTP) and a decline in antioxidants ASC and

microglial morphology. (C) the microglial staining in ipsilateral hemispheres of 0 Gy-GBM143, and 2 Gy-GBM143 compared with that of ipsilateral hemispheres of two separate mice cranially irradiated with 20 Gy-single dose; however, not injected with any human-GBM PDX line. (D) (i) Site of GBM143 injection at IH of mice having received ± cranial irradiation (ii) Stages of microglial activation observed in experimental setting.

*(Continued)*

FIGURE 5 | outcome. The box marked with yellow outline (irradiated by RT) shows metabolic changes in the radiated brain stromal microenvironment, with rise in energy carriers ATP and GTP, and reduction in levels of antioxidants, ascorbate and glutathione, and the cellular processes affected by them. With IR, continued and excess rise in levels of energy carriers and expense of antioxidants within stromal cells of the brain can lead to altered extracellular milieu. Pathophysiological changes in the extracellular milieu, which can be immediate or long term caused by the radiation are enlisted in purple box. These alterations would collectively contribute to radiated stroma and GBM cell interactions that are permissive to GBM growth and, aggressive recurrence. Translational relevance of the study and its insights gained from pre-radiated brain microenvironment to prevent secondary and recurrent GBM spread is illustrated in Figure S5.

GSH to associate with the observed tumor phenotype. Independent groups have reported metabolic alterations in GBM cells to be pro-tumorigenic (21–25). We show for the first time a comprehensive view over the metabolomic alterations in the pre-radiated brain administered with high-dose IR (equivalent to late effects of hypo-fractionated dose), in vivo, that associate with tumor proliferation, migration, and inflammatory phenotype. These observations suggest an unprecedented role of the pre-radiated brain microenvironment on aggressive GBM recurrence, with, sustained and progressive metabolic stresses to worsen GBM outcome.

### FUTURE DIRECTION

The role of antioxidants in compromising the therapeutic effect of RT and pro-oxidants in sensitization to RT has long been debated (123–134). Radiation therapy mediates its effects directly or indirectly by production of ROS; thereby, causing oxidative damage to macromolecules and induction of apoptosis. Therefore, increased expression of antioxidant peptides in tumors have been thought to reduce the cytotoxic effects of RT, and GSH inhibition is proposed to have a therapeutic advantage in sensitizing cells to RT (73, 135). Ascorbate can act as a pro-oxidant in acidic microenvironments, such as tumors (136); thus, it may function as a radio-sensitizer for GBM cells and a radioprotector for normal cells post-RT (137, 138). While discrepancies remain regarding ASC's role as a radio-sensitizer or radio-protector in GBM, its potential as an anticancer agent has been reviewed (139–143).

Our study demonstrates an immediate effect of prior exposure to high-dose irradiation in the non-tumor/untransformed brain cells as a decrease in antioxidant levels, including GSH and ASC, consistent with their utilization to neutralize RT-induced free radicals. The depletion of these antioxidants can lead to further acute or chronic oxidative stress, altering the brain TME, which may contribute to the enhanced aggressiveness of recurrent tumors. While radiation-induced oxidative stress is necessary for DNA damage in tumor cells, this study raises the question if GSH and ASC administration after completion of radiation or primary treatment regime could help mitigate the radiation-induced metabolic stress in the microenvironment. If the post-radiation redox state contributes to tumor aggressiveness, there may be an opportunity to attenuate the RT-associated aggressiveness of recurrent GBM, enhancing the long-term safety of brain radiation treatment for glioblastoma (translational relevance illustrated in **Figure S5**).

## DATA AVAILABILITY STATEMENT

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

### ETHICS STATEMENT

The animal study was reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) at Mayo Clinic, Rochester.

### AUTHOR CONTRIBUTIONS

KG and TB led the project, contributed to experimental design, review, and discussion. TB supervised and supported KG. KG, YX, and BC carried-out mice tumor experiments. KG conducted survival studies and performed immunostainings. BC supervised KG on irradiator operation. IO assisted KG. KG and JJ collaborated to analyze images. SM and IV conceived <sup>1</sup>H-NMR protocol. KG, IV, and SZ performed <sup>1</sup>H-NMR and GC-MS studies, and data analysis. Metabolomics core provided support with LC-MS and data analysis. All authors contributed to experiments and research execution. Figures provided by KG and IV. Illustrations created by KG. All authors contributed to manuscript writing, research, editing, and final review.

### FUNDING

Funding support (TCB) was provided by NIH K12 NRDCP, NINDS NS19770-01, Mayo Clinic Cancer Center, Brains Together for a Cure, the Mayo Clinic Grand Forks Career Development Program and Regenerative Medicine Minnesota. Additionally, this work was supported by the Mayo Clinic Metabolomics Resource Core grant (U24DK100469) and the Mayo Clinic Metabolomics Resource Core NMR developmental funds. The authors acknowledge the editing and research assistance of Superior Medical Experts.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2020.00535/full#supplementary-material

Figure S1 | Tumor growth assessed post-moribund for cranially irradiated mice (A–E): (A) GBM143 PDX line obtained from flank tumor, cultured *in vitro* for 3 weeks. Images acquired in three independent fields (F1–F3), using transmitted light microscopy (at 10X) indicate morphology of cells to be branched, neuroglia-like, interspersed with enlarged polygonal cells. (B) Scheme for the slicing strategy: Mice brain was sectioned into four equidistant pieces (∼1.8 mm apart); Slices were made from each in the order of being 5µm thick coronal slices from the cerebral hemisphere only, Rostral to caudal for 22 slides, so as to cover a depth of 120µm from each of the four tissue pieces. These slices were arranged onto the glass slides, such that each slice on a slide is obtained from one of the respective four brain pieces, sectioned equidistantly. Two slides (1 and 22) were stained with H&E and evaluated for tumor growth. Tumor positive area was detected in slices obtained from two out of four sectioned pieces for most of the mice brain samples. Percent positive H and E staining was assessed for each. (C) Illustration showing arrangement of the slices on a glass slide, and evaluation of percent positive H&E. (D) Relative H&E staining as observed for slices obtained from 0 Gy, 10 Gy, and 20 Gy. Dot-plot for the overall tumor burden estimated in these groups. (E) Scheme illustrating steps involved in performing single cell count: mice brain coronal sections are stained for h-Lamin A+C –Cy3 (and Ki67–Cy5), for both 0 Gy and 20 Gy. A defined region is selected and masked (area-squared in white). This masked area-image in single channels is imported into cell profiler software and cropped. This cropped image is used as the input image, pipeline for nuclei detection run, and single cell count obtained. Similar steps are performed for a defined region selected at center of corpus callosum for h-LaminA+C staining (images in box, on right). (F) Effects of radiation induced alterations on GBM outcome: Scheme of experiment for survival analysis in athymic nude mice groups, 0 Gy and 20 Gy irradiated.

Figure S2 | (A) Scheme for experiment involving Proton–Nuclear Magnetic Resonance spectroscopy (1H-NMR) and Gas chromatography–mass spectrometry (GC-MS). (B) <sup>1</sup>H-NMR: Multivariate analysis for C57BL/6 mice, having groups as indicated. Supervised Orthogonal Partial Least Square-Discriminate Analysis (OPLS-DA) to show further separation of 0 Gy, with irradiated group, irradiation (IR) (20 and 4 Gy × 10); (i) Total variable importance in the projection (VIP) values (ii) Predicted VIP values. Parameters involved in group separation using multivariate analysis in M1–M7 models are listed in the Table S1. (C) Heatmaps for GC-MS data: (i) Heatmap for relative abundance of metabolites (i.e., normalized total peak area of metabolites for all mice within each group) between athymic nude mice groups, 0 Gy and 20 Gy. (ii) Heatmap for relative abundance of metabolites averaged for each group (i.e., normalized total peak

### REFERENCES


area for metabolites, averaged for all mice within each group), between C57BL/6 mice grouped indicated.

Figure S3 | Proton–Nuclear Magnetic resonance spectroscopy (1H-NMR): The graphs show, relative abundance of metabolites between C57BL/6 mice groups. Group Comparison: Aged (24 mo), Aged-Obese (24 mo) verses control (0 Gy), & radiated (20 Gy, 4 Gy × 10). The significantly altered metabolites are categorized as per their molecular type or biological pathway involvement. Statistical significance is represented as <sup>∗</sup>*p* < 0.05; ∗∗*p* < 0.01; ∗∗∗*p* < 0.001, ∗∗∗∗*p* < 0.0001.

Figure S4 | Gas chromatography–mass spectrometry (GC-MS): Group Comparison: Aged (24 mo), Aged-Obese (24 mo) verses control (0 Gy), and radiated (20 Gy, 4 Gy × 10). The graphs show, relative abundance of metabolites between C57BL/6 mice groups. The significantly altered metabolites are categorized as per their molecular type or biological pathway involvement. Statistical significance is represented as <sup>∗</sup>*p* < 0.05; ∗∗*p* < 0.01; ∗∗∗*p* < 0.001, ∗∗∗∗*p* < 0.0001.

Figure S5 | Translational Neuro-Oncology: The model illustrates sequential alterations that may contribute to tumor recurrence post-primary treatment regime. Standard of care for glioblastoma multiforme (GBM) involves tumor resection, radiation therapy (RT) and chemotherapy (Temozolomide, TMZ). Residual tumor cells or glioblastoma stem cells (after primary treatment regime) have the ability to migrate away from initial site, if their surrounding microenvironment becomes liberal for it. Radiation induced alterations in brain parenchyma and its extracellular microenvironment (or tumor stromal compartment), include metabolic changes such as reduced antioxidants, increase in energy carriers, neuroinflamation, and others. These changes can dramatically remodel the pre-radiated brain stroma, making it permissive for tumor cells to re-grow and migrate to distant sites forming new foci; thereby, causing tumor recurrence and spread. Future therapeutic interventions to prevent secondary tumor growth may harness these insights to leverage the potential of radiation therapy, and better treatment with use of cell proliferation and migration inhibitors and metabolic modulators to advance GBM care.

Table S1 | Model parameters used for multivariate analysis of <sup>1</sup>H-NMR data.


143. Alexander MS, Wilkes JG, Schroeder SR, Buettner GR, Wagner BA, Du J, et al. Pharmacologic ascorbate reduces radiation-induced normal tissue toxicity and enhances tumor radiosensitization in pancreatic cancer. Cancer Res. (2018) 78:6838–51. doi: 10.1158/0008-5472.CAN-18-1680

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

Copyright © 2020 Gupta, Vuckovic, Zhang, Xiong, Carlson, Jacobs, Olson, Petterson, Macura, Sarkaria and Burns. 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.

# TGFβ Signaling Increases Net Acid Extrusion, Proliferation and Invasion in Panc-1 Pancreatic Cancer Cells: SMAD4 Dependence and Link to Merlin/NF2 Signaling

Raj R. Malinda, Katrine Zeeberg, Patricia C. Sharku, Mette Q. Ludwig, Lotte B. Pedersen, Søren T. Christensen and Stine F. Pedersen\*

*Section for Cell Biology and Physiology, Department of Biology, Faculty of Science, University of Copenhagen, Copenhagen, Denmark*

#### Edited by:

*Daniel McVicar, National Cancer Institute (NCI), United States*

#### Reviewed by:

*Yasumasa Kato, Ohu University, Japan Cinzia Antognelli, University of Perugia, Italy*

> \*Correspondence: *Stine F. Pedersen sfpedersen@bio.ku.dk*

#### Specialty section:

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

Received: *01 December 2019* Accepted: *14 April 2020* Published: *07 May 2020*

#### Citation:

*Malinda RR, Zeeberg K, Sharku PC, Ludwig MQ, Pedersen LB, Christensen ST and Pedersen SF (2020) TGF*β *Signaling Increases Net Acid Extrusion, Proliferation and Invasion in Panc-1 Pancreatic Cancer Cells: SMAD4 Dependence and Link to Merlin/NF2 Signaling. Front. Oncol. 10:687. doi: 10.3389/fonc.2020.00687* Pancreatic ductal adenocarcinoma (PDAC) is a major cause of cancer-related death, with a 5-year survival of <10% and severely limited treatment options. PDAC hallmarks include profound metabolic acid production and aggressive local proliferation and invasiveness. This phenotype is supported by upregulated net acid extrusion and epithelial-to-mesenchymal transition (EMT), the latter typically induced by aberrant transforming growth factor-β (TGFβ) signaling. It is, however, unknown whether TGFβ-induced EMT and upregulation of acid extrusion are causally related. Here, we show that mRNA and protein expression of the net acid extruding transporters Na+/H<sup>+</sup> exchanger 1 (NHE1, SLC9A1) and Na+, HCO<sup>−</sup> 3 cotransporter 1 (NBCn1, SLC4A7) are increased in a panel of human PDAC cell lines compared to immortalized human pancreatic ductal epithelial (HPDE) cells. Treatment of Panc-1 cells (which express SMAD4, required for canonical TGFβ signaling) with TGFβ-1 for 48 h elicited classical EMT with down- and upregulation of epithelial and mesenchymal markers, respectively, in a manner inhibited by SMAD4 knockdown. Accordingly, less pronounced EMT was induced in BxPC-3 cells, which do not express SMAD4. TGFβ-1 treatment elicited a SMAD4-dependent increase in NHE1 expression, and a smaller, SMAD4-independent increase in NBCn1 in Panc-1 cells. Consistent with this, TGFβ-1 treatment led to elevated intracellular pH and increased net acid extrusion capacity in Panc-1 cells, but not in BxPC-3 cells, in an NHE1-dependent manner. Proliferation was increased in Panc-1 cells and decreased in BxPC-3 cells, upon TGFβ-1 treatment, and this, as well as EMT *per se*, was unaffected by NHE1- or NBCn1 inhibition. TGFβ-1-induced EMT was associated with a 4-fold increase in Panc-1 cell invasiveness, which further increased ∼10-fold upon knockdown of the tumor suppressor Merlin (Neurofibromatosis type 2). Knockdown of NHE1 or NBCn1 abolished Merlin-induced invasiveness, but not that induced by TGFβ-1 alone. In conclusion, NHE1 and NBCn1 expression and NHE-dependent acid extrusion are upregulated during TGFβ-1-induced EMT of Panc-1 cells. NHE1 upregulation is SMAD4-dependent, and SMAD4-deficient BxPC-3 cells show no change in pH<sup>i</sup> regulation. NHE1 and NBCn1 are not required for EMT *per se* or EMT-associated proliferation changes, but are essential for the potentiation of invasiveness induced by Merlin knockdown.

Keywords: NHE1, SLC9A1, NBCn1, SLC4A7, Merlin, proliferation, invasion, PDAC

### INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) is one of the most devastating cancers globally (1). The exceedingly poor prognosis for PDAC patients reflects a combination of late detection, rapid local invasiveness, and a severe lack of reliable biomarkers and efficacious treatment schemes (2). PDAC is associated with extensive metabolic changes, and PDAC tumors are accordingly highly acidic (3). While PDAC genotypes are highly complex, the most widely characterized driver mutations are activating KRAS mutations, inactivating p53 tumor suppressor mutations, and inactivation or loss of the cyclin-dependent kinase inhibitor 2A (CDKN2A, P16INK4) and the transforming growth factor β (TGF-β) effector, SMAD4 (4, 5).

TGFβ signaling involves the binding of a TGFβ dimer (TGFβ-1,−2, or−3, of which TGFβ-1 is most ubiquitous) to the TGFβ receptor types I and II (TGFβRI and –II; the former also known as ALK5). This results in formation of a heterotetrameric receptor complex, where TGFβRII phosphorylates and activates TGFβRI. TGFβRI in turn phosphorylates the transcription factors SMAD2/3, which bind to the co-SMAD, SMAD4, to form a hetero-trimeric protein complex that enters the nucleus to control gene expression. This complex may further interact with a variety of other transcription factors, which are necessary cofactors for SMAD-dependent gene regulation (6, 7). TGFβ ligands also signal through SMAD-independent pathways, including mitogen-activated protein kinases, small GTPases, and the phosphatidyl-inositol-3-kinase (PI3K)-AKT-mTOR pathway (6, 7).

In non-cancer epithelial cells and in premalignant cells, TGFβ signaling is consistently cytostatic, blocking cell cycle progression by increased expression of cyclin-dependent kinase (CDK) inhibitors. However, in many cancer cells, this is overridden by strong CDK activation by other pathways, causing TGFβ to be pro-tumorigenic (6). Accordingly, TGFβ signaling has been shown to stimulate cell motility, invasion, and proliferation, and limit antitumor immune response, and TGFβRI inhibition can revert these effects (8–10). Both pro- and antitumorigenic, highly genotype-dependent roles of TGFβ signaling were demonstrated in PDAC cells (4, 11–13). Illustrating the importance of TGFβ signaling in this cancer, a recent study showed that almost 50% of PDAC patient tumors exhibited mutations in TGF-β signaling components. While SMAD4 inactivating mutations are most common, mutations in SMAD3, TGFβ receptor type I (TGFBR1) and−2 (TFGBR2) are also reported (4).

TGFβ signaling is a major driver of epithelial-to-mesenchymal transition (EMT), a process with key roles in metastasis and chemotherapy resistance (6, 8, 11, 14–16). In PDAC, TGFβinduced EMT has been reported to involve SMAD4-dependent (17) and -independent (18) signaling, however, the process is incompletely understood.

Solid tumors are characterized by an often profoundly acidified extracellular pH (pHe), a neutral or slightly increased intracellular pH (pHi), and a greatly increased rate of acid extrusion (19, 20). The latter occurs because the acid generated by the high, predominantly glycolytic, metabolism of tumor cells is actively extruded from the cancer cells by specific transporters. These transporters, including the Na+/H<sup>+</sup> exchanger NHE1 (SLC9A1) and the Na+, HCO<sup>−</sup> 3 cotransporters NBCn1 (SLC4A7) and NBCe2 (SLC4A5) confer additional advantages to the cancer cells, including stimulation of proliferation, survival, and invasiveness, leading to increased tumor growth and metastasis (21–24). In particular NHE1 is important for cell motility and invasiveness, which are key downstream events in EMT (25). Directly implying a link to TGFβ, NHE1 is implicated in fibronectin release in a manner rescued by TGFβ-1 (26).

We therefore hypothesized that net acid extruding proteins are regulated by TGFβ signaling in human PDAC cells and contribute to its downstream effects. We here show that TGFβ-1 induced EMT of Panc-1 cells is associated with increased protein levels of NHE1 and NBCn1 as well as increased pH<sup>i</sup> , whereas smaller changes were observed in SMAD4-deficient BxPC-3 cells, which show only a very modest EMT. This difference between the two cell lines is corroborated in the opposite effects of TGFβ-1 on proliferation, which is increased in Panc-1 and decreased in BxPC-3 cells. Furthermore, knockdown of the tumor suppressor Merlin potentiates TGFβ-1-induced Panc-1 cell invasiveness in a manner dependent on both NHE1 and NBCn1. We propose that acid-extruding transporters are novel players in TGFβ-1-induced EMT in PDAC cells.

### MATERIALS AND METHODS

### Antibodies and Reagents

Primary antibodies used in western blot analysis were: mouse anti-β-actin, mouse anti-α-tubulin, and mouse anti-α-smooth muscle actin (α-SMA), all from Sigma-Aldrich; goat polyclonal anti-CTGF and mouse anti-NHE1 (clone 54), anti-Poly-ADP Ribose Polymerase (PARP), anti-cleaved PARP (Asp214) and anti-pSer807/811-Rb, all from Santa Cruz Biotechnology; mouse anti-dynactin 1 (DCTN1) and mouse anti-E-cadherin, from BD Biosciences; rabbit anti-GAPDH, rabbit anti-Histone 3, rabbit anti-Merlin, mouse anti-p53, mouse anti-Ki67, and rabbit anti-β-catenin, all from Cell Signaling. Mouse anti- Ki-67 was from Dako (Glostrup, Denmark) and rabbit polyclonal anti-NBCn1 was a kind gift from Jeppe Prætorius, Aarhus University, Denmark. Secondary antibodies used in western blotting were horseradish-peroxidase-conjugated goat polyclonal anti-mouse or anti-rabbit and rabbit polyclonal anti-goat from Dako. Secondary antibodies for immunofluorescence analysis were AlexaFluor488-conjugated donkey anti-mouse or antirabbit, and AlexaFluor568-conjugted donkey anti-mouse or anti-rabbit, all from Invitrogen. Recombinant human TGFβ-1 (PHG9214) was from Life Technologies. Mouse monoclonal antibody against Proliferating cell nuclear antigen (PCNA) was from Cell Signaling Technology.

### Cell Culture

Human PDAC cell lines MIAPaCa-2, Panc-1, BxPC-3, and AsPC-1 were acquired from American Type Culture Collection (ATCC, Rockville, MD, USA) and maintained in RPMI medium 1640+GlutaMAXTM-I or Dulbecco's Modified Eagle's medium (DMEM)+GlutaMAXTM-I (both from Gibco) supplemented with 10% (v/v) fetal bovine serum and 100 U/ml penicillin and 100µg/mL streptomycin at 37◦C in a humidified atmosphere of 5% CO2. MIAPaCa-2 cell medium was further supplemented with 2.5% (v/v) heat-inactivated horse serum. Immortalized human pancreatic ductal epithelial (HPDE H6c7) cells were a kind gift from Dr. Ming-Sound Tsao at Ontario Cancer Institute, Toronto, Canada (27, 28) and were cultured in kerantinocyte basal medium supplemented with epidermal growth factor and bovine pituitary extract.

### siRNA-Mediated Knockdown

siRNAs employed were: NHE1 siRNA: ON-TARGET SMART pool (Thermo Scientific); NBCn1 siRNA (SASI\_Hs01\_00030755, Sigma-Aldrich) sense sequence 5′ -CAUUAACUGGGAUUG CCUA-3′ , Merlin siRNA (SASI\_Hs\_01\_00188862, Sigma) sense sequence 5′ -CCUCAAAGCUUCGUGUUAA-3; SMAD4 (EHU018671 esiRNA siRNA mixture, Sigma). A 19-bp scrambled oligomer (sense: 5′ -AGGUAGUGUAAUCGC CUUG-3′ ) (Eurofins MWG Operon, Ebersberg, Germany) was used for mock transfection. Cells were seeded to ∼40% confluency in the relevant culture dishes and transfected with NHE1 siRNA (100 nM), NBCn1 siRNA (25 nM), Merlin siRNA (50 nM) or mock siRNA (50 nM), using Lipofectamine (Invitrogen) transfection reagent, according to the manufacturer's specifications. 48 h after transfection, cells were serum starved, and 24 h later, exposed to TGFβ treatment (or corresponding control conditions) for another 48 h before analysis for the relevant experiment as indicated below.

### RT-qPCR Analysis

Total RNA was isolated using NucleoSpin <sup>R</sup> RNA II (Macherey-Nagel, Germany) according to the manufacturer's instructions, reverse-transcribed using Superscript III Reverse Transcriptase (Invitrogen, Carlsbad, CA) and cDNA transcripts were amplified by qPCR using the SYBR Green technique (Applied Biosystems, Cheshire, UK). Amplification was performed in triplicate in an ABI7900 qPCR machine, using 40 cycles of (95◦C for 30 s, 60◦C for 1 min, 72◦C for 30 s). Primer sequences were: NHE1-fw: 5′ -CACACCACCATCAAATACTTCC-3′ , NHE1-rv: 5 ′ -GAACTTGTTGATGAACCAGGTC-3′ ; NHE2-fw: TTG GAGAGTCCCTGCTGAATGATG, NHE2-rv: tcagctgtgatgt aggacaaataactg, NBCn1-fw: 5′ -GCAAGAAACATTCTGACC CTCA-3′ , NBCn1-rv: 5′ -GCTTCCACCACTTCCATTACzCT, NBCe2-fw: atcttcatggaccagcagatcac, NBCe2-rv: tgcttggctggcatc aggaag. mRNA expression was quantified using the comparative threshold cycle (Ct) method, using β-actin as reference gene (fw 5′ - AGCGAGCATCCCCCAAAGTT-3′ , rv 5′ -GGGC ACGAAGGCTCATCATT-3′ ), and is given relative to that in HPDE cells.

### SDS-PAGE and Western Blotting

Cells were seeded to 60–70% confluency in 6-well culture dishes, lysed in lysis buffer [1% SDS, 10 mM Tris-pH 7.5, Na3VO<sup>4</sup> 1 mM, and complete protease inhibitor cocktail (Roche)] and homogenized by sonication. SDS-PAGE was performed using Bio-Rad CriterionTM TGXTM precast 10% gels in Tris/Glycine buffer. Gels were run for 1 h at 150 V, followed by transfer to Trans-Blot <sup>R</sup> TurboTM 0.2µm nitrocellulose membranes (Bio-Rad). Protein transfer was evaluated by Ponceau red staining, followed by blocking in 5% dry milk in TBST (0.01 M Tris/HCl, 0.15 M NaCl, 0.1% Tween 20) for 30 min at room temperature, incubation with primary antibodies overnight at 4◦C, and finally washing in TBST and incubation with HRP-conjugated secondary antibodies for 2 h at room temperature. Blots were developed using the FUSION-Fx chemiluminescence system (Vilber Lourmat). Images were processed in Adobe Photoshop and band intensities were quantified using UN-SCAN-IT gel 6.1 software.

### Immunofluorescence Microscopy Analysis

Cells were grown on glass coverslips in 6-well culture dishes, fixed in 4% paraformaldehyde for 15 min at room temperature, washed in icecold PBS, and permeabilized for 12 min in permeabilization buffer (0.2 % Triton X-100, 1% BSA in PBS). Unspecific fluorescence was blocked by a 30 min incubation in PBS plus 2% BSA, followed by incubation with primary antibody for 1½ h at room temperature, three washes in 2% BSA blocking buffer, and incubation with secondary antibodies diluted in 2% BSA blocking buffer) for 45 min, washing in PBS containing 2% BSA, a 5 min incubation with (4',6-Diamidino-2-Phenylindole, Dihydrochloride) (DAPI) for nuclear staining, extensive washing in PBS and mounting in N-propyl-gallate mounting media (2% w/v in PBS/glycerin). Fluorescence images were captured on a fully motorized Olympus BX63 upright microscope with an Olympus DP72 color, 12.8-megapixel, 4.140 × 3.096-resolution camera and with a fully motorized and automated Olympus IX83 Inverted microscope with a Hamamatsu ORCA-Flash 4.0 camera (C11440-22CU). The software used was Olympus CellSens dimension, which is able to do deconvolution on captured z stacks, and images were processed for publication using Adobe Photoshop CS6.

### Real-Time Imaging of Intracellular pH

Measurements of pH<sup>i</sup> were carried out essentially as described previously (21). Briefly, cells seeded in WillCo glass-bottom dishes (WillCo Wells, Amsterdam, the Netherlands) were loaded with 2′ ,7′ -bis-(2-carboxyethyl)-5-(and-6)-carboxyfluorescein acetoxymethyl ester (BCECF-AM, 1.6µM) in growth medium for 30 min at 37◦C. Cells were washed once in HCO<sup>−</sup> 3 containing Ringer solution [118 mM NaCl, 25 mM NaHCO3, 5 mM KCl, 1 mM MgSO4, 1 mM Na2HPO4, 1 mM CaCl2, 3.3 mM 3-(N-morpholino)propanesulfonic acid (MOPS), 3.3 mM 2- [Tris(hydroxymethyl)-methylamino]-ethanesulfonic acid (TES), 5 mM HEPES, pH 7.4], placed in a 37◦C imaging chamber equipped with gas and solute perfusion, at the stage of a Nikon Eclipse Ti microscope, and imaged using a 40X/1.4 NA objective and EasyRatioPro imaging software (PTI, NJ, USA). Emission was measured at 520 nm after excitation at 440 and 485 nm. Acidification was induced by exposure to 20 mM NH4Cl for 10 min. Calibration to pH<sup>i</sup> values was performed using the high K <sup>+</sup>/nigericin technique and a 4-point linear calibration curve.

### BrdU Proliferation Assay

Eighty percentage confluent Panc-1 and BxPC3 cells were trypsinized and resuspended in growth medium. Cells were seeded in 96-well plates (Celllstar <sup>R</sup> , cat # 655090). After 24 h incubation at 37◦C/5% CO2, plates were washed in PBS and 200 µl serum free medium was added. After 24 h, the cells were incubated with 10 ng/mL TGFβ, and/or 10µM of the NHE1 inhibitor cariporide. Forty-eight hours later, a Cell Proliferation ELISA, BrdU (chemiluminescent) kit (Roche, cat # 11 669 915 001) was used for determining the proliferation status of the cells, according to the manufacturer's instructions. 20 µl BrdU labeling solution was added to each well and the plates were incubated for 4 h, followed by a fixation/denaturing step, and incubation with a peroxidase-coupled mouse monoclonal anti-BrdU antibody. The plates were washed extensively, incubated with substrate solution containing luminol, 4-iodophenol and peroxide for 4 min on a shaker, and luminescence measured using a BMG FLUOstar OPTIMA Microplate Reader.

### Invasion Assay

Cell invasion was assessed using growth factor reduced corning <sup>R</sup> Matrigel <sup>R</sup> invasion chambers, 24-well plate and 8.0µm pore size (Corning, BioCoat, MA, USA). Prior to the experiment, invasion chambers containing Matrigel were rehydrated by adding 500 µl serum free medium for 2 h at 37◦C. Forty-eight hours after transfection with the indicated siRNAs, cells were serum starved, and 24 h later, treated or not with TGFβ for 48 h, before being washed with sterile PBS, trypsinized, washed 3 x in serum free medium, and 50,000 cells in this medium seeded into the upper chamber. Experiments were always done in duplicate. The lower chamber was filled with 10% serum containing medium. Chambers were incubated for 22 h at 37◦C/5% CO<sup>2</sup> to allow cells to invade through the Matrigel. Non-invaded cells from the upper surface of the chamber were removed with a cotton swab. Invaded cells on the lower surface of the chamber were fixed for 30 min in ice-cold absolute methanol, and filters were stained with 30% Giemsa solution (Sigma-Aldrich) for 30 min and mounted on glass slides. Invaded cells (four images per filter per condition) were counted using bright-field microscopy.

## Data Analysis and Statistics

Data are shown as individual representative experiments or as means of at least three independent experiments, with standard error of mean (SEM) error bars. Statistical significance was tested using Student t-test or one- or two-way ANOVA followed by Tukey's multiple comparison tests as appropriate, using GraphPad prism 6.

### RESULTS

### Expression of Net Acid Extruding Transporters Is Increased in PDAC Cells

Our previous in silico analysis of RNA sequencing data from pancreatic cancer patient tissue indicated that a number of SLC9 and SLC4 family transporters are upregulated in PDAC patient tumor tissue compared to normal pancreatic epithelium (29). From these, we selected for widely expressed net acid extruders for initial analyses: two Na+/H<sup>+</sup> exchangers, NHE1 and−2 (SLC9A1-2), and two Na+,HCO<sup>−</sup> 3 cotransporters (NBCs), NBCe2 (SLC4A5) and NBCn1 (SLC4A7). The relative mRNA levels of the transporters were determined by RT-qPCR analysis in immortalized pancreatic epithelial (HPDE) cells (27, 28), and in a panel of human PDAC cell lines of different genotypes: MIAPaCa-2, Panc-1, BxPC-3, and AsPC-1 cells. All PDAC cell lines studied exhibited increased mRNA levels of one or more of these transporters compared to HPDE cells, with the specific transporters upregulated differing between cell lines (**Figures 1A,B**). Thus, BxPC-3 and AsPC-1 cells showed elevated expression of NHE1 and−2, and MIAPaCa-2 cells predominantly showed upregulation of NBCn1 and NBCe2, while both NHE2 and NBCn1 were upregulated in Panc-1 cells. For further analyses, we focused on NHE1 and NBCn1, which play central roles in the development of PDAC and other cancers (21, 22, 24, 30–33). Consistent with the mRNA data, western blot analysis revealed that the NHE1 protein level was significantly increased in BxPC-3 cells and the NBCn1 protein level in MIAPaCa-2 and Panc-1 cells, compared to that observed in HPDE cells (**Figures 1C,D**).

These data show that net acid extruding proteins are upregulated in PDAC cells compared to normal pancreatic ductal epithelial cells, and that the specific pattern of upregulation is cell type dependent.

### TGFβ-1 Treatment Elicits a SMAD4-Dependent EMT in PDAC Cells

BxPC-3 and Panc-1 cells, which display a high levels of NHE1 and NBCn1 expression, respectively, were chosen for further analysis. We first assessed the ability of TGFβ ligand to induce EMT in the two cell lines. Cells were serum-starved for 24 h, followed by treatment with 10 ng/ml human recombinant TGFβ-1 for 48 h. After lysis, cells were subjected to western blot analysis for the epithelial marker E-cadherin as well as the mesenchymal markers α-smooth muscle actin (α-SMA) and connective tissue growth factor (CTGF) (**Figure 2A**). In Panc-1 cells, TGFβ-1 treatment significantly reduced the level of E-cadherin, and increased that of α-SMA and CTGF, consistent with induction of

FIGURE 1 | mRNA and protein expression of net acid extruding transporters is upregulated in PDAC cells compared to normal pancreatic ductal epithelial cells. (A,B) Immortalized pancreatic ductal epithelial (HPDE) cells and a panel of PDAC cell lines: MIAPaCa-2, Panc-1, BxPC-3, and AsPC-1 cells, were grown to about 80% confluence, lysed and mRNA expression of NHE1 and−2 (A) and NBCe2 and NBCn1 (B) analyzed by RT-qPCR. Data (mean with S.E.M. error bars) are normalized to the expression level in HPDE cells in the same experiment (shown as a dotted line). (C,D) HPDE, MIAPaCa-2, Panc-1, BxPC-3, and AsPC-1 cells were grown to about 80% confluence, lysed and protein levels of NHE1 (C) and NBCn1 (D) analyzed by Western blotting. Data (mean with S.E.M. error bars) are normalized to the β-actin loading control and to the expression level in HPDE cells in the same experiment (shown as a dotted line). *n* = 3–4 independent experiments per condition in all panels. \* , \*\* One-way ANOVA, *p* < 0.05 and 0.01, respectively, against the level in HPDE cells.

EMT (**Figures 2A,B**). In BxPC-3 cells, TGFβ-1 treatment elicited a small yet significant decrease in E-cadherin of about 25%, which was not accompanied by detectable changes in α-SMA and CTGF expression (**Figures 2A,C**). Immunofluorescence analysis was performed to assess the impact of TGFβ-1 treatment on cell morphology and protein localization in Panc-1 cells. TGFβ-1 treatment induced a marked internalization of both E-cadherin and β-catenin (arrows), a strong, predominantly membraneassociated increase in α-SMA, and increased intracellular CTGF expression (**Figure 2D**).

These data show that TGFβ-1 treatment elicited a marked EMT in Panc-1 cells, and a modest and partial EMT in BxPC-3 cells.

BxPC-3 cells are SMAD4-deficient (34), which has previously been shown to be responsible for their partial resistance to TGFβ-1-induced EMT (14), although the absolute requirement for SMAD4 for TGFβ-1-induced EMT is controversial (14, 35). To assess the role of SMAD4 in TGFβ-1-induced EMT in Panc-1 cells, we knocked down SMAD4 by siRNA in Panc-1 cells, followed by TGFβ-1 treatment for 48 h, and western blotting for E-cadherin and CTGF (**Figures 2E–H**). In mocktransfected cells, TGFβ-1 treatment almost doubled the level of SMAD4. Confirming efficient knockdown, the SMAD4 protein level was reduced by more than 80% in SMAD4 siRNA-treated cells (**Figures 2E,F**). Notably, the E-cadherin level was more than doubled by SMAD4 knockdown in the absence of TGFβ-1, and this was only marginally reduced by TGFβ-1-induced increase in CTGF expression was abolished in SMAD4-depleted cells (**Figure 2G**). Conversely, SMAD4 knockdown had no effect on the CTGF level under control conditions yet abolished the increase in CTGF induced by TGFβ-1 treatment (**Figure 2H**).

Collectively, these data show that TGFβ-1 induced EMT in Panc-1 cells is strongly dependent on SMAD4 and suggest that a basal level of SMAD4 signaling in the absence of TGFβ-1 treatment is responsible for the low basal E-cadherin level in these cells.

### NHE1 Expression Is Increased in a SMAD-Dependent Manner During TGFβ-1-Induced EMT

We next asked whether TGFβ-1 treatment altered NHE1 and NBCn1 expression in Panc-1 and BxPC-3 cells. The protein expression level of NHE1 increased by about 60%, and that of NBCn1 by about 40%, following TGFβ-1 treatment of Panc-1 cells (**Figures 3A,B**), and a similar, albeit non-significant trend was seen in BxPC-3 cells (**Figures 3A,C**). The mRNA level of NHE1 also tended to be upregulated by TGFβ-1 in Panc-1 cells, and that of NBCn1 in BxPC-3 cells (**Figures 3C,D**). The TGFβ-1-induced increase in NHE1 expression was strongly reduced by SMAD4 knockdown in Panc-1 cells (**Figures 3E,F**), whereas the modest increase in NBCn1 expression was not affected by SMAD4 knockdown (**Figures 3E,G**). This suggests that increased NHE1 expression is mainly mediated by canonical TGFβ signaling, whereas the effect of TGFβ-1 on NBCn1 expression appears to be mediated by non-canonical signaling events. Both transporters localize predominantly to the plasma membrane in both cell types. Interestingly, TGFβ-1 treatment caused a characteristic redistribution of NBCn1 to membrane ruffles in BxPC-3 cells (**Figure 3I**) whereas there was no detectable redistribution in Panc-1 cells (**Figure 3H**).

To determine whether NHE1 and NBCn1 played a role in the EMT process per se, we knocked down each transporter in Panc-1 cells and subjected the cells to 48 h of TGFβ-1 treatment as above, followed by western blotting for E-cadherin and CTGF. Although the protein level of both transporters was essentially abolished upon siRNA treatment, this did not detectably affect TGFβ-1–induced E-cadherin downregulation or CTGF upregulation (**Supplementary Figure 1**).

These results show that TGFβ-1 induced EMT is associated with increased expression of NHE1 and NBCn1 and increased localization of the transporters at the plasma membrane where they could increase net acid extrusion from the cells. However, neither transporter is required for induction of EMT per se.

### TGFβ-1 Increases Steady State pH<sup>i</sup> and NHE1-Dependent Acid Extrusion in Panc-1 Cells

To directly determine whether the changes in transporter expression upon TGFβ-1-induced EMT are associated with altered pH<sup>i</sup> homeostasis, cells were subjected to TGFβ-1 treatment for 48 h, loaded with the pH<sup>i</sup> sensitive fluorophore BCECF-AM, and steady state pHi determined by live imaging of BCECF fluorescence. Experiments were carried out in the presence of CO2/HCO<sup>−</sup> 3 to allow contributions from bicarbonate-dependent transporters such as NBCn1. TGFβ-1 treatment significantly increased steady state pH<sup>i</sup> in Panc-1 cells, from an average value of 7.06 ± 0.105 in control cells to 7.26 ± 0.043 after TGFβ-1 treatment (**Figure 4A**). In contrast, the steady state pH<sup>i</sup> of BxPC-3 cells was 7.01 ± 0.062 in controls and 7.09 ± 0.066 after TGFβ-1 treatment, not significantly different (**Figure 4B**).

To evaluate whether cellular capacity for acid extrusion was increased by TGFβ-1 treatment, we next determined the pH<sup>i</sup> recovery rate. Cells were pretreated with or without TGFβ-1 for 48 h as above, and subjected to an NH4Cl-prepulse to acidify the cells: after equilibration in normal CO2/HCO<sup>−</sup> 3 saline, cells were perfused with 15 mM NH4Cl. Upon dissociation, NH<sup>3</sup> rapidly enters the cells by diffusion. Its association with cellular H<sup>+</sup> to NH<sup>+</sup> 4 causes near-instantaneous alkalinization, followed by slow return toward steady state pH<sup>i</sup> as NH<sup>+</sup> 4 from the solution enters the cells via ion transporters, shifting the equilibrium. When cells are again perfused with normal CO2/HCO<sup>−</sup> 3 saline, all NH<sup>3</sup> rapidly diffuses out, and an excess of free H<sup>+</sup> is left in the cells. The rate of recovery from this acidification is a measure of net acid extrusion capacity, determined as the slope of the initial, linear part of the curve under conditions where starting pH<sup>i</sup> is similar (36). Experiments were again carried out in the presence of CO2/HCO<sup>−</sup> 3 . **Figures 4C,D** shows representative traces of pH<sup>i</sup> over time from the maximal acidification. As seen, TGFβ-1 treatment markedly increased the pH<sup>i</sup> recovery rate in Panc-1 cells (**Figure 4C**) but not in BxPC-3 cells (**Figure 4D**). NHE1 and NBCn1 activity is posttranslationally regulated (37,

FIGURE 2 | TGFβ-induces a predominantly SMAD4-dependent EMT. (A) Panc-1 and BxPC-3 cells were serum starved for 24 h, followed by 48 h growth with or without 10 ng/ml human recombinant TGFβ-1 for 48 h as indicated. Cells were lysed and subjected to western blotting for E-cadherin, α-SMA, and CTGF, using GAPDH and DCTN1 as loading markers. Representative western blots are shown. (B,C) Summarized, quantified data from experiments as in (A), for Panc-1 cells (B) and BxPC-3 cells (C). Data (mean with S.E.M. error bars) were normalized to loading control and to the level in the absence of TGFβ-1 (Ctrl.), shown by the dotted line. *n* = 3–8 independent experiments per condition. \*, \*\*, \*\*\*\* One-way ANOVA, *p* < 0.05, 0.01, 0.0001, respectively, against the level in non-TGFβ-1 treated control cells. (D) Immunofluorescence microscopy analysis of the localization and expression level of E-cadherin, β-catenin, α-SMA and CTGF in Panc-1 cells. Nuclei are stained using DAPI. The images shown are representative of at least three independent experiments per condition. Arrows indicate the distinct plasma membrane localization of E-cadherin and β-catenin under control conditions and the shift in localization upon TGFβ-1 treatment. (E–H) Effect of SMAD4 knockdown in Panc-1 *(Continued)* FIGURE 2 | cells on E-cadherin and CTGF protein levels. E, representative western blot, F-H, SMAD4, E-cadherin, and CTGF protein levels, normalized to mock ctrl. and shown as mean with S.E.M. error bars and individual data points. Relative to mock siRNA-treated controls, SMAD4 expression was 1.8 ± 0.55 in mock siRNA + TGFβ-1 treated cells, 0.18 ± 0.066 in SMAD4 siRNA treated cells, and 0.18 ± 0.063 in SMAD4 siRNA + TGFβ-1 treated cells. β-actin and Histone 3 (H3) are used as loading markers. Data are from four independent experiments per condition. \*, \*\* One-way ANOVA, *p* < 0.05, 0.01, respectively.

38), hence, their contributions to pH<sup>i</sup> regulation are not predicted by their protein levels. Notably, in both Panc-1 and BxPC-3 cells, pH<sup>i</sup> recovery in both control- and TGFβ-1 treated cells was abolished by the specific NHE1 inhibitor cariporide regardless of (**Figures 4E,F**).

Collectively, these results show that TGFβ-1-induced EMT in Panc-1 cells is associated with increased steady state pH<sup>i</sup> and increased NHE1-dependent acid extrusion capacity.

### TGFβ-1 Differentially Regulates Cell Proliferation in Panc-1 and BxPC-3 Cells

Depending on the cell type and context, TGFβ signaling can counteract or stimulate cell proliferation, and this has been ascribed at least in part to SMAD2/3 signaling, which rely on SMAD4 (4, 6). We therefore assessed the impact of TGFβ-1 treatment on cell proliferation in Panc-1 and BxPC-3 cells. Cells treated for 48 h with or without TGFβ-1 were lysed and blotted for phosphorylated retinoblastoma protein (p-pRb) and proliferating cell nuclear antigen (PCNA) as markers of cell cycle entry and progression, respectively. Notably, TGFβ-1 treatment increased the p-pRb level in Panc-1 cells yet decreased it in BxPC-3 cells (**Figures 5A,B**), and similar results were obtained for PCNA (**Figures 5A,C**). This was confirmed by IFM analysis using Ki-67 as a proliferation marker: The fraction of Ki-67 positive cells was decreased in both cell types—although most dramatically in Panc-1 cells—by 48 h of serum starvation, and was increased in Panc-1 cells, yet decreased in BxPC-3 cells by TGFβ-1 (**Figures 5D–F**). Strikingly, TGFβ-1 treatment also increased p53 expression in Panc-1 cells, but not in BxPC-3 cells (**Figure 5G**). Co-staining for p53 and p-pRb confirmed these data and revealed that elevation of nuclear staining for p-pRb and p53 was detected within the same cells (**Supplementary Figure 2**). BrdU incorporation analysis confirmed that TGFβ-1 treatment increased proliferation of Panc-1 cells, yet decreased that of BxPC-3 cells, and showed that proliferation was unaffected by inhibition of NHE1 (10µM cariporide) or NBCs (10µM S0859), under both basal and TGFβ-1-treated conditions (**Figures 5H,I**).

Taken together, these results show that TGFβ-1 treatment stimulates proliferation of Panc-1 cells yet inhibits that of BxPC-3 cells. This occurs in parallel with a TGFβ-1-induced increase in p53 expression in the Panc-1 cells and is unaffected by inhibition of NHE1 or NBCn1.

### Loss of Merlin Stimulates TGFβ-1-Induced Invasiveness in an NHE1- and NBCn1-Dependent Manner

Previous studies have demonstrated the involvement of acid-base transporters in cell motility and invasiveness, through roles of the transporters in cell adhesion, cytoskeletal dynamics,and matrix degradation (19, 25, 39–41). We therefore asked whether the TGFβ-1-induced upregulation of NHE1 and NBCn1 contributed to TGFβ-1-induced invasiveness of Panc-1 cells.

Stimulation with TGFβ-1 increased invasion of Panc-1 cells through Matrigel more than 4-fold (**Figure 6B**), consistent with the strong TGFβ-1-induced EMT induction in these cells (**Figure 2**). siRNA-mediated knockdown of NHE1 further increased TGFβ-1-induced invasiveness, whereas knockdown of NBCn1 or both transporters in combination had no effect (**Figures 6A,B**). The unexpected exacerbation of invasiveness by NHE1 knockdown prompted us to ask whether the impact of NHE1 on PDAC cell invasiveness was genotype-dependent. The tumor suppressor Merlin (aka Neurofibromatosis type 2, NF2), which is downregulated in many PDAC tumors (42), was reported to regulate EMT (43) and cell motility (44), and was recently shown to regulate NHE1 in melanoma cells (45). siRNA-mediated knockdown of Merlin (**Figure 6A**) increased basal invasion 4-fold and almost doubled TGFβ-1-induced invasiveness (**Figure 6C**). Notably, in Merlin-depleted cells, knockdown of either NHE1 or NBCn1 abolished the increase in invasion (**Figure 6C**).

These results show that Merlin depletion increases basal invasion and potentiated TGFβ-1-induced invasiveness. Invasion induced by Merlin depletion, but not that induced by TGFβ-1 alone, is dependent on NHE1 and NBCn1.

### DISCUSSION

Pancreatic cancer has one of the most ominous mortality rates of any cancer globally, and the relative burden of disease is expected to increase over the next decade (1). PDAC tumors frequently exhibit abundant secretion of TGFβ from both stromal and tumor cells (46). Another characteristic of PDAC is extensive EMT (14, 15), the roles of which in PDAC development are incompletely understood (11, 16).

An emerging hallmark of solid tumors is profound dysregulation of pH homeostasis and upregulation of net acid-extruding transport proteins which play important roles in cancer development (19, 20). Given the extreme acid-base homeostasis of the pancreas under physiological conditions (3), PDAC is particularly interesting in this context. Here, we asked whether TGFβ-1-induced EMT in PDAC cells is associated with upregulation of net acid extruding transporters. We focused on NHE1 and NBCn1, which we and others have shown to play important roles in cancer development (21–24), and found that these transporters were upregulated in PDAC cell lines compared to normal controls. The great majority of patient PDAC tumors show activating KRAS mutations, and about 50% show mutations in TGFβ pathway components,

*(Continued)*

FIGURE 3 | non-TGF-β treated control cells. Data are mean with S.E.M. error bars, of 8-11 independent experiments per condition. (E–G) Effect of SMAD4 knockdown on NHE1 and NBCn1 protein levels in Panc-1 cells. (E) representative western blot, (F,G) NHE1 and NBCn1 protein levels, normalized to mock ctrl. and shown as mean with S.E.M. error bars and individual data points. Data shown are from 4 independent experiments per condition. For SMAD4 knockdown efficiency, see Figure 2E and Figure 2 legend. \*\*, \*\*\* One-way ANOVA, *p* < 0.05, 0.01, respectively. (H,I) IFM analysis illustrating the localization and expression level of native NHE1 (red) and NBCn1 (green) in Panc-1 (H) and BxPC-3 (I) cells. Nuclei are stained using DAPI. Images shown represent at least three independent experiments per condition. Scale bars are 20µm, and 10 µm in the zoomed images.

most commonly SMAD4 loss or inactivating mutations (4). For further analysis we therefore selected Panc-1 cells, which harbor an activating KRAS mutation and express wild type SMAD4, and BxPC-3 cells, which are KRAS wild type and express a truncated, defective version of SMAD4 (34). Consistent with previous reports (14, 15, 35), 48 h TGFβ-1 treatment robustly induced EMT characteristics, i.e., downregulation of E-cadherin, upregulation of α-SMA and CTGF, and internalization of β-catenin in Panc-1 cells, compared to a partial EMT in BxPC-3 cells where only E-cadherin expression was detectably altered. Accordingly, SMAD4 knockdown in Panc-1 cells largely abolished EMT. EMT has been reported to involve both SMAD4-dependent (17) and–independent (18) processes. TGFβ-induced E-cadherin downregulation in Panc-1 cells and other PDAC cell lines was reported to be nearly completely dependent on SMAD3 and SMAD4, as well as the EMT-associated transcription factors SLUG and SNAIL, which are activated downstream from SMAD4 (35). The SMAD3-4 complex binds and activates the SNAIL promoter, in turn leading to the SNAIL-dependent E-cadherin downregulation (47). However, non-canonical TGFβ pathways including ERK1/2 were also reported to contribute to EMT (35), consistent with our finding of partial EMT induction in BxPC-3 cells (34).

A central finding of this work was that TGFβ-1-induced EMT was associated with upregulation of NHE1, and to a lesser extent NBCn1, expression in Panc-1 cells. Consistent with this, NHE1 was recently proposed to be upregulated by ZEB1, a transcription factor involved in driving EMT (48). Knockdown of the NHE1 or NBCn1 had no detectable effect on EMT induction per se (E-cadherin, CTGF and α-SMA levels). However, in congruence with the transporter upregulation, EMT induction was associated with an increase in steady state pH<sup>i</sup> and increased capacity for net acid extrusion in Panc-1 cells but not in BxPC-3 cells. Acid extrusion was potently inhibited by the NHE1 inhibitor Cariporide in both cell lines, despite the presence of CO2/HCO<sup>−</sup> 3 and the relatively high expression of NBCn1 in the Panc-1 cells. Both cell lines also express NHE2 at rather high levels (**Figure 1A**). With reported Ki values for Cariporide of 0.05µM for NHE1 and 3µM for NHE2 in transfected fibroblasts (49), it seems likely that both isoforms might contribute to this recovery (see also below). Notably, the stimulatory effect of TGFβ on pH<sup>i</sup> regulation in Panc-1 cells differs from the reported inhibitory effect of TGFβ on pH<sup>i</sup> regulation in noncancer hepatocytes (50). Thus, it appears that, fully in line with the opposite effects of TGFβ signaling on proliferation and survival in cancer- and non-cancer cells (6, 8), TGFβ attenuates pH<sup>i</sup> regulation in normal cells, yet stimulates it in some cancer cells.

TGFβ-1 treatment significantly increased proliferation of the SMAD4-positive Panc-1 cells, while decreasing it in the SMAD4 deficient BxPC-3 cells. Paradoxically, at the same time, TGFβ treatment increased p53 expression in Panc-1 cells but not in BxPC-3 cells. Thus, clearly, the presence of SMAD4 and TGFβ-1-induced p53 signaling does not prevent a pro-proliferative effect of the ligand. Both TGFβ and acid extruding proteins are important regulators of cell proliferation/death balance, and EMT has recently been recognized to play important roles also in control of cancer cell survival (11, 16). We therefore reasoned that transporter upregulation might regulate the balance between pro-death and pro-growth effects of TGFβ in PDAC, yet, inhibition of NHE1 or NBCs did not affect proliferation of either cell line under these conditions. It remains possible that such effects might be uncovered in the severely nutrient-deprived and hypoxic tumor microenvironment, where NHE1 has been proposed to play a role in nutrient uptake in PDAC cells via macropinocytosis (51).

Given the roles of pH<sup>i</sup> and pH<sup>e</sup> in general, and NHE1 and NBCn1 in particular, in cancer cell motility and invasion (19, 25, 39–41), we asked whether NHE1 and NBCn1 impacted TGFβ-1-induced invasiveness. Forty-eight hours of TGFβ-1 treatment robustly increased invasiveness of Panc-1 cells and this was not abolished by transporter knockdown; in fact, siRNA-mediated depletion of NHE1 modestly increased invasiveness, a finding reminiscent of our demonstration that counter to its general role in favoring motility and invasiveness (25), NHE1 inhibition actually increased motility of p95HER-overexpressing breast cancer cells, while inhibition of NBCs had no effect (52). The lack of effect of NHE1 knockdown shown here contrasts the conclusion of a recent report using 10µM cariporide to inhibit NHE1 (32). Since NHE1 knockdown was highly efficient in our study, it seems possible that the effect of cariporide may reflect a contribution of NHE2, which, as noted above, would also have been inhibited at this concentration. The tumor suppressor Merlin exhibits reduced expression in PDAC patients (42, 53), is regulated by TGFβ (54), has been assigned a role in regulation of EMT in ARPE-19 cells (43), and in regulation of NHE1 in melanoma cells (45). We therefore hypothesized that Merlin depletion would impact TGFβ-1-induced invasiveness and its regulation by NHE1 and NBCn1. Indeed, Merlin knockdown in itself robustly increased invasiveness, and under these conditions, knockdown of either NHE1 or NBCn1 decreased invasiveness, a tendency seen under both basal and TGFβ-stimulated conditions. This indicates that the roles of NHE1 and NBCn1 may be particularly important in cancers with Merlin downregulation. It is well documented that the roles of pH regulatory transporters in invasion involve effects on focal adhesion strength and turnover, cytoskeletal dynamics, and matrix degradation, downstream of

pH (pH<sup>i</sup> ) was monitored by real-time imaging analysis in Panc-1 (A,C,E) and BxPC-3 (B,D,F) cells after loading of cells with BCECF-AM. Experiments were conducted at 37◦C, and under CO2/HCO<sup>−</sup> 3 buffered conditions. (A,B) Steady state pH<sup>i</sup> was increased by TGFβ-1 in Panc-1 but not in BxPC-3 cells. The graphs show the pH<sup>i</sup> in paired control and TGFβ-1 treated cells from each experiment. The mean pH<sup>i</sup> values were, for Panc-1 cells, 7.06 ± 0.105 in control cells and 7.26 ± 0.043 after TGFβ-1 treatment, and in BxPC-3 cells, 7.01 ± 0.062 in controls and 7.09 ± 0.066 after TGFβ-1. Data are based on 5 (Panc-1) and 6 (BxPC-3) independent biological experiments for each condition. \* significantly different, *p* < 0.05, paired *t*-test. (C–F) The pH<sup>i</sup> recovery rate was determined as the initial rate of recovery after NH4Cl-prepulse-induced intracellular acidification. Where indicated, cariporide (10µM) was present during the recovery phase to inhibit NHE1. (C,D) Representative traces with SD error bars, (E,F) means with S.E.M. error bars of 3–5 independent experiments per condition. \*\*, \*\*\* Paired *t*-test, *p* < 0.01 and 0.001, respectively against corresponding conditions (Ctrl, TGFβ) in absence of cariporide).

transporter-mediated changes in pH<sup>i</sup> and pericellular pH<sup>e</sup> (41). While the precise roles of the transporters in Panc-1 cell invasion were not further studied here, this is supported by the increased pH<sup>i</sup> and acid extrusion capacity in TGFβ-stimulated Panc-1 cells also demonstrated here.

A limitation of this study is that we did not study the correlation of NHE1 and NBCn1 with EMT markers, TGFβsignaling pathway components, and Merlin expression in patient tumor tissue from primary pancreatic tumors and metastases. Future work should address such correlations to evaluate the

/or Merlin as indicated, starved for 48 h, and treated or not with TGFβ-1 for 48 h, followed by Western blotting to assess protein expression levels. Data represent three independent experiments. (B) Panc-1 cells were treated with siRNA against NHE1 or NBCn1 as indicated, treated or not with TGFβ-1 as above for 48 h, and seeded at 50,000 cells/well in in serum-free medium in Matrigel-coated Boyden chambers. The lower chamber contained 10% FBS. Twenty-two hours later, experiments were terminated and the number of invaded cells determined. (C) As (B), except including knockdown of merlin as indicated. *n* = 3–15 independent experiments, each done in duplicate, per condition. \*, \*\*, \*\*\*\*) *p* < 0.05, 0.01, and 0.0001, respectively, 2-way ANOVA, ctrl. vs. TGFβ; #, ## *p* < 0.05, 0.01, respectively, 2-way ANOVA between the indicated conditions.

relevance of TGFβ-mediated regulation of these transporters to invasiveness in patients.

In conclusion, we show here that NHE1 and NBCn1 expression and NHE-dependent acid extrusion are upregulated during TGFβ-1-induced EMT of Panc-1 cells. NHE1 upregulation is SMAD4-dependent, and SMAD4-deficient BxPC-3 cells show no change in pH<sup>i</sup> regulation. The difference between Panc-1 and BxPC-3 cells is corroborated in opposite effects of TGFβ-1 on cell proliferation, which is increased in Panc-1 and decreased in BxPC-3 cells by treatment with this ligand. Knockdown of Merlin strongly potentiates TGFβ-1-induced Panc-1 cell invasiveness in a manner dependent on acid-extruding transporters. We suggest that these transporters are novel players in the events induced during TGFβ-1-induced EMT in PDAC cells.

### DATA AVAILABILITY STATEMENT

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

### AUTHOR CONTRIBUTIONS

SP, SC, and LP developed the concept. SP, SC, RM, and KZ designed the experiments. RM, PS, and KZ performed the experiments for **Figures 2A–D**, **3A–D,H–I**, and RM the experiments for **Figure 6** and **Supplementary Figure 1**. SP performed the experiments in **Figure 4**. Experiments in **Figures 5A–G** and **Supplementary Figure 2** were performed by PS, and **Figures 5H–I** by ML. Data analysis and figures were done by SP, RM, KZ, PS, and ML. RM and SP wrote the manuscript with comments and inputs from all co-authors. All authors have seen and approved the final version of the manuscript.

### FUNDING

This work was supported by the Marie Curie Initial Training Network IonTraC (grant agreement no. 289648) and by the Hartmann Foundation. RM was supported by a PhD fellowship

### REFERENCES


from the Government of India and the Department of Biology, University of Copenhagen.

### ACKNOWLEDGMENTS

HPDE cells were a kind gift from Dr. Ming-Sound Tsao, Ontario Cancer Institute, Toronto, Canada (27, 28), and antibody against NBCn1 was a kind gift from Prof. J. Praetorius, Aarhus University. We gratefully acknowledge S. C. Kong for performing parts of the experiments for **Figure 1**, and the excellent technical assistance from K. F. Mark.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fonc. 2020.00687/full#supplementary-material


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

Copyright © 2020 Malinda, Zeeberg, Sharku, Ludwig, Pedersen, Christensen and Pedersen. 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.

# Dynamically Shaping Chaperones. Allosteric Modulators of HSP90 Family as Regulatory Tools of Cell Metabolism in Neoplastic Progression

Carlos Sanchez-Martin<sup>1</sup> , Stefano A. Serapian<sup>2</sup> , Giorgio Colombo2,3 \* and Andrea Rasola<sup>1</sup> \*

<sup>1</sup> Dipartimento di Scienze Biomediche, Università di Padova, Padua, Italy, <sup>2</sup> Dipartimento di Chimica, Università di Pavia, Pavia, Italy, <sup>3</sup> Istituto di Chimica del Riconoscimento Molecolare, CNR, Milan, Italy

#### Edited by:

Alessandra Castegna, University of Bari Aldo Moro, Italy

#### Reviewed by:

Didier Picard, Université de Genève, Switzerland Ciro Leonardo Pierri, University of Bari Aldo Moro, Italy

#### \*Correspondence:

Giorgio Colombo g.colombo@unipv.it Andrea Rasola andrea.rasola@unipd.it

#### Specialty section:

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

Received: 27 January 2020 Accepted: 10 June 2020 Published: 16 July 2020

#### Citation:

Sanchez-Martin C, Serapian SA, Colombo G and Rasola A (2020) Dynamically Shaping Chaperones. Allosteric Modulators of HSP90 Family as Regulatory Tools of Cell Metabolism in Neoplastic Progression. Front. Oncol. 10:1177. doi: 10.3389/fonc.2020.01177 Molecular chaperones have recently emerged as fundamental regulators of salient biological routines, including metabolic adaptations to environmental changes. Yet, many of the molecular mechanisms at the basis of their functions are still unknown or at least uncertain. This is in part due to the lack of chemical tools that can interact with the chaperones to induce measurable functional perturbations. In this context, the use of small molecules as modulators of protein functions has proven relevant for the investigation of a number of biomolecular systems. Herein, we focus on the functions, interactions and signaling pathways of the HSP90 family of molecular chaperones as possible targets for the discovery of new molecular entities aimed at tuning their activity and interactions. HSP90 and its mitochondrial paralog, TRAP1, regulate the activity of crucial metabolic circuitries, making cells capable of efficiently using available energy sources, with relevant implications both in healthy conditions and in a variety of disease states and especially cancer. The design of small-molecules targeting the chaperone cycle of HSP90 and able to inhibit or stimulate the activity of the protein can provide opportunities to finely dissect their biochemical activities and to obtain lead compounds to develop novel, mechanism-based drugs.

Keywords: chaperones, HSP90, TRAP1, mitochondria, tumor metabolism, allosteric inhibitor, ATP-competitive inhibitors, anti-neoplastic strategies

## INTRODUCTION

Chaperones are molecular machines that assist folding, conformational changes and subcellular trafficking of proteins and control their degradation following aggregation, unfolding or misfolding. Fine and orchestrated tuning of these processes is carried out by different chaperone families and leads to maintenance and quality control of the proteome, an extremely complex and vital task for cells (1). Heat Shock Protein 90 (HSP90) proteins are chaperones that exert their regulatory functions on the structure and activity of a variety of diverse client proteins, thus integrating signaling and metabolic circuitries and acting as crucial components in consenting flexible adaptations of cells to environmental changes and stresses (2).

Exposure to harmful environmental stimuli is a common event in the process of tumor growth. Fluctuations in pH, oxygen, or nutrient availability prompt a profound rewiring of the metabolic circuitries of neoplastic cells and major changes in the homeostasis of their proteome (proteostasis) (3, 4). These noxious conditions also affect biochemical functions confined in specific subcellular compartments, such as protein folding in the endoplasmic reticulum (ER) (5, 6) as well as the bioenergetic functions of mitochondria (7, 8). Activation of organellerestricted signaling pathways and metabolic adaptations can subtly regulate the equilibrium among death, dormancy, and aggressiveness of tumor cells (9–11). In order to cope with these stresses and sustain pro-oncogenic biological routines, including growth, proliferation, invasion, metastasis and evasion from death stimuli, most cancer cells overexpress HSP90 family members (12, 13). The various paralogs of HSP90 proteins can play a key role at the crossroads of these multiple cellular functions in the different cellular districts, namely HSP90 in the cytosol, Grp94 in the ER and TRAP1 in mitochondria (14–16); extracellular HSP90 is also involved in cell-tocell communication (17). Induction of Hsp90 family protein expression contributes to the adaptations of the metabolic machineries in tumor cells and has been associated with cancer progression, resistance to chemotherapy and poor prognosis (12, 18). Adaptability to stress conditions is linked to specific subsets of clients, and the range of functional flexibility of the chaperone and of potential activities of its interactors are further expanded both by post-translational modifications and by the recruitment of other chaperones and of co-chaperones (19– 21). In this context, the development of drugs targeting HSP90 components has emerged as a promising anti-neoplastic strategy.

Here, we report our views on molecular design strategies aimed to act on the circuitries in which HSP90 family members play a key role in cancer cells. In particular, we focus on cytosolic HSP90 and its mitochondrial paralog TRAP1, as Grp94 has a more specialized role in the maturation process of particular secretory and membrane-bound proteins clients, such as immunoglobulins, integrins, and Toll-like receptors (14). We will discuss interventions that range from the use of allosteric modulators of chaperone functions, to the targeting of proteinprotein interactions involved in the assembly of functional complexes. We also discuss possible perspectives in combining the use of molecules that target HSP90 complexes with the use of other antineoplastic compounds, with a particular focus on the control of metabolic vulnerabilities in cancer models.

### STRUCTURE AND FUNCTION OF HSP90 MOLECULAR CHAPERONES

Molecular chaperones of the HSP90 family are essential cell constituents, making up 1–2% of the proteome. Their expression can be further stimulated by stress, and in tumor cells they can reach up to the 4–7% of the expressed proteome, thus shaping all biological processes required for neoplastic progression (**Figure 1**). HSP90 exists as a homodimer and each individual chain consists of three globular domains (22). Structures reporting different full length chaperone isoforms can be found at the following pdb codes: 2ioq, 2iop, 2cg9, 2o1v, 2o1u, 4job, 4ipe, 4iyn, 5uls, 5tvu, 5tvx, 6d14, 5tth, 5tvw.

The structures show the common organization in a Nterminal ATP-binding domain (N-domain), a middle domain (M-domain) involved in ATP hydrolysis, and a C-terminal domain (C-domain) responsible for HSP90 dimerization and for interactions with several co-chaperones. HSP90, TRAP1, and Grp94 have a mutual sequence identity of about 30–40%, which reflects in the high structural similarity and alignability of their individual domains (23–25). However, the preferential relative orientation of the domains in the crystal structures solved so far varies significantly depending on the protein, cellular compartment, and organism (26), yielding a global root mean square deviation (RMSD) of atomic positions of at least 7Å.

HSP90 chaperones manifest their functions by promoting the folding and tuning the activity of a plethora of clients endowed with highly diverse structures, cellular localizations and functions. The two main cytosolic HSP90 isoforms, HSP90α and HSP90β, have an interactome that includes more than 400 putative clients (https://www.picard.ch/HSP90Int/index. php), making them central modulators of at least a dozen of important biochemical pathways, including stress regulation, protein folding, DNA repair, kinase signaling, cell survival and metabolism (2, 12). HSP90 effects on clients encompass facilitating the formation of specific protein conformations, as in the case of kinase activation (27), prompting the assembly of multiprotein complexes (28), stabilizing the bindingcompetent conformation of ligand receptors, and regulating protein dynamics and conformational state ensembles (29). Client stability depends on the chaperone, and its inhibition induces proteasomal degradation of client proteins.

Dimers of HSP90 family proteins undergo a complex functional cycle that might allow them to adapt to different client proteins. ATP binding elicits a series of conformational changes (**Figure 2**) leading to the "closed conformation" of the chaperone in which ATP hydrolysis occurs. Induction of the closed state is the rate-limiting step of the reaction. ATP binding has a much lower affinity than ADP binding (K<sup>D</sup> ∼400µM vs. ∼10µM), indicating that under physiological conditions of nucleotide concentrations, cytosolic Hsp90 primarily populates two states that are absent in ATP-regenerating conditions: either ADP bound to both arms, or ATP bound to one arm and ADP bound to the opposite arm (30). A NTD loop termed the "lid" region closes over the ATP-bound active site. NTDs then dimerize and associate with the M-domains, prompting ATP hydrolysis (31). This step is instrumental for dissociation of the two NTDs and the subsequent release of ADP and inorganic phosphate (Pi); eventually, HSP90 returns to the open (apo) conformation.

A dynamic equilibrium exists among the different conformations of HSP90. X-ray crystallography, small-angle X-ray scattering (SAXS) solution data and kinetic measurements have led to the proposal of a general functional mechanism based on global conformational modulations triggered by ATP binding and hydrolysis, which integrates an array of structural information (25). In the absence of nucleotide, various conformations co-exist. ATP binding shifts the chaperone to

transcriptase; ERK 1/2, extracellular signal-regulated kinase 1/2; HER2, human epidermal growth factor receptor 2; ALK, anaplastic lymphoma kinase; CDK, cyclin-dependent kinase; CyP-D, cyclophilin D; Epha2, ephrin type-A receptor 2; IFIT3; Interferon induced protein with tetratricopeptide repeats 3.

a partially closed state, and then into a closed conformation; in the case of TRAP1, this is asymmetric and significantly strained, leading to buckling of the MD:CTD interface (32). The hydrolysis of ATP is sequential and deterministic and determines the conformational modulation of the MD:CTD region (32). This region has a key role in client binding (33) and is close to

completion of the cycle. Prostaglandin E synthase 3 (p23; also known as PTGES3) slows the HSP90 ATPase cycle by stabilizing the twisted state.

the allosteric C-terminal binding site. Upon ATP hydrolysis, the strain is relieved to yield a symmetric closed state (23). In vitro experiments demonstrate that although the fundamental conformational states are well-conserved among species and paralogs, equilibria and kinetics are unique for every HSP90 homolog (26), suggesting adaptations to the specific needs of clients in each subcellular environment.

In cells, HSP90 acts as a nucleating site for the assembly of networks of stable multiprotein complexes that show tumor-specific traits of physical and functional integration absent in normal cells (34, 35). Such large complexes act to enhance biochemical and metabolic pathways required to bear conditions encountered during malignant transformation. Mechanistically, co-chaperones select stochastically distributed HSP90 conformers that meet functional needs and structurally organize complexes for client activation (e.g., Cdc37 for kinases), or either increase (e.g., Aha1) or slow down (e.g., p23) ATPase rates of HSP90. Some co-chaperones, for instance Aha1 and Cdc37, are overexpressed in cancer and can be post-translationally modified by HSP90 client enzymes, generating reciprocal regulatory mechanisms of the chaperone machinery (36). By using the extraordinary power of state-of-the-art cryoEM, the Agard lab revealed the structures of two very different client HSP90 complexes, namely HSP90:Cdc37:Cdk4 (37) and HSP90:Hsp70:Hop:GR (38). Expectedly, the multiprotein functional assemblies are quite dynamic, which explains why they have eluded crystallization.

The rate of ATP hydrolysis by HSP90 is low, thus the HSP90 cycle may be differently tuned within different tissues or subcellular compartments by complex post-translational modifications (PTMs) that include phosphorylation, sumoylation, acetylation, S-nitrosylation, oxidation, and ubiquitination (20, 36, 39) (**Table 1**). We are far from understanding the effect of individual PTMs. In general, HSP90 phosphorylation, predominantly on Ser residues, but also on Thr and Tyr residues (68), slows down the chaperone conformational cycle, affecting maturation of clients and interactions with co-chaperones (20, 69). Co-chaperones broaden the functional range of HSP90, either modulating its chaperone cycle or enabling the recruitment of specific subsets of clients, thus providing a suitable folding platform for each client, or even carrying out both activities (12). Co-chaperone binding is also regulated by HSP90 acetylation, whereas S-nitrosylation in the CTD inhibit HSP90 chaperone cycle and activity (65, 68).


Aha1, HSP90 ATPase homolog 1; Cdc37, cell division cycle 37 homolog; CK1, casein kinase 1; CK2, casein kinase 2; CHIP, C-terminus of Hsp70-interacting protein; DNA-PK, DNAdependent protein kinase; eNOS, endothelial nitric oxide synthase; GSK3-β, glycogen synthase kinase beta 3; HDAC6, histone deacetylase 6; MMP-2, matrix metalloproteinase-2; Mps1, monopolar spindle-1; PKA, protein kinase A; PKCγ, protein kinase C gamma; Pnck, pregnancy-upregulated non-ubiquitous calmodulin kinase; PP5, serine/threonine-protein phosphatase 5.

Furthermore, PTMs can function as allosteric switch points that regulate interdomain communication between the two protomers (65, 69).

Despite this level of depth and sophistication in the knowledge of the roles of various players in the chaperone cycle, the factors that determine whether a protein is a HSP90 client are still elusive. HSP90 might facilitate conformational rearrangements in clients, or it might sequester them, thus avoiding their proteasomal degradation (70, 71).

Furthermore, protein quality control is a compartmentalized process characterized by peculiar features in the various subcellular regions, where different and specific networks of chaperones are present. In mitochondria, which house essential metabolic pathways, such as the tricarboxylic acid (TCA) cycle, the oxidative phosphorylation (OXPHOS) machinery and branches of amino acid, lipid, and nucleotide metabolic pathways, the paralog of the HSP90 chaperone family is TRAP1 (15, 72). TRAP1 shares the same domain structure of HSP90, but lacks a charged linker between middle and Cterminal domains and displays a long N-terminal extension called "strap" that extends between protomers in the closed state and inhibits its function at low temperatures. During TRAP1 chaperone cycle, ATP binding induces a dramatic structural change from the apo, open state to a closed, asymmetric structure, with one protomer buckled and the other one in a straight conformation (25, 73–75). The subsequent hydrolysis of the two ATP molecules bound to the TRAP1 dimer gives off the energy required for client remodeling. Hydrolysis of the first ATP swaps protomer symmetry and rearranges the client-binding site, causing structural changes in client conformation, whereas the second ATP is used to induce the formation of a compact ADP state of the chaperone, which releases the client and eventually the ADP molecules (32). Interestingly, work by the Agard lab has established that the asymmetric theme in the mechanisms of conformational dynamics is a general characteristic of the Hsp90 family (25, 32, 74). We are just beginning to understand PTM regulation of TRAP1 (**Table 2**), whereas no co-chaperones are known and the number of its known clients remains quite small.

### HSP90 CHAPERONES IN CANCER

Tumor cells are exposed to a variety of stresses that can directly hit polypeptide conformation and functionality, including unbalance in redox equilibrium caused by a profound rewiring of their metabolic circuitries and by inconstant oxygen availability (82), thus leading to a potential increase in oxidative stress. Moreover, genomic instability in a framework of relentless proliferation can lead to a high risk of synthesis of misfolded proteins. HSP90 molecular chaperones are central hubs of complex biological pathways and strongly induced by hypoxia, shortage of nutrients, high rate of DNA replication and expression of mutated proteins, thus acting at several levels to block cell death and to promote proliferation under the harsh conditions of neoplastic progression (22).

Indeed, overexpression of HSP90 has been observed in a variety of cancer types, including medulloblastoma, pancreatic, ovarian, breast, lung, and endometrial cancer, oropharyngeal squamous cell carcinoma and multiple myeloma, and high HSP90 levels are associated with poor prognosis in lung, esophageal and bladder cancer, melanoma and in several forms of leukemia (13). Most identified HSP90 clients are proteins related to biological processes dysregulated in cancer, such as signal transduction, survival, growth and invasiveness of cells and include steroid hormone receptors, both wild-type and mutant forms of the tumor suppressor p53, telomerase, hypoxia-inducible factor 1α (HIF1α) (12) and kinases, which display a continuous range of binding affinities for HSP90 (83). Some kinases would require HSP90 to stabilize their open conformation in order to efficiently bind ATP, whilst others seem to only need HSP90 for initial folding (27, 84) and would perform their enzymatic activity without HSP90 assistance. Studies with closely related pairs of client/non-client kinases, like the client v-Src and the non-client c-Src, which share 98% sequence identity, suggest that HSP90 dependence requires a combination of factors, including folding cooperativity and subtle changes in the overall stability and compactness of clients (85). HSP90 can also be secreted in a variety of tumor cells under the regulation of HIF-1α. In the extracellular matrix and on cell surfaces, HSP90


ERK1/2, extracellular signal-regulated kinases 1 and 2; NOSs, nitric oxide synthases; PINK1, PTEN-induced kinase 1; SIRT3, sirtuin 3.

decreases the tumor-suppressing effects of TGFβ and modulates cell migration and invasiveness (17), for instance by interaction with matrix metalloproteases (86–88).

Further layers of complexity exist in the interplay between HSP90 chaperones and cancer. Some HSP90 clients, like p53, may change the chaperone and co-chaperone networks by inducing the expression of co-chaperone subsets (89). The "epichaperome" is a functionally connected network of HSP70 and HSP90 machineries, which includes co-chaperones and is present in more than 50% of tumors (35), where it expands and integrates chaperone activities and promotes tumor survival (12).

Cancer cells incur high level of mitochondrial functional changes in order to maintain the required levels of ATP, reducing equivalents and metabolic intermediates, and exposure to fluctuating levels of oxygen and to high amounts of ROS can hamper proper protein folding and lead to mtDNA mutations, thus damaging mitochondrial structure and function (10, 90). Under these conditions TRAP1, the mitochondrial HSP90 paralog, could contribute to maintain an adequate quality control and to preserve mitochondrial functions. TRAP1 expression is higher in many tumors compared to surrounding nonmalignant tissues and was shown to correlate with progression, metastasis and disease recurrence in prostate and breast cancer, hepatocellular and colorectal carcinoma and non-small cell lung cancer (15, 91). In mitochondria, TRAP1 provides resistance to oxidative stress (18, 92), possibly counteracting the effects of several chemotherapeutics, and inhibits opening of the permeability transition pore (PTP), a cell deathinducing mitochondrial channel composed by the ATP synthase holoenzyme and that can be induced by a ROS surge (93). Thus, TRAP1 exerts a pro-neoplastic function by counteracting ROS-induced, PTP-mediated cell death. However, ROS effects on tumor growth are multifaceted, as oxidative stress can favor genetic instability and aggressiveness of tumors in advanced stages. Consequently, the anti-oxidant activity of TRAP1 could hamper growth in specific tumor types or stages as in cervical carcinoma, clear cell renal cell carcinoma and high-grade ovarian cancer, where TRAP1 expression inversely correlates with tumor grade (81, 94). TRAP1 also down-regulates the activity of both cytochrome c oxidase, the complex IV of the respiratory chain (81), and of succinate dehydrogenase (SDH) (95), which oxidizes succinate to fumarate at the crossroad between OXPHOS and the TCA cycle (**Figure 3**). Hence, TRAP1 participates in the metabolic switch of tumor cells toward aerobic glycolysis, i.e., decreased OXPHOS activity paralleled by enhanced glucose utilization (96). Importantly, SDH inhibition increases intracellular succinate levels, and succinate acts as an oncometabolite in several ways (97). It competitively inhibits α-ketoglutarate–dependent dioxygenases that include prolyl hydroxylases (PHDs), the JmjC domain-containing demethylases (KDMs) and the TET (10–11 translocation) family of 5 methylcytosine hydroxylases (7). PHD inhibition stabilizes the transcription factor HIF1α, increasing invasiveness, angiogenesis and further metabolic changes in tumor cells (98), whereas inhibition of KDMs, which hydroxylate lysine residue on histones, and of TETs, which induce DNA demethylation of CpG islands near gene promoters, prompts complex epigenetic rearrangements in neoplastic cells (**Figure 3**). Oncogenic kinase pathways directly target TRAP1, as it is both Tyr-phosphorylated in a Src-dependent way and Ser-phosphorylated by ERK1/2, favoring cytochrome oxidase inhibition and enhancing TRAP1 inhibition of SDH activity, respectively (78, 81), whereas Snitrosylation elicits TRAP1 degradation (76) and decreases its ATPase activity (77) (**Table 2**).

### INHIBITING HSP90 AS AN ANTI-NEOPLASTIC STRATEGY

To a superficial analysis, HSP90 would appear an unlikely target for anti-neoplastic drugs, as it is highly expressed in all cell types. However, HSP90 inhibitors tend to accumulate in tumors and are more toxic in most cancer cells than in their non-transformed counterparts (99). This could depend on HSP90 induction and/or ectopic localization in many tumor types, where it can undergo selective PTMs and can interact with a specific landscape of co-chaperones and client proteins, creating multimolecular complexes restricted to tumor cells. For instance, HSP90 binds inhibitors more strongly when it is part of epichaperome complexes (35). Some HSP90 clients express oncogenic mutations that can change their association pattern with HSP90. This can create interactions that are selectively druggable and enhance HSP90 affinity for inhibitors (100). Moreover, oncoproteins could become addicted to HSP90 in order to maintain their mutated and potentially unstable conformations, rendering disruption of this interaction particularly toxic for neoplastic cells (101).

To date, more than 50 clinical trials have been performed or are under way with several HSP90 inhibitors, but expectations of evolution toward therapeutic application have been largely frustrated. In most cases, the anti-neoplastic efficacy of HSP90 inhibition has been modest, only inducing a transient growth arrest that is reverted after drug removal, or adverse effects have been recorded, leading to termination or suspension of clinical trials. Possible reasons of these failures include a compensatory induction in the expression of other HSPs, in particular HSP70, off-target effects in patients caused by the multiplicity of biological functions regulated by HSP90 and insufficient stratification of patients enrolled in the studies (12).

### ATP-Competitive Inhibitors

Since HSP90 functions depend on its ATPase activity, most drugs have been developed as competitive inhibitors targeting the active site and competing with ATP for binding the protein. The underlying hypothesis is that disrupting the enzymatic activity of HSP90 would reverberate on the chaperone protein folding machinery, simultaneously affecting multiple oncoproteins that are essential to the proliferation and maintenance of cancer cells. The molecular basis is that ATP-ADP exchange regulates and determines a well-balanced, functionally-oriented conformational equilibrium: outcompeting the nucleotide by drug-like ligands will expectedly unbalance conformational dynamics, leading to a blockage of correct biological activities.

The unusual mode of ATP binding to HSP90 allows specific inhibition of HSP90 chaperone activity by chemical compounds. Indeed, the base and the sugar of ATP are lodged in the NTD binding pocket in a "kinked" conformation, with the phosphates pointing outwards and the γ-phosphate becoming buried only when MD and NTD associate (102). Competitive inhibitors in the ATP pocket block ATP hydrolysis and subsequently hamper the closure of the N-terminus of the dimer, thus inhibiting the HSP90 chaperone cycle (103) (**Table 3**)**.** Prototypical molecules of this class are the benzoquinone geldanamycin (GA) and the macrolide radicicol, whose selectivity has been widely discussed elsewhere (121, 122). Both compounds demonstrate strong toxicity on tumor cells (20). GA was the first HSP90 inhibitor to be evaluated as an antitumor agent. In neoplastic cells GA induces apoptosis, inhibits cell migration associated with FAK and HGF activity, as well as angiogenesis and epithelialmesenchymal transition by down-regulating VEGF receptor, HIF-1α and NF-κB signaling (123–126). However, GA preclinical TABLE 3 | Key data concerning orthosteric Hsp90 inhibitors Geldanamycin, Radicicol, and PU-H71.


<sup>a</sup>Chosen bioassays are limited to those that are directly comparable or similar in nature. Only human chaperones and cell lines have been considered. The list is merely indicative: a more complete list is available from the PubChem Database (https://pubchem.ncbi.nlm.nih.gov/compound/5288382; https://pubchem.ncbi.nlm.nih.gov/compound/radicicol; https:// pubchem.ncbi.nlm.nih.gov/compound/pu-h71 [accessed May 24, 2020]). <sup>b</sup>Residues with at least one atom detected to be closer than 3.0 Å to any atom in the ligand. Hydrogen atoms included in the analysis. <sup>c</sup>Human chaperone targets only. <sup>d</sup>Active and/or recruiting clinical trials only, as listed in the PubChem Database (https://pubchem.ncbi.nlm.nih.gov/compound/ 5288382; https://pubchem.ncbi.nlm.nih.gov/compound/radicicol; https://pubchem.ncbi.nlm.nih.gov/compound/pu-h71 [accessed May 24, 2020]). <sup>e</sup>Documented, non-chaperone human targets only. <sup>f</sup>Antiproliferation/growth inhibition of SkBr3 breast cancer cell line. <sup>g</sup>Antiproliferative activity against HCT116 cells (luminescence assay). <sup>h</sup>Hsp90 competition with fluorescent geldanamycin (fluorescence assay). <sup>i</sup>Hsp90 binding (Western Blot on MCF-7 cell lysate). <sup>j</sup>Hsp90 binding (surface plasmon resonance analysis). <sup>k</sup>Hsp90 competition with fluorescent geldanamycin (fluorescence assay on SkBr3 cell lysate).

trials were discontinued due to its hepatotoxicity, poor solubility and in vivo instability. Similarly, radicicol cannot be used as a drug as it is not stable (13, 19). Therefore, several radicicol and GA derivatives were developed (127, 128). Tanespimycin (17-N-allylamino-17-demethoxygeldanamycin, 17-AAG) is a GA derivative that induces cell cycle arrest and apoptosis in cancer cells. Tanespimycin entered a Phase III clinical trial for multiple myeloma, but its development was halted because of poor solubility and poor oral bioavailability and lapsed patent protection, with a prolonged disease stabilization in several tumor types, without any tumor regression (13). Another GA derivative, Alvespimycin (dimethylaminoethylamino-17 demethoxygeldanamycin, 17-DMAG) demonstrated anti-tumor activity and improved solubility in water, but dose-limiting side effects were recorder during various clinical trials (129, 130) (around 2000 X ray structures can be found at the RCSB protein databank (https://www.rcsb.org/). A new generation of GA derivatives is currently under evaluation, such as retaspimycin hydrochloride (IPI-504) or ganetespib. These molecules are well-tolerated and effective in cells, with reduced liver and cardiovascular toxicity (131, 132). Ganetespib (STA-9090), a small molecule inhibitor containing a triazole moiety that binds to the ATP-binding pocket of HSP90, is the most promising second generation HSP90-targeting compound (133). Its potent anti-tumor activity was translated into several clinical studies, demonstrating efficacy both in monotherapy and in combination with other drugs in various cancer types. Ganetespib produced significant single agent activity in ALK-driven cancer models, however only transient responses were reported in patients with KRAS mutant tumors due to rapid development of resistance (134). Second generation radicicol derivatives, such as NVP-AUY922 (luminespib, VER-2296) or AT13387 (Onalespib) showed strong efficacy both pre-clinically and in clinical trials, some of which are ongoing (103). Other inhibitors targeting the HSP90 ATP-binding pocket include purine analogs that can be administered orally (135), substituted resorcinols and compounds featuring a substituted benzamide substructure. Many of these molecules have entered clinical trials (136).

The nucleotide-binding pocket of the NTD is considered the most conserved structural component among HSP90 family members, impeding the rational design of paralogselective inhibitors targeting it. Therefore, a strategy used for inhibiting TRAP1 was to exploit its subcellular localization, and mitochondria-permeable GA derivatives were conceived and synthesized. Gamitrinibs (Geldanamycin mitochondrial matrix inhibitors) are 17-AAG derivatives linked to either guanidinium repeats or triphenylphosphonium (TPP), used as mitochondriotropic moieties (137). These molecules induced mitochondrial PTP opening and tumor cell death in mouse models of prostate cancer (138). Similarly, TPP was linked to the purine-scaffold Hsp90 inhibitor PU-H71PU-H71 (120) to develop a mitochondria-targeting conjugate, SMTIN-P01. Indeed, the co-crystal structures of PU-H71 in complex with either HSP90 or TRAP1 highlighted slight differences in the ATP-binding pocket of the two chaperones. The Leu172-Phe201 sequence is disordered only in the TRAP1 binding site, and the two flanking residues (Asn171 and Gly202) have different configurations (103, 139), providing a molecular basis for the differentiation of ligands to target specifically one isoform. Accordingly, TRAP1 inhibitors without mitochondrial delivery vehicles were also reported, showing a better binding to TRAP1 than to Hsp90 (139).

In spite of this wealth of efforts, so far ATP-competitive HSP90 inhibitors have not met clinical expectations and none of them has been approved for cancer treatment. All compounds showed toxicity (e.g., liver or ocular toxicity) and/or absence of convincing anticancer efficacy (13). Possible reasons include only partial inhibition of the target client proteins, P-glycoproteindependent efflux from target cells, requirement for reductive metabolism to reach full activation (17-AAG), off-target effects on biochemical pathways not specific of tumor cells (2). Moreover, the drug concentration required to outcompete ATP and induce client degradation is often the same as that needed to induce the heat shock response (HSR). HSR is based on the activation of the transcription factor HSF1, which leads to overexpression of multiple heat shock proteins, including HSP70, HSP40, and HSP27. As HSR is a survival mechanism, it can be detrimental in an anti-cancer therapy, and trying to avoid it determines dosage, toxicity and tolerance problems (136).

These limitations could be overcome if one were able to disentangle the intricacies in the functional mechanisms of different paralogs together with their relationships to metabolic and/or signaling regulation. Chemical interventions based on using ad hoc designed molecules to perturb a specific aspect of the HSP90 functional spectrum and directly report on the consequences of this perturbation would represent ideal tools. The diversity of conformations, protein-protein interactions, and functions involved in HSP90 mechanisms makes the "onedrug-fits-all" perspective unrealistic. On the other hand, that very diversity may provide a greater number of drug discovery opportunities thanks to the variety of structural and chemical motifs involved in conformational regulation and proteininteraction phenomena.

### Allosteric Inhibitors

The perspective of developing different strategies to target various aspects of the multifaceted HSP90 complexes can expectedly generate novel types of chemical intervention, unveiling new chemical tools for the investigation of biological mechanisms and/or novel candidates for therapeutic applications. In this conceptual framework, increasing appreciation is being given to the potential of allosteric modulators (**Table 4**). Allostery defines the feature of proteins to undergo a modulation of affinity toward a primary binding event caused by binding an "effector" a different distant position called the allosteric site. This modulation may cause increase or decrease of protein activity (ATP in Hsp90 processing for instance) and its downstream effects in the cell. Allosteric modulators represent an interesting opportunity for drug development in HSP90-related metabolic circuitries for several reasons: on the one hand, they permit to finely tune and regulate both the enzymatic functions and the interactions of HSP90, and on the other hand they may facilitate the selective targeting of different chaperone isoforms. The latter option would be highly desirable in the development of drugs that need to perturb the function of one specific paralog of the protein, active in specific pathologic conditions and/or in specific subcellular compartments. This type of chemical tools could selectively regulate/disrupt the functions of paralogs in a controlled way, shedding light on the correlations between the induced perturbation and the consequent biological activities, and laying the bases for novel mechanism-driven therapeutic interventions. Allostery is the prime mechanism by which achieving fine protein regulation via the activation of specific conformational states that meet functional requirements. Modifications at one site, caused for instance by ligand binding, are propagated through the protein, shifting the structural population with the modulation/perturbation of dynamic states that encode specific functions. In this context, the atomistic understanding of allosteric mechanisms provides the basis for the development of new drug candidates. In HSP90, the region at the border between the M-Domain and the C-terminal domain, located at 60Å from the ATP-site, has been shown to host a druggable allosteric site (152–156).

Experimental evidence for the possibility to target this site initially came from the Neckers group, who demonstrated the interaction between HSP90 and coumarin (140, 141). In particular, they observed that the coumarin antibiotics Novobiocin and Chlorobiocin caused the impairment of HSP90 chaperone functions by disrupting the interactions with the large group of TPR-containing co-chaperones. Importantly, despite being used at high concentrations, these molecules did not show the toxicity of ATP-competitive ligands related to the induction of the HSR (12). Novobiocin causes moderate antiproliferative effects on tumor cell models, down-regulating the expression of important HSP90-dependent clients, including Raf-1, erbB2, mutant p53, and v-Src (141). Since this seminal demonstration of the importance and druggability of alternative binding sites, significant synthetic efforts have been dedicated to improving the activities of coumarin-based allosteric molecules. In this context, the Blagg group demonstrated the possibility of substituting the carbohydrate moiety and the phenyl substituents around the coumarin scaffold by more accessible groups. The TABLE 4 | Summary of the most important compounds identified as allosteric modulators of HSP90.


(Continued)

TABLE 4 | Continued


<sup>a</sup>Expression levels of HSP90 client proteins. <sup>b</sup>Anti-proliferation assay. <sup>c</sup>Luciferase refolding. <sup>d</sup> Isothermal titration calorimetry.

coumarin was also substituted by a series of biphenyl- or bicyclic scaffolds that permitted to explore the structure-activity relationships of a large number of derivatives. This series of allosteric inhibitors showed anticancer activities reaching the mid/low nanomolar range (143, 144). However, the antineoplastic efficacy of this family of inhibitors and their molecular mechanisms of action remain unclear (103), and no carboxy-terminal inhibitor has reached clinical trials up to now (85).

We have recently used a molecular dynamics-based strategy to identify an allosteric pocket distal to the ATPase site of TRAP1, which allowed the rational design and testing of small molecule compounds that target it (157). We have also found that the same allosteric domain can host the bis-dichloroacetate ester of the vegetal derivative honokiol DCA (HDCA) (158). All these molecules inhibit TRAP1 with a high selectivity over HSP90, abolishing TRAP1-dependent down-regulation of SDH activity in cancer cells and their in vitro tumorigenic growth (157, 158). Therefore, they constitute potential leads that can be used to better dissect TRAP1 biochemical functions and to conceive novel antineoplastic approaches.

Targeting Protein-Protein Interactions (PPIs) as sources of new leads is another interesting strategy to hit HSP90 chaperone function. PPIs are often less well-conserved than active sites, making easier to achieve selectivity (159). In general, PPIs tend to modulate the activity of the interacting proteins, rather than inducing on/off effects. Thus, PPI targeting compounds could flexibly titrate chaperone activity in the context of specific cochaperone-client complexes, lowering the possibility of inducing off-target effects. However, not all PPIs are equally druggable, as the surfaces of contact usually have a larger buried surface area than enzyme-active sites, making it difficult to identify small molecules capable to block them (160).

Furthermore, as the pharmacological inhibition of clients or the downregulation of co-chaperone levels was shown to hypersensitize cells to HSP90 inhibitors, the perspective of combining drugs acting on different levels of regulation machineries is gaining increasing traction. Several compounds have been described that lead to the modulation of co-chaperone binding to HSP90. Examples include molecules like derrubone, withaferin A, and celastrol, which block CDC37 binding to HSP90. However, none of these compounds entered therapeutics so far because of insufficient efficacy in clinical studies. This could depend on several factors: variability and flexibility of HSP90 interactome in different cancer types and stages, off-target effects and HSR induction (161). Alternative interventions may involve direct targeting of PPI interfaces, disrupting HSP90-cochaperone interfaces, as recently reported (162).

### Allosteric Activators

Together with inhibition, a viable approach to controlling the metabolic implications of HSP90 and its paralogs entails the use of allosteric activators of the ATPase and of the conformational dynamics of the chaperone. If we consider the dynamic nature of the HSP90 chaperone network, whereby different HSP90 conformations are stabilized by interactions with different multi-protein complexes, allosteric activators can expectedly select/favor a subset of HSP90 structures that may have preferential binding to a selected population of interactors (**Figure 4**). Starting from original methods of molecular dynamics (MD) simulation analysis, we were able to design a series of activators capable of modifying the biochemical properties of the chaperone, as well as its activities in cells (149, 150, 163). Conjugation to mitochondrial targeting moieties showed that designed activators could show activity also on TRAP1, modifying the activities of the client SDH, with an impact on the energy metabolism of targeted cells (150). In the context of activators, Prodromou and coworkers discovered dihydropyridines able to stimulate ATPase activity. Stimulation was shown to reverberate in a compromised ability to chaperone, which

consequently induced the HSR in Alzheimer's disease cellular models (151).

The validity of the allosteric approach in studying the metabolic implications of chaperone stems from the fact that the overall activity of HSP90 is not simply shutdown but rather modulated. Ligand binding at a site far from the active site can be propagated through the protein, modifying the dynamic and/or structural populations with the modulation of dynamic states that encode specific interactions. This can in turn translate in the reshaping of surfaces that control interactions with other proteins (co-chaperones and clients), potentially favoring the binding of one partner over other alternatives. In this context, the controlled perturbation of interactions determines an overall modification of the dynamics in the protein networks that underlie signaling and metabolic pathways, providing novel intervention opportunities. Allosteric modulation of the chaperone enzymatic and conformational activities is thus directly coupled to phenotypic effects, through the finely-tuned modification of networks of PPIs.

### COUPLING HSP90 TARGETING WITH OTHER ANTI-NEOPLASTIC APPROACHES

The lack of translation to the clinical practice of HSP90 inhibitors clearly indicates the need of novel strategies to exploit cancer cell sensitivity to chaperone targeting. A promising treatment strategy is the combination therapy, in which HSP90 directed molecules are associated with other chemotherapeutics, and the combined action on different targets potently and selectively elicits the death of malignant cells (**Table 5**). Some examples already exist. A large scale phase III clinical trial (Galaxy-2) in advanced lung cancer evaluated the effects of combining ganetespib and anti-microtubule agent docetaxel in either KRAS mutant or KRAS wild type non-small cell lung cancer (NSCLC) patients. This trial failed to demonstrate any benefit in terms of progression free survival or overall survival (170). It was then demonstrated that resistance to ganetespib and to the combined treatment with docetaxel in KRAS mutant NSCLC patients was caused by hyperactivation of ERK1/2 p90RSK-mTOR signaling pathway and by circumventing the G2-M checkpoint arrest of the cell cycle (171, 172). These observations suggest that combining ganetespib with ERK1/2, p90RSK, or CDC25C inhibitors could overcome the observed resistance (131). In perspective, one would like to define the biochemical features of the tumor and the fine mechanisms of chaperone/client interaction in order to design HSP90 targeting-molecules that modulate specific signaling hubs, such as transcriptional or epigenetic regulation, maintenance of DNA integrity or bioenergetic circuitries. Such a re-shaping of HSP90 inhibition, aimed at avoiding the deleterious effects of a global damage to cellular proteostasis, would imply selecting


TABLE 5 | Preclinical and clinical studies where HSP90 inhibitors were combined with other anti-cancer drugs.

different agents for combination usage and reconsidering dosing strategies, adapting them to various cancer patient subsets (173). Novel usage options for HSP90 inhibitors are provided by the combination with targeted therapies or with immune therapy approaches, and preliminary data are encouraging [reviewed in (174)]. Combination of the multi-kinase inhibitor sorafenib with tanespimycin demonstrated efficacy in melanoma and renal cancer patients. HER2-positive, metastatic breast cancer patients treated with tanespimycin showed an improved clinical outcome. Ganetespib induced tumor regression in melanoma xenografts when supplied together with the MEK inhibitor TAK-733 and with the BRAF(V600E) inhibitor vemurafenib. Recently, HSP90 inhibitors were also tested in combination with immunotherapy, and approaches combining inhibitors of immune checkpoints and of HSP90 have been assayed. The anti-PD-L1 antibody STI-A1015 elicited higher therapeutic efficacy in melanoma and colon cancer cell models when combined with ganetespib as compared to the monotherapy regimens. In the case of TRAP1, its genetic inhibition prompts an increase in oxygen consumption rate and a decrease in extracellular acidification rate (78, 95), indicative of a metabolic rewiring toward OXPHOS coupled with a down-modulation of glycolysis. Therefore, it would be promising to identify highly selective TRAP1 inhibitors that do not affect Hsp90 activity. In principle, these compounds could be associated with OXPHOS-targeting molecules (175) in order to induce a bioenergetic catastrophe in tumor cells.

Design of new molecules, such as conjugated compounds that link established chemotherapeutic drugs with HSP90 inhibitors, may open further therapeutic windows. This strategy exploits the observation that HSP90 inhibitors highly accumulate in cancer cells, and could therefore act as cargoes for chemotherapeutics, thereby increasing their efficacy while reducing toxicity. The bifunctional compound STA-8666 consists of an HSP90 inhibitor and a topoisomerase inhibitor (SN-38) and demonstrated antitumor effects and lower systemic toxicity in preclinical studies on different cancer models (176, 177). Finally, the simultaneous inhibition of HSP90 and specific protein kinases appears to be another promising avenue to reduce drug resistance (12).

### CONCLUSIONS

Understanding the fine details that regulate the function of hub proteins as central as HSP90 and TRAP1 for biochemical and metabolic pathways is a highly challenging task. In this context, chemical biology approaches based on the use of small molecules represent attractive means to expand the reach of our investigations of the complex biology of this chaperone. Designed molecules have the potential to induce variable functional responses by inhibiting or stimulating a certain activity and thus directly focus on the molecular mechanism they are perturbing. In terms of investigation of biological processes, such data must be combined to complementary methods rooted in molecular biology such as pull-downs, genetic screens, CRISPRi, biochemical assays, proteomic and interactomic analyses, expression of mutant proteins where strategically positioned residues are modified. This integrated approach would improve our understanding of the roles of HSP90 in cancer metabolism at different levels. In a more complex scenario, regulation of chaperone networks should also be molecularly investigated during orchestrated responses of the cell to a variety of noxious stimuli such as compartmentalized unfolded protein responses.

Overall, we propose that a molecular understanding of the biology of chaperones through small molecule compounds is important for both fundamental and practical reasons. On the fundamental side, they would illuminate the determinants of biochemical mechanisms. On the practical side, selective chemical tools can expectedly be evolved into compounds that might modulate selected chaperones (and their isoforms) under specific conditions of stress. Finally, combinatorial therapies could aim at simultaneously exposing tumors to specific damaging agents while blunting the activity of protecting chaperones, thus setting the stage for the definition of innovative anti-neoplastic strategies.

### AUTHOR CONTRIBUTIONS

CS-M, GC, and AR contributed conception and design of the paper. CS-M and SS wrote sections of the manuscript and prepared figures. All authors

### REFERENCES


contributed to manuscript revision, read, and approved the submitted version.

### ACKNOWLEDGMENTS

GC and AR gratefully acknowledge the financial support of Associazione Italiana Ricerca Cancro (AIRC grant IG 2017/20019 to GC and IG 2017/20749 to AR) and of Neurofibromatosis Therapeutic Acceleration Program (NTAP). AR was also supported by grants from Piano for Life OdV and Linfa OdV.


dimer closure through a coiled-coil intermediate. Proc Natl Acad Sci USA. (2016) 113:2952–7. doi: 10.1073/pnas.1516167113


a multiscale analysis of MD simulations. PLoS Comput Biol. (2012) 8:e1002433. doi: 10.1371/journal.pcbi.1002433


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

Copyright © 2020 Sanchez-Martin, Serapian, Colombo and Rasola. 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.