Abrogating GPT2 in triple‐negative breast cancer inhibits tumor growth and promotes autophagy

Uncontrolled proliferation and altered metabolic reprogramming are hallmarks of cancer. Active glycolysis and glutaminolysis are characteristic features of these hallmarks and required for tumorigenesis. A fine balance between cancer metabolism and autophagy is a prerequisite of homeostasis within cancer cells. Here we show that glutamate pyruvate transaminase 2 (GPT2), which serves as a pivot between glycolysis and glutaminolysis, is highly upregulated in aggressive breast cancers, particularly the triple‐negative breast cancer subtype. Abrogation of this enzyme results in decreased tricarboxylic acid cycle intermediates, which promotes the rewiring of glucose carbon atoms and alterations in nutrient levels. Concordantly, loss of GPT2 results in an impairment of mechanistic target of rapamycin complex 1 activity as well as the induction of autophagy. Furthermore, in vivo xenograft studies have shown that autophagy induction correlates with decreased tumor growth and that markers of induced autophagy correlate with low GPT2 levels in patient samples. Taken together, these findings indicate that cancer cells have a close network between metabolic and nutrient sensing pathways necessary to sustain tumorigenesis and that aminotransferase reactions play an important role in maintaining this balance.


| INTRODUCTION
Breast cancer is a heterogeneous disease and despite successes in targeted therapy of some subtypes, treatment still remains a clinical challenge. 1 In particular, patients with triple-negative breast cancer (TNBC) subtype, histopathologically defined as negative for protein expression of estrogen receptor, progesterone receptor and human epidermal growth factor 2, are known to have the worst prognosis and currently lack any targeted therapy. Breast cancer like several other tumor diseases undergo global metabolic shifts in order to sustain their growth and survival. 2,3 Elucidating metabolic changes in breast cancer, therefore, provides an opportunity to potentially overcome current therapeutic challenges. Metabolic reprogramming plays a crucial role in tumorigenesis by fulfilling the biosynthetic and bioenergetic demands for rapid growth and maintaining the redox state of tumor cells. Activated aerobic glycolysis, also known as "the Warburg effect," is the most commonly reported metabolic characteristic in tumors. 4 In addition to the Warburg effect, glutamine addiction is commonly observed in several cancer types as glutamine is the second most important carbon source after glucose. This fuels the tricarboxylic acid (TCA) cycle to provide both energy and biosynthetic precursors. 5 Glutamine is broken down in the cells and distributed into different nonessential amino acids via transaminase reactions to produce α-ketoglutarate (α-KG), a key metabolite of the TCA cycle. Recent reports have shown that aminotransferase enzymes, like glutamate pyruvate transaminase 2 (GPT2), which catalyzes the reversible reaction between glucose-derived pyruvate and glutamine-derived glutamate to produce alanine and α-KG, are essential for tumor growth. 6,7 In addition to providing building blocks, metabolites can regulate signaling in cancer cells to promote cell growth. High cellular levels of certain amino acids including glutamine, can activate the mechanistic target of rapamycin complex 1 (mTORC1) which controls cell growth. [8][9][10][11] Conversely, nutrient starvation and impaired mTORC1 activity can induce self-eating in cells, a process known as macroautophagy (referred to as autophagy from here on), in order to promote cell survival. 9,12,13 Our study reveals that TNBC tumors have increased GPT2 expression compared to other subtypes. While perturbations of this aminotransferase reaction decrease TCA cycle metabolites in accordance with its role in anaplerosis, we also observed a decrease in glutamine uptake and rewiring of glucose carbon atoms in cells concomitant with abrogated alanine synthesis. Furthermore, we found that the observed metabolic changes result in an inactivation of mTORC1 as well as in the induction of autophagy both in vitro and in vivo. This connection between GPT2 and nutrient sensing pathways (autophagy and mTORC1) was also present in patients suggesting that there is a strong network between these pathways which collectively supports tumorigenesis. Loss of GPT2 inhibits tumor growth also in vivo, corroborating GPT2 as a critical enzyme in oncogenesis in TNBC by playing a crucial role in maintaining a balance between opposing nutrient sensing mechanisms.

| Patient samples
Tissue lysates from fresh-frozen breast cancer patient samples were derived from the multicenter prospective Prognostic assessment in What's new?
Glutamate pyruvate transaminase 2 (GPT2) serves a key role in glutaminolysis, a major feature of metabolic reprogramming in cancer, and is known to be upregulated in certain tumor types. In this study, GPT2 was found to be significantly upregulated in aggressive breast cancers, especially triple negative breast cancer (TNBC). Its downregulation, meanwhile, inhibited growth in triple-negative MDA-MB-468 cells. In addition, GPT2 inhibition reduced cellular glutamate uptake, an effect linked to GPT2 modulation of the mTORC1 pathway and autophagy activity. The findings highlight relationships between metabolic enzymes, nutrient signaling and autophagy in breast cancer, potentially opening up new therapeutic opportunities.
routine Application (PiA, NCT01592825) as previously described. 14 Breast cancer subtypes were defined to histopathological characteristics such as receptor status and grading according to the St Gallen classification and by von Minckwitz and colleagues. 15,16 Expression of proteins was analyzed by reverse phase protein array (RPPA).
Cells were cultured at 37 C in a humidified atmosphere and 5% CO 2 .
Both parental cell lines as well as the GPT2 knock outs have been repeatedly authenticated using single nucleotide polymorphism (SNP) profiling (Multiplexion GmbH, Friedrichshafen, Germany). All experiments were performed with mycoplasma-free cells.

| Small interfering ribonucleic acid and inhibitor treatment
Prior to small interfering ribonucleic acid (siRNA) transfections, cells were seeded to 80% confluency. After overnight cultivation, transfections were performed with RNAiMax (Invitrogen AG, Carlsbad, Kalifornien) according to the manufacturer's instructions. siRNAs (Dharmacon, Lafayette, Colorado) were used at a final concentration of 10 nM (sequences : Table S1).

| Microarray
Genome-wide gene expression profiling was performed using HumanHT-12 v4 BeadChips (Illumina, San Diego, California). Raw probe intensities were background-corrected using negative control probes and a normal + exponential (normexp) convolution model. An offset value (=16) was added to the data in order to prevent negative gene expression in the background correction step. Data were then normalized via quantile normalization using the negative and positive probes. 18 Control probes were removed and intensities log2-transformed after normalization. The German Cancer Research Center (DKFZ) microarray core facility performed sample preparation, RNA quality control and hybridization.
The lumi R package 19 was used for processing, quality control, background correction and robust spline normalization of gene expression data. The limma R package 20 was used for testing of differential gene expression. Functional analysis and gene prioritization were performed using BioInfoMiner 21 (https://bioinfominer.com) as well as by pathway enrichment analysis and visualization using the gene set enrichment analysis (GSEA) 22 module on the GenePattern platform. 23

| Immunoblotting
Protein isolates were denatured using 4× Roti Load (Roth-Chemie GmbH, Karlsruhe, Germany) for 5 minutes at 95 C and 20 μg of total protein was separated on in 12.5% SDS PAGE. Proteins were blotted onto PVDF Immobilon-P membranes (Merck, Darmstadt, Germany).
Membranes were blocked for 60 minutes at room temperature (RT) in Rockland blocking buffer, then primary antibodies (see Table S4) were incubated overnight at 4 C in Rockland blocking buffer. After three washing steps with TBS-T, secondary IRDye680 or IRDye800 conjugated antibodies (LI-COR, Lincoln, Nebraska) were diluted 1:10000 in Tris-buffered saline supplemented with Tween-20 (TBS-T) was purchased from Pierce (Thermo Scientific, Waltham, MA) and incubated with the membranes for 45 minutes at RT. Membranes were scanned and analyzed with an Odyssey scanner and Odyssey 2.1 software, respectively (LI-COR). For Western blot quantification, local background subtraction and β-actin normalization was performed.

| RPPA profiling
The method used in this study has been described previously. 24 Briefly, cell lysates were adjusted to a total protein concentration of 2 μg/μL, mixed with 4× SDS sample buffer (10% glycerol, 4% SDS, 10 mM DTT, 125 mM Tris-HCl, pH 6.8) and denatured at 95 C for 5 minutes. Lysates were spotted on nitrocellulose-coated glass slides (Grace-Biolabs, Bend, Oregon) using an Aushon 2470 contact spotter (Aushon BioSystems, Billerica, Massachusetts). Spotted slides were incubated with blocking buffer (Rockland Immunochemicals, Gilbertsville, Pennsylvania) in TBS (50%, v/v) containing 5 mM NaF and 1 mM Na 3 VO 4 for 2 hours at room temperature, prior to incubation with target-specific primary antibodies at 4 C overnight.
Primary antibodies were detected with either Alexa Fluor 680-coupled goat anti-mouse or anti-rabbit IgG, each in a 1:8000 dilution (Life Technologies, Darmstadt, Germany). In addition, every ninth slide was stained with the Fast Green FCF protein dye for total protein quantification, as described before. 24

| Intracellular extraction
One milliliter chloroform was added to 5 mL of methanolic cell extracts, shaken for 60 minutes at 4 C and centrifuged at 4149g for 15 minutes at 4 C for phase separation (methanol-chloroform-water extraction). Polar phases were collected and dried under vacuum. Cell extracts were stored at −20 C until preparation for GC-MS analysis. The extracellular extracts were stored at −20 C until preparation for GC-MS analysis. Extracellular amino acids were measured accordingly using UPLC.

| GC-MS analysis
A quantification dilution series was treated in parallel with the extracts. Derivatization was carried out as previously described with modifications. 26

| Statistical analysis and graphical illustrations
Unless otherwise mentioned, data are presented as mean ± SD. Statistical analyses were performed applying the unpaired two-tailed Student's t test and P values <.05 were considered statistically significant. P values <.05, <.01 and <.001 are indicated with one, two and three asterisks, respectively, in respective figures. All graphs were generated using the GraphPad Prism Software and illustrated via Inkscape v 1.0.1.
Additional materials and methods information can be found in Appendix S1.

| GPT2 is upregulated in TNBC
Upregulated glutamine metabolism is one of the characteristics of the TNBC subtype of breast cancer. 29,30 Along these lines, we found the enzyme GPT2 to be significantly higher expressed in TNBC tumors as compared to the other breast cancer subtypes ( Figure 1A) within a proteomic data set of a prospective multicenter cohort of 800 breast cancer patients, the PiA cohort. 14 GPT2 protein expression correlated also with tumor grade ( Figure 1B Figure 1F) thus validating the expression data from the cell line panel. 33 In addition, MCF7 cells also had lower GPT2 protein expression in comparison to MDA-MB-468, corroborating the mRNA data ( Figure 1G). Accordingly, both intracellular and extracellular alanine levels were lower in MCF7 compared to the MDA-MB-468 cell line ( Figure 1H,I). Tracing the conversion of uniformly labeled 13 C glucose revealed that label incorporation into the alanine pool was significantly higher in MDA-MB-468, further validating the above obtained data ( Figure S1D). Taken together these data suggest that upregulated alanine metabolism correlates with increased tumor aggressiveness and is particularly elevated in the triple-negative subtype of breast cancer, both in vitro and in vivo.

| GPT2 downregulation inhibits cell growth in triple-negative MDA-MB-468 cells
Next, we explored whether GPT2 has an impact on tumor cell growth.
To this end we followed three complementary approaches. First, we transiently knocked down GPT2 using a pool of siRNAs and observed a reduction in the relative numbers of MDA-MB-468 (Figure 2A,B) but not of MCF7 cells ( Figure S3A Protein analysis confirmed that the inhibitor did not affect GPT2 protein levels ( Figures 2E and S3E). Yet, cell growth was more strongly affected in MDA-MB-468 compared to MCF7 cells despite using a lower concentration of the inhibitor (Figures 2F and S3F,G). These data suggest that MDA-MB-468 cells are more dependent on the GPT2-catalyzed pathway than MCF7 cells. Finally, we knocked out  In contrast, the TCA cycle intermediates, including α-KG, fumarate and malate were strongly decreased upon GPT2 inhibition, while pyruvate and citrate did not change significantly ( Figure 3D).

| GPT2 perturbation decreases glutamine uptake and TCA cycle intermediates in TNBC cells
Decreases in TCA cycle intermediates were also seen in cell lines where GPT2 had been knocked out ( Figure S4C). Given that the α-KG that is generated in the reaction catalyzed by GPT2 directly feeds into the TCA cycle thus bypassing pyruvate and citrate, these data confirm that the observed effect on the TCA cycle is a result of blockage of the GPT2-catalyzed transaminase reaction. In contrast to MDA-MB-468, no significant changes in the uptake of the carbon sources, glucose and glutamine, and lactate secretion were observed in MCF7 cells having been treated with the GPT2 inhibitor ( Figure S4A). The TCA cycle intermediates also did not exhibit similar dramatic changes F I G U R E 2 Abrogation of GPT2 inhibits cell growth in MDA-MB-468. MDA-MB-468 cells were transfected with a pool of four siRNAs targeting GPT2 (or corresponding nontargeting control) and then GPT2 was quantitated by qRT-PCR and western-blot at RNA and protein levels, respectively, A, and for changes in relative cell numbers, B. GPT2 inhibitor effects on extracellular alanine levels, C,D, GTP2 protein expression, E, and relative cell number, F, were analyzed. G, GPT2 was knocked out using two different sgRNAs (KO1, KO2) and GPT2 protein levels and impact on extracellular alanine levels, H, as well on cell proliferation, I, assessed. Data are represented as mean ± SD, n = 3 biological replicates unless stated otherwise. **P < 0.01, ***P < .001, t test. KO, knockout; qRT-PCR, quantitative real-time polymerase chain reaction; sgRNA, single-guide ribonucleic acid is highly dependent on the GPT2-catalyzed pathway to sustain cell growth.

| GPT2 inhibition rewires glucose metabolism
Given that glucose uptake was only slightly increased upon GPT2 inhibition while lactate secretion was not elevated ( Figure 3A-C), we hypothesized that the carbon atoms from glucose might lead to a buildup of cellular pyruvate levels. This might then enter into the TCA cycle and thereby compensate for the decreased input of glutaminederived carbon atoms. As we indeed observed a trend toward increased pyruvate production upon inhibitor treatment ( Figure 3D), we next traced the conversion of u-13 C-glucose within the TCA cycle intermediates. 13   Previous studies have shown that amino acid starvation, especially of glutamine, triggers stress pathways and autophagy of cells. 8,12,34 Thus, we sought to unravel whether mTORC1 activity was compromised also in our cell systems. For this purpose, mTORC1 activity was blocked by amino acid deprivation followed by amino acid repletion at different time points. MDA-MB-468 GPT2-WT cells demonstrated F I G U R E 5 GPT2 downregulation impairs mTORC1 activity and increases autophagy in triple-negative breast cancer cells MDA-MB-468. A, Pathway analysis of differentially expressed genes from MDA-MB-468 cells transfected with siRNA targeting GPT2 or corresponding nontargeting control. B,C, Western blot analysis of phospho-p70S6K (Thr389) and total 4EBP1 protein levels (left panels), and quantification of phospho-p70S6K (Thr389) (right panels) of MDA-MB-468 WT cells and two GPT2 KO clones having been deprived of amino acid for 50 minutes followed by replenishment of amino acids for the indicated times. D, GSEA analysis of microarray data of MDA-MB-468 cells transfected with siRNA targeting GPT2 (or corresponding nontargeting control) showing core enriched distribution of macroautophagy genes, FDR = 9.8%. E, Western blot analysis of p62 protein levels (top) and quantification (bottom) in MDA-MB-468 WT and two independent clones with GPT2 KO. F, Quantification of GFP-LC3 puncta (autophagosome numbers) in WT and two independent GPT2 CRISPR/Cas9 KO clones (normalized to WT control-treated cells) (n = 2 biological replicates). Western blot analysis of P62 protein levels and, G, relative protein expression of p62 protein in MDA MB 468 WT and two independent GPT2 CRISPR/Cas9 KO clones, H, treated with Baf A1 (100 nM), torin 1 (Tor1; 250 nM) or DMSO control for 3 hours. Western blot images are representative of three independent experiments. I, Relative luminescence levels (normalized to WT control treated cells) in two independent GPT2 CRISPR/Cas9 KO clones stably transfected with a HiBiT-LC3 reporter and treated with DMSO control, starvation media (HBSS) or Torin 1 (250 nM) for 3 hours. Data are represented as mean ± SD, n = 3 biological replicates unless stated otherwise. ***P < 0.001, ** P < 0.01, t-test. Baf A1, bafilomycin A1; KO, knockout; mTORC1, mechanistic target of rapamycin complex 1. FDR, false discovery rate; GSEA, gene set enrichment analysis; HBSS, Hank's Balanced Salt Solution F I G U R E 6 Legend on next page. faster recovery after amino acid repletion compared to KO cells, as shown by the increase in phosphorylation of the mTORC1 substrate p70s6K ( Figure 5B,C) as well as an enrichment of hyperphosphorylated 4EBP1 protein ( Figures 5B,C and S6D). These data suggested that KO of GPT2 compromises mTORC1 activity.

| GPT2 downregulation compromises mTORC1 activity and induces autophagy in TNBC cells
Gene set enrichment analysis of the gene expression data sets revealed that knockdown of GPT2 also exhibited an enhanced autophagy profile in vitro ( Figure 5D). mTORC1 is one of the major autophagy regulators and its inhibition by amino acid starvation has been described to induce autophagy. 9 Thus, we next explored the status of the autophagy pathway upon GPT2 modulation. To this end, we assessed the levels of the autophagy marker p62 by western blot, as well as of autophagosome formation using the GFP-LC3 reporter in GPT2-WT and KO cells. Indeed, p62 protein levels were decreased in both GPT2 KO cell lines ( Figure 5E) while the basal number of GFP-LC3 puncta was elevated ( Figure 5F), suggesting an increase in autophagy activity in the absence of GPT2 and therefore corroborating our findings from global gene expression analysis. All cells responded similarly to autophagy modulation by starvation with HBSS or inhibition of mTOR by Tor1. The relative levels of p62 were consistently lower in the KO cells; however, this was also because the basal levels of p62 were decreased in GPT2 KO cells ( Figure 5G,H). We confirmed these data using the luminescence HiBiT-tagged LC3 reporter 35

| GPT2 KO suppresses tumor growth and induces autophagy in vivo
To assess the physiological significance of our in vitro results, we next explored the effects a loss of GPT2 has in vivo. To this end, MDA-MB-468 WT and GPT2 KO cells were orthotopically injected into NSG mice. Indeed, GPT2 KO xenografts exhibited significantly and drastically slower tumor growth ( Figures 6A and S6A) as well as reduced tumor weight ( Figures 6D and S6B) compared to the WT cells, with GPT2 KO1 having a stronger effect than GPT2 KO2 cells.
KO of GPT2 in the tumors was confirmed by western blot and this was accompanied by decreased alanine levels ( Figure 6B,E). In agreement with the in vitro data ( Figure 3D), citrate levels remained unchanged while fumarate levels showed a trend toward decreasing in GPT2 KO xenografts compared to the WT, while malate levels were not affected ( Figures 6F and S6C). Also consistent with our in vitro results, p62 levels were lower in the GPT2 KO xenografts, suggesting an induction of autophagy in GPT2 KO tumors as well ( Figure 6G).
Finally, we tested a potential correlation between GPT2 expression and autophagy in patients and quantified the levels of autophagy-related proteins in tumors of the PiA cohort. 14 Indeed, the expression level of p62 was positively correlated with GPT2 in TNBC tumors ( Figure 6H), but not in the other subtypes ( Figure S6E). As expected, the phosphorylated forms of mTOR and p70s6k proteins strongly correlated with GPT2 expression in TNBC tumors, whereas the total proteins did not ( Figure 6H). Interestingly, the positive correlation of GPT2 expression we observed with phosphorylated mTOR as well as p70S6K proteins was even stronger in the other subtypes of breast cancer, suggesting that the link between mTOR signaling and GPT2 could be relevant in breast cancer in general. These findings thus substantiate the connection between mTORC1 activity and glutaminolysis in breast cancer, while that between GPT2 and autophagy might be specific for the TNBC subtype ( Figure S6E). To our knowledge, this is the first report (a) providing quantitative analysis of GPT2 in breast cancer patients at the protein level; (b) presenting in vivo data from CRISPR GPT2 KO cells; and (c) uncovering the correlation between GPT2 and autophagy as well as mTOC1 activity in vitro, in vivo and in patients.

| DISCUSSION
Metabolic reprogramming is an essential part of tumorigenesis, which is prominently characterized by the Warburg effect and glutamine anaplerosis. 2,3,36 Glutamine-dependent growth is an important feature of several aggressive cancers including the TNBC subtype. 2,3,36,37 In our study, we found that GPT2, which is part of a network of enzymes involved in glutaminolysis, is significantly upregulated in aggressive breast cancers, especially in the triple-negative subtype. Its expression negatively correlates with overall patient survival. Of note, this is the first study showing GPT2 expression not only at the mRNA but also at  14 Data are represented as mean ± SD, n = 4 biological replicates unless stated otherwise. ***P < .001 t test. Correlation coefficients (r) and significance were calculated using Pearson's correlation test. GC-MS, gas chromatography-mass spectrometry; PiA, Prognostic assessment in routine Application; UPLC, ultra-performance liquid chromatography; TNBC, triple-negative breast cancer the protein levels. These clinical findings corroborate our in vitro data showing that perturbation of GPT2 expression in triple-negative MDA-MB-468 cells slows down cell growth. Blocking the alanine aminotransferase reaction results in a reduced glutamine uptake. This is in line with previous studies having shown that the alanine aminotransferase reaction positively correlates with glutamine uptake. 6,7,38 Cancer cells utilize aminotransferase reactions to feed glutaminederived carbons into the TCA cycle via α-KG along a major route of glutamine catabolism. Here, we have observed that inhibition of GPT2 results in reduced α-KG pools and that this reduction is concurrent with a decrease in the pool of TCA cycle intermediates downstream of α-KG, thus confirming this transamination reaction to be an important anaplerotic route. 6,7,38 We cannot completely rule out targeting also of other enzymatic reactions by the chemical inhibitor we applied. 39 However, the strong overlap of metabolic and phenotypic effects we observed upon inhibition and knockdown/KO of GPT2 suggests that, indeed, the main activity of β-chloro-L-alanine seems to be on the inhibition of the alanine aminotransferase reaction that is catalyzed by GPT2 in our system. This conclusion is supported by the finding that GPT1, a paralog of GPT2 and a second target of β-chloro-L-alanine, is neither expressed in the MDA-MB-468 cell line nor in other breast cancer cell lines at similar levels as GPT2.
GPT2 is a pivotal enzyme linking glutamine catabolism and glycolysis. 6 However, the fate of carbon atoms in glucose has not been explored thus far in conditions of abrogated GPT2 activity. Our data illustrate that inhibition of GPT2 results in an increased carbon flow from glucose into the TCA cycle possibly to compensate for the reduction in TCA cycle metabolites. We also show that despite having a weaker impact on luminal cell line MCF7, GPT2 inhibition induces similar effects on glucose rewiring also in these cells. Interestingly, we found that the TCA cycle intermediates deriving from pyruvate carboxylation were significantly increased upon GPT2 inhibitor treatment in MDA-MB-468 cells suggesting that PC activity is elevated upon GPT2 inhibition. Furthermore, combinatorial targeting of GPT2 and PC had an even stronger effect on cell growth compared to the individual treatments. PC has been shown to play a key role in anaplerosis under glutamine starvation conditions, 40 and another study showed that lung metastases from breast cancer switch from the glutamine to PC anaplerotic routes in response to changes in the tumor microenvironment. 41 These studies support our finding that tumor cells tend to upregulate pyruvate carboxylation upon reduced glutamine uptake, potentially to support biosynthetic processes. However, our data suggest that this mechanism is not sufficient to maintain cellular homeostasis. Nutrient starvation is counteracted by different cellular mechanisms, one major process being autophagy (self-eating) where the cells undergo partial proteolysis enabling them to replenish their pool of amino acids. 12,42,43 However, only few amino acids can trigger autophagy, including glutamine. [8][9][10][11] Our data illustrate this connection as we found an enrichment of autophagy pathway genes and an induction of autophagosome formation in GPT2 KO cells. Amino acid deprivation also leads to inactivation of mTORC1 thereby triggering autophagy. [9][10][11]13 Consistent with this, we found that KO of GPT2 results in decreased mTORC1 activity as seen by the reduced phosphorylation levels of downstream effectors of the pathway, p70s6k and 4EBP1. 44,45 Our in vitro findings were substantiated upon in vivo testing, as growth of xenograft tumors was massively impeded in the absence of GPT2. The even stronger growth impairment seen in vivo might be due to several factors including longer duration of the experiment where cells undergo many more rounds of replication. Furthermore, glutamine could become a limiting factor in vivo as physiological levels are lower than in cell culture. 10,46 More research is needed to investige these and potential other factors affecting tumor growth in conditions where GPT2 activity is abrogated. The observed reduction in p62 levels in the KO tumors indicates that autophagy activation likely occurs as an adaptation to the loss of GPT2 also in vivo. These results are further supported by patient data where we could show a strong correlation between GPT2 protein levels and the autophagy and mTORC1 pathways specifically in TNBCs, while the link between GPT2 and mTORC1 seems to be relevant even across all subtypes of breast cancer. Taken together our results demonstrate that there is a robust network connecting mTORC1, autophagy and glutaminolytic pathways with GPT2 alanine aminotransferase activity that likely helps to maintain amino acid levels in TNBC cancer and thereby promotes proliferation as well as sustains cell growth.
In conclusion, our findings show that GPT2 has a prominent function particularly in TNBC, the most aggressive subtype lacking any targeted therapy. While its role in aiding glutamine-dependent anaplerosis has been mostly established, our study demonstrates that perturbing this pathway leads to the rewiring of glucose utilization within tumor cells that involves the activation of an alternative anaplerotic pathway catalyzed by pyruvate carboxylase. Combinatorial targeting of PC with glutamine metabolism inhibitors might thus be an interesting strategy for the treatment of breast cancer similar to glioblastoma cells where pyruvate carboxylase has been shown as an alternative route for carbon delivery to the TCA cycle under conditions of glutamine starvation and has been discussed as a route for resistance to inhibition of glutaminolysis. 40 Our study shows for the first time that GPT2 modulates the counterbalancing forces of autophagy and mTORC1 activity not only in vitro, but also in vivo and in patients. Higher GPT2 levels in patients correlate with markers of inhibited autophagy and poorer outcome and it is plausible to hypothesize that new treatments for TNBC patients with elevated GPT2 might include modulators of the GPT2-mTORC1-autophagy pathway.