Glutamine-Directed Migration of Cancer-Activated Fibroblasts Facilitates Epithelial Tumor Invasion

Cancer invasion is a complex process in which tumor cells interact with the extracellular matrix and create migration tracks that are used for collective or single-cell movement (1). This process is also influenced by the tumor microenvironment (TME), a blend of different types of cells activated by the tumor and required for the acquisition of many cancer hallmarks (2). Among the components of the TME, a role for the cancer-activated fibroblasts (CAF) supporting invasion of epithelial cells has been well-stablished (3–9). CAFs are activated by signals derived from the tumor, such as TGFβ or platelet-derived growth factor; this activation requires the expression of Snail1 transcriptional factor (8–11). Accordingly, alike other mesenchymal cells such as mouse embryonic fibroblasts (MEF) or mesenchymal stem cells (MSC), CAFs depleted in Snail1 cannot be activated: upon stimulation with TGFβ, they do not express specific markers of activated fibroblasts and do not enhance epithelial cell invasion (9–11).

Glutamine (Gln) is the most abundant amino acid in plasma and the major carrier of nitrogen between organs (12). Because of its high utilization by tumor cells, a drop in Gln concentration compared with the healthy tissue has been observed in different neoplasms. Since the initial report published more than 70 years ago (13) several groups have observed this Gln reduction in different tumors (as recent publications; see refs. 14, 15). Moreover, decreased Gln levels have been detected in the central part of melanoma and breast-engrafted tumors when compared with their periphery (16). Tumor cells utilize Gln for ATP production and for providing intermediates for macromolecular synthesis; it contributes to the supply not only of carbon, but also of nitrogen, required for the biosynthesis of compounds such as nucleotides, glucosamine 6-phosphate, and nonessential amino acids (12, 17–18). Moreover, it also contributes to cell protection through the synthesis of glutathione and NADPH (12). Most of Gln effects are dependent on its conversion to glutamate (Glu) and ammonia by glutaminase (GLS). Glu is then shuttled to the tricarboxylic acid cycle through the generation of α-ketoglutarate by glutamate dehydrogenases and converted to various nonessential amino acids via specific amino transferases (12, 17–18). High glutaminolysis and Gln “addiction” are particularly relevant in many aggressive forms of human cancers and are stimulated by c-Myc and K-ras oncogenes (19, 20).

We have isolated different CAFs and epithelial tumor cell lines (ePyMT) from breast tumors obtained from MMTV-PyMT transgenic mice (21). In this murine model, Snail1 is detected in CAFs and its depletion decreases tumor development and invasion (9). When comparing the nutrient needs of different ePyMT, we observed that mesenchymal cells present a much higher requirement for Gln than epithelial cells. This higher Gln dependence is also detected when CAFs and other mesenchymal cells are compared with epithelial tumor cell lines. Moreover, this Gln requirement promotes both migration and invasion of CAFs and other mesenchymal cells following a Gln gradient and support epithelial tumor cell invasion in the same conditions.

Authenticated cell lines were obtained from the European Collection of Authenticated Cell Cultures or the ATCC and supplied by the Cancer Cell Line Repository from IMIM. All cell lines were used for no more than 20 passages and routinely tested by PCR to verify that they remained Mycoplasma-free. Cells were grown in DMEM (Gibco) supplemented with 4.5 g/L glucose, 1 mmol/L sodium pyruvate, 2 mmol/L Gln, 100 IU/mL penicillin, 100 mg/L streptomycin, and 10% FBS (Gibco) and maintained at 37°C in a humid atmosphere containing 5% CO₂. All the experiments were performed on cells incubated for at least 24 hours in DMEM 0.5% FBS, 0.1 or 1 g/L glucose (Gibco), and 0.2 or 2 mmol/L Gln. The generation of AT3 from MMTV-PyMT murine tumors (21, 22) has been reported (23); BTE136 cells were isolated from these tumors after microdissection and digestion with collagenase P (1 mg/mL) on a gentle MACS Dissociator (Miltenyi Biotec). Dispersed cells were centrifuged, washed with medium with FBS, filtered (70 μmol/L and 40 μmol/L), and immediately seeded in p-100 plates in DMEM, plus 10% FBS, the abovementioned reagents, gentamycin (50 μg/mL), and fungizone (0.5 μg/mL). Cells were passaged 20 times before being transferred to the standard culture medium. The generation of HT29 M6 KO for Snail1 gene using CRISPR-Cas9 technology has been previously reported (9). MSC, either Snail1^+/− (considered Snail1 wild-type) or Snail1^−/− (Snail1 KO) were previously established in our laboratory from the Snail1^Flox/− mice (11) by retroviral transduction of a plasmid encoding the Cre recombinase or a control vector. Wild-type, Akt1 KO, and Akt2 KO MEF were kindly provided by Dr. Morris Birnbaum (University of Pennsylvania, Philadelphia, PA). The generation of CAFs (CAF1875) from breast tumors from MMTV-PyMT mice has been previously reported as well as the depletion of Snail1 in these cells (9). The production of transfectant cells is described in the Supplementary Data.

The viability of tumor and fibroblastic cell lines in glucose and Gln-depleted cultures was determined by crystal violet staining. Fifty (tumor cells) or ten (fibroblasts) thousand cells per well were seeded in 24 multi-well plates in complete growth medium and incubated for 24 hours to enable cell adhesion. Cells were then washed with PBS twice and culture medium was replaced for DMEM 1 mmol/L sodium pyruvate, 0.5% FBS, with either high (1 g/L) or low (0.1 g/L) glucose and high (2 mmol/L) or low (0.2 mmol/L) Gln supplementation. Cell viability was assessed after 48 hours postdeprivation by crystal violet staining. Sensitivity to GLS1 inhibition was assessed using the same procedure and incubating the cells with the indicated doses of CB-839 (Cayman Chemical Company) or dimethyl sulfoxide (DMSO) for 48 hours. The relative half-maximal inhibitory response (IC₅₀) of CB-839 was calculated for each cell line by nonlinear curve fits using GraphPad software.

Transwells (3422, Costar) were coated with 50 μL of Matrigel (0.5 μg/μL; 354230, Corning) for invasion or left uncoated for migration assays and incubated for 1 hour at 37°C. For invasion, 1 × 10⁵ epithelial cells or 2 × 10⁴ fibroblasts were seeded on Matrigel-coated Transwells in DMEM plus FBS (0.5%) and the indicated concentration of Gln in a final volume of 150 μL. In coculture experiments, RFP- or dTomato-labeled epithelial and GFP-labeled fibroblasts were used instead. After 4 hours at 37°C, the bottom chamber was filled with the control medium, always in low FBS (0.5%) unless otherwise specified. When indicated, a medium with a lower concentration of glucose (0.1 g/L instead of 1 g/L utilized as standard) was used in the top chamber. In other experiments, the bottom chamber was supplemented with lactate (20 mmol/L) or FBS (10%). Invasion was stopped at 48 hours (epithelial cells) or 12 hours (fibroblasts); cells were washed with PBS and fixed with p-formaldehyde (PFA; 4%) for 20 minutes. Cells at the top side of the Transwell membrane were removed with a cotton swab and the membrane with the invading cells was stained with DAPI and mounted for microscopy analysis. Five random photos (10×) of each membrane were taken to analyze the area of invasion. Only RFP- or dTomato-, GFP-labeled cells were analyzed in coculture experiments. Quantification of invading cells was performed with ImageJ software. The differences in Gln concentration between the top and the bottom compartments were detected at least for 6 hours (Supplementary Table S1).

Transwell inserts were coated with 50 μL of a dense matrix composed by 3 μg/μL Matrigel (354230, Corning) and 2 μg/μL collagen I (354249, Corning) and polymerized at 37°C for 1 hours. CAFs (5 × 10⁴) were seeded on top. Upon cell adhesion, culture media was removed and cells were covered with 50 μL of the same matrix. Following matrix polymerization, media were added to the top and bottom compartments and invasion was assessed in high Gln (both compartments filled with DMEM 1 g/L glucose, 0.5% FBS, 2 mmol/L glutamine) or a Gln gradient (same medium with 0.2 mmol/L Gln in the top compartment). TGFβ was added in both compartments. Media and TGFβ treatment were refreshed every 24 hours. After 3 days, the cultures were fixed with PFA (4%) and processed by standard methods for histology analysis. Matrices were mounted in paraffin blocks in transversal orientation and sections were made at 3 μm and stained with hematoxylin and eosin.

Cells were seeded in DMEM 1 g/L glucose, 0.5% FBS, 2 mmol/L (high), or 0.2 mmol/L (low) Gln at a final density of 3–5 × 10⁶ cells/mL in the central compartment of chemotaxis μ-slides (80326, IBIDI), according to manufacturer's instructions. After an overnight incubation to allow cell adhesion, the two adjacent reservoirs were filled with DMEM with either high or low Gln, and Gln gradients were generated by the addition of 30 μL of DMEM 4 mmol/L Gln in one of the low Gln reservoirs. In this device, the Gln gradient was stable for 24 hours (Supplementary Table S1). TGFβ was added to all the compartments of the device when specified. Right after generating the gradient, cell migration was recorded by life cell imaging in a Zeiss cell observer microscope, which took pictures in different random areas every 5 minutes for 12 hours. Single-cell coordinates at each time point were tracked by the ImageJ Manual-Tracking plugin. The average speed of migration was calculated as the ratio of distance migrated (μm) per min.

Alternatively, slides were fixed with PFA (4%) and analyzed by immunofluorescence after 6 hours of migration. To maintain the integrity of the culture, all the following reagents were introduced to the device by the two lateral reservoirs. All the procedure was performed at room temperature. Cells were permeabilized with PBS plus 0.1% Triton X-100 for 15 minutes. Following two PBS washes, samples were blocked with 3% BSA in PBS for 2 hours and incubated with the primary antibody for 90 minutes. Slides were washed with PBS twice and incubated with fluorescent secondary antibodies (Alexa 488 and 555-conjugated anti-rabbit or mouse IgGs; Thermo Scientific) for 1 hour. After two more PBS washes, samples were stained with DAPI (25 μg/mL) for 15 minutes and mounted with Fluoromount G (Southern Biotech). Random pictures of the migration front were taken in a Leica TCS-SP5 confocal microscope.

One-hundred microliters of AT3 cells (5 × 10⁶) plus MEF (5 × 10⁵) when indicated were diluted in Matrigel (1:1) and subcutaneously injected in SCID mice. When tumors reached 0.5 cm³, we implanted two pellets of the polycaprolactone-polyurethane polymer Actifit (Orteq; 4 × 4 × 3 mm) soaked in Gln (20 mmol/L) or PBS at opposing sides of the tumor. Animals were euthanized three days later. The position of each polymer within the tumor was labeled with green (+Gln) or yellow (+PBS) ink; polymers were removed and samples processed for histochemical analysis. Hematoxylin and eosin–stained tumor slides were scanned with Aperio CS2 Scan (ScanScope) and quantified with QuPath software (https://qupath.github.io/). For each condition, tumors were divided in two and the percentage of area occupied by muscle and adipocytes trapped inside the tumor was calculated. These experiments were approved by the Ethical Committee for Animal Research from the PRBB and Generalitat de Catalunya.

Human breast tumor samples were obtained from Parc de Salut MAR Biobank (MARBiobanc, Barcelona, Spain). Their study was performed according to declaration of Helsinki guidelines and approved by the Ethical Committee for Clinical Research from PRBB and patients gave written informed consent to the use of their samples. Samples were fixed with PFA (4%) at room temperature, dehydrated, and paraffin-embedded according standard procedures. Sections (2.5 μm) were prepared and, after standard deparaffination and rehydration, antigen unmasking was carried immersing the sections in Tris EDTA buffer pH 9 and boiling for 15 minutes. Samples were blocked during 2 hours in Tris-buffered saline (TBS) plus FBS (1%) and BSA (1%), and incubated with primary antibodies overnight (TRAF6 1/40; Akt2, 1/100). Signal was amplified with EnVision+ System HRP Labeled Polymer (anti-rabbit, DAKO) and visualized with the DAB kit (DAKO). Invasive and noninvasive areas of the tumor were determined by an expert pathologist and microphotographed with Olympus BX61 microscope at IMIM Microscopy Unit. Five random stromal zones of noninvasive areas or the invasive front were analyzed. Stromal cells presenting a cellular asymmetrical distribution of TRAF6 or Akt2 staining with respect to the cell nucleus were identified and designated as polarized cells. Graphs shown the density of those cells, represented as the number of polarized cells per area (px²) of stroma. Statistical significance was determined by Student t test using GraphPad v6 software.

Other methods are described in the Supplementary Information.

Gln concentration is reduced in different tumors (14–15). We confirmed that MMTV-PyMT murine breast tumors (21–22) show a lower ratio between Gln and Glu or other metabolites in tumors than in plasma (Supplementary Fig. S1A). We determined the Gln sensitivity of two tumor cell lines obtained from these MMTV-PyMT breast tumors. These two lines present a different morphology being AT3 more epithelial than BTE136 and expressing higher levels of E-cadherin and CK14 (Fig. 1A). In this analysis, we maintained both cell types in a low percentage of serum (0.5%) to eliminate the effects of growth, and in 1 g/L of glucose, to more adequately mimic its physiologic concentration in serum. AT3 was much more resistant than BTE136 to a decrease of Gln from 2 mmol/L, the standard concentration in cell culture medium, to 0.2 mmol/L (Fig. 1B, top, first and second columns). In contrast, the sensitivity to a drop in glucose from 1 g/L to 0.1 g/L was similar in both cell lines (Fig. 1B, low). CAFs obtained from the same tumors also displayed a high sensitivity to Gln deprivation, an effect that was not altered by Snail1-deletion or by incubation with TGFβ (Fig. 1B, top).

We extended our study to other epithelial tumor cells and mesenchymal cells, such as MSCs, because these cells contribute to the formation of CAFs (4, 24, 25). MSCs and MEFs were also dependent on Gln for surviving, whereas tumor cells were much less affected by a decrease in the concentration of this amino acid (Fig. 1B, top). Depletion of Snail1 in CAFs (Fig. 1A) did not significantly affect the sensitivity to Gln. In accordance with the Gln requirement of CAFs and not of epithelial tumor cells, only CAFs underwent apoptosis in Gln-low medium as determined by analyzing the levels of cleaved caspase-3 (Fig. 1C). A higher apoptosis was also detected in low Gln when comparing mesenchymal BTE136 with epithelial AT3 PyMT cells. The higher sensitivity to Gln deprivation was not associated to lower glutamine synthase expression in Gln-low conditions (Fig. 1C).

In the conditions used in these assays (0.5% of FBS; 1 g/L of glucose), CAFs minimally proliferate in medium containing 2 mmol/L Gln; a drop of Gln until 0.5 mmol/L slightly decreased cell number that was severely affected in 0.2 or 0 mmol/L Gln (Supplementary Fig. S1B). In contrast, only a total depletion of Gln compromised viability of epithelial HT29 M6 cells. Because most Gln is metabolized to Glu by glutaminase1 (GLS), we checked the effect of CB-839, a widely used selective inhibitor of GLS (26), on CAFs viability. Addition of CB-839 severely decreased the viability of CAFs or BTE136 PyMT tumor cells, whereas it did not significantly alter that of HT29 M6 or AT3 PyMT, both tumor cells exhibiting an epithelial phenotype (Fig. 1D).

Because CAFs present a high dependence on Gln, we reasoned that when challenged with a Gln concentration-gradient, they might migrate toward this amino acid. To examine this, cells were added to the top compartment of a Boyden Chamber in DMEM plus 0.5% FBS containing 1 g/L of glucose and 0.2 mmol/L (Gln-low) or 2 mmol/L Gln (Gln-high). Bottom chamber was filled with the same medium but with 2 mmol/L Gln. As shown in Fig. 2A, CAFs migrated to a higher extent when the top chamber was Gln-low and only when they were stimulated with TGFβ, suggesting that migration requires the previous fibroblast activation. These results were also reproduced in MSCs (Fig. 2B). Snail1 depletion in CAFs and MSCs prevented TGFβ stimulation of migration (Fig. 2A and B), in accordance with the required function of Snail1 in fibroblast activation. Gln-driven migration was also assessed in tumor cell lines. Only mesenchymal BTE136 tumor cells migrated better following the Gln gradient than in high Gln; the rest of the tumor cell lines analyzed, including AT3 cells, did not show any significant difference between the two conditions (Supplementary Fig. S2A and S2B).

MSC migration was also determined by time-lapse microscopy using chemotaxis μ-slides (IBIDI). Cells were seeded in the central chamber of these slides in the presence of TGFβ and exposed to a Gln gradient, using two different concentrations of Gln (0.2 and 2 mmol/L) in the two lateral chambers. As shown in Fig. 2C and Supplementary Video S1, MSC preferentially moved toward the Gln-high medium when exposed to the gradient; migration was less directional when they were grown on high Gln (Supplementary Video S2). The number of cells that migrated to the high-Gln chamber was higher when cells were subjected to the Gln gradient than when grown in high Gln (Fig. 2D; quantification in Fig. 2E). The absolute migration speed was slightly higher in cells moving in the Gln gradient than in high Gln (Fig. 2F).

Fibroblasts also invade Matrigel in response to changes in Gln concentration. CAFs moved toward the Gln-high lower compartment when the top compartment was Gln-low; this effect required stimulation with TGFβ (Fig. 2G). Similar results were obtained with MSCs (Fig. 2H). As in the migration assays, elimination of Snail1 in MSCs or CAFs prevented Gln-dependent Matrigel invasion (Fig. 2G and H). This directional invasion was not observed in a glucose gradient when we used a 10-fold lower concentration of glucose (0.1 g/L) in the top than in the bottom chamber. Lactate supplementation in the bottom chamber did not enhance invasion either (Fig. 2G).

CAF invasion was also evaluated when these cells were seeded between two layers of matrix (Fig. 2I). When Gln was high in the two compartments, cells invaded the matrix in both directions, both as single cells and in small aggregates. The same morphology was observed in Gln gradients but in this case cells moved only toward the high Gln compartment. As expected, because it requires matrix degradation, Gln-driven CAF invasion was prevented by incubation with GM6001 (GM), a general metalloprotease inhibitor (Supplementary Fig. S3A).

We determined whether the capability of mesenchymal cells to invade toward Gln might be transferred to epithelial tumor cells when both cells are cocultured. This has been reported for HT29 M6 invasion when directed by a gradient in FBS: coculture with CAFs or MSCs increased HT29 M6 invasion (see also Supplementary Fig. S3B; ref. 9). For these studies, we specifically labeled tumor cells with RFP and assessed invasion in the cocultures only of RFP-positive cells. As in previous assays, FBS concentration was maintained at 0.5% in both compartments. As shown in Fig. 3A and B, invasion of HT29 M6 was not significantly different in high Gln or in a Gln-gradient, but it was remarkably increased by coculture with CAFs and MSCs only in Gln gradients. The stimulation by MSCs was similar to that we have previously reported when invasion was driven by an FBS gradient (Supplementary Fig. S3B, last two bars). A similar result was obtained when migration was determined because HT29 M6 migration in a Gln gradient was also increased by coculture with MSCs (see Supplementary Fig. S2B). MSCs did not enhance HT29 M6 invasion when the coculture was exposed to glucose or lactate gradients (from 0.1 to 1 g/L or from 0 to 20 mmol/L, respectively; Supplementary Fig. S3B).

We also analyzed other tumor cell lines (AT3 and BTE136 PyMT cells) and four other breast or colon tumor cells (T47D, MCF7, SW480, and HCT116). The more mesenchymal BTE136 cells invaded following the Gln gradient (Fig. 3C, bars 1 and 2) in contrast to AT3 (Fig. 3C, bars 3 and 5) and in accordance with their higher Gln dependence (see Fig. 1). Other epithelial tumor cells did not show Gln-driven invasion (Supplementary Fig. S3C). However, invasion of AT3 and the rest of the cell lines were stimulated by coculture with CAFs or MSCs (Fig. 3C; Supplementary Fig. S3C). Cocultures of AT3 with CAFs, or SW480 and T47D tumor cells with MSCs invaded better in Gln gradients than when Gln was maintained high in both compartments. Invasion of MCF7 and HCT116 cells was stimulated similarly by MSC in both conditions.

We also checked the relevance of the TGFβ/Snail1 axis in tumor invasion. The broadly used TGFβ receptor inhibitor SB505124 (SB) significantly decreased the action of MSCs on HT29 M6 invasion, although it did not affect the basal HT29 M6 (Fig. 3B). Snail1 depletion in MSCs and in CAFs promoted a similar inhibition (Fig. 3A and B). HT29 M6 invasion was also blocked by the metalloprotease inhibitor GM (Supplementary Fig. S3D).

The sensitivity of the coculture to TGFβ inhibitors and Snail1 expression in MSCs and CAFs suggested that these cells were activated by HT29 M6, as we have reported previously (9). We specifically analyzed MSC invasion in cocultures using GFP-labeled MSCs. The presence of HT29 M6 tumor cells increased MSC invasion; this stimulation was higher when cells were exposed to a Gln gradient (Supplementary Fig. S3E). The TGFβ receptor inhibitor SB decreased the HT29 M6-dependent enhancement of MSC invasion that was also blocked by GM inhibitor (Supplementary Fig. S3E).

We also determined whether the invasion in cocultures was associated to an epithelial-to-mesenchymal transition (EMT) of tumor cells. Taking advantage of the different origin of CAFs (murine) and HT29 M6 (human), we analyzed the expression in these cells of genes related to EMT: no differences were detected in the expression of CDH1, SNAI1, SNAI2, TWIST1, or ZEB1 in HT29 M6 cultured in high or low Gln, either in the presence of MSCs or not (Supplementary Fig. S4A and S4B). The lack of requirement for Snail1 in HT29 M6 cells was also verified using CRISPR/Cas9-edited Snail1 KO HT29 M6 cells. In the presence of MSCs, these tumor cells responded to the Gln gradient identically than control cells (Supplementary Fig. S4C).

We set an animal tumor model to study the physiologic relevance of Gln-driven invasion. Tumors were generated by subcutaneous grafting of AT3 cells; when they were initially detected, before reaching the endpoint size, two inert pellets were implanted at opposed sides of the tumor. The animals were euthanized three days later and the tumors were studied. Invasion was much higher toward the Gln-soaked pellet (labeled with green ink) than to the control, PBS-treated pellet (labeled in yellow; Fig. 3D and E). Coxenografting of AT3 cells with MEFs increased invasion at both sides but the Gln-high margin always presented greater cell infiltration (Fig. 3D and E).

We also studied the molecular basis underlying the directional migration of the activated fibroblasts toward Gln. First, we analyzed the activation of CAFs by TGFβ in low-Gln or high-Gln medium. TGFβ triggered rapid responses in CAFs such as Smad2 phosphorylation, Snail1 protein upregulation, and transcription of several CAF genes (Supplementary Fig. S5A–S5D). Some of these responses, such as transcription of USP27X, Serpine1 or Adamts16, S6 phosphorylation, and synthesis of Snail1 and other late CAF markers were not observed in low Gln; thus, CAFs were not properly activated by TGFβ in low Gln medium. Titration of the Gln concentration required for this activation indicated that 0.5 mmol/L was sufficient for TGFβ to promote an upregulation in Snail1 protein and Serpine1 and Adamts16 transcription as efficient as in 2 mmol/L Gln (Supplementary Fig. S5B and S5D).

Snail1 function and CAF activation require Akt (11). Accordingly, TGFβ-stimulated CAF invasion was potently inhibited by the Akt antagonist MK-2206 (MK; Supplementary Fig. S6A; ref. 27). Moreover, Akt is activated in the leading edge of migrating cells (28). We analyzed the distribution of Akt isoforms in cells migrating in a Gln gradient. A significant percentage of CAFs moving toward Gln showed a preferential localization of Akt2 in the leading front of these cells; a representative image of these cells is shown in Fig. 4A. This asymmetrical Akt2 distribution was not observed in cells migrating in high Gln conditions even though they were stimulated by TGFβ (Fig. 4A). CAF activation was also required because Akt2 was not polarized in cells without TGFβ or lacking Snail1 (Fig. 4A). Akt2 localization was not affected by GLS inhibition (Supplementary Fig. S6B), suggesting that it was not dependent on Gln-derived metabolites. Akt1 did not display a similar asymmetrical intracellular localization and neither mTOR kinase did (Fig. 4B). Cells with polarized Akt2 distribution were also observed in MSCs and MEFs activated by TGFβ in Gln gradients (Fig. 4C). In contrast to AT3 epithelial tumor cells, BTE136 cells displayed asymmetrical Akt2 correlating with their capability to migrate following a Gln gradient (Supplementary Fig. S6C). The polarized distribution of Akt2 in MEFs also correlated with a preferential distribution of αSMA in the leading edge of migrating cells (Fig. 4D).

Genetic elimination of Akt2 (Fig. 5A) slightly decreased Snail1 protein upregulation caused by TGFβ (Fig. 5B). In contrast, MEF Akt2 KO did not accumulate αSMA in the leading front (Supplementary Fig. S7A). Remarkably, depletion of Akt2 (and not of Akt1) prevented the increase in migration caused by TGFβ (Fig. 5C) demonstrating that Akt2 works downstream of Snail1. This deficiency in cell migration detected in MEF Akt2 KO was rescued by ectopic transfection of full-length Akt2 and not by an Akt2 mutant lacking the pleckstrin homology-domain (Akt2-ΔPH; Fig. 5C). The ectopic full-length Akt2 also showed an asymmetrical distribution when expressed in MEF Akt2 KO, in contrast to Akt2-ΔPH, which was mainly nuclear (Supplementary Fig. S7B).

The effect of Gln depletion on Akt2 activity was analyzed. We did not detect a significant upregulation of Akt2 Ser474 phosphorylation in CAF upon TGFβ stimulation; decreasing Gln concentration up to 0.2 mmol/L did not affect this parameter either (Fig. 5D). Akt2 activity was determined in CAFs assessing the phosphorylation of its substrate GSK3α by the immunoprecipitated kinase. Akt2 activity was upregulated by TGFβ only when cells were cultured in high Gln and not in low Gln (Fig. 5E and F). These results were validated immunoprecipitating Akt from MEFs depleted of Akt1: also in these cells, high Gln was required for TGFβ stimulation of the activity of the only Akt isoform present, Akt2.

Finally, and in accordance with the low migration exhibited by MEF Akt2 KO, these cells were unable to stimulate invasion of HT29 M6 when cocultured in a Gln gradient, in contrast to the action of control or Akt1-depleted MEFs (Fig. 5G).

Next, we investigated Gln-sensitive proteins that might mediate the polarized distribution of Akt2. TRAF6 is an ubiquitin ligase that activates Akt producing its multi-ubiquitination (29). Compared with other proteins associated to Akt, TRAF6 presents the relevant feature of being controlled by nutrients: the amino acid-dependent phosphorylation of p62/SQSTM1 potentiates its interaction with TRAF6 and the binding of the complex to different TRAF6 substrates (29, 30). As Akt2, TRAF6 also displayed a polarized distribution in a significant proportion of CAFs when these cells were exposed to a Gln gradient; TRAF6 was more abundant in the migration front and colocalized with Akt2 (Fig. 6A; Supplementary Fig. S7C); this TRAF6 localization required CAF activation because it was prevented by Snail1 depletion or the absence of TGFβ (Supplementary Fig. S7C). TRAF6 was downregulated using two different shRNAs; both populations showed similar levels of Akt2 than the control and slightly lower levels of phosphorylated Akt2 (Fig. 6B). Snail1 was similarly activated by TGFβ and Smad2 phosphorylation was not affected by TRAF6 downmodulation (Fig. 6C). As TRAF6 produces Akt multiubiquitination (31), we studied this modification on Akt2. TRAF6 downmodulation decreased the amount of Akt2 labeled with histidine-tagged ubiquitin (Fig. 6D). Akt2 ubiquitination was not sensitive to TGFβ but it was markedly downregulated in low Gln medium when compared with high Gln, in accordance with previous results indicating that TRAF6 is controlled by nutrients. TRAF6 was required for Akt2 activity because TRAF6 downregulation by the two different shRNAs decreased GSK3α phosphorylation by immunoprecipitated Akt2 (Fig. 6E).

Downregulation of TRAF6 also altered the cellular distribution of Akt2 that was no longer detected in the leading edge of polarized cells (Fig. 6F). In addition, these cells presented a defective αSMA localization (Supplementary Fig. S7D) and a lower invasion following a Gln gradient (Fig. 6G).

We also determined whether Akt2/TRAF6 asymmetrical distribution was observed in human tumors. An analysis of human breast tumors revealed the presence of elongated fibroblastic cells with a polarized distribution of Akt2 and TRAF6 (Fig. 6H). These polarized cells were more abundant in the areas of invasion (Fig. 6I).

Finally, we assessed the relevance of p62/SQSTM. This protein was also downregulated in CAFs using two different shRNAs (Fig. 7A). Similarly to TRAF6, cells with lower p62 expression also downregulated phosphorylated Akt2 (Fig. 7A) and exhibited a similar activation of Snail1 and pSmad2 by TGFβ as control cells (Fig. 7B). p62 also displayed an asymmetrical distribution in CAFs migrating toward Gln but not in cells maintained in high Gln and showed a remarkable colocalization with TRAF6 and Akt2 (Fig. 7C; Supplementary Fig. S8A–S8C). p62 downregulation affected the cellular distribution of Akt2 and TRAF6 because their polarization was lost in these cells in Gln gradients (Fig. 7C). Finally, and in accordance with the altered polarization of CAFs, p62 interference also prevented CAF invasion in Gln gradients (Fig. 7D).

Tumors are considered complex tissues composed by different types of cells that cross-talk and cooperate with epithelial transformed cells. Among the cells of this tumor microenvironment, CAFs have received a special attention because they support growth, invasion, and immune resistance of tumor cells. We have focused on the cooperative effects of tumor cells and CAFs on invasion and migration. CAFs exhibited a higher dependence on Gln than tumor cells; accordingly, CAFs are more sensitive to GLS inhibition than epithelial tumor cells. Similar results have been obtained by others when comparing high- versus low-invasive ovarian cancer cell lines: high invasion is related to an increased Gln dependence (32). In addition, mesenchymal lung cancer cells with low E-cadherin expression are more sensitive to a GLS inhibitor than high E-cadherin epithelial tumor cells (33). Supporting these results, a GLS inhibitor, CB-839 (Telaglenastat), is in phase II clinical trials: it is well tolerated and has a synergistic effect with other drugs on different types of tumors, particularly on triple-negative breast tumors. It would be interesting to find out if this compound has a more pronounced effect in fibrotic tumors.

Gln is considered an essential amino acid in cancer cells that rely on this amino acid for different metabolic processes. Through its conversion to Glu and α-ketoglutarate, Gln is driven to the tricarboxylic acid cycle for energy generation; it is also a precursor of several nonessential amino acids through the action of specific amino transferases and participates in the synthesis of glutathione to eliminate reactive oxygen species (12, 18, 34). Probably for its high utilization by tumor cells, Gln depletion has been observed in the central part of tumors generated xenografting melanoma or breast tumor cells when compared with their periphery (13). However, most cancer cells can rewire their metabolism to adapt to low Gln conditions (35). According to our hypothesis, because CAFs are more sensitive to low Gln levels, the gradient in this amino acid established between tumoral areas with high proliferation and the external zones might drive CAF escape from the tumor. When moving to Gln-richer territories, CAFs would create new tracks that would be used by tumor cells to migrate. Furthermore, some results of our work suggest that CAFs might promote additional effects because these cells also stimulate epithelial tumor migration when cocultured. Therefore, besides producing more proteases to open these tracks, CAFs might stimulate tumor cell invasion by the secretion of diffusible molecules enhancing migration. These factors are being characterized in our lab.

Both CAF invasion and CAF-stimulated tumor cell invasion require their activation by cytokines or growth factors derived from the tumor cells, activation that is dependent on Snail1 expression (9–11). This CAF activation also needs the action of Akt2. This protein kinase works downstream of Snail1 because its depletion does not substantially prevent Snail1 upregulation. Several reports indicate that Snail1 activates Akt during EMT (36–38). Among the members of this family, Akt2 plays the most relevant role in mesenchymal cells. For instance, Akt2 is increased during EMT, whereas Akt1 is downregulated (39, 40). In tumor cells, Akt2 is required for the increased invasion and resistance to chemotherapeutic agents provided by EMT (41, 42). Other results have also demonstrated a relevant function of Akt2 in fibroblast invasion (43). Here we show that Akt2 genetic depletion in MEFs precludes their migration stimulated by TGFβ. When moving toward Gln, Akt2 is localized in the leading front in a remarkable percentage of activated cells, whereas it is more widely distributed in cells that do not exhibit a preferential direction of migration, or in cells that have not been activated. Akt activation in specific compartments, such as invadopodia, is required for migration and invasion because it promotes the compartment-specific phosphorylation of several substrates (44). Invadopodia formation needs not only Akt activation but its specific localization in this structure; remarkably, a constitutively active form of Akt not targeted to invadopodia decreases migration (45).

Gln controls both Akt2 activity and localization, two parameters that require the action of TRAF6. TRAF6 is an E3 ligase that causes Lys63-dependent polyubiquitination, a modification that is not related to degradation but to cell signaling and protein traffic (46). Besides acting on Toll-like receptor (47), TRAF6 interacts and participates in the activation of Src, TAK1, mTORC1, Akt, and PI3K (29, 30, 48–50). Actually, polyubiquitination of mTOR by TRAF6 depends on its nutrient-dependent interaction with p62 (30), establishing a link between amino acid supply and Akt activation. Moreover, through the modulation of TAK1, Akt, and PI3K, TRAF6 is involved in TGFβ signaling and is needed for TGFβ-induced cell migration (50). This TRAF6-mediated activation of PI3K and Akt is independent on TGFβ receptor kinase activity (50), in contrast to the signals inducing Snail1 expression.

In our cells, TRAF6 is necessary for Akt2 activation but it does not seem to be sufficient, because Akt polyubiquitination is not substantially altered by TGFβ. According to our current view, Akt2 activation is sensitive to Gln at two different levels. First, Gln is necessary for the synthesis of Snail1 that provide signals required for activation (38); second, Gln is also needed for Akt2 polyubiquitination by TRAF6/p62. The convergence of these two signals would promote the redistribution of active Akt2 to the migration front, where all its substrates are present and will enable its action on fibroblast migration and invasion.

Metabolic stress, which is consequence of the high tumor cell proliferation and the poor nutrient supply, has been considered as a possible cause of invasion. This has been attributed to the accumulation of lactate in the tumor core that directly potentiates tumor cell migration (51) or the expression of HIF1α by the hypoxic tumoral conditions that promote an EMT in epithelial tumor cells (52). In this work, we described that the Gln deficiency observed in some areas of the tumor also facilitates invasion in a more indirect fashion, acting mainly on CAF invasion and enabling the subsequent movement of tumor cells away from tumor regions poor in this amino acid. It remains to be established whether tumor cells with a mesenchymal phenotype, which are very sensitive to Gln depletion, will also enhance the invasion of the more abundant epithelial tumor cells. In any case, these results show a close collaboration between epithelial and mesenchymal tumor cells in response to nutrient deprivation with relevance in tumor invasion.

A. Mestre-Farrera reports grants from Secretaria d'Universitats i Recerca del Departament d'Empresa i Coneixement de la Generalitat de Catalunya during the conduct of the study. M. Quintela-Fandino reports personal fees from MSD and MEI Pharma and grants from Pfizer and Bayer outside the submitted work. M. Duñach reports grants from Ministerio de Ciencia, Innovación y Universidades-Agencia Estatal de Investigación (Retos de Investigación) and FEDER during the conduct of the study. A. García de Herreros reports grants from Ministerio de Ciencia e Innovación-Agencia Estatal de Investigación (Retos de Investigación) and FEDER and grants from Instituto Carlos III during the conduct of the study. No disclosures were reported by the other authors.

A. Mestre-Farrera: Conceptualization, investigation, methodology, data curation, writing-original draft. M. Bruch-Oms: Investigation, methodology. R. Peña: Data curation, investigation, visualization, methodology, project administration. J. Rodríguez-Morató: Investigation, methodology. L. Alba-Castellón: Conceptualization, methodology. L. Comerma: Resources. M. Quintela-Fandino: Resources. M. Duñach: Resources, funding acquisition, project administration. J. Baulida: Conceptualization, methodology. O.J. Pozo: Data curation, methodology. A. García de Herreros: Conceptualization, formal analysis, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.

We thank Drs. J. Moscat for advice, J. Yélamos, M. Birnbaum, and L. Tío for cell lines and reagents and M. Iglesias for assistance. This study was funded by grants awarded by Ministerio de Ciencia, Innovación y Universidades-Agencia Estatal de Investigación (Retos de Investigación) and FEDER (SAF2016-76461-R and PID2019-104698RB-I00 to A.G. de Herreros; RTI2018-099719-B-100 to M. Duñach). We also acknowledge support from the Instituto Carlos III (PIE15/00008). A. Mestre-Farrera was funded by a Predoctoral FI Contract by the Secretaria d'Universitats i Recerca del Departament d'Empresa i Coneixement de la Generalitat de Catalunya (FI-DGR 2016). M. Bruch-Oms and L. Alba-Castellón were recipients of FPI contracts awarded by Ministerio de Ciencia y Tecnologia.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.