Collagen I promotes cancer cell survival via amino acid import and mTORC1/S6 activation

Abstract

Invasive breast and pancreatic cancer cells thrive within a collagen I-rich, poorly perfused extracellular matrix (ECM) network, necessitating robust metabolic adaptation to endure nutrient deficiency, such as glucose starvation. Here we demonstrate that collagen I is critical for the survival and growth of breast and pancreatic cancer cells. Mechanistically, collagen I promotes α2 β1 integrin-dependent S6 phosphorylation by the mammalian target of rapamycin complex 1 (mTORC1) and drives the membrane localisation of the LAT1−4F2hc amino acid transporter. This process ensures a sustained intracellular essential amino acid supply, further fuelling mTORC1 signalling and limiting autophagy. This collagen I-driven pathway is essential for cancer cell survival, as inhibiting the activity of α2 β1 integrin or the LAT1-4F2hc transporter significantly reduces cell growth and invasion in both 2D and 3D models. Finally, the clinical relevance of these transporters is underscored by the significant upregulation of LAT1-4F2hc expression in basal-like breast and pancreatic cancer patients, correlating with poor prognosis and drug resistance. Collectively, our findings highlight that targeting the LAT1−4F2hc transporter might represent a highly promising therapeutic strategy to limit cancer cell growth and invasion in highly fibrotic and nutrient-deprived tumours.

How do invasive breast and pancreatic cancer cells thrive in collagen-rich environments under nutrient stress? This study shows that collagen I promotes mTORC1 activation and drives the localisation of the LAT1-4F2hc amino acid transporter in the membrane, ensuring a sustained intracellular amino acid supply.

Introduction

Breast cancer is characterised by a desmoplastic reaction, driven by the accumulation of fibrillar collagens, especially type I [1], which is associated with poorer clinical outcomes [2–4]. In healthy mammary tissue, type I collagen is essential for tissue development and cellular organisation during mammary gland branching [5–8]. However, upon cancer initiation, type I collagen undergoes structural alterations that significantly impact ECM stiffness and elasticity, with fibres becoming thicker and more aligned [9–11]. This reorganisation, particularly the radial alignment of collagen fibres, facilitates cancer cell migration and invasion, further promoting metastasis [10–12]. A similar fibrotic response is observed in pancreatic cancer, where fibrillar collagens, primarily types I and III, comprise over 80% of the ECM in pancreatic ductal adenocarcinoma (PDAC) tumours, where they play a significant role in tumour progression. These collagens provide not only structural support but also act as a reservoir for growth factors and influence the behaviour of both cancer and stromal cells through their organisation, crosslinking, and dynamic changes. Such remodelling can drive key tumorigenic processes including cell proliferation, angiogenesis, invasion, metastasis, resistance to apoptosis, and reactivation from dormancy [13–15].

Altered cellular metabolism is a hallmark of cancer. The cancer cell metabolic phenotype is influenced by both cell-intrinsic and extrinsic factors within the tumour microenvironment (TME), including nutrient availability [16–19]. Unlike normal cells with functional vasculature, cancer cells often grow in poorly perfused environments with limited nutrient access. To survive and sustain high proliferation rates, they reprogram metabolic pathways to optimise nutrient acquisition and utilisation [20,21]. Central to this reprogramming is the uptake of glucose (Glc) and glutamine, serving as carbon sources and electron donors [22]. While intrinsic metabolic regulation is increasingly understood, the specific effects of TME-derived nutrient constraints remain poorly characterised.

In this study, we showed that collagen I partially rescued invasive breast cancer and PDAC cell growth under Glc starvation, without affecting cell proliferation in complete media or serum deprivation. This was due to a reduction in cell death in breast cancer cells and a combination of increased survival and cell division in PDAC cells. Mechanistically, collagen I promoted α2β1 integrin-dependent phosphorylation of S6 by the mammalian target of rapamycin complex 1 (mTORC1) in Glc deprived-conditions, but not in complete media. In addition, collagen I increased the plasma membrane localisation of both the light chain (SLC7A5) and the heavy chain (SLC3A2) of the amino acid transporter LAT1-4F2hc, which control essential amino acid uptake. Inhibition of either α2β1 integrin or LAT1-4F2hc prevented collagen I-dependent cell growth and invasion, in both 2D and 3D contexts, including primary mouse tumour organoids. Finally, the co-expression of SLC3A2 and SLC7A5 was significantly upregulated in basal-like breast cancer and pancreatic cancer patients, correlating with poor disease outcomes and drug resistance. Therefore, this work suggests that the inhibition of LAT1-4F2hc could represent a promising therapeutic strategy to limit cancer cell growth and invasion in highly fibrotic and nutrient-deprived breast and pancreatic tumours.

Results

Collagen I promoted the growth/survival of breast and PDAC cancer cells under glucose starvation

Breast and PDAC tumours are characterised by a fibrotic and nutrient-deprived TME. To investigate the effect of collagen I on cellular responses to Glc and pyruvate starvation (hereafter referred to as Glc starvation), we cultured breast cancer cells (MDA-MB-231) and pancreatic cancer cells (PANC1) on either 2 mg/ml polymerised collagen I or uncoated plastic in complete media or under Glc starvation for up to 5 days. We observed that the presence of collagen I, while not affecting cell growth in complete media (Fig 1A and 1C), significantly increased MDA-MB-231 (Fig 1B) and PANC1 (Fig 1D) cell numbers under Glc deprivation, to approximately 50% of full media values.

Fig 1. Collagen I supported cell survival under Glc starvation.

(A–D) MDA-MB-231 (A, B) or PANC1 cells (C, D) were seeded on plastic (P) or 2 mg/ml collagen I (Coll) for 5 days in complete media (A, C) or under Glc starvation (B,D), fixed and stained with Hoechst 33342. Images were collected by ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N = 3 independent experiments. **p < 0.01, ****p < 0.0001 two-way ANOVA, Tukey’s multiple comparisons test. (E, F) MDA-MB-231 and PANC1 cells were seeded or plastic, 2 mg/ml collagen I (Coll) for 6 days under glucose starvation (Glc) or in complete media (Com), every 2 days cells were treated with PI and Hoechst 33342 for 1 h, imaged with ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N = 3 independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 two-way ANOVA, Tukey’s multiple comparisons test. (G, H) MDA-MB-231 and PANC1 cells were seeded on 2 mg/ml collagen I (Coll) or plastic (P), in complete media (Com) or under Glc starvation (Glc). MDA-MB-231 cells were incubated with EdU at day 6 post starvation, fixed and stained with Hoechst 33342 and Click iT EdU imaging kit at day 8. PANC1 cells were incubated with EdU at day 5 post starvation, fixed and stained with Hoechst 33342 and EdU Cell Proliferation imaging kit at day 6. Images were collected by ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N = 3 independent experiments. *p < 0.05, **p < 0.01, ****p < 0.0001 one-way ANOVA Kruskal–Wallis, Dunn’s multiple comparisons test. All the raw data associated with this figure are available in S1 Data.

To determine whether collagen I-driven growth was associated with carcinoma progression, we took advantage of the MCF10 series of cell lines, composed of nontransformed mammary epithelial (MCF10A), noninvasive ductal carcinoma in situ (MCF10A-DCIS), and invasive (MCF10CA1) cells. In contrast to MDA-MB-231 cells, collagen I did not support the growth of MCF10A (S1A Fig) and MCF10A-DCIS (S1B Fig) cells under Glc starvation. However, collagen I promoted the growth of MCF10CA1 cells after 6 days of Glc deprivation (S1C Fig). Consistently, we obtained similar results in SW1990 pancreatic cancer cells, where the presence of collagen I significantly increased cell numbers compared to plastic (S2A Fig). Together, these data indicate that collagen I-mediated cell growth under Glc starvation is associated with carcinoma progression.

To elucidate whether the observed increase in cell numbers was due to enhanced survival or increased division rates, we performed cell death and 5-ethynyl-2′-deoxyuridine (EdU) incorporation assays. Propidium Iodide (PI) is a cell-impermeable nuclear dye, which is excluded from healthy cells, while it can penetrate damaged or dying cells. The quantification of the percentage of PI-positive cells indicated that MDA-MB-231 cells exhibited a higher death rate on plastic under Glc starvation compared to complete media at all time points. Moreover, collagen I significantly reduced cell death under Glc starvation at day 4, with a similar trend at day 5, although not statistically significant (Fig 1E). Similarly, PANC1 cell death was significantly higher under Glc starvation compared to complete media on plastic, while the presence of collagen I completely prevented cell death at all time points (Fig 1F). Collagen I did not affect viability in both MDA-MB-231 and PANC1 cells in complete media (Fig 1E and 1F). EdU is a thymidine analogue that is incorporated into the DNA during the S phase of the cell cycle. As expected, we detected a significant reduction in the percentage of EdU-positive cells under Glc starvation compared to complete media on plastic in both MDA-MB-231 and PANC1 cells (Fig 1G and 1H). While there was no significant difference in EdU incorporation in MDA-MB-231 cells in the presence or absence of collagen I under Glc deficiency (Fig 1G), there was a significantly higher percentage of EdU-positive PANC1 cells cultured on collagen I compared to plastic (Fig 1H), indicating an increased division rate. Collagen I did not affect EdU incorporation in complete media in either cell line (Fig 1G and 1H). Moreover, the presence of collagen I significantly increased MCF10CA1 EdU incorporation (S1D Fig) and reduced the percentage of MCF10CA1 cells positive for the apoptosis marker cleaved caspase 3/7 (S1E Fig). Similarly, collagen I promoted EdU incorporation (S2B Fig) and reduced cell death in SW1990 cells (S2C Fig).

In summary, our findings demonstrate that the presence of collagen I increased MDA-MB-231, MCF10CA1, SW1990, and PANC1 cell numbers under Glc starvation. This effect was attributed to a reduction in cell death for all cell lines, with a concomitant increase in the division rate in PANC1, MCF10CA1, and SW1990 cells, but not in MDA-MB-231 cells. Moreover, the observation that collagen I did not affect the growth of noninvasive cancer and normal mammary epithelial cells suggests that this survival mechanism is associated with invasiveness and/or cancer progression.

α2β1 integrin was required for collagen I-dependent cell growth

We have previously demonstrated that ECM internalisation supported breast cancer cell growth under amino acid starvation [23]. To investigate whether collagen I-dependent cell growth/survival under Glc starvation was also mediated by its internalisation and subsequent lysosomal degradation, we assessed the ability of MDA-MB-231 cells to uptake collagen I under starvation. Interestingly, we found that collagen I endocytosis was significantly increased in Glc-depleted media, both in the presence and absence of the lysosomal protease inhibitor E64d (S3 Fig). We have previously shown that glutaraldehyde-mediated crosslinking prevented ECM extracellular degradation and internalisation [23]. To determine the requirement of collagen I uptake in promoting cell growth, MDA-MB-231 and PANC1 cells were seeded on chemically crosslinked collagen I under Glc starvation. While MDA-MB-231 cells exhibited comparable growth on crosslinked collagen I as on control collagen I (Fig 2A), crosslinking further stimulated PANC1 cell growth (Fig 2B), indicating that the supportive effect of collagen I in Glc deprived–conditions was not primarily dependent on its internalisation.

Fig 2. Collagen I chemical crosslinking did not impair cell proliferation under starvation.

MDA-MB-231 (A) and PANC1 cells (B) were seeded on 2 mg/ml collagen I (Coll) or crosslinked collagen I (X-coll) for 8 or 6 days, respectively. Cells were fixed and stained with Hoechst 33342. Images were collected by an ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N = 3 independent experiments. ***p < 0.001 two-way ANOVA, Tukey’s multiple comparisons test. All the raw data associated with this figure are available in S2 Data.

Cells interact with the ECM through plasma membrane receptors of the integrin family, including α2β1 integrin, one of the most characterised collagen I receptors [24]. Upon ligand binding, integrins trigger the activation of adhesion signalling pathways, which promote cell proliferation and survival [25]. Therefore, we examined the role of α2β1 integrin by treating the cells with BTT-3033, an inhibitor that prevents α2β1 integrin collagen binding, and monitoring cell proliferation. Under Glc starvation, BTT-3033 treatment elicited a dose-dependent inhibition on MDA-MB-231 and PANC1 cell proliferation (Fig 3A). In complete media, while BTT-3033 did not affect MDA-MB-231 cell numbers on plastic (S4A Fig), α2β1 integrin pharmacological inhibition significantly diminished cell growth in MDA-MB-231 cells seeded on collagen I and in PANC1 cells on both plastic and collagen I (S4B, S4E and S4F Fig). Consistently, α2 integrin downregulation significantly decreased cell numbers in both cell lines, under Glc deprivation and in complete media, with a more prominent effect in the presence of collagen I compared to plastic (Figs 3B, S4C, S4D, S4G and S4H). In contrast, β1 integrin knockdown did not affect MDA-MB-231 cells but significantly reduced PANC1 cell numbers only in the presence of collagen I in both Glc starvation and complete media (Figs 3B, S4G and S4H). Immunofluorescence staining confirmed efficient downregulation of both α2 and β1 integrin protein levels in both cell lines (S5 Fig). Similarly, both pharmacological inhibition and siRNA-mediated knock down of α2 integrin significantly impaired MCF10CA1 cell growth on collagen I under Glc starvation, while western blotting confirmed efficient α2 integrin downregulation (S6 Fig). Furthermore, to extend our findings to a more physiologically relevant context, we evaluated the impact of BTT-3033 and α2 integrin knockdown in 3D models. 3D spheroids were generated, embedded in a mixture of collagen I and Geltrex and grown under Glc starvation for 2 days (MDA-MB-231 cells) or 6 days (PANC1 cells). This discrepancy in time frame is due to the difference in invasiveness between the two cell lines. MDA-MB-231 spheroids were highly invasive and extended long multicellular protrusions. Both α2 integrin pharmacological inhibition and siRNA-mediated protein downregulation significantly inhibited MDA-MB-231 spheroid invasion (Fig 3C and 3D). Consistently, we previously found that α2β1 integrin was required for cancer cell migration and invasion in complete media [26], indicating that α2β1 integrin controls motility in a nutrient-independent manner. In contrast, PANC1 spheroids mostly grew in size, with some cells invading in the surrounding ECM. Consistently, α2 integrin inhibition and knockdown dampened spheroid growth under Glc starvation (Fig 3E and 3F). Moreover, primary tumour organoids were generated by growing E0771 mouse triple negative breast tumours into a mixture of collagen I and Geltrex in Glc-free media, in the presence or absence of BTT-3033 for 3 days. While control organoids grew over time, with strands of cells invading the surrounding ECM, both 2.5 μM and 5 μM BTT-3033 completely prevented organoid growth and opposed cell invasion. Image analysis indicated that α2 integrin pharmacological inhibition significantly reduced organoids’ size and invasion compared to the control group (Fig 3G and 3H). These results collectively indicate that the collagen I receptor α2β1 integrin modulated collagen I-dependent cell growth in breast and pancreatic cancer cells, in both 2D and 3D contexts.

Fig 3. α2β1 integrin was required for cancer cell growth and invasion under Glc starvation.

(A) MDA-MB-231 and PANC1 cells were seeded on 2 mg/ml collagen I and treated with 5 μM, 10 μM BTT-3033 (BTT) or DMSO (Ctrl) for 4 days under Glc starvation. (B) MDA-MB-231 and PANC1 cells were seeded on 2 mg/ml collagen I and transfected with an siRNA targeting β1 integrin (β1 si), an siRNA targeting α2 integrin (α2 si) or a nontargeting siRNA control (Nt si) for 4 days under Glc starvation. Cells were fixed and stained with Hoechst 33342. Images were collected by ImageXpress micro and analysed by MetaXpress software. (C) MDA-MB-231-GFP spheroids were generated by the hanging drop method, embedded in 3 mg/ml collagen I and Geltrex (50:50) mixture and starved in Glc-free media for 2 days, in the presence of 10 µM BTT-3033 (BTT) or DMSO (Ctrl). Live images were collected by a Nikon A1 Confocal microscope. Bar, 200µm. (D) MDA-MB-231-GFP cells were transfected with an siRNA targeting α2 integrin (α2 si) or a nontargeting siRNA control (Nt si) for 24 h. Spheroids were generated by the hanging drop method, embedded in 3 mg/ml collagen I and Geltrex (50:50) mixture and starved in Glc-free media for 3 days. Live images were collected by a Nikon A1 Confocal microscope. Bar, 200 µm. (E) PANC1 spheroids were generated by the hanging drop method, embedded in 3 mg/ml collagen I and Geltrex (50:50) mixture and starved in Glc-free media for 6 days in the presence of 10 μM BTT-3033 (BTT) or DMSO (Ctrl). Spheroids were imaged live every 2 days with an Olympus E450 microscope. (F) PANC1 cells were transfected with an siRNA targeting α2 integrin (α2 si) or a nontargeting siRNA control (Nt si) for 24 hours. Spheroids were generated by the hanging drop method, embedded in 3 mg/ml collagen I and Geltrex (50:50) mixture and starved in Glc-free media for 6 days. Live images were collected by an Olympus E450 microscope. Bar, 250 µm. (G, H) E0771 mouse tumour organoids were grown in a 3 mg/ml collagen I and Geltrex (50:50) mixture and starved in Glc-free media for 3 days in the presence of 2.5 μM, 5 μM BTT-3033 (BTT) or DMSO (Ctrl). Spheroids were imaged live every day with an Olympus E450 microscope. Bar, 250 µm. Data are presented as mean ± SEM, N = 3 independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, (A, B and H) Kruskal–Wallis, Dunn’s multiple comparisons test, (C–G) Two-way ANOVA, Tukey’s multiple comparisons test. All the raw data associated with this figure are available in S3 Data.

Collagen I modulated mTOR signalling and autophagy

It is well established that Glc controls nutrient signalling, through the regulation of mTORC1 and AMP-activated protein kinase (AMPK). Under Glc starvation, mTORC1 activity is inhibited, leading to the activation of autophagy, an intracellular degradation process to support nutrient acquisition. This is a crucial adaptation where cells shift from growth and anabolic processes promoted by active mTORC1 signalling to autophagy-mediated catabolism for survival and nutrient recycling [27]. To determine if the increased cell numbers and survival rates observed on collagen I under Glc deficiency were mediated by changes in nutrient signalling, we assessed mTORC1 signalling and autophagy. As a readout for mTORC1 activation, we stained the cells for the phosphorylated form of S6 (pS6), a well-established mTORC1 downstream target. Surprisingly, we did not detect changes in S6 phosphorylation in response to 1-day Glc starvation compared to complete media, while collagen I consistently increased the phosphorylation of S6 in MDA-MB-231 cells, regardless of nutrient availability (Fig 4A). BTT-3033 treatment significantly lowered S6 phosphorylation in the presence of collagen I to the same level as in cells seeded on plastic (Fig 4A), indicating that α2 integrin was required for collagen I-driven phosphorylation of S6. Similarly, the presence of collagen I significantly increased S6 phosphorylation detected by western blotting after 2 h Glc starvation in both MDA-MB-231 (Fig 4B) and PANC1 cells (Fig 4C). Interestingly, the phosphorylation of another mTORC1 downstream target, 4EBP1, was significantly reduced under Glc starvation, independently on the substrate the cells were seeded on (S7A Fig). Moreover, collagen I did not affect mTOR phosphorylation upon 24 h Glc starvation, and we detected a trend towards reduced total mTOR levels on collagen I compared to plastic (S7B Fig), although this was not statistically significant. S6 can also be phosphorylated by p90 Ribosomal S6 Kinases, downstream of mitogen activated protein kinase (MAPK) signalling [28]. To determine the contribution of mTORC1 to collagen I-driven S6 phosphorylation under Glc starvation, we incubated the cells with the mTORC1 inhibitor Rapamycin. pS6 levels were significantly reduced in both MDA-MB-231 and PANC1 cells (S7C Fig), suggesting that mTORC1 contributed to collagen I-promoted S6 phosphorylation under Glc starvation. These data suggest that adhesion signalling might affect only a subset of mTORC1 targets. Consistently, S6, but not 4EBP1, phosphorylation has been reported to be associated with focal adhesions [29]. To determine whether mTORC1 activity was required for collagen I-driven cell growth under Glc starvation, breast and pancreatic cancer cells were grown in the presence of Rapamycin for 4 days. We detected a statistically significant reduction in both MDA-MB-231 (Fig 4D) and PANC1 (Fig 4E) cell numbers in the presence of Rapamycin compared to DMSO control, indicating that mTORC1 signalling was required for ECM-dependent cell growth under Glc starvation.

Fig 4. Collagen I promoted S6 phosphorylation in cancer cells under Glc starvation.

(A) MDA-MB-231 cells were seeded on 2 mg/ml collagen I (Coll, +) or plastic (P, −) under complete or Glc starvation media for 1 day, in the presence of 10 μM BTT-3033 (BTT) or DMSO (Ctrl). Cells were fixed and stained with Hoechst 33342 (blue), Phalloidin (grey), and pS6 (magenta). Images were collected with a ZEISS LSM980 Airyscan2 confocal microscope and analysed by Fiji/ImageJ software. Bar, 20 μm. Data are presented as mean ± SEM, N = 3 independent experiments (the bigger dots represent the mean of individual fields of view). (B, C) MDA-MB-231 (B) and PANC1 (C) cells were grown on 2 mg/ml collagen I (Coll) or plastic (P) for 1 day, then starved in Glc-free media for 2 hours. Lysates were collected and the levels of phosphorylated S6 (pS6), GAPDH and tubulin were measured by Western blotting. Data are presented as mean ± SEM, N = 4 independent experiments. (D, E) MDA-MB-231 (C) and PANC1 (D) cells were seeded on 2 mg/ml collagen I and grown under glucose starvation in the presence of 100 nM Rapamycin (Rap) or DMSO (Ctrl) control for 4 days. Cells were fixed and stained with Hoechst 33342 as mean ± SEM, N = 3 independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, Kruskal–Wallis, Dunn’s multiple comparisons test (A); *p < 0.05, ****p < 0.0001 Mann–Whitney test (B–E). All the raw data associated with this figure are available in S4 Data.

To investigate whether collagen I controlled autophagy, we used MDA-MB-231 cells stably expressing a GFP-LC3 reporter. During autophagy, microtubule-associated protein 1A/1B-light chain 3 (LC3) is converted from a soluble form (LC3-I) to a membrane bound form (LC3-II) which localises to the membrane of autophagosomes, double-membrane vesicles that engulf cellular components [30]. Therefore, if autophagy is induced, GFP-LC3 is recruited on autophagosomes, resulting in a punctate distribution, while when autophagy is off, GFP-LC3 is mostly diffused in the cytosol (Fig 5A). As anticipated, 1-day Glc starvation on plastic significantly increased the number of LC3-positive puncta, indicating an induction of autophagy. Consistent with the fact that ECM attachment prevents autophagy [31], collagen I reduced the number of LC3-positive structures under Glc starvation, to the same levels as cells in complete media (Fig 5A and 5B). The presence of BTT-3033 resulted in a small, but not statistically significant, increase in LC3 puncta, indicating that α2 integrin-mediated adhesion did not mediate collagen I-dependent autophagy inhibition under Glc starvation. As expected, the inhibition of VPS34, a known autophagy regulator, significantly reduced GFP-LC3 puncta (Fig 5B). The decrease in LC3 puncta in the presence of collagen I could represent an increase in autophagic flux, where autophagosomes are being degraded faster, or a decrease in autophagosome formation. To address this, we used the LC3-RFP-GFP reporter to measure autophagic flux, based on the different sensitivities of its fluorescent proteins to lysosomal acidity. GFP fluorescence is quenched at acidic lysosomal pH, while RFP remains stable. When autophagosomes fuse with lysosomes, the GFP signal decreases faster than RFP, making the RFP/GFP ratio an indicator of autophagic activity [32]. The stable ratio of LC3-RFP (autolysosome) to LC3-Yellow (autophagosome) between plastic and collagen I suggested that the increased LC3 vesicular recruitment in plastic-cultured cells was not due to disrupted lysosomal degradation, nor the reduction in LC3 puncta on collagen I was caused by increased degradation (S8 Fig). Consistently, western blot analysis of LC3-II protein levels in MDA-MB-231 and PANC1 cells showed that cells cultured on plastic exhibited significantly higher LC3-II levels compared to those on collagen I after 1 or 3-day Glc starvation, respectively. Similarly to our imaging results, inhibition of α2β1 integrin did not significantly affect LC3-II levels. Moreover, the mTORC1 inhibitor Rapamycin also did not influence LC3-II, suggesting that the reduction in autophagy observed in the presence of collagen I was not mediated by increased mTORC1 activity (Fig 5C and 5D).

Fig 5. Collagen I inhibited autophagy under Glc starvation.

(A) MDA-MB-231 cells stably expressing GFP-LC3 were seeded on 2 mg/ml collagen I (Coll, +) or plastic (P, −) under complete or Glc starvation media for 1 day, in the presence of 10 μM BTT-3033 (BTT), 4 μM VPS34-IN1 (VPS) or DMSO (Ctrl). Cells were fixed and stained with Hoechst 33342 (blue) and Phalloidin (grey). Images were collected with a ZEISS LSM980 Airyscan2 confocal microscope and analysed by Fiji/ImageJ software. Bar, 20 μm. (B) Data are presented as mean ± SEM, N = 3 independent experiments (the bigger dots represent the mean of individual fields of view). MDA-MB-231 (C) and PANC1 (G) cells were seeded on 2 mg/ml collagen I (+) or plastic (−) under Glc starvation. Cells were treated with DMSO (Ctrl), 100 nM Rapamycin (Rap) or 10 μM BTT-3033 (BTT). Cell lysates were collected after one (C) or 3 days (D) of starvation. Lysates were collected and the levels of LC3 and GAPDH were measured by Western blotting. Data are presented as mean ± SEM, n ≥ 4 independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, Kruskal–Wallis, Dunn’s multiple comparisons test (B, E–G); ****p < 0.0001 Mann–Whitney test (C, D). All the raw data associated with this figure are available in S5 Data.

Collectively, our data indicate that cells cultured on collagen I under Glc starvation maintain an anabolic-like status, characterised by higher S6 phosphorylation and lower autophagy, whereas cells on plastic are driven towards catabolism. Interestingly, α2β1 integrin is required for S6 phosphorylation, but not for autophagy inhibition, suggesting alternative regulatory mechanisms.

Collagen I increased intracellular amino acid content

It is well established that cancer cells rewire their metabolism under Glc starvation, by increasing their reliance on alternative fuels, such as amino acids [27]. In addition, several amino acids promote mTORC1 activity [33]. Therefore, we hypothesised that collagen I might result in changes in intracellular metabolite levels. To address this, we performed nontargeted mass spectrometry in cells grown under Glc starvation for 1 day (Fig 6A and 6D) and detected significant metabolic differences in MDA-MB-231, PANC1, MCF10CA1, and SW1990 cells cultured on collagen I compared to plastic (Figs 6B, 6E, S9A, S9B, S9D and S9E). Furthermore, MDA-MB-231 cells grown on chemically crosslinked collagen I showed no significant difference in metabolic content relative to untreated collagen I (S10 Fig), indicating that collagen I internalisation was not responsible for the collagen I-induced changes in intracellular metabolites. Metabolic pathway enrichment analysis on the upregulated metabolites on collagen I identified several pathways related to amino acid metabolism in MDA-MB-231 and PANC1 cells (Fig 6C and 6F), as well as in MCF10CA1 and SW1990 cells (S9C and S9F Fig), suggesting increased intracellular amino acid levels. To confirm this, we conducted targeted mass spectrometry during a four-day Glc starvation period. Overall, the levels of most amino acids were increased on collagen I compared to plastic. In particular, essential amino acids, including leucine/isoleucine, lysine, methionine, phenylalanine, threonine, tryptophan, and valine, were upregulated on collagen I in MDA-MB-231 and PANC1 cells, from day one to day four of starvation (Fig 6G and 6H). It is worth noting that the day 4 data might be influenced by the fact that we observed increased cell death on plastic compared to collagen I. To determine the source of the upregulated metabolites, we inhibited autophagy by treating MDA-MB-231 and PANC1 cells with a VPS34 inhibitor. Most amino acid levels were unaffected by the VPS34 inhibitor, except for the following nonessential amino acids: aspartate, glycine, and glutamine in MDA-MB-231 cells (Fig 6G), and alanine, aspartate, glycine, glutamine, and serine in PANC1 cells (Fig 6H). Interestingly, these amino acids were also the most abundant in cells on plastic, consistent with our observation that autophagy was elevated on plastic compared to collagen I. Therefore, we ruled out autophagy as the main source for elevated amino acid levels on collagen I. Moreover, intracellular amino acid levels were not changed in either MDA-MB-231 or PANC1 cells treated with the α2β1 integrin inhibitor BTT-3033 (Fig 6G and 6H), indicating that integrin/ECM binding was not required for collagen I-driven increased amino acid content under Glc starvation. Given that all upregulated amino acids are essential and present in the culture media, we hypothesised that the cells might rely on extracellular amino acid import.

Fig 6. Collagen I increased intracellular amino acid levels.

(A, D) Metabolomics workflow. MDA-MB-231 and PANC1 cells were plated on plastic or 2 mg/ml collagen I (Coll) for 1 day in Glc free media. Metabolites were extracted and quantified by non-targeted mass spectrometry. Volcano plot (B, E) and enriched metabolic pathways (C, F) are shown. These datasets are deposited in ORDA (https://doi.org/10.15131/shef.data.21608490). (G) MDA-MB-231 and (H) PANC1 cells were plated on plastic or 2 mg/ml collagen I for up to 4 days in Glc free media in the presence of 4 μM VPS34-IN1 or 10 μM BTT-3033 where indicated. Metabolites were extracted at day 1, 2, and 4 and intracellular amino acid levels were quantified by targeted mass spectrometry. The fold change relative to collagen I day 1 is presented. All the raw data associated with this figure are available in S6 Data.

Collagen I increased LAT1-4F2hc protein levels to increase intracellular essential amino acid levels

The majority of the upregulated amino acids (leucine/isoleucine, methionine, phenylalanine, threonine, tryptophan, and valine), excluding threonine, are known substrates for the Large Neutral Amino Acid Transporter 1 (LAT1). LAT1-4F2hc is a heterodimer consisting of the common heavy chain, SLC3A2, and its specific light chain, SLC7A5 (Fig 7A). To explore whether LAT1-4F2hc contributed to collagen I-dependent intracellular amino acid accumulation, we measured the expression of SLC3A2 and its associated light chains, SLC7A5, SLC7A6, and SLC7A8. qPCR analysis showed that the mRNA levels of SLC3A2 and SLC7A5 were strongly elevated by Glc starvation on plastic and collagen I in both MDA-MB-231 and PANC1 cells, while SLC7A6 and SLC7A8 expression was overall lower and not affected by the starvation (Fig 7B and 7C). We then assessed SLC3A2 and SLC7A5 membrane distribution by performing immunofluorescence staining of nonpermeabilised cells. We observed significantly higher SLC3A2 and SLC7A5 protein levels in both MDA-MB-231 (Fig 7D and 7E) and PANC1 (Fig 7F and 7G) cells grown on collagen I compared to plastic under Glc starvation. In complete media, we found a small, but statistically significant, increase in the plasma membrane targeting of SLC3A2 in MDA-MB-231 cells (S11A Fig) and SLC7A5 in PANC1 cells (S11D Fig), while SLC3A2 localisation in PANC1 cells (S11C Fig) and SLC7A5 localisation in MDA-MB-231 (S11B Fig) where not affected by the presence of collagen I. These results collectively indicate that collagen I mostly controls LAT1-4F2hc membrane localisation, rather than its mRNA expression, suggesting that this might promote amino acid import under Glc starvation. To determine whether this process was dependent on α2β1 integrin, we measured SLC7A5 localisation in the presence of BTT-3033. While we detected a small, but statistically significant, reduction in the transporter membrane targeting in MDA-MB-231 cells (S11E Fig), SLC7A5 localisation was not affected in PANC1 cells (S11F Fig), suggesting that α2β1 integrin is not the major regulator of collagen I-mediated SLC7A5 localisation. Moreover, siRNA-mediated knockdown of SLC3A2 and SLC7A5 alone or in combination resulted in lower threonine, phenylalanine, valine, leucine/isoleucine, tryptophan and tyrosine level in MDA-MB-231, as detected by targeted mass spectrometry (Fig 7H). In PANC1 cells, the combined downregulation of SLC3A2 and SLC7A5 reduced the intracellular levels of all the amino acid tested (Fig 7I). To determine whether these changes in intracellular amino acid content could be due to defective import, 13C-Tyrosine uptake was measured by targeted mass spectrometry in MDA-MB-231 cells seeded on collagen I under Glc starvation (S12A Fig). Consistently with our previous results, we detected a small, but statistically significant reduction in intracellular 13C-Tyrosine levels in the presence of the LAT1-4F2hc inhibitor D-Phenylalanine [34,35], while pharmacological inhibition of α2β1 integrin did not have any effect (S12B Fig). Together these data suggest that collagen I might increase intracellular amino acid levels, at least in part, by promoting LAT1-4F2hc-mediated import, in an α2β1 integrin independent manner.

Fig 7. LAT1-4F2hc was required for collagen I-mediated increased in intracellular amino acid.

(A) Schematic representation of SLC3A2-containing amino acid transporters. (B) MDA-MB-231 and (C) PANC1 cells were seeded on 2 mg/ml collagen I (Coll) or plastic (P) under complete or Glc-free media for 24 h, mRNA was extracted and the levels of SLC3A2, SLC7A5, SLC7A6, SLC7A8, and GAPDH as housekeeping gene were quantified by SYBR-green qPCR. Data are presented as 2^-ΔCT. Data are presented as mean ± SEM, N = 3 independent experiments, ****p < 0.0001 Kruskal–Wallis, Dunn’s multiple comparisons test. (D, E) MDA-MB-231 and (F, G) PANC1 cells were seeded on 2 mg/ml collagen I (Coll) or plastic (P) under Glc starvation for 1 or 2 days, respectively. Cells were fixed and stained with (D, F) Hoechst 33342 (magenta), SLC3A2 (cyan) and Phalloidin Alexa Fluor 555 (grey) or (E, G) Hoechst 33342 (cyan), SLC7A5 (red) and Phalloidin Alexa Fluor 555 (grey). Images were collected with a ZEISS LSM980 Airyscan2 microscope and analysed by Fiji/ImageJ software. Bar, 20µm. Data are presented as mean ± SEM, N = 3 independent experiments (the bigger dots represent mean intensity of each field of view); **p < 0.01, ***p < 0.001 Mann–Whitney test. (H, I) MDA-MB-231 and PANC1 cells were transfected with an siRNA targeting SLC3A2 (3A2), an siRNA targeting SLC7A5 (7A5) or a combination of siRNA targeting SLC3A2 and SLC7A5 (3A2+7A5) and grown on 2 mg/ml collagen I under Glc starvation for 1 or 3 days, respectively. Metabolites were extracted and analysed by targeted mass spectrometry. The fold change relative to collagen I is presented. N = 3 independent experiments. All the raw data associated with this figure are available in S7 Data.

LAT1-4F2hc supported S6 phosphorylation and reduced autophagy under glucose starvation, to promote cell growth

As intracellular amino acids are well defined regulators of mTORC1 activity [33], we monitored S6 phosphorylation and autophagy upon SLC3A2 downregulation (Fig 8A, 8F and 8K). We found that SLC3A2 knockdown significantly reduced S6 phosphorylation, to a similar extent than upon α2 integrin downregulation (Fig 8B, 8G and 8L), and enhanced LC3-II levels (Fig 8C, 8H and 8M) in MDA-MB-231 cells after 1 day of Glc starvation, as well as in PANC1 cells after 2 and 3 days of Glc starvation. Consistent with our pharmacological inhibitor results, α2 integrin knockdown did not affect LC3-II levels (Fig 8C, 8H and 8M). Moreover, α2 integrin and SLC3A2 downregulation did not influence each other levels, while the siRNA-mediated knockdown significantly reduced the expression of the target proteins (Fig 8D, 8E, 8I, 8J, 8N and 8O). This suggests that SLC3A2 mediated collagen I-dependent S6 phosphorylation and autophagy inhibition in cancer cells under Glc starvation. To rule out compensatory effects due to the long-term starvation, we also assessed S6 phosphorylation after 2-h Glc starvation, in the presence of BTT-3033 or D-phenylalanine. Consistently with our previous data, we found that both α2β1 integrin and LAT1-4F2hc inhibition significantly reduced S6 phosphorylation under Glc deprivation (S13A Fig). Interestingly, neither treatment affected pS6 levels in cells grown on collagen I in complete media (S13B Fig), indicating that α2β1 integrin and LAT1-4F2hc specifically control S6 phosphorylation on collagen I under Glc starvation.

Fig 8. SLC3A2 was required for collagen I-induced mTORC1 activation and autophagy inhibition.

MDA-MB-231 (A–E) and PANC1 (F–O) cells were transfected with an siRNA targeting SLC3A2 (3A2 si), an siRNA targeting α2 integrin (α2 si) or a non-targeting siRNA control (Nt si) and grown on 2 mg/ml collagen I in Glc-free media for one (A–E), two (F–J) and 3 days (K–O). Protein level of phosphorylated S6 (pS6, B,G,L), LC3II (C,H,M), SLC3A2 (D,I,N) and α2 integrin (E,J,O) were measured via Western blotting. Data are presented as mean ± SEM, N ≥ 4 independent experiments. *p < 0.05, **p < 0.01 Kruskal–Wallis, Dunn’s multiple comparisons test. All the raw data associated with this figure are available in S8 Data.

We therefore wanted to investigate the role of SLC3A2 and SLC7A5 in supporting cell growth/survival on collagen I. In MDA-MB-231 cells, on the one hand, siRNA-mediated knockdown of SLC3A2 significantly decreased cell numbers under Glc starvation (Fig 9A), but not in complete media (Fig 9B). On the other hand, SLC7A5 downregulation resulted in a small, but statistically significant, reduction in cell numbers under both Glc starvation (Fig 9C) and complete media (Fig 9D). In PANC1 cells, SLC3A2 knockdown significantly reduced cell numbers in complete media (Fig 9G), but not under Glc starvation (Fig 9F). In contrast, SLC7A5 downregulation significantly reduced cell numbers under both conditions (Fig 9H and 9I). Western blotting confirmed efficient downregulation of SLC7A5 upon siRNA transfections in both MDA-MB-231 (Fig 9E) and PANC1 (Fig 9J) cells. When cells were seeded on plastic, the downregulation of SLC7A5, but not SLC3A2, significantly reduced MDA-MB-231 and PANC1 cell number in complete media (S14A and S14C Fig), while both transporters did not affect the growth of either cell line under Glc starvation (S14B and S14D Fig). This indicates that SLC7A5 plays a broader role in controlling cell proliferation, regardless of nutrient and ECM availability. In the 3D spheroid model, SLC3A2 and SLC7A5 double knockdown resulted in a small, but statistically significant, reduction in MDA-MB-231 spheroid invasion after 3 days of Glc starvation (Fig 9K). Additionally, SLC3A2 and SLC7A5 double knockdown reduced PANC1 spheroid size after 6 days of Glc starvation (Fig 9L). To determine the role of LAT1-4F2hc in primary tumour cells, E0771 mouse triple negative breast organoids were grown into a mixture of collagen I and Geltrex in Glc-free media in the presence or absence of D-phenylalanine for 2 days. Consistent with our 3D spheroid results, LAT1-4F2hc pharmacological inhibition significantly reduced organoids’ growth and invasion compared to the control group (Fig 9M and 9N). Together, these results demonstrate that SLC3A2 and SLC7A5 are required for breast and pancreatic cancer cell proliferation and invasion, in both 2D and 3D models, to a different extent depending on the cell type.

Fig 9. SLC3A2 and SLC7A5 were required for cell growth in 2D and 3D systems.

MDA-MB-231 (A–D) and PANC1 (F–I) cells were transfected with an siRNA targeting SLC3A2 (3A2 si), an siRNA targeting SLC7A5 (7A5 si) or a nontargeting siRNA control (Nt si) and grown on 2 mg/ml collagen I under complete or Glc-free media for 4 days. Cells were fixed and stained with Hoechst 33342. Images were collected by an ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N = 3 independent experiments. *p < 0.05, **p < 0.01 and ****p < 0.0001 Mann–Whitney test. MDA-MB-231 (E) and PANC1 (J) cells were transfected with an siRNA targeting SLC7A5 (7A5 si) or a nontargeting siRNA control (Nt si) for 3 days. Lysates were collected and the levels of SLC7A5 and GAPDH were measured by Western blotting. (K) MDA-MB-231-GFP cells were transfected with a combination of siRNAs targeting SLC3A2 and SLC7A5 (3A2+7A5 si) or a nontargeting siRNA control (Nt si) for 24 hours, spheroids were generated by the hanging drop method and embedded in 3 mg/ml collagen I and Geltrex (50:50 ratio) for 3 days under Glc starvation. Images were collected by a Nikon Confocal A1 microscope. The invasion area was quantified with Fiji/ImageJ. Bar, 200 µm. Data are presented as mean ± SEM, N = 2 independent experiments. (L) PANC1 cells were transfected as in K, spheroids were generated by the hanging drop method and embedded in 3 mg/ml collagen I and Geltrex (50:50 ratio) for 6 days under Glc starvation. Live images were collected every day by an Olympus E450 microscope. Bar, 250 µm. Data are presented as mean ± SEM, N = 2 independent experiments. (M, N) E0771 mouse tumours organoids were grown in a 3 mg/ml collagen I and Geltrex (50:50) mixture and starved in Glc-free media for 2 days in the presence or absence of 50 mM D-phenylalanine (D-Phe). Spheroids were imaged live every day by an Olympus E450 microscope. Bar, 250 µm. Data are presented as mean ± SEM, N = 3 independent experiments. *p < 0.05, **p < 0.01 and ****p < 0.0001, (I,J,K) Two-way ANOVA, Tukey’s multiple comparisons test, (L) Mann–Whitney test. All the raw data associated with this figure are available in S9 Data.

Intracellular amino acid content was not affected by collagen I under serum starvation

To determine whether collagen I-driven elevated amino acid levels supported cell survival under additional nutrient limitation conditions, we assessed the response of breast and pancreatic cancer cells to serum starvation. MDA-MB-231 and PANC1 cells were grown on plastic or collagen I for up to 8 days in the absence of serum. While the presence of collagen I resulted in a significant increase in MDA-MB-231 cell number (S15A Fig), PANC1 cell numbers were reduced on collagen I compared to plastic (S15B Fig), suggesting a cell type-specific response. Focussing on MDA-MB-231 cells, we next assessed whether collagen I-driven cell growth under serum deprivation was also associated with S6 phosphorylation and increased intracellular amino acid pools. Interestingly, collagen I promoted the phosphorylation of S6 (S15C Fig), while the intracellular levels of all the essential amino acid tested, with the exception of methionine, were not affected (S14D Fig). Together, these data suggest that the mechanisms through which collagen I support cell growth/survival are different depending on nutrient availability.

LAT1-4F2hc elevated expression correlated with poor prognosis and reduced chemotherapy response in breast and pancreatic cancer patients

To examine whether LAT1-4F2hc may have clinical implications, we looked at the relationship between the co-expression of SLC3A2 and SLC7A5 and disease outcome in breast and pancreatic cancer patients. RNA sequencing data showed overexpression of SLC3A2 and SLC7A5 in basal-like breast cancer (Fig 10A) and pancreatic adenocarcinoma patients (Fig 10C). Correspondingly, concomitant high expression of both genes correlated with poor overall survival (Fig 10B) and disease-free survival (Fig 10D) in basal-like breast and pancreatic cancer patients, respectively. To determine whether amino acid transporters may impact on chemoresistance in breast cancer patients, we looked at the correlation between the co-expression of SLC3A2 and SLC7A5 and chemotherapy response using transcriptomic data from 3,104 breast cancer patients [36]. We found that the two genes were highly expressed in chemoresistance (nonresponder) tumours (Fig 10E) and the area under the ROC curve (AOC) was 0.628 (Fig 10F), classifying them as weak biomarkers with potential use in prediction of chemotherapy treatment [36].

Fig 10. SLC3A2 and SLC7A5 co-expression was upregulated in breast and pancreatic tumours and correlated with poor prognosis and chemotherapy resistance.

(A, C) RNA sequencing data from basal-like breast tumours (n = 135) and normal breast tissue (n = 291) or pancreatic adenocarcinoma tumours (n = 179) and normal pancreatic tissue (n = 171) for the co-expression of SLC3A2 and SLC7A5. *p < 0.05. Data were obtained from gepia2.com. (B) Overall survival of basal-like breast cancer patients with high (red) or low (blue) co-expression of SLC3A2 and SLC7A5. Data were directly generated using gepia2.com (https://gepia2.cancer-pku.cn/). (D) Disease free survival of pancreatic adenocarcinoma patients with high (red) or low (blue) co-expression of SLC3A2 and SLC7A5. Data were obtained from gepia2.com. (E, F) RNA sequencing data and ROC analysis for the co-expression of SLC3A2 and SLC7A5 from chemoresistant (Nonresponder) and chemosensitive (Responder) breast tumours. (E) p = 1.4e-6, Mann–Whitney test. (D) AUC = 0.628; ROC p value = 2.5e-07. Data were directly generated in ROCplot.com (https://rocplot.com/).

Discussion

Here, we showed that the ECM component collagen I supported the growth/survival of invasive breast cancer (MDA-MB-231 and MCF10CA1) and pancreatic cancer cells (PANC1 and SW1990) under combined Glc and pyruvate starvation. It is possible that the lack of both metabolites contributes to the effects described here. In contrast, noninvasive breast cancer (MCF10DCIS) and nontransformed mammary epithelial cells (MCF10A) showed no such growth advantage in the presence of collagen I. These findings suggest that the ability to utilise collagen I for survival under nutrient deprivation is a mechanism acquired during cancer progression and associated with invasiveness. We also observed a varying effect on cell behaviour; in PANC1, SW1990, and MCF10CA1 cells, collagen I promoted both cell division and survival, whereas in MDA-MB-231 cells, it supported survival, without promoting cell division. PANC1 cells are classified as Glucose Insensitive Cells (GIC) due to their strong antioxidant capacity, which confers high tolerance to reactive oxygen species (ROS) and enables them to maintain stable ATP levels. Conversely, MDA-MB-231 cells are considered Glucose Sensitive Cells (GSC) due to a lower tolerance to metabolic stress, characterised by a failure of ROS regulation and a more pronounced decrease in ATP levels [37]. Therefore, we hypothesise that the fundamental role of collagen I engagement via integrin receptors is to provide a potent survival signal that activates anti-apoptotic pathways, such as FAK/PI3K/Akt signalling [38]. Cell proliferation is extremely energy-intensive and requires a high supply of ATP and building blocks [39]. In this context, GICs like PANC1 cells are fully capable of supporting the increased energy demand and higher ROS production inherent to rapid growth, resulting in a successful increase in proliferation. Meanwhile, GSCs, such as MDA-MB-231 cells, are limited by their metabolic fragility, which prevents them from sustaining the intense energy expenditure required for rapid, active proliferation, thus resulting in an unchanged division rate.

We demonstrated that the mechanism through which collagen I supported cell growth/survival under Glc starvation was independent from ECM internalisation. Our recent work demonstrated that breast cancer cells relied on ECM uptake followed by lysosomal degradation to grow under amino acid starvation [23], suggesting that collagen I promotes cell growth/survival through different mechanisms depending on the type of nutrient limitation. Furthermore, our results contrast with studies showing that soluble albumin or collagen did not support pancreatic cancer cell growth under Glc deficiency [40,41]. However, it has been shown that α2β1 integrin has higher affinity for fibrillar collagen I, compared to soluble collagen [42], suggesting that matrix collagen I might be required to trigger α2β1 integrin-dependent cell survival. Indeed, here we showed that inhibition of α2β1 integrin significantly reduced cell growth and invasion in both breast and pancreatic cancer cells under Glc starvation.

Mechanistically, we showed that collagen I promoted 2 separate pathways: S6 phosphorylation by mTORC1, a central sensor of nutrient availability [43], and the stimulation of amino acid transporter membrane localisation, leading to increased intracellular amino acid content (Fig 11). Interestingly, prolonged Glc starvation reduced 4EBP1 phosphorylation, but not S6 in breast cancer cells. While Glc deprivation has been established to inhibit mTORC1 signalling [27], this response has mostly been characterised after short-term Glc limitation (30 min to 3 h). Therefore, it is possible that the longer starvation used in this study might result in additional cellular adaptations, leading to the reactivation of S6 phosphorylation, but not 4EBP1. Consistent with this, S6K phosphorylation has been shown to be induced by Glc starvation in muscle cells [44]. However, collagen I promoted S6 phosphorylation, both upon 2-h and 24-h Glc starvation, while 4EBP1 and mTOR phosphorylation were not affected. mTORC1 was previously shown to control S6, but not 4EBP1 phosphorylation in colorectal cancer [45]. S6 phosphorylation was consistently reduced by α2β1 integrin pharmacological inhibition or siRNA-mediated knockdown. These results are in line with previous studies demonstrating that integrin-ECM binding activates the Akt/mTORC1/S6K/pS6 pathway [46,47].

Fig 11. Working model.

Under Glc starvation, collagen I promoted α2β1 integrin-dependent S6 phosphorylation by mTORC1 in breast and pancreatic cancer cells. Collagen I also increased the membrane levels of the LAT1-4F2hc amino acid transporter heavy chain, SLC3A2 (3A2), and light chain, SLC7A5 (7A5), resulting in increased intracellular amino acid levels. This resulted in S6 phosphorylation and autophagy inhibition, leading to increased survival, proliferation and invasion, in 2D and 3C contexts.

Glc starvation is a well-documented inducer of autophagy, a process of cellular self-digestion [48]. While this held true for cells grown on plastic, we observed that cells cultured on collagen I exhibited reduced LC3 recruitment to autophagosomes under Glc starvation. Integrin activation has been shown to prevent autophagy via several pathways, including the PI3K-Akt/mTOR [49] and Src/FAK [50] pathways, which deactivate AMPK, a major autophagy regulator [51]. However, this relationship is not always straightforward, as evidenced by studies where fibronectin secreted from skeletal muscle activated hepatic autophagy in an α5β1 integrin-dependent manner [52]. Interestingly, our results show a more complex regulatory landscape. The α2β1 integrin inhibitor, BTT-3033, and the mTORC1 inhibitor, Rapamycin, did not significantly alter autophagy in cells seeded on collagen I under Glc starvation, suggesting that the observed reduction in autophagy in the presence of collagen I is not directly mediated by α2β1 integrin or mTORC1 activation. This implies that, while integrin signalling and mTORC1 are essential for collagen I-dependent growth/survival, the downstream suppression of autophagy may be mediated by an independent pathway. As a distinct collagen I receptor, discoidin domain receptor 2 (DDR2) offers a parallel signalling route to integrins, with evidence showing DDR2-mediated autophagy inhibition in adventitial fibroblasts [53]. Thus, DDR2 could provide a redundant, nonintegrin-dependent mechanism for guaranteeing the suppression of catabolic autophagy necessary for proliferation and survival.

The activation of mTORC1 has been shown to occur through several signalling pathways. One such mechanism is triggered by the presence of essential amino acids in the cells, particularly leucine [43]. It is well established that amino acid-dependent activation of mTORC1 is primarily mediated by their cellular uptake by plasma membrane transporters. The LAT1-4F2hc complex is central to this mechanism [35,54,55]. In our study, in both breast and pancreatic cancer cells under Glc starvation, collagen I resulted in increased intracellular essential amino acids, including isoleucine, leucine, phenylalanine, methionine, tryptophan, tyrosine, and valine, known substrates for the LAT1-4F2hc transporter [34]. We demonstrated that collagen I increased the membrane localisation of both SLC3A2 and SLC7A5 in breast and pancreatic cancer cells under Glc deficiency, which provides a mechanistic explanation for the increased intracellular amino acid content observed. SLC3A2 is critical for the proper function and localisation of SLC7A5, acting as an ancillary protein that supports SLC7A5 traffic to the plasma membrane. The N-glycosylation of SLC3A2 is crucial for its own stability and successful transport, which in turn ensures the correct localisation of SLC7A5 [56–58]. Moreover, the association with SLC3A2 is fundamental for the transport activity of SLC7A5, as SLC7A5 alone cannot efficiently transport amino acids [58]. Consistently, our data show that knocking down both SLC3A2 and SLC7A5 has synergistic effects on intracellular amino acid levels. Additionally, downregulation of SLC3A2 decreased the phosphorylation of S6, an mTORC1 downstream target, and increased the autophagy marker LC3-II in both MDA-MB-231 and PANC1 cells. This is consistent with previous observations showing that LAT1 inhibition decreased S6 phosphorylation and promoted autophagy [35,59].

SLC3A2 and integrin crosstalk has been previously reported. SLC3A2 is known to physically associate with integrins, particularly with the cytoplasmic domain of β1 integrin in mammalian cells and βPS in Drosophila, and this interaction drives adhesion signalling [60–63]. Moreover, SLC3A2 has been shown to regulate integrin distribution in polarised cells [64], while it is currently unknown whether integrin activation controls SLC3A2 function/localisation.

It remains unclear whether the α2β1 integrin and the LAT1-4F2hc pathways are activated independently upon collagen I adhesion or are interlinked, although the evidence presented here points towards independent activation. Indeed, α2β1 integrin pharmacological inhibition did not affect amino acid levels. Previous studies have shown that the phosphorylation of 4EBP1 downstream of mTORC1 transcriptionally regulates amino acid transporters by controlling the translation and stability of the transcription factor ATF4 [65]. However, we did not observe any difference in the expression of SLC3A2 and SLC7A5 in the presence of collagen I compared to plastic, and α2 integrin knockdown did not alter SLC3A2 total protein level. Additionally, there was no difference in 4EBP1 phosphorylation between collagen I and plastic, suggesting that this mechanism might not be at play in our context. We can hypothesise that collagen I binding might affect LAT1-4F2hc membrane trafficking and localisation, potentially mediated by another collagen receptor. Indeed, α2β1 integrin pharmacological inhibition only slightly reduced SLC7A5 membrane targeting in MDA-MB-231, but not in PANC-1 cells. Detailing the specific molecular targets within this pathway is now the critical next step and promises to uncover further novel therapeutic targets against matrix-driven diseases. Interestingly, collagen I-dependent regulation of intracellular amino acid concentration does not appear to be a general starvation response, as we did not observe any difference in intracellular essential amino acid content upon serum deprivation.

Evidence in the literature supports the role of both SLC3A2 and SLC7A5 in promoting tumour growth and survival, as their elevated expression enables cancer cells to thrive in a hostile tumour microenvironment. Upregulated expression of SLC3A2 in osteosarcoma and gliomas promotes tumour growth by activating the PI3K/Akt pathway [66] and is linked to enhanced malignancy and poor prognosis [67]. Similarly, high expression of SLC7A5 is a common feature across several cancers [68]. In bladder cancer, it is associated with poor prognosis and enhanced proliferation, migration, and invasion [69], while in pulmonary adenocarcinoma, it predicts patient prognosis [70]. The inhibition of SLC7A5 has shown promise in suppressing cancer progression. In melanoma, knockdown or inhibition of SLC7A5 suppresses metastasis and proliferation by downregulating the mTOR signalling pathway [71]. In mouse models of colorectal cancer, Slc7a5 deletion was found to prevent tumour growth and metastasis. Consistently with our findings, this was coupled with a reduction in S6 phosphorylation (but not 4EBP1) and an increase in autophagy. Interestingly, Slc7a5 expression was found to be upregulated by Kras mutation [45]. Given the fact that the cell lines used in this study harbour KRAS mutations, this raises the possibility that the collagen I-driven regulation of SLC7A5 shown here could also be downstream of KRAS, and future work will be needed to determine this. The combined expression of SLC7A5 and SLC3A2 has also been identified as a potential predictor for response to endocrine therapy in oestrogen-receptor-positive breast cancer, where SLC7A5 expression was also correlated with genes related to proliferation and hypoxia [72,73]. In our study, knocking down SLC3A2 and SLC7A5 in invasive breast cancer cells inhibited 2D cell growth and 3D cell invasion. In PDAC cells, knocking down SLC7A5, but not SLC3A2, decreased growth in 2D, while the double knockdown of both genes decreased growth in 3D spheroids. It is important to consider that SLC3A2 functions as a chaperone for other amino acid antiporters, notably SLC7A11. The literature suggests that SLC7A11-mediated glutamate efflux depletes intracellular glutamate, thereby limiting the cell’s ability to utilise glutamine-derived carbon (via α-ketoglutarate) to maintain the tricarboxylic acid cycle in Glc-depleted conditions [74]. This suggests that in the 2D PANC1 model, SLC3A2 knockdown may be affecting a compensatory metabolic mechanism mediated by SLC7A11 or other SLC3A2-dependent transporters, which ultimately offsets the growth reduction expected from decreased amino acid uptake.

In conclusion, our study demonstrated that collagen I partially rescued breast and pancreatic cancer cell growth under Glc deficient conditions, via a combination of two pathways. Firstly, α2β1 integrin was required to sustain S6 phosphorylation downstream of mTORC1, thereby promoting an anabolic state. Secondly and concurrently, the cells exhibited higher levels of the LAT1-4F2hc (SLC7A5-SLC3A2) transporter on the plasma membrane, resulting in increased intracellular amino acid pools in the presence of collagen I under Glc starvation. This resulted in reduced autophagy and sustained S6 phosphorylation. Therefore, targeting LAT1-4F2hc-dependent amino acid transport could represent a novel therapeutic strategy to prevent the growth and progression of highly fibrotic and nutrient-deprived tumours.

Methods

Reagents

Primary antibodies for mTOR, phospho-mTOR Ser2448, Phospho-S6 Ribosomal Protein ser235/236 and LC3A/B, alpha-Tubulin (DM1A) were from Cell Signalling Technology; α2 integrin (CD49b)-FITC conjugated and β1 integrin (CD29)-Alexa Fluor 488 conjugated from BioLegend; α2 integrin for Western blotting from BD-bioscience; SLC3A2 (CD98) and SLC7A5 (LAT1) from Proteintech and GAPDH from SANTA CRUZ Biotechnology. Secondary antibodies Alexa-Fluor 594 anti-Rabbit IgG and Alexa-Fluor 488 anti-Rabbit IgG were from Cell Signalling; IRDye 800CW (anti-mouse IgG), IRDye 680CW (anti-rabbit IgG), and DRAQ5 were from LI-COR. Alexa Fluor TM 488, 555 and 647 Phalloidin, Click-iT EdU Imaging Kits, PI, NHS-Fluorescein and Hoechst 33342 were from Invitrogen. Rat tail Collagen I (high concentration) was from Corning. All media and dialysed FBS were from Gibco. E64d (Aloxistatin) was from AdooQ Bioscience. BTT-3033 and VPS34-IN were from Cambridge Bioscience. D-phenylalanine was from Santa Cruz Biotechnology. Rapamycin was from Fluorochem.

Cell culture

MDA-MB-231, PANC1 and SW1990 cells were maintained in high glucose Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% foetal bovine serum (FBS) and 1% penicillin/streptomycin (PS). MCF10A, MCF10A-DCIS and MCF10CA1 cell lines, a gift from Professor Giorgio Scita (IFOM, Milan), were cultured in DMEM/F12. MCF10A-DCIS cells were supplemented with 5% Horse serum (HS), 20 ng/ml epidermal growth factor (EGF), and 1% PS. MCF10A and MCF10CA1 cells were grown with 5% HS, 20 ng/ml EGF, 0.2 mg/ml hydrocortisone, 10 μg/ml insulin, and 1% PS. All cell lines were incubated at 37 °C with 5% CO₂ and passaged every 3–4 days.

To culture E0771 mouse triple negative breast organoids, 24-well plates were coated with a 10 µl drop of 10.3 mg/ml Matrigel, polymerised for 10 min at 37 °C to make a dome. Organoids were diluted in 10.3 mg/ml Matrigel and 10µl of Matrigel containing organoids were seeded on top of the dome and incubated upside down at 37 °C for 30 min. Then 1 ml of advanced DMEM/ F12 supplemented with 1% PS, GlutaMAX (1% vol/vol), HEPES (1% vol/vol), hEGF (5 ng/ml), hydrocortisone (0.3 μg/ml) and 10 µM Y-27632 was added. Organoids were incubated at 37 °C with 5% CO₂ and passaged every 7 days.

ECM preparation

Collagen I was diluted in ice-cold PBS to final concentrations of 2 mg/ml and incubated at 37 °C with 5% CO₂ for 3 hours to polymerise.

ECM cross-linking

Polymerised collagen I was treated with 10% glutaraldehyde for 30 min at room temperature (RT). Following two PBS washes, the glutaraldehyde was quenched with 1 M glycine for 20 min at RT, followed by two additional PBS washes. Crosslinked collagen I was then stored overnight at 37 °C with 5% CO₂ in PBS.

Starvation conditions

Glucose- and pyruvate-free media DMEM was supplemented with 10% dialysed FBS (DFBS) for MDA-MB-231, PANC1, and SW1990 cells. Complete media were supplemented with 10% DFBS. Glucose- and pyruvate-free media DMEM/F12 was supplemented with 5% HS, 10 μg/mL insulin, and 20 ng/mL EGF for MCF10A cells; 2.5% HS, 1% PS, 20 ng/ml EGF, 10 μg/ml insulin, 0.2 mg/ml hydrocortisone for MCF10CA1; 5% HS and 20 ng/mL EGF for MCF10A-DCIS cells. Serum-free DMEM was used for serum starvation.

Proliferation assay

96-well plates were coated with collagen I at a concentration of 2 mg/ml, 15 µl per well. Three wells were designated as technical replicates for each experimental condition. 10³ MDA-MB-231, MCF10A, MCF10A-DCIS, PANC1 cells/well, 2x10³ SW1990 cells/well, and 400 MCF10CA1 cells/well were seeded in their respective complete growth media. After a 5-h incubation at 37 °C in a 5% CO₂ atmosphere, the media were replaced with 200 µl of glucose starvation media or complete media with 10% DFBS. Where indicated, the starvation media also contained 5, 10, or 15 µM BTT-3033 or DMSO as a vehicle control. Inhibitors were replenished every 2 days for a total duration of 4 days. Cells were subsequently fixed by the addition of 4% paraformaldehyde (PFA) for 15 min RT, followed by two washes with PBS. Nuclear staining was performed using either DRAQ5 or Hoechst 33342. 5 μM DRAQ5 in PBS was applied to the cells for 1 h at RT with gentle rocking. Cells were then washed twice with PBS for 30 min to minimise background fluorescence before imaging. DRAQ5 fluorescence was detected using a Licor Odyssey Sa instrument via the 700 nm channel with 200 mm resolution. Signal intensity (calculated as total intensity minus total background) for each well was quantified using Image Studio Lite software. Alternatively, cells were fixed directly with 4% PFA containing 10 µg/ml Hoechst 33342 for 15 min, followed by two PBS washes. Images were acquired with an ImageXpress Micro using a 2x objective, ensuring full well coverage of each well. Image analysis was conducted using MetaXpress and Costume Module Editor (CME) software at the Sheffield RNAi Screening Facility (SRSF)

EdU incorporation assay

96-well plates were coated with 2 mg/ml collagen I. 10³ MDA-MB-231 cells/well, 2 x 10³ PANC1 and SW1990 cells/well and 400 MCF10CA1 cells/well were seeded in complete growth media. After a 5-h incubation at 37 °C with 5% CO₂, the complete growth media was replaced with 200 µl of glucose starvation media. At designated time points (day six post-starvation for MDA-MB-231 cells, day five for PANC1, day seven for SW1990 cells and day four for MCF10CA1), cells were incubated with 5 μM EdU for 1–2 days, at 37 °C with 5% CO₂. Cells were then fixed using 4% PFA containing 10 µg/mL Hoechst 33342 for 15 min at RT and subsequently permeabilized with 0.25% Triton X-100 for 5 min. EdU detection was achieved by incubating cells with the Click-iT EdU Alexa Fluor 555 detection cocktail (Invitrogen) or EdU Cell Proliferation Image kit (Antibodies.com) for 30 min at RT with gentle rocking. Following two PBS washes, cells were stored in PBS for imaging. Images were acquired using an ImageXpress Micro system with a 2x objective, and subsequent quantification was performed using MetaXpress and CME software.

PI live-cell imaging

MDA-MB-231, SW1990, and PANC1 cells were seeded in 96-well plates coated with 2 mg/ml collagen I in complete growth media. Following a 5-h incubation period at 37 °C with 5% CO₂, the complete growth media was replaced with glucose starvation media. At day 4, 5, 6 or 7 post-starvation, cells were co-incubated with 1 µg/ml PI and 5 µg/mL Hoechst 33342 for 30 min at 37 °C with 5% CO₂. Cells were then washed once with PBS and maintained in phenol red-free media for imaging. Images were acquired live using an ImageXpress Micro system with a 10x objective, and subsequent quantification was performed with MetaXpress and CME software.

Cleaved Caspase 3/7 staining

MCF10CA1 cells were seeded 96-well plates coated with 2 mg/ml collagen I in complete growth media. Following a 5-h incubation at 37 °C with 5% CO₂, the complete growth media was replaced with glucose starvation media. On day 6 post-starvation media was changed to PBS containing 5 μM Cell Event Caspase-3/7 Green Detection Reagent for 1 h and 30 min. Cells were fixed and stained with Hoechst 33342. Images were collected using an Image Xpress micro system with a 10x objective and quantified with MetaXpress and CME software.

Immunofluorescence

For immunofluorescence staining of α2 integrin, β1 integrin and pS6 1 x 10⁴ cells were seeded in 96-well plates. For SLC3A2 and SLC7A5 staining, 2 x 10⁵ cells were seeded in 3.5 cm glass-bottom dishes. Cells were fixed with 4% PFA for 15 min, followed by two washes with PBS. For surface staining of α2 integrin, β1 integrin, SLC3A2, and SLC7A5, cells were not permeabilized. Instead, they were directly incubated with 3% bovine serum albumin (BSA) in PBS for 1 h at RT for blocking. For pS6 staining, cells were permeabilized with 0.25% Triton X-100 for 5 min at RT, followed by washes and subsequent blocking with 3% BSA in PBS for 1 h at RT. Following blocking, cells were incubated with primary antibodies (1:200 dilution) for 2 h at RT, followed by two PBS washes. Cells were then incubated with an Alexa Fluor 488 or Alexa Fluor 555 anti-rabbit IgG secondary antibody (1:1000) and Phalloidin Alexa Fluor 555 or Alexa Fluor 647 (1:1000) for 1 h at RT, followed by two final PBS washes. Cells were stained with 5 µg/mL Hoechst 33342 and kept in PBS for imaging. α2 and β1 integrin images were collected by an ImageXpress Micro system with a 10x objective, while SLC3A2, SLC7A5, and pS6 images were collected by a Zeiss LSM980 Airyscan 2 System with a 40x 1.2 NA water (SLC3A2 and SLC7A5) or a 20x 0.8 NA (pS6) objective.

ECM uptake

3.5 cm glass-bottomed dishes were coated with 2 mg/ml collagen I and incubated for 3 h at 37 °C with 5% CO₂ for polymerisation. Collagen I was then labelled with 10 µg/ml Fluorescein succinimidyl ester (NHS) for 1 h at RT on a gentle rocker. Following labelling, 1 x 10⁵ MDA-MB-231 cells/dish were seeded in complete growth media. After a 5-hour incubation at 37 °C with 5% CO₂, the complete growth media was replaced with 1 ml of glucose starvation media. Cells were treated with either 20 μM E64d or DMSO. Cells were fixed on day three by adding 4% PFA for 15 min at RT. Fixed cells were permeabilised with 0.25% Triton X-100 for 5 min and then incubated with Phalloidin Alexa Fluor 555 (1:400 in PBS) for 10 min to visualise actin. Dishes were mounted with 2–3 drops of Vectashield mounting medium containing DAPI and sealed with parafilm, then stored at 4 °C. Cells were visualised using a Nikon A1 confocal microscope equipped with a 60x 1.4 NA oil immersion objective. Collagen I uptake index was quantify with Fiji/ImageJ [75] as in Nazemi and colleagues, 2024 [23].

Western blotting

MDA-MB-231, MCF10CA1, and PANC1 cells were lysed with lysis buffer (50 mM Tris (pH 7) and 1% SDS) and transferred to Qiashredder columns (Qiagen) to remove the DNA. Lysates were diluted 4:1 in NuPAGE-LDC sample buffer and run on 4% to 15% Mini Protean TGX gels in SDS-PAGE running buffer (25 nM Tris base, 192 mM glycine, and 0.1% SDS) at 100 V for 1 hour and 15 min. Proteins were transferred to Immobilon-FL PVDF membrane (MERCK) in transfer buffer (10% Towbin buffer, 20% methanol) at 100 V for 75 min. Membranes were blocked in TBS-T with 5% w/v skimmed milk or 5% w/v BSA. Then, membranes were incubated with 1:1,000 GAPDH or 1:5000 α-tubulin together with primary antibodies for LC3, pS6, S6, α2 integrin and SLC3A2 1:1000 in TBS-T with 5% w/v skimmed milk or 5% w/v BSA overnight at 4 ºC. Membranes were incubated with secondary antibodies, IRDye 800CW anti-mouse IgG for α2 integrin, GAPDH and α-tubulin (1:30,000) and IRDye 680CW anti-rabbit IgG for LC3, pS6, S6 and SLC3A2 (1:20,000) in TBS-T with 0.01% SDS, for 1 h at RT on the rocker. Membranes were washed 2 times in TBS-T for 15 min on the rocker at RT. Images were taken by a Licor Odyssey Sa system. Band intensity was quantified with Image Studio Lite software.

Nontargeted metabolite profiling

MDA-MB-231, MCF10CA1, SW1990, and PANC1 cells were cultured on collagen I-coated 6-well plates in complete growth media. Following a 5-h incubation, the media was changed to glucose starvation media for up to 4 days. Metabolite extraction involved washing cells with ice-cold PBS, followed by incubation with a cold 5 MeOH: 3 AcN: 2 H₂O solution. After centrifugation, samples were analysed by electrospray ionisation mass spectrometry (Waters G2 Synapt) at the Sheffield Faculty of Science Biological Mass Spectrometry Facility (biOMICS), with data acquired in both positive and negative ion modes. Three technical replicates per sample were run, and only peaks consistently present across all replicates were included. Data were binned (0.2 amu m/z), and putative metabolites were identified using the HumanCyc database. Data analysis was performed using Perseus software (v1.5.6.0) and used Student t test (SO = 0.1, false discovery rate (FDR) = 0.05 and p < 0.05), and metabolic pathway analysis was conducted with MetaboAnalyst 5.0 (p < 0.05).

Targeted metabolomics

Metabolites extraction was performed as above. A mass spectrometer Waters Synapt G2-Si coupled to Waters Acquity UPLC was used to separate alanine, asparagine, aspartic acid, arginine, cysteine, glutamic acid, glycine, glutamine, histidine, isoleucine, lysine, leucine, phenylalanine, methionine, serine, proline, tryptophan, threonine, tyrosine, and valine with separation was carried out on a Waters BEH C18 column (50 x 2.1 mm) at 40 °C, with an injection volume of 5 μl. The mobile phases consisted of 0.1% formic acid (A) and 0.1% acetonitrile (B). The flow rate was 0.4 ml/min, and the gradient program was as follows: 1%–35% B (0–3 min), 35%–99% B (3–6 min), 99% B (6–6.9 min), 99%–1% B (6.9–7 min), and 1% B (7–8 min). The samples were run in positive and negative modes. To identify the targeted metabolites, the retention time and the mass of the compound were matched with the standards. To account for analytical variation, all peak intensities were normalised to the total ion count (TIC) for each sample prior to statistical analysis. This was done by dividing each individual ion intensity by the sum of all detected ion intensities within the respective sample. A minimum of three independent biological replicates were performed, with one well of the 6-well plate dedicated to each experimental condition per replicate. For 13C-Tyrosine uptake experiments, cells were seeded on 2 mg/ml collagen I in complete media for 3 h, media was changes to glucose starvation media, in the presence of 10μM BTT-3033, 50mM D-Phenylalanine or DMSO control and cells were incubated overnight. Cells were washed with PBS, incubated with PBS containing 10% DFBS for 15 min then 13C-Tyrosine was added for 30 min. Metabolites were extracted and analysed as above.

siRNA Transfection

For cell proliferation experiments, Dharmacon ON-TARGETplus siRNA smart pools (150 nM, 20 µl per well) were mixed with Dharmafect 1 (DF1, 0.24 µl in 19.76 µl DMEM, 20 µl per well) and incubated in a 96-well plate for 30 min at RT. Following this, MDA-MB-231 and PANC1 cells were seeded in 60 µl of 10% FBS DMEM (antibiotic-free), while MCF10CA1 cells were seeded in 5% HS DMEM/F12, establishing a 30 nM final siRNA concentration. Cells were maintained overnight at 37 °C and 5% CO₂. Media was then changed to 200 µl of either complete (with DFBS) or glucose starvation media for up to 4 days. Cells were fixed, stained with 10 µg/ml Hoechst 33342, and imaged using ImageXpress Micro. Images were analysed with MetaXpress and CME software.

For western blotting, 3D spheroids and metabolite profiling, in a 6-well plate, 10 µl of 5 µM siRNA was mixed with 190 µl of Opti-MEM per well (25 nM final concentration). Separately, 2 µl of DF1 was combined with 198 µl of Opti-MEM and incubated for 5 min at RT. The Opti-MEM/DF1 mixture (200 µl) was then added to the siRNA solution and incubated on a rocker for 20 min. Finally, cells in 1.6 ml of media were added to each well. After 24 hours of incubation at 37 °C and 5% CO₂, the media was replaced with 3 ml of glucose starvation media (Table 1).

Table 1. siRNA smart pools used in this study.

Type	Target	Ref No.
si-GENOME siRNA smartpool	Human SLC7A5	M-004953-01-0005
ON-TARGETplus siRNA smartpool	Human SLC3A2	L-003542-00-0005
ON-TARGETplus siRNA smartpool	Human ITGA2	L-004566-00-0005
ON-TARGETplus siRNA smartpool	Human ITGB1	L-004506-00-0005
ON-TARGETplus siRNA	Non-targeting siRNA #4	D-001810-04-20

RT-QPCR

mRNA was extracted from MDA-MB-231 and PANC1 cells using the Qiagen RNeasy Mini kit, following manufacturer’s instructions. Cells grown on plastic were trypsinised, while those on collagen I were scraped; both were then pelleted, washed, and snap-frozen. Lysis in Buffer RLT and purification through spin columns, followed by washes, yielded purified mRNA. mRNA was then used for cDNA synthesis with the High-Capacity cDNA Reverse Transcription Kit (Fisher), where 1 μg of mRNA per sample was reverse transcribed in a 20 μl reaction containing RT buffer, dNTP mix, random primers, and MultiScribe Reverse Transcriptase. A negative reverse transcriptase control was included. The synthesised cDNA was stored at −80 °C. Finally, qPCR analysis involved preparing a master mix with QuantiNova SYBR Green PCR Kit and QuantiTect Primer Assay. 7 μl of this mix was combined with 3 μl of 1:100 diluted cDNA (5 ng/μl final concentration) in a 384-well plate. -RT and blank water controls were included for all target genes. Samples were run on a Quantstudio 12K flex real-time PCR system for 40 cycles, with a melting curve analysis to confirm product specificity. Target gene expression was calculated using the 2^−ΔCt method, normalising to GAPDH as the housekeeping gene (Table 2).

Table 2. RT-QPCR primers used in this study.

Primer	Assay name	Cat No.
GAPDH	Hs_GAPDH_1_SG	QT00052248
SLC3A2	Hs_SLC3A2_1_SG	QT00085897
SLC7A5	Hs_SLC7A5_1_SG	QT00089145
SLC7A6	Hs_SLC7A6_1_SG	QT00002674
SLC7A8	Hs_SLC7A8_1_SG	QT00006384

Stable cell line generation

MDA-MB-231 cells were plated in 6-well plates and allowed to grow to high confluency. On day 2, a Lipofectamine 2000-DNA complex was prepared by combining diluted Lipofectamine (5 μl LF in 250 μl Opti-MEM per well) with the 2.5 μg pEGFP-C1+ mRFP LC3 plasmid in 250 μl Opti-MEM, incubated 20 min, and then added to cells in antibiotic-free DMEM. After overnight incubation, the transfection solution was replaced with complete media. Cells were grown in the presence of 1 mg/ml G418 2–3 days post-transfection, when cells reached 50%–70% confluency. GFP- and RPF-positive single cell clones were sorted using a Sony MA900 cell sorter, selecting for cells with high intensity for both RFP and GFP.

3D Spheroid Generation

MDA-MB-231, stably expressing GFP as described in Nazemi and colleagues, 2024 [23], and PANC1 cell 3D spheroids were generated via the hanging drop method. Each 20 μl drop contained 500 cells suspended in a mixture of 4.8 mg/ml methylcellulose and 20 μg/ml soluble collagen I. After 48 h, spheroids were embedded in a matrix mix composed of 3 mg/ml collagen I and 3 mg/ml Geltrex. To avoid spheroids sinking to the bottom of the plate, the plate was slowly turned upside-down and incubated for 30 min at 37 °C. Then 1 ml media was added to each well.

For drug treatment studies, the media was changed to glucose starvation media 24 hours post-embedding, and spheroids were subsequently treated with either 10 μM BTT-3033 or DMSO control. In siRNA knockdown experiments, cells were transfected for 24 h prior to spheroid formation and embedding in the collagen I and Geltrex matrix.

MDA-MB-231 spheroids were imaged using a Nikon A1 confocal microscope with CFI Plan Fluor 10x (NA 0.3) objective. For drug treatment assays, imaging occurred from day 1 to day 3 post-embedding. For siRNA knockdown experiments, imaging was performed on day 1 and day 4 post-embedding. The invasion area was quantified by subtracting the spheroid’s core area from its total area.

PANC1 spheroids were imaged every 2 days up to day 6 post-starvation by an Olympus E450 microscope with 10x objective, with drug treatments being refreshed every 2 days.

Mouse tumour organoid invasion

E0771 mouse triple negative breast tumour organoids were grown in 3 mg/ml collagen I and 3 mg/ml Geltrex (1:1). 24-well plates were coated with a 10 µl drop of 10.3 mg/ml Matrigel, polymerised for 10 min at 37 °C to make a dome. 10 µl of organoids in Geltrex-collagen I mixture were added on top of the dome and incubated upside down at 37 °C for 30 min. Then 1 ml of advanced DMEM/F12 supplemented with 1% PS, GlutaMAX (1% vol/vol), HEPES (1% vol/vol), hEGF (5 ng/ml), hydrocortisone (0.3 μg/ml), and 10 µM Y-27632 was added and organoids were incubated at 37 °C with 5% CO_2. After 1 day, the media was changed to glucose starvation media. Cells were treated with DMSO (Ctrl), 2.5 or 5 µM BTT-3033 or 50 mM D-phenylalanine. Cells were imaged live every day up to 3 days with an Olympus E450 microscope with 4x and 10x objectives. Size of organoids was measured by quantifying the core area of each organoid. Invasion length was quantified by measuring the length of invasive branches. The resulting data points reflect individual measurements, with each point in the graph representative of a single protrusion’s length.

Expression, survival, and ROC analysis

RNA sequencing data from basal-like breast cancer, pancreatic adenocarcinoma, normal breast and pancreatic tissue and survival analysis were performed in gepia2 (http://gepia2.cancer-pku.cn/#index), using TCGA/GTEx data. The link between gene expression and response to therapy (including taxane, anthracycline, ixabepilone, CMF, FAC, and FEC) was analysed in ROC plotter (https://www.rocplot.com/), using transcriptome-level data of breast cancer patients [36]. Breast cancer datasets were identified in GEO (https://www.ncbi.nlm.nih.gov/gds), using the platform IDs “GPL96”, “GPL570”, and “GPL571”. Jetset probes were used (https://services.healthtech.dtu.dk/services/jetset/).

Statistical analysis

Graphs were generated using GraphPad Prism software (versions 10). To compare two datasets, a Mann–Whitney U test was employed. For more than two datasets with a single independent variable, one-way ANOVA (Kruskal–Wallis, Dunn’s multiple comparisons test) was applied. When two independent variables were present, two-way ANOVA (Tukey’s multiple comparisons test) was utilised. Metabolomics data were analysed in Perseus software.

Supporting information

S1 Fig. Collagen I supported the growth of invasive breast cancer cells, but not non-invasive or non-transformed mammary epithelial cells.

(A–C) MCF10A, MCF10A-DCIS and MCF10CA1 cells were seeded on plastic (P) or 2 mg/ml collagen I (Coll) for 6 or 8 days under Glc starvation, fixed, stained with DRAQ5, and imaged with a Licor Odyssey Sa system. Signal intensity was calculated by Image Studio Lite software. Data are presented as mean ± SEM, N = 3 independent experiments. ****p < 0.0001 two-way ANOVA, Tukey’s multiple comparisons test. (D) MCF10CA1 cells were seeded on 2 mg/ml collagen I (Coll) or plastic (P) under Glc starvation. Cells were incubated with EdU at day 4 post starvation, fixed and stained with Hoechst 33342 and Click iT EdU imaging kit at day 6. (E) MCF10CA1 were seeded on plastic (P) or 2 mg/ml collagen I (Coll) for 6 days. Cells were fixed, stained for cleaved Cas3/7 and Hoechst 33342. Images were collected by ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N ≥ 3 independent experiments. ***p < 0.001, ****p < 0.0001 Mann–Whitney test. All the raw data associated with this figure are available in S10 Data.

(TIF)

S2 Fig. Collagen I supported cell growth and prevented cell death in SW1990 pancreatic cancer cells.

(A) SW1990 cells were seeded on 2 mg/ml collagen I (Coll) or plastic (P) for 7 days under Glc starvation, fixed, stained with DRAQ5 and imaged with a Licor Odyssey Sa system. Signal intensity was calculated by Image Studio Lite software. Data are presented as mean ± SEM, N = 3 independent experiments. ***p < 0.001 two-way ANOVA, Tukey’s multiple comparisons test. (B) SW1990 cells were seeded on 2 mg/ml collagen I (Coll) or plastic (P) under Glc starvation, cells were incubated with EdU at day 7 post starvation, fixed and stained with Hoechst 33342 and Click iT EdU imaging kit at day 7. (C) SW1990 cells were seeded as in B and treated with PI at day 7. Images were collected by ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N ≥ 3 independent experiments. *p < 0.05, **p < 0.01, Mann–Whitney test. All the raw data associated with this figure are available in S11 Data.

(TIF)

S3 Fig. Glucose starvation promoted collagen I internalisation.

(A) MDA-MB-231 cells were seeded on 2 mg/ml NHS-Fluorescein labelled collagen I (yellow) under complete (Com) or Glc starvation media in the presence of DMSO or 20μM E64d for 3 days. Cells were fixed and stained for actin (magenta) and nuclei (cyan). Images were collected by Nikon Confocal A1 microscope. Bar, 20 μm. (B) Collagen I uptake index was measured with Fiji/ImageJ. Data are presented as mean ± SEM, n = 3 independent experiments (the bigger dots represent the mean of individual experiments). **p < 0.01, ****p < 0.0001. One-way ANOVA Kruskal–Wallis, Dunn’s multiple comparisons test. All the raw data associated with this figure are available in S12 Data.

(TIF)

S4 Fig. α2β1 integrin affected cell growth in complete media in a cell line-dependent manner.

MDA-MB-231 (A, B) and PANC1 (E, F) cells were seeded on plastic (A,C) or on 2 mg/ml collagen I (B,D) and treated with 5 μM, 10 μM BTT-3033 (BTT) or DMSO (Ctrl) for 4 days in complete media. MDA-MB-231 (C, D) and PANC1 (G, H) cells were seeded on plastic (C,G) or on 2 mg/ml collagen I (D,H) and transfected with an siRNA targeting β1 integrin (β1 si), an siRNA targeting α2 integrin (α2 si) or a nontargeting siRNA control (Nt si) for 4 days in complete media. Cells were fixed and stained with Hoechst 33342. Images were collected by ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N = 3 independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001Kruskal–Wallis, Dunn’s multiple comparisons test. All the raw data associated with this figure are available in S13 Data.

(TIF)

S5 Fig. α2β1 integrin-targeting siRNAs significantly reduced protein expression.

(A) MDA-MB-231 and (B) PANC1 cells were transfected with a nontargeting control siRNA (Nt si), an siRNA targeting α2 integrin (α2 si) or an siRNA targeting β1 integrin (β1 si) for 4 days. Cells were fixed and stained with Hoechst 33342 (blue) and α2 integrin or β1 integrin (grey). Images were collected by ImageXpress micro. Bar, 50 μm. Integrin signal intensity was quantified by MetaXpress software. Data are presented as mean ± SEM, N = 3 independent experiments. ****p < 0.0001 Mann–Whitney test. All the raw data associated with this figure are available in S14 Data.

(TIF)

S6 Fig. α2 integrin was required for MCF10CA1 cell growth.

(A) MCF10CA1 cells were seeded on 2 mg/ml collagen I and grown under glucose starvation in the presence of 15 μM BTT-3033 (BTT) or DMSO (Ctrl) control for 4 days. (B) MCF10CA1 cells were transfected with a nontargeting control siRNA (Nt si) or an siRNA targeting α2 integrin (α2 si) and grown on 2 mg/ml collagen I under glucose starvation for 4 days. Cells were fixed and stained with Hoechst 33342. Images were collected by ImageXpress micro. Data are presented as mean ± SEM, N = 3 independent experiments. ****p < 0.0001 Mann–Whitney test. (C) MCF10CA1 cells were transfected as in B, lysed and the levels of α2 integrin and GAPDH were measured by Western blotting. All the raw data associated with this figure are available in S15 Data.

(TIF)

S7 Fig. Collagen I did not affect the phosphorylation of mTOR or 4EBP1.

MDA-MB-231 cells were seeded on 2 mg/ml collagen I (+) or plastic (−) under complete (A) or Glc deficient media (A, B) for 1 day. Lysates were collected and the levels of phosphorylated 4EBP1 (p4EBP1, A), mTOR, phosphorylated mTOR (pmTOR, B) and GAPDH were measured by Western blotting. Data are presented as mean ± SEM, N ≥ 4 independent experiments. *p < 0.05 Kruskal–Wallis, Dunn’s multiple comparisons test. (C) MDA-MB-231 and PANC1 cells were seeded on 2 mg/ml collagen I in the presence of 100 nM Rapamycin or DMSO control for 1 day. Lysates were collected and the levels of phosphorylated S6 (pS6), GAPDH and tubulin were measured by Western blotting. Data are presented as mean ± SEM, N = 4 independent experiments. *p < 0.05 Mann–Whitney test. All the raw data associated with this figure are available in S16 Data.

(TIF)

S8 Fig. Collagen I did not affect autophagosome maturation.

MDA-MB-231 cells stably expressing GFP-RFP-LC3 were seeded on 2 mg/ml collagen I (Coll) or plastic (P) under complete or Glc deficient media for 1 day. Cells were fixed and stained with Hoechst 33342 and Phalloidin Alexa Fluor 647. Images were collected with a ZEISS LSM980 Airyscan2 confocal microscope and analysed by Fiji/ImageJ software. Data are presented as mean ± SEM, N = 3 independent experiments (the bigger dots represent the mean of individual experiments). Non-significant, Kruskal–Wallis, Dunn’s multiple comparisons test. All the raw data associated with this figure are available in S17 Data.

(TIF)

S9 Fig. Collagen I increased intracellular amino acid levels in MCF10CA1 and SW1990 cells under glucose starvation.

(A) Metabolomics workflow. MCF10CA1 and SW1990 cells were plated on plastic or 2 mg/ml collagen I (Coll) in Glc-free media for 1 day. Metabolites were extracted and quantified by non-targeted mass spectrometry. Volcano plot (B, D) and enriched metabolic pathways (C, E) are shown. These datasets are deposited in ORDA (https://doi.org/10.15131/shef.data.21608490).

(TIF)

S10 Fig. Collagen I crosslinking did not affect intracellular metabolite levels.

(A) Metabolomics workflow. (B) MDA-MB-231 cells were plated on 2 mg/ml collagen I (Coll) or 10% glutaraldehyde-crosslinked 2 mg/ml collagen I (X-linked Coll) for 1 day in Glc free media. Metabolites were extracted and quantified by nontargeted mass spectrometry. Volcano plot is shown. This dataset is deposited in ORDA (https://doi.org/10.15131/shef.data.21608490).

(TIF)

S11 Fig. Collagen I marginally affected SLC3A2 and SLC7A5 localisation in complete media and α2β1 integrin was not the main regulator of collagen I-driven SLC7A5 localisation under Glc starvation.

(A, B) MDA-MB-231 and (C, D) PANC1 cells were seeded on 2 mg/ml collagen I (Coll) or plastic (P) in complete media for 1 or 2 days, respectively. Cells were fixed and stained with Hoechst 33342 (magenta or cyan), SLC3A2 (cyan) or SLC7A5 (red) and Phalloidin Alexa Fluor 555 (grey). (E) MDA-MB-231 and (F) PANC1 cells were seeded on 2 mg/ml collagen I under Glc-free media in presence or absence of 10 μM BTT-3033 (BTT) for 1 or 2 days, respectively. Cells were fixed and stained with Hoechst 33342 (cyan), SLC7A5 (red) and Phalloidin Alexa Fluor 555 (grey). Images were collected with a ZEISS LSM980 Airyscan2 microscope and analysed by Fiji/ImageJ software. Bar, 20 µm Data are presented as mean ± SEM, N = 3 independent experiments (the bigger dots represent mean intensity of each field of view); *p < 0.0209, Mann–Whitney test. All the raw data associated with this figure are available in S18 Data.

(TIF)

S12 Fig. LAT1-4F2hc, but not α2β1 integrin, was required for tyrosine import.

MDA-MB-231 cells were seeded on 2 mg/ml collagen I (Coll) in Glc-free media for 24 h in the presence of 10 μM BTT-3033 (BTT), 50 mM D-Phenylalanine (D-Phe), or DMSO control, incubated with 13C Tyrosine (13C-Tyr) for 30 mins in PBS, metabolites were extracted and 13C-Tyr was measured by targeted mass spectrometry. N = 6 independent experiments, *p < 0.05 Kruskal–Wallis, Dunn’s multiple comparisons test. All the raw data associated with this figure are available in S19 Data.

(TIF)

S13 Fig. LAT1-4F2hc and α2β1 integrin were required for S6 phosphorylation in the presence of collagen I under Glc starvation.

MDA-MB-231 cells were grown on 2 mg/ml collagen I for 1 day, pretreated with 10 μM BTT-3033 for 24 hours and starved in Glc-free media (A) or kept in complete media (B) for 2 h, in the presence of 10 μM BTT-3033 (BTT), 50 mM D-Phenylalanine (D-Phe), or DMSO (Ctrl). Protein level of phosphorylated S6 (pS6), GAPDH and tubulin were measured via Western blotting. Data are presented as mean ± SEM, N = 4 independent experiments. *p < 0.05 Mann–Whitney test (A) or Kruskal–Wallis, Dunn’s multiple comparisons test (B, C). All the raw data associated with this figure are available in S20 Data.

(TIF)

S14 Fig. SLC7A5, but not SLC3A2, was required for cell growth on plastic in complete media.

MDA-MB-231 (A, B) and PANC1 (E–H) cells were transfected with an siRNA targeting SLC3A2 (3A2 si), an siRNA targeting SLC7A5 (7A5 si) or a non-targeting siRNA control (Nt si) and grown on plastic under complete (A,C) or Glc-free media (B,D) for 4 days. Cells were fixed and stained with Hoechst 33342. Images were collected by an ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N = 3 independent experiments. *p < 0.05, **p < 0.01 and ***p < 0.001 Kruskal–Wallis, Dunn’s multiple comparisons test. All the raw data associated with this figure are available in S21 Data.

(TIF)

S15 Fig. Collagen I supported cell growth and S6 phosphorylation in breast cancer cells under serum starvation.

(A, B) MDA-MB-231 (A) or PANC1 cells (B) were seeded on plastic (P) or 2 mg/ml collagen I (Coll) for 5 days under serum starvation (FBS-free), fixed and stained with DRAQ5. Images were collected by a Licor Odyssey Sa system and analysed by Image Studio software. Data are presented as mean ± SEM, N = 3 independent experiments. *p < 0.05, ***p < 0.001 two-way ANOVA, Tukey’s multiple comparisons test. (C) MDA-MB-231 cells were grown on plastic (P) or 2 mg/ml collagen I (Coll) for 1 day under serum starvation, fixed, stained for phosphorylated S6 (pS6), imaged with ImageXpress micro and analysed by MetaXpress software. Data are presented as mean ± SEM, N = 3 independent experiments. ***p < 0.001 Mann–Whitney test. (D) MDA-MB-231 cells were grown on 2 mg/ml collagen I (Coll) or plastic (P), under serum starvation for 1 day. Metabolites were extracted and analysed by targeted mass spectrometry. The fold change relative to collagen I is presented. N = 3 independent experiments. All the raw data associated with this figure are available in S22 Data.

(TIF)

S1 Raw Images. Original western blot image used to prepare Figs 4B, 4C, 5C, 5D, 8A, 8F, 8K, 9E, 9J, S6C, S7A, S7B, S7C and S13.

(PDF)

S1 Data. Numerical data used for the generation of the graphs presented in Fig 1.

(XLSX)

S2 Data. Numerical data used for the generation of the graphs presented in Fig 2.

(XLSX)

S3 Data. Numerical data used for the generation of the graphs presented in Fig 3.

(XLSX)

S4 Data. Numerical data used for the generation of the graphs presented in Fig 4.

(XLSX)

S5 Data. Numerical data used for the generation of the graphs presented in Fig 5.

(XLSX)

S6 Data. Numerical data used for the generation of the graphs presented in Fig 6.

(XLSX)

S7 Data. Numerical data used for the generation of the graphs presented in Fig 7.

(XLSX)

S8 Data. Numerical data used for the generation of the graphs presented in Fig 8.

(XLSX)

S9 Data. Numerical data used for the generation of the graphs presented in Fig 9.

(XLSX)

S10 Data. Numerical data used for the generation of the graphs presented in S1 Fig.

(XLSX)

S11 Data. Numerical data used for the generation of the graphs presented in S2 Fig.

(XLSX)

S12 Data. Numerical data used for the generation of the graphs presented in S3 Fig.

(XLSX)

S13 Data. Numerical data used for the generation of the graphs presented in S4 Fig.

(XLSX)

S14 Data. Numerical data used for the generation of the graphs presented in S5 Fig.

(XLSX)

S15 Data. Numerical data used for the generation of the graphs presented in S6 Fig.

(XLSX)

S16 Data. Numerical data used for the generation of the graphs presented in S7 Fig.

(XLSX)

S17 Data. Numerical data used for the generation of the graphs presented in S8 Fig.

(XLSX)

S18 Data. Numerical data used for the generation of the graphs presented in S11 Fig.

(XLSX)

S19 Data. Numerical data used for the generation of the graphs presented in S12 Fig.

(XLSX)

S20 Data. Numerical data used for the generation of the graphs presented in S13 Fig.

(XLSX)

S21 Data. Numerical data used for the generation of the graphs presented in S14 Fig.

(XLSX)

S22 Data. Numerical data used for the generation of the graphs presented in S15 Fig.

(XLSX)

Abbreviations

AMPK

activated protein kinase

BSA

bovine serum albumin

CME

Costume Module Editor

DDR2

discoidin domain receptor 2

DMEM

Dulbecco’s Modified Eagle’s Medium

ECM

extracellular matrix

EdU

ethynyl-2′-deoxyuridine

EGF

epidermal growth factor

FACS

Fluorescence-Activated Cell Sorting

FBS

foetal bovine serum

FDR

false discovery rate

GIC

Glucose Insensitive Cells

GSC

Glucose Sensitive Cells

HS

Horse serum

LAT1

large neutral amino acid transporter

LC3

microtubule-asosciated protein 1A/1B-light chain 3

MAPK

mitogen activated protein kinase

PDAC

pancreatic ductal adenocarcinoma

PI

Propidium Iodide

PS

penicillin/streptomycin

ROS

reactive oxygen species

RT

room temperature

SRSF

Sheffield RNAi Screening Facility

TIC

total ion count

TME

tumour microenvironment.

Data Availability

All relevant data are within the paper and its Supporting information files. Metabolomics raw data are available at ORDA via https://doi.org/10.15131/shef.data.21608490.

Funding Statement

MN, BY and ER are funded by Cancer Research UK (C52879/A29144; https://www.cancerresearchuk.org/). ER is also funded by Breast Cancer Now (2023.11PR1656; https://breastcancernow.org/) and Yorkshire Cancer Research (YCRSPF\2024\100093; https://www.yorkshirecancerresearch.org.uk/). The Wolfson Light Microscopy Facility was funded by the Wellcome Trust, grant WT093134AIA. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.