Growth factors maintain intratumoral heterogeneity and drive therapeutic resistance in triple-negative breast cancer

TL;DR

This study shows that growth factors like HGF and NRG1 help triple-negative breast cancer resist treatment and maintain tumor diversity.

Contribution

The paper reveals that growth factors maintain tumor heterogeneity and resistance to trametinib in TNBC.

Findings

HGF and neuregulin 1 drive therapeutic resistance in TNBC.

HGF inhibitors can restore sensitivity to trametinib.

High HGF expression in patients correlates with poor outcomes and increased mesenchymal markers.

Abstract

Triple-negative breast cancer (TNBC) shows considerable intratumoral heterogeneity, which contributes to therapeutic resistance. Recent studies show that targeted therapeutics can steer TNBC toward homogeneous, drug-resistant states, but little is understood about how the microenvironment modulates these responses. We report studies to determine how components of the microenvironment impact response to trametinib and cellular heterogeneity. We find that multiple microenvironmental factors, including HGF and neuregulin 1, can drive therapeutic resistance and that treatment with hepatocyte growth factor (HGF) inhibitors restores trametinib sensitivity. Interestingly, treatment with these ligands reverses trametinib-induced homogeneity, restoring heterogeneity to levels comparable to baseline both in vitro and in vivo. Analysis of patient data demonstrates that TNBC with high HGF expression levels has a poor outcome and increased expression of basal and mesenchymal state markers. Our data suggest that common growth factors drive therapeutic resistance and maintain tumor heterogeneity in TNBC, and that co-targeting these factors may improve therapeutic response.

Devlin et al. demonstrate that growth factors like HGF and NRG1 can cause resistance to targeted inhibitors like trametinib in TNBC. These growth factors function by maintaining low levels of proliferation as well as the intratumoral heterogeneity that is characteristic of TNBC.

Figures

A](#F1)). EdU incorporation showed a more limited set of ligands that enhanced proliferation, which consisted mainly of EGF family ligands and HGF ([Figure 1B](#F1)). ECMs that enhanced proliferation under trametinib treatment again included collagen 1, FN1, laminin, and vitronectin ([Figure 1B](#F1)). Of note, the subtle changes in cell number and EdU incorporation we observed contrasted with our previous findings in HER2^+^ breast cancer cell lines.^[15](#R15)^ Specifically, in HCC1143 cells, a single dominant driver of resistance to therapy was not observed, and the effects of the ECM molec

B](#F2)). In contrast, these factors were only active in SUM149PT and MDAMB468 cells at the lowest concentration, giving rise to significantly higher cell numbers than PBS-only controls with 25 nM trametinib ([Figure 2B](#F2)). Interestingly, we also observed that the combination of HGF and NRG1β was less effective in MDAMB468 than either of the ligands alone ([Figure 2B](#F2)). Additionally, we studied the response in these cell lines using live cell imaging. Again, we observed that these factors could confer resistance to trametinib in all three cell lines ([Figure 2C](#F2)) as the cells gre

D](#F3)). Treatment with trametinib led to a highly significant decrease in heterogeneity compared to the control condition. Treatment with NRG1β was able to somewhat restore the heterogeneity following trametinib treatment, but the heterogeneity remained significantly lower than in untreated controls. Most interestingly, HGF treatment by itself or in combination with NRG1β was able to fully restore heterogeneity from the trametinib treatment condition, such that it was not significantly different from baseline controls ([Figure 3D](#F3)). To measure how much ligand treatment helps to reverse

A](#F4)). Quantifying the data, we saw that 30% of cells at baseline expressed both MET and KRT14. Following treatment with trametinib, this double-positive population increased to 68% ± 0.018% (standard error of the mean, SEM) of the population ([Figure 4B](#F4)). Treatment with NRG1β and trametinib maintained the KRT14^+^/MET^+^ population at 68% ± 0.002%, but trametinib and HGF treatment decreased the double-positive cell population to 47% ± 0.013% ([Figure 4B](#F4)). Treatment with both ligands in the presence of trametinib had an intermediate number of double-positive cells (53% ± 0.019%

D](#F5)), we chose to use HGF as the ligand. We obtained custom implantable pellets that contained ~1 μg of HGF that uniformly release ligand over 21 days (equivalent to ~48 ng of HGF released daily). We implanted 2 pellets per mouse to increase the daily HGF released to ~96 ng per mouse. We orthotopically implanted 2 × 10^6^ SUM149PT cells into both the left and right abdominal mammary fat pads of 18 immune-deficient SCID mice. After ~14 days, we could detect small palpable lesions in 26 of the glands. At this point, mice were randomized into 4 groups and were treated as follows: (1) placebo

Full text

INTRODUCTION

Triple-negative breast cancer (TNBC) is a clinically defined subtype of breast cancer that lacks expression of hormone and HER2 receptors and has a poor prognosis.^1–3^ A subset of TNBC that belongs to the intrinsic molecular subtype known as basal breast cancer^1,4,5^ is characterized by a high degree of intratumoral heterogeneity at both clonal^6^ and differentiation state levels.^7,8^ It is thought that this heterogeneity and associated plasticity play a central role in therapeutic resistance in basal tumors, as resistant subclones can rapidly emerge from the heterogeneous population under therapeutic pressure.^9–13^ Indeed, TNBC tends to be more resistant to therapy than its non-TNBC counterparts, resulting in poorer clinical outcomes in the basal/TNBC subgroup compared to the luminal subtype tumors.^2^ A recent study from researchers in our Cancer Systems Biology Consortium group confirmed that heterogeneity and plasticity can drive resistance.^10^ In this study, they showed that small-molecule inhibitors targeting MEK kinase (mitogen-activated protein kinase kinase) or phosphatidylinositol 3-kinase (PI3K) resulted in steering the cells into quiescent basal or luminal and mesenchymal-positive lineages that were resistant to therapy.^10^ Removal of therapeutic pressure led to the rapid reestablishment of heterogeneity and the re-initiation of active cell cycling. Therefore, it was concluded that drug combinations, and in particular those that prevented differentiation state switching, were required to overcome intratumoral heterogeneity and allow for effective therapeutic intervention in heterogeneous TNBC.^10^

In parallel, our group was developing a technology called microenvironment microarrays (MEMA) to study the impact of the microenvironment on the phenotypes of cancer cells.^14,15^ We used this MEMA platform to show that HER2^+^ breast cancer cells could be rendered resistant to HER2-targeted therapeutics by the presence of different growth factors like hepatocyte growth factor (HGF) and neuregulin 1 (NRG1) and that these factors operated in a subtype-specific manner.^15^ However, how such factors might impact tumor heterogeneity and drug response in TNBC was not known.

We report here the application of our MEMA technology to the TNBC cells previously tested with targeted inhibitors to determine how the presence of different microenvironmental factors might impact tumor heterogeneity and/or drug response. We found that several factors, including HGF and NRG1, could increase cell proliferation and render cells more resistant to the MEK inhibitor trametinib, which is in clinical trials for use in TNBC.^16^ Interestingly, combinations of HGF and NRG1 resulted in greater resistance than treatment with either of the single growth factors, consistent with the concept that these factors are operational in subsets of the heterogeneous cell population. Surprisingly, under trametinib treatment, the growth factors restore tumor heterogeneity, rather than driving the outgrowth of specific basal or luminal subpopulations. We demonstrate that treatment with drug combinations aimed at blocking the activity of HGF was capable of reversing drug resistance and driving the cells back into a homogeneous state. Finally, we show that these factors are operational in xenograft models of TNBC and that elevated expression of these factors in TNBC patients is associated with both increased tumor heterogeneity and worse clinical outcomes, suggesting clinical relevance to patients. These data suggest that growth factors from the tumor microenvironment can maintain tumor cell proliferation and heterogeneity under therapeutic pressure, leading to increased drug resistance, but that targeted inhibition of their receptors reverses these effects, restoring drug sensitivity and decreasing tumor heterogeneity.

RESULTS

MEMA platform reveals influence of microenvironment factors on heterogeneity and therapeutic response

We made use of the basal-subtype cell line HCC1143 to study heterogeneity and resistance to the targeted therapeutic trametinib. Basal-like TNBC cell lines like HCC1143 show significant phenotypic heterogeneity when examining differentiation state markers, unlike claudin-low TNBC cell lines like MDAMB231 and Hs578T that homogeneously express mesenchymal markers.^10^ We chose trametinib, an MEK inhibitor, since it is an approved drug for cancer treatment,^17^ which is under investigation for use in TNBC^16^ and had been shown to drive heterogeneous HCC1143 cells into a KRT5/14^+^ basal-like state while also inhibiting the growth of the cells.^10^ We used our MEMA technology to test the impact of defined microenvironmental signals on HCC1143 TNBC cells under three conditions: (1) low serum (0.1% fetal bovine serum) conditions to assess microenvironment impact on cell growth; (2) standard growth conditions to assess microenvironment impact on cell heterogeneity and growth enhancement; and (3) standard growth conditions plus trametinib, to assess microenvironment impact on drug response, cell heterogeneity, and differentiation state. In all three experiments, cells were grown for three days on the MEMA, treated with 5-ethynyl-2’-deoxyuridine (EdU) 1 h prior to fixation, then fixed and stained by immunofluorescence (IF) for EdU incorporation, and examined KRT14 (basal) and VIM (mesenchymal) differentiation state markers.

HCC1143 cells grown on MEMA under low-growth conditions showed continued growth of cells, with only a modest increase in cell numbers, with a small number of ligands. In particular, the addition of HGF, fibroblast growth factors (FGF-1 and -6), or epidermal growth factor (EGF) family ligands (EGF, NRG1β, and AREG) gave rise to significantly higher cell numbers compared to no ligand (PBS only) control conditions (see Figure S1A). Similarly, we were able to observe ligand conditions that altered the KRT14:VIM ratio, with BMP proteins increasing the ratio, while EGF family ligands lowered the ratio (Figure S1B).

Cells grown on MEMA under standard media conditions showed modest changes in either growth rate or differentiation state compared to the vehicle controls (Figures S2A and S2B). With respect to growth as measured by average cell number, we observed that NRG1α and NRG1β significantly enhanced cell growth, as did IFN-γ, while BMPs had variable effects on cell growth (BMP5 significantly enhanced growth, while BMP6 significantly inhibited growth of HCC1143 cells). As with the low serum conditions, differentiation state was most significantly altered by treatment with EGF, HGF, and TGF-β1, which drove cells into a more basal-like lineage, signified by a higher KRT14:VIM ratio, while cells grown in the presence of BMPs became more mesenchymal-like lineage, denoted by a lower KRT14:VIM ratio (Figure S2B).

In our final MEMA assay, we investigated the impact of different microenvironmental conditions on the response of TNBC HCC1143 cells to trametinib treatment. We treated HCC1143 cells growing on the MEMA platform with 100 nM trametinib for 3 days, then fixed, stained, and imaged the cells. We observed a number of ligands and extracellular matrix (ECM) combinations that gave rise to modest but significant increases in cell numbers, including EGF family ligands (EGF, NRG1α, NRG1β, and AREG), HGF, FGFs (FGF-1 and -6), and several TGF/BMP family members (TGFβ1, TGFβ2, BMP-2, -3, -6, and -7; see Figures 1A and 1C). ECMs that enhanced cell numbers included fibronectin 1 (FN1), laminin, vitronectin, and, in particular, multiple ECM combinations containing collagen I (Figure 1A). EdU incorporation showed a more limited set of ligands that enhanced proliferation, which consisted mainly of EGF family ligands and HGF (Figure 1B). ECMs that enhanced proliferation under trametinib treatment again included collagen 1, FN1, laminin, and vitronectin (Figure 1B). Of note, the subtle changes in cell number and EdU incorporation we observed contrasted with our previous findings in HER2^+^ breast cancer cell lines.^15^ Specifically, in HCC1143 cells, a single dominant driver of resistance to therapy was not observed, and the effects of the ECM molecules were much more evident than in the HER2^+^ cells. Consistent with our findings in the absence of trametinib, ligands were able to modulate the KRT14:VIM ratio (Figure 1D). However, the entire curve was shifted upward to a higher KRT14:VIM ratio, consistent with trametinib driving KRT14 expression.^10^ The average log_2_ KRT14:VIM ratio in the trametinib-treated cells was greater than 1 and was significantly higher than the average log_2_ ratios in the low and normal serum conditions by ANOVA (p < 0.0001). NRG1-SMDF and BDNF significantly increased the basal expression, while HGF, EGF, AREG, NRG1β, and NRG1α were among the ligands that decreased the KRT14:VIM ratio (Figure 1D).

Trametinib response in the presence of growth factor combinations

Given these findings, we postulated that, in contrast to the previous study of microenvironment-mediated resistance in HER2^+^ cell lines,^15^ the lower levels of resistance conferred by microenvironmental factors against trametinib in HCC1143 might be due to the intrinsic heterogeneity of the TNBC cells. Thus, these growth factors might only confer resistance to subsets of cells that express the necessary receptor. We hypothesized that the growth factors HGF and NRG1 might be acting on specific subtypes of cells within the heterogeneous population (e.g., basal and luminal differentiation state-positive cells, respectively), consistent with our previous observations of subtype-specific effects of these factors in HER2^+^ cell lines. This hypothesis would also predict that the effects of NRG1 and HGF would be additive, as they would likely be acting within different subsets of the heterogeneous tumor population. To test this and validate our MEMA findings, we treated HCC1143 cells with trametinib under baseline conditions or in the presence of NRG1β, HGF, or a combination of NRG1β plus HGF (Figure 2A). We confirmed that NRG1β and HGF can confer resistance to trametinib in HCC1143 cells, and also demonstrated that the effects of these ligands were additive, as the two together gave rise to higher levels of resistance than either alone. Indeed, the GR50 (50% growth rate, represents dose of trametinib required to inhibit growth of the cells by 50% ^18^) was 6.0 nM under baseline conditions but increased significantly to 15.5 nM if HGF was present (p < 0.05), 24.9 nM if NRG1β was present (p < 0.05), and 52.6 nM if both HGF and NRG1β were added (p < 0.05). Replicate experiments (Figure S3A) were consistent with these findings.

We next confirmed that these factors were active in conferring resistance to additional TNBC cell lines and not simply specific to HCC1143. We first treated BT20, MDAMB468, and SUM149PT TNBC cells with trametinib at baseline to establish GR50 values for these cell lines (Figures S2B–S2D). Cells were treated for three days, then fixed and stained with DAPI, and imaged to quantify the cell numbers. SUM149PT cells were the most sensitive, with a GR50 value of 40 ±10 nM, while MDAMB468 and BT20 had similar responses (GR50 values of 180 ± 25 and 150 ± 25 nM, respectively). We then treated the cells with HGF and/or NRG1β in the presence of trametinib from 25 to 200 nM. The presence of the resistance-conferring ligands led to significant increases in the number of cells compared to baseline conditions in all three cell lines. In BT20, these factors gave rise to resistance across the entire range of trametinib concentrations (25–200 nM; see examples in Figure 2B). In contrast, these factors were only active in SUM149PT and MDAMB468 cells at the lowest concentration, giving rise to significantly higher cell numbers than PBS-only controls with 25 nM trametinib (Figure 2B). Interestingly, we also observed that the combination of HGF and NRG1β was less effective in MDAMB468 than either of the ligands alone (Figure 2B). Additionally, we studied the response in these cell lines using live cell imaging. Again, we observed that these factors could confer resistance to trametinib in all three cell lines (Figure 2C) as the cells grew to higher confluence in the presence of ligand compared to PBS controls. These results suggest that the ability of HGF and NRG1β to confer resistance to trametinib in TNBC cells is a general phenomenon, not specific to HCC1143.

The live cell imaging data also suggested that the impact of growth factor on proliferation in the presence of trametinib might be an immediate, not adaptive, response. To investigate this further, we performed additional live cell imaging experiments where we measured cell growth in the presence of growth factors and trametinib for different time periods. We treated cells with trametinib for 96 h continuously. For some cells, we had growth factors (HGF, NRG1β, or the combination) present for the entire 96 h time course. As expected, trametinib inhibited growth compared to untreated control, but the addition of growth factors significantly restored growth (Figure 2D). In particular, if HGF was added (either alone or in combination), the growth curves were indistinguishable from untreated controls, while NRG1β led to a more modest but still significant growth restoration (Figure 2D). After 48 h of treatment with trametinib alone, we found that the addition of ligand led to an immediate change in growth rate, such that by 96 h, there was a significant difference in the number of cells present between the trametinib controls and the cells that received ligand starting at 48 h (Figure 2D).

Impact of HGF and NRG1β on differentiation state

We next sought to understand the impact of HGF and NRG1β on the differentiation state in HCC1143 cells. As mentioned above, we previously found that these factors showed subtype specificity, with HGF only active in the basal subtype and NRG1β only active in the luminal subtype of HER2^+^ cells.^15^ Thus, we reasoned that HGF might steer the heterogeneous HCC1143 cells into a more basal-like state, while NRG1β might steer them into a more luminal-like state. We treated HCC1143 cells for 24 h with HGF, NRG1β, or HGF + NRG1β both in the presence and absence of 50 nM trametinib, then fixed and stained the cells either for KRT14 (basal), KRT19 (luminal), and VIM (mesenchymal) expression, or alternatively KRT14 (basal) and MET (HGF receptor) expression, and performed image analysis. When we quantified the cell numbers, we confirmed that the treatment with either ligand alone or in combination when trametinib was absent resulted in a significant increase in cell number compared to vehicle control (PBS/DMSO) (Figure 3A). As expected, trametinib treatment resulted in loss of cells, which was rescued by the addition of ligand (Figure 3A).

Visual inspection of images of the cells stained for differentiation states indicated that in the absence of trametinib, treatment with HGF (either alone or in combination with NRG1β) resulted in a reduction in the basal keratin positive cells (KRT14^+^/KRT19^+^, KRT14^+^/VIM^+^, KRT14^+^/KRT19^+^/VIM^+^, or KRT14^+^ cells; KRT14^+^ cells are green in Figure 3B) in HCC1143, with an expansion of the mesenchymal state (VIM^+^, red cells in Figure 3B). In contrast, NRG1β treatment alone did not differ significantly from vehicle control (PBS/DMSO)-treated cells (Figure 3B). When trametinib was added, the cells were driven into a basal-like state (KRT14^+^), consistent with previous findings (Figure 3B). Surprisingly, addition of NRG1β or HGF to trametinib-treated cells did not appear to expand the luminal (blue shaded, KRT19^+^) or basal states, respectively, but instead appeared to restore the heterogeneity of the cell population (Figure 3B). We performed image analysis to quantify the different cell states, which showed that at baseline, 45.6 ± 5.4 (standard deviation, SD) % of the cell population contained KRT14^+^ cells (KRT14^+^, KRT14^+^/KRT19^+^, KRT14^+^/VIM^+^, or KRT14^+^/VIM^+^/KRT19^+^; see Figure 3C). Treatment with HGF or HGF in combination with NRG1 in the absence of trametinib led to a small decrease in KRT14^+^ cell content (Figure 3C; HGF: 33.4 ± 7.1%, p < 0.0001; HGF + NRG1: 39.0 ± 7.2%, p < 0.005). Treatment with NRG1 alone did not result in a significant change compared to control (48.7 ± 5.6%, p = 0.07). Treatment with trametinib significantly expanded the KRT14^+^ population, accounting for 72.4 ± 4.6% of the total cell population (Figure 3C, p < 1 × 10^−17^). Consistent with our visual observations, treatment with HGF in the presence of trametinib significantly decreased KRT14^+^ cell content compared to trametinib alone, restoring the levels closer to those of untreated control conditions (HGF treatment: 56.2 ± 5.8%, p < 5 × 10^−8^; NRG1 + HGF treatment: 58.6 ± 6.0%, p < 1 × 10^−8^).

We calculated the Shannon entropy to assess cellular heterogeneity, collapsing the basal-like, luminal, and mesenchymal states into all eight possible combinations of cell states. Shannon entropy measures diversity within a population, so higher Shannon entropy index values correspond to higher levels of diversity/heterogeneity, and lower values represent a more homogeneous cellular population.^19^ Treatment with NRG1β or HGF + NRG1β did not significantly alter the heterogeneity of the cell population, but HGF treatment reduced the heterogeneity significantly (Figure 3D). Treatment with trametinib led to a highly significant decrease in heterogeneity compared to the control condition. Treatment with NRG1β was able to somewhat restore the heterogeneity following trametinib treatment, but the heterogeneity remained significantly lower than in untreated controls. Most interestingly, HGF treatment by itself or in combination with NRG1β was able to fully restore heterogeneity from the trametinib treatment condition, such that it was not significantly different from baseline controls (Figure 3D). To measure how much ligand treatment helps to reverse the loss of heterogeneity resulting from tramentinib treatment, we also calculated the Kullback-Leibler divergence, taking all 8 possible states of the vehicle control condition’s mean values into account as a reference. A divergence of 1 state (or 0 bit) means no difference can be found, and theoretically, the divergence value can become infinite. Trametinib caused the largest divergence from the control condition (Figure 3E). We compared the divergence of each of the treatment conditions compared to the PBS + trametinib-treated cells. All of the DMSO vehicle conditions (i.e., not treated with trametinib) showed significantly lower divergence values than the PBS + tramentinib-treated conditions, indicating that they were closer to baseline than the trametinib-treated cells. Treatment of trametinib-treated cells with NRG1 had no significant effect on divergence, but treatment with HGF (either alone or in combination with NRG1) significantly reduced the divergence.

We also examined the relationship between KRT14 (basal marker, green) and MET (HGF receptor, red) expression by IF. We observed co-expression of both KRT14 and MET in cells at baseline and HGF-treated cells. Interestingly, this co-expression increased following trametinib treatment (Figure 4A). Quantifying the data, we saw that 30% of cells at baseline expressed both MET and KRT14. Following treatment with trametinib, this double-positive population increased to 68% ± 0.018% (standard error of the mean, SEM) of the population (Figure 4B). Treatment with NRG1β and trametinib maintained the KRT14^+^/MET^+^ population at 68% ± 0.002%, but trametinib and HGF treatment decreased the double-positive cell population to 47% ± 0.013% (Figure 4B). Treatment with both ligands in the presence of trametinib had an intermediate number of double-positive cells (53% ± 0.019% of the population). We calculated the correlation between MET and KRT14 expression, and found that at baseline, it was positive (R = 0.624), with a modest increase when ligand was present (Figure 4C). Interestingly, treatment with trametinib significantly increased the correlation (R = 0.808, p < 0.001 compared to controls), while HGF treatment reduced the correlation (R = 0.770, p < 0.05; Figure 4C).

Signaling pathway activation

We next performed western blot analysis to understand how HGF and NRG1β activated downstream signaling in HCC1143 cells, both at baseline and in the presence of trametinib. As expected, in the absence of trametinib, HGF (either alone or in combination with NRG1β) strongly activated MEK-ERK signaling, as measured by an increase in p-ERK levels (Figure 5A). PI3K pathway signaling activation also occurred with HGF treatment, as indicated by elevated levels of p-AKT, but was not as strong as the p-ERK signal (Figure 5A). In contrast, NRG1β treatment had little impact on p-ERK levels but instead led to activation of PI3K signaling (Figure 5A). However, in the presence of trametinib, which inhibits MEK activity,^7^ HGF treatment only had a very small impact on p-ERK levels. Instead, all ligand treatments (HGF, NRG1β, or HGF + NRG1β) strongly activated p-AKT and PI3K signaling (Figure 5A) when trametinib was present.

Targeting ligand activation to restore drug sensitivity

We previously found that targeting the receptors activated by resistance-conferring growth factors restored sensitivity to anti-HER2 agents in HER2-positive breast cells. Thus, we reasoned that targeting the receptors in TNBC cells should also result in restoration of sensitivity to trametinib. We treated cells for 72 h with HGF, trametinib, and crizotinib, a small-molecule dual tyrosine kinase inhibitor that includes targeting MET,^20^ then fixed, stained with DAPI, and performed image analysis. As expected, HGF treatment enhanced the growth of these cells at baseline, trametinib treatment reduced cell numbers, and HGF conferred partial resistance to trametinib (Figure 5B). Crizotinib monotherapy did not significantly impair the growth of HCC1143 cells compared to untreated controls, but crizotinib in the presence of HGF significantly reduced cell numbers (Figure 5B). Finally, crizotinib treatment in combination with trametinib in the presence of HGF not only restored trametinib sensitivity but also further enhanced cell killing by significantly reducing the cell number compared to cells treated with trametinib alone (Figure 5B).

Impact of ligands on heterogeneity and clinical behavior in TNBC patients

Our data suggests that the presence of some growth factors, like NRG1β and HGF, can enhance the growth of TNBC cells, resulting in elevated cell numbers in the presence of targeted inhibitors like trametinib. Furthermore, the presence of these ligands drives continued tumor heterogeneity, a state that has been postulated to play a role in therapeutic resistance.^7^ To determine if there was any evidence for such activity in patient samples, we analyzed The Cancer Genome Atlas (TCGA) breast cancer dataset.^21^ We identified a subset of 257 TNBC tumors based on low levels of expression of estrogen receptor, progesterone receptor, and HER2 from RNA sequencing analysis of breast tumors. Within this subset, we further examined the expression of NRG1β, HGF, and differentiation state markers (the basal markers KRT5 and KRT14, the luminal marker KRT19, and the mesenchymal marker VIM). We median-centered the log_2_-transformed expression values (relative to the entire TCGA breast tumor dataset) and then performed subset analysis. We defined these subsets based on the expression of HGF and NRG1, where expression was deemed high (more than 3-fold higher than the median of all samples), low (more than 3-fold lower), or normal (between 3-fold higher and 3-fold lower). This resulted in 7 groups: (1) both HGF and NRG1 high (n = 9); (2) high HGF but normal or low NRG1 (n = 18); (3) high NRG1 but normal or low HGF (n = 57); (4) a “normal-expression” group in which levels of both HGF and NRG1 were between 3-fold higher and lower (n = 83); (5) low levels of HGF but normal levels of NRG1 (n= 36); (6) low levels of NRG1 (n = 30); and (7) low levels of both HGF and NRG1. We then examined the expression levels of KRT5, KRT14, KRT19, and VIM in each of the groups (see Figure 5C) and looked for significant differences compared to the normal expression group. The group that had high expression of both HGF and NRG1 had significantly higher levels of expression of basal keratins (KRT5 and KRT14) and VIM (Figure 5C). The same pattern was also observed for NRG1 high tumors, although the levels were not as high (Figure 5C). In contrast, tumors with high HGF had significantly higher levels of VIM expression and significantly lower levels of the luminal marker KRT19 (Figure 5C). Tumors with low NRG1 had significantly lower KRT5 and KRT14, while the group with both low HGF and low NRG1 had significantly lower KRT19 and VIM (Figure 5C), although these just achieved significance. These data suggest that there are significantly different levels of HGF and NRG1 expressed in TNBC patients, and that they are associated with the expression of subtype-specific differentiation state markers.

We next sought to understand whether there was clinical significance to the expression of these ligands. We obtained clinical outcome data from TCGA breast cancer patients from the Genomic Data Commons portal (https://portal.gdc.cancer.gov/) and identified clinical follow-up data for 234 of the 257 TNBC samples. We performed a survival analysis comparing the groups to the “normal-expression” expression subgroup, which we used as the control. Patients with high HGF levels had significantly worse outcomes than patients with “normal” levels of expression (Figure 5D). Although the number of patients with high NRG1 and HGF was small, we also found that this group had significantly worse outcomes compared to the control group (Figure S4). None of the other groups had significantly different outcomes compared to the control group (Figure S4).

Impact of ligands in TNBC xenografts

We next sought to determine if ligands could drive resistance to therapy in TNBC xenografts treated with trametinib. For these studies, we chose to use the SUM149PT cell line, as we had previously observed more uniform growth without necrosis in xenografts with this line compared to HCC1143 xenografts. Since the TCGA data that we analyzed showed that HGF had a greater impact on outcomes in TNBC than NRG1 (Figure 5D), we chose to use HGF as the ligand. We obtained custom implantable pellets that contained ~1 μg of HGF that uniformly release ligand over 21 days (equivalent to ~48 ng of HGF released daily). We implanted 2 pellets per mouse to increase the daily HGF released to ~96 ng per mouse. We orthotopically implanted 2 × 10^6^ SUM149PT cells into both the left and right abdominal mammary fat pads of 18 immune-deficient SCID mice. After ~14 days, we could detect small palpable lesions in 26 of the glands. At this point, mice were randomized into 4 groups and were treated as follows: (1) placebo pellet implanted, daily vehicle treatment for 21 days (control mice, N = 4); (2) HGF pellet implanted, vehicle treatment (N = 5); (3) placebo pellet, 2 mg/kg daily trametinib treatment for 21 days (N = 4); and (4) HGF pellet, 2 mg/kg daily trametinib treatment for 21 days (N = 5). Trametinib treatment started one day after the pellets were implanted. One mouse in group 4 showed significant weight loss within the first 3 days of trametinib treatment and had to be euthanized, leaving 4 mice in the group. All other mice completed the full course of treatment. The day after the final treatment with trametinib, the mice were euthanized, and the tumors were excised for analysis. We found that the tumor weights of the trametinib-treated mice were significantly smaller than tumors from the vehicle-treated mice (Figure 6A). However, the HGF and trametinib-treated tumors were significantly larger than the placebo and trametinib-treated tumors (Figure 6A). Interestingly, the HGF plus vehicle tumors were also slightly larger than the placebo plus vehicle tumors, but the difference did not reach significance levels. We were interested to determine if the differentiation state changed in these tumors in the same fashion as it did in the in vitro treated cells. We stained formalin-fixed, paraffin-embedded sections with DAPI to stain nuclei and KRT14 to detect cells with basal differentiation. The cells were then segmented and quantified using QuPath.^22^ Visually, we observed that trametinib treatment increased KRT14 basal marker expression in the tumors as expected, but that the levels appeared to be lower in tumors co-treated with HGF (Figure 6B). Quantification of the levels of expression of KRT14 in the tumors from the different treatment groups revealed that trametinib significantly increased KRT14 expression, but that it was once again significantly lower in HGF and trametinib-treated mice than in placebo plus trametinib treatment (Figure 6C). The data suggest that HGF impacts the growth of TNBC tumors in response to trametinib and alters their differentiation state, in accordance with our in vitro results.

DISCUSSION

TNBC continues to be one of the most difficult forms of breast cancer to treat, with some of the lowest 5- year survival rates of any of the breast cancer subtypes.^3^ The poor outcomes for TNBC patients are thought to be due to the lack of therapeutic options, along with the heterogeneity and plasticity of the disease, which enables rapid evolution in response to therapeutic pressure, leading to resistance, and the lack of targeted inhibitors approved for treatment.^8^ Members of our Cancer Systems Biology Consortium previously showed that heterogeneous TNBC cell lines like HCC1143 were capable of undergoing state switching in response to different targeted therapeutics, leading to a quiescent, resistant population that would resume growth when therapeutic pressure was removed.^10^ We sought to understand how factors from the microenvironment might impact both the therapeutic response and heterogeneity when treated with some of the same therapeutic agents, utilizing our MEMA platform.

Interestingly, we found that some of the same growth factors that we previously identified that conferred resistance to HER2-targeted inhibitors in HER2^+^ breast cell lines^15^ were operational in TNBC, namely HGF and NRG1β. Unlike the strong resistance we observed in HER2^+^ breast cells when we treated them with a single ligand, the effects of the growth factors on therapeutic resistance were more modest. We reasoned that this could be due to the heterogeneity of the TNBC cells, since we had previously shown that the factors worked in a subtype-specific manner. Indeed, treatment with a combination of the two growth factors resulted in greater resistance than when treated with a single growth factor in all the cell lines we tested, with the exception of MDAMB468. Together, our data suggest that these factors are likely acting on subsets of the cell population, but in a manner distinct from the HER2^+^ subtype specificity we previously observed, where NRG1 was only operational in luminal subtype cells, while HGF only acted on basal subtype cells. ^15^ We expected that treatment with HGF would lead to expansion of the basal subtype cells, while NRG1β would lead to expansion of luminal subtype cells, based on this work in HER2^+^ breast cells. Surprisingly, we found that HGF treatment in the absence of the drug led to a decrease in basal marker-positive cells in HCC1143 cells. When the drug was present, HGF maintained the overall heterogeneity of the HCC1143 population, but did not drive expansion of the basal cell population. This was unexpected, as trametinib alone drives cells into a more basal state, and we had hypothesized that HGF would further expand that basal state. Indeed, we observed that the co-expression of basal markers (KRT14) and the receptor for HGF (MET) was highly correlated, and that this correlation increased with treatment. However, it appears that the action of trametinib to block MEK-ERK signaling results in the activation of PI3K pathway signaling in response to MET treatment instead, which may result in the expansion of the non-basal cells to maintain heterogeneity. Indeed, it has been shown that basal subtype breast cancer cells are more reliant on MEK pathway signaling,^23^ while ER^+^ and in particular, luminal B subtype cells, favor PI3K signaling pathways.^24^ However, trametinib treatment blocks MEK signaling, which has been shown to re-activate PI3K signaling.^23^ Thus, the ability of HGF to activate AKT signaling in the presence of trametinib is likely due to the removal of the MEK-PI3K feedback inhibition loop^23^ and drives the expansion of the luminal-like population, again pointing to the plasticity of these cells and their ability to adapt under therapeutic pressure. Interestingly, NRG1β treatment resulted in a reduction of the luminal marker KRT19, although the KRT14/KRT19 double-positive population expands when trametinib is present. It appears that the growth factors are utilized by cancer cells to both grow and maintain a heterogeneous differentiation state population associated with the plasticity that allows rapid evolution in response to different therapeutic pressures, and thus lack the distinct subtype specificity that we had previously observed in HER2^+^ cells.

Multiple studies have demonstrated that resistance is often an adaptive response, with resistance emerging over time (reviewed in Labrie et al.^25^). Our data suggest that in the case of growth factor-mediated resistance with HGF and NRG1β, the resistance is an immediate response. The heterogeneity and plasticity of the TNBC cells we used show that the growth rate changes as soon as the growth factor is administered, which is inconsistent with an adaptive response. This suggests that many TNBC tumors are poised to respond to increased growth factor expression, and thus, treatment with therapeutics may be countered if high levels of growth factors are present.

Our in vivo work supports the findings that HGF treatment can lead to resistance and impact the differentiation state of TNBC. However, the effects of HGF in this experiment were modest, most likely due to the low amount of HGF delivered by the pellet and the fact that HGF has a high affinity for proteoglycans on the surface of most cells, which limits how much it diffuses.^26^ While the pellets were implanted subcutaneously, it is likely that only a fraction of the HGF made it to the tumors. We hope to perform follow-up work in which we co-inject fibroblasts that are engineered to express HGF along with the TNBC cells to more definitively show the in vivo effects of HGF on TNBC growth and therapeutic response. Importantly, we also confirmed that there was an association between the expression of both growth factors, differentiation state, and patient outcome in TNBC patient samples from TCGA. In particular, we observed that TNBC patients with elevated levels of HGF had significantly worse outcomes compared to TNBC patients with lower levels of these growth factors. This was also associated with elevated levels of the mesenchymal marker VIM, and in the cases that also expressed high levels of NRG1, elevated KRT5 and KRT14. All the TCGA breast samples were collected prior to treatment,^21^ and so the impact of growth factor expression in the setting of therapy was not assessed. We postulate that high levels of expression of these growth factors in TNBC patients undergoing treatment will result in maintenance of a heterogeneous state and continued growth despite therapeutic pressure. Furthermore, high levels of expression of both MET and HER3 receptors have been shown to be associated with a poor prognosis in TNBC,^27,28^ consistent with this notion. Furthermore, these factors have been associated with resistance to other targeted inhibitors in other tumor types like HER2^+^ breast cancer^15,29^ and melanoma,^30^ suggesting they may confer broad resistance to therapy across multiple cancers.

It has previously been observed that cells undergo a differentiation state switching in response to targeted therapeutics like trametinib, with luminal-expressing cells converting to basal-expressing cells that are more resistant to the therapy.^10^ In contrast, when growth factors like HGF and NRG1β are present, we see no evidence for state switching. Instead, the cells continue to grow and maintain the differentiation state heterogeneity, particularly when both growth factors are present. This suggests that in patients, high levels of expression of resistance-conferring growth factors would result in more aggressive disease that could continue to grow even in the presence of therapy, whereas tumors with low levels of expression of these factors would enter a more homogeneous quiescent state that would only resume growth once therapeutic pressure was removed. This further suggests that targeting growth factor receptors in combination with the primary targeted therapeutic would likely have greater efficacy and slow tumor growth and progression in patients with high levels of expression of resistance-conferring growth factors like HGF and NRG1β. In keeping with this hypothesis, we observed that targeted inhibition of MET receptor with crizotinib resulted in restoration of tramentinib sensitivity.

In summary, we have shown here that growth factors from the tumor microenvironment, like HGF and NRG1β, are capable of maintaining TNBC cells in a heterogeneous state and allow them to grow even when treated with targeted therapeutics like trametinib. Mining TCGA data demonstrates that these growth factors are expressed in patient samples, and that differentiation state and clinical outcome are clearly associated with their expression. Our data suggest that using drugs against growth factors or their receptors in combination with the primary therapeutic target may improve efficacy and patient outcomes for patients with TNBC.

Limitations of the study

We have demonstrated that HGF and NRG1 can both impact the response to trametinib in TNBC. There are several limitations to our study. First, we focused on a single targeted agent, trametinib. However, we feel it is likely that these growth factors may have broader effects on therapeutic resistance, as resistance with these ligands has been observed using targeted inhibitors in HER2^+^ breast cancer^15,29^ and melanoma,^30^ among others. Second, most of the work that we performed was in vitro. While we were able to validate our findings in a xenograft model, we feel that additional follow-up studies that optimize the delivery of HGF (and NRG1), as described above in the discussion, would be beneficial to more definitively establish the role of these ligands in resistance in TNBC.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, James E. Korkola (korkola@ohsu.edu).

Materials availability

This study did not generate new, unique reagents.

STAR★METHODS

EXPERIMENTAL MODELS AND SUBJECT DETAILS

Breast cancer cell lines

Breast cancer cell lines derived from human female tumors were used in this study. The cell lines HCC1143, BT20 and MDAMB468 were obtained from American Type Culture Collection (ATCC), Manassas, VA. SUM149PT was provided by Steve Ethier. Each cell line was genotyped to ensure accurate identity, and regularly screened for mycoplasma infection. Cell lines were maintained in their respective medium and serum concentration as recommended by originator specifications at 37° C in 5% CO2 in a humidified incubator as described previously.^33^

Mice for xenograft studies

Immune compromised NOD.Cg-Prkdc^scid^ Il2rg^tm1Wjl^/SzJ (NSG) mice (Jackson Labs) were obtained at 8 weeks of age for use in the xenograft studies.

METHOD DETAILS

Experimental design

The number of independent biological replicates of each experiment (N) performed are given in the figure legends. Where appropriate the mean, standard deviation (SD), or standard error of the mean (SEM) were calculated as indicated. There was no blinding of any experimental data, and no sample-size estimation or randomization were used in standard drug treatment experiments. Protein combination printing locations in MEMA experiments, and drug treatment plates in Figures 2A and S2 were randomized. Experimental results were reproduced in at least three technical replicates (TR), and included either n = 3 sample replicates, or were reproduced with at least 3 biological replicates (BR). No data were excluded from published results.

Drug treatment

Trametinib and crizotinib (Selleckchem) were confirmed by mass spec and used at the concentrations indicated in figure legends. DMSO (ThermoFisher) and human IgG isotype control (Abcam) concentrations were equivalent to the highest dose of the respective drug used in each experiment. Treatment durations were as indicated in respective figure legends. Cells were treated in 96 and 384 multiwell plates with soluble drug and ligand combinations added to their medium, then fixed for fluorescent imaging and quantification (described below). Each treated cell line was seeded at an experimentally determined concertation so that untreated control wells would reach 80% confluency by the end of the treatment period. Drug combination studies and CTG assays in Figure 4F were performed as previously reported^34,35^ in randomized replicates.

MicroEnvironment MicroArrays

MEMAs were generated in 8-well cell culture plates as described previously. A detailed description of the methodology and a list of the ECM components, soluble ligands, and their concentrations is currently available at the Synapse MEP-LINCs website (https://www.synapse.org/#!Synapse:syn2862345/wiki/72486). Proteins on the MEMA were chosen because of their reported involvement at sites of local and metastatic disease, and their capacity to elicit a biological effect in in vitro assays. The proteins included in the library represent components of lymphocytic infiltrates, stroma, blood and lymphatic system, local extracellular matrix, macrophages, and endothelium. Each matrix protein was printed in ~15 replicate random locations. We added soluble ligand to the wells, so an entire MEMA experiment comprised eight plates (seven ligands plus a PBS control well per plate; thus, 8 plates comprised all 56 ligands tested). 2.5 × 10^5^ cells of each cell line were added to replicate arrays for 15 min, after which unbound cells were removed with a growth medium wash. Arrays were cultured in RPMI medium with 10% fetal bovine serum (FBS) for 12 h at 37° C in 5% CO2 in a humidified incubator. One set of arrays was treated with 250 nm trametinib, one set with DMSO, and one set was transferred to low serum (0.1% FBS) conditions. Following this, appropriate concentrations of soluble ligands were added to duplicate sets of arrays. Arrays were returned to incubate for 71 h, after which 1 μM EdU was added to the medium for 1 h. Cells were then fixed in 2% PFA at RT, and stored at 4° C in PBS.

After fixation, EdU detection and immunofluorescent histochemistry (IHC) was performed as described below. Arrays were imaged on a customized automated high content fluorescence microscope platform (Nikon HCA), and resulting image data was output to an OMERO image database. ^36^ Cells were segmented and intensity levels were calculated using CellProfiler.^31^

Immunofluorescent histochemistry and fluorescent imaging

Array-bound and well-bound cells were fixed in 2% PFA for 15 min at RT following respective treatments. Cells were then permeabilized with .3% Triton X-100 for 25 min at RT. Cell primary antibody staining was performed with KRT14 (Abcam, 1:200), VIM (Dako, 1:200), KRT19 (Dako, 1:200), MET (R&D Systems, 3 μg/mL working concentration), and DAPI (ThermoFisher, 1:10,000). Secondary antibody staining was performed with IgG3 Alexa Fluor 488 (ThermoFisher, 1:200), and IgG1 Alexa Fluor 555 (ThermoFisher, 1:200). Well plates were imaged on the GE InCell 6000 platform, and image analysis and cell count quantification were performed on the GE InCell Analyzer software package. Size gating of nuclei was used to exclude apoptotic cells, and EdU positivity was determined as nuclei having a mean fluorescent intensity above an experimentally consistent threshold (this threshold was defined using single cell parametric analysis plotting total DAPI intensity against mean EdU intensity). All fluorescent imaging studies were performed at consistent intensity and gain settings across experiments.

EdU incorporation

Cells were incubated with 1 μM EdU for 1 h prior to fixation. Cells were fixed, permeabilized, and stained with Click-iT Plus EdU Alexa Fluor 647 HCS Assay Kit (ThermoFisher) following manufacturers recommended protocol.

Live cell imaging

Live-cell imaging experiments were performed on the IncuCyte ZOOM platform with HCC1143, SUM149PT, MDAMB468 and BT20 cells using phase contrast. Cells were plated for ~16h and then appropriate concentrations of drug or ligand NRG1β or HGF were added to their medium at time zero of the time course. Cells were imaged every 2–3 h (4 images per well), and Incucyte proprietary image analysis software quantified detected percent confluence. Live-cell time course experiments had N = 2 biological replicates in each experiment, and all had N > 3 technical replicates with consistent results.

Protein expression by western immunoblots

For Western blots, cell lysates were collected using Nonidet-P40 lysis buffer supplemented with Halt protease and phosphatase inhibitor cocktail (Thermo Scientific). Immunodetection of proteins was carried out using standard protocols for equal amounts of protein loaded in SDS gels (as determined by BCA protein abundance assays). The antibodies panAKT (clone C67E7), pAKT (S473, clone D9E), ERK1/2 (clone 137FS), and pERK1/2 (T202/Y204, clone D13.14.4E) were all purchased from Cell Signaling Technologies. Immunoblots were imaged on the LI-COR Odyssey platform, and quantified using LI-COR Image Studio Lite.

Mouse xenograft experiments

Immune compromised NOD.Cg-Prkdc^scid^ Il2rg^tm1Wjl^/SzJ (NSG) mice (Jackson Labs) were obtained at 8 weeks of age. After ~2 weeks of acclimation, 1 × 10^6^ SUM149PT cells were implanted orthotopically into the left and right mammary fat pads of 18 mice under isofluorane anesthesia. Mice were treated with meloxicam prior to surgery and 24 h post-surgery as an analgesic. Tumors were allowed to develop for 2–3 weeks, at which time small palpable lesions were detectable in 26/36 implanted glands. The mice were randomized into four groups (HGF pellet+ vehicle (N = 5); HGF pellet + trametinib (N = 5); placebo pellet + vehicle (N = 4); placebo pellet + trametinib (N = 4)). Two slow release pellets were implanted into each mouse (placebo or HGF containing 25 μg per pellet). The following day we began daily treatment for 21 days with 2 mg/kg trametinib (Selleck Chemicals) in 13% DMSO and 40% PEG400 (Polysciences, Inc) by oral gavage. After 3 weeks of trametinib-treatment, mice were euthanized and the tumors were carefully excised and fixed in 2% paraformaldehyde. Excised tumors were weighed to compare sizes then submitted to the OHSU histopathology core for embedding and sectioning. Sectioned tumors were used for immunofluorescence staining for KRT14 levels (Abcam, 1:200) and detected using anti-IgG3 Alexa Fluor 488 (ThermoFisher, 1:200) and DAPI for detection of nuclei. Images were captured on a Zeiss fluorescent microscope and image segmentation and quantification of KRT14 positive cells was performed using QUPath software.^22^

QUANTIFICATION AND STATISTICAL ANALYSIS

Information on biological replicates (indicated as N) and technical replicates can be found in the respective figure legends. The reported statistics used sample means, standard deviation, standard error of the mean (SEM), and p-values obtained from unpaired parametric t-tests of sample sizes of equivalent variance (unless otherwise noted in figure legends) and ANOVA analyses with Tukey’s correction. Dunnette’s test was run with the DunnettTest function from the R package DescTools using the PBS spots as the controls. All reported cell assays had at least 3 technical replicates. Kaplan-Meier log rank analysis was used to compare patient outcomes based on expression levels of HGF and NRG1.

For analysis of MEMA data, we preprocessed and normalized using open-source R software available from (https://github.com/HeiserLab/TNBC_MEMA; https://doi.org/10.5281/zenodo.17581995). The cell count in each MEMA spot was based on the DAPI stained nuclei. EdU intensity was auto-gated to label cells as EdU^+^ and the proportion of EdU^+^ cells in each spot was reported to measure proliferation. The per-cell intensity values were median summarized to the spot level. Each intensity and morphology signal was independently RUV normalized in a series of matrices with arrays as the rows and spots as the columns.^37,38^ The RUV controls were the residuals created by subtracting the replicate median from each spot value. After RUV normalization, bivariate LOESS normalization was applied to the normalized residuals using the array row and array column as the independent variables. After normalization, the ~15 replicates of each condition were median summarized to the MEP level. Major findings from the MEMA were recapitulated in at least 3 experimental replicates. Exact replicate count and standard error for each condition are available in supplemental MEMA files linked to in Data Availability.

For single cell differentiation state analysis, the experimental annotation was taken from MEMA level1 data and merged with MEMA level0 and relevant columns were extracted. For each cell total intensity values were calculated and log2 transformed. Cells without nuclei were excluded. Additionally, outliers were excluded from the data by rrscaling (Hunt er al. R cran package rrscale version 1.0 run on R version 4.1.2) and log2 transformed data was filtered for cells not removed by the rrscale algorithm. Analysis was done in Python version 3.10.12. We made use of the altair version 5.1.2 (plotting), matplotlib version 3.8.0 (plotting), pandas version 2.1.0 (calculation, plotting), numpy version 1.23.4 (caculation), and scipy version 1.11.2 (Entropy and Kullback Leibler divergence calculation). For Entropy calculation, the mean values from the PBS DMSO condition were taken for reference. Protein expressed above the mean were regarded positive, proteins expressed below the mean value were regarded negative. Entropy and Kullback-Leibler values were power transformed to retrieve values in unit states rather than bits. For more detailed information and complete reproducibility all raw data, normalized data, normalization and analysis code, and the resulting plots and reports are available as stated under Data Availability (see below).

MEMA image segmentation

MEMA images were segmented with unpublished code that is similar to the method described in https://doi.org/10.1007/978-1-4939-9773-2_24.

TCGA RNAseq analysis

We log_2_ transformed and median centered TCGA RNAseq data. We examined expression of estrogen receptor, progesterone receptor, and HER2 to identify samples that we defined as TNBC based on low levels of expression of these genes. Expression of the differentiation state markers KRT5 and KRT14 (basal markers), KRT19 (luminal marker) and VIM (mesenchymal) marker were examined within these TNBC samples and subsets were compared based on HGF and NRG1 levels to determine associations with outcome.

If image segmentation

Stained tissue sections from xenograft experiments were segmented and quantified using QUPath software.^22^

Data and software availability

MEMA data for cell numbers and EdU incorporation for HCC1143 cells under standard, low serum, and trametinib-treated conditions is available from the M2CH portal at synapse (Synapse data: https://www.synapse.org/#!Synapse:syn21578893) and quantification of IF data for keratins, vimentin, and MET levels is available at Gitlab (Gitlab data: https://doi.org/10.5281/zenodo.17586670).Custom code that was used for processing MEMA data is available at Github (https://doi.org/10.5281/zenodo.17581995) and custom code used for analyzing IF data is available at Gitlab (https://doi.org/10.5281/zenodo.17586670).

Supplementary Material

1

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2025.116826.