Stiffness-dependent MSC homing and differentiation into CAFs – implications for breast cancer invasion

Cancer-associated fibroblasts (CAFs) drive cancer development by stimulating tumor growth and invasion via enhanced extracellular matrix (ECM) production, stromal stiffening and matrix metalloproteinase (MMP)-mediated ECM remodeling (Tomasek et al., 2002; Das et al., 2017), as well as by physically pulling cancer cells out of the primary tumor (Labernadie et al., 2017). Although most CAFs express the myofibroblast marker α smooth muscle actin (αSMA, also referred to as ACTA2), considerable phenotypic heterogeneity exists within CAFs, which is indicative of the existence of multiple CAF subtypes with variable expression of fibroblast activation protein (FAP), fibroblast-specific protein (FSP1, also known as S100A4) and platelet-derived growth factor receptor β (PDGFRβ, also known as PDGFRB) (Mishra et al., 2008; Su et al., 2018). Although the origin of CAFs remains unclear, several studies have identified mesenchymal stem cells (MSCs) as potential source (Karnoub et al., 2007; Weber et al., 2015), which home to tumor sites in response to various factors in the tumor stroma (Chaturvedi et al., 2012).

Many epithelial cancers have increased stromal stiffness, due to enhanced collagen deposition, and alignment (Calvo et al., 2013). Although MSCs are known to respond to matrix stiffness and differentiate into multiple lineages (Engler et al., 2006), the influence of stiffness on MSC differentiation into CAFs is less understood. A recent study has demonstrated that when exposed to conditioned medium secreted by cancer cells cultured on tissue culture plastic, MSCs differentiate into CAFs in a stiffness-dependent manner, driving cancer proliferation (Ishihara et al., 2017). However, we have recently shown that the secretome of cancer cells is stiffness dependent (Patwardhan et al., 2021). In this work, we address the importance of stiffness-dependent crosstalk between MSCs and MDA-MB-231 breast cancer cells in mediating MSC chemotaxis and differentiation into CAFs, as well as the involvement of MSC-derived CAFs (hereafter referred to as MSC-CAFs) in cancer invasion. By analyzing previously published single-cell RNA sequencing (scRNAseq) data from individuals with breast cancer (Pal et al., 2021), we demonstrate that triple-negative breast cancer (TNBC) tumors harbor a population of cells exhibiting both MSC and CAF markers (referred to hereafter as the MSC/CAF signature). Using polyacrylamide gels of stiffnesses mimicking normal, pre-metastatic and metastatic stroma, we show that MSC chemotaxis and MSC differentiation into CAFs are optimal on gels with a stiffness of 2 kPa, which mimic pre-metastatic stroma, and that the differentiation process occurs in a TGFβ (herein referring to TGFB1)- and contractility-dependent manner, resulting in stable differentiation of MSC-CAFs. We further demonstrate that MSC-CAFs from 2 kPa gels promote cancer cell proliferation, invasion and shear stress resistance. Finally, we analyze scRNAseq data from circulating tumor cells (CTCs) and metastatic tumors, and establish the presence of cells expressing the MSC/CAF signature in those samples. In addition to illustrating the role of MSCs in mediating cancer metastasis via stiffness-dependent homing to tumors and differentiation into CAFs, our results implicate the pre-metastatic niche as driver of cancer progression.

Since TNBC tumors have inter- and intra-tumor cellular heterogeneity, we evaluated whether they harbor MSCs by analyzing scRNAseq data from 20 individuals with breast cancer (Pal et al., 2021). In total, 64,550 cells from eight individuals with TNBC and 57,135 cells from individuals with other breast cancer subtypes (four BRCA1 pre-neoplastic, four HER2 positive and four ER positive) were grouped. Seurat clustering (Satija et al., 2015) identified transcriptionally distinct populations, which were tested for MSC markers including CD73 (also referred to as NT5E), CD90 (also referred to as THY1) and CD106 (also referred to as VCAM1) (Maleki et al., 2014). Out of 11 clusters in the TNBC group (Fig. 1A), cluster 7 expressed MSC markers (Fig. 1B, left panel). In contrast, cells in the non-TNBC group were divided into 11 clusters (Fig. 1C), none of which expressed MSC markers (Fig. 1D). Various ECM genes (COL1A1 and COL1A2) and the MSC marker CD90 were expressed in cluster 7 (Fig. 1A; Fig. S1) (Roson-Burgo et al., 2016; Tang et al., 2017) indicating that cells in this cluster represent MSCs that invaded the tumor stroma. Gene ontology (GO) analysis (Wu et al., 2021b) revealed that cluster 7 is enriched in pathways related to MSCs (Fig. 1E). Cluster 7 also expressed CAF markers including ACTA2, FAP, CAV1, CD29 (also known as ITGB1), TAGLN and COL1A2 (Fig. 1B, right panel; Fig. S1C). Within this cluster, in comparison to a sizeable subpopulation of cells expressing membrane markers, a smaller subset of cells expressed only contractile markers. A small fraction of cells co-expressed both contractile and membrane markers (Fig. S1B). Invasion (MMP2), ECM stiffening (LOX) and bone metastasis [CTGF (CCN2), IGF1] markers were also expressed by cells in cluster 7 (Fig. S1D). Cell type deconvolution of bulk transcriptomic profiles from The Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA) dataset (Cancer Genome Atlas Network, 2012) was performed to confirm the presence of MSCs. Data were divided into groups consisting of 194 TNBC and 899 non-TNBC samples (Cancer Genome Atlas Network, 2012). The deconvolution of immune and stromal cell types demonstrated the presence of MSCs as one of the cell types, and MSCs were found to be significantly abundant in the TNBC samples (Fig. 1F).

To characterize the lineage relationship between cells expressing the MSC/CAF signature in TNBC, a differentiation trajectory for cluster 7 cells was reconstructed using MARGARET (Pandey and Zafar, 2022), which inferred a bifurcating trajectory represented as a directed graph consisting of five cell-state clusters (Fig. S2A,B), two of which were detected as terminal states (cell-state clusters 1 and 2). The cells were pseudotemporally ordered by selecting the cell-state cluster 0, which comprised THY1-positive, NT5E-positive, CD14-negative, CD25 (IL2RA)-negative and CD4-negative cells, as the initial population (Fig. S2C,D). Two branches emerged from the starting population, one branch gave rise to cells (cell-state clusters 4 and 1) co-expressing MSC markers (THY1, VCAM1) as well as markers corresponding to membrane associated CAFs (COL1A1, FAP, THBS2) (Fig. S2D,E). The second branch led to a terminal state (cell-state cluster 2) expressing contractile CAF markers (ACTA2, TAGLN, CNN1) (Fig. S2D,E). This suggests that the cells expressing MSC markers differentiate into different CAF subtypes expressing membrane-associated genes and contractile genes, whose lineage-specific expression trends are shown in Fig. S2E.

We further tested another previously published TNBC scRNAseq dataset for the presence of MSCs (Wu et al., 2020). Seurat clustering analysis from five individuals with TNBC identified 19 clusters, of which three clusters (clusters 6, 11 and 15) expressed different MSC markers (CD90, VCAM1, PDGFRβ) (Fig. S3A) along with expression of CAF and invasion markers (Fig. S3B). The presence of MSCs in TNBC tumors was experimentally evaluated by staining for the MSC marker Stro-1 (Lv et al., 2014) and the CAF marker FAP (Han et al., 2020) in grade 3 tumor tissue sections from four individuals with TNBC (Fig. 1G; Fig. S1E). Three samples were positive for Stro-1 at the tumor and the stromal site, indicating the presence of MSCs at the tumor; however, only two were positive for FAP. Taken together, these results suggest that MSCs are recruited to the primary tumor and differentiate into CAFs exhibiting distinct phenotypes.

The existing literature and our histological analysis raise the possibility of MSC recruitment to the tumor site mediated by cancer cell-secreted soluble factors (Ridge et al., 2017; Brennen et al., 2013). We hypothesized that cells expressing the MSC/CAF signature are MSCs that home to the primary tumor and undergo differentiation into CAFs. To determine the extent to which ECM stiffening associated with cancer progression may influence MSC homing at the site of the primary tumor, we probed the influence of the stiffness-dependent cancer cell secretome on MSC chemotaxis. Chemotaxis in response to factors secreted by MDA-MB-231 cells cultured on polyacrylamide gels mimicking the stiffness of normal mammary stroma (0.5 kPa), pre-metastatic tumor stroma (∼2 kPa) and metastatic tumor stroma (∼5 kPa) (Paszek et al., 2005; Levental et al., 2009) was studied using a microfluidic device with two side channels connected by transverse channels (Fig. 2A) (Saxena et al., 2018). MSCs mixed with three-dimensional (3D) collagen (of various concentrations; Fig. 2B) were seeded on one side channel, and only collagen was introduced to the other side channel. Gradient generation across the transverse channels filled with 3D collagen (2 mg/ml) was visualized using FITC–dextran, which revealed gradient establishment within an hour (Fig. 2C). After polymerization of the 3D collagen, cancer-conditioned medium (CCM) from MDA-MB-231 cells on different stiffness gels or tissue culture plastic was introduced at the cell-free side, and plain medium was added on the cell-containing side. Tracking of MSC nuclei along the transverse channels revealed fastest motility in the presence of CCM from MDA-MB-231 cells on 2 kPa gels (Fig. 2D). Taken together, these results suggest that CCM from MDA-MB-231 cells on 2 kPa gels optimally induces MSC chemotaxis.

To next probe the effect of the stiffness-tuned cancer cell secretome on MSC fate, MSCs were cultured for 7 days on 0.5, 2 and 5 kPa polyacrylamide gels using conditioned medium (CM; composed of CCM from an MDA-MB-231 culture of matching substrate stiffness and fresh medium mixed in a 1:1 ratio) (Fig. 3A). Whereas spreading after 1 day was stiffness dependent with no effect of CM addition, we found that CM increased spreading of MSCs on 2 kPa and 5 kPa gels after 7 days (Fig. 3B, panels i and ii). However, CM altered the motility of MSCs on 2 kPa gels on day 1 itself, with further increases detected for MSCs on 2 kPa and 5 kPa gels at day 7 (Fig. 3B, panel iii). Additionally, the highest proliferation was observed on 2 kPa gels over 7 days (Fig. S4A). Higher MSC spreading and motility of MSCs on 2 kPa gels in the presence of CM correlated with increased cytoskeletal organization and cell stiffness (Fig. S4B). We found that expression of the MSC marker Stro-1 was downregulated after 7 days on both 2 kPa and 5 kPa gels in the presence of CM (Fig. S4C), which led us to hypothesize that the observed changes indicate MSC differentiation into CAFs. Indeed, expression of the CAF markers αSMA and FAP (Mishra et al., 2008; Su et al., 2018), as well as the nuclear marker lamin-A (LMNA), was elevated in cells on 2 kPa gels in the presence of CM (Fig. 3C). Additionally, immunostaining for αSMA and LMNA revealed highest expression in the 2 kPa with CM condition (Fig. 3D). Upregulation of CAF markers in cells on 2 kPa gels was associated with high TGFβ levels in cells on 2 kPa gels in the presence of CM (Fig. 3E). Inhibition of TGFβ using a neutralizing antibody (Fig. S4D, panel i) led to ∼50% drop in αSMA levels for cells on 2 kPa and 5 kPa gels in the presence of CM and to ∼30% drop in LMNA levels for cells on 2 kPa gels in the presence of CM (Fig. 3F; Fig. S4D, panels ii and iii). Collectively, these results suggest that MSC differentiation into CAFs is driven by TGFβ in the CCM, with the highest levels of TGFβ secreted by cells on 2 kPa gels.

In line with the condition in which we observed highest αSMA expression, MSCs possessed high levels of phosphorylated myosin light chain (pMLC) when cultured on 2 kPa gels in the presence of CM (Fig. 4A) and exerted increased tractions (Fig. 4B; Fig. S4E). Since stiffness-dependent differentiation requires non-muscle myosin II-dependent stiffness sensing (Engler et al., 2006), to probe the role of contractility in MSC differentiation into CAFs, MSCs were cultured with or without CM in the presence of blebbistatin (Bleb) for 7 days (Fig. 4C, panel i). Bleb treatment led to a drop in the proliferation capability of MSCs (Fig. S4G) and to significant decreases in the levels of αSMA, LMNA and pMLC in the presence of CM (Fig. 4C, panels ii and iii; Fig. S4F). However, no changes in αSMA or LMNA expression in response to Bleb treatment were observed in the absence of CM. To assess differentiation stability, MSC-CAFs were cultured for 5 days in low-serum medium without CM (Fig. 4D, panel i); we observed marginal reductions in αSMA and LMNA levels for cells on 2 kPa gels, whereas αSMA levels in cells on 5 kPa gels were found to drop significantly (Fig. 4D, panel ii). Next, to next probe the sensitivity of MSC-CAFs to contractile inhibition, they were cultured for 3 days with Bleb in the absence of CM (Fig. S5A, panel i). Remarkably, pMLC and αSMA levels remained relatively unchanged for cells on 2 kPa gels but dropped significantly for cells on 5 kPa gels (Fig. S5A, panels ii and iii). Taken together, these results suggest that 2 kPa stiffness induces stable differentiation of MSC-CAFs in a contractility-dependent manner.

To probe the role of CAFs in modulating cancer cell behavior, MDA-MB-231 cells were cultured on 2 kPa and 5 kPa gels in the presence of conditioned medium from the corresponding MSC-CAFs (MSC-CAFCM) (Fig. 5A). Although MSC-CAFCM induced increased cell spreading on both 2 kPa and 5 kPa gels (Fig. 5B), highest proliferation was observed on 2 kPa gels in the presence of MSC-CAFCM over 7 days (Fig. 5C). To investigate the role of MSC-CAFs in cancer invasion, we prepared spheroids consisting of MDA-MB-231 cells alone (S_M) or MDA-MB-231 cells together with MSC-CAFs from either 2 kPa or 5 kPa gels (S_MF2 and S_MF5, respectively) (Fig. 5D). All spheroids were similar in size, with most cells being viable (Fig. S5B), and S_MF2 heterospheroids exhibited maximum outward cell scattering when implanted in 3D collagen gels (Fig. 5E). Labeling of MSC-CAFs and MDA-MB-231 cell with two different dyes revealed clear enrichment of MSC-CAFs at the outer periphery of spheroids (at day 0) in S_MF2 spheroids relative to S_MF5 spheroids (Fig. S5C). After 2 days of invasion, the invasive front of S_MF2 spheroids comprised primarily of MSC-CAFs from 2 kPa gels, whereas the invasive front of S_MF5 spheroids included both MSC-CAFs from 5 kPa gels and MDA-MB-231 cells (Fig. 5F). The high invasiveness of S_MF2 spheroids correlated with the high MMP2 expression in MSC-CAFs from 2 kPa gels, with no difference in MMP9 expression observed for MSC-CAFs from either 2 kPa gels or 5 kPa gels (Fig. 5G). Maximum degradation of collagen (i.e. fluorescence of degraded DQ-collagen) was observed for MSC-CAFs from 2 kPa gels, which we attribute to the high MMP2 expression in those cells (Fig. 5H). Taken together, these results suggest that, in addition to promoting cancer cell proliferation, MSC-CAFs from 2 kPa gels drive cancer invasion via increased matrix degradation.

Cancer metastasis involves entry into the vasculature, adhesion to the endothelium and subsequent extravasation. Integrin profiling of MSC-CAFs revealed upregulated of expression of β1 and β3 integrins (ITGB1 and ITGB3, respectively) in MSC-CAFs from 2 kPa gels, and of α5 integrins (ITGA5) in MSC-CAFs from 5 kPa gels, following 7 days treatment with CM (Fig. 6A,B; Fig. S5A, panel iv). However, expression of all other α integrin subunits tested remained the same across both 2 kPa and 5 kPa MSC-CAFs in the presence of CM (Fig. S5A, panel iv). Consistent with this, in a simple adhesion assay on fibronectin-coated coverslips (Fig. 6C), the fraction of S_MF2 spheroids that remained attached was higher than that for S_M and S_MF5 spheroids (Fig. 6D). To assess the adhesion strength, spheroids attached for 2 h were incubated with trypsin (Fig. 6E). The time required for detachment was determined based on spheroid movement from their initial location after trypsin addition (Fig. 6F, panel i). Remarkably, the detachment duration, τ_detach, a measure of adhesion strength, was longest for S_MF2 spheroids (Fig. 6F, panel ii). To test whether S_MF2 spheroids also exhibited greater shear resistance, flow experiments were performed using a straight channel device with an intermediate port for introducing spheroids (Fig. 6G). Experiments were performed using a syringe pump after 10–12 h of spheroid adhesion. For a physiologically relevant flow rate of 400 μl/min (Kucukal et al., 2018), simulations predicted shear stresses of ∼20 dyn/cm² (∼2 Pa) within the middle port at the position of the spheroids (x=800–1100 μm; Fig. 6H). Consistent with the measurements of τ_detach, the time taken by spheroids to detach under shear (τ_shear), was highest for S_MF2 spheroids (Fig. 6I). Collectively, these results suggest that increased integrin expression in MSC-CAFs from 2 kPa gels makes S_MF2 spheroids more adhesive and shear resistant.

To evaluate whether CTC clusters harbor MSCs and CAFs, we analyzed scRNAseq data from CTCs isolated from individuals with breast cancer and from mammary-specific polyomavirus middle T antigen overexpression (MMTV-PyMT) mouse models (Szczerba et al., 2019). CTCs were clustered into three groups expressing the MSC markers CD29, CD166 (also known as ALCAM) and CD73 (all clusters); CD105 (also known as ENG; clusters 0 and 1); VCAM1 (cluster 0); and CD45 (also known as PTPRC; cluster 1); as well as the CAF markers ACTA2, FSP1, CAV1 and TAGLN (Fig. 7A; Fig. S6A). Additionally, CTCs expressed CD44, LMNA, MMP9 and bone metastasis marker CTGF. Moreover, epithelial–mesenchymal transition (EMT) scoring of CTCs revealed high EMT expression in cluster 2, but very low expression in the other two clusters (Fig. S7). Interestingly, cells in cluster 0 and 1 expressed high levels of TLN2 and MYH10. Whereas cells with a high EMT score might correspond to cancer cells that have undergone EMT, the combination of low EMT marker expression and prominent MSC and CAF marker expression point to the possibility that cells with the MSC/CAF signature are present in CTC clusters.

Since CD44 interactions drive multicellular aggregation in TNBCs (Liu et al., 2017), and because we observed prominent CD44 expression in all CTC clusters, we therefore checked CD44 expression in MSC-CAFs from 2 kPa and 5 kPa gels. CD44 expression was significantly increased in both 2 kPa and 5 kPa MSC-CAFs, with highest levels in MSC-CAFs on 2 kPa gels, compared to the level of expression in MSCs (Fig. 7B). Although CD44 expression was higher in MDA-MB231 cancer cells than in MSC-CAFs (Fig. S5D), the intensity of CD44 staining in the two cell types was comparable in the invasive front of S_MF2 spheroids (Fig. S5E).

Finally, to test whether CTCs expressing the MSC/CAF signature were associated with metastasis, we integrated scRNAseq data from primary and metastatic tumors (total 375 cells) of breast cancer (Davis et al., 2020) with the data from 61 CTCs (Szczerba et al., 2019). To eliminate batch effects, anchor-based integration using Seurat (Stuart et al., 2019) was used, which resulted in clustering of the cells into two groups: cluster 0 consisted of 54 CTCs, 121 primary tumor cells and 129 metastatic tumor cells (Fig. S6B), whereas cluster 1 had 7 CTCs, 71 primary tumor cells and 54 metastatic tumor cells. Cluster 0 showed enrichment for MSC (CD73), CAF (ACTA2, CAV1) and other (CD44 and LMNA) markers. This analysis suggests that CTCs expressing the MSC/CAF signature are also associated with primary and metastatic tumor cells expressing the same markers.

In this study, by analyzing publicly available scRNAseq data, we correlated the aggressiveness of TNBCs with the presence of cells exhibiting an MSC/CAF signature. We then showed that stiffness-modulated CCM regulates MSC chemotaxis and their differentiation into CAFs, with MSC-CAFs that differentiated on 2 kPa gels having increased stromal invasion and resistance to shear stresses. Finally, analysis of publicly available scRNAseq data of CTCs suggested that CTC clusters harbor cells exhibiting the MSC/CAF signature. Based on our findings, we propose a model wherein cancer progression is associated with MSC homing to tumors at the pre-metastatic stage, differentiation into MSC-CAFs, and subsequent MSC-CAF-driven invasion and metastasis (Fig. 7C). This suggest that the combination of physical and chemical cues present in the pre-metastatic niche actively drives cancer progression.

The recent literature suggests that soluble factors secreted by cancer cells recruit MSCs to the primary tumor, and that the recruited MSCs subsequently differentiate into CAFs (Barcellos-de Souza et al., 2016). CAFs represent one of the most abundant cell types in the heterogeneous tumour microenvironment and are categorized based on differential expression of certain markers (Costa et al., 2018). Based on bulk and single-cell RNA analysis, a recent study has categorized CAFs into ‘fibrotic’ and ‘contractile’ CAFs, with ‘fibrotic’ CAFs characterized by high FAP expression and ‘contractile’ CAFs exhibiting elevated levels of αSMA expression (Vashisth et al., 2021). In line with these observations, our scRNAseq and trajectory analysis suggests that MSCs differentiate into membrane-associated and contractile CAF states. However, the microenvironmental heterogeneity (in terms of physical cues) determining the MSC differentiation trajectories remains unclear. By probing stiffness-mediated bi-directional crosstalk between MSCs and cancer cells, we establish the importance of the pre-metastatic niche in generating a mixed population. MSCs migrate in response to tumor-released soluble factors (Shinojima et al., 2013), and MDA-MB-231 cells have been reported to release levels of TGFβ that are 5-fold higher than those released by MCF-7 cells (Guerrero et al., 2010). The faster MSC chemotaxis induced by CCM from MDA-MB-231 cultures on 2 kPa gels observed in this study may be attributed to elevated TGFβ levels. Another study has shown that sphingosine-1-phosphate, which is secreted by MDA-MB-231 cells to a greater extent on softer substrates (Ko et al., 2016), helps MSCs exit from the bone marrow to the blood (Kong et al., 2014). Thus, increased MSC chemotaxis in response to CCM from 2 kPa cultures might be driven by multiple factors in the CCM, warranting a detailed proteomic analysis.

Microenvironmental heterogeneity between low- and high-grade tumors has been shown to induce formation of different subtype of CAFs (Bernard et al., 2019), which can be identified based on the differential expression of markers including integrin β1, αSMA, FSP1, FAP, PDGFRβ and CAV1 (Costa et al., 2018). While our results suggest that MSC homing and stiffness-dependent differentiation into MSC-CAFs represents one of the mechanisms by which such heterogeneity can arise, plasticity of tissue-resident fibroblasts might also contribute to the heterogeneity, and this is an area that requires a thorough investigation. A recent study based on differential expression of markers has shown that CCM induces MSC differentiation into CAFs with high levels of αSMA expression (Ishihara et al., 2017). However, the CCM used in that study was from cancer cells on tissue culture plastic, which is not physiologically relevant. Here, we show that stiffness-dependent CCM influences the expression profile of MSC-CAFs, with a high-αSMA and high-FAP signature observed for cells on 2 kPa gels, which could be attributed to higher TGFβ levels inducing increased contractility (Talele et al., 2015). Similar to our findings, a recent study has demonstrated TGFβ-induced differentiation of fibroblasts into high-FAP CAFs on soft substrates (Avery et al., 2018). High expression of FAP and its involvement in matrix turnover (Fan et al., 2016) correlates with metastasis and poor prognosis in several tumors (Errarte et al., 2016). MSC differentiation into high-FAP CAFs on 2 kPa gels thus highlights the role of the pre-metastatic niche in driving cancer progression.

Since TGFβ converts fibroblasts into CAFs, blocking of TGFβ in vivo inhibits their differentiation (Grauel et al., 2020). We found that the higher TGFβ levels in CCM from cultures on 2 kPa gels generated MSC-CAFs with higher αSMA expression, and this was blocked in the presence of a TGFβ-neutralizing antibody, highlighting the role of stiffness-dependent TGFβ signaling in MSC differentiation into CAFs. TGFβ and YAP–TAZ signaling induce the expression of actin-associated proteins (Malmstrom et al., 2004), leading to increased cell contractility and thereby establishing a positive-feedback loop for maintaining the CAF phenotype (Meyer-Ter-Vehn et al., 2006). Reduced differentiation of MSC-CAFs upon Bleb treatment of MSCs illustrates the importance of stiffness-sensing for MSC differentiation into MSC-CAFs expressing high levels of αSMA and LMNA. Our findings are consistent with previous studies that have shown that contractility and stiffness can drive increased expression of LMNA, which in turn regulates αSMA expression (Dingal et al., 2015; Swift et al., 2013). Additionally, Bleb inhibits conversion of latent TGFβ to its active form, which is required for CAF differentiation (Wipff et al., 2007). CAFs are known to display plasticity (Öhlund et al., 2017), which might depend on the extent of differentiation. The high LMNA expression we observed in MSC-CAFs from 2 kPa gels, combined with their relative insensitivity to Bleb, indicates stable differentiation, which might be attributable to further epigenetic changes (Vizoso et al., 2015). A thorough investigation of these pathways and mechanisms is beyond the scope of this work and will be explored in the future.

CAFs regulate cancer cell proliferation, invasion and metastasis via cell–cell interactions and the secretion of soluble factors and chemokines (Wang et al., 2017). MMP-mediated ECM remodeling by MSC-CAFs might contribute to cancer cell proliferation either by releasing cancer cells from contact inhibition or activating integrin signaling via efficient integrin engagement (Das et al., 2017). Increased proliferation of cancer cells might also be driven by several CAF-secreted factors including SDF1 (also known as CXCL12), FGF1 and uPA (also known as PLAU) (Fiori et al., 2019). A recent study has demonstrated the presence of prosaposin in conditioned medium from MSCs, which promotes cancer cell proliferation via AKT signaling activation on stiff substrates (Ishihara et al., 2017). In contrast to this study, we found that higher cancer cell proliferation on 2 kPa gels can be attributed to the use of MSC-CAFCM rather than conditioned medium from MSCs. In vivo, CAFs with high expression of αSMA and FAP drive cancer progression by LOX-mediated matrix stiffening (Jolly et al., 2016) along with production and alignment of fibronectin, which can be attributed to increased CAF contractility (Erdogan et al., 2017). The high invasiveness of S_MF2 spheroids observed in our study may be attributed to high expression of αSMA and MMPs, proteolytic degradation, and ECM remodeling by the highly contractile MSC-CAFs from 2 kPa gels, as well as to the physical pulling of cancer cells via heterotypic adhesions (Labernadie et al., 2017). On the other hand, homophilic adhesions contributed by CD44-positive fibroblasts might promote survival and drug resistance of cancer cells (Liu et al., 2017).

Metastasis is mediated by CTC clusters, which circulate in the blood, adhere to the endothelium and subsequently extravasate (Aceto et al., 2014). Recent studies have documented the presence of CAFs in CTC clusters (Duda et al., 2010; Ao et al., 2015). Our analysis of scRNAseq data also suggests the presence of CAFs in CTC clusters. CAFs in CTCs confer shear resistance to prostate cancer cells via intercellular contacts and soluble factors, thereby maintaining their viability and proliferative capacity (Ortiz-Otero et al., 2020). Consistent with this, our S_MF2 spheroids exhibited increased adhesion and shear resistance. The prominent expression of CD44 in CTC clusters not only contributes to maintaining cohesion between cells, but also mediates endothelial adhesion (Zhang et al., 2019). ECM proteins, such as β1 and β3 integrins, also mediate CTC adhesion via binding to fibronectin on endothelial cells (Danen et al., 2002; Osmani et al., 2019). Recent reports have suggested that upon exposure to shear stresses, tumor cells in CTCs convert into cancer stem-like cells, which influences their survival and metastatic potential (Choi et al., 2019). The role of CAFs in this process needs further exploration.

Different tumors exhibit different secondary organ preferences during metastasis. Primary breast tumors metastasize to several organs, including lungs, liver and bone, with bone being the preferred site (Nguyen et al., 2009). Fenestrated endothelial cells and a discontinuous basal lamina makes it easier for CTCs to metastasize to bone marrow (Aird, 2007). A seminal study has demonstrated that MSC-enriched TNBCs exhibit bone metastasis with CAFs secreting high levels of IGF1 (Zhang et al., 2013). Consistent with this, we found expression of IGF1 in cells expressing the MSC/CAF signature, which might lead to activation of PI3K–AKT signaling, increasing their preference for the bone (Zhang et al., 2013). In addition, bone colonization is mediated by β3 integrins expressed by CAFs and cancer cells that bind to osteoblasts (Liapis et al., 1996), which might determine whether these cells undergo dormancy or develop a secondary tumor (Haider et al., 2020).

In conclusion, our study demonstrates a stiffness-dependent crosstalk between MSCs and breast cancer cells that could drive cancer invasion through increased MSC homing, differentiation into MSC-CAFs and subsequent MSC-CAF-mediated cancer invasion. Future studies targeting this crosstalk at the pre-metastatic stage might help identify treatment approaches to slow the spread of cancer. The roles of MMP2 in maintenance of the MSC-CAF state and regulation of cancer cell proliferation represent two prospective future directions.

The previously published and publicly available scRNAseq datasets used in this study were as follows: human breast cancer, NCBI GEO series GSE161529 (Pal et al., 2021) and European Nucleotide Archive (ENA) under the accession code PRJEB35405 (Wu et al., 2020); TCGA, NCBI GEO accession GSE2109 (Cancer Genome Atlas Network, 2012); CTC and MMTV-PyMT, NCBI GEO accession GSE109761 (Szczerba et al., 2019). For the human breast cancer datasets, we first performed pre-processing by filtering out cells that have unique feature counts over 2500 or less than 200. The cells containing greater than 5% mitochondrial counts were further removed. The cells remaining after the pre-processing steps were subject to normalization and scaling using the Seurat V3 package (Satija et al., 2015). For clustering the cells, we first selected the top 2000 most variable genes using the FindVariableFeatures function in Seurat. Principal component analysis (PCA) was performed for reducing the dimensions of gene expression data to the top 20 principal components. For visualizing the cells, the PCA space representation of the cells was projected in two dimensions using the Uniform Manifold Approximation and Projection (UMAP) algorithm. The FindNeighbors and FindClusters functions in Seurat V3 were used for clustering the cells in PCA space using the Shared Nearest Neighbor algorithm. The resulting cell clusters were visualized in UMAP space using the DimPlot function in Seurat. The expression of marker genes by cells in the clusters was visualized using the FeaturePlot function. The differentially expressed genes in each cluster were inferred using the FindAllMarkers function in Seurat. The DoHeatMap function was used for generating the expression heatmap for cell clusters and top 20 differentially expressed genes for all clusters.

The differentiation trajectory of the cells expressing MSC and CAF markers was reconstructed using MARGARET (Pandey and Zafar, 2022). In order to reconstruct the trajectory, first the lower-dimensional embeddings for the selected cells were computed using the metric learning approach in MARGARET [train_metric_learner() function], which also partitioned the cells into clusters. Next, the undirected cluster graph was computed based on the connectivity measure calculated using the compute_undirected_cluster_connectivity() function in MARGARET. The pseudotime for the cells was calculated using the compute_pseudotime() function by selecting the cell-state cluster expressing MSC markers (THY1 and NT5E) as the initial population. The compute_trajectory_graph_v2() function was used for learning the final directed graph that delineates the possible trajectory of the cells. The cell pseudotimes were visualized using the plot_pseudotime() function, and the expression of the marker genes on the trajectory was visualized using the plot_gene_expression() function. The expression trends for the marker genes were visualized using the plot_lineage_trends() function.

For integration of scRNAseq data from eight individuals with TNBC and 12 individuals with non-TNBC, and for the integration of the CTC and metastatic cancer datasets, we first pre-processed and normalized the datasets using the Seurat V3 package (Stuart et al., 2019). For each dataset, we selected the top 2000 highly variable genes using the FindVariableFeatures function in Seurat V3 with the ‘vst’ selection method. For batch effect elimination, anchors were inferred using the FindIntegrationAnchors function and, based on the inferred anchors, the datasets were integrated using the IntegrateData function. The integrated data were scaled, and dimension reduction, clustering and visualization steps were followed as described above.

For the integrated TNBC cells, the cluster with the enrichment of MSC markers [CD73 (NT5E), CD90 (THY1), CD106 (VCAM1)] was selected for pathway enrichment and GO analysis. For this cluster of cells, first the differentially expressed genes were identified using Seurat. After that, GO enrichment analysis of the significantly upregulated genes was performed using ClusterProfiler v4.0.5 (Wu et al., 2021b), and biological processes related to MSCs were visualized using ggplot2 (version 3.3.5).

To determine the cellular abundance of MSCs from bulk RNA sequencing data, we performed cell type deconvolution of the TCGA-BRCA bulk RNA sequencing dataset (from 194 TNBC and 899 non-TNBC samples; Cancer Genome Atlas Network, 2012) using xCell (v1.1.0; Aran et al., 2017). For cell type deconvolution using xCell, fragments per kilobase of exon per million (FPKM) data from the TCGA-BRCA dataset was used along with the xCell standard 64 cell-type signatures. Samples with deconvolution P-values greater than 0.05 were discarded. For the rest of the samples, the xCell enrichment scores (higher xCell enrichment score is proportional to the cell type abundance) for MSCs were compared for TNBC and non-TNBC groups using a two-tailed heteroscedastic t-test.

The relevant genes associated with EMT were obtained from gene module information from a single-cell atlas of human breast cancers (Wu et al., 2021a). For each cell, an EMT module score was calculated using the Seurat ‘AddModuleScore’ function. This function quantifies the activity of each gene module by summarizing the expression levels of the genes within the module. The resulting module scores provide a measure of the module's activity in each cell. UMAP was used to visualize cells coloured according to their EMT score.

Immunohistochemistry (IHC) was performed on archived histologically confirmed cases of TNBC (n=4). Archived formalin-fixed paraffin-embedded (FFPE) specimens of breast cancer cases were retrieved for the study. Sections of 4 μm thickness were taken on positively charged slides, and IHC was performed manually using the polymer technique (Kim et al., 2016). Unstained sections were deparaffinized in limonene followed by rehydration in a series of ethanol baths (100%, 90%, 75% and 50%). Heat-induced retrieval of antigen epitopes (HIER) was performed in citrate buffer (pH 6.0) using pressure cooker at 15 psi (∼103 kPa) for 30 min. Samples were then incubated for 60 min at room temperature and further processed using a REAL Envision detection system (DAKO, Glostrup, Denmark) including peroxidase/3-3-diaminobenzidine tetrahydrochloride (DAB). Details of IHC are listed in Table S2. All breast cancer patient samples used in this study were collected in compliance with ethical guidelines of Tata Memorial Hospital and with informed consent from the patients. The study protocol was approved by Tata Memorial Hospital, and all necessary clearances were obtained prior to sample collection and analysis. The study was conducted according to the principles expressed in the Declaration of Helsinki.

Primary human MSCs were obtained from Lonza (cat# PT-2501) and MDA-MB-231 human breast cancer cells were from the National Centre for Cell Science (NCCS), Pune, India. MSCs were maintained in complete medium consisting of low-glucose Dulbecco's modified Eagle's medium (DMEM; Himedia, cat# AL006A) supplemented with 16% fetal bovine serum (FBS; Thermo Fisher Scientific, cat# 12662029), 1% penicillin-streptomycin (pen-strep; Sigma, cat# P4333) and 1% L-glutamine (Thermo Fisher Scientific, cat# 35050061). MDA-MB-231 cells were cultured in high-glucose DMEM (Himedia, cat# AL006) with 10% FBS (Himedia, cat# 1112), 1% pen-strep and 1% L-glutamine. For TGFβ inhibition experiments, a TGFβ-neutralizing antibody (R&D systems, cat# MAB1835) was added at a concentration of 1.25 μg/ml. Contractility inhibition experiments were carried out by adding the non-muscle myosin II inhibitor blebbistatin (5 μM; Sigma, cat# B0560).

Polyacrylamide gels of various stiffnesses (0.5, 2 and 5 kPa) were fabricated on 3-APTMS-functionalized (Sigma, cat# 440159) glass coverslips by combining 40% acrylamide and 2% bis-acrylamide in specific ratios as described previously (Tse and Engler, 2010). After UV crosslinking the gels with Sulfo-SANPAH (Thermo Fisher Scientific, cat# 22589) in 50 mM HEPES buffer (SRL chemicals, cat# 63732), gels were functionalized with 25 μg/ml of type-I collagen (Thermo Fisher Scientific, cat# A104830) at 4°C overnight followed by washing with PBS.

For collection of MDA-MB-231 secreted cancer-conditioned medium (CCM) on polyacrylamide gels and tissue culture plastic, cells seeded at a density of 12,000 cells/cm² were cultured for 48 h. The collected CCM was centrifuged (300 g, 10 min) and filtered through a 0.22 μm-pore filter prior to use for experiments. For preparation of MSC-CAF-conditioned medium (MSC-CAFCM), MSCs were cultured on polyacrylamide gels for 7 days in the presence of CCM. After 7 days, differentiated MSCs were washed with PBS and cultured with MSC complete medium for the next 24 h (Fig. 5A). The medium was then collected, centrifuged and filtered, as described above, for further use. Levels of TGFβ in CCM were assayed using the TGFβ1 DuoSet ELISA kit (R&D Systems, cat# DY240), using unconditioned medium as the blank control.

For studying CCM-induced MSC chemotaxis, we modified our previously designed device (Saxena et al., 2018). Specifically, medium containing MSCs (prestained with Hoechst 33342) were mixed with different concentrations of 3D collagen (1, 2 and 3 mg/ml; BD Biosciences, cat# 354249) and introduced through one of the channel inlets. Medium containing only collagen was introduced through the other inlet. After allowing MSCs to adhere onto the 3D collagen gel (∼6 h), stiffness-modulated CCM was introduced as a chemokine at the opposite channel inlet. For measuring cell motility, time-lapse microscopy was performed for 12 h using an inverted microscope equipped with an onstage incubator (Evos FL Auto, Life Technologies). Images were acquired every 15 min. Cell motility was quantified using the manual cell tracker plugin in ImageJ (NIH). FITC–dextran (Sigma cat# 46945) was used to monitor gradient establishment in the device.

For differentiation experiments, MSCs were seeded on polyacrylamide gels at a density of 800 cells/cm². At 12 h after seeding, CCM collected from MDA-MB-231 cultures on gels of 0.5, 2 and 5 kPa stiffness were added onto the respective MSC cultures on gels of matching stiffness in 1:1 ratio (i.e. CCM:fresh medium=1:1), with MSC complete medium used as a control, for a duration of 7 days. The medium was replenished after every 48 h. For expression profiling of CAF-associated markers, integrins and MMPs, quantitative reverse transcription real-time PCR (qRT-PCR) was performed as described elsewhere (Livak and Schmittgen, 2001). Total RNA was isolated using RNeasy Mini Kit (Qiagen, cat# 74104). A total of 1 μg RNA was used in cDNA synthesis using a ProtoScript II Reverse Transcription kit (NEB, cat# e6560) as per the manufacturer's instructions. Primer details are provided in Table S1.

For immunostaining, MSCs were fixed with ice-cold 4% paraformaldehyde (pH 7) in the presence of permeabilization buffer (0.5% Triton X-100) for 1 min to remove soluble cytoplasmic proteins. After fixation, cells were washed with cytoskeleton stabilizing buffer (CSB) (60 mM PIPES, 27 mM HEPES, 10 mM EGTA, 4 mM magnesium sulphate heptahydrate, pH 7) three times, and then incubated with blocking buffer (2% BSA) for 30 min at 4°C. Cells were then incubated overnight at 4°C with primary antibody diluted in blocking buffer. Primary antibodies used were: anti-αSMA antibody (mouse monoclonal; Sigma, cat# A2547; 1:400 dilution), anti-LMNA antibody (mouse monoclonal; Sigma, cat# Ab8980; 1:400 dilution), anti-pMLC (rabbit; Cell Signaling Technology, cat# 3671S; 1:400 dilution), anti-Stro-1 (mouse; Abcam, cat# ab102969; 1:400 dilution), anti-CD44 (mouse; Novus, cat# 8E2F3; 1:400 dilution) and anti-ITGB1 (rabbit; Abcam, cat# ab183666; 1:400 dilution). The following day, cells were washed with CSB thrice and then incubated with the respective secondary antibody (Life Technologies, cat# A11061 and Abcam, cat# ab175473; 1:1000 dilution), Alexa Fluor 488–phalloidin (Life Technologies; 1:400 dilution) and Hoechst 33342 (Life Technologies; 1:1000 dilution) for 2 h at room temperature. Cells were imaged using a laser-scanning confocal microscope (LSM 710, Zeiss 40× magnification), Olympus inverted fluorescence microscope (40× or 60× magnification) or Zeiss spinning-disc confocal microscope (20× magnification). Intensity of cells was quantified for multiple cells across three independent experiments for each condition using ImageJ (NIH).

For cell proliferation measurements, MSCs cultured on polyacrylamide gels in the presence and absence of CCM were imaged (15 random images) at days 1, 2, 3, 5 and 7. Population doubling was calculated by dividing the average number of cells per frame on a given day by the average number of cells per frame on day 1. Cell spreading area was determined by manually drawing the outline of individual cells using the polygonal tool in ImageJ (NIH).

For traction force microscopy measurements, 5 kPa gels were prepared on 22×22 mm² coverslips. Once the gels solidified, a thin layer of 25 μl of 5 kPa gel with 1 μm rhodamine-labeled fluorescent beads (Fluka; 1:50 dilution) was added over the previously prepared gel. ECM coating was performed as described above. At 24 h after cell seeding, the cells were imaged at selected locations for phase contrast and red fluorescence. Cells were then lysed using Triton-X 100 without moving the plate containing the gels and red fluorescence images were again captured. For calculating the magnitude of traction forces, MATLAB code from J. P. Butler was used (Butler et al., 2002).

For collagen degradation assays, gels were prepared and coated with DQ-collagen (1:20 with normal collagen; Invitrogen, cat# D12060) overnight at 4°C. After culturing MSCs and MSC-CAFs on these substrates for 48 h, cells were fixed using 4% paraformaldehyde, and then stained with Texas Red–phalloidin and Hoechst 33342. Fluorescence images (degraded DQ-collagen, green; F-actin, red; nucleus, blue) were captured at 60× objective magnification using an Olympus IX83 microscope. Integrated intensity of degraded DQ-collagen per cell was quantified using NIH ImageJ and normalized with respect to that of MSCs at day 1.

For measuring stiffness of cells, samples were probed with 10 kHz silicon nitride pyramidal probes mounted on an atomic force microscope (MFP-3D AFM, Asylum). At least 50 force indentation curves were recorded per condition.

MSCs cultured with CCM on 2 kPa and 5 kPa gels were detached using TrypleE (Life Technologies, cat# 12604013), pelleted and resuspended in medium. Cancer cells from tissue culture plastic dishes were also trypsinized and resuspended in medium. Homospheroids were generated using MDA-MB-231 cells alone (S_M), heterospheroids were generated by mixing MSC-CAFs from 2 kPa or 5 kPa gels with MDA-MB-231 cells in a 1:1 ratio (S_MF2 and S_MF5, respectively) (Fig. 5F). Single spheroids were generated using the hanging drop method (Leung et al., 2015) by incubating cells in medium containing rat-tail collagen I (Thermo Fisher Scientific, cat# A104830) at 37°C, 5% CO­₂ for 48 h. Cell viability within spheroids was assessed by carrying out live/dead staining using Calcein AM (Life Technologies, cat# C3099) and propidium iodide (PI; Himedia, cat# TC-252). For spheroid invasion experiments, spheroids were embedded in 1.5 mg/ml 3D collagen gels by mixing collagen with 10× PBS and DMEM at 4°C followed by incubation for collagen gel formation. After 30 min, wells were flooded with medium. Images were captured at different timepoints starting from 0 h to 7 days. The extent of invasion was quantified by measuring the area invaded by cells migrating out from the spheroids and normalized with respect to invasion of MDA-MB-231 spheroids (i.e. S_M). To assess the spatial position of MSC-CAFs in heterospheroids, MSC-CAFs were stained with CellTracker Red (Thermo Fisher Scientific, cat# C34552) and MDA-MB-231 cells with CellTracker Green (Thermo Fisher Scientific, cat# C2925). Spatial distribution of MSC-CAFs and cancer cells was observed 48 h after implanting spheroids in collagen gels.

For spheroid adhesion experiments, spheroids were seeded on fibronectin-coated (Sigma, cat# F1141) glass coverslips. Images were taken at 0 h to count the number of spheroids seeded, and samples were then incubated for 2 h. Coverslips were washed with PBS after 2 h to remove non-adherent/weakly-attached spheroids, and then imaged to count the remaining spheroids. Percentage spheroid attachment was calculated across all conditions. For estimating the strength of adhesion, trypsin de-adhesion experiments were performed wherein after 2 h of adhesion onto fibronectin-coated coverslips, spheroids were incubated with trypsin-EDTA (Hi-media, cat# T001; 1×) and imaged at 10 s intervals till the spheroids start to detach, i.e. t=τ* (Soumya et al., 2014a,b). The time required for detachment, τ_detach, in minutes, corresponds to the time when there was a visible shift observed in the centroid position of the spheroid, as shown in the pseudocolor images (pink and blue) in Fig. 6F.

For spheroid shear experiments, a straight channel 400 μm in width with three ports was designed (Fig. 6G). The shear stress profile in the zone of interest was simulated using COMSOL MultiPhysics (version 5.2) assuming laminar flow conditions. The device designs were printed on to a transparency mask using a printer with resolution 600 dpi and higher. The design was patterned on a silicon wafer by photolithography with SU-8 (2050) photoresist (MicroChem). Polydimethylsiloxane (PDMS) devices were fabricated by pouring PDMS (Sylgard 184) onto the master, curing them in an oven, punching the ports using a 3 mm biopsy punch, and bonding to glass using plasma oxidation. The devices were coated with fibronectin prior to seeding spheroids in the middle well and allowing them to adhere for 10–12 h. After sealing the spheroid ports, medium from the inlet was flown at a rate of 400 μl/min. τ_shear corresponds to the time duration for which individual spheroids remained attached under shear. This was determined based on change in the spheroid position before and after flow was started. The position of individual spheroids is depicted in pseudocolor (pink and blue) in Fig. 6I.

All statistical analyses were performed using Origin 9.1, with P<0.05 considered to be statistically significant. Based on the normality of data assessed using the Kolmogorov–Smirnov normality test, one-way ANOVA were performed to assess statistical significance and Fisher post-hoc tests were used to compare the means.

We thank the Centre for Nanoelectronics, IIT Bombay for providing lithography facilities, and Industrial Research and Consultancy Centre (IRCC) for confocal (LSM and spinning disc) and AFM central facilities.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261145.reviewer-comments.pdf