Abstract
Alterations of the extracellular matrix (ECM), including both mechanical (such as stiffening of the ECM) and chemical (such as variation of adhesion proteins and deposition of hyaluronic acid (HA)) changes, in malignant tissues have been shown to mediate tumor progression. To survey how cells from different tissue types respond to various changes in ECM mechanics and composition, we measured physical characteristics (adherent area, shape, cell stiffness, and cell speed) of 25 cancer and 5 non-tumorigenic cell lines on 7 different substrate conditions. Our results indicate substantial heterogeneity in how cell mechanics changes within and across tissue types in response to mechanosensitive and chemosensitive changes in ECM. The analysis also underscores the role of HA in ECM with some cell lines showing changes in cell mechanics in response to presence of HA in soft substrate that are similar to those observed on stiff substrates. This pan-cancer investigation also highlights the importance of tissue-type and cell line specificity for inferences made based on comparison between physical properties of cancer and normal cells. Lastly, using unsupervised machine learning, we identify phenotypic classes that characterize the physical plasticity, i.e., the distribution of physical feature values attainable, of a particular cell type in response to different ECM-based conditions.
Similar content being viewed by others
Introduction
The stiffness of normal tissues, usually characterized by a shear or Young’s modulus, varies over a wide range depending on the type of tissue, ranging from ~100 Pa for bone marrow or fat to near MPa for some muscles and even more for cartilage1,2. For a specific tissue, animal, and age, the elastic modulus is a tightly controlled quantity, but it undergoes large changes in some diseases such as fibrosis and cancer, especially when the tissues are mechanically stressed as they are in vivo3. The effects of changes in the mechanical properties of tissues on the cells within them are increasingly implicated both in the normal development of tissues and in the pathological functions of cells in various disease states4,5,6. In addition to changes in physical properties, changes in tissue stiffness triggered by injury or disease can lead to chemical changes in the extracellular matrix (ECM) including its most abundant constituents such as collagen, fibronectin, and hyaluronic acid. In cancer biology, the increased stiffness of several types of tumors, most prominently breast, liver, and colorectal tumors, has motivated studies of mechanical sensing by both cancer and stromal cells, which have yielded accruing evidence that changes in cellular environment both in vitro and in vivo can affect the structure and function of cells in a manner that promotes malignancy7,8. The combination of physical and chemical changes in ECM can direct tumor growth and dispersion9. Apart from altered ECM conditions in the primary tissue site, metastasizing cancer cells will often encounter substantially different ECM chemical composition and mechanical stiffness in metastatic tissue sites (such as lung to brain, breast to bone metastasis10). Therefore, it is crucial to understand the response of the cells to changes in ECM stiffness and chemistry.
Precisely how cells detect or respond to mechanical signals, and how that response depends on the chemistry of the ECM is only beginning to be revealed, with many proteins implicated in the capacity of cells to respond to mechanical cues11. It is also not clear whether a response to mechanical cues is a general feature of most cell types or whether it is different in normal and malignant cells. Studies have found differing responses to mechanical environments even among closely related cell types12. For example, pluripotent cells derived from human bone marrow can be directed to differentiate into highly distinct cell types by manipulating their substrate elastic modulus to levels that span the range from very soft tissues like the brain to much stiffer tissues like muscle or bone13. In contrast, a similar study of murine embryonic stem cells showed very little dependence of the cells on substrate stiffness14. In the context of cancer, a reasonable expectation might be that since one of the hallmarks of the malignant transformation is the ability to grow in soft agar15, cancer cells might have much less response to mechanical signals than the normal cells from which they derived, which typically are much more limited in their proliferation rates on substrates with normal physiological stiffness16. One of the earliest quantitative studies of stiffness sensing in fibroblasts showed that whereas NIH 3T3 fibroblasts responded very strongly to changes in substrate elastic modulus, by altering their morphology, their cytoskeletal assembly, and the degree of protein tyrosine phosphorylation, transfection of these cells with H-Ras eliminated the response of the cells to stiffness, producing a phenotype similar to that seen on plastic even when cells were grown on very soft substrates17,18. However, a total loss of mechanical sensitivity is not a universal feature of cancer cells11,12, and even highly abnormal cells such as HeLa cells respond robustly to changes in substrate elastic modulus19,20.
Thus, to gain a perspective on the relative importance of substrate characteristics on the structure, and motility of cancer cells on substrates with defined stiffness and chemical composition, 25 different human cancer cell lines derived from tumors in breast, colon, brain, ovary, pancreas, prostate, and skin tissues were studied along with 5 immortalized but non-tumorigenic cell lines from the same set of tissues. Using this pan-cancer mechanobiology dataset, we inferred how the cellular mechanics (adherent area, cell shape, cell stiffness, and motility) are influenced by varying biologically realistic ECM physicochemical properties that capture cell-ECM interactions for major solid tumor types. We find heterogeneity to be the norm in how different cell types respond in terms of change in cellular mechanics to mechanosensitive and chemosensitive changes in ECM conditions. Our results also caution against generalizing comparison of normal vs cancer cell mechanics across and even within tissue types. Lastly, we show that behavior of different cell lines across various substrates can be grouped into physical feature-specific classes that define cellular plasticity in terms of the range of physical feature values a specific cell line can attain in response to a specific substrate.
Results
The pan-cancer dataset provides a rich array of cellular physical measurements (spread area, shape (circularity and aspect ratio), cell stiffness, and motility) across 30 cell lines spanning 8 different tissue types on varying substrates which consist of polyacrylamide (PAAm) gels with elastic moduli of 500 Pa or 30 kPa coated with either fibronectin (500 Pa FN, 30 kPa FN) or collagen I (500 Pa Coll, 30 kPa Coll), cross-linked hyaluronic acid gels with elastic modulus of 500 Pa coated with fibronectin (HA FN) or collagen I (HA Coll), or glass slides (Fig. 1a). All gels were greater than 100 µm in height, well above the distance over which a cell might sense the support beneath it21. Out of the 30 cell lines, 25 are cancer cell lines and 5 are immortalized but non-malignant (normal) cell lines (Fig. 1b). Live cell imaging at 24 h was used to measure cell area, aspect ratio, and circularity. For example, Fig. 1c shows brightfield microscopy images of the T98G glioblastoma cell line along with median values of its measured physical features across different substrates. Motility was assessed by tracking individual cells during time lapse movies. Cell stiffness was quantified using atomic force microscopy, yielding measurements of Young’s modulus. It is important to note that, unlike cell morphology and motility data at single cell level, cell stiffness data involves measurements from multiple cells but obtained from different locations on each cell. The cell stiffness measurements from AFM can vary as a result of position within the cell22,23. Analyzing the relationship between the physical features, we found cell area to be highly correlated with shape features (aspect ratio and circularity) (Supplementary Fig. 1). Cell stiffness is not correlated with any of the other physical properties, and cell speed is weakly correlated to morphology (Supplementary Fig. 1).
a 30 cell lines (including 25 cancer and 5 normal cell lines) that span eight different tissue types were grown on seven different substrates and physical properties of the cells (namely, area, shape (circularity and aspect ratio), stiffness, and motility) were measured. 500 Pa Coll, 500 Pa FN, 30 kPa Coll, 30 kPa FN: polyacrylamide (PAAm) gels with elastic moduli of 500 Pa or 30 kPa coated with either fibronectin or collagen I. HA Coll, HA FN: cross-linked hyaluronic acid gels with elastic modulus of 500 Pa coated with fibronectin or collagen I. Schematic created in BioRender. Parihar, K. (2025) https://BioRender.com/rvjr4y4. b Tables listing the tissue type and cell lines considered. (N) refers to non-malignant (normal) cell lines. c Examples of cell area, aspect ratio, stiffness, and speed (median values) on different substrates for T98G along with phase-contrast image of the cells.
Heterogeneity in sensitivity to changes in ECM stiffness and composition
One of the primary ways cells respond to ECM alterations is through changes in their physical properties. We thus asked, how sensitive are the morphology, cell stiffness, and cell speed of different cell lines to change in substrate conditions? The effect of substrate sensitivity on the phenotypic features is analyzed using the relative change in the median value of the feature when moving from one substrate to another. Due to skewness observed in the data, we used the median value as a population-level representative of a cell line’s physical behavior on a particular substrate. We analyze both mechanosensitive and chemosensitive changes in ECM conditions. The alterations considered include (i) increase in stiffness of PAAm-based substrate (30k–500 Pa Coll and 30k-500Pa FN), (ii) 500 Pa PAAm-based to 500 Pa HA-based substrates (HA-500Pa Coll and HA-500Pa FN), (iii) change in integrin ligand on HA-based substrate (HA Coll-FN), on soft (500 Pa Coll-FN) and stiff (30 kPa Coll-FN) PAAm-based substrates, (iv) 500 Pa HA-based to stiff 30 kPa PAAm-based substrate (30 kPa-HA Coll and 30kPa-HA FN) and (v) stiff PAAm-based substrate to extremely high, supraphysiological stiffness of glass (Glass-30kPa Coll and Glass-30kPa FN).
Increase in ECM stiffness of PAAm-based substrate (30k–500 Pa) significantly increases area in melanoma cells, with a similar effect observed on soft HA-based substrates as compared to soft PAAm-based substrates (HA-500Pa) (Fig. 2a). Melanoma cells (SK-MEL-2, WM266-4, MeWo) also have greater spread area on soft HA-based substrates as compared to stiff PAAm-based substrates (30 kPa-HA Coll and 30kPa-HA FN; Supplementary Fig. 2a). In contrast, change from 500 Pa PAAm to 500 Pa HA based substrate (HA-500Pa) causes a decrease in area for several cell lines in other tissue types (Fig. 2a). Interestingly, even on stiff PAAm-based substrate, some of the cell lines (22Rv1, SW620, U-87, Capan-1, and SKOV-3) show a decrease in spread area as compared to soft PAAm-based substrate (30k–500 Pa, Fig. 2a). Corresponding to an increase or decrease in area with change in ECM conditions, the shape of the cell becomes more elongated (higher aspect ratio or lower circularity) or more rounded (lower aspect ratio or higher circularity) respectively (Supplementary Figs. 4, 5).
For each cell line, the ratio of the median values for a cell area, b cell stiffness, and c cell speed as a measure of the phenotypic sensitivity to substrate change. (30k–500 Pa Coll: 30 kPa Coll/500 Pa Coll, 30k-500Pa FN: 30 kPa FN/500 Pa FN, HA-500Pa Coll: HA Coll/500 Pa Coll, HA-500Pa FN: HA FN/500 Pa FN). (N) refers to non-malignant (normal) cell lines. See also Supplementary Figs. 2–5. ***p < 0.01; **p < 0.05; *p < 0.1, adjusted for multiple testing using Benjamini-Hochberg procedure. For each cell line, phenotypic sensitivity to substrate change is calculated only if there are at least 25 data points (n ≥ 25) for the physical feature of interest on both the substrates. See Supplementary Tables 1–3 for the exact value of n for the cell lines.
In contrast to the substrate effects on cell area, the stiffness of melanoma cells does not change significantly with changes in ECM stiffness or in presence of HA (Fig. 2b). Colon cells, on the other hand, show an increase in cell stiffness on stiffer PAAm-based substrates (30k–500 Pa) as well as for HA-500Pa FN alteration (Fig. 2b). Notably, some of the cell lines (Caov-3, NCI-H2126) also show a decrease in cell stiffness with increasing ECM stiffness (Fig. 2b). As compared to changes observed in area, a switch in the type of integrin ligand (HA Coll-FN, 500 Pa Coll-FN, 30 kPa Coll-FN) does not have significant effect on cell stiffness (Supplementary Fig. 2). Going from gel-based substrates to glass leads to a significant increase in cell stiffness across several cell lines, however even in this case cell lines with no significant change in cell stiffness as well as the opposite behavior of decreasing cell stiffness can be seen (Supplementary Fig. 2b). For cell speed as the phenotype of interest, increase in ECM stiffness (30k–500 Pa) can lead to either increase or decrease in migration speeds in a cell line dependent manner across tissue types (Fig. 2c). A similar mix of increase and decrease in cell speed across cell lines is observed in all the other ECM alterations considered in the analysis (Fig. 2c and Supplementary Fig. 3).
Altogether, there is substantial heterogeneity in the mechanosensitive and chemosensitive response of cell lines across and within tissue types to changes in ECM composition and stiffness. In particular, the changes in different physical properties (morphology, cell stiffness, and cell speed) vary considerably compared to each other.
Difference between normal and cancer cells in terms of physical response depends on the substrate conditions and the cell lines being compared
Comparing the median and distribution of physical feature values across all substrates revealed some notable differences between cancer and normal cell lines (Supplementary Fig. 6, 7). The normal breast cell lines are stiffer as compared to their cancer counterparts, which is in contrast to the normal lung cell line being softer than cancer cells (Supplementary Fig. 7a). The normal cell lines (namely, hTERT-HPNE, NL20, and hTERT-HME1) have greater median migration speeds and a much broader distribution than their cancer counterparts in the respective tissue types (Supplementary Fig. 7b). Other observations to note are the outlier-like behavior of the pancreatic normal cell line hTERT-HPNE that has a much larger area and highly non-circular shape (i.e., high aspect ratio) (Supplementary Fig. 6, 7c), and the pancreatic cancer cell line Capan-1 that has much greater cell stiffness as compared to all the other cell lines (Supplementary Fig. 7a, d).
Next, we assessed how the cancer and normal cell behavior differ within each of the substrate types by comparing the median values of the physical features. The normal breast cell line hTERT-HME1 has greater median area than all the cancer cell lines on 500 Pa Coll and 30 kPa PAAm substrates but not on HA substrates (Fig. 3a). The normal breast cell line MCF10A, on the other hand, has median area smaller than or similar to HCC1937 on all substrates and greater median area than the other 3 cancer cell lines on 500 Pa FN, 30 kPa, and glass substrates (Fig. 3a). Similar cell line and substrate dependence in cancer vs normal comparison can be seen for lung and prostate tissues as well. The normal lung cell line NL20, depending on the substrate type, has either similar or smaller median spread area than the lung cancer cell line NCI-H2126 (Fig. 3b). However, as compared to NCI-H2087, NL20 has either similar or larger median spread area depending on the substrate type (Fig. 3b). The normal prostate cell line RWPE-1 has smaller median spread area than the cancer cell lines DU145 and PC-3 on all substrates, but as compared to LnCaP, RWPE-1 has smaller median spread area on all except 500 Pa Coll and 30 kPa Coll substrates (Fig. 3c). The pancreatic normal cell line hTERT-HPNE has substantially higher spread area, not only compared to other pancreatic cell lines but all the other cell lines, on 500 Pa PAAm, 30 kPa PAAm, and glass substrates with reduced though still high area on HA substrates (Supplementary Fig. 8a), underscoring its outlier-like behavior noted before in across substrate comparison. In the context of aspect ratio, pancreatic, breast, and lung normal cell lines have higher aspect ratio than the corresponding cancer cell lines on 500 Pa and 30 kPa PAAm substrates (with a few exceptions) (Supplementary Figs. 8b, 9a, b). In contrast, the prostate normal vs cancer cell line comparison of aspect ratio is much more cell line dependent (Supplementary Fig. 9c).
a, d Breast, b, e lung, and c, f prostate cell lines on soft (500 Pa) and stiff (30 kPa) PAAm Coll and FN substrates, soft (500 Pa) HA substrates coated with Coll and FN, and glass. (N) refers to non-malignant (normal) cell lines. See also Supplementary Figs. 8 and 9. ***p < 0.01; **p < 0.05; *p < 0.1, adjusted for multiple testing using Benjamini-Hochberg procedure. The number of measurements for the physical feature of interest is at least 25 (n ≥ 25) for each cell line on a particular substrate. See Supplementary Tables 1 and 2 for the exact value of n for the cell lines.
In terms of cell stiffness, all the breast cancer cell lines are softer than MCF10A on all the substrates and softer than hTERT-HME1 on all except HA substrates (Fig. 3d). In contrast, the lung cancer cell lines are stiffer than the normal cell line NL20 on all the substrates (Fig. 3e). For prostate tissue, similar to area, the cancer vs normal comparison with cell stiffness as the metric is cell line dependent on all substrates with the exception of HA Coll where normal cell line RWPE-1 has lower median cell stiffness than all the cancer cell lines (Fig. 3f). In the case of pancreas, the outlier-like behavior of Capan-1 cancer cell line, noted before in across substrate comparison, in terms of its cell stiffness being substantially higher as compared to other cell lines (pancreatic and other tissues) can be seen on all the substrates (Supplementary Fig. 8c).
Within substrate comparisons for cell motility as the metric of response shows a more nuanced picture than the prior observation from across substrate comparison of breast, lung and pancreatic normal cell lines having greater migration speeds than their cancer counterparts. Normal breast hTERT-HME1 cells are faster than all the cancer cells only on 500 Pa Coll, 30 kPa, and glass substrates, whereas comparison of cancer cell lines with breast normal cell line MCF10A are much more cell line dependent on all the substrates (Fig. 4a). The normal lung cell line NL20 is substantially faster than the cancer cell lines on all the substrates (Fig. 4b). Pancreatic normal cell line hTERT-HPNE shows higher median migration speed than both the cancer cell lines only on the 30 kPa FN, HA Coll, HA FN, and glass substrates (Fig. 4c). The normal prostate cell line RWPE-1 is faster than the non-metastatic cancer cell line 22Rv1 on all except HA Coll substrate, whereas, compared to LNCaP and DU145 (metastatic cell lines24,25), RWPE-1 cells are slower and faster on HA and 30 kPa substrates respectively (Fig. 4d).
a Breast, b lung, c pancreatic, and d prostate cell lines on soft (500 Pa) and stiff (30 kPa) PAAm Coll and FN substrates, soft (500 Pa) HA substrates coated with Coll and FN, and glass. (N) refers to non-malignant (normal) cell lines. ***p < 0.01; **p < 0.05; *p < 0.1, adjusted for multiple testing using Benjamini-Hochberg procedure. For each cell line on a particular substrate, n ≥ 25 cells. See Supplementary Tables 3 for the exact value of n for the cell lines.
In addition to speed, directional persistence is an important property of cell migration. To determine migratory persistence for cell lines on different substrates, we calculated directional autocorrelation as a function of time26 and then used it to estimate decorrelation time up to which migration can be viewed as directionally persistent. For breast, the normal cell line hTERT-HME1 has higher persistence than cancer cell lines only on 500 Pa Coll and 30 kPa Coll (Supplementary Fig. 15a). The lung and pancreatic normal cell lines (NL20 and hTERT-HPNE) have more persistent migration on all the substrates as compared to their cancer counterparts, though persistence for hTERT-HPNE decreases substantially on HA Coll (Supplementary Fig. 15b, c). Among prostate cell lines, the normal cell line RWPE-1 shows more persistence than cancer cell lines on 500 Pa Coll, 30 kPa Coll, 30 kPa FN, and HA FN substrates.
Altogether, the results show that for each physical feature as the metric of response, there is a lack of generalization between tissue types and different substrates for comparisons between normal and cancer cells. Further, within a tissue type, there are cases of both cell line independent (like normal lung cells being faster and more persistent than cancer cells, breast normal cells being softer than cancer cells) and cell line dependent comparisons between normal and cancer cells for metric of response being any of the physical features.
Distinct mechanotypes characterize the physical response of cells to ECM-based cues
The highly variable response of cell lines to substrate changes prompted us to ask if there are potential underlying similarities in how different cell lines phenotypically behave on different substrates. We first performed principal component analysis (PCA) using the median values of the five physical features for each of the cell line-substrate pairs (Sup. Fig. 10). There were no clear clusters in the PCA plots except the separation of outlier-like cases mentioned in preceding section for spread area and cell stiffness, namely hTERT-HPNE (normal, pancreas) and Capan-1 (cancer, pancreas) respectively. This suggests that the potential underlying similarities in phenotypic behavior across substrates might be physical feature-specific instead of being defined by a combination of them.
Next, we used all the measurements (instead of only median values as measure of response) to compare physical feature-specific behavior of cell lines across the substrates. Figure 5a shows the analysis workflow employing an unsupervised machine learning framework using consensus clustering27,28 to identify phenotypic classes for each physical feature. For comparison of cell lines within and across substrates (i.e., comparing cell line-substrate pairs), we used Wasserstein-1 distance29,30. The number of phenotypic classes for each of the physical feature is determined based on the proportion of ambiguous clustering (PAC)31 and Calinski-Harabasz Index (CHI)32. That is, the optimal number of clusters corresponds to low PAC and high CHI. Additionally, given the inherent uncertainty in unsupervised clustering, we also separate out possible boundary cases (cell line-substrate pairs) which do not have a high confidence score for being part of one cluster over another (see methods for details).
a Schematic representation of the unsupervised machine learning approach. ECDF: Empirical cumulative density function. Schematic created in BioRender. Parihar, K. (2025) https://BioRender.com/e521y03. Mechanotypes for cell b area, c stiffness, and d speed, whereby the heatmaps show the phenotypic class for each cell line-substrate pair and the KDEs correspond to characteristic density function for each class. The numeric values shown in the heatmaps correspond to median values of the physical feature for each cell line-substrate pair. (N) refers to non-malignant (normal) cell lines. See also Supplementary Figs. 11–13. Note that in this analysis only those cell line-substrate pairs are considered which have at least 25 data points (n ≥ 25) for the physical feature of interest. See Supplementary Tables 1–3 for the exact value of n for the cell lines.
Based on this unsupervised learning methodology, 5 classes (mechanotypes) for area and aspect ratio, 4 for cell stiffness, and 3 and 2 mechanotypes respectively for circularity and motility were identified (Fig. 5b–d, Supplementary Fig. 11–13). The outlier-like cases of area (hTERT-HPNE) and cell stiffness (Capan-1), noted in previous section, separate out into a distinct class. Also, the melanoma cell line MeWo on glass is separated out as a single class for aspect ratio due to its much higher values as compared to other cell line-substrate pairs. Thus, we primarily have 4 mechanotypes each for area and aspect ratio, and 3 mechanotypes for cell stiffness. We also compared the performance of Kolmogorov-Smirnov (KS) distance, a commonly used metric for comparing distributions, in identifying relevant clusters. As compared to Wasserstein-1 distance, unsupervised clustering based on KS distance was unable to separate out the outlier-like behavior of hTERT-HPNE in area and Capan-1 in cell stiffness (Supplementary Fig. 14).
The difference between mechanotypes can be seen from their respective kernel density estimates (KDEs), obtained by combining data from all the cell line-substrate cases falling in that class (the boundary cases have not been included in the density estimation). More specifically, we can see that certain cell lines have more plasticity in terms of the range of values they can attain for a specific physical feature as compared to others on a particular substrate and across substrates. Each substrate has at least one cell line within each of the mechanotypic classes for each physical feature. Within each tissue type, there is a mix of 2 or more mechanotypes with some interesting patterns. Among breast tissue cell lines, the normal cell lines (hTERT-HME1 and MCF10A) and highly aggressive basal-like cell lines33,34 (MDA-MB-231 and HCC1937) mainly have intermediate to high area mechanotypes, whereas less aggressive luminal-like cell lines33,34 (T-47D and MCF7) primarily have the lowest area mechanotype (Fig. 5b). The colon-derived cells are largely part of only the lowest area mechanotype (blue KDE in Fig. 5b) and belong to the intermediate to high cell stiffness mechanotypes (green and black KDEs in Fig. 5c). Similarly, ovary cells primarily show an intermediate cell stiffness mechanotype (green KDE in Fig. 5c) along with lower area mechanotypes (blue and green KDEs in Fig. 5b). Skin cells, on the other hand, show intermediate to high cell stiffness mechanotypes, but with a more varied range of area mechanotypes as compared to colon and ovary cells. In contrast to these tissues, the low cell stiffness mechanotype (blue KDE in Fig. 5c) is primarily present in lung and prostate tissues.
Both low and high cell speed mechanotypes (Fig. 5d) show a broad range of migratory persistence. For instance, while the high cell speed mechanotype comprises cell lines (hTERT-HPNE and RWPE-1) showing some of the most persistent migration, the high cell speed mechanotypes of MCF7 and MDA-MB-231 cell lines have some of the lowest persistence among all cell lines (Supplementary Fig. 16). The melanoma cell line SK-MEL-2 has low cell speed mechanotype but shows higher persistence as compared to other melanoma cell lines (A375, WM266-4) which have high cell speed mechanotype (Supplementary Fig. 16). The high cell speed mechanotype LnCaP and DU145 prostate cancer cell lines have similar persistence as the low cell speed mechanotype 22Rv1 prostate cancer cell line (Supplementary Fig. 16). Even within the same cell line SW620, the low and high cell speed mechanotypes on different substrates have similar persistence (Supplementary Fig. 16). The data, thus, suggests migratory persistence to be not necessarily dependent on the cell speed mechanotype.
It is important to note the possibility that a cell line may stay within the same mechanotypic class across two substrate conditions but change its primary sampling region within the distribution leading to a significant change in physical property. For instance, SW620 stays within the low area (blue KDE in Fig. 5b) and intermediate cell stiffness (green KDE in Fig. 5c) mechanotype in both 500 Pa Coll and 30 kPa Coll but shows a significant decrease in median area and increase in median cell stiffness with an increase in ECM stiffness from 500 Pa Coll to 30 kPa Coll (Fig. 2a, b).
In summary, the unsupervised learning-based analysis shows that the versatility in the physical response of cells to ECM composition and stiffness can be categorized into specific classes (mechanotypes) mapping to narrow, intermediate, and broader distributions of physical feature values. To alter specific aspects of cellular mechanics with change in substrate conditions, the cells can stay within the mechanotype but change the region they predominantly sample from in the corresponding distribution, or simply switch between mechanotypes.
Discussion
Cellular function, normal or pathological, is influenced by interactions with the microenvironment4. Here, we studied how differing ECM mechanics and chemical compositions impact the physical response of cancer and normal cell lines from 8 different tissue types. Conventionally, it has been expected that cells will spread more and increase in stiffness with increasing substrate stiffness. However, accruing evidence suggests a more complicated picture12. While increase in spread area on stiffer ECM has been previously shown for fibroblasts35, endothelial cells36,37,38,39, hepatic stellate cells40,41,42,43, and myocytes44,45,46,47,48, insensitivity to change in ECM stiffness or a decrease in area on stiffer ECM has also been reported for several other cell types14,49,50,51,52. These cellular responses are further contingent on the type and density of the integrin ligand available for adhesion53,54,55. In the context of cell stiffness, similar contrasting behaviors of increase in cell stiffness or insensitivity to change in ECM stiffness depending on the cell type have also been reported13,14,54,56,57,58. In the case of motility, it has been argued that the migration speed of cells has a maximum speed at a particular substrate stiffness and thus, speed can increase or decrease with changing ECM stiffness depending on which side of the peak speed cell is on59,60,61. The ECM stiffness for peak speed appears to differ among cell types and is further dependent on the ligand density59,61. Specifically for cancer cells, studies have found differing physical responses to change in substrate stiffness17,49,58,62,63,64. Consistent with these prior studies, our results of 30 distinct cancer and normal cell lines also show a diverse trend within and across tissue types in mechanosensitive response to increasing ECM stiffness, with the changes in morphology, cell stiffness, and motility further dependent on integrin ligand type. While the conventional picture of increasing area and cell stiffness could be seen more consistently across cell lines in the case of going from gel-based matrix to extremely high, supraphysiological stiffness of glass, even in this case there were some exceptions.
Hyaluronic acid (or hyaluronan, HA), a soft polymeric glycosaminoglycan present in ECM, plays an important role in development, wound healing, and cancer progression65,66,67,68. Prior studies have found that myocytes, fibroblasts, glioma, and hepatocellular carcinoma cells can respond to soft HA matrices coated with integrin ligands similarly to their response to stiff substrates in terms of increase in spread area and cell stiffening54,69,70,71. Importantly, these studies also found that the response was integrin ligand dependent and not universal across cell types54,70,71. To ascertain whether this extends to other cell types, we analyzed how the physical response of cancer and normal cell lines change on integrin ligand (fibronectin or collagen I) coated soft HA-based substrates as compared to soft and stiff inert PAAm-based substrates. Most melanoma cell lines responded with as much or even more increase in spread area on soft HA-based substrates as compared to stiff PAAm-based substrates. However, similar to mechanosensitivity, overall there was heterogeneity in the response (insensitive or significant increase/decrease) to the presence of HA across cell lines and integrin ligand types for all the physical features (area, shape, cell stiffness, and motility). While the observed effect of HA to alter cell response is not universal, our results nevertheless emphasize an important aspect, often not reported in previous studies, of considering the role of ECM mechanics in conjunction with HA when studying how a particular cell type may respond in vivo.
Mechanistic understanding of how the presence of HA, especially with an integrin ligand, impacts cellular mechanics is sparse. HA interacts with cells through membrane receptors, like CD44 and RHAMM, and these linkages can regulate cell adhesion and motility70,72,73. Prior studies suggest potential augmentation of integrin-mediated signaling with HA-mediated signaling leading to cellular response on soft matrix otherwise only seen on stiff matrix54,70,71. In particular, Mandal et al. showed that the hepatocellular carcinoma cell line (Huh7) spread as much on soft HA matrices as on stiff matrices and that decreasing PIP2 levels or inhibiting PIP2 or PIP3 function reduced spreading and adhesion of cells to HA but not on stiff substrates71. Combined with the reports that proteins involved in actin polymerization, such as formins74,75 and N-WASP76, and focal adhesion proteins, such as alpha-actinin77 and vinculin77,78, can be activated by either force or binding phosphoinositides, cells adhering to soft substrates containing both an integrin ligand and HA may form focal adhesions and undergo cytoskeleton reorganization (similar to that observed on stiff substrates) depending on the synthesis and spatial distribution of phosphoinositides. More work is thus needed to understand the mechanisms underlying the interplay between HA and integrins in mediating cellular mechanics across different cell types.
Tissue specific comparisons of cancer and normal cell lines showed that for each of the physical features as a metric of response the differences are cell line specific rather than general. Some interesting trends though can be observed at the tissue level. Among breast cell lines, normal cells were consistently stiffer than cancer cells on all substrates except on HA matrices where one normal cell line (MCF10A) remained stiffer but other normal cell line (hTERT-HME1) had comparable cell stiffness with cancer cells. This is consistent with a prior study which found cancer cells derived from breast, bladder, cervix, and pancreatic tumors to be softer than their normal counterparts49. However, it is important to note that the study used collagen-coated glass coverslip as opposed to the much broader range of substrate types used in our analysis49. Further, in contrast to breast, normal lung cells were softer than their cancer counterparts across different substrate types. These normal lung cells also had much higher migration speeds than cancer cells on all the substrates. In breast, however, one normal cell line (hTERT-HME1) had higher motility than cancer cell lines on several but not all substrates, whereas the other normal cell line (MCF10A) had similar or lower speeds than cancer cells. Altogether, the results show that to study differences/similarities between cancer and normal cells using physical properties as the metric, it is important to consider both substrate conditions (mechanical and chemical) and the cell types being compared.
Most prior studies rely on gauging differences in cellular behavior on different substrates by comparing the sample mean or median values of the metric of response. Their behavior can also be analyzed more holistically by comparing the distribution of values that cell types show on different substrates. To ascertain potential underlying patterns in the distributional response of cell lines across varying ECM conditions, we used unsupervised machine learning to identify phenotypic classes (mechanotypes) for each of the physical features. These mechanotypes are characterized by the distribution of physical feature values that cells can attain, given a specific cell type (defined here by cell line) and substrate condition. Cancer cells exploit cellular plasticity (such as epithelial-to-mesenchymal plasticity and drug-tolerant states) in adjusting to an unfavorable metabolic environment, evade immune attack, spread from a primary to metastatic site, and escape the toxic effects of anticancer drugs79,80. Extending this idea in the context of cell mechanics, the mechanotypes identified can be viewed as a measure of mechanical plasticity for a cell line on a particular substrate. That is, the distribution for each class describes plasticity of the cells in terms of physical feature values attainable.
Each cell line showed capability of switching between mechanotypes with change in substrate conditions for at least one of the physical features. This is particularly important in the context of cancer which is accompanied by extensive remodeling of the ECM8 and thus, depending on the local microenvironment cancer cells may equip themselves with means to attain required mechanical plasticity for fitness advantage. While what this fitness advantage (if any) depending on mechanotype may entail remains to be clarified, our results provide some hints in regard to the potential influence of mechanical plasticity. For example, consider the case of colon cancer cell lines, SW480 and SW620, that were established from the same patient, whereby SW480 was established from the primary site and SW620 was derived from a metastatic site81. While both show the same area mechanotype on most of the substrates, SW480 has a higher area mechanotype than SW620 on 30 kPa FN and Coll. They also show differing mechanotypes in terms of motility on 500 Pa Coll and 30 kPa Coll, and HA FN with SW620 having higher motility. SW620 also has a relatively softer cell mechanotype than SW480 across all the substrate conditions. Along similar lines, Tsujita et al. had reported that low-invasive breast cancer cells (MCF7) had higher plasma membrane tension as compared to highly metastatic breast cancer cells (MDA-MB-231)82. Consistent with this observation, we can see that MCF7 has a relatively stiffer mechanotype as compared to the MDA-MB-231 on all substrates except HA FN. Additionally, a clear distinction between breast normal and cancer cell lines can be seen in terms of cell stiffness, with normal cell lines having a high cell stiffness mechanotype and cancer cell lines showing low to intermediate cell stiffness mechanotypes (Fig. 5c). Further, among the prostate metastatic cell lines a shift from low to high area mechanotype can be observed for low to highly metastatic24,25 (LNCaP to DU145 to PC-3) cell lines for all except HA FN substrate. This analysis thus shows how the phenotypic plasticity in terms of physical characteristics of a cell type is inextricably linked to the ECM conditions. This is particularly important in the context of tumor progression, during which cancer cells experience substantial changes in the ECM mechanics and chemistry of their primary tissue and also encounter different ECM conditions in the metastatic tissue sites. Unraveling how cancer cells are able to exploit the ECM-based cues to alter their physical plasticity might provide therapeutic targets that could, for example, aid in abrogating cellular mechanical softness which helps cancer cells evade killing by cytotoxic T cells and thus, potentiate immunotherapy83.
In summary, this work underscores the fact that how cells process extracellular mechanical signals can be highly distinct between cell types or even for the same cell type on chemically distinct substrates leading to a wide range of physical responses. The mechanistic underpinning for the heterogeneity in the physical response of cell lines to change in substrate conditions and the phenotypic classes (mechanotypes) that we have documented here remains to be clarified. Future work will focus on relating the cellular microstates (characterized by key genes/proteins) underlying the mechanical and morphological responses to ECM-based cues. An important limitation of our study to note is that the pan-cancer mechanobiology dataset used here has been derived from subconfluent cells on a flat substrate. Another limitation of using the substrates we produce is that neither collagen nor fibronectin are fibrillar when they bind to the gel surface and so do not reproduce possible effects of placing integrin ligands in a linear array or providing a substrate with non-linear viscoelasticity. One advantage of the method, though, is to remove yet another variable of spatial patterning that might confound our results, and previous studies have shown that the responses of at least some cell types to integrin ligand-coated PAAm gels is similar to those in 3D and to fibrillar networks51,84. However, differences from the observed behavior may arise due to these key ECM properties85,86,87. For example, glioblastoma cells in 3D agarose gels with collagen I reduce their migration speeds with increasing stiffness of the gel88, but the same cells on polyacrylamide gel with collagen I migrated faster with increasing stiffness of the gel89. Our study’s setup also does not capture the temporal evolution of cell-substrate interactions which are important in patho-physiological and cancer-related processes. In particular, cells can themselves alter ECM structure over time by generating dynamic pulling forces and matrix metalloproteinases90,91, leading to wide-ranging cell-substrate profiles over time. Nevertheless, considering the lack of generalizability in cellular response to mechanical stimuli, future studies focusing on delineating how cells sense mechanical stimuli and how that is translated to phenotypic change should take into consideration the possibility of differing intrinsic characteristics (such as adhesion protein expression and type, molecular control of cytoskeletal assembly92,93, and vesicle trafficking for delivery and retrieval of transmembrane protein complexes94) leading to varied mechanosensitive responses, along with the augmentative role that chemical components of ECM such as HA can play in a cell’s response.
Methods
Cell growth
All cell lines (Table 1) were from American Biological Culture Collection (ATCC; Manassas, VA) and grown according to ATCC standard operating procedures (SOPs), which has been deposited on Figshare (https://doi.org/10.6084/m9.figshare.28916828). Media and reagents used for cell culture were either provided by ATCC or purchased from an outside supplier as per ATCC SOP instructions. Thawed cells were taken to be P0. Cells were passaged until P3 and plated on polyacrylamide or HA gels at single cell density (25,000 cells/gel). Experiments were performed 24 h after cell plating.
Polyacrylamide gel fabrication
Polyacrylamide gels were created by using a combination of acrylamide and bisacrylamide from Bio-Rad Laboratories (Hercules, CA), along with water and NHS (N-hydroxysuccinimide ester) from Sigma (St. Louis, MO) dissolved in toluene (Fisher Scientific, Waltham, MA). Briefly, NHS was dissolved in toluene to create a saturated solution. The solution was then spun for 5 min on a tabletop centrifuge to remove any undissolved NHS. The NHS-toluene solution was then added to the water/polyacrylamide/bisacrylamide mixture and vortexed for 10 s. To make the 30 kPa polyacrylamide gels, a final concentration of 13.5% acrylamide was used, while 500 Pa polyacrylamide gels had a final concentration of 3.5%. Both gels had a bisacrylamide concentration of 0.33%. Acrylamide and bisacrylamide were from BioRad Laboratories (Hercules, CA). After vortexing the solution, we centrifuged at 1000 rpm in a tabletop centrifuge to separate the toluene from the polyacrylamide solution. We then removed the polyacrylamide solution from under the toluene layer, being careful not to disturb the toluene. We then aliquoted out the polyacrylamide. To begin polymerization we then added to a final concentration of 0.3% and 0.06% of Temed and APS (ammonium persulfate solution) respectively. Our solution was immediately pipetted onto functionalized coverslips, and a siliconized coverslip was added on top. Gels were allowed to polymerize for 20 min. Top coverslips were removed, and gels were rinsed twice for 5 min each in PBS. To crosslink protein to the surface of the gel, the gels were incubated in a 0.1 mg/ml solution of collagen (BD Biosciences, San Jose, CA) or fibronectin (EMD Millipore, Billerica, MA) in HEPES pH 8.0 either overnight at 4 °C or at RT for 4 h. Gels were then rinsed and kept in PBS for storage. Note that we used at least 100 μl of 0.1 mg per ml collagen or fibronectin to ensure surface coating density of a minimum of 2200 ng per cm2, which has previously been shown to be a saturating amount of surface ligands that ensure maximal spreading of cells as a function of ligand concentration95.
Hyaluronic acid gel fabrication
HA gels were created by polymerizing thiol-modified hyaluronic acid with ExtraLink (PEGDA) (Ascendance Biotechnology, Alameda, CA) according to the manufacturer’s instructions. A 1 ml vial of lyophilized HA was reconstituted with 875 μl of degassed, deionized water for 30 min at 37 °C. After 30 min, 125 μl of a 1 mg/ml solution of either collagen or fibronectin was added to the vial. Gels were polymerized at a ratio of 1 part ExtraLink to 4 parts HA/protein solution. The desired volume of this mixture was pipetted on a glutaraldehyde-functionalized glass cover slip and then a siliconized coverslip was placed on top. Gels were allowed to polymerize for 20 min before removing the top coverslip and storing in PBS. To be consistent with polyacrylamide gels in terms of the ligand concentration, the amount of integrin ligands was chosen that provided ~2000 ng/cm2 on the surface of the hyaluronic acid gels.
Cell seeding
To seed cells, gels in PBS were sterilized for 30 min under direct UV light. PBS was aspirated, warmed sterile media was added to the gels, and cells were added to single cell density (25,000 cells/18 mm gel). Cells were allowed to adhere to the gels for 24 h before experiments. After 24 h, cells were imaged for morphology, single vs touching analysis, and motility. These analyses all used images taken on a Leica DMIRE2 microscope at 10× magnification. Also at 24 h, cells were measured by AFM for stiffness.
Cell morphology using light microscopy
Three metrics of cell morphology were determined using bright field imaging of live cells at 24 h on each of the 7 substrates. Because of the need to image live cells without genetic or chemical perturbation, fluorescence microscopy cannot be used, and the contrast between cell and substrate, especially for soft gels, is too low to allow automated edge tracing algorithms to work. Therefore, each cell was traced manually from images taken using a 10× lens and digital magnification to provide a pixel to micron ratio of 1.5. This resolution was optimal for allowing sufficient accuracy of tracing the cell contour while enabling simultaneous imaging of 5–10 single cells per field, based on our previous measurements using this method. Cells from two gel replicates per condition were traced, with up to 50 cells per replicate. The public domain NIH-developed software imageJ was used for tracing and to calculate three metrics: adherent area, aspect ratio, and circularity.
Cell stiffness by atomic force microscopy (AFM)
AFM measurements were conducted at room temperature using a Bioscope DAFMLN-AM head (Bruker, Santa Barbara, CA) mounted on an Axiovert 100 microscope (Zeiss, Thornwood, NY). The spring constants of triangular silicon nitride cantilevers with a 1 µm diameter silica bead attached (Novascan, Ames, IA) were determined by resonance measurements (0.04–0.1 N/m). Indentations were made on three distinct areas of the cell avoiding the nucleus and the cell periphery. To quantify cellular stiffness (Young’s modulus), the first 500 nm of indentation into the cell was fit to the Hertz model for a sphere making contact with a homogenous, elastic half space
where f is the force applied to the cell, k is the spring constant of the cantilever, d is the deflection of the cantilever, E is the Young’s modulus, R is the radius of the bead, δ is the indentation into the cell and ν is the Poisson’s ratio of the cell (assumed to be 0.5). 10–15 cells were analyzed at three different points for each of the 7 substrate conditions per cell line.
Motility
Cells were tracked, using the ImageJ cell tracker plugin, in time lapse movies made for periods between 1 and 3 h taken after 24 h of incubation of the substrates. The centroid of each cell was tracked by localization of its nucleus at 5 min intervals. Speed was determined by dividing the total distance a cell moved by the time over which the cell was tracked. To analyze migration persistence, the directional autocorrelation as a function of time26 was calculated using R package celltrackR96. The correlation coefficient is the cosine of the pairwise angle difference (θ) over different time intervals Δt between the normalized displacement vectors describing a migration trajectory. Supplementary Figs. 17–21 show the directional autocorrelation for the cell lines on different substrates, whereby mean and confidence intervals for correlation coefficients at a specific Δt were obtained using all the single cell trajectories of the particular cell line-substrate pair. The decorrelation time was estimated as the time point when cos(θ) becomes less than 0.2. The decorrelation time is taken as the proxy for migratory persistence of a cell line on a particular substrate. For RWPE-1 cell line on 30 kPa Coll, hTERT-HPNE cell line on 500 Pa Coll, 500 Pa FN and HA FN, and Panc-1 cell line on 30 kPa FN, the directional autocorrelation did not fall below 0.2 even at maximum Δt that could be computed from their cell trajectories (Supplementary Figs. 18, 19). In these cases, we used the total trajectory time as the measure of migratory persistence.
Unsupervised clustering
The unsupervised clustering workflow was applied separately to each physical feature (area, aspect ratio, circularity, cell stiffness, and motility). Wasserstein-1 distance29,30 was used as the pairwise distance metric between cell line-substrate combinations and was computed using the R package maotai. The Wasserstein-1 distance measures the discrepancy between two distributions in terms of the area between their respective cumulative density functions (CDFs). Unlike Kullback-Lieibler (KL) divergence, a widely used measure for comparing probability distributions, Wasserstein-1 distance is symmetric and satisfies the triangular inequality, two fundamental properties for a distance to be considered as a metric. As compared to Jansen-Shannon divergence (symmetrized version of KL-divergence), Wasserstein-1 distance can be computed without estimating the probability density functions using KDEs, thus avoiding the need to choose the bandwidth parameter of the smoothing kernel which could have significant effect on the clustering. Another distance metric often used to compare distribution is the Kolomogrov-Smirnov (KS) distance. The KS distance is defined as the largest absolute difference between the two empirical CDFs evaluated at any point and is bounded between 0 and 1. In contrast, Wasserstein-1 distance takes into consideration differences across the complete span of CDFs and is particularly good for distributions containing a significant amount of data in long tails. Thus, KS distance can level off due to its boundedness even if the distributions are considerably apart, whereas Wasserstein-1 distance will increase substantially which would aid in better separation of distinct distributions.
Consensus clustering method27 was used to determine the number and membership of possible classes for each physical feature. This unsupervised clustering method has been extensively used in cancer research to identify molecular subclasses based on omics data97,98,99. Clustering was performed using R package ConsensusClusterPlus28. In our analysis, the input to consensus clustering was the n × n matrix containing pairwise Wasserstein-1 distance between distributions, where n signifies the number of cell line-substrate pairs having at least 25 measurements. Note that a cell line-substrate pair corresponds to a particular cell line on a particular substrate. Consensus clustering involves repeated clustering of subsampled sets to determine the stability of a specified number of cluster count (k). For each of the physical features, we used 0.8 subsampling rate (i.e., 80% of n cell line-substrate pairs selected at random) and 1000 repetitions for cluster count k = 2, 3, …, 9. Partition around medoids (a form of k-medoids clustering method) was used to cluster the subsampled distance matrices. For each value of cluster count k, consensus matrix M is computed where element Mij is called consensus index for (ith cell line-substrate pair, jth cell line-substrate pair) defined as
Then, for each k, a final agglomerative hierarchical clustering using 1−Mij values as distance is used to group cell line-substrate pairs into k consensus clusters. The optimal number of clusters, among the tested cluster counts k = 2, 3, …, 9, was determined using the PAC31 and CHI32. PAC corresponds to the fraction of Mij values between 0.1 and 0.9, which can be viewed as how many (ith cell line-substrate pair, jth cell line-substrate) pairs neither have high enough consensus (Mij > 0.9) to be considered in the same cluster nor low enough consensus (Mij < 0.1) to be considered in different clusters. CHI, on the other hand, compares the between-cluster separation with within-cluster dispersion. Thus, the optimal cluster count corresponds to low PAC and high CHI values. Hierarchical clustering can be used as a visualization tool to view a consensus matrix with high consensus index cell line-substrate pairs grouped together27. We used this to visualize the consensus matrices corresponding to the optimal cluster count (Supplementary Figs. 11−13 ).
Consensus indices can be used to determine item consensus values for each of the cell line-substrate pairs (items). Item consensus values correspond to the mean consensus of an item with all items in a particular cluster and thus, each cell line-substrate pair would have k item-consensus values corresponding to each cluster at a particular k. The cluster for which a cell line-substrate pair has the highest item-consensus value is considered its membership class. To further account for the inherent uncertainty in unsupervised clustering, after determining the optimal cluster count k, we identified the cell line-substrate pairs that do not show strong membership to its assigned class. That is, we considered the cell line-substrate pairs that have the highest item-consensus value of less than 0.8 as boundary cases. These boundary cases were instead viewed as being members of either of the two classes for which they have two largest item-consensus values (both less than 0.8) based on the available data. The same framework described here for Wasserstein-1 distance based clustering was used for KS distance based clustering. The heatmap visualizations for cluster membership of each cell line-substrate pair were created using ComplexHeatmap100,101.
Statistical analysis
A two-sided permutation test was used to assess statistical significance of the ratio of median values of a particular physical feature on two different substrate conditions being greater than 1, and of the difference between cancer cell lines and their normal counterpart(s) within the same tissue type in terms of the median value of the physical features. The Benjamini-Hochberg procedure102 was used to adjust the p-values for multiple testing.
Data availability
All the measurement data for morphology (area, aspect ratio, and circularity), cell stiffness, and migration trajectories used in this study is available in the Figshare repository, https://doi.org/10.6084/m9.figshare.28916828.
Code availability
The underlying code for this study is available in Github and can be accessed via this link https://github.com/kparihar13/pson_cell_line_mechanobiology_analysis. All analyses reported in this study used the statistical software R (v.4.4.2).
References
Discher, D. E. et al. Matrix mechanosensing: from scaling concepts in ’omics data to mechanisms in the nucleus, regeneration, and cancer. Annu. Rev. Biophys. 46, 295–315 (2017).
Levental, I., Georges, P. C. & Janmey, P. A. Soft biological materials and their impact on cell function. Soft Matter 3, 299–306 (2007).
Perepelyuk, M. et al. Normal and fibrotic rat livers demonstrate shear strain softening and compression stiffening: a model for soft tissue mechanics. PLoS ONE 11, e0146588 (2016).
DuFort, C. C., Paszek, M. J. & Weaver, V. M. Balancing forces: architectural control of mechanotransduction. Nat. Rev. Mol. Cell Biol. 12, 308–319 (2011).
Discher, D. E., Janmey, P. & Wang, Y. Tissue cells feel and respond to the stiffness of their substrate. Science 310, 1139–1143 (2005).
Mammoto, T., Mammoto, A. & Ingber, D. E. Mechanobiology and developmental control. Annu. Rev. Cell Dev. Biol. 29, 27–61 (2013).
Mohammadi, H. & Sahai, E. Mechanisms and impact of altered tumour mechanics. Nat. Cell Biol. 20, 766–774 (2018).
Winkler, J., Abisoye-Ogunniyan, A., Metcalf, K. J. & Werb, Z. Concepts of extracellular matrix remodelling in tumour progression and metastasis. Nat. Commun. 11, 1–19 (2020).
Oudin, M. J. & Weaver, V. M. Physical and chemical gradients in the tumor microenvironment regulate tumor cell invasion, migration, and metastasis. Cold Spring Harb. Symp. Quant. Biol. 81, 189–205 (2016).
Gao, Y. et al. Metastasis organotropism: redefining the congenial soil. Dev. Cell 49, 375–391 (2019).
Chin, L., Xia, Y., Discher, D. E. & Janmey, P. A. Mechanotransduction in cancer. Curr. Opin. Chem. Eng. 11, 77–84 (2016).
Janmey, P. A., Fletcher, D. A. & Reinhart-King, C. A. Stiffness sensing by cells. Physiol. Rev. 100, 695–724 (2020).
Engler, A. J., Sen, S., Sweeney, H. L. & Discher, D. E. Matrix elasticity directs stem cell lineage specification. Cell 126, 677–689 (2006).
Poh, Y.-C., Chowdhury, F., Tanaka, T. S. & Wang, N. Embryonic stem cells do not stiffen on rigid substrates. Biophys. J. 99, L19–L21 (2010).
Shin, S. I., Freedman, V. H., Risser, R. & Pollack, R. Tumorigenicity of virus-transformed cells in nude mice is correlated specifically with anchorage independent growth in vitro. Proc. Natl. Acad. Sci. USA. 72, 4435–4439 (1975).
Klein, E. A. et al. Cell-cycle control by physiological matrix elasticity and in vivo tissue stiffening. Curr. Biol. 19, 1511–1518 (2009).
Wang, H.-B., Dembo, M. & Wang, Y.-L. Substrate flexibility regulates growth and apoptosis of normal but not transformed cells. Am. J. Physiol.-Cell Physiol. 279, C1345–C1350 (2000).
Plunkett, C. et al. H-Ras transformation of mammary epithelial cells induces ERK-mediated spreading on low stiffness matrix. Adv. Healthc. Mater. 9, e1901366 (2020).
Lee, S., Hong, J. & Lee, J. Cell motility regulation on a stepped micro pillar array device (SMPAD) with a discrete stiffness gradient. Soft Matter 12, 2325–2333 (2016).
Wei, Q. et al. Mechanotargeting: mechanics-dependent cellular uptake of nanoparticles. Adv. Mater. 30, 1707464 (2018).
Butler, J. P., Tolić-Nørrelykke, I. M., Fabry, B. & Fredberg, J. J. Traction fields, moments, and strain energy that cells exert on their surroundings. Am. J. Physiol. Cell Physiol. 282, C595–C605 (2002).
Kuznetsova, T. G., Starodubtseva, M. N., Yegorenkov, N. I., Chizhik, S. A. & Zhdanov, R. I. Atomic force microscopy probing of cell elasticity. Micron Oxf. Engl. 1993 38, 824–833 (2007).
Thomas, G., Burnham, N. A., Camesano, T. A. & Wen, Q. Measuring the mechanical properties of living cells using atomic force microscopy. J. Vis. Exp. 50497 https://doi.org/10.3791/50497 (2013).
Molter, C. W. et al. Prostate cancer cells of increasing metastatic potential exhibit diverse contractile forces, cell stiffness, and motility in a microenvironment stiffness-dependent manner. Front. Cell Dev. Biol. 10, 932510 (2022).
Wu, X., Gong, S., Roy-Burman, P., Lee, P. & Culig, Z. Current mouse and cell models in prostate cancer research. Endocr. Relat. Cancer 20, R155–R170 (2013).
Gorelik, R. & Gautreau, A. Quantitative and unbiased analysis of directional persistence in cell migration. Nat. Protoc. 9, 1931–1943 (2014).
Monti, S., Tamayo, P., Mesirov, J. & Golub, T. Consensus clustering: a resampling-based method for class discovery and visualization of gene expression microarray data. Mach. Learn. 52, 91–118 (2003).
Wilkerson, M. D. & Hayes, D. N. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics 26, 1572–1573 (2010).
Panaretos, V. M. & Zemel, Y. Statistical aspects of Wasserstein distances. Annu. Rev. Stat. Its Appl. 6, 405–431 (2019).
Peyré, G. & Cuturi, M. Computational optimal transport: with applications to data science. Found. Trends® Mach. Learn. 11, 355–607 (2019).
Șenbabaoğlu, Y., Michailidis, G. & Li, J. Z. Critical limitations of consensus clustering in class discovery. Sci. Rep. 4, 6207 (2014).
Caliński, T. & Harabasz, J. A dendrite method for cluster analysis. Commun. Stat. 3, 1–27 (1974).
Wong, S. W. et al. Rapamycin synergizes cisplatin sensitivity in basal-like breast cancer cells through up-regulation of p73. Breast Cancer Res. Treat. 128, 301–p313 (2011).
Privat, M. et al. A high expression ratio of RhoA/RhoB is associated with the migratory and invasive properties of basal-like Breast Tumors. Int. J. Med. Sci. 17, 2799–2808 (2020).
Pelham, R. J. & Wang, Y. Cell locomotion and focal adhesions are regulated by substrate flexibility. Proc. Natl. Acad. Sci. USA 94, 13661–13665 (1997).
Yeung, T. et al. Effects of substrate stiffness on cell morphology, cytoskeletal structure, and adhesion. Cell Motil. 60, 24–34 (2005).
Califano, J. P. & Reinhart-King, C. A. Substrate stiffness and cell area predict cellular traction stresses in single cells and cells in contact. Cell. Mol. Bioeng. 3, 68–75 (2010).
Andresen Eguiluz, R. C., Kaylan, K. B., Underhill, G. H. & Leckband, D. E. Substrate stiffness and VE-cadherin mechano-transduction coordinate to regulate endothelial monolayer integrity. Biomaterials 140, 45–57 (2017).
Bravo, B. et al. Opposite effects of mechanical action of fluid flow on proangiogenic factor secretion from human adipose-derived stem cells with and without oxidative stress. J. Cell. Physiol. 232, 2158–2167 (2017).
Ganesan, M. et al. Matrix stiffness regulate apoptotic cell death in HIV-HCV co-infected hepatocytes: importance for liver fibrosis progression. Biochem. Biophys. Res. Commun. 500, 717–722 (2018).
Georges, P. C. et al. Increased stiffness of the rat liver precedes matrix deposition: implications for fibrosis. Am. J. Physiol. - Gastrointest. Liver Physiol. 293, 1147–1154 (2007).
Li, Z. et al. Transforming growth factor-β and substrate stiffness regulate portal fibroblast activation in culture. Hepatology 46, 1246–1256 (2007).
Olsen, A. L. et al. Hepatic stellate cells require a stiff environment for myofibroblastic differentiation. Am. J. Physiol. Gastrointest. Liver Physiol. 301, G110–G118 (2011).
Chopra, A., Tabdanov, E., Patel, H., Janmey, P. A. & Kresh, J. Y. Cardiac myocyte remodeling mediated by N-cadherin-dependent mechanosensing. Am. J. Physiol. Heart Circ. Physiol. 300, H1252–H1266 (2011).
Brown, X. Q., Ookawa, K. & Wong, J. Y. Evaluation of polydimethylsiloxane scaffolds with physiologically-relevant elastic moduli: interplay of substrate mechanics and surface chemistry effects on vascular smooth muscle cell response. Biomaterials 26, 3123–3129 (2005).
Dasbiswas, K., Majkut, S., Discher, D. E. & Safran, S. A. Substrate stiffness-modulated registry phase correlations in cardiomyocytes map structural order to coherent beating. Nat. Commun. 6, 6085 (2015).
Peyton, S. R., Raub, C. B., Keschrumrus, V. P. & Putnam, A. J. The use of poly(ethylene glycol) hydrogels to investigate the impact of ECM chemistry and mechanics on smooth muscle cells. Biomaterials 27, 4881–4893 (2006).
Bhana, B. et al. Influence of substrate stiffness on the phenotype of heart cells. Biotechnol. Bioeng. 105, 1148–1160 (2010).
Lin, H.-H. et al. Mechanical phenotype of cancer cells: cell softening and loss of stiffness sensing. Oncotarget 6, 20946–20958 (2015).
Wullkopf, L. et al. Cancer cells’ ability to mechanically adjust to extracellular matrix stiffness correlates with their invasive potential. Mol. Biol. Cell 29, 2378–2385 (2018).
Georges, P. C., Miller, W. J., Meaney, D. F., Sawyer, E. S. & Janmey, P. A. Matrices with compliance comparable to that of brain tissue select neuronal over glial growth in mixed cortical cultures. Biophys. J. 90, 3012–3018 (2006).
Koser, D. E. et al. Mechanosensing is critical for axon growth in the developing brain. Nat. Neurosci. 19, 1592–1598 (2016).
Engler, A. J., Richert, L., Wong, J. Y., Picart, C. & Discher, D. E. Surface probe measurements of the elasticity of sectioned tissue, thin gels and polyelectrolyte multilayer films: correlations between substrate stiffness and cell adhesion. Surf. Sci. 570, 142–154 (2004).
Pogoda, K. et al. Soft substrates containing hyaluronan mimic the effects of increased stiffness on morphology, motility, and proliferation of glioma cells. Biomacromolecules 18, 3040–3051 (2017).
Reimer, M., Petrova Zustiak, S., Sheth, S. & Martin Schober, J. Intrinsic response towards physiologic stiffness is cell-type dependent. Cell Biochem. Biophys. 76, 197–208 (2018).
Byfield, F. J. et al. Absence of filamin A prevents cells from responding to stiffness gradients on gels coated with collagen but not fibronectin. Biophys. J. 96, 5095–5102 (2009).
Kasza, K. E. et al. Filamin A is essential for active cell stiffening but not passive stiffening under external force. Biophys. J. 96, 4326–4335 (2009).
Solon, J., Levental, I., Sengupta, K., Georges, P. C. & Janmey, P. A. Fibroblast adaptation and stiffness matching to soft elastic substrates. Biophys. J. 93, 4453–4461 (2007).
Peyton, S. R. & Putnam, A. J. Extracellular matrix rigidity governs smooth muscle cell motility in a biphasic fashion. J. Cell. Physiol. 204, 198–209 (2005).
Stroka, K. M. & Aranda-Espinoza, H. Endothelial cell substrate stiffness influences neutrophil transmigration via myosin light chain kinase-dependent cell contraction. Blood 118, 1632–1640 (2011).
Shebanova, O. & Hammer, D. A. Biochemical and mechanical extracellular matrix properties dictate mammary epithelial cell motility and assembly. Biotechnol. J. 7, 397–408 (2012).
Tilghman, R. W. et al. Matrix rigidity regulates cancer cell growth and cellular phenotype. PloS ONE 5, e12905 (2010).
Tang, X. et al. Mechanical force affects expression of an in vitro metastasis-like phenotype in HCT-8 cells. Biophys. J. 99, 2460–2469 (2010).
Munevar, S., Wang, Y. & Dembo, M. Traction force microscopy of migrating normal and H-ras transformed 3T3 fibroblasts. Biophys. J. 80, 1744–1757 (2001).
Rooney, P. & Kumar, S. Inverse relationship between hyaluronan and collagens in development and angiogenesis. Differentiation 54, 1–9 (1993).
Ananthanarayanan, B., Kim, Y. & Kumar, S. Elucidating the mechanobiology of malignant brain tumors using a brain matrix-mimetic hyaluronic acid hydrogel platform. Biomaterials 32, 7913–7923 (2011).
Chanmee, T., Ontong, P. & Itano, N. Hyaluronan: cancer and cancer metastasis. in Glycoscience: Biology and Medicine (eds. Taniguchi, N., Endo, T., Hart, G. W., Seeberger, P. H. & Wong, C.-H.) 1411–1417. https://doi.org/10.1007/978-4-431-54841-6_191 (Springer Japan, 2015).
Sato, N., Kohi, S., Hirata, K. & Goggins, M. Role of hyaluronan in pancreatic cancer biology and therapy: once again in the spotlight. Cancer Sci. 107, 569–575 (2016).
Chopra, A. et al. Reprogramming cardiomyocyte mechanosensing by crosstalk between integrins and hyaluronic acid receptors. J. Biomech. 45, 824–831 (2012).
Chopra, A. et al. Augmentation of integrin-mediated mechanotransduction by hyaluronic acid. Biomaterials 35, 71–82 (2014).
Mandal, K. et al. Soft hyaluronic gels promote cell spreading, stress fibers, focal adhesion, and membrane tension by phosphoinositide signaling, not traction force. ACS Nano 13, 203–214 (2019).
Turley, E. A., Noble, P. W. & Bourguignon, L. Y. W. Signaling properties of hyaluronan receptors*. J. Biol. Chem. 277, 4589–4592 (2002).
Gares, S. L. & Pilarski, L. M. Balancing thymocyte adhesion and motility: a functional linkage between β1 Lntegrins and The Motility Receptor RHAMM. J. Immunol. Res. 7, 094616 (2000).
Ramalingam, N. et al. Phospholipids regulate localization and activity of mDia1 formin. Eur. J. Cell Biol. 89, 723–732 (2010).
Kozlov, M. M. & Bershadsky, A. D. Processive capping by formin suggests a force-driven mechanism of actin polymerization. J. Cell Biol. 167, 1011–1017 (2004).
Kawamura, K. et al. N-WASP and WAVE2 acting downstream of phosphatidylinositol 3-kinase are required for myogenic cell migration induced by hepatocyte growth factor. J. Biol. Chem. 279, 54862–54871 (2004).
Fukami, K., Endo, T., Imamura, M. & Takenawa, T. alpha-Actinin and vinculin are PIP2-binding proteins involved in signaling by tyrosine kinase. J. Biol. Chem. 269, 1518–1522 (1994).
Rubashkin, M. G. et al. Force engages vinculin and promotes tumor progression by enhancing PI3K activation of phosphatidylinositol (3,4,5)-triphosphate. Cancer Res. 74, 4597–4611 (2014).
Yuan, S., Norgard, R. J. & Stanger, B. Z. Cellular plasticity in cancer. Cancer Discov. 9, 837–851 (2019).
Pérez-González, A., Bévant, K. & Blanpain, C. Cancer cell plasticity during tumor progression, metastasis and response to therapy. Nat. Cancer 4, 1063–1082 (2023).
Leibovitz, A. et al. Classification of human colorectal adenocarcinoma cell lines. Cancer Res. 36, 4562–4569 (1976).
Tsujita, K. et al. Homeostatic membrane tension constrains cancer cell dissemination by counteracting BAR protein assembly. Nat. Commun. 12, 5930 (2021).
Liu, Y. et al. Cell softness prevents cytolytic T-cell killing of tumor-repopulating cells. Cancer Res. 81, 476–488 (2021).
Tran, K. A. et al. Matching mechanical heterogeneity of the native spinal cord augments axon infiltration in 3D-printed scaffolds. Biomaterials 295, 122061 (2023).
Chaudhuri, O., Cooper-White, J., Janmey, P. A., Mooney, D. J. & Shenoy, V. B. Effects of extracellular matrix viscoelasticity on cellular behaviour. Nature 584, 535–546 (2020).
Tran, K. A. et al. Dynamic tuning of viscoelastic hydrogels with carbonyl iron microparticles reveals the rapid response of cells to three-dimensional substrate mechanics. ACS Appl. Mater. Interfaces 13, 20947–20959 (2021).
Saraswathibhatla, A., Indana, D. & Chaudhuri, O. Cell–extracellular matrix mechanotransduction in 3D. Nat. Rev. Mol. Cell Biol. 1–22 https://doi.org/10.1038/s41580-023-00583-1 (2023).
Ulrich, T. A., Jain, A., Tanner, K., MacKay, J. L. & Kumar, S. Probing cellular mechanobiology in three-dimensional culture with collagen-agarose matrices. Biomaterials 31, 1875–1884 (2010).
Ulrich, T. A., de Juan Pardo, E. M. & Kumar, S. The mechanical rigidity of the extracellular matrix regulates the structure, motility, and proliferation of glioma cells. Cancer Res. 69, 4167–4174 (2009).
Mak, M. Impact of crosslink heterogeneity on extracellular matrix mechanics and remodeling. Comput. Struct. Biotechnol. J. 18, 3969–3976 (2020).
Xiao, H. et al. Proteolysis and contractility regulate tissue opening and wound healing by lung fibroblasts in 3D microenvironments. Adv. Healthc. Mater. 13, 2400941 (2024).
Alisafaei, F. et al. Vimentin is a key regulator of cell mechanosensing through opposite actions on actomyosin and microtubule networks. Commun. Biol. 7, 1–13 (2024).
Patteson, A. E., Carroll, R. J., Iwamoto, D. V. & Janmey, P. A. The vimentin cytoskeleton: when polymer physics meets cell biology. Phys. Biol. 18, 011001 (2020).
Grant, B. D. & Donaldson, J. G. Pathways and mechanisms of endocytic recycling. Nat. Rev. Mol. Cell Biol. 2009 109 10, 597–608 (2009).
Engler, A. et al. Substrate compliance versus ligand density in cell on gel responses. Biophys. J. 86, 617–628 (2004).
Wortel, I. M. N. et al. CelltrackR: an R package for fast and flexible analysis of immune cell migration data. ImmunoInformatics 1–2, 100003 (2021).
Verhaak, R. G. W. et al. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell 17, 98–110 (2010).
Hoadley, K. A. et al. Cell-of-origin patterns dominate the molecular classification of 10,000 tumors from 33 types of cancer. Cell 173, 291–304.e6 (2018).
Robertson, A. G. et al. Comprehensive molecular characterization of muscle-invasive bladder cancer. Cell 171, 540–556.e25 (2017).
Gu, Z., Eils, R. & Schlesner, M. Complex heatmaps reveal patterns and correlations in multidimensional genomic data. Bioinformatics 32, 2847–2849 (2016).
Gu, Z. Complex heatmap visualization. iMeta 1, e43 (2022).
Benjamini, Y. & Hochberg, Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J. R. Stat. Soc. Ser. B Methodol. 57, 289–300 (1995).
Acknowledgements
The data was obtained with support from the National Cancer Institute and Leidos Biomedical Research, Inc. under contract 15X008 with Frederick National Laboratory for Cancer Research. This study has also received funding from the National Institutes of Health under R35GM136259, EB01775309, and U01CA250044.
Author information
Authors and Affiliations
Contributions
PAJ designed the experimental mechanobiology research. K.P., J.N., P.A.J., and R.R. designed the computational analysis research. K.C. managed the experimental research project. K.C., L.C., M.E.M., M.G.M., A.S.v.O., A.H., E.E.C., P.A.G., M.D., and T.L. performed cell culture and imaging experiments. F.J.B. performed atomic force microscopy measurements. D.V.I. performed experiments and assisted in data interpretation. K.P., J.N., P.A.J., and R.R. designed the data analysis and machine learning research. K.P. performed statistical analysis and machine learning. K.P., P.A.J., and R.R. wrote the article.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Parihar, K., Nukpezah, J., Iwamoto, D.V. et al. Tissue-dependent mechanosensing by cells derived from human tumors. npj Biol. Phys. Mech. 2, 19 (2025). https://doi.org/10.1038/s44341-025-00023-5
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s44341-025-00023-5
