Collective cell durotaxis emerges from long-range intercellular force transmission

The ability of cells to follow gradients of extracellular matrix stiffness—durotaxis—has been implicated in development, fibrosis, and cancer. Here, we found multicellular clusters that exhibited durotaxis even if isolated constituent cells did not. This emergent mode of directed collective cell migration applied to a variety of epithelial cell types, required the action of myosin motors, and originated from supracellular transmission of contractile physical forces. To explain the observed phenomenology, we developed a generalized clutch model in which local stick-slip dynamics of cell-matrix adhesions was integrated to the tissue level through cell-cell junctions. Collective durotaxis is far more efficient than single-cell durotaxis; it thus emerges as a robust mechanism to direct cell migration during development, wound healing, and collective cancer cell invasion.

The ability of cells to follow gradients of extracellular matrix stiffness-durotaxis-has been implicated in development, fibrosis, and cancer. Here, we found multicellular clusters that exhibited durotaxis even if isolated constituent cells did not. This emergent mode of directed collective cell migration applied to a variety of epithelial cell types, required the action of myosin motors, and originated from supracellular transmission of contractile physical forces. To explain the observed phenomenology, we developed a generalized clutch model in which local stick-slip dynamics of cell-matrix adhesions was integrated to the tissue level through cell-cell junctions. Collective durotaxis is far more efficient than single-cell durotaxis; it thus emerges as a robust mechanism to direct cell migration during development, wound healing, and collective cancer cell invasion.
T he ability of living cells to migrate following environmental gradients underlies a broad range of phenomena in development, homeostasis, and disease (1,2). The bestunderstood mode of directed cell migration is chemotaxis, the well-established ability of cells to follow gradients of soluble chemical cues (1). Some cell types are also able to follow gradients in the stiffness of their extracellular matrix (ECM), a process known as durotaxis (3)(4)(5)(6)(7)(8)(9)(10). Durotaxis has been implicated in development (11), fibrosis (12), and cancer (13), but its underlying mechanisms remain unclear.
Most of our understanding of directed cell migration has been obtained in single isolated cells. However, fundamental processes during development, wound healing, tissue regeneration, and some forms of cancer cell invasion are driven by directed migration of cell groups (14)(15)(16). Cell-cell interactions within these groups provide cooperative mechanisms of cell guidance that are altogether inaccessible to single cells (14)(15)(16)(17)(18)(19)(20). Here, we investigated whether cell groups undergo collective durotaxis and the cooperative nature of underlying mechanisms.
Asymmetric expansion is not attributable to cell proliferation, because it was unaffected by inhibition of cell division and because the number of divisions in the cluster was independent of substrate stiffness ( fig. S4). Importantly, it is not attributable to additive contribution of singlecell durotaxis either, because single MCF-10A cells did not durotax in isolation (Fig. 2, A to C and E, and movie S2). Taken together, these data establish an unanticipated mode of collective durotaxis driven by an emergent property of the cell cluster.
Even if single MCF-10A cells did not durotax in isolation, they exhibited faster randomly oriented velocity on stiffer gels (Fig. 2D) (13). In a closepacked system, this feature could explain collective durotaxis because volume exclusion would force cells to move persistently away from the cluster at a higher speed on the stiffer edge. To test the contribution of this mechanism, we perturbed cell-cell junctions independently of close packing by knocking down a-catenin using small interfering RNA (siRNA). Like control cells, SCIENCE sciencemag.org 9    isolated cells depleted of a-catenin migrated faster on stiffer regions but did not durotax ( fig. S5, A to D, and movie S3). Unlike control clusters, however, close-packed clusters depleted of acatenin expanded nearly symmetrically (Fig. 2, F to I; fig. S5, E to M; and movie S4), thereby indicating that cell-cell adhesions are required for collective cell durotaxis. These findings rule out mechanisms solely based on local stiffness sensing and point to a long-range mechanism involving cell-cell adhesion.
We next studied whether this long-range mechanism involves transmission of physical forces across the cluster. To this end, we implemented a generalized traction force microscopy algorithm to map forces exerted by cells on substrates with arbitrary stiffness profiles (supplementary text 1). Clusters migrating on uniform gels (Fig. 3, A and C, and movie S5) and on gradient gels (Fig.  3, B and D, and movie S6) exhibited similar traction force maps. The highest tractions were localized at the edges and pointed toward the midline of the cluster, whereas relatively lower tractions in the bulk showed no particular orientation. To average out fluctuations and retain systematic traction patterns, we computed x-t kymographs of traction component T x (Fig. 3, I to J). Kymographs on uniform and gradient gels revealed two traction layers of similar magnitude and opposite sign at both edges of the clusters and negligible average tractions in the bulk. Interestingly, soft edges had cell-substrate forces similar to those of stiff edges but smaller and denser cell-matrix adhesions, consistent with previous findings that force levels are not necessarily linked to a specific adhesion size ( fig. S6, A and  B) (23). Unlike traction forces, substrate displacements on gradient gels were nearly one order of magnitude higher on the soft edge than on the stiff one (Fig. 3, E, F, K, and L, and movies S5 and S6).
To compute force transmission within the monolayer, we used monolayer stress microscopy and focused on the normal component of the stress tensor in the direction of expansion s xx (24,25), which we hereafter refer to as intercellular tension (Fig. 3, G and H, and movies S5 and S6). Kymographs showed that intercellular tension increased up to a plateau within the first few cells at the monolayer edges and remained roughly constant in the monolayer bulk thereafter (Fig. 3, M and N).
Our force measurements establish that the monolayer expands by generating contractile traction forces of equal magnitude at both edges and that these forces are transmitted across the cluster. To explore how this physical scenario might lead to collective durotaxis, we developed a model that integrates clutchlike cell-ECM dynamics at focal adhesions (23,(26)(27)(28), long-range force transmission through cell-cell junctions, and actin polymerization at monolayer edges (Fig. 4, A and B, and supplementary text 2). For a monolayer attached to a substrate of uniform stiffness, the model predicts symmetric expansion; actin polymerization exceeds acto-myosin contraction to the same extent on both edges. By contrast, on stiffness gradients, the substrate deforms and opposes polymerization more on the soft edge, tilting monolayer expansion toward the stiff one (Fig. 4B).
We next used the model to produce testable predictions. First, the model predicts that durotaxis should decrease upon reducing the difference in substrate deformation between the soft and stiff side (Fig. 4C). We confirmed this prediction by reducing the slope of the stiffness gradient (Fig. 4D), seeding clusters on progressively stiffer regions of the gradient gels (Fig. 4D), reducing cluster size ( fig. S7), inhibiting myosin contractility with blebbistatin (Fig. 4, E to H; fig. S8; and movie S7), and reducing cell adhesion by decreasing fibronectin coating ( fig. S9). Notably, clusters placed on the softest region of the gradient, where the model predicts highest durotaxis, exhibited directed migration toward the stiff edge as a single unit rather than asymmetric expansion ( fig. S10 and movie S8). Second, impairing long-range force transmission should abrogate durotaxis. This prediction was confirmed by knocking down a-catenin (Fig. 2, F to I; fig.  S11; and movie S9) and by laser-ablating clusters in the direction parallel to the midline (Fig. 4, I to L; fig. S12; and movie S10), which drastically reduced both intercellular force transmission and durotaxis. Third, the clutch mechanism implies that actin retrograde flow should be faster on the soft edge and that the differences in the speeds of actin flow and monolayer expansion between edges should be of the same order. This was indeed the case ( fig. S6, C and D). Finally, the model predicts that in response to an extremely steep gradient, even single cells should generate sufficiently large differences in substrate deformation between their leading and trailing edges to enable durotaxis (10). As predicted, single MCF-10A cells exhibited weak but significant durotaxis in response to a step gradient (433.6 ± 43.6 kPa/mm) ( fig. S13). Given the close agreement between model predictions and experiments, we conclude that local stiffness sensing at the cluster edges and long-range force transmission through intercellular junctions are sufficient to explain the phenomenology of collective cell durotaxis.
Recent experimental and theoretical research has emphasized that some collective systems are more effective at responding to environmental gradients than their isolated constituents (19,20,29). This emergent phenomenon, often alluded to as "collective intelligence," has been observed in cell clusters during chemotaxis (19,20), fish schools during phototaxis (29), and human groups during online gaming (30). In the context of these phenomena, collective durotaxis is unique in that the very same machinery that senses the attractanti.e., the actomyosin cytoskeleton-is responsible for propulsion toward it. As such, collective durotaxis might be the most rudimentary, and perhaps most primitive, mechanism by which a collective living system responds to a gradient. Rudimentary or not, collective durotaxis is robust, is general, and dramatically boosts single-cell responses, providing a new mechanism to organize directed cell migration during development, wound healing, and collective cancer cell invasion.  fig. S12 and movie S10.
Tumor genetics guides patient selection for many new therapies, and cell culture studies have demonstrated that specific mutations can promote metabolic phenotypes. However, whether tissue context defines cancer dependence on specific metabolic pathways is unknown. Kras activation and Trp53 deletion in the pancreas or the lung result in pancreatic ductal adenocarinoma (PDAC) or non-small cell lung carcinoma (NSCLC), respectively, but despite the same initiating events, these tumors use branched-chain amino acids (BCAAs) differently. NSCLC tumors incorporate free BCAAs into tissue protein and use BCAAs as a nitrogen source, whereas PDAC tumors have decreased BCAA uptake. These differences are reflected in expression levels of BCAA catabolic enzymes in both mice and humans. Loss of Bcat1 and Bcat2, the enzymes responsible for BCAA use, impairs NSCLC tumor formation, but these enzymes are not required for PDAC tumor formation, arguing that tissue of origin is an important determinant of how cancers satisfy their metabolic requirements.
T he development of new cancer therapeutics relies on underlying genetic features to identify sensitive patients (1). Mutations in both KRAS and TP53 are common genetic events found in tumors arising from many tissues, and cancers with these mutations are often difficult to treat (2,3). These genetic events, as well as others associated with cancer, contribute to the metabolic changes that support biomass accumulation and cancer cell proliferation (4). Oncogenic RAS signaling increases glucose and glutamine consumption to support anabolic processes including nucleotide, lipid, and nonessential amino acid biosynthesis and can also drive extracellular protein and lipid scavenging (5). TP53 mutations increase glucose consumption and glycolytic flux, whereas inactivation of TP53 renders cancer cells more dependent on serine uptake and metabolism (6).
KRAS and TP53 mutations are found in most human pancreatic tumors (7) and are also common in lung adenocarcinoma (8). What is known of how mutant KRAS or disruption of TP53 affects cancer metabolism is based on cell culture studies in defined medium, although in vivo nutrient availability varies widely between tissues, and vasculature changes can limit nutrient access within tumors (9,10). The inability to model these differences in culture has therefore limited understanding of how tissue of origin influences tumor metabolism (11). Furthermore, environment can influence metabolic phenotypes in vitro (12)(13)(14), and metabolic dependencies in vivo can differ from those found in vitro (15). Metabolic differences between tumor types may also result from cell-autonomous effects, and tumor metabolic gene expression more closely resembles that of its tissue of origin than that of other tumors (16). The same oncogenic driver can also cause different metabolic phenotypes in lung and liver tumors (17). This raises the possibility that tumor type is a major determinant of some tumor metabolic dependencies in vivo.
Elevated plasma branched-chain amino acid (BCAA) levels are found in early pancreatic ductal adenocarinoma (PDAC) but not in non-small cell lung carcinoma (NSCLC), even when the tumors are initiated by the same genetic events (18). To confirm that tumor tissue of origin influences whole-body BCAA metabolism, we used LSL-Kras G12D/+ ;Trp53 flox/flox (KP) mice. We crossed KP mice to mice harboring a Cre-recombinase allele driven by a Pdx-1 promoter (KP -/-C model) (19) or delivered viral Cre to the lungs of these mice (20) in order to generate models of PDAC and NSCLC, respectively. Consistent with prior reports (18), mice with early PDAC have increased levels of plasma BCAAs, whereas mice with early NSCLC exhibit decreased plasma BCAA levels (fig. S1, A to D). When cells derived from these tumors