Extracellular matrix as a driver for intratumoral heterogeneity

All cells within an organ cooperate to constitute its organotypic ECM. Therefore, ECMs of different organs differ in their chemical composition and physical properties. Stromal cell-synthesized ECM is predominantly fibrillar with a microporous architecture and predominantly constituted of collagens [11]. On the other hand, glandular epithelial cells secrete ECM that is rich in laminins, non-fibrillar collagen IV and sulfated proteoglycans that constitute a sheet-like nanoporous superstructure, known as the basement membrane (BM) (figure 1(A)) [12]. Early biochemical studies conducted by Vaheri and co-workers have shown that tumor cells secrete characteristic ECM mixtures: irrespective of their tissue of origin, mesenchymal-like cancer lines tend to secrete collagen-rich ECM, similar to untransformed stromal cells, whereas epithelioid cancer cells secrete laminin-rich ECM [13]. In fact, early pathological studies indicate that cancers can broadly be identified based on the ECM associated with them. For example, neural tumors and sarcomas in pediatric patients can be distinguished from each other by their secretion of laminin-collagen IV-fibronectin pattern and collagen (except collagen V) pattern, respectively [14].

With the advent of sophisticated omic technologies, ECM and associated gene products have been actively included within transcript signatures reported to be cognate to specific tumors (figure 1(B)). For example, Yuzhalin and co-workers propose that a set of ECM proteins including collagens like COL1A1, COL10A1, and COL11A1, and proteoglycans such as aggrecan and biglycan are elevated across alimentary cancers as well as breast and lung cancers [15]. Signatures based on transcription of ECM genes have also been proposed to prognose the fate of non-small cell lung cancers [16]. Notably, tumor ECM is synthesized not just by tumor epithelia but also by associated stromal cells such as fibroblasts and macrophages [17]. Such untransformed stromal cells are recruited to the tumor through morphogenetic cues such as epidermal growth factor and transforming growth factor-β. Furthermore, analyses carried out on triple-negative breast cancer xenografts and isogenic invasion-diverse cancer lines show distinct ECM mixtures secreted by stromal and cancer cells within tumors formed by them [18].

The deposition of a dense fibrillar ECM around a growing tumor, known as a desmoplastic reaction, is one of the notable characteristics of cancers such as pancreatic ductal adenocarcinoma and invasive ductal carcinoma of breast, and has been used as a histopathological cue to prognose the disease long before cancer-related changes in ECM composition, chemistry and rheology were understood [19]. Desmoplastic changes are the result of collagen deposition and signify a more aggressive progression [20]. Collagen patterning and fibrillogenesis are also sensitive to other ECM constituents: sulphated proteoglycans such as decorin and biglycan are known to bind to collagen using their protein cores, whereas their hydrophilic glycosaminoglycan (GAG) chains alter the kinetics and geometry of fibrillogenesis [21, 22]. Chondroitin and dermatan sulfate polymers may also show self-association, which provides insights into how different levels of such GAGs could alter their miscibility with polymerizing collagen fibrils, leading to heterogeneously patterned and constituted ECM across spatial scales [23].

Cancer cells are intimately involved in the enzymatic degradation of protein and glycan constituents of the ECM [11]. ECM degradation creates spaces for cells to migrate; in addition, the degraded components of ECM also may potentiate invasion (figure 1(B)). Although matrix metalloproteinases (MMPs), represent the most well studied family of secreted ECM proteases, the ability to degrade matrix extends to other family of proteases as well. For example, cysteine proteases such as cathepsins are involved in the degradation of BM proteins like nidogen, collagen IV and tenascin as well as stromal proteins such as fibronectin [24]. Serine and aspartate proteases not only degrade ECM, some of its members are also involved in activating MMPs [25]. Proteases are secreted both by cancer cells as well as cancer-associated stromal fibroblasts. Their effect on cancer progression is highly context-specific: whereas proteases have been shown to spur progression, there may be distinct ways through which proteases may also impede invasion and metastasis [26]. In addition to proteins, the degradation of glycan components of ECM is mediated by glycosidases. The heparan sulfate cleaving endoglycosidase heparanase is frequently overexpressed in tumors wherein its heightened canonical activity results in the release of chemotactic factors sequestered within the ECM resulting in greater cancer invasion [27].

An increasing number of papers implicate protease-independent ECM modifying mechanisms for invasion of cancer cells. Such processes involve the deformation of BM mediated through the proliferative or migrative force of cancer cells: cancer invadopodia, transient and cyclically extended cellular protrusions act both as locally concentrated hotspots for protease activity as well as ECM-deforming pore-forming agents [28]. Not just cancer cells, even cancer-associated fibroblasts deform BM in a metalloprotease-agnostic manner by exerting contractile forces on the BM molecules and altering their physical properties [29]. These observations suggest how the tandem interplay of chemical and mechanical contributions to ECM remodeling is a norm rather than an exception in the context of tumor microenvironmental dynamics.

ECM is subject to extensive biochemical alterations by the cells of the tumor microenvironment (figure 1(B)). These could include dysregulation of otherwise physiological posttranslational modifications that are essential to their secretion and assembly. One of the best-known examples is the hydroxylation of procollagen mediated by lysyl hydroxylases and the crosslinking of collagen fibrils through lysyl oxidases: such processes are crucial to the assembly of stromal collagen ECM during tissue homeostasis [11]. However, an overexpression of such enzymes in cancers leads to stiffening of tumor ECM exacerbating the proliferation and migration of cancer cells (the role of lysyl oxidases in cancer has been extensively reviewed in [30]).

In fact, cancer cells regulate the sulfation of GAG chains of proteoglycans both through upregulation of sulfating enzymes and through downregulation of desulfating counterparts, resulting in altered ECM and a tumor microenvironment promoting invasion and metastasis [31, 32]. Another important matrix constituent, hyaluronic acid, a GAG is subject to changes in size because of the interplay between its biosynthetic enzyme hyaluronic acid synthase and its degrading enzyme hyaluronidase [33]. As a result, high molecular weight hyaluronic acid (HMWHA) can be broken down into low molecular weight hyaluronic acid (LMWHA): whereas HMWHA is inhibitive to cancer cell migration, LMWHA promotes it. The sculpting of the ECM by transformed and associated cells of the tumor microenvironment results in topological alterations that have been showcased using methods such as second harmonic generation microscopy (SHG). In fact, SHG imaging of ECM alterations have been used to exhibit the histopathological distinctions in untransformed tissues, benign- and malignant-tumors of the ovaries as well as to generate collagen signatures that can prognose progression of specific cancers (please see a detailed treatment of the various methodologies used to study the biological, chemical and physical characteristics of the ECM in the context of the tumor microenvironment in box 1) [34].

Unlike elastic structures which deform upon stress impingement and relax to their original state upon stress release, biological structures dissipate a proportion of energy that has been exerted to deform them. Therefore, the mechanical behavior of tissues and tumors can be characterized as being viscoelastic, meaning they show properties of both viscous and elastic materials, typically by deforming initially like an elastic material followed by a viscous material-like time-dependent mechanical response and energy dissipation [46]. Viscoelastic materials can be characterized quantitatively using both a storage modulus (which can be measured as the ratio of the elastic stress to strain, indicating in effect, the capacity of a material to store energy), and a loss modulus (which measures the energy lost per cycle of sinusoidal deformation) [47].

To a considerable extent, geometries of arrangement and interaction between elements constituting materials influence their mechanical properties. This is evident especially for polymeric materials, such as the ECM making them one of the predominant determinants of tissue- or tumor-mechanical behaviors [46]. For instance, the presence of loose polymeric ends within their network, the concurrent presence of weaker polymeric networks overlain on existent covalently linked networks, or the conversion of long chain polymers into short chains with lesser entanglement and connections predispose to a viscoelastic behavior, wherein the deformation is transient and is known as non-viscoplastic type of viscoelasticity [48]. In contrast, the presence within the network of interspersed weak crosslinks or even entirely constituting the network, allows for transient breaking and rejoining of polymers to form new network architectures (leading to residual deformations) and is called viscoplastic viscoelasticity [49, 50].

Despite some recent progress in deciphering the effect of mechanical changes in ECM on cell phenotype, driven mainly by our ability to tune the storage and loss moduli independently of each other, (see next section), how cells alter such mechanical properties endogenously is relatively unclear. That tumor ECM is stiffer in nature (and shows elevated storage modulus) can be explained by a greater degree of covalent crosslinking mediated by remodeling enzymes that are secreted by cancer cells as well as their associated stromal cells as described in the previous section [51]. On the other hand, the concurrent presence, within collagen fibrillar architectures of weaker polymeric networks brought about by aggregative sulphated proteoglycans may also alter the loss moduli of such tumor matrix microenvironments [23]. Synthesis of a collagen-rich ECM by cancer cells and their associated fibroblasts to replace the non-fibrillar BM ECM potentially alters the nonlinearity of tumor microenvironmental elastic response [47]. The viscoplastic nature of fibrillar collagen I matrices is used to advantage by migrating cancer cells that reorient the fibers parallel to their egress leading to the tumor associated collagen signatures that have been described for cancer migration [52].

Like their non-malignant counterparts, cancer cells show distinct behaviors in different ECM milieu (figure 1(C)). Such milieu may differ broadly in three ways: through different compositional chemistries of ECMs, through variations in their architectures (fiber arrangements), and through altered mechanical-rheological behaviors. Although differences in each of these categories are concurrent with, and consequential to, changes in other two, most investigations have focused on demonstrating the consequences of variations of one category on cancer cell phenotype.

A recent study in high grade serous ovarian cancer rigorously demonstrates that substrata of different ECM molecules or their combinations (including, but not limited to collagen I, IV and VI, laminin, and fibronectin) modulate the response of the cancer cells to chemotherapy as well as apoptotic cues in a FAK- and β1 integrin-pMLC-YAP signaling-dependent manner [56]. Recent studies show that behaviors such as migration and chemoresistance are sensitive to the nanospacing of the ECM ligands that engage their cognate receptors on the cell surface, suggesting that such plasticity is linked to dynamical changes in signaling landscapes [57]. Ligand density may have profound effects on decisions made by cancer cells to switch between dormant and activated states. Elegant experiments using hydrogels with tunable ligand concentrations and mechanical behaviors of matrices indicate that increased ligand densities and a degradable ECM encourage activation of dormant cancer cells, whereas increased ECM degradability by itself results in restricted survival and cellular dormancy [58, 59]. Similar results have also been observed for switches between growth and dormancy for ER⁺ breast cancer cells [60]. Such approaches explore the contributions of ligand density to cancer cell phenotype by tuning the former and observing variations in the latter; however, there are fewer studies that explore endogenous densities of bound ligands within tumor-like microenvironments. Efforts in this direction have focused on measuring adhesion with cell substrata to estimate concentrations of ECM ligands that engage cells causing them to latch on to the ECM surface [61]. Even so, measuring adhesion within 3D ECM microenvironments is challenging. Mooney's group has used Forster resonance energy transfer signals between fluorescein-painted cells embedded in Rhodamine-conjugated RGD-alginate gels to calculate bound ligand densities [62]. A second approach has used fluorescently tagged integrin domain constructs that can measure specific integrin ligand densities within collagen-gelatin hybrid scaffolds using multiphoton microscopy [61].

Recent investigations demonstrate how altering the ECM architecture may modulate cancer cell traits. Studies using reconstituted in vitro collagen I scaffolds show that as fiber diameter is increased, the invasiveness of breast cancer cells, MDA-MB-231 which employ mesenchymal migration and MCF-7, which employ amoeboidal migration, increases [63]. On the other hand, collagen fiber alignment has been shown to regulate not just migration, but also MMP secretion and invadopodia kinetics [64]. The presence of dermatan sulfates in collagen I scaffolds alters their alignment properties (as detected using second harmonic imaging) leading to changes in network-like behaviors of the aggressive SKOV3 ovarian cancer cells [21]. Macroscale architectural differences in tissues may deeply influence cancer progression: Paul and co-workers show how the disordered blood vessel topography of the caudal vascular plexus in zebrafish predispose colonization of intravasated breast and mammary cancer cells relative to the linear topographic arrangement of a less preferred organotypic site: the brain [65]. Investigating organ-specific differences in ECM compositions that may regulate vascular topographies is likely to provide rich insights into architectural modulation of cancer organotropism.

In addition to architecture, the mechanical properties of ECM have shown to intimately regulate cell behavior in both physiological and pathological contexts [46]. The state of ECM crosslinking as mediated by cells of the tumor microenvironment has contextual effects on transductive cues guiding cancer progression. When stromal ECM constituents, such as fibrillar collagens are crosslinked, they stiffen and induce integrin clustering and migratory signaling cognate to focal adhesion formation and PI3K activity [66]. Activation of FAK/Src pathway by a stiffer stromal ECM in an integrin signaling dependent manner rewires the signaling to drive cancer cell proliferation in primary colorectal and breast tumors [51]. The effect of changes in storage modulus on cell spreading and proliferation have shown that softer substrata are inhibitive to such processes. In addition to compositional complexity, stiffness of ECM too tunes phenotypes such as proliferation, migration, progenitory potentials and sensitivity to drugs in ovarian and breast cancers [21, 67]. In the context of ovarian cancer, stiffness of the ECM microenvironment has been observed to increase cell migration in a FAK-dependent manner and disaggregation of spheroids constituted from the aggressive SKOV3 cell lines [68]. However, some of these studies have typically been conducted with covalently crosslinked elastic substrata such as polydimethylsiloxane and polyacrylamides, whereas ECM is mechanically viscoelastic and viscoplastic.

Chaudhuri and co-workers have sought to examine the changes in behaviors of transformed and untransformed cells grown on covalently (elastic) and ionically (viscoplastic) cross-linked alginate gels: they observed that the cell spreading on soft viscoelastic gels exceeds that on elastic gels with similar storage moduli [69]. In one study comparing the response of normal and cancer cells, primary hepatocytes and hepatocellular carcinoma cells show opposite behaviors when cultured on soft viscoelastic substrata: whereas hepatocytes spread to a lower extent (relative to their spreading on elastic substrata), cancer cells spread faster [70]. This has been attributed to differential rates of binding kinetics between cell surface ligands and substrata, but should ideally be tested using isogenic lines of benign and malignant cells. Analytical approaches combined with experiments have further shown that an intermediate viscosity in softer gels engenders spreading of cells, whereas at higher stiffnesses, cell behavior is agnostic to variations in loss moduli [71].

As tumor cells migrate through ECM, the latter acts as a mechanical barrier to discern differences in physical properties of cells such as their size, shape, and rigidity (figure 1(D)). For example, a more crosslinked BM tends to restrict invadopodia formation and impedes migration [72]. Dissemination of cancer cells depends also on the plasticity of cellular populations to switch between various modes of dissemination. Migration constitutes a spectrum, whose extremes consist of collective cell migration and dispersed single-cell migration [73]. Collective cell invasion involves ensembles of cells moving as 'leaders' and 'followers' and has been proposed to require greater cell–cell adhesion and matrix degradation as requisites for migration. However, even mesenchymal cells, with decreased expression of the epithelial markers such as E-cadherin are known to migrate collectively [74]. This is consistent with recent studies that show that despite low cadherin expression, tumor cells of the invasive lobular carcinoma of breast invade in a collective manner: Ilina and co-workers have proposed a crucial role for the ECM as a confining environment that in the context of cell density, regulates jamming-unjamming transition irrespective of E-cadherin expression [75]. On the other hand, when cell adhesion is minimal and pore sizes between stromal ECM fibers are narrow, a dispersed 'mesenchymal' mode of migration is evidenced. Such a process is typically associated with the ability of cells to actively adhere and de-adhere from ECM. Another variant, amoeboid dispersed migration typically eschews high levels of cell–ECM and cell–cell adhesion [73]. The effects of the ECM as a physical confining agent have profound effects on cellular behavior within spheroidal ensembles. Delarue and co-workers have dissected multiscalar mechanisms of regulation of cell proliferation in confined colorectal-, breast-, and sarcoma cancer spheroids. The compressive stress generated by the confinement decreases cellular volume at the spheroidal core, followed by the expression of the proliferation inhibitor p27^Kip1, and finally a cell cycle arrest [76].

The different behaviors of migrating cancer cells can be assimilated into a single framework in the form of a phase diagram, wherein spheroidal cores resemble solid phase, streams of connected migrating cells resemble liquid phase, whereas dispersed mesenchymal cells represent gaseous phases. These could be primarily understood to be a function of the cells to attach to one another but also the dynamics of confinement provided by the ECM surrounding them. Kang and co-workers propose that unjamming transition explains the gas-like dispersion seen for cancer cells in sparsely distributed stromal-like environments (low density collagen) but liquid-like dispersion in dense counterparts (high density collagen) follows a jamming transition [77].

Chaudhuri and others propose a framework wherein the biochemical, architectural, and mechanical properties of the ECM can be integrated together to assess how it influences cancer cell migration. The confining ability of the peritumor ECM is not just a function of the fiber lengths and pore sizes (a function of the polymerization process of ECM macromolecules) and the balance between diffusive degradative (proteolytic and glycosidic) and anti-degradative biochemical cues, but also the mechanical properties of the ECM [47]. If the ECM fibers are sufficiently plastic, their smaller pores can be remodelled even in a protease-independent fashion using invadopodia resulting in tumor cell migration through wider pores [72]. The greater propensity for LMWHA to allow cancer cell migration could be driven by their greater degree of viscoplasticity driven by weaker linkages between smaller polymers along with greater porosity. More elastic matrices (rendered through covalent cross linking of hyaluronic acid polymers and glycol-heparin) on the other hand, have been found to be inhibitive to the growth and migration of glioblastoma spheroids.