Engineering the Extracellular Matrix to Model the Evolving Tumor Microenvironment

Summary

Graphical Abstract

Open in a separate window

Highlights

Introduction

The extracellular matrix (ECM) was one of the first elements of the microenvironment shown to be altered in tumors relative to normal tissue. Composed of fibrillar proteins, glycoproteins, and proteoglycans, the ECM is an organizational structure for tissues, provides cues to direct cellular signaling, and serves as a reservoir for growth factors (Eble and Niland, 2019). Early studies demonstrated that the ECM could direct tumor cell behavior, with increases in proliferation for cells cultured on ECM compared with tissue culture plastic (Vlodavsky et al., 1980). Clinical evidence also supports a role for ECM in cancer risk and prognosis across multiple tumor types. For example, women with elevated levels of collagen in their breasts have a greater than 4-fold increased risk of developing breast cancer (McCormack and dos Santos Silva, 2006). In ovarian, lung, breast, and colorectal cancer, patterns of ECM gene expression have been identified that correlate with poor prognosis (Cheon et al., 2014; Lim et al., 2017; Pearce et al., 2018; Yuzhalin et al., 2018).

Although tumors initiate in the native ECM of the primary or metastatic site, this matrix will change throughout tumor progression. Cancer cells frequently have altered production of ECM proteins or matrix remodeling enzymes (e.g., matrix metalloproteinases [MMPs], lysyl oxidases) (Eble and Niland, 2019). In addition, proliferation and activation of fibroblasts to a cancer-associated fibroblast (CAF) phenotype results in substantial ECM deposition (Sahai et al., 2020). Utilizing human tumor cells in a mouse xenograft host, it was recently demonstrated that tumor cells produce most of the ECM modulating enzymes, stromal cells produce most of the glycoproteins, and both components produce collagens (Hebert et al., 2020). Characterization of the tumor ECM by immunohistochemical staining or RNA-based analysis has demonstrated that ECM composition around the primary tumor and metastatic sites undergoes radical changes during tumor progression. For example, many studies have documented upregulation of collagen I in primary breast, liver, lung, and esophageal tumors and in metastatic ovarian cancer (Fang et al., 2019; Fogg et al., 2020; van 't Veer et al., 2002; Zhang et al., 2018). Hyaluronic acid (HA) is also significantly elevated in a wide range of cancers (e.g., breast, prostate, bladder, colon) and is associated with an increased likelihood of metastasis and poor prognosis (Lokeshwar et al., 2014). In a tissue microarray of patients with breast cancer, higher epithelial fibronectin expression was observed in patients with overall worse survival (Bae et al., 2013). Finally, alterations in laminin composition are common in many cancers, although these findings vary significantly with laminin isoform. For example, laminin-111 is decreased in breast tumors compared with normal breast tissue (Gudjonsson et al., 2002), whereas laminin-332 is upregulated in many cancers (including breast) and is associated with poor prognosis (Guess and Quaranta, 2009).

More recent studies have begun to address the limitations of identifying individual protein differences through immunostaining by using proteomics to describe the tissue and tumor matrisome. For instance, comparison of primary colon cancer and liver metastases revealed 23 proteins that were shared between tumor sites but were not detected in normal colon or liver (Naba et al., 2014b). Proteomic studies have also revealed differences in the matrisome of tumors with different metastatic potential. For example, ECM differences were determined from the primary tumors generated by xenografts of MBA-MB-231 breast cancer cells compared with LM2 breast cancer cells, which have higher metastatic potential (Naba et al., 2014a). In a similar manner, biopsies from more advanced omental metastases in ovarian cancer showed higher expression of glycoproteins (fibrinogen and fibronectin), along with increased proteoglycans, secreted factors, and affiliated proteins relative to smaller metastases (Pearce et al., 2018). More recent studies using proteomic analysis of the ECM have identified tenascin-C as a prognostic marker in lung cancer (Gocheva et al., 2017) and SERPINB1 as a component of brain tropism in breast cancer metastasis (Hebert et al., 2020). In a similar manner, in-depth characterization of changes in the matrisome of other tumor types may identify new therapeutic targets.

As the cellular and matrix components change with an evolving tumor microenvironment, the mechanics at the tissue and cellular levels also evolve. At the tissue level, the change in mechanics is characterized by an increase of bulk stiffness due to increased deposition of fibrillar matrix proteins, such as collagen I, as a part of a fibrotic response (Lopez et al., 2011; Pearce et al., 2018). In mouse breast cancer models, in situ atomic force microscopy (AFM) measurements showed that stiffness increased with tumor progression and ECM and tumor-associated vasculature were the drivers of increasing stiffness (Lopez et al., 2011). In patients with colorectal cancer, metastases to the liver were stiffer than primary tumors in the same patient owing to activation of highly contractile fibroblasts in the metastatic microenvironment and their deposition of fibrillar matrix (Shen et al., 2020). Stiffness can reinforce fibroblast activation, creating a pro-tumorigenic, positive-feedback loop. These findings support that a mature tumor microenvironment continually enhances fibrosis, stiffening the mechanics of the environment. When the tumor microenvironment is examined at a cellular level, it becomes apparent that these mechanical changes also result from increased matrix cross-linking and reorganization of fiber architecture, alterations that can physically guide cells and influence their behaviors. For example, in breast cancer, it has been seen that collagen cross-linking by lysyl oxidase promotes integrin clustering and focal adhesion formation, which, in turn, promote tumor progression via increased cell invasion (Levental et al., 2009).

Examining the impact of changes in ECM composition and mechanics in vivo remains challenging owing to the complexity and cost of animal models, as well as the need to induce these matrix changes in a tumor-localized manner. In contrast to genetically engineered models of oncogenic drivers of cancer, most models of altered ECM remain systemic rather than subject to site-specific (e.g., Cre/Lox) or temporal (e.g., TetR) activation. The potential to use a site-specific Cre/Lox system has been demonstrated with subcutaneous versican depletion in mice; however, to our knowledge this is the only published report of these techniques to study the role of ECM in cancer progression (Fanhchaksai et al., 2016). In this review, we will describe recent advances in engineering methodologies that are being used to examine the effects of tumor-associated alterations in the ECM, including changes in tissue-level architecture, ECM composition, and tissue- and cellular-level mechanics. These methods can be used to reconstruct the microenvironment of different tumor types to determine how the altered ECM drives tumor progression and subsequently identify mechanisms to slow or stop these effects.

Engineered ECM Models Mimic Cancer-Associated Changes in Tissue-Level Architecture

Normal tissues have multi-cellular organizations that are key to their function and are disrupted in cancer. Pioneering studies from the group of Mina Bissell demonstrated that questions about tissue architecture could be examined in vitro, as cells had different morphologies when cultured on top versus embedded in basement membrane (Figure 1). This was extended to cancer cells, demonstrating that blocking interactions with the ECM could revert a malignant phenotype (Weaver et al., 1997) or make tumor cells susceptible to apoptosis (Wang et al., 2002). Building from this approach, networks of endothelial cells have been formed in fibrin gels to examine the influence of different stimuli on tumor cell intravasation (Zervantonakis et al., 2012) and extravasation (Escribano et al., 2019; Song et al., 2018). In recent work, this platform was extended to incorporate an embedded “tumor,” an endothelial-lined vessel, and patterned growth-factor releasing particles to generate chemotactic gradients (Meng et al., 2019). Such multi-component models are an important step to faithfully recreate tumors in vitro.

Open in a separate window

Figure 1

Examples of In Vivo Architectures Reproduced In Vitro Using ECM

Structures resembling mammary ducts form when mammary epithelial cells are cultured in 3D gels of basement membrane. Endothelial cells embedded in fibrin can generate vessel structures; when coupled with tumor cells embedded in the matrix or flown through the vessel, the processes of intravasation or extravasation can be examined. Finally, when collagen I is gelled around molds, structures with different architectures can be formed and lined with cells, such as the model of DCIS shown here. Created with BioRender.com.

Of course, many tumors become fibrotic as they progress, such that culture in collagen gels may be more physiologically relevant than basement membrane. Collagen I gels also have greater mechanical integrity than basement membrane or fibrin, which has enabled engineers to “pre-pattern” the tissue structure through microfabricated molds inserted into gelling matrix and examine the effects of different macro-structures on cell behavior (Nelson et al., 2006). Using a method developed to generate lumens to mimic blood or lymph vessels (Jimenez-Torres et al., 2016), a recent model of ductal carcinoma in situ (DCIS, the earliest stage of breast cancer) has been described (Figure 1). In this model, a lumen with a diameter of approximately 300 μm is lined with normal mammary epithelial cells and then left empty in the center (normal) or filled with breast cancer cells (DCIS) (Ayuso et al., 2018). This lumen-generating method has been modified to generate models of ovarian cortical inclusion cysts, demonstrating that, of the two proposed cells of origin for ovarian cancer, only fallopian tube epithelial cells are able to invade into the surrounding matrix (Fleszar et al., 2018). As cortical inclusion cysts range in size and are believed to resolve over time, a recent paper modeled the range of curvatures observed in human cysts and determined that as curvature increased (i.e., the cyst diameter decreased) the fallopian tube epithelial monolayer transitioned from stable to invasive (Fleszar et al., 2019).

Biomaterial Approaches Enable Modeling of Tumor-Associated Changes in ECM Composition

Although the studies highlighted above utilized ECM to generate physiologically and pathophysiologically relevant structures in vitro, they did not ask how changes in the ECM influence tumor progression. To better understand the role of ECM composition changes in cancer progression, biomaterials-based approaches have been used to build in vitro models of the tumor ECM microenvironment. Here, we will discuss the design and use of 3D scaffolds derived from decellularized ECM, intact ECM molecules, and functional fragments of ECM molecules such as minimal adhesion sequences (Figure 2). To construct a scaffold, these materials are either directly gelled or incorporated into materials that are modified to enable cross-linking (e.g., crosslinker-modified poly(ethylene glycol) [PEG], gelatin, and HA).

Open in a separate window

Figure 2

Example Methods Used to Generate ECM Scaffolds with Different Compositions

Created with BioRender.com.

Bioengineering Approaches to Study the Mechanics of the Tumor ECM

Increased deposition of fibrillar collagen is a hallmark of tumorigenesis and is associated with increased bulk stiffness; thus, many early in vitro systems aimed at modeling mechanical changes during tumorigenesis consisted of hydrogels formed using increasing concentrations of collagen I (Baker et al., 2009; Provenzano et al., 2009). Although increased collagen I density is reflective of the native tumor microenvironment, this approach does not permit tissue stiffness to be studied independently; collagen concentration affects not only stiffness but also fiber density, which is known to influence cell behavior (Berger et al., 2017; Lewis et al., 2018). Perhaps more importantly, the relatively weak non-covalent bonds that form collagen I gels provide only a limited range of elastic moduli (0.1–4 kPa), which is substantially less than the range of stiffnesses seen in most solid cancers (0.4–60 kPa) (Berger et al., 2017; Kawano et al., 2015). To overcome these limitations, various biomaterials approaches have been developed to examine the impact of bulk and local mechanics (Figure 3).

Open in a separate window

Figure 3

Comparison of Tissue-Level and Cellular-Level Mechanical Changes During Tumorigenesis

At the tissue level, mechanical changes during tumorigenesis are characterized by an increase in bulk stiffness due to increased vasculature, deposition of fibrillar ECM proteins, and a loss of tissue elasticity. At the cellular level, mechanical changes are facilitated through the remodeling of ECM architecture, including increased fiber alignment and cross-linking, which results in heterogeneity. Created with BioRender.com.

The Engineered Matrix In Vitro Is Dynamic, as Is the Extracellular Matrix In Vivo

An important finding that has arisen from numerous studies is the confirmation that the initial engineered matrix that is presented to the tumor cells is quickly remodeled, as has been shown for other cell types (Loebel et al., 2019). However, the feedback that is activated depends on the initial scaffold composition. For example, glioblastoma cells were observed to produce higher levels of HA when cultured in gels generated from GelMA alone compared with gels composed of GelMA copolymerized with MeHA (Chen et al., 2017). Cells cultured on top of ECM mimics of different tissues have been used to screen for environments that support breast cancer cell dormancy (Barney et al., 2020). In this system, cells that generated a stable, fibrillar network of fibronectin maintained a dormant state, which could then be reversed through MMP-2 mediated degradation of this network (Barney et al., 2020). Additionally, stiffness can independently regulate matrix remodeling. Glioblastoma cells cultured in a PEG-based scaffold (modified with RGD peptides and MMP-degradable sequences) with interpenetrating HA upregulated HA synthase 1 and collagenase MMP-1 in stiff (26 kPa) versus soft (1 kPa) gels (Wang et al., 2014). Recently, our group demonstrated that Mena, an actin regulatory protein, was upregulated when breast cancer cells were embedded in stiff (12 kPa) but not soft (2 kPa) gels and that the increase in Mena facilitated organization of cell-secreted EDA-fibronectin to overcome the inhibition to invasion observed in the stiff matrix (Berger et al., 2020).

Future Directions

Owing to mounting evidence that ECM fibers are critically influential in tumor cell behavior and observations that omission of fibers may result in incorrect predictions of cellular outcomes (Berger et al., 2017; Davidson et al., 2020), an emphasis on fiber-tunable scaffolds has been emerging in the evolution of tumor tissue engineering. Many techniques allow alteration of fiber size independently of overall collagen concentration or bulk Young's modulus (Carey et al., 2012; Oh et al., 2020; Seo et al., 2020), although these approaches have been limited by issues such as poor control over fiber size modulation, non-physiological fiber sizes, or use of specialized equipment. There has been recent progress on this front, wherein simple modifications to the common collagen gelation protocol yielded physiological fiber sizes (Gong et al., 2020), although there remains a need for accessible technologies capable of modulating fiber features in a controlled manner. Another limitation that remains across these fiber-focused studies is that they are typically conducted with collagen-only materials and have not included other ECM cues prevalent in the tumor environment (e.g., fibronectin, laminin, HA). Moving forward, there is an immediate opportunity to blend fiber-tuning techniques with copolymerization of ECM cues, thereby creating multi-component scaffolds with independently tunable fiber density, fiber thickness, ECM composition, and bulk modulus. Interpreting these complex, multivariate relationships will be aided by computational approaches (Kreeger et al., 2018).

Studies that examine the influence of fiber features on cell behavior have also raised the intriguing question of whether the biological identity of the fiber even matters, or if it is the physical features of the structure that dominates the delivery of ECM information to the cells (Baker et al., 2015). Synthetic materials have been used to produce controlled, fibrillar environments to study fibrosis (Baker et al., 2015; Davidson et al., 2020), and these platforms may represent another path forward in advancing tumor-mimicking research. For example, a modified, non-native sugar (dextran vinylsulfone) can be electrospun to create scaffolds with mechanical tunability at both the single fiber and bulk scaffold level, while also allowing modulation of fiber density and alignment (Davidson et al., 2020). These fibers can be functionalized with ECM-derived peptides or proteins to increase their biomimicry. This type of system has the potential to address many of the challenges of teasing apart the relative contributions of changes in architecture, composition, and mechanics on tumor progression.

There is also room for growth in developing dynamic scaffold environments that can undergo user-controlled temporal changes in ECM stiffness or composition. Dynamic stiffening of scaffolds can be achieved by multiple means (Madl and Heilshorn, 2018) and has been used to examine the role of matrix stiffness in tumor progression (Ondeck et al., 2019). Additionally, recent years have seen the development of numerous bio-orthogonal scaffold manipulations (Madl and Heilshorn, 2018) that enable dynamic modification of scaffold composition, which have not yet been applied in the context of cancer. A wide range of cytocompatible modification chemistries has been employed to temporally control the in situ addition, removal, and reversible patterning of ECM cues (DeForest and Anseth, 2011; Madl and Heilshorn, 2018; Shadish et al., 2019). As we learn more about the ECM changes that accompany tumorigenesis, these highly flexible techniques may be implemented to create defined, temporally tunable microenvironments that allow researchers to probe complex questions regarding the signals and mechanisms that drive tumor progression. Of course, one remaining complication is the difference in time scales between tumor ECM changes (e.g., days) and chemical modification of biomaterials (e.g., minutes).

Finally, although the aforementioned platforms are able to mimic many elements of the native ECM environment, the only scaffold that can truly recapitulate the complexity of the ECM is the native tissue itself. Although ex vivo organ cultures are typically seen as lacking tunability, opportunities exist to modulate the ECM in these intact tissues. Approaches to selectively deplete ECM components have been explored for other applications (Rodriguez et al., 2011). Meanwhile, although ECM enrichment of ex vivo tissues has not previously been described, the successful application of gene delivery to in vivo structures (Mueller et al., 2018) suggests that “ECM editing” of organ cultures is an achievable goal. The ECM complexity of these “top-down” scaffolds would complement the defined environments provided by “bottom-up” approaches. Together, the emerging technologies discussed in this section have the potential to provide robust tumor-mimicking platforms that will significantly advance our understanding of tumor biology and progression.

Acknowledgments

References