Amoeboid–mesenchymal transition and the proteolytic control of cancer invasion plasticity

Significance

In vivo, cancer cells exhibit invasion plasticity as they traverse extracellular matrix (ECM) barriers in collective or single-cell fashion while displaying either mesenchymal or amoeboid phenotypes. Current dogma holds that invasion plasticity enables cancer cells to reversibly transition between proteinase -dependent and -independent mechanisms, thereby complicating efforts to therapeutically target invasion programs. To the contrary, we uncover a requirement for proteolytic ECM remodeling—regardless of invasive strategy, when cancer cells infiltrate tissue barriers assembled in vitro or in vivo. A single membrane-anchored metalloproteinase, MMP14/MT1-MMP, is responsible for driving both amoeboid and mesenchymal tissue-invasive activity whose importance is highlighted in analyses of human breast cancer tissue where proteinase-expressing carcinoma cells are surrounded by tracts of proteolyzed type I collagen.

Abstract

Invasion plasticity allows malignant cells to toggle between collective, mesenchymal, and amoeboid phenotypes while traversing extracellular matrix (ECM) barriers. Current dogma holds that collective and mesenchymal invasion programs trigger the mobilization of proteinases that digest structural barriers dominated by type I collagen, while amoeboid activity allows cancer cells to marshal mechanical forces to traverse tissues independently of ECM proteolysis. Here, we use cancer spheroid-3-dimensional matrix models, single-cell RNA sequencing, and human tissue explants to identify the mechanisms controlling mesenchymal versus amoeboid invasion. Unexpectedly, collective/mesenchymal- and amoeboid-type invasion programs—though distinct—are each characterized by active tunneling through ECM barriers, with expression of matrix-degradative metalloproteinases. CRISPR/Cas9-mediated targeting of a single membrane-anchored collagenase, MMP14/MT1-MMP, ablates tissue-invasive activity while coregulating cancer cell transcriptional programs. Though changes in matrix architecture, nuclear rigidity, and metabolic stress as well as the presence of cancer-associated fibroblasts are proposed to support amoeboid activity, none of these changes restore invasive activity of MMP14-targeted cancer cells. While a requirement for MMP14 is bypassed in low-density collagen hydrogels, invasion by the proteinase-deleted cells is associated with nuclear envelope and DNA damage, highlighting a proteolytic requirement for maintaining nuclear integrity. Nevertheless, when cancer cells confront explants of live human breast tissue, MMP14 is again required to support invasive activity. Corroborating these results, spatial transcriptomic and immunohistological analyses of human breast cancers identified MMP14 expression in tissue-infiltrating carcinoma cells that were further juxtaposed with proteolyzed type I collagen fragments, underlining the pathophysiologic importance of this proteinase in directing invasive activity in vivo.

Invasion plasticity is a hallmark of cancer cells capable of assuming collective, mesenchymal, or amoeboid phenotypes as they infiltrate tissues (1–5). Mesenchymal-type phenotypes, elicited by leader cells during collective invasion or individual cells during single-cell invasion, assume elongated shapes, while exerting contractile and protrusive forces on surrounding tissues (4, 6, 7). By contrast, amoeboid phenotypes are characterized by spheroid-shaped cells that invade in a more fluid fashion as discrete single-cell units (1, 8, 9). Independent of phenotype, tissue-invasive cancer cells are confronted by the extracellular matrix (ECM), a heterogeneous composite of structural proteins, proteoglycans, and glycosaminoglycans (10). Though complex, the ECM is dominated structurally by type I collagen, a triple-helical fibrillar protein that is deposited variably in a tissue-specific manner (11). Current dogma holds that mesenchymally invading cancer cells traverse ECM barriers by mobilizing proteolytic enzymes that actively degrade collagen-rich tissues (4, 12–14). By contrast, amoeboid cancer cells are thought to navigate similar tissues in a proteinase-independent fashion by using mechanical force to remodel the ECM, or alternatively, by distorting cell shape to infiltrate matrix pores (1, 4, 8, 15–18).

Here, we characterize the requirements of mesenchymal versus amoeboid cancer cell invasion in relation to collagen density, pore size, and alignment as well as the modifying effects of nuclear rigidity, nutrient stress, hypoxia, and cancer-associated fibroblasts. Unexpectedly, we find that while mesenchymal and amoeboid invasion mobilize distinct transcriptional programs, each activate a comparable set of ECM-degradative proteolytic enzymes, including the membrane-anchored metalloproteinase, MMP14/MT1-MMP. Following CRISPR/Cas9-mediated deletion of MMP14, cancer cells lose all tissue-invasive activity in dense matrix constructs while remaining unable to access proteinase-independent invasive programs. In low-density collagen matrices, however, invasive activity proceeds independently of MMP14, but under these conditions, the proteinase nevertheless plays a role in maintaining nuclear integrity. Nevertheless, when model constructs are replaced with human tissue explants, only MMP14-expressing cancer cells actively tunnel through native matrix. Finally, analyses of human breast cancer spatial transcriptomics and tissue biopsies confirm that invasive lesions are uniformly found in association with MMP14 expressing-carcinoma cells decorated with tracks of type I collagen degradation products. Hence, while invasion plasticity enables cancer cells to toggle between mesenchymal and amoeboid phenotypes, MMP14-dependent ECM proteolysis plays the dominant role in driving tissue-invasive activity.

Results

Characterization of 3D Mesenchymal vs. Amoeboid Cancer Cell Invasion.

Tumor spheroids comprised of human HT1080 fibrosarcoma cells embedded in 3D type I collagen hydrogels (2.2 mg/mL) with an average pore size of ~2 to 4 µm² and rigidity of ~162 Pa, actively infiltrate the surrounding matrix over a 72-h culture period while displaying a largely collective/mesenchymal invasion program (Fig. 1 A–D and Movie S1). Characteristic of a mesenchymal phenotype (6), HT1080s display a mixed population of collective and single cells with elongated cell shapes, associated with local strain on collagen fibers and activated β1-integrin (Fig. 1 E–I). As a decrease in β1-integrin activity induces amoeboid behavior, we next cultured HT1080 spheroids in the presence of the β1-integrin blocking antibody, 4B4 (4, 12, 17). Under these conditions, 4B4-treated spheroids lose collective/mesenchymal features, but maintain invasive activity as individual spheroid-shaped cells, exhibiting elevated cortical actin and phospho-myosin light chain (pMLC) (Fig. 1 A–L and Movie S2). Costaining of the plasma membrane and actin cytoskeleton highlight dynamic bleb formation and retraction in amoeboid cells (Fig. 1 M and O) (19). Real-time cell-tracking analyses confirm that control and 4B4-treated amoeboid cells exhibit comparable cellular speed and directional persistence (Fig. 1 P–R and Movies S3 and S4).

Fig. 1.

Characterization of mesenchymal versus amoeboid cancer cell 3D invasion. (A) Representative three-dimensional fluorescent images of control and 4B4-treated (2 µg/mL) HT1080 spheroids embedded in 2.2 mg/mL type I collagen hydrogels after 72-h culture and labeling with phalloidin and DAPI. The Inset shows phase contrast images of spheroids at the time of embedding. (B) High magnification images of control and 4B4-treated HT1080 cells invading through fluorescently labeled type I collagen hydrogels. Cells were labeled with CellMask-647 and Hoechst. (C) Quantification of HT1080 invasion with and without 4B4-treatment. (D) Quantification of the percentage of single vs. collective invasive cells. (E) Quantification of the circularity of individually invading tumor cells. (F) Representative fluorescent images of collagen alignment observed at the leading edge of invasive tumor cells. Bottom images are colored based on fiber orientation. (G) Representative fluorescent images of active ß1-integrin staining in invading tumor cells. (H) Quantification of collagen alignment, by coherency of fluorescently labeled collagen fibers adjacent to invading cells. (I) Quantification of detected ß1-integrin immunofluorescent staining. (J) Immunofluorescent images of invading control and amoeboid cells labeled with phalloidin/DAPI and stained for pMLC. Intensity profiles, depict the fluorescent intensity by channel along the line designated in the image. (K) Quantification of the peak cortical actin staining intensity in invading cells. (L) Quantification of pMLC fluorescence in invading cells. (M) Quantification of membrane blebs observed in single z-slice images of invading cells. (N) Representative three-dimensional z-stack reconstructions of mesenchymal and amoeboid cells, labeled with CellMask-647 and Hoechst. Red arrows indicate membrane blebs on the cell surface. (O) Representative fluorescent images of invading cells stained for CellMask-647, SPY555-FastAct, and Hoechst. Right panels display inverted single channel images for the specified dye. Purple arrows indicate membrane blebs without associated actin structures while orange arrows indicate blebs with actin structures. (P) Endpoint images of invading HT1080 spheroids expressing GFP-NLS protein used to track nuclei during invasion. (Q and R) Quantification of mean speed and mean directional change rate detected during real-time imaging of HT1080 invasion. P-value < 0.05 (*), <0.01 (**), <0.001 (***), and < 0.0001 (****).

Single-Cell Mapping of the Mesenchymal and Amoeboid Invasion Transcriptome.

To characterize the transcriptional programs underlying 3D mesenchymal versus amoeboid invasion, control and 4B4-treated spheroids were subjected to single-cell RNA sequencing. After a 72-h culture period, cells are detected in a continuum of states, ranging from stationary cells confined to the boundaries of the spheroid, to those migrating away from the spheroid body (Fig. 2A). Following cell cycle regression, initial clustering yielded three populations for both mesenchymal and amoeboid conditions (Fig. 2A). Consistent with the progression of cells from stationary to motile states, visualization of gene expression associated with a cell migration gene set (20) demonstrated a gradual increase in migratory behavior that distinguished highly motile cells from low motility/nonmotile populations (Fig. 2A and SI Appendix, Fig. S1 A–C). Analyses of gene ontology (GO) biological process pathways in merged control and 4B4-treated samples show upregulated pathways associated with migration as well as tumor–ECM interactions concentrated in cluster 3 (SI Appendix, Fig. S1D). Upon comparing the motile populations of the amoeboid versus mesenchymal samples, 3,227 genes are upregulated, and 917 genes down-regulated (Fig. 2B; top up/down DEGs). Top distinguishing genes by phenotype included FOXD1, GAS6, STMN1, and TENM2 in mesenchymal cells and SPP1, ROMO1, ME1, and FN1 in the amoeboid cell group (Fig. 2C). Interrogating a previously described GO amoeboid gene set (20), and identifying critical amoeboid invasion-associated genes, such as RHOA (1) and SLC7A11 (21), further supported the amoeboid-like status of 4B4-treated cells (Fig. 2D). In addition, key pathways assigned to the amoeboid phenotype, such as hypoxia-like HIF1-alpha signaling (4, 12, 22), TGF-beta signaling (23) (Fig. 2D) and PI3K/PKB-related pathway activation (8) were also elevated in the amoeboid population (Fig. 2E). Meanwhile, proliferation, translation, and apoptotic processes were down-regulated in 4B4-treated cells, possibly contributing to the lower energetic requirements and stress resistance of amoeboid cells (Fig. 2E) (4, 21). Complementing pathway analyses, amoeboid cells exhibited lower proliferative potential as demonstrated by fewer G2/M phase cells, decreased PCNA levels, and lower cell number recovery following a 72-h invasion period without accompanying changes in cell death (SI Appendix, Fig. S1 E–H). Unexpectedly, while examining transcripts of migration-associated matrix metalloproteinases (MMPs), believed to play little, if any, role in amoeboid behavior, we find similar expression levels during amoeboid-type versus mesenchymal-type invasion, including the membrane-anchored MMPs, MMP14, MMP16, MMP17, and MMP24 as well as the secreted MMP, MMP2 (Fig. 2 D and F).

Fig. 2.

Single-cell transcriptomic characterization of mesenchymal and amoeboid invasion. (A) Representative fluorescent images of HT1080 spheroids at 72 h postembedding, when single-cell RNA sequencing was performed. UMAP plots demonstrate initial clustering results and expression profiles of a cell migration gene set. (B) Top differentially expressed genes between invading control cells and invading 4B4-treated cells. (C) Box plots of top DEGs specific to each condition. Red lines indicate median values. (D) Violin plots comparing gene expression of specified genes and gene sets between control and 4B4-treated invading cells. (E) Analysis of GO biological processes comparing invading amoeboid and mesenchymal cells. (F) Dotplot of MMP expression for control and 4B4-treated HT1080 cells. P-value < 0.05 (*), <0.01 (**), <0.001 (***), and <0.0001 (****).

Mesenchymal- and Amoeboid-Type Cell Invasion Requires MMP-Dependent Activity.

As mesenchymal and amoeboid invasion are associated with MMP expression, we questioned whether structural changes might be detected in the traversed collagen framework. Following embedding of spheroids in fluorescently labeled collagen hydrogels, matrix-free tunnels are found in the wake of both invading mesenchymal cells and amoeboid cells (Fig. 3 A and B). Moreover, as several MMPs are able to cleave collagen fibrils at a single site located approximately ¾ of the distance from its N terminus (24), an antibody specific for this MMP-derived cleavage product, termed ¾-collagen, was utilized to detect MMP-dependent collagenolysis (25). Indeed, regardless of invasive strategy, both mesenchymal and amoeboid cells generated ¾-collagen fragments (Fig. 3 A and C).

Fig. 3.

MMP-dependent mesenchymal and amoeboid invasion. (A) Representative immunofluorescent images of invading control and 4B4-treated HT1080 cells in fluorescently labeled collagen, stained for ¾-collagen degradation products. Yellow lines indicate tunnels formed in the collagen hydrogel. (B and C) Quantification of tunnel diameter and fluorescent intensity of cell-associated ¾-collagen fluorescence. (D) Fluorescent images of HT1080 spheroids after 72-h culture in collagen in the presence of pan-specific MMP inhibitors BB-94 (5 µM) or GM6001 (10 µM). High magnification images and inverted single-channel images show ¾ collagen detected at the cell–collagen interface. (E) Quantification of spheroid invasion in the presence and absence of each MMP inhibitor. (F) Representative images of invasive spheroids of MDA-MB-231 and SUM159 cells after 72-h of invasion in collagen hydrogels. (G and H) Quantification of single vs collectively invading cells and invasive cell circularity. (I) Immunofluorescence images of spontaneously mesenchymal and amoeboid MDA-MB-231 cell invasion and associated ¾ collagen degradation products. (J) Fluorescent images of breast cancer spheroids treated with the indicated MMP inhibitor. (K) Quantification of MDA-MB-231 and SUM159 spheroid invasion in the absence or presence of each MMP inhibitor. P-value < 0.05 (*), <0.01 (**), <0.001 (***), and <0.0001 (****).

To determine whether collagenolytic activity is a necessary component of invasion, control or β1-integrin targeted HT1080 spheroids were cultured in 3D hydrogels in the presence of either of two broad-spectrum MMP inhibitors, BB-94 or GM6001 (26). Contrary to reports of a “switch” to proteinase-independent invasion in the face of proteinase inhibition (2, 8, 14, 15, 27), control HT1080 cells exhibited a complete block in invasion upon treatment with either MMP inhibitor (Fig. 3 D and E). While MMP-inhibited cells are able to extend thin filopodia-like protrusions and even release invasive cytoplasts (28) into the surrounding matrix, movement of intact cell bodies is inhibited completely (Fig. 3 D and E and Movies S5 and S6). Importantly, 4B4-treated, amoeboid HT1080 cells are likewise unable to engage in proteinase-independent invasion (Fig. 3 D and E). Loss of invasive potential in both cases is further associated with abrogated detection of ¾-collagen degradation products (Fig. 3 C and D). A similar complete block in invasion was observed using BB-94 in a “2.5D” invasion assay, where cells are plated atop a 3D collagen hydrogel and allowed to invade into the subjacent matrix (SI Appendix, Fig. S2 A and B) (29). To address whether MMP-inhibition is only sufficient to disrupt the initial migration away from the spheroids or if it would likewise halt the forward penetration of cells already committed to invasion, BB-94 was added after the spheroids were allowed to infiltrate the surrounding matrix for 48 h. Following an additional 48-h culture period in the presence of BB-94, invading cells halted in place, allowing lagging cells to “fill-in” behind leading cells, but without progression of the leading edge (SI Appendix, Fig. S3 A–C). Finally, as amoeboid activity has additionally been shown to be induced by nutrient stress (4, 12, 22), HT1080 spheroids in 3D culture were subjected to low glutamine, low glucose, or hypoxic conditions (SI Appendix, Fig. S4A). Though varying degrees of a shift to single-cell amoeboid invasion are observed (SI Appendix, Fig. S4 A–D), cells remain reliant on MMPs as illustrated by the complete inhibition observed in the presence of BB-94 (SI Appendix, Fig. S4 A and B).

To confirm that a similar reliance on MMP activity is observed in cancer cells capable of spontaneously engaging amoeboid phenotypes, we noted that the human breast cancer cells lines, MDA-MB-231 and SUM159, generate spheroids that likewise invade extensively into the surrounding type I collagen matrix, but with cells inherently exhibiting a mixture of mesenchymal and amoeboid phenotypes (30–32) (Fig. 3F). In this system, ~20% of MDA-MB-231 and ~25% of SUM159 cells at the invasive front display amoeboid activity while the remainder migrate in a collective fashion (Fig. 3 F–H). Under these conditions, the collagen-invasive activity of both amoeboid and mesenchymal cells is again associated with the generation of ¾-collagen degradation products (Fig. 3I). Moreover, despite eliciting spontaneous amoeboid/collective phenotypes, invasion by both breast cancer cell lines is completely inhibited by BB-94 or GM6001 (Fig. 3 J and K and Movies S7 and S8). Following treatment of these lines with the 4B4 antibody, MDA-MB-231 and SUM159 cells demonstrated a partial shift to an amoeboid phenotype (SI Appendix, Fig. S5 A–C), but invasive programs remained totally reliant on MMP activity (SI Appendix, Fig. S5 A and D).

MMP14 Directs Both Mesenchymal and Amoeboid 3D Invasion.

As mesenchymal and amoeboid invasion was associated with the expression of both secreted and membrane-anchored MMPs, we sought to identify the MMP(s) responsible for ¾-collagen generation and invasion. To begin, we treated 3D-embedded spheroids with endogenous inhibitors of MMP activity, i.e., either TIMP-1 or TIMP-2 (33). TIMP-1, an effective inhibitor of secreted MMPs (33), did not affect invasion while TIMP-2, an inhibitor of both secreted and membrane-tethered MMPs (33), strongly blocks invasion of HT1080, MDA-MB-231, or SUM159 cells (SI Appendix, Fig. S6 A–C). Of the MMPs detected, the membrane-tethered MMP, MMP14/MT1-MMP, is known to function as an effective pericellular type I collagenase (34–37). To monitor endogenous MMP14 production and trafficking in real-time, we generated a knock-in MMP14 allele with an mCherry sequence inserted into the extracellular juxtamembrane domain of the proteinase (Fig. 4A). Using these knock-in lines, live cells cultured atop collagen hydrogels actively trafficked MMP14^mCherry-KI containing vesicles throughout the cell body (Fig. 4B). In mesenchymal or amoeboid cells, perinuclear accumulation of MMP14^mCherry-KI is observed with substantial amounts of MMP14^mCherry-KI additionally found in vesicles distributed throughout the cytoplasm as well as the cell periphery (Fig. 4C and SI Appendix, Fig. S7). In nonpermeabilized cells and following immunostaining with an mCherry-specific antibody, MMP14^mCherry-KI exclusively outlines the cell surface (Fig. 4C and SI Appendix, Fig. S7). When visualized as a 3D-reconstruction, the surface presentation of the proteinase is highlighted in both mesenchymal and amoeboid cells (Fig. 4D).

Fig. 4.

MMP14 directs mesenchymal and amoeboid 3D invasion. (A) Depiction of the constructs used to generate MMP14-mCherry knock-in HT1080 and MDA-MB-231 cells. (B) Representative 3D images of an HT1080^{MMP14-mCherryKI} cell transfected with GFP-LifeAct atop a thin collagen gel. (C) Immunofluorescent images of permeabilized and nonpermeabilized HT1080 cells stained with an mCherry-specific antibody to label total vs. cell surface MMP14-mCherry proteins. (D) 3D reconstructions of z-stack images of nonpermeabilized control and 4B4-treated HT1080 cells stained for mCherry. (E) Representative images of control and MMP14-KO spheroids after 72 h of invasion. (F) Quantification of spheroid invasion with CRISPR targeting of MMP14 via two separate guide RNAs. (G) Immunofluorescence staining of ¾-collagen associated with control and MMP14-targeted spheroids. Inverted images depict ¾-collagen staining alone. (H) Quantification of ¾-collagen staining at the tumor cell–collagen interface. P-value < 0.05 (*), <0.01 (**), <0.001 (***), and <0.0001 (****).

To assess the impact of MMP14 activity on each invasive program, CRISPR/Cas9 editing was used to delete its expression in HT1080, MDA-MB-231, and SUM159 cells (SI Appendix, Fig. S8A). Following targeting, all MMP14-deleted cell lines retain normal proliferative and motile activities in standard 2D culture (SI Appendix, Fig. S8 B and C). By contrast, spheroids prepared from knockout HT1080 cells—either in the absence or presence of β1-integrin targeting—are devoid of 3D collagen-invasive activity, while generating only trace levels of the ¾-collagen degradation products despite inserting filopodia-like protrusions into the surrounding matrix (Fig. 4 E–H and Movie S9). Similar results are observed with MMP14-targeted MDA-MB-231 or SUM159 spheroids (Fig. 4E and SI Appendix, Fig. S8 D and E and Movie S10). To rule out the possibility that an MMP14-derived cleavage product yields promotility signals lacking in MMP14-null spheroids, or alternatively, that MMP14-null cells generate inhibitory signals that impede wild-type invasion, control and MMP14-null tumor spheroids were assembled separately and then cultured in close proximity within 3D collagen hydrogels in order to expose each population to soluble factors generated by the opposing population. However, coculture neither rescues the invasive potential of MMP14-null cells nor interferes with the activity of control spheroids (SI Appendix, Fig. S9A). Only when control and MMP14-null cancer cells are comixed into single spheroids were MMP14-null cells observed invading in 3D, presumably, and as described previously (38), by following control cells acting as leader-cells (SI Appendix, Fig. S9B). Taken together, these results are most consistent with a parsimonious model wherein MMP14 functions as a cis-acting proteinase that drives invasive activity by cleaving barrier-forming ECM components. Indeed, wild-type MMP14-expressing cancer cells are unable to infiltrate peptide-crosslinked PEG hydrogels unless MT1-MMP-scissile bonds are incorporated into the matrix (39) (SI Appendix, Fig. S10 A and B). In turn, wild-type cell invasion into synthetic hydrogels incorporating cleavable moieties is blocked in the presence of pan-specific MMP inhibitors or when MMP14 is specifically deleted (SI Appendix, Fig. S10 A and B).

Collagen Cross-Linking Rather than Pore Size Dictates Proteinase-Independent Invasive Potential.

A body of work suggests that proteinases are not required when average pore sizes exceed that of ~10% of the cell’s nuclear dimensions, i.e., roughly 7 μm² for HT1080 and MDA-MB-231 cells (5, 14). Pepsinized collagen (atelocollagen), lacking the nonhelical telopeptide domains critical for regulating fibrillogenesis and intermolecular collagen crosslinks (37), has frequently been used to generate larger pore constructs conducive to the study of proteinase-independent invasion (8, 12, 22, 23, 40–42). Indeed, atelocollagen-based constructs yield hydrogels with an average pore size of ~17um², well in excess of the proposed nuclear size restrictions of HT1080 or MDA-MB-231 cells (14) (Fig. 5 A and B). Under these conditions, both MMP14-wild-type and MMP14-null HT1080 and MDA-MB-231 spheroids exhibit strong invasive potential (Fig. 5 C and D and SI Appendix, Fig. S11 A and B), highlighting the ability of MMP14-null cells to retain intact motile responses in a permissive matrix environment. Interestingly, despite forgoing the requirement for matrix proteolysis in atelocollagen hydrogels during invasion (Fig. 5 C and D and SI Appendix, Fig. S11 A and B), ¾-collagen degradation products are still detected as wild-type, but not MMP14-targeted, cells invade (SI Appendix, Fig. S12). While proteinase-independent invasion of atelocollagen concur with previous reports (8, 13–15), it remains unclear as to whether proteinase-independent invasion arises as a consequence of the increase in matrix pore size, or the absence of the collagen crosslinks characteristic of native tissues (43). As such, we generated telopeptide-intact type I collagen hydrogels formed at low temperatures to allow for the generation of large pore constructs comparable to those observed in atelocollagen matrices (Fig. 5 A and B) (14, 44). While both control HT1080 and MDA-MB-231 spheroids invade robustly in the high porosity native collagen gels, MMP14-null cancer cells display almost no invasive potential despite the larger pore size, aside from small numbers of cells that only partially separate from the spheroid but remain near the central cell mass (Fig. 5 C and D and SI Appendix, Fig. S11 A and B). Hence, even in type I collagen matrices with pore sizes exceeding those previously held to be permissible to proteinase-independent invasion, mesenchymal as well as amoeboid activity remains MMP14-dependent when covalent crosslinks remain intact.

Fig. 5.

MMP14-dependent cancer cell–collagen interactions. (A) Confocal reflectance microscopy images of collagen hydrogels prepared using acid-solubilized rat collagen gelled at 37 or 8 °C and pepsin-extracted bovine collagen (PureCol) gelled at 37 °C. All hydrogels contain 2.2 mg/mL collagen. (B) Quantification of average collagen pore sizes in each condition. (C) Representative fluorescent images of invading spheroids in the indicated collagen conditions. (D) Quantification of spheroid invasion in the indicated collagen conditions. P-value < 0.05 (*), <0.01 (**), <0.001 (***), and <0.0001 (****).

Despite the absolute requirement for MMP14 in driving invasion in native collagen hydrogels, the potentially complex impact of its activity on transcriptional programs remains undefined. As such, high-throughput RNA sequencing was performed on wild-type vs. MMP14-targeted MDA-MB-231 spheroids upon initial spheroid formation as well as following 24 and 48 h embedded in native type I collagen hydrogels (SI Appendix, Fig. S13 A and B). Rather than becoming progressively more quiescent, immobile MMP14-null cells upregulate transcript levels of a large family of genes relative to invading controls (SI Appendix, Fig. S13C). Focusing on the differences observed following 48 h culture in 3D collagen (with 941 genes upregulated and 280 downregulated in MMP14-KO cells compared to controls), while genes associated with proliferation are in fact downregulated in MMP14-null cells, transcripts related to cell–cell adhesion, cell–ECM adhesion, and cell migration are all elevated in the MMP14-KO cells, suggesting an enhanced effort of targeted cells to mount an invasive response while being limited by their inability to cleave surrounding collagen molecules (SI Appendix, Fig. S13 D and E). Of note, comparing all spheroid conditions, expression of other cancer cell-associated MMPs remains largely unaffected in MMP14-null cells, and while MMP2, MMP9, and MMP10 levels increase in the knockout cells, these changes are unable to rescue invasive activity (SI Appendix, Fig. S14).

Cancer Cell– and Microenvironment-Intrinsic Modulation of 3D Invasion Programs.

Independent of MMP14 activity, recent reports suggest that changes in either cancer cell phenotype or the surrounding tumor microenvironment can modulate the matrix remodeling requirements necessary for 3D invasion. Regarding cell-intrinsic mechanisms, the nucleus is the most rigid intracellular organelle whose mechanical properties dictate the ability of motile cells to traverse limiting pore dimensions (14, 18, 45). Nuclear rigidity and deformability are largely controlled by the nuclear lamina proteins, lamin A and C (products of the LMNA gene), the targeting of which has been demonstrated to significantly decrease nuclear stiffness (46, 47). Therefore, we sought to determine whether LMNA-deleted cells acquire proteinase-independent invasive activity (14, 16, 46, 48, 49). Following LMNA targeting in either HT1080 or MDA-MB-231 cells (SI Appendix, Fig. S15A), cells transmigrate fixed 8.0 µm filter pores comparably to controls, but when confronted with subnuclear 3.0 µm pores, LMNA-targeted cancer cells demonstrate enhanced transmigration potential (SI Appendix, Fig. S15B). Further supporting a decrease in nuclear stiffness, LMNA-targeted cells invading in 3D collagen display enhanced nuclear deformation relative to controls as assessed by nuclear elongation during invasion (SI Appendix, Fig. S15 C and D). However, unlike Transwells, passageways found in collagen hydrogels are not aligned in register, thereby forcing migrating cells to negotiate a more tortuous path. Indeed, in 3D collagen culture, while LMNA knockout HT1080 spheroids display invasive activity in the presence or absence of β1-integrin targeting, invasion is completely blocked following MMP inhibition with comparable results obtained using LMNA-targeted MDA-MB-231 spheroids (SI Appendix, Fig. S15 E and F).

Independent of cancer cell-intrinsic changes in phenotype, alterations in chemotactic signals, matrix composition, fluid viscosity, or alignment of collagen fibrils can modulate migratory activity (50–59). While EGF and SDF1 can serve as potent cancer cell chemoattractants (50, 51), neither stimulus rescues invasive activity in MMP14-null cancer cells (SI Appendix, Fig. S16 A and B). Type I collagen hydrogels impregnated with exogenous fibronectin (52, 53) or collagen VI (54) have each been reported to enhance 3D cancer cell invasion programs with similar activity assigned to increases in fluid viscosity (31). However, under each of these conditions, MMP14-targeted cells remain unable to infiltrate the surrounding hydrogel (SI Appendix, Fig. S16 A and B). Of note, the physical alignment of collagen fibrils, particularly when tension-induced, has been reported to be conducive for proteinase-independent invasion (55, 56, 60). As such, using a mechanical tool to generate tension-induced focal collagen alignment, control or MMP14-targeted spheroids were juxtaposed in the central region of the aligned gels (SI Appendix, Fig. S17A). Control cancer cell spheroids preferentially invade parallel with the aligned fibers, while MMP14-targeted cells remain fixed in place and unable to invade (SI Appendix, Fig. S17 B and C). Finally, invasion-incompetent cancer cells have been reported to interact with accessory cell populations, particularly cancer-associated fibroblasts (CAFs) (52, 57–59), which alternatively remodel the ECM via either mechanical or proteolytic processes, thereby allowing cancer cells to passively infiltrate surrounding tissues in trans. However, in contrast to studies that describe CAF-led tumor cell invasion by generating spheroids containing an admixture of tumor cells and CAFs (52, 57), we sought to more accurately recapitulate cancer cell–stromal cell interactions that occur when invasive tumors infiltrate a CAF-enriched interstitial matrix. To this end, wild-type or MMP14 knockout MDA-MB-231 spheroids were embedded in collagen hydrogels seeded with patient-derived breast CAFs (61). Consistent with their reported invasion-promoting properties, coembedding wild-type spheroids with CAFs significantly enhances the invasive capacity of control cancer cells (SI Appendix, Fig. S17 D and F). However, upon coculture with MMP14-null MDA-MB-231 spheroids, CAFs are incapable of rescuing cancer cell invasion despite the fibroblasts’ ability to locally proteolyze surrounding collagen fibers (SI Appendix, Fig. S17 D–G). Together, these results underline an inescapable reliance on cancer cell MMP14 to drive invasive activity through native collagen barriers.

MMP14-Dependent Maintenance of Nuclear integrity.

In vivo, cancer cells may encounter collagenous barriers across a range of densities with increases in collagen concentrations and rigidity proposed to trigger the generation of more aggressive tumor phenotypes (62–64). As such, cancer cell spheroids were embedded in collagen hydrogels at 1.0 mg/mL, 2.2 mg/mL, and 4.0 mg/mL where matrix rigidity and pore size range from ~20 Pa to ~400 Pa, and from ~1 µm² to ~30 µm², respectively (SI Appendix, Fig. S18 A and B). Despite reports that increases in matrix rigidity induce cancer cells to assume more aggressive phenotypes (62–64), the invasive activity of HT1080 and MDA-MB-231 spheroids decreases as a function of collagen concentration (SI Appendix, Fig. S17 H and I). Further supporting an enhanced barrier effect at increasing collagen concentrations, while cells invading through low density, 1 mg/mL collagen exhibit higher percentages of single-cell invasion, there is a shift to collective tumor cell invasion as collagen concentrations increase (SI Appendix, Figs. S17H and S18C), an observation consistent with reports of the occurrence of cell–cell jamming (13, 65). Furthermore, as reduced invasion in increasingly dense matrix barriers has been associated with a smaller overall spheroid footprint (66–68), we additionally assessed changes related to cell death and proliferation via a Live/Dead Assay and FUCCI reporter system (69). While cells in all collagen concentrations exhibited little to no cell death (SI Appendix, Fig. S19 A and B), proliferation trended downward (by a maximum of ~25%) with increasing collagen density (SI Appendix, Fig. S19 C and D), contributing to the overall reduction in spheroid footprint observed in SI Appendix, Fig. S17H. Interestingly, while invasion of MMP14-null cancer cells is completely abrogated in higher-density collagen gels, significant invasive activity is retained in low-density collagen matrices (SI Appendix, Fig. S17 H and I), a finding supportive of earlier work describing proteinase-independent invasive activity when collagen concentration is lowered (70). Interestingly, despite the apparently dispensable role for MMP14, collagen-infiltrating cancer cells continue to actively degrade pericellular collagen in low-density gels (SI Appendix, Fig. S20). Furthermore, while invading wild-type cancer cells display normal nuclear shapes in either low- or high- density collagen gels, MMP14-null tumor cells invade the low-density matrices in tandem with marked distortions in nuclear shape and position, as indicated by nuclear pinch points and the ratio of front and rear protrusion lengths, coincident with attendant increases in DNA damage, indicated by γH2AX staining (SI Appendix, Fig. S17 J–N). Of note, neither control nor MMP14-null cancer cells display similar changes in 2D culture (SI Appendix, Fig. S21A). Consistent with the premise that in the absence of proteolytic activity even low-density collagen matrices “force” nuclei to undergo significant shape changes, nuclear deformations observed in BB-94-treated or MMP14-null invading cells is strongly associated with increased nuclear envelope rupture, visualized by bursts of GFP-NLS fluorescent signals appearing in the cytoplasmic compartment (SI Appendix, Fig. S17O and Movies S11 and S12). Hence, while MMP14 is not required to support invasion per se in low-density collagen hydrogels, the collagenolytic widening of matrix pores is required to maintain nuclear integrity.

Proteinase Requirements at the Cancer Cell–Stromal Interface of Human Tissue.

As collagen hydrogels cannot fully recapitulate the structural complexity of the native interstitial matrix, we utilized live, human breast tissue explants to assess the relative importance of MMP14 in supporting infiltration of stromal barriers. To avoid creating a damaged passageway through the tissue at an injection site, we used a 2.5D invasion format for these studies with fluorescently labeled MDA-MB-231 cells cultured atop the stromal surface. While control cells readily invaded the tissue over a 6-d culture period, invasion is completely blocked when MMP activity is inhibited with BB-94 or when MMP14 is deleted (Fig. 6 A–C). Hence, when substituting human tissue for type I collagen hydrogels, a requirement for MMP14 activity in driving invasive activity is retained. Given these findings, we next sought to assess human cancer tissues for the presence of MMP14 expression. Using publicly available bulk RNA sequencing data comparing 2,881 normal and 4,484 tumor samples from common solid tumor types (71), MMP14 is consistently detected across the various tissue types and is significantly elevated in tumor tissues (SI Appendix, Fig. S22 A and B). As bulk sequencing does not specifically evaluate tumor cell MMP14 expression or activity, we sought to evaluate MMP14 mRNA and protein expression levels, specifically in invasive breast cancer lesions. Using publicly available spatial transcriptomic data from breast cancer patients (72), we find that the majority of sites positive for the epithelial marker, KRT7, are likewise MMP14-positive (>70%) (Fig. 6D). A metagene identifying MMP14-positive epithelial cells (MMP14+/KRT7+) was used to map areas predicted to contain invasive regions in an unbiased fashion (Fig. 6E). Indeed, regions annotated based on histology as invasive cancer, consistently displayed high MMP14/KRT7 metagene z-scores (Fig. 6F and SI Appendix, Fig. S23, invasive regions outlined in red). Of note, varying levels of the MMP14/KRT7 metagene were also detected in in situ cancer regions present in 3 of the 8 patient samples (Fig. 6F and SI Appendix, Fig. S23, in situ regions outlined in black).

Fig. 6.

Proteinase requirements at the cancer cell–stromal interface of human breast tissue. (A) Cartoon depicting experimental setup for ex vivo human breast tissue invasion assays. (B) Representative immunofluorescence images of cross-sections of human breast tissue samples following 6-d of MDA-MB-231 cancer cell invasion. (C) Quantification of tumor cell invasion into human breast tissue, grouped by individual patient sample (n = 5). Average values for each sample were compared between conditions using a paired t test. (D) Table summarizing the percentages of KRT7-positive spots also positive for MMP14, based on normalized gene expression, within invasive cancer regions in a spatial transcriptomic dataset from 8 patient samples. (E) Spatial plots depicting an example of KRT7 and MMP14 z-scores compared to the z-scores for the combined KRT7/MMP14 metagene. (F) Two representative examples of annotated tissue sections compared to the corresponding KRT7/MMP14 metagene z-score plot. Red outlines depict invasive cancer regions and black outlines depict in situ cancer lesions. (G) Representative immunofluorescent images of normal and malignant human breast tissues, stained for the indicated antibodies. The bottom row depicts sites of colocalization between type I collagen staining and ¾-collagen staining. (H) Quantification of collagen degradation detected in human tissues. Measurements taken from 13 clinical samples with ~10 regions assessed in each sample. P-value < 0.05 (*), <0.01 (**), <0.001 (***), and <0.0001 (****).

Finally, as spatial transcriptomic data are nevertheless limited to detection of MMP14 mRNA, we further aimed to determine whether the joint appearance of MMP14 protein and degraded type I collagen fragments could be found in invasive lesions. In comparisons between normal breast tissue, in situ/noninvasive tumor regions, and frankly invasive lesions, we detect MMP14 in both normal and malignant mammary epithelial cells (Fig. 6G). However, upon staining of the basement membrane protein, type IV collagen, we confirm continuous basement membrane surrounding both normal and noninvasive malignant regions (73), suggesting that epithelial cells in these areas do not establish direct contact with the underlying interstitial matrix (Fig. 6G). However, in areas devoid of basement membrane staining, signaling the emergence of invasive carcinomas (74), MMP14⁺ cancer cells are found in association with large deposits of ¾-collagen degradation products (Fig. 6 G and H). By contrast, in normal breast tissue or noninvasive carcinomas surrounded by an intact basement membrane, type I collagen degradation fragments are seldom observed as confirmed by quantitative analysis (Fig. 6 G and H). Interestingly, as invasive lobular carcinomas (ILCs) have been reported to display more amoeboid-like, single-cell invasion programs (75, 76), we specifically assessed whether MMP14 and ¾-collagen were detected in both subtypes of invasive breast cancer. Indeed, invasion-associated MMP14 and ¾-collagen degradation products are readily detected in both IDC (n = 9) and ILC (n = 3) (Fig. 6G).

Discussion

MMP14/MT1-MMP Drives Mesenchymal and Amoeboid Invasion.

Invasion plasticity imbues cancer cells with the ability to penetrate tissue barriers by reversibly adopting any one of a range of distinct motility programs that do—or do not—rely on proteolytic activity (1, 2, 4, 77, 78). As such, we began our analyses by segregating a single cancer cell line into two discrete populations dominated by either mesenchymal-like collective or amoeboid patterns of invasion. Unexpectedly, the motile fractions of each of these distinct phenotypes expressed a similar cohort of matrix-degrading metalloproteinases while generating matrix-cleared tunnels lined by type I collagen degradation products in a process dependent on MMP activity (36). Further work identified MMP14/MT1-MMP as the dominant, if not sole, pericellular collagenase driving invasive activity. Though MMP14-mediated collagen cleavage was required for invasion and proven to be responsible for almost all collagen degradation, weak residual ¾-collagen staining was detected in MMP14-KO samples. We therefore cannot rule out that other MMPs may participate in type I collagenolysis (e.g., MMP1, MMP13, or MMP15) but nevertheless remain incapable of supporting 3D invasion (35). Though MMP14 is also able to activate a number of secreted MMPs, including MMP2, MMP8, and MMP13, each of these secreted proteinases are sensitive to inhibition by TIMP1 (33), an anti-proteinase that failed to block MMP14-dependent invasive activity despite the use of supraphysiologic concentrations. This result is consistent with earlier work from our group where i) Mmp14-dependent fibroblast invasion in collagen hydrogels was unaffected by deleting Mmp2, Mmp8, Mmp13, or MMP15) and ii) invasion-incompetent COS cells acquire collagen-invasive activity only following transfection with an MMP14 expression vector (35, 79).

Inconsistencies Surrounding Mesenchymal–Amoeboid Transition.

A singular role for MMP14 appears inconsistent with i) the oft-described ability of neoplastic cells to undergo a “mesenchymal*–amoeboid transition” following MMP inhibition (1, 15, 16, 22, 42) ii) the characterization of mechanical processes whereby cancer cells physically tear collagen fibrils in the absence of collagenolytic activity (8, 32), iii) the ability of cells to mobilize a “nuclear piston” to traverse mechanically plastic nanoporous structures (78) and iv) the failure of synthetic MMP inhibitors in clinical trials (80). However, several issues require further consideration.

First, with regard to an amoeboid–mesenchymal switching mechanism, the original model describing this process (15) was subsequently revised with the discovery that cancer cells are unable to adopt an amoeboid phenotype when encountering rigid matrix barriers with subnuclear pore sizes unless proteolytic remodeling was engaged (14). In this regard, a number of recent studies have utilized pepsin-extracted type I collagen matrices to achieve pore sizes permissive for proteinase-independent invasion (12, 13, 18, 22, 40, 42, 45, 81). Using this nonphysiologic material, we likewise find that cancer cell invasive activity is unaffected by either MMP inhibitors or following MMP14 targeting. While these results are consistent with claims that proteolytic requirements for matrix-invasive activity can be bypassed if pore size is sufficiently large, the absence of telopeptides affects the mechanical properties of the assembled hydrogels (82). Whereas native collagen hydrogels behave as viscoelastic solids, pepsin-extracted collagen hydrogels are devoid of covalent crosslinks and assume plastic material characteristics (82, 83). Indeed, when we increased the pore size of hydrogels reconstituted from telopeptide-intact collagen to match that of telopeptide-deleted collagen, proteinase-independent invasion could no longer be maintained. Hence, without increasing mechanical plasticity, even pore sizes far in excess of nuclear dimensions cannot efficiently support proteinase-independent invasion.

Second, a membrane blebbing process has recently been described that allows melanoma cells to invade 3D hydrogels by mechanically fragmenting collagen fibers (8, 32). However, in our work, neither amoeboid fibrosarcoma nor breast carcinoma cells displayed similar activity. It is also worth noting that the bulk of the experiments reported by Driscoll et al. were performed in telopeptide-deleted collagen hydrogels (8). Moreover, while a limited subset of experiments in this report, and others (22, 84), describe proteinase-independent invasion in telopeptide-intact collagen gels, marked variability in the levels of covalent collagen cross-linking can be found in commercially produced collagen preparations. While recent studies have also described the ability of normal as well as neoplastic cells to create migration paths in viscoelastic and/or plastic matrices via mechanical processes alone (8, 78, 82), neither native collagen hydrogels nor live tissue explants display the physical properties necessary to support proteinase-independent invasion. Importantly, our results should not be construed to suggest that specific tissues may not be permissive for proteinase-independent invasion, but rather that type I collagen-rich tissues require the mobilization of MMP14, a finding consistent with our recent studies of breast cancer progression in vivo where local invasion and metastatic activity were obviated when carcinoma cell MT1-MMP was deleted genetically (85).

Finally, the importance of MMP-dependent cancer invasion programs in vivo has long been questioned given the early failure of synthetic MMP inhibitors in clinical trials (86) However, it is now clear that these inhibitors were largely ineffective due to high toxicity and low efficacy (37, 86, 87).

Consideration of Cell Intrinsic and Extrinsic Influences on Protease Dependence During Invasion.

While a number of cell-intrinsic and -extrinsic microenvironmental factors have been reported to modulate cancer cell invasion programs (46, 47, 50–59), none of these elements restored invasive activity to MMP14 knockout cells in intact collagen hydrogels. Perhaps most surprising in these efforts is the fact that CAFs, though capable of promoting the invasive activity of wild-type cancer cells and degrading pericellular type I collagen, did not confer invasive activity to MMP14 knockout cancer cells. However, previous reports i) did not examine the ability of wild-type CAFs to direct the invasive activity of MMP14 knockout cancer cells (52, 57, 58), ii) alternatively generated hybrid spheroids or cell mixtures that contain both cancer cells and CAFs (57, 59), a construct that may not accurately recapitulate cancer cell–stromal cell interactions in vivo or iii) used collagen-Matrigel hydrogels (57–59) where Matrigel interferes with type I collagen fibrillogenesis while displaying highly plastic mechanical properties in its own right (88, 89). Further, wild-type cancer cell spheroids cannot rescue the invasive activity of juxtaposed MMP14 knockout cells, ruling out the possibility of exosome-dependent MMP14 transfer or the release of MMP14-derived promotility factors (90, 91). These observations are additionally supported in our recent work where Mmp14-deleted murine mammary carcinoma cells lost all tissue-invasive potential in vivo, despite the presence of MMP14-positive CAFs in the surrounding stroma (85).

As matrix concentration increases, matrix pore size and rigidity as well as ligand density vary in a coupled fashion. Several elegant bioengineering tools using synthetic hydrogels have been developed to deconvolute the relative effects of each of these physical variables on cell behavior (92–96). In vivo, however, these properties do not change in singular fashion. As type I collagen levels increase or decrease in pathophysiologic settings, attendant changes in matrix density, pore size, and rigidity occur in a mechanically coupled manner, rendering an experimental approach where the collagen concentration is varied in vitro most relevant to the in vivo setting (93). Indeed, though a body of evidence supports the proposition that increased matrix rigidity promotes aggressive characteristics in cancer cells (62–64, 97), we and others find that cancer cell invasive potential decreases markedly as collagen concentration increases, a result consistent with reports that type I collagen acts as a barrier to the spread of neoplastic cells in vivo (85, 98–101). Of note, despite assessing a relatively small range of rigidities (20 to 400 Pa) that remain below that of the more rigid matrices described elsewhere (~1.2 kPa) (63, 98), we nevertheless note a stark contrast in cell behavior attributed to the overall barrier effects of increased collagen density. By contrast, decreasing collagen concentration augmented invasive activity in our studies, providing our first example of MMP14-independent invasion through a crosslinked matrix barrier. In this scenario, however, the migratory capacity maintained in the absence of MMP14-mediated collagenolysis only proceeded at the expense of nuclear integrity. Of note, whereas reports have described loss of lamin B1 and accrual of DNA damage in MMP14-targeted cells in 2D culture (102, 103), we do not detect similar changes in our MMP14 knockout cells under these conditions (SI Appendix, Fig. S21 A and B).

Human Tissue Assays and Patient Data Support MMP14 as a Critical Driver of Cancer Cell Invasion.

Regardless of hydrogel construction, in vitro models cannot recapitulate the complex stromal cell–ECM environment encountered in vivo (10). Using live human tissue explants, however, we confirmed a critical role for MMP14 in conferring breast carcinoma cells with invasive activity. We do note that occasional areas of invasion were observed with MMP14 knockout carcinoma cells or even in the presence of pan-specific MMP inhibitors. However, collagen concentrations may vary in association with neoplastic sites and the deposition of other ECM components, ranging from fibronectin and laminin to fibrin, that are sensitive to a wide range of proteolytic enzymes must be considered (104, 105). Likewise, the ability of cancer cells to locate matrix-poor zones that allow proteinase-independent infiltration over limited distances cannot be ruled out and may provide an explanation for amoeboid-like movement observed in short-term imaging of tumor sites in vivo (15, 106, 107). However, caution should be exercised in terms of ascribing proteinase-free cancer cell movement to earlier descriptions of large, collagen-free pores observed in vivo as assessed by second harmonic generation imaging (108, 109). While this technique can resolve larger diameter type I collagen fibers, small diameter fibrils remain invisible as do most ECM macromolecules save for type I-III collagens (110, 111). Indeed, using nanoparticle diffusion to assess in vivo matrix properties of malignant tissues, pore sizes have been reported to be even smaller than those typically observed in collagen hydrogels generated in vitro (93). Moreover, recent studies demonstrate that fast moving breast cancer cells observed in vivo are devoid of metastatic potential relative to slower moving matrix-degradative carcinoma cells (112).

Finally, in seeking to define a role for MMP14 in the clinical setting, we leveraged a spatial transcriptomic dataset to probe human cancer lesions for MMP14 expression, specifically within invasive regions. These data revealed a strong association between MMP14 expression and invading tumor cell populations. Of note, MMP14 expression was not limited to cancer cells, and could also be detected in other annotated areas containing stromal and immune populations. These data were not unexpected, as multiple cell populations, including fibroblasts (113), endothelial cells (114), and macrophages (115), are capable of expressing MMP14. As these data were limited to MMP14 at the transcript level, we further performed immunohistological analyses of human breast cancer tissues, finding that MMP14⁺ cancer cells are found within invasive lesions decorated with ¾-collagen degradation products. It should be noted, however, that type I collagen degradation products were found in association with MMP14 expression at the protein level only in areas devoid of an intervening basement membrane. Similar to our findings here, we have previously described the constitutive expression of epithelial Mmp14 in the normal mouse mammary gland (85), but the proteolytic activity of the proteinase (i.e., its expression in latent vs active form and the level of endogenous inhibitors) has not been previously determined. Moreover, as active MMP14 can degrade basement membranes (115), it seems most likely that the proteinase is either confined to intracellular compartments, exocytosed in latent form or complexed with endogenous inhibitors (116). In any case, though an association between MMP14 expression and type I collagen degradation does not prove causation, our findings in vivo coupled with our in vitro analyses support the contention that MMP14 is essential for cancer cell invasion programs operative in type I collagen-rich stromal tissues. While cancer cells do exhibit mesenchymal–amoeboid transitions, amoeboid- type invasion remains reliant on MMP14 activity. Presently, inhibiting MMP14 activity specifically in cancer cell populations is not feasible. Nevertheless, identifying the transcriptional programs that drive its expression or the processes that control its exocytosis to the cell surface are deserving of further study.

Materials and Methods

Cell Culture.

HT1080, MDA-MB-231, SUM159, and HEK293T cell lines used in this study were all acquired through ATCC. Detailed methods of cell culture conditions and in vitro assays are available in SI Appendix.

Human Breast Tissue Invasion and Histology.

Deidentified human breast tissue samples were harvested immediately following surgery, suspended in RPMI 1640 (Gibco, 11875135) with penicillin and streptomycin, and shipped on ice by The Cooperative Human Tissue Network (CHTN) to the laboratory. Detailed methods can be found in SI Appendix.

Quantification and Statistics.

For each comparison between two groups, statistical analysis was performed, and P values were calculated with an unpaired two-tailed Student’s t test with Welch’s correction using GraphPad Prism 7 software. Additional details can be found in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Movie S1.

Phase-contrast visualization of invading control HT1080 spheroid from 24 to 48 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S2.

Phase-contrast visualization of invading 4B4-treated HT1080 spheroid from 24 to 48 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S3.

Spinning-disk confocal imaging of control HT1080^GFP-NLS spheroid invasion with Trackmate tracing from 1 hour post embedding to 48 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S4.

Spinning-disk confocal imaging of 4B4-treated HT1080^GFP-NLS spheroid invasion with TrackMate tracing from 1 hour post embedding to 48 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S5.

Phase-contrast visualization of invading control HT1080 spheroid from 1 hour to 72 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S6.

Phase-contrast visualization of invading BB94-treated HT1080 spheroid from 1 hour to 72 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S7.

Phase-contrast visualization of invading control MDA-MB-231 spheroid from 1 hour to 72 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S8.

Phase-contrast visualization of invading BB94-treated MDA-MB-231 spheroid from 1 hour to 72 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S9.

Phase-contrast visualization of invading MMP14-KO HT1080 spheroid from 1 hour to 72 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S10.

Phase-contrast visualization of invading MMP14-KO MDA-MB-231 spheroid from 1 hour to 72 hours post-embedding in acid solubilized rat tail collagen (2.2mg/ml, 37°C, pH 7.4).

Movie S11.

Spinning-disk confocal imaging of control MDA-MB-231^GFP-NLS spheroid invasion from 24 to 48 hours post-embedding in low-density acid solubilized rat tail collagen (1.0mg/ml, 37°C, pH 7.4).

Movie S12.

Spinning-disk confocal imaging of BB94-treated MDA-MB-231^GFP-NLS spheroid invasion from 24 to 48 hours post-embedding in low-density acid solubilized rat tail collagen (1.0mg/ml, 37°C, pH 7.4).

Data, Materials, and Software Availability

Single-cell and bulk RNA sequencing will be deposited in Gene Expression Omnibus (GEO). Previously published data were used for this work [TNMplot.com Data (Large-scale Human Bulk RNA-seq Data) (71) and Spatial Transcriptomic Data (72)]. All other data are included in the manuscript and/or SI Appendix.