Collective invasion can be led by breast cancer cells expressing basal epithelial markers, typified by keratin-14 (KRT14). We analyzed gene expression data from The Cancer Genome Atlas and demonstrated a significant correlation between a KRT14+ invasion signature and a stromal-mediated extracellular matrix (ECM) organization module. We then developed a novel coculture model of tumor organoids with autologous stromal cells. Coculture significantly increased KRT14 expression and invasion of organoids from both luminal and basal murine breast cancer models. However, stromal cell conditioned medium induced invasion but not KRT14 expression. Cancer cells released TGFβ and that signaling pathway was required for stromal cell–induced invasion and KRT14 expression. Mechanistically, TGFβ induced NOX4 expression in stromal cells and NOX4 inhibition reduced invasion and KRT14 expression. In summary, we developed a novel coculture model and revealed dynamic molecular interactions between stromal cells and cancer cells that regulate both basal gene expression and invasive behavior.

Implications:

Fibroblasts within mammary tumors can regulate the molecular phenotype and invasive behavior of breast cancer cells.

Visual Overview:

http://mcr.aacrjournals.org/content/molcanres/18/11/1615/F1.large.jpg.

This article is featured in Highlights of This Issue, p. 1613

Invasion is a fundamental early step in metastasis, which is the dominant cause of cancer mortality. Cancer cells can invade surrounding tissues through a continuum of single-cell amoeboid, single-cell mesenchymal, or collective invasion strategies (1). Recent studies across multiple tumor types suggest that collective invasion is more frequently observed than single-cell invasion in patient tumor samples (2). Collective invasion has also been implicated as the source of circulating tumor cell (CTC) clusters, which have been demonstrated to be more efficient at seeding metastases than single cells (3, 4). Detection of CTC clusters also correlates with poor outcomes in patients (3). Therefore, understanding the molecular mechanisms that regulate collective invasion may help develop therapeutic strategies to limit or prevent invasion and metastases.

Collective invasion in invasive ductal carcinoma (IDC) can be led by specialized cancer cells that retain a basal epithelial phenotype (5). These leader cells are identified by expression of basal epithelial genes (e.g., KRT14 and TP63) and the absence of myoepithelial markers (e.g., ACTA2, MYH11; ref. 5). KRT14+ cells are detected in tumors from diverse breast cancer subtypes and KRT14 expression correlates with poor survival independent of tumor size or hormone receptor status (6). Furthermore, KRT14 expression has been shown in mouse models to be required for both invasion and distant metastasis (4, 5, 7). However, it is increasingly clear that tumor-level phenotypes typically arise through reciprocal interactions among the cancer cells, the extracellular matrix (ECM), and various stromal cell populations (8). Indeed, multiple groups have reported guidance of cancer cell invasion by macrophages (9) and fibroblasts (10–12).

The aim of this study was to identify mechanisms regulating the activation of the KRT14+ basal invasion program. We utilized a systems biology approach and analyzed existing transcriptomic data from human tumors. This analysis identified a group of ECM-related genes that were highly correlated with a breast cancer invasion gene expression signature defined by KRT14+, suggesting that stromal cells could regulate the KRT14+ basal epithelial invasion phenotype. We then developed a novel three-dimensional (3D) culture model to test the effects of autologous stromal cells on both invasion and KRT14 expression.

Bioinformatics analysis

RSEM normalized RNA sequencing (RNA-seq; V2) data with accompanying clinical traits, ABSOLUTE tumor purity measurements, and reverse phase protein microarray (RPPA) signature scores were downloaded from the Ciriello and colleagues study (13). Gene expression data were log2 (normalized counts + 1) transformed, normalized for tumor purity, and z-score scaled. Weighted gene correlation network analysis (WGCNA) was performed using the WGCNA R package as described previously (14). The KRT14+ gene signature was defined using the sum of z-scores for genes significantly upregulated [FDR Padj < 1e-6 and log2 (fold change) > 3] in KRT14+ versus KRT14 tumor cells, as detected by RNA-seq in a previous study (4). Full details of all bioinformatics analyses are provided in the Supplementary Materials and Methods section.

Mouse lines and breeding

Animal use was conducted in accordance with protocols approved by the Johns Hopkins Institutional Animal Care and Use Committee. The FVB-Tg(C3-1-Tag)cJeg/JegJ (C31-Tag), FVB/N-Tg(MMTV-PyVT)634Mul/J (MMTV-PyMT), and B6.129(Cg)-Gt(ROSA)26Sortm4(ACTB-tdTomato,-EGFP)Luo/J (mT/mG) mouse lines were acquired from the Jackson Labs. The TgN(ActbECFP)1Nagy(CK6/ECFP; β-actin::CFP; cyan fluorescent label) transgenic line (15) was acquired from the Hadjantonakis Lab (MSKCC, New York, NY). All lines were backcrossed and maintained on the FVB/NCrl (Charles River Laboratories) background.

Tumor organoid and stromal cell isolation

Tumor organoids were isolated from primary tumors as described previously (5), with the ECM prepared as in ref. 16. Stromal cells were isolated during the differential centrifugations that are used to separate organoids from the single cells. The supernatant from these centrifugation steps was collected and centrifuged at 1,500 rpm for 5 minutes. The cell pellet was resuspended in cell depletion buffer (PBS + 2% FCS + 1 mmol/L EDTA), counted and diluted to 1 × 108 cells/mL. Biotin-conjugated antibodies targeting CD326 (EpCAM; G8.8; 2 μg/mL; Biolegend), CD45 (30-F11; 2 μg/mL; Biolegend), and erythroid cells (TER-119; 1 μg/mL; Stemcell Technologies) were used to deplete epithelial/immune/erythroid cells using the EasySep Mouse Streptavidin RapidSpheres Isolation Kit (Stemcell Technologies), as per the manufacturer's instructions. The efficiency of cell depletion was assessed by flow cytometry using standard protocols. Details of this analysis are provided in Supplementary Materials and Methods.

Cell culture

All cell cultures were maintained in 37°C humidified incubators at 5% CO2. DMEM/F12 supplemented with Gluta-MAX, insulin/transferrin/selenium, and penicillin/streptomycin was used as base media for all cultures. Organotypic cultures were carried out as described previously (5) with some alterations; full details are provided in Supplementary Materials and Methods. For contractility assays, 150,000 stromal cells were embedded in 350 μL 1.5 mg/mL collagen gels, cast in 24-well plates. The gels were incubated for 1 hour to set. 1 mL of base media was added and incubated overnight, then the gels were detached from the wells. After 72 hours (C31-Tag) or 110 hours (MMTV-PyMT), gels were collected and weighed.

Organotypic culture staining and analysis

For end-point analysis at 72 hours, organotypic cultures were fixed, stained, and imaged using a confocal microscope as described previously (5), please see Supplementary Materials and Methods for full details.

Image analysis was carried out in an automated manner using custom image processing macros. Full details are provided in the Supplementary Materials and Methods.

A KRT14+ gene expression signature correlates with genes expressed by stromal cells involved in ECM organization

We previously identified conserved basal protein expression in cells leading collective breast cancer invasion (7). The most conserved marker for invasive leader cells was KRT14 expression and we used RNA-seq to define the broader gene expression differences between KRT14+ and KRT14 cancer cells (6). We now apply this KRT14+ signature to examine RNA-seq data from human IDC samples (The Cancer Genome Atlas; n = 490) using WGCNA (13). WGCNA describes patterns in gene expression, identifying groups of genes (modules) that are highly correlated, to suggest genes that could be functionally related.

Our analysis identified 12 modules corresponding to critical biological processes within these tumors (Fig. 1A; Supplementary Table S1), including previously described features of the IDC intrinsic molecular subtypes (17). For example, a “DNA replication” module was identified, which correlated with RPPA profile scores for proliferation (r = 0.85, P = 3.1e-131) and cell-cycle (r = 0.80, P = 1.1e-107) signatures (Supplementary Table S1). This module was most highly expressed in the basal PAM50 subtype. Increased expression was also seen in Luminal B compared with Luminal A subtypes, consistent with the well-documented variation in proliferative capacity between these subtypes (ref. 17; Supplementary Fig. S1A). Three modules were significantly correlated with estrogen receptor IHC status (r > 0.59, P < 7.0e-47; Supplementary Table S1), including an “Estrogen receptor signaling” module. This module was upregulated in luminal breast cancer subtypes and contained numerous previously described markers of luminal breast cancer (XBP1, GATA3, BCL2, ERB4, and SLC39A6 (ref. 17; Supplementary Fig. S1A).

Figure 1.

WGCNA of primary tumor samples from human invasive ductal breast cancer reveals a highly significant correlation between a KRT14+ tumor cell gene signature and the expression of genes involved in ECM organization. A, Graphical representation of the topological overlap between genes WGCNA where each node (gene) is colored by their module assignment within the network and sized by their connectivity within the assigned module. Modules are labeled by the most prominent gene ontology term enrichment (see Supplementary Table S1 for details). B, Linear Regression analysis comparing the ECM organization module eigengene with KRT14+ signature score. C, Network map (equivalent to that shown in A) where each node is colored to represent correlation (Pearson r) to the KRT14+ gene signature.

Figure 1.

WGCNA of primary tumor samples from human invasive ductal breast cancer reveals a highly significant correlation between a KRT14+ tumor cell gene signature and the expression of genes involved in ECM organization. A, Graphical representation of the topological overlap between genes WGCNA where each node (gene) is colored by their module assignment within the network and sized by their connectivity within the assigned module. Modules are labeled by the most prominent gene ontology term enrichment (see Supplementary Table S1 for details). B, Linear Regression analysis comparing the ECM organization module eigengene with KRT14+ signature score. C, Network map (equivalent to that shown in A) where each node is colored to represent correlation (Pearson r) to the KRT14+ gene signature.

Close modal

We next tested for correlations between our KRT14+ defined basal signature and the WGCNA gene expression modules. This identified a highly significant correlation with the “ECM Organization” module (module eigengene correlation: r = 0.68, P = 5.3e-64; Fig. 1B and C; Supplementary Table S2). This correlation was consistently observed across PAM50 subtypes (Supplementary Fig. S1B). Moreover, the “ECM organization” module correlates with Luminal B, Her2, and Basal PAM 50 subtypes (Supplementary Fig. S1C). Expression of genes in the “ECM Organization” module suggests the intratumoral accumulation of cancer-associated fibroblasts (CAF), as these cells are a prominent stromal component of IDC tumors that both secrete and remodel ECM (18). Furthermore, the module contained numerous previously described CAF markers [ACTA2 (19), FAP (20), NOX4 (21), LOX (22)]. Gene set enrichment analysis (GSEA) also showed significant enrichment of genes associated with invasion-associated desmoplasia, regulated by CAFs (ref. 23; Supplementary Fig. S1D; NES = 2.9, Padj = 1.2e-3). Taken together, our analysis reveals a correlation between the basal epithelial transcriptional program expressed in KRT14+ breast cancer cells and gene expression markers typical of CAF-associated ECM remodeling, across molecular subtypes of breast cancer.

Stromal cells directly enhance collective invasion and KRT14 expression in primary tumor organoids

The correlation between basal gene expression and CAF markers led us to hypothesize that stromal cells could regulate the KRT14+ basal phenotype and thereby enhance cancer invasion. To test this hypothesis, we developed a novel protocol to isolate autologous tumor organoids and stromal cells from primary tumors and combine them in 3D organotypic cultures (Fig. 2A). We note that the fixation and quantification of invasion was done at an earlier time point compared with our past publications, resulting in a relatively lower amount of invasion in the monocultures. This gave us more sensitivity to detect stromal-mediated changes in invasive behavior. Tumor organoids were isolated as previously (5), and concurrently stromal cells were isolated by depleting immune (CD45+), epithelial (EpCAM+), and erythroid (TER119+) cells from the single-cell fraction that is normally discarded during organoid isolation (Fig. 2A). This procedure resulted in approximately 95% stromal cell enrichment, as validated by flow cytometry (Fig. 2B).

Figure 2.

Direct tumor–stroma coculture increases tumor cell invasion and KRT14 expression. A, Schematic describing the experimental set-up. B, Flow cytometry validation of epithelial (EpCAM+) and immune cells (CD45+) depletion from stromal cells. C, Representative confocal images of MMTV-PyMT primary tumor organoids, expressing a membrane Tomato (mT) reporter and stained for KRT14, cultured in collagen gels for 72 hours either in monoculture or in coculture with stromal cells (the segmented organoid outline is shown in white). Scale bar, 50 μm. The associated quantifications of organoid invasion and KRT14 expression are shown below. Statistical testing was performed using a Mann–Whitney test comparing organoids isolated from three tumors, ****, P < 0.0001. D, MMTV-PyMT organoids were cultured either in monoculture, in monoculture with stromal cell conditioned media, or in coculture. Left, Representative confocal images of organoids stained for pan-cytokeratin and KRT14. Scale bar, 50 μm. Quantification of invasion and KRT14 expression from three different tumors is shown on the right. ****, P < 0.0001; ns, not significant (P > 0.05; ANOVA test). Micrographs are displayed as maximum intensity Z-projections of 45 μm in Z. Circ−1, inverse circularity; FC, fold change; MFI, mean fluorescence intensity.

Figure 2.

Direct tumor–stroma coculture increases tumor cell invasion and KRT14 expression. A, Schematic describing the experimental set-up. B, Flow cytometry validation of epithelial (EpCAM+) and immune cells (CD45+) depletion from stromal cells. C, Representative confocal images of MMTV-PyMT primary tumor organoids, expressing a membrane Tomato (mT) reporter and stained for KRT14, cultured in collagen gels for 72 hours either in monoculture or in coculture with stromal cells (the segmented organoid outline is shown in white). Scale bar, 50 μm. The associated quantifications of organoid invasion and KRT14 expression are shown below. Statistical testing was performed using a Mann–Whitney test comparing organoids isolated from three tumors, ****, P < 0.0001. D, MMTV-PyMT organoids were cultured either in monoculture, in monoculture with stromal cell conditioned media, or in coculture. Left, Representative confocal images of organoids stained for pan-cytokeratin and KRT14. Scale bar, 50 μm. Quantification of invasion and KRT14 expression from three different tumors is shown on the right. ****, P < 0.0001; ns, not significant (P > 0.05; ANOVA test). Micrographs are displayed as maximum intensity Z-projections of 45 μm in Z. Circ−1, inverse circularity; FC, fold change; MFI, mean fluorescence intensity.

Close modal

Following isolation, tumor organoids were cultured in collagen gels ± stromal cells for 72 hours (Supplementary Video 1). Individual organoids were imaged by confocal microscopy, then segmented and quantitatively analyzed using a custom image processing macro. Invasion was measured as 1/circularity to reflect the transition from a circular shape in noninvasive organoids to a progressively less circular shape following invasion. KRT14 protein levels within each organoid were quantified using immunofluorescence as a marker for the broader basal epithelium program (4, 5). Coculture with stromal cells significantly increased invasion, KRT14 expression, and the percentage of organoid area positive for KRT14 in MMTV-PyMT organoids (Fig. 2C).

To identify the type of interaction involved in stromal cell–mediated increases in tumor organoid invasion and KRT14 expression, we compared tumor organoids cocultured with stromal cells and organoids in monoculture supplemented with stromal cell conditioned media. Conditioned media supplementation induced a significant increase in invasion compared with monoculture controls but did not induce a significant increase in KRT14 levels (Fig. 2D). Moreover, the magnitude of the increase in invasion induced by coculture was significantly greater than that induced by conditioned media (Fig. 2D). These data suggest that stromal cell induction of KRT14 expression involves juxtacrine signaling, while invasion can be induced by juxtacrine signaling and soluble factors.

Stromal cells are in close physical contact with invasive organoids

Previous research has demonstrated diverse mechanisms by which stromal cells can increase cancer invasion, including paracrine signaling (19), ECM remodeling (22), and directly leading collective invasion (10). To investigate the function of stromal cells within our 3D cocultures, we performed time-lapse differential interference contrast (DIC) microscopy. Visual inspection revealed stromal cells to be highly motile within the collagen I matrix (Supplementary Video 1). They also directly interacted with and remodeled collagen I fibers (Supplementary Video 2). DIC images allowed real-time analysis of stromal cell dynamics; however, without cell type–specific labels, it was difficult to unambiguously determine the location and type of cancer cell–stroma interactions. We therefore isolated organoids from MMTV-PyMT (β-actin::CFP) tumors and cocultured them with stromal cells from MMTV-PyMT (mT/mG; red fluorescent), using mice of comparable age and tumors of comparable size. KRT14 immunofluorescence was imaged by confocal microscopy to quantify the location of stromal cells in relation to the organoids. Stromal cells localized to the bifurcation point between an invasion strand and the body of the organoid (∼80%), adjacent to invading epithelial cells (∼50%) and at the leading edge (∼30%; Supplementary Fig. S2A and S2B). The numbers add to >100% due to the presence of stromal cells in more than one location in a typical organoid. Taken together, the location and behavior of stromal cells in these cocultures suggests that their regulation of invasion may be mediated by cell–cell contacts, in addition to matrix remodeling.

Their close physical proximity led us to consider the possibility of contact-mediated juxtracrine signaling between stromal cells and invasive organoids. The Notch pathway was an attractive candidate pathway both based on its general role in regulating cell fate decisions among adherent cells and the fact that we previously showed that Jag1 (coding NOTCH receptor ligand Jagged-1) was significantly upregulated in KRT14+ cells compared with KRT14 cells (4). The role of NOTCH signaling in the tumor microenvironment is not fully understood but Jag1 has been shown to promote fibroblast activation in prostate cancer (24). Therefore, we hypothesized that NOTCH signaling may be involved in the stromal cell–mediated invasion and KRT14 expression observed in our model system. To test this, we treated organoid monocultures and organoid–stromal cell cocultures with the γ-secretase inhibitor DAPT. DAPT had no significant effect on stromal cell–mediated increases in organoid invasion or KRT14 expression. However, DAPT significantly increased KRT14 expression when organoids were grown in monoculture (Supplementary Fig. S2C). These results suggest that the Notch pathway is not involved in stromal cell–mediated KRT14 expression and invasion but may limit KRT14 expression in the absence of stromal cells. We were therefore motivated to consider the role of other signaling pathways in regulating invasion.

Inhibition of TGFβ-induced NOX1/4 signaling blocks ECM remodeling and prevents the stromal induction of invasion and KRT14 expression

To explore alternative signaling mechanisms, we examined the WGCNA ECM Organization module. GSEA identified significant enrichment of genes upregulated in TGFβ-treated fibroblasts (Supplementary Fig. S3A). Therefore, we examined the role of TGFβ signaling in our coculture model. Inhibition of TGFβ receptor I (TGFβRIi), resulted in a significant decrease in organoid invasion but an increase in KRT14 expression in monoculture (Fig. 3A, consistent with ref. 25). In contrast, TGFβRIi induced a significant decrease in both stromal cell–induced invasion and KRT14 expression (Fig. 3A). These results reveal that TGFβ signaling regulates the ability of stromal cells to induce invasion and KRT14 expression. Further studies will be required to disentangle the distinct contributions of stromal versus epithelial TGFβ signaling on KRT14 expression.

Figure 3.

TGFβ–NOX4–ROS pathway regulates organoid invasion and KRT14 expression. A, MMTV-PyMT organoids, cultured either alone or in coculture with stromal cells, were treated with TGFβRIi at 1 μmol/L or with a vehicle (DMSO) as control. Left, Representative confocal images of organoids invasion (shown by the organoid outline in white) and KRT14 expression. Scale bar, 50 μm. Right, The associated quantifications of invasion and KRT14 intensity from three different tumors. *, P < 0.0332; **, P < 0.0021; ****, P < 0.0001 (ANOVA test). B, Stromal cells were treated with either DMSO as control, TGFβRIi (1 μmol/L), TGFβ1 (2 nmol/L), or the combination of both TGFβRIi (1 μmol/L) and TGFβ1 (2 nmol/L). NOX4 expression was analyzed by immunoblot and β-actin was used as a loading control. The bar graph represents the quantification of NOX4 expression from three independent experiments. *, P = 0.0332; **, P = 0.0021 (ANOVA test). C, MMTV-PyMT organoids, either in monoculture or coculture and treated or not with 40 μmol/L NOX1/4 inhibitor (GKT831). Left, Representative confocal images of organoid invasion (organoid outline in white) and KRT14 expression. Scale bar, 50 μm. Right, Quantification of invasion and KRT14 intensity from three different tumors. ***, P < 0.0002; ****, P < 0.0001 (ANOVA test). D, MMTV-PyMT organoids, either in monoculture or in coculture, were treated with ROS inhibitor (NAC) at 1 mmol/L or with a vehicle (DMSO) as control. Left, Representative confocal images of organoid invasion (shown by the organoid outline in white) and KRT14 expression. Scale bar, 50 μm. Right, Quantification of invasion and KRT14 intensity from three different tumors. ns, not significant (P > 0.05); *, P < 0.0332; ****, P < 0.0001 (ANOVA test). Micrographs are displayed as maximum intensity Z-projections of 45 μm in Z. Circ−1, inverse circularity; FC, fold change; MFI, mean fluorescence intensity.

Figure 3.

TGFβ–NOX4–ROS pathway regulates organoid invasion and KRT14 expression. A, MMTV-PyMT organoids, cultured either alone or in coculture with stromal cells, were treated with TGFβRIi at 1 μmol/L or with a vehicle (DMSO) as control. Left, Representative confocal images of organoids invasion (shown by the organoid outline in white) and KRT14 expression. Scale bar, 50 μm. Right, The associated quantifications of invasion and KRT14 intensity from three different tumors. *, P < 0.0332; **, P < 0.0021; ****, P < 0.0001 (ANOVA test). B, Stromal cells were treated with either DMSO as control, TGFβRIi (1 μmol/L), TGFβ1 (2 nmol/L), or the combination of both TGFβRIi (1 μmol/L) and TGFβ1 (2 nmol/L). NOX4 expression was analyzed by immunoblot and β-actin was used as a loading control. The bar graph represents the quantification of NOX4 expression from three independent experiments. *, P = 0.0332; **, P = 0.0021 (ANOVA test). C, MMTV-PyMT organoids, either in monoculture or coculture and treated or not with 40 μmol/L NOX1/4 inhibitor (GKT831). Left, Representative confocal images of organoid invasion (organoid outline in white) and KRT14 expression. Scale bar, 50 μm. Right, Quantification of invasion and KRT14 intensity from three different tumors. ***, P < 0.0002; ****, P < 0.0001 (ANOVA test). D, MMTV-PyMT organoids, either in monoculture or in coculture, were treated with ROS inhibitor (NAC) at 1 mmol/L or with a vehicle (DMSO) as control. Left, Representative confocal images of organoid invasion (shown by the organoid outline in white) and KRT14 expression. Scale bar, 50 μm. Right, Quantification of invasion and KRT14 intensity from three different tumors. ns, not significant (P > 0.05); *, P < 0.0332; ****, P < 0.0001 (ANOVA test). Micrographs are displayed as maximum intensity Z-projections of 45 μm in Z. Circ−1, inverse circularity; FC, fold change; MFI, mean fluorescence intensity.

Close modal

NADPH oxidase 4 (NOX4) was identified among the WGCNA ECM organization module hub genes and is known to mediate TGFβ's role in converting fibroblasts to tumor-promoting myofibroblasts (21, 26, 27). Therefore, we hypothesized that NOX4 may be involved in the TGFβ-dependent effects on stromal cell–mediated organoid invasion and KRT14 expression. We selected NOX4 for further study based on its significance in this computational analysis, its mechanistic plausibility in regulating key aspects of invasion in this system (25), and the ready availability of antibodies and a potent inhibitor. In support of this hypothesis, we demonstrated that tumor organoids secrete TGFβ (Supplementary Fig. S3B) and that stromal cell expression of NOX4 is TGFβRI dependent, when treated with either recombinant TGFβ or organoid conditioned media (Fig. 3B; Supplementary Fig. S3C). To test the role of NOX4 in stromal cell–mediated invasion and KRT14 expression we treated monocultures and cocultures with a NOX1/4 inhibitor (GKT831). This treatment did not affect monocultures, but significantly reduced the ability of stromal cells to promote invasion and KRT14 expression (Fig. 3C). We repeated this experiment using siRNA targeting NOX4, which produced approximately 55% knockdown of NOX4 expression and significantly reduced organoid invasion (Supplementary Fig. S3D–S3F). CAF/myofibroblast mediated ECM remodeling is known to facilitate tumor cell invasion (22). Therefore, we analyzed whether NOX4 was responsible for stromal cell–mediated collagen remodeling, using a gel contraction assay, which showed that NOX4 inhibition reduced stromal cell contractility (Supplementary Fig. S3G). Those results were also validated in the murine C31-Tag basal breast cancer model (Supplementary Fig. S3G and S3H). NOX4′s primary function is reactive oxygen species (ROS) production (28). Therefore, we treated monocultures and cocultures with the antioxidant N-acetylcysteine (NAC) to determine whether nonspecific ROS inhibition would phenocopy NOX4 inhibition. NAC significantly reduced stromal cell–induced organoid invasion. However, at 1 mmol/L, it had no significant effect on KRT14 expression (Fig. 3D).

In this study, we used WGCNA as a hypothesis-generating platform to suggest mechanisms regulating the KRT14+ basal invasive phenotype. These efforts suggested a role for CAFs, based on correlation between a Krt14+ gene signature and gene expression modules identified in TCGA data. We then used primary 3D organotypic culture to test this hypothesis and to reveal a novel mechanism for stromal regulation of cancer cell invasion. Juxtacrine signaling mediated by stromal cells induces tumor cells to increase activation of a basal invasive program. These observations were facilitated by development of a novel method to coculture tumor organoids and autologous stromal cells in 3D, which enabled tumor–stromal interactions to be analyzed in a physiologically relevant setting. Furthermore, we demonstrated that these protumorigenic effects of stromal cells are sensitive to inhibition of NOX1/4 signaling.

Paracrine signaling from stromal cells has been shown previously to increase tumor cell invasion, with multiple demonstrated molecular mediators of these signaling interactions (e.g., HGF; ref. 19). Prior studies have predominantly utilized transwell assays to monitor tumor cell invasion, which precludes investigation of collective invasion. This technical difference in assays may account for why we observed a more prominent role for juxtacrine signaling and suggests that different mechanisms of tumor-stromal interaction may drive different modes of tumor cell invasion.

Fibroblasts have been shown to lead collective invasion of squamous cell carcinoma cells, by creating tracks through the ECM that cancer cells could not efficiently create on their own (10). Similar fibroblast leader cells have not been described in IDC; instead CAFs have been shown to regulate breast cancer invasion at a distance, through Hippo signaling–dependent remodeling of ECM (12). Our analysis of CAF localization revealed stromal cells at both the leading edge of invasion strands and at the junction connecting the strand to the organoid body, suggesting multiple roles beyond generating tunnels. Our study is the first to demonstrate regulation of KRT14+ basal phenotype by the stromal microenvironment. However, KRT14+ epithelial cells are capable of leading collective invasion in the absence of stromal cells in organoids from both mouse models, suggesting both cancer cell–intrinsic and stromal-dependent mechanisms of induction of the KRT14+ basal phenotype and collective invasion (5). We also acknowledge that our isolation method cannot distinguish the relative contributions of normal versus CAFs to this induction of invasive phenotype and behavior. The autologous coculture model system developed here should prove broadly useful to investigate this signaling interaction further in breast cancer and also to extend these analyses to other solid tumors.

Stromal cells may also regulate the KRT14+ basal phenotype and invasion indirectly, through ECM remodeling (12). Time-lapse imaging revealed stromal cell remodeling of collagen I fibers in cocultures. Invasive KRT14+ tumor islands are also more frequently found in regions with aligned and elongated fibrillar collagen in vivo in mouse models (5). The CAFs most representative of the ECM Organization module, those activated by TGFβ, have also been shown to regulate collagen structure in the stroma of multiple solid tumors (29). Furthermore, collagen cross-linking and alignment correlates with poor prognosis in breast cancer, regulating invasion and metastases (22).

Strategies to therapeutically target cancer invasion are critically needed in multiple cancers. Our study suggests that targeting the fibroblastic stroma through NOX4 inhibition could reduce invasion and therefore metastasis, though further work in animal models is needed. Further analysis of the ECM organization module hub genes (Supplementary Table S2) could also identify alternative candidate targets.

C.J. Hanley reports grants from CRUK during the conduct of the study; in addition, C.J. Hanley has a patent for WO2019086579, use of Nox inhibitors for treatment of cancer (named coinventor) issued, and the NOX1/4 inhibitior (GKT831) was kindly donated by Genkyotex. G.J. Thomas reports nonfinancial support from Genkyotex (supplied NOX1/4 inhibitor GKT137831) during the conduct of the study, in addition, G.J. Thomas has a patent for WO201908657, use of NOX inhibitors for the treatment of cancer issued (coinventor). A.J. Ewald reports grants from NIH/NCI, BCRF/Pink Agenda, and Commonwealth Foundation during the conduct of the study and other from ImmunoCore (spouse is employee), and MedImmune (spouse was formerly an employee), outside the submitted work; in addition, A.J. Ewald has a patent for molecular signatures of invasive cancer subpopulations (US 10545133) issued. No potential conflicts of interest were disclosed by the other authors.

C.J. Hanley: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. E. Henriet: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. O.K. Sirka: Conceptualization, investigation, methodology, writing-review and editing. G.J. Thomas: Conceptualization, resources, supervision, funding acquisition, writing-original draft, writing-review and editing. A.J. Ewald: Conceptualization, resources, supervision, funding acquisition, writing-original draft, writing-review and editing.

The authors thank Joel Bader for helpful comments on the manuscript. C.J. Hanley and G.J. Thomas received support for this project through grants from Cancer Research UK. A.J. Ewald received support for this project through grants from: the Breast Cancer Research Foundation/Pink Agenda (BCRF-19-048), the Commonwealth Foundation, and the NIH/NCI (U01CA217846, U54CA2101732, 3P30CA006973). The results from Fig. 1 are in part based upon data generated by the TCGA Research Network: http://cancergenome.nih.gov/. The GKT831 compound was kindly provided by Genkyotex.

1.
Te Boekhorst
V
,
Preziosi
L
,
Friedl
P
. 
Plasticity of cell migration in vivo and in silico
.
Annu Rev Cell Dev Biol
2016
;
32
:
491
526
.
2.
Bronsert
P
,
Enderle-Ammour
K
,
Bader
M
,
Timme
S
,
Kuehs
M
,
Csanadi
A
, et al
Cancer cell invasion and EMT marker expression: a three-dimensional study of the human cancer-host interface
.
J Pathol
2014
;
234
:
410
22
.
3.
Aceto
N
,
Bardia
A
,
Miyamoto
DT
,
Donaldson
MC
,
Wittner
BS
,
Spencer
JA
, et al
Circulating tumor cell clusters are oligoclonal precursors of breast cancer metastasis
.
Cell
2014
;
158
:
1110
22
.
4.
Cheung
KJ
,
Padmanaban
V
,
Silvestri
V
,
Schipper
K
,
Cohen
JD
,
Fairchild
AN
, et al
Polyclonal breast cancer metastases arise from collective dissemination of keratin 14-expressing tumor cell clusters
.
Proc Natl Acad Sci U S A
2016
;
113
:
E854
63
.
5.
Cheung
KJ
,
Gabrielson
E
,
Werb
Z
,
Ewald
AJ
. 
Collective invasion in breast cancer requires a conserved basal epithelial program
.
Cell
2013
;
155
:
1639
51
.
6.
de Silva Rudland
S
,
Platt-Higgins
A
,
Winstanley
JH
,
Jones
NJ
,
Barraclough
R
,
West
C
, et al
Statistical association of basal cell keratins with metastasis-inducing proteins in a prognostically unfavorable group of sporadic breast cancers
.
Am J Pathol
2011
;
179
:
1061
72
.
7.
Cheung
KJ
,
Ewald
AJ
. 
A collective route to metastasis: seeding by tumor cell clusters
.
Science
2016
;
352
:
167
9
.
8.
Tabassum
DP
,
Polyak
K
. 
Tumorigenesis: it takes a village
.
Nat Rev Cancer
2015
;
15
:
473
83
.
9.
Roussos
ET
,
Condeelis
JS
,
Patsialou
A
. 
Chemotaxis in cancer
.
Nat Rev Cancer
2011
;
11
:
573
87
.
10.
Labernadie
A
,
Kato
T
,
Brugues
A
,
Serra-Picamal
X
,
Derzsi
S
,
Arwert
E
, et al
A mechanically active heterotypic E-cadherin/N-cadherin adhesion enables fibroblasts to drive cancer cell invasion
.
Nat Cell Biol
2017
;
19
:
224
37
.
11.
Gaggioli
C
,
Hooper
S
,
Hidalgo-Carcedo
C
,
Grosse
R
,
Marshall
JF
,
Harrington
K
, et al
Fibroblast-led collective invasion of carcinoma cells with differing roles for RhoGTPases in leading and following cells
.
Nat Cell Biol
2007
;
9
:
1392
400
.
12.
Calvo
F
,
Ege
N
,
Grande-Garcia
A
,
Hooper
S
,
Jenkins
RP
,
Chaudhry
SI
, et al
Mechanotransduction and YAP-dependent matrix remodelling is required for the generation and maintenance of cancer-associated fibroblasts
.
Nat Cell Biol
2013
;
15
:
637
46
.
13.
Ciriello
G
,
Gatza
ML
,
Beck
AH
,
Wilkerson
MD
,
Rhie
SK
,
Pastore
A
, et al
Comprehensive molecular portraits of invasive lobular breast cancer
.
Cell
2015
;
163
:
506
19
.
14.
Langfelder
P
,
Horvath
S
. 
WGCNA: an R package for weighted correlation network analysis
.
BMC Bioinformatics
2008
;
9
:
559
.
15.
Hadjantonakis
AK
,
Macmaster
S
,
Nagy
A
. 
Embryonic stem cells and mice expressing different GFP variants for multiple non-invasive reporter usage within a single animal
.
BMC Biotechnol
2002
;
2
:
11
.
16.
Nguyen-Ngoc
KV
,
Shamir
ER
,
Huebner
RJ
,
Beck
JN
,
Cheung
KJ
,
Ewald
AJ
. 
3D culture assays of murine mammary branching morphogenesis and epithelial invasion
.
Methods Mol Biol
2015
;
1189
:
135
62
.
17.
Yersal
O
,
Barutca
S
. 
Biological subtypes of breast cancer: prognostic and therapeutic implications
.
World J Clin Oncol
2014
;
5
:
412
24
.
18.
Orimo
A
,
Gupta
PB
,
Sgroi
DC
,
Arenzana-Seisdedos
F
,
Delaunay
T
,
Naeem
R
, et al
Stromal fibroblasts present in invasive human breast carcinomas promote tumor growth and angiogenesis through elevated SDF-1/CXCL12 secretion
.
Cell
2005
;
121
:
335
48
.
19.
Marsh
D
,
Suchak
K
,
Moutasim
KA
,
Vallath
S
,
Hopper
C
,
Jerjes
W
, et al
Stromal features are predictive of disease mortality in oral cancer patients
.
J Pathol
2011
;
223
:
470
81
.
20.
Kraman
M
,
Bambrough
PJ
,
Arnold
JN
,
Roberts
EW
,
Magiera
L
,
Jones
JO
, et al
Suppression of antitumor immunity by stromal cells expressing fibroblast activation protein-alpha
.
Science
2010
;
330
:
827
30
.
21.
Hanley
CJ
,
Mellone
M
,
Ford
K
,
Thirdborough
SM
,
Mellows
T
,
Frampton
SJ
, et al
Targeting the myofibroblastic cancer-associated fibroblast phenotype through inhibition of NOX4
.
J Natl Cancer Inst
2018
;
110
:
109
20
.
22.
Levental
KR
,
Yu
H
,
Kass
L
,
Lakins
JN
,
Egeblad
M
,
Erler
JT
, et al
Matrix crosslinking forces tumor progression by enhancing integrin signaling
.
Cell
2009
;
139
:
891
906
.
23.
Kim
H
,
Watkinson
J
,
Varadan
V
,
Anastassiou
D
. 
Multi-cancer computational analysis reveals invasion-associated variant of desmoplastic reaction involving INHBA, THBS2 and COL11A1
.
BMC Med Genomics
2010
;
3
:
51
.
24.
Su
Q
,
Zhang
B
,
Zhang
L
,
Dang
T
,
Rowley
D
,
Ittmann
M
, et al
Jagged1 upregulation in prostate epithelial cells promotes formation of reactive stroma in the Pten null mouse model for prostate cancer
.
Oncogene
2017
;
36
:
618
27
.
25.
Padmanaban
V
,
Krol
I
,
Suhail
Y
,
Szczerba
BM
,
Aceto
N
,
Bader
JS
, et al
E-cadherin is required for metastasis in multiple models of breast cancer
.
Nature
2019
;
573
:
439
44
.
26.
Cucoranu
I
,
Clempus
R
,
Dikalova
A
,
Phelan
PJ
,
Ariyan
S
,
Dikalov
S
, et al
NAD(P)H oxidase 4 mediates transforming growth factor-beta1-induced differentiation of cardiac fibroblasts into myofibroblasts
.
Circ Res
2005
;
97
:
900
7
.
27.
Hecker
L
,
Vittal
R
,
Jones
T
,
Jagirdar
R
,
Luckhardt
TR
,
Horowitz
JC
, et al
NADPH oxidase-4 mediates myofibroblast activation and fibrogenic responses to lung injury
.
Nat Med
2009
;
15
:
1077
81
.
28.
Shiose
A
,
Kuroda
J
,
Tsuruya
K
,
Hirai
M
,
Hirakata
H
,
Naito
S
, et al
A novel superoxide-producing NAD(P)H oxidase in kidney
.
J Biol Chem
2001
;
276
:
1417
23
.
29.
Mellone
M
,
Hanley
CJ
,
Thirdborough
S
,
Mellows
T
,
Garcia
E
,
Woo
J
, et al
Induction of fibroblast senescence generates a non-fibrogenic myofibroblast phenotype that differentially impacts on cancer prognosis
.
Aging
2016
;
9
:
114
32
.