ΔNp63/p73 Drive Metastatic Colonization by Controlling a Regenerative Epithelial Stem Cell Program in Quasi-mesenchymal Cancer Stem Cells

Summary

Cancer stem cells (CSCs) may serve as the cellular seeds of tumor recurrence and metastasis and can be generated via epithelial-mesenchymal transitions (EMTs). Isolating pure populations of CSCs is difficult because EMT programs generate multiple alternative cell states and phenotypic plasticity permits frequent interconversions between these states. Here, we used cell-surface expression of integrin β4 (ITGB4) to isolate highly enriched populations of human breast CSCs and identified the gene regulatory network operating in ITGB4⁺ CSCs. Specifically, we identified ΔNp63 and p73, the latter of which transactivates ΔNp63, as centrally important transcriptional regulators of quasi-mesenchymal CSCs that reside in an intermediate EMT state. We found that the transcriptional program controlled by ΔNp63 in CSCs is largely distinct from the one that it orchestrates in normal basal mammary stem cells and, instead, more closely resembles a regenerative epithelial stem cell response to wounding. Moreover, quasi-mesenchymal CSCs repurpose this program to drive metastatic colonization via autocrine EGFR signaling.

eTOC Blurb

Quasi-mesenchymal cancer stem cells (CSCs) are known to be highly metastatic. Lambert et al. demonstrate that ΔNp63 and p73 are key regulators of the quasi-mesenchymal CSC state. They show that these transcription factors control a regenerate epithelial stem cell program that drives metastatic colonization.

Graphical Abstract

Introduction

Cancer stem cells (CSCs) have been identified in numerous solid malignancies ^1–4 and are thought to serve as the cellular seeds of both post-therapeutic recurrence at sites of primary tumor formation and, following dissemination to distant anatomical sites, of metastatic colonies ^4–8. The epigenetic nature of the controls governing the formation of these cells is indicated by the bi-directional interconversions between CSCs and non-CSCs within tumor cell populations ^9–12.

Variations of the cell-biological program termed epithelial-mesenchymal transition (EMT), drive epithelial cells into a range of more mesenchymal phenotypic states ¹³. We and others demonstrated that activation of the EMT program can lead to substantial increases in stem cells within carcinomas ^14,15 and in normal epithelial tissues ¹⁶. However, demonstration of the precise relationship between the EMT and the stem cell state has been complicated by several factors, among them the various manifestations of the EMT program ¹⁷, which can generate cells residing in a number of intermediate, partially mesenchymal cell states ^13,18–21. Ongoing phenotypic plasticity that enables extensive interconversions between the various intermediate quasi-mesenchymal states ^13,19,22,23, further complicates attempts to isolate and characterize pure populations of CSCs.

Recent studies indicate that CSCs reside within an intermediate, quasi-mesenchymal phenotypic state ^24–28. Indeed, quasi-mesenchymal carcinoma cells have been shown to initiate metastases with high efficiency ²⁹, whereas more fully mesenchymal cells are limited in their colonization capacity and must proceed through some form of a mesenchymal-epithelial transition (MET) in order to form macrometastatic lesions ^30–33. However, it remains unclear precisely how quasi-mesenchymal carcinoma cells differ from their highly mesenchymal counterparts and the underlying mechanisms that enable quasi-mesenchymal CSCs to initiate metastatic colonization are similarly obscure.

Because of the challenges in isolating pure populations of CSCs, the epigenetic configuration and underlying gene regulatory network of metastatic CSCs residing in a quasi-mesenchymal state(s) have yet to be resolved, including the critical regulators that establish and maintain their transcriptional network. In earlier work, the combination of CD44^high/CD24^low cell-surface markers was used to isolate mesenchymal carcinoma cells that exhibit stem cell activity ^14,15, but this mesenchymal cell population is now known to harbor mixtures of cells residing at various points along the EMT spectrum. Recent work from our laboratory described an alternative, improved strategy for isolating populations of CSCs based on the cell-surface expression of integrin β4 (ITGB4) ²⁸. In particular, ITGB4^hi triple-negative human breast cancer cells were found to be highly enriched for CSCs and, at the same time, to display cellular and molecular features indicating their residence in an intermediate, quasi-mesenchymal cell state.

Here, we employed the SUM159 ITGB4^hi human breast cancer cell line as a representative model for quasi-mesenchymal CSCs. This allowed us to define the detailed transcriptional network that supports the CSC state, to assess how this CSC network relates to the corresponding one operating in normal mammary stem cells, and determine how this epigenetic program enables metastatic colonization.

Results

p63 and p73 are transcriptional drivers of quasi-mesenchymal metastatic CSCs

Integrin-β4 (ITGB4; also termed CD104) is a cell-surface protein that can be used to isolate quasi-mesenchymal cancer stem cells (CSCs) ²⁸. In the SUM159 human breast cancer line, ITGB4 can be used as a marker to separate two phenotypically stable cell populations – ITGB4^lo and ITGB4^hi – that exhibit little, if any, interconversion. ITGB4^lo cells were found to be highly mesenchymal and weakly tumorigenic, whereas ITGB4^hi cells were shown to reside in a quasi-mesenchymal state and to readily form tumors upon implantation at limiting dilutions ²⁸. These properties, especially their limited phenotypic plasticity, made these cell populations a tractable experimental model to interrogate the underlying regulatory programs that distinguish quasi-mesenchymal CSCs from more mesenchymal carcinoma cells.

We first mapped the enhancer landscape and measured chromatin accessibility in both cell populations, using H3K27ac ChIP-seq and ATAC-seq, respectively (Figure 1A). The overall abundance and genomic distributions of this histone mark were similar between the ITGB4^lo and ITGB4^hi cells (Supplementary Figure 1A). Nonetheless, we identified 751 H3K27ac peaks that were significantly enriched in ITGB4^hi cells and 561 significantly enriched in ITGB4^lo cells (Supplementary Figure 1B). In order to identify sites of open, accessible chromatin, we employed ATAC-seq analysis ³⁴, which revealed 5206 peaks that were uniquely present in the ITGB4^hi cells and 4832 that were unique to the ITGB4^lo cells (Supplementary Figure 1C).

Figure 1. p63 and p73 are transcriptional drivers of quasi-mesenchymal metastatic CSCs.

(A) Experimental approach used in this study.

(B) Motif enrichment scores across 547 open chromatin regions specific to ITGB4^hi cells.

(C) Enrichment of TP63 and TP73 motifs in the indicated H3K27ac regions.

(D) Clustergram heatmap showing ITGB4^hi- and ITGB4^lo-specific H3K27ac ChIP-seq peaks (left) and p63/p73 binding at these sites in ITGB4^hi cells (right).

(E) Venn diagram comparing p73 and p63 binding sites defined by ChIP-seq.

(F) Average gene expression of H3K27ac-associated genes bound or unbound by p63 (left) or p73 (right) in ITGB4^hi cells.

(G) Differential super-enhancers (SEs) in the ITGB4^hi vs. ITGB4^lo cells. Statistically significant SEs are shown in blue.

(H) ALDH activity in ITGB4^lo and ITGB4^hi cells. Representative flow cytometry plots (above) and quantification (below). Data are ± SEM (n = 3), ** p ≤ 0.01.

(I) Metastatic colonization of ITGB4^lo and ITGB4^hi cells. Representative lung images (top) and quantification (below). Data are ± SEM (n = 3), * p ≤ 0.05. Scale bars, 1 mm.

(J) Immunohistochemistry (IHC) staining for p63 and p73 in representative ITGB4^hi lung metastases. Scale bars, 100 μm.

See also Figure S1.

Integrating the enhancer landscape and chromatin accessibility datasets allowed us to identify potential master regulators of the transcriptional program operating in the quasi-mesenchymal ITGB4^hi cells (Figure 1A). In particular, we focused on a unique set of 547 open chromatin regions, each encompassed within a larger DNA region of H3K27ac-positivity present specifically in the ITGB4^hi cells. We predicted transcription factor (TF) binding events within these regions and then calculated the enrichment of these predicted binding sites relative to all genomic regions marked with H3K27ac. The three most significantly enriched motifs all belonged to members of the p53 family of transcription factors (p53, p63, and p73), which share similar DNA recognition motifs. These motifs were enriched nearly 4-fold above the next most-enriched TF motif and were enriched exclusively in ITGB4^hi-specific H3K27ac peaks (Figure 1B and 1C).

Since the SUM159 cells used in these analyses harbor a mutation in the DNA-binding domain of TP53 that is thought to impair its ability to bind DNA ³⁵, we focused our subsequent experiments on the actions of p63 and p73. Of note, p63 is known to function as a key regulator of adult stem cells in stratified epithelial tissues ³⁶, including those in the mammary gland, where it is expressed by basal epithelial cells ³⁷.

Transcript and protein levels of p63 and p73 were higher in ITGB4^hi cells relative to ITGB4^lo cells (Supplementary Figure 1D and 1E). Most ITGB4^hi cells stained positive for both markers in their nuclei, while ITGB4^lo cells had a smaller fraction of cells that were p73-positive and essentially lacked detectable p63 (Supplementary Figure 1F). Therefore, the majority of cells in the ITGB4^hi compartment associated with CSC activity co-expressed p63 and p73.

We next performed ChIP-seq analysis for p63 and p73 in the ITGB4^hi cells. Consistent with the motif enrichment analysis, there was a marked enrichment of both p63 and p73 binding at enhancers and promoters in ITGB4^hi cells (Figure 1D), with 49% of ITGB4^hi-specific H3K27ac peaks being bound by p63 and 57% bound by p73 (Supplementary Figure 1G); a majority of these sites (64.3%) were bound by both TFs (Figure 1E). p73 was bound to distinct loci in ITGB4^lo cells compared to ITGB4^hi cells (Supplementary Figure 1H). There was no enrichment of p63 or p73 at H3K27ac peaks that were specific to the ITGB4^lo cells or in the regions that were shared in common between the two cell populations (Supplementary Figure 1G). In ITGB4^hi cells, p63 and p73 bound primarily to intergenic enhancer regions and, to a lesser extent, proximal and distal promoters (Supplementary Figure 1I). The genes associated with p63 and p73 binding in ITGB4^hi CSCs were expressed at significantly higher levels in these cells relative to other H3K27ac associated-genes not bound by these factors (Figure 1F), suggesting that p63/p73 were functioning primarily as transcriptional activators.

Super-enhancers (SEs) have been implicated in the control of cell-type specific transcriptional programs ^38–40. Using the H3K27ac ChIP-seq data, we mapped SEs that were activated in ITGB4^lo and ITGB4^hi cells (see Methods). The TP73 locus contained one of the few SEs specifically activated in ITGB4^hi cells and the TP63 locus was marked by a strong SE as well (Figure 1G). Moreover, we also found that p63 was bound, among other sites, to the enhancer site of its own encoding gene as well as to the gene encoding p73; a similar pattern of behavior applied to the p73 gene, with p73 binding at the TP63 and TP73 loci being particularly strong (Supplementary Figure 1J). Hence, p63 and p73 appeared to function in auto-regulatory positive-feedback loops, mirroring patterns of autoregulation that are characteristic of other cell type-specific master transcription factors (Supplementary Figure 1K). Collectively, these results implicated p63/p73 in the regulation of the chromatin landscape and the resulting transcriptional program of ITGB4^hi quasi-mesenchymal CSCs.

Aldehyde dehydrogenase (ALDH) activity has been proposed as a marker of CSCs that reside closer toward the epithelial end of the EMT spectrum ^41,42. Indeed, we found that the ITGB4^hi cells contained ~14-fold more ALDH-positive cells than ITGB4^lo cells (Figure 1H). We sorted ITGB4^hi cells on the basis of ALDH activity and, in freshly prepared protein lysates, found higher levels of p63 and the EMT-TF Snail in ALDH⁺ cells compared to the ALDH⁻ population (Supplementary Figure 1L). However, extensive plasticity among the ALDH⁻ and ALDH⁺ subpopulations in ITGB4^hi cells precluded further functional comparisons.

Since CSCs have been implicated in metastatic progression, we tested the ability of ITGB4^hi and ITGB4^lo cells to spawn lung metastases after tail-vein injection, which gauges the ability of cells to complete the later stages of the invasion-metastasis cascade, including survival in the circulation, extravasation, and the subsequent proliferative expansion of metastatic colonies. Only ITGB4^hi cells were able to readily form macrometastatic colonies (Figure 1I). The metastases formed by the ITGB4^hi cells stained strongly positive for both p63 and p73 (Figure 1J). Hence, ITGB4^hi cells represented a population of quasi-mesenchymal CSCs with a strongly elevated metastatic colonization ability and possessed a unique chromatin landscape that we speculated was coordinated in large part by the actions of the TF(s) p63 and/or p73.

ΔNp63 and p73 are Required to Maintain CSCs in a Quasi-mesenchymal Cell State

The TP63 and TP73 genes each encode two main alternative isoforms: a full-length version carrying an intact N-terminal transactivation domain (TA) and a truncated form lacking the N-terminal TA domain (ΔN). First, we determined which specific isoforms of p63 and p73 were expressed in the ITGB4^hi cells, since the different isoforms are known to exert different functional effects ⁴³. Using isoform-specific qPCR⁴⁴, we confirmed that ΔNp63 was the major isoform expressed by ITGB4^hi CSCs (Supplementary Figure 2A), similar to the known role of this isoform in the stem cells of normal stratified epithelial tissues ³⁶. In contrast, the ITGB4^hi cells expressed two predominant p73 isoforms that could be readily resolved by western blot analysis and, we suspected, represented the expression of the TA and ΔN versions of p73 (Supplementary Figure 1E). This use of CRISPR/Cas9 isoform-specific gene knockouts supported this notion (Supplementary Figure 2B).

Compared to a non-targeting (NT) control, knockout of ΔNp63, of p73, or both (Supplementary Figure 2C) led to a clear shift of the quasi-mesenchymal ITGB4^hi cells toward the ITGB4^lo highly mesenchymal state (Figure 2A). Knockout of the larger p73 isoform using sgRNAs directed at exon 2 or 3, or complete knockout of both p73 isoforms using sgRNAs targeting the shared exon 4 had similar effects on ITGB4 status (Supplementary Figure 2D). Hence, in all subsequent experiments we proceeded with the p73 sgRNA that targeted the exon shared by both isoforms of this protein.

Figure 2. Loss of ΔNp63 and p73 impairs the post-extravasation proliferation of CSCs.

(A) Representative flow cytometry plots in the indicated ITGB4^hi CRISPR-KO cells.

(B) Western blot analysis of the indicated proteins in control or KO cells that were successively sorted for loss of ITGB4 expression.

(C) Quantification of ALDH-positive cells in ITGB4^hi control (NT) or KO cells. Data are ± SEM (n = 3), ** p ≤ 0.01, *** p ≤ 0.001.

(D) Metastatic colonization of ITGB4^hi control or KO cells. Representative lung images (left) and quantification (right). Data are ± SEM (n = 10), *** p ≤ 0.001, **** p ≤ 0.0001. Scale bars, 1 mm.

(E) IHC staining for ΔNp63 and p73 in metastatic lung lesions. Scale bars, 100 μm.

(F) Quantification of luminescence after injection of ITGB4^hi NT or dKO cells (n=8, days 0–2, n=4, days 3–10) and representative IVIS images (right).

(G) Representative IF staining for tdTom and CD31 in lung cryosections and quantification of extravasated ITGB4^hi NT (n=29 cells from 2 mice) and dKO cells (n=28 cells from 2 mice) 48 hours after tail-vein injection (right). Scale bars, 20 μm.

(H) Representative IF staining for tdTom and Ki67 in lung cryosections and quantification of Ki67-positive cells 48 hours or 10 days after injection of ITGB4^hi NT or dKO cells. 48hr: ITGB4^hi NT (n=41 cells from 2 mice), dKO (n=79 cells from 2 mice). 10 days: ITGB4^hi NT (n=166 cells from 2 mice), dKO (n=26 cells from 2 mice). Scale bars, 20 μm.

(I) Representative IF staining for tdTom and CD31 in lung cryosections 10 days after tail-vein injection of ITGB4^hi dKO cells. Scale bar, 20 μm.

(J) GSEA of single-cell RNA-seq data comparing ITGB4^hi NT and dKO cells isolated from the lungs of mice 3 days after tail-vein injection. Positive enrichment indicates higher expression in ITGB4^hi dKO cells.

(K) t-SNE plot of tdTom-positive ITGB4^hi NT and dKO cells isolated from the lungs.

Corresponding plots show expression of genes of interest.

See also Figure S2 and S3.

Since the CRISPR/Cas9 knockout cells were generated in a pooled fashion, we suspected that these cell populations harbored a mix of cells that had undergone various degrees of gene knockout. Indeed, when we sorted the knockout populations on the basis of ITGB4 expression, we found that the cells that persisted in the ITGB4^hi state retained expression of ΔNp63 and/or p73 (Supplementary Figure 2E), indicating that these cells had escaped CRISPR-mediated knockout. We therefore sorted for the ITGB4^lo fraction to enrich for cells with knockout of ΔNp63 and/or p73 in both the single- and double-knockout (dKO) cell lines (Figure 2B) and used these sorted cells for all subsequent experiments.

Consistent with the binding of ΔNp63 and p73 to their respective enhancers, we found that knockout of either TF led to a corresponding decrease in the expression of the other, although this effect was far more pronounced with knockout of p73, which led to undetectable ΔNp63 protein expression (Figure 2B). This suggested that, at least in the ITGB4^hi cells, p73 lies upstream of ΔNp63, acting as a potent activator of ΔNp63 transcription.

Since ITGB4 is a known transcriptional target of ΔNp63 and p73 ^45,46, we undertook to determine if the shift toward the ITGB4^lo cell-surface phenotype observed following knockout of ΔNp63/p73 was truly indicative of a change in cell state rather than simply a loss of expression of a direct transcriptional target. In fact, ITGB4^hi cells that had suffered knockout of either ΔNp63 and/or p73 became highly mesenchymal in nature, as gauged by their morphology (Supplementary Figure 2F), the high-level expression of canonical EMT markers, and increased expression of the EMT-inducing transcription factor Twist (Supplementary Figure 2G). Moreover, the ALDH⁺ population was significantly decreased in the knockout lines, mirroring the abundance of this population in the highly mesenchymal ITGB4^lo state (Figure 2C and Supplementary Figure 2H). Notably, loss of ΔNp63 and/or p73 had no effect on cell proliferation in 2D culture (Supplementary Figure 2I) and did not compromise the ability of these cells to grow as tumorspheres (Supplementary Figure 2J). Altogether, these observations further substantiated the notion that ΔNp63/p73 are essential to maintain residence of these breast carcinoma cells in a quasi-mesenchymal state and that, in their absence, CSCs spontaneously lapse into a highly mesenchymal phenotypic state with associated loss of stemness.

ΔNp63 and p73 are Required for the Metastatic Ability of Cancer Stem Cells

We proceeded to test the relative abilities of the various cell populations to colonize the lungs in experimental metastasis assays. Whereas control ITGB4^hi NT cells were highly efficient in generating metastatic colonies, cells lacking ΔNp63 and/or p73 were severely compromised in their ability to form metastases (Figure 2D).

Cells that ostensibly suffered knockout of p73 could still form an appreciable number of macrometastases. However, these metastases invariably stained positive for ΔNp63 (Figure 2E), suggesting that the growths derived from cells that had somehow re-activated ΔNp63 even in the absence of p73, which remained undetectable by IHC in the metastatic lesions (Figure 2E). These data indicated that while p73 could act upstream of ΔNp63, it was the downstream actions of ΔNp63 that appeared be the critical governor of ongoing residence of cells in the CSC state, enabling metastatic outgrowth in the lung.

We next determined with greater precision which of the multiple steps of metastasis formation depended upon ΔNp63/p73 in the experimental metastasis assay used above. We used IVIS bioluminescent imaging to track ITGB4^hi NT and dKO cells immediately after tail-vein injection (30 minutes post-injection) and followed their fate for 10 days thereafter. The abundance of NT and dKO cells in the lung was comparable between days 1 and 3, with a reduction in luminescence in the dKO group only being evident at day 7 (Figure 2F). These results were supported using flow cytometry to estimate the number of tdTom⁺ cells in the lungs at day 2 and day 10 (Supplementary Figure 3A and 3B).

Immunofluorescent staining 2 days after injection revealed the majority of both NT (75.9%) and dKO (82.1%) cells had successfully extravasated (Figure 2G). These results echoed the in vitro behavior of these cells, as cells with KO of ΔNp63 and/or p73 exhibited comparable, or even slightly enhanced, rates of invasion in vitro through endothelial cells layers (Supplementary Figure 3C). These data demonstrated that dKO cells seeded the lung parenchyma with comparable efficiency as the control NT cells, and focused our attention instead on the post-extravasation behavior of such cells.

We assessed the proliferative potential of NT control or dKO cells after seeding the lung by quantifying the number of Ki67-positive tdTom⁺ cancer cells at 2- and 10-days post-injection. Two days after injection, the numbers of Ki67-positive cells were similar in NT and dKO groups (34.1% vs. 32.9%, respectively), but by 10 days only 7.7% of dKO cells stained positive for Ki67 whereas 20.5% of NT cells were positive (Figure 2H). We did not detect significant levels of apoptosis, as assessed by cleaved caspase 3 staining, in either group at days 2 or 10 (Supplementary Figure 3D). These observations indicated that the ability of carcinoma cells to sustain proliferation in the lung after extravasation, a process that lies at the heart of metastatic colonization, was compromised in CSCs lacking ΔNp63/p73. Importantly, this pronounced metastatic defect could not be attributed simply to the loss of surface ITGB4 expression in the KO cells because overexpression of ITGB4 in parental ITGB4^lo cells did not significantly alter their colonization potential (Supplementary Figure 3E).

We noted that occasional dKO cells were capable of entering a dormant state, becoming closely affiliated with and spreading along the vascular capillaries of the lung (Figure 2I), a behavior that has previously been associated with metastatic dormancy ⁴⁷. These cells could persist within the lung for extended periods of time, being detectable even 5 weeks after injection (Supplementary Figure 3E). Single-cell RNA-sequencing of NT and dKO cells isolated from the lungs 3 days after injection revealed the hallmark EMT gene set to be the top positively enriched gene set in the dKO cells (Figure 2J). Thus, in the parenchymal microenvironment of the lung, dKO cells maintained their highly mesenchymal nature, as evidenced by strong expression of the Twist1 and Prrx1 EMT-TFs together with the absence of cytokeratin 14 expression (Figure 2K). Taken together, these results supported the notion that CSCs lacking ΔNp63/p73 remain trapped in highly mesenchymal cell state in which cell proliferation within the lung parenchyma cannot be actively maintained and results, instead, in a small fraction of these carcinoma cells entering into metastatic dormancy.

Wishing to extend and refine these findings, we used NAMEC8 cells, another model of triple-negative breast cancer ^28,48. These cells, which reside in a mesenchymal state, derive from genetically defined, experimentally immortalized human mammary epithelial cells ⁴⁹. We proceeded to transform the NAMEC8 cells with oncogenic Ras and further engineered them to express a doxycycline-inducible version of ΔNp63. We then sorted the ITGB4^lo fraction of NAMEC8 cells using FACS (Supplementary Figure 3G) and induced ΔNp63 expression with doxycycline treatment.

Induced expression of ΔNp63 in ITGB4^lo NAMEC8 cells led to a clear shift toward the ITGB4^hi state compared to control treated cells (Supplementary Figure 3H). Importantly, while control-treated ITGB4^lo NAMEC8 cells were unable to generate macroscopic metastases, expression of ΔNp63, which we maintained in vivo, endowed these cells with a strongly enhanced capacity for metastatic colonization (Supplementary Figure 3I). These observations using genetically defined human breast cancer cells and experimentally induced ΔNp63 expression directly reinforced the notion that ΔNp63 is a key determinant of residence in the quasi-mesenchymal phenotypic state, which in turn is associated with strong increases in stemness as indicated by the greatly enhanced ability of cells to launch metastatic colony formation.

ΔNp63/p73 are required for maintenance of the CSC enhancer landscape and transcriptome

We next examined how loss of these TFs impacted the chromatin landscape and gene regulatory program of ITGB4^hi CSCs. Enhancer mapping in control and KO cells using H3K27ac ChIP-seq showed that KO of ΔNp63, p73 or both, led to a collapse of the ITGB4^hi CSC-specific enhancer landscape with a corresponding activation of ITGB4^lo-specific enhancer elements (Figure 3A and 3B). This resulted in large-scale changes to the transcriptome with 483 genes coordinately up-regulated and 358 genes down-regulated across the three knockout cell lines (Figure 3C). Consistent with our previous observations, gene set enrichment analysis (GSEA) of RNA-seq data revealed that the top gene signatures positively enriched in the ΔNp63/p73 knockout lines were associated with activation of a strong EMT program (Figure 3D). Interestingly, however, the other gene sets negatively enriched upon ΔNp63/p73 KO were associated with various inflammatory signatures (Figure 3D). Hence, in addition to influencing mesenchymal polarization and stemness, ΔNp63/p73 appeared to control the expression of a number of inflammation-associated genes in ITGB4^hi CSCs.

Figure 3. ΔNp63 regulates a distinct transcriptional program in CSCs and normal basal cells.

(A) Clustergram heatmap showing ITGB4^hi- and ITGB4^lo-specific H3K27ac ChIP-seq peaks (left) in the indicated ITGB4^hi CRISPR KO cells (right).

(B) Metagene plot showing H3K27ac ChIP-seq reads in parental ITGB4^lo and ITGB4^hi cells (left) and ITGB4^hi cells with the indicated CRSIPR KOs (right).

(C) Heatmap of the top differentially expressed genes (log₂ fold-change ±1, p ≤ 0.01) in control (NT) ITGB4^hi cells compared to the indicated KO lines (left). Venn diagram of genes downregulated across the three KO cell lines (right).

(D) GSEA based on RNA-seq data comparing ITGB4^hi NT and dKO cells. Positive enrichment indicates higher expression in ITGB4^hi dKO cells.

(E) Venn diagram showing ΔNp63 target genes in ITGB4^hi cells and MCF10A cells ⁵⁴. Top gene ontology (GO) terms of the ITGB4^hi-specific ΔNp63 targets are shown below.

(F) Venn diagram showing ΔNp63 target genes in ITGB4^hi cells and two subpopulations of CD10⁺ myoepithelial/basal ⁵⁵.

(G) GSEA of RNA-seq data comparing ITGB4^hi NT and dKO cells using the indicated epithelial wounding signatures.

See also Figure S4.

ΔNp63 regulates a distinct transcriptional program in CSCs and normal basal cells

Of great interest is the biological origin of CSCs: do carcinoma cells assemble a stem-cell program de novo that is unique to neoplastic cells or do they simply exploit and adapt with minor revisions a stem-cell program operative in the corresponding normal stem cell compartment? The observations above suggested that quasi-mesenchymal CSCs might rely on a transcriptional program that is closely related to that of normal epithelial stem cells since, as mentioned previously, ΔNp63 is also a key transcriptional regulator of untransformed epithelial stem cells, including those in the mammary gland ³⁷.

On a global level, through use of the Epigenome Roadmap data ⁵⁰, we found that the enhancers activated in quasi-mesenchymal ITGB4^hi CSCs overlapped with H3K27ac regions associated with normal keratinocytes, mammary epithelial cells, and esophageal cells (Supplementary Figure 4A and 4B), in line with the known role of ΔNp63 in these cell types. In contrast, the enhancers that were activated in the highly mesenchymal ITGB4^lo non-CSCs overlapped with those found in prototypical mesenchymal cell types, such fibroblasts, mesenchymal stem cells, and osteoblasts (Supplementary Figure 4A and 4B). Hence, CSCs residing in a quasi-mesenchymal cell state, retain at least some of the enhancers actively operating in the corresponding normal tissue, whereas carcinoma cells that have progressed toward more extreme mesenchymal states shed these enhancers and instead activate enhancers associated with bona fide mesenchymal cells. Moreover, KO of ΔNp63 and/or p73 in ITGB4^hi cells led to reduced expression of a number of genes that have been implicated in basal mammary stem cells, including SLUG ¹⁶, PROCR ⁵¹, DLL1 ⁵², and BCL11B ⁵³ (Supplementary Figure 4C), suggesting, that certain parallels exist between the ΔNp63-driven transcriptional programs operating in quasi-mesenchymal CSCs and in normal basal stem cell.

Nonetheless, we wished to make a more precise assessment of the target genes controlled by ΔNp63 in normal and neoplastic mammary stem cells. Thus, we compared the genomic binding locations of ΔNp63 in ITGB4^hi cells with its previously mapped binding locations in non-neoplastic, immortalized MCF10A human mammary basal cells ⁵⁴. Although MCF10A cells may represent an imperfect basal cell model, this allowed us to define high-confidence ΔNp63 target genes in both cell types by employing two criteria: (1) ΔNp63 ChIP-seq peaks located near a gene (−11kb to +5.1kb of a TSS) and (2) gene downregulation upon loss of ΔNp63 expression (FDR <0.05 and log₂ fold-change >1.5).

We found that ΔNp63 controlled an almost entirely distinct set of target genes in the quasi-mesenchymal CSC population when compared to those bound in the untransformed MCF10A mammary epithelial cells (Figure 3E). ITGB4^hi-specific ΔNp63-bound target genes in the breast cancer cells were enriched for pathways corresponding to TNF-α signaling, developmental biology, and cytokine signaling (Figure 3E). Using previously published gene expression data from murine mammary epithelial cells ³⁷, we confirmed that a representative selection of ITGB4^hi-specific ΔNp63 targets from the top scoring gene sets were not more highly expressed in basal cells compared to luminal cells (Supplementary Figure 4D).

This suggested that many of the genes regulated by ΔNp63 in CSCs are not targets of this TF in normal basal cells. Further support for this notion was provided by an additional comparison of ΔNp63 targets in ITGB4^hi cells with previously defined ΔNp63 targets in CD10⁺ myoepithelial/basal cells isolated from normal human breast tissues ⁵⁵. Here, too, we observed minimal overlap of ΔNp63 target genes between ITGB4^hi CSCs and either the CD10⁺CD44⁻ and CD10⁺CD44⁺ subpopulations (Figure 3F). Of note, CD10⁺CD44⁺ cells were shown to have high ALDH activity and potential progenitor activity ⁵⁵; still there were few shared ΔNp63 target genes between this subpopulation of myoepithelial/basal cells and ITGB4^hi CSCs. Altogether, this indicated that although quasi-mesenchymal ITGB4^hi CSCs exhibit some resemblance to their normal stem cell counterparts in terms of the overall enhancer landscape, the specific transcriptional circuits controlled by ΔNp63 in normal mammary stem cells and CSCs are actually quite distinct.

We were intrigued by the various inflammatory and cytokine gene signatures that were prominent features of the ΔNp63-controlled transcriptional program in ITGB4^hi carcinoma cells, suggesting to us that CSCs might adopt a transcriptional program associated with wounding and regeneration. Indeed, there was strong overlap between the genes downregulated in ITGB4^hi cells upon ΔNp63/p73 KO and transcriptional signatures associated with wounding and regeneration across various epithelial tissues. These signatures included those associated with the response of epidermal cells to skin wounding ⁵⁶, with a transitional cell state linked with alveolar regeneration following lung injury ⁵⁷, and with a damage-induced “revival” stem cell population in the intestine ⁵⁸ (Figure 3G). In experimental models of epidermal wound healing, spatially distinct cellular regions emerge that include a migratory “leading edge” adjacent to the wound as well as a lagging “proliferative hub” of activated stem cells ⁵⁹. Interestingly, ITGB4^hi CSCs manifested hybrid transcriptional features of both the leading edge and proliferative hub that were dependent on ΔNp63 expression (Supplementary Figure 4D). Hence, the transcriptional program orchestrated by ΔNp63 in CSCs closely resembles a regenerative epithelial response to tissue damage in contrast to the tonic stem cell program expressed by basal mammary epithelial cells.

ΔNp63 is sufficient to activate the CSC transcriptional circuit

Given the apparent involvement of ΔNp63/p73 in maintaining residence in the CSC state, we next sought to determine if ΔNp63 and/or p73 was sufficient to establish the CSC gene regulatory network in non-CSCs. Accordingly, we forced expression of ΔNp63, TAp73, or both TFs, in ITGB4^hi ΔNp63/p73 dKO cells using CRISPR-resistant expression vectors to do so (Figure 4A).

Figure 4. ΔNp63 is sufficient to activate the CSC transcriptional circuit.

(A) Western blot analysis of ΔNp63 and p73 in ITGB4^hi cells ectopically expressing the indicated vectors.

(B) Clustergram heatmap showing ITGB4^hi- and ITGB4^lo-specific H3K27ac ChIP-seq peaks in the indicated ITGB4^hi dKO overexpression cells.

(C) Metagene plot showing H3K27ac ChIP-seq peaks in the indicated ITGB4^hi dKO overexpression cells.

(D) Flow cytometry analysis of the indicated ITGB4^hi dKO overexpression cells.

(E) Volcano plots showing genes differentially expressed between ITGB4^hi dKO cells expressing the indicated overexpression vectors compared to cells expressing an empty vector (EV). Genes with significant expression changes are highlighted in blue (down) or red (up).

(F) Venn diagram showing the genes downregulated in all ITGB4^hi KO lines and those upregulated upon re-expression of ΔNp63 in dKO cells. (log₂ fold-change ±1, p ≤ 0.01).

(G) Western blot analysis of ΔNp63 and p73 in ITGB4^lo cells ectopically expressing the indicated vectors.

(H) Clustergram heatmap showing ITGB4^hi- and ITGB4^lo-specific H3K27ac ChIP-seq peaks in control or ΔNp63-expressing ITGB4^lo cells.

(I) Metagene plot showing H3K27ac ChIP-seq peaks in control or ΔNp63-expressing ITGB4^lo cells.

(J) Volcano plot showing genes differentially expressed between ITGB4^lo ΔNp63 and control EV cells.

(K) Flow cytometry analysis of control or ΔNp63-expressing ITGB4^lo cells.

(L) ALDH activity in ITGB4^lo cells expressing a doxycycline-inducible ΔNp63 expression vector.

(M) Representative lung images after injection of ITGB4^lo cells expressing a doxycycline-inducible ΔNp63 expression vector (left) and quantification of macrometastatic lesions (right).

Data are ± SEM (n = 3 in −dox group and n=4 in +dox group). Scale bars, 1 mm.

Forced expression of ΔNp63 alone, but not TAp73, was sufficient to re-establish many of the ITGB4^hi enhancer elements that were lost after ΔNp63/p73 dKO (Figure 4B and 4C). Moreover, forced expression of ΔNp63 on its own could switch cells to the ITGB4^hi state and led to marked transcriptional changes, whereas expression of TAp73 essentially had no effect on the transcriptome of the dKO cells (Figure 4D and 4E). This demonstrated that ΔNp63, although it apparently lacks a transactivation domain, can nonetheless drive the transcriptional activation of CSC target genes. Of the 109 genes that became significantly transcriptionally activated upon expression of ΔNp63 in dKO cells, 73.4% of these were genes coordinately downregulated in the three knockout lines (Figure 4F). Also, concomitantly induced expression of ΔNp63 and TAp73 in dKO cells induced changes that were comparable to those resulting from ΔNp63 expression alone, indicating that the observed transcriptional effects of dual ΔNp63/p73 overexpression were mediated almost entirely by ΔNp63.

Forced expression of ΔNp63 in ITGB4^lo cells had similar, albeit less dramatic effects on the enhancer and transcriptional network as well as conversion to the ITGB4^hi state (Figures 4G–K). Notably, ITGB4^lo cells expressing a doxycycline-inducible version of ΔNp63 harbored a higher percentage of ALDH⁺ cells (Figure 4L) and showed some ability to form macrometastatic lesions in the lung, although this result was not statistically significant (Figure 4M). This suggested that efficient transition of ITGB4^lo cells to the quasi-mesenchymal state might be enhanced by other transcription factors in addition to ΔNp63.

Altogether, these observations provided direct evidence that the ΔNp63 isoform, not p73, is the proximal TF primarily responsible for enforcing activation of the CSC transcriptional circuit and strengthened the candidacy of ΔNp63 as a centrally acting master regulator of residence in the quasi-mesenchymal CSC state.

ITGB4^hi CSCs rely on autocrine stimulation of the EGFR pathway to drive metastatic colonization

We found that ITGB4^hi cells with KO of ΔNp63/p73 also displayed a strong negative enrichment for a gene signature associated with EGF response (Figure 5A), which was notable because EGFR signaling is a key mediator of the re-epithelialization phase of wound healing response ⁶⁰. Indeed, relative to the more mesenchymal ITGB4^lo cells, ITGB4^hi cells displayed substantially elevated phosphorylation of the EGFR and downstream ERK phosphorylation, which was lost upon knockout of ΔNp63 and/or p73 (Figure 5B). Consistent with these findings, we also found that forced expression of ΔNp63 in ITGB4^lo NAMEC8R cells was able to activate EGFR signaling (Supplementary Figure 5A). Both ITGB4^lo and ITGB4^hi dKO cells displayed surface levels of EGFR comparable to those seen in ITGB4^hi CSCs (Supplementary Figure 5B), implying that the differential activation of the EGFR in these two cell types could reflect differences in the availability of the cognate ligands rather than in receptor protein levels.

Figure 5. ITGB4^hi CSCs are dependent on EGFR signaling for metastatic colonization.

(A) GSEA of RNA-seq data comparing ITGB4^hi NT and the indicated KO lines using an EGF response signature.

(B) Western blot analysis of EGFR and ERK activation in the indicated parental or KO cell lines.

(C) Representative qRT-PCR analysis of the indicated EGFR ligands in the ITGB4^lo and ITGB4^hi cells. Data are ± SEM, ** p ≤ 0.01, *** p ≤ 0.001, **** p ≤ 0.0001.

(D) Relative expression of the indicated EGFR ligands (from RNA-seq data) in control ITGB4^hi cells (NT) or KO cell lines. Data are ± SEM (n=3), * p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001,

(E) Western blot analysis of EGFR and ERK activation in serum-starved ITGB4^lo cells treated with media conditioned from ITGB4^hi cells. Negative control (−)cells were maintained in serum-free media.

(F) ITGB4^lo or ITGB4^hi tumorsphere growth after treatment with DMSO or erlotinib (1 μM). Representative images (left) and quantification (right). Data are ± SEM (n = 3), ** p ≤ 0.01. Scale bars, 100 μm.

(G) Correlation of TP63 expression with a 4 gene EGFR ligand score (TGFA, EREG, AREG, EPGN) in breast cancer specimens from the TCGA database.

(H) Representative IF staining for tdTom and pEGFR in lung cryosections prepared 48 hours after tail-vein injection of the indicated cell lines. Scale bars, 20 μm.

(I) Metastatic colonization of ITGB4^hi control or EGFR KO cells. Representative lung images (left) and quantification (right). Data are ± SEM (n = 5), ** p ≤ 0.01. Scale bars, 1 mm.

(J) Metastatic colonization of ITGB4^lo cells expressing a doxycycline-inducible TGFA expression vector. Representative lung images (left) and quantification (right). Data are ± SEM (n = 4 in −dox group and n=3 in +dox group), * p ≤ 0.05. Scale bars, 1 mm.

See also Figure S5.

In fact, we found that ITGB4^hi cells expressed significantly elevated transcript levels of a number of EGFR ligands, including TGFA, AREG, EREG, and EPGN (Figure 5C). Moreover, knockout of ΔNp63 and/or p73 led to a strong reduction in the expression of these ligands (Figure 5D), suggesting that stimulation of EGFR signaling in ITGB4^hi CSCs was maintained through autocrine secretion of one or more EGFR ligands. In further support of this idea, the highly mesenchymal ITGB4^lo cells could be induced to activate EGFR and downstream ERK signaling by treatment with media conditioned from ITGB4^hi cells (Figure 5E).

To test the functional importance of EGFR signaling to cells residing in the quasi-mesenchymal state, we treated ITGB4^lo and ITGB4^hi cells with erlotinib, a pharmacologic inhibitor of EGFR. Although we observed minimal effects when used in standard 2D monolayer culture (Supplementary Figure 5C), erlotinib treatment in tumorsphere culture strongly reduced the number of ITGB4^hi-derived spheres but not the number of ITGB4^lo spheres (Figure 5F). Treatment with erlotinib did not alter ITGB4 surface levels or ΔNp63 expression (Supplementary Figures 5D and 5E).

Consistent with these observations, we found a significant correlation between the expression of TP63 and a combined EGFR ligand score (comprising TGFA, AREG, EREG and EPGN) in patient samples from the breast TCGA dataset (Figure 5G). In addition, we analyzed single-cell RNA-sequencing data of circulating tumor cells isolated from breast cancer patients ⁶¹ and found expression of these ligands by a subset of circulating tumor cells (Supplementary Figure 5F). These observations suggested that the connection between ΔNp63 and the expression of EGFR ligands may be extrapolated to cases of human breast cancer.

Encouraged by these results, we next evaluated whether ITGB4^hi CSCs also depended on EGFR signaling for metastatic colonization. In accord with the in vitro experiments described above, we noted that ITGB4^hi NT control cells showed activation of EGFR in the lungs of mice 48 hours after tail-injection, as evidenced by numerous phospho-EGFR-positive membrane puncta, while dKO cells showed a strong reduction is such phospho-EGFR puncta (Figure 5H).

We then established ITGB4^hi cells with knockout of EGFR (Supplementary Figure 5G) in order to determine its essentiality in the ΔNp63-positive CSCs. EGFR KO had no effect on cell viability in vitro, cell morphology, ΔNp63 or p73 protein expression, ALDH positivity, or ITGB4 cell-surface levels (Supplementary Figure 5H–5K). However, EGFR KO completely ablated the ability of ITGB4^hi CSCs to colonize the lung (Figure 5I). Conversely, we found that overexpression of a single EGFR ligand, TGFA (Supplementary Figure 5L), strongly enabled metastatic colonization by the highly mesenchymal ITGB4^lo cells (Figure 5J), indicating that autocrine EGFR signaling functionally distinguishes metastatic quasi-mesenchymal CSCs from carcinoma cells residing in a highly mesenchymal state, which are non-metastatic and instead may enter into a dormant state.

Discussion

Activation of an EMT program in carcinoma cells has been linked to both metastatic progression ^8,13 and the formation of CSCs ²⁴. However, EMT program generates a number of intermediate, quasi-mesenchymal cell types with differing phenotypes and functions ^13,19,23. Multiple lines of evidence indicate that CSCs with high metastatic potential reside in an intermediate, quasi-mesenchymal state ^18,26,29, but such cells have proven difficult to isolate and study due to the extensive phenotypic plasticity that is characteristic of the various cell types generated by an EMT program. Here, we took advantage of a ITGB4^hi CSC population, which resides stably in a quasi-mesenchymal state, allowing us to undertake an in-depth analysis of how the chromatin configuration and enhancer landscape of these cells differs from those of the highly mesenchymal ITGB4^lo cells.

As we discovered in a completely unbiased search for TFs that control the quasi-mesenchymal CSC state, our analysis pointed clearly to p63 and p73 as key regulators that orchestrate the central transcriptional program of quasi-mesenchymal breast CSCs. The wide-ranging functions of these TFs in cancer have been debated for many years and a loose dichotomy has emerged with the N-terminal truncated isoforms (ΔN) generally having oncogenic functions while the full-length isoforms (TA) tend to be tumor-suppressive ^43,62. Our data indicated that p63, specifically the ΔNp63 isoform, is the TF proximally responsible for coordination of the CSC transcriptional program; the effects resulting from loss of p73 appear to be primarily due to its ability to transactivate the gene encoding ΔNp63. We were not able to distinguish unique roles for the different p73 isoforms, but we found that both isoforms seemed to be expressed in ITGB4^hi CSCs. Although ΔNp63 has previously been implicated as a key factor in both normal and neoplastic mammary stem cells ^37,63, our work adds to our understanding of how ΔNp63 functions in CSCs in two notable ways.

First, we found that a major function – perhaps the primary function – of ΔNp63 is to maintain CSCs in a quasi-mesenchymal cell state. Loss of either ΔNp63, p73, or both was sufficient to disrupt the CSC transcriptional circuit and caused carcinoma cells to lapse into a highly mesenchymal state that was associated with a loss of stem cell activity and an inability to generate lung metastases. Second, although these findings seemed, on the surface, to suggest a strong parallel between the normal mammary stem cell program and the one operating in breast CSCs, we were surprised to find that ΔNp63 regulated an almost completely distinct set of target genes in CSCs compared to normal mammary basal stem cells. This indicated that ΔNp63 controls normal and neoplastic epithelial stem cells in mechanistically different ways.

Indeed, our data suggested that the CSC program more closely approximates a regenerative epithelial response to wounding that is associated with various inflammatory mediators, cytokines, and activation of EGFR signaling. This is consistent with other observations that have linked inflammatory cytokines to CSCs ^64–66 as well as recent reports that have found similarities in the epithelial transcriptional circuits activated in the wounding response and during metastatic progression ^67–69. One intriguing possibility is that CSCs preferentially co-opt a transcriptional circuit normally reserved for activated stem cells that proliferate rapidly in response to tissue damage, in contrast to the program that operates in a tonic fashion and controls quiescent stem cells and tissue homeostasis ⁷⁰.

The biochemical mechanisms responsible for ΔNp63 target selection in breast CSCs remain unknown, but conceivably could involve interactions with components of the SWI/SNF complex ⁶⁹ or NF-kB pathway ⁷¹, similar to previous reports in head and neck squamous cell carcinoma. Additionally, we found that ectopic expression of ΔNp63 resulted in only limited conversion of ITGB4^lo cells to the quasi-mesenchymal state, suggesting that other collaborating transcription factors may be required in order for highly mesenchymal carcinoma cells to efficiently transition to the CSC state.

Our study also provides one explanation for the high metastatic competence of quasi-mesenchymal carcinoma cells. We found that ΔNp63 was necessary for autocrine stimulation of EGFR, which sustained proliferation after extravasation into the lung parenchyma. Clusters of circulating tumor cells (CTCs) that disseminate in aggregate are also known to initiate metastases with high efficiency ⁷² and the EGFR ligand epigen (EPGN) is a key signal that promotes the metastatic growth of CTC clusters ⁷³. Hence, the determinants of successful metastatic colonization may be quite similar for quasi-mesenchymal CSCs, which likely exhibit a single-cell mode of dissemination, and CTC clusters, which disseminate collectively. In fact, both seem to rely on a common mechanism for colonization that is centered on stimulation of EGFR.

It is also notable that highly mesenchymal carcinoma cells, which display strong PDGFR signaling but weak EGFR signaling ⁴⁸, are compromised in their ability to generate macrometastatic colonies, suggesting potential qualitative or quantitative differences in these two mitogenic pathways, at least with respect to metastatic colonization. Indeed, we found that simply activating autocrine EGFR signaling in highly mesenchymal carcinoma cells, achieved via expression of TGFA, enabled metastatic colonization.

ΔNp63 has been found to promote metastasis by altering the stromal microenvironment ^74,75, but we suspect such stromal interactions may come into play at later phases of metastatic colonization, long after the initial requirement for proliferation that is mediated by EGFR signaling.

Through a detailed analysis of chromatin access and enhancer activation we have elucidated the underlying transcriptional network that distinguishes highly metastatic quasi-mesenchymal CSCs from more extreme mesenchymal cells that lack stem cell activity and have limited colonization potential. Our findings revealed that the transcriptional circuits operative in breast CSCs are substantially different from those associated with normal mammary stem cells, provide a mechanistic explanation for the high metastatic competence of quasi-mesenchymal cells, and suggest potential dependencies and vulnerabilities for specifically targeting cells residing in the CSC state.

Limitations of the Study

This study used a limited number of breast cancer cell lines to provide an in-depth characterization of the nature and function of the gene regulatory circuit controlled by ΔNp63 and p73 in quasi-mesenchymal CSCs. However, the specific transcriptional program controlled by these factors could vary within and across cancer types. Also, because we focused on human cancer cells, our metastasis models involved immunocompromised mice and therefore excluded potentially relevant interactions between CSCs and components of the immune system.

STAR★METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Robert Weinberg (weinberg@wi.mit.edu).

Materials Availability

Plasmids and cell lines generated for this study are available upon request from the lead contact.

Data and Code Availability

• Sequencing data have been deposited at GEO and are publicly available as of the date of publication. The accession number for the ChIP-seq and ATAC-seq reported in this paper is GSE172407. The accession numbers for the RNA-seq and single-cell RNA-seq reported in this paper are GSE171629, GSE171626 and GSE171628.

• No original code was generated for this study.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell Lines

Isolation of SUM159 ITGB4^lo and ITGB4^hi cells was described previously ²⁸. Both lines were cultured in F12 supplemented with 5% inactivated fetal bovine serum (IFS), hydrocortisone (1 μg/mL), insulin (5 μg/mL) and 5% penicillin/streptomycin. 293T cells used for viral production were maintained in DMEM supplemented with 10% IFS and 5% penicillin/streptomycin. NAMEC8 cells were cultured as previously described ⁴⁸. Cells lines were confirmed to be mycoplasma-free.

Mice and Animal Procedures

A colony of NOD scid gamma (NOD.Cg-Prkdscid Il2rgtm1 Wjl/SzJ, originally from Jackson Laboratory) mice was maintained in-house and used for all xenograft experiments. Animals were housed in the Whitehead Institute animal facility under standard conditions, including 12-hour light/dark cycle, temperature of 18–23°C and humidity of 40–60%. Mice were housed in sterilized cages, with food (irradiated chow) and water available ad libitum. All procedures and experiments described here were approved by the Massachusetts Institute of Technology Committee on Animal Care (CAC). Male NSG mice, aged 6–10 weeks, were used for metastatic colonization assays.

METHOD DETAILS

Plasmids and Cloning

Single-guide RNAs were designed using software from the Broad Institute, Genetic Perturbation Platform (https://portals.broadinstitute.org/gppx/crispick/public) and cloned into lenti-CRISPR-V2-puro (Addgene plasmid # 52961), a gift from Feng Zhang ⁷⁹ or lenti-CRSIPR-V2-GFP vector (Addgene plasmid # 82416), a gift from David Feldser ⁸⁰, as previously described ⁷⁹. Cells were labeled with a Luciferase-tdTomato-encoding lentiviral plasmid ⁷⁷. For rescue experiments, a plasmid encoding ΔNp63alpha (Addgene plasmid # 26979), a gift from David Sidransky ⁸¹, was used to subclone ΔNp63 into the Gateway entry vector pENTR1A (Thermo Fisher); an entry vector (pENTR223) encoding TAp73 was obtained from the PlasmID Repository at Harvard Medical School (Clone ID: HsCD00399879). Silent mutations of the PAM sites were made in the entry vectors using the Q5 Site-directed Mutagenesis kit according to the manufacturer’s protocols (New England Biolabs) and subsequently recombined into the Gateway destination vector pLenti6/V5-DEST (Thermo Fisher). Doxycycline-induced vectors for ΔNp63 (FLAG/blasticidin) and TGFA (EGFP) were designed and manufactured by VectorBuilder and co-transduced with pLenti-CMV-rtta3-hygro (Addgene plasmid #26730). All plasmids were verified by sequencing.

Lentiviral Production and Generation of Cell Lines

Lentivirus was produced by co-transfection of 293T cells with the viral vector and the packaging plasmids pMD2.G (Addgene plasmid # 12259) and psPax2 (Addgene plasmid # 12260) as previously described ⁸². Cells were infected in the presence of 2 μg/mL polybrene (Millipore Sigma) and subsequently selected with puromycin (2 μg/mL), blasticidin (10 μg/mL) or hygromycin (250 μg/mL). Cells infected with viral constructs bearing a fluorescent tag were selected by cell sorting.

Fluorescence-Activated Cell Sorting (FACS)

Single-cell suspensions of ~ 1 × 10⁶ cells were washed with ice-cold FACS buffer (PBS⁻ + 2% IFS) and resuspended in 100 μL FACS buffer for staining. Cells were stained with antibodies for 30 minutes at 4°C and washed one time prior to analysis for sorting. Cell were analyzed on a BD Fortessa or LSRII instrument and sorted on a BD FACSAria, both running FACSDiva software (BD Biosciences). FlowJo (FlowJo, LLC) software was used for the analysis of FACS data. The following antibodies were used for flow cytometry: CD44-PeCy7 (1:500, Thermo Fisher), ITGB4-eFluor660 (1:50, Thermo Fisher), and EGFR (1:100, BD Biosciences). ALDH activity was assessed using the ALDEFLUOR kit (STEMCELL Technologies) according to the manufacturer’s protocol (5 μL/test of ALDEFLUOR reagent and DEAB), with cells incubated for 30 minutes at 37°C prior to analysis.

Quantitative RT-PCR

RNA was extracted with TRIzol (Thermo Fisher) and purified using the RNeasy Plus Kit (Qiagen). cDNA was synthesized with the High-capacity cDNA Reverse Transcription Kit (Thermo Fisher) using oligo-dT primers (Thermo Fisher) and an RNase inhibitor (New England Biolabs). Quantitative real-time PCR was performed with the SYBR Green Mastermix (Roche) using a LightCycler 480 II (Roche).

Western Blotting

Cells were washed in ice-cold PBS⁻ and lysed on ice with RIPA (Millipore Sigma) containing protease (Roche) and phosphatase inhibitors (Roche). Protein lysates were cleared by centrifugation at 14,000g for 15 minutes and resolved on NuPAGE Novex 4–12% Bis-Tris gels as described by the manufacturer (Thermo Fisher). Proteins were transferred to PVDF membranes and blocked with TBS-T containing 5% non-fat dry milk for 30 minutes prior to incubation with primary antibodies overnight at 4°C. Blots were washed in TBS-T, incubated with HRP-conjugated secondary antibodies (Cell Signaling Technologies) and detected using Western Lightning Plus-ECL chemiluminescence substrate (PerkinElmer). Condition media was collected from cells grown in serum-starved media for 24hr and was cleared by centrifugation. The following antibodies were used: anti-p63α (CST 13109, 1:1000), anti-ΔNp63α (CST 67825, 1:1000) anti-p73 (CST 14620, 1:1000), anti-GAPDH (CST 5174, 1:10000), anti-FLAG (Millipore Sigma F1804, 1:1000), anti-E-cadherin (CST 3195, 1:1000), anti-N-cadherin (CST 13116, 1:1000), anti-fibronectin (BD 610078, 1:1000), anti-Slug (CST 9585, 1:1000), anti-Snail (CST 3879, 1:1000), anti-Twist (Abcam ab50887, 1:1000), anti-Zeb1 (CST 3396, 1:1000), anti-phosphoEGFR^Tyr1068 (CST 3777, 1:1000), anti-EGFR (CST 4267, 1:1000), anti-phospho-ERK^{Thr202/Tyr204} (CST 4370, 1:1000), anti-ERK (CST 4695, 1:1000), anti-COX IV (CST 4850, 1:5000), and anti-beta-tubulin (CST 2128, 1:5000).

Immunostaining

For cellular immunofluorescence, cells were cultured in tissue culture-treated 8-well chamber slides (Thermo Fisher). Cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton-X, and blocked in PBS- with 3% horse serum (Vector Laboratories). Cells were incubated overnight at 4°C with the following primary antibodies: anti-p63α (CST 13109, 1:800), anti-ΔNp63 (CST 67825, 1:400) anti-p73 (CST 14620, 1:250) and secondary staining was performed using the TSA amplification kit according the manufacturer’s protocol (PerkinElmer). Fixed paraffin-embedded tissue sections were deparaffinized with Histo-Clear II (National Diagnostics) and rehydrated through a descending alcohol series. Antigen retrieval was performed with a homemade or commercial (Dako) citrate buffer (pH 6.1). For immunohistochemistry, incubatation with HRP-conjugated secondary antibodies was performed using the VECTASTAIN Elite ABC Kit (Vector Laboratories) and detected with ImmPACT DAB peroxidase substrate kit (Vector Laboratories). Slides were counterstained with hematoxylin QS (Vector Laboratories). For immunofluorescence, tissue sections were incubated with secondary antibodies conjugated to Alexa Fluor-488, -555, or -633 (Biotium, 1:400) and slides were mounted with Prolong anti-fade with NucBlue (Thermo Fisher). For visualization and staining of carcinoma cells following tail-vein injection, the lungs were perfused and embedded in O.C.T. compound (Sakura), and snap frozen in liquid nitrogen as previously described ⁷⁶. Cryosections (20 μM) were thawed overnight, fixed in 4% paraformaldehyde, and stained as described above for cellular immunofluorescence with the following primary antibodies: anti-CD31-Alexa Fluor657 (BioLegend 102416, 1:100), anti-Ki67 (CST 9129, 1:400), anti-cleaved-caspase 3 (CST 9661, 1:400), anti-phospho-EGFR^Tyr1068 (CST 3777, 1:1000). Images were acquired on a Zeiss inverted fluorescent microscope and analyzed with the ZEN imaging software (Zeiss).

Metastasis Assays

For lateral tail-vein injections, male NSG mice were injected with 5 × 10⁵ cells resuspended in 100 μl PBS⁻ and lungs were analyzed for visible macroscopic metastases formation 4–6 weeks after injection using a Leica MZ12 fluorescence dissection microscope. Lungs were fixed in 10% neutral buffered formalin, embedded in paraffin, sectioned (5 μM), and stained with H&E. Tumor cell extravasation was assessed on cryosections prepared 48 hours after tail-vein injections. Tdtom⁺ cells were scored as extravasated if they were completely outside a capillary vessel and/or if there was no overlap between their nuclei and CD31⁺ vessels. To examine tdTom⁺ carcinoma cells in the lungs with flow cytometry, lungs were minced and incubated in DMEM with 1.5 mg/ml collagenase A and 0.01% DNase for 30 minutes at 37°C and dissociated using a gentleMACS tissue dissociator (Miltenyi Biotec). RBCs were lysed by resuspension in RBS Lysis Buffer (BioLegend) and cells were passed through a cell strainer (BD Biosciences) to obtain a single-cell suspension prior to flow cytometric analysis. For metastasis experiments involving cells carrying doxycycline-inducible constructs, expression was induced in vitro for 5–7 days prior to tail-vein injection and mice were maintained on a doxycycline-containing diet (625 mg/kg, Teklad/Envigo) for the duration of the experiment.

Bioluminescent Imaging

Mice were imaged using an IVIS Spectrum imaging system following intraperitoneal injection with D-luciferin (165 mg/kg, PerkinElmer) and the data analyzed using Living Image software (PerkinElmer).

ChIP-seq

ChIP-seq was performed as described ^83,84. Briefly, cells were crosslinked with 1% formaldehyde in PBS for 8 minutes and quenched in 1mM glycine. Cells were washed sequentially in lysis buffer 1 (140 mM NaCl, 1 mM EDTA, 50 mM HEPES, 10% glycerol, 0.5% NP-40, 0.25% Triton-X-100), lysis buffer 2 (10 mM TRIS [pH 8.0], 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA) and sonication buffer (10 mM TRIS [pH 8.0], 1 mM EDTA, 0.1% SDS) containing 1x protease inhibitor (Thermo Fisher 78430) and 1mM sodium butyrate on ice for 10 minutes. Cells were sonicated for 5 minutes total (Covaris E220) and supernatants collected. Sheared chromatin was diluted 1:1 in dilution buffer (300 mM NaCl, 2 mM EDTA, 50 mM TRIS [pH 8.0], 1.5% Triton-X, 0.1% SDS) and incubated with anti-H3K27ac (Abcam, ab4729), anti-p63 (CST 13109) or anti-p73 (Bethyl A300-126a) conjugated Protein G Dynabeads (Thermo Fisher) overnight (8–16 hours, with rotation) at 4°C, and then washed two times with wash buffer 1 (20 mM TRIS [pH 8.0], 150 mM NaCl, 2 mM EDTA, 1% Triton-X, 0.1% SDS, 0.05% DOC), two times with high-salt wash buffer (20 mM TRIS [pH 8.0], 500 mM NaCl, 2 mM EDTA, 1% Triton-X, 0.1% SDS, 0.05% DOC), one time with LiCl buffer (250 mM LiCl, 10 mM TRIS 8.0, 1mM EDTA, 0.5% NP-40) and one time with TE-NaCl buffer (10 mM Tris [pH 8.0], 1 mM EDTA, 50 mM NaCl). Samples were eluted from beads for 1 hr at 65°C in elution buffer (50 mM TRIS [pH8.0], 10 mM EDTA, 1% SDS) and the supernatant reverse-crosslinked at 65°C for 8 hours. Samples were diluted 1:1 with TE buffer (50 mM TRIS [pH 8.0], 1 mM EDTA) and treated with RNase A (0.2 mg/ml) for 2 hours at 37°C and then Proteinase K (0.2 mg/ml) for 2 hours at 55°C. DNA was isolated by phenol-chloroform extraction and ethanol precipitation. DNA libraries were prepared using Swift Biosciences ChIP kit (Cat# 21024) according to the manufacturer’s instructions using primers and amplification instructions from KAPA Biosystems HIFI library amp kit (Cat# KK2611) and were sequenced on an Illumina HiSeq.

Enhancer calling

Enhancers and were called and quantified from H3K27ac ChIP-seq data as done previously ⁸⁴. Briefly, peaks were called using MACS v1.4 ⁷⁸ using a p-value cutoff of 1e-9. Enhancers were then formed from these peaks by merging them together if they were within 12.5kb of each other. In order to compare between multiple sample replicates, we created a merged enhancer map by merging together any overlapping enhancers between the ITGB4^hi and ITGB4^lo replicates. We then calculated a score for each enhancer as before ⁸⁴, by calculating the read coverage in each enhancer (with input subtraction) and fitting to a negative binomial distribution. The scores were divided by the point at which the fit distribution was at 0.975, such that those scores below 1 were called “typical enhancers” and those scores above 1 were called “super-enhancers”. This produced an enhancer score for each enhancer in each replicate. These were then quantile normalized. The enhancer scores for replicates of the same state were averaged together to produce an enhancer score for each cell state.

Differential enhancers

Differential enhancers were determined using the merged enhancer map of all replicates. We then counted the number of unique reads from the IP sample overlapping each enhancer to use in a differential analysis. The R package DESeq2 ⁸⁵ was used to determine enhancers with differential read coverage in the ITGB4^hi vs. ITGB4^lo state, using default parameters. We called differential super-enhancers by filtering the differential enhancers to those enhancers with an enhancer score of at least 1 in at least one replicate. A new adjusted p-value (FDR) was calculated using the Benjamini & Hochberg method on only these super-enhancers. Those super-enhancers with a FDR<0.05 and a log₂ fold-change of at least 1 were called differential, and were further filtered for those with an enhancer score of at least 1 in the differential cell state.

Differential H3K27ac peaks

We created a merged map of H3K27ac peaks by merging the peaks from all the replicates from the ITGB4^hi and ITGB4^lo states. The number of reads overlapping each peak from each replicate (IP only) were then counted. We created a score for each peak by normalizing the number of reads in each peak using the function varianceStabilizingTransformation from DESeq2. Only reads with a score of at least the median in at least one sample were used for downstream analysis. We then performed a differential analysis using DESeq2 on the un-normalized counts to find peaks differential between ITGB4^hi and ITGB4^lo states. We used an FDR cutoff of 0.05 and a log₂ fold-change cutoff of 1 to call differential peaks.

Enhancer linking to genes

Enhancers were linked to genes using the Pearson correlation of the H3K27ac ChIP-seq signal at the enhancers with all promoters (+5kb to −1kb) within 500kb of the enhancer. Promoter reads were normalized using and the varianceStabilizingTransformation function from DESeq2 and the quantile normalized enhancer score was used for enhancers. This was done across a set of breast cancer cell lines and patient samples. Only positive correlations were kept. Then, promoters were filtered for those either with a correlation of at least 0.5 and p<0.05 or within 10kb of the enhancer. Finally, only those promoters with a correlation within 0.1 of the best correlated promoters to that enhancer were linked to the gene.

p63/p73 ChIP-seq analysis

p63/p73 ChIP-seq peaks were called using MACS2 and IDR with pseudo-replicates as described above in the ATAC-seq peak calling section. An IDR threshold of 0.01 was used. p63/p73 peaks were linked to genes by defining a gene regulatory region for each gene. This was defined as the basal regulatory region (−5kb to +1kb) around the TSS of each gene, plus up to 10kb beyond the basal regulatory region, up to the next gene’s basal regulatory region. If a p63/p73 peak was located in the gene regulatory region of the gene, it was said to bind that gene. For the clustergram heatmap, we used 100bp windows in the 5kb around the center of each of the H3K27ac peaks in the merged map. The number of reads per million (RPM) mapped reads in each window was calculated for each IP replicate for H3K27ac and p63/p73. The replicates of H3K27ac in ITGB4^hi or ITGB4^lo were averaged together. The heatmap shows the log₂(RPM+0.1).

Expression of p63/p73 bound genes

Genes were considered bound by p63 or p73 if a ChIP-seq peak for the TF was present in the gene regulatory region of the gene as defined above. A Student’s t-test was used to compare the expression of p63/p73 bound vs. unbound genes, only considering genes with an H3K27ac enhancer linked to the gene.

Roadmap enhancer analysis

Enhancers based on H3K27ac for all cell types in the Epigenome Roadmap ⁵⁰ were compared to the H3K27ac enhancers from this study: those activated in ITGB4^hi cells, activated in the ITGB4^lo cells, or unchanged. The fraction of the enhancers from this study overlapping each cell type was calculated.

ATAC-seq

ATAC-seq was performed as described ⁸⁶. Briefly, cells were treated in culture medium with 200 U/ml DNase (Worthington cat# LS002007) at room temperature for 30 minutes. Cells were pelleted and washed twice with cold PBS. Fifty thousand cells were centrifuged, washed with 50 μl of cold PBS and resuspended in cold lysis buffer. Cells were pelleted for 10 minutes at 500×g, 4°C and placed on ice. Cells were combined with transposition mix (25 μl TD [2x reaction buffer, 2.5 μl TDE1, Tn5 Transposase, 22.5 μl nuclease-free H2O]) from Illumina Tagment DNA TDE1 Enzyme and Buffer Kit (Cat# 20034198) and incubated at 37°C for 30 minutes. Immediately following transposition, DNA was purified using a Qiagen MinElute Reaction Cleanup Kit according to the manufacturer’s protocol. DNA was amplified by PCR using the NEBNext High-Fidelity 2x PCR Master Mix using 1 cycle of 5 minutes 72°C, 30 seconds 98°C. Amplified DNA was purified using Qiagen MinElute Reaction Cleanup Kit and sequenced on an Illumina HiSeq.

ATAC-seq analysis

We first trimmed off an adapter sequence in the ATAC-seq reads and then aligned them to hg19 using Bowtie2. Duplicate reads were then removed before peak calling. Peak calling was performed using MACS2 ⁷⁸ and IDR (https://sites.google.com/site/anshulkundaje/projects/idr). Aligned reads from each replicate of a given state were combined. Pseudo-replicates from these combined reads were then used for peaking calling with MACS2. IDR was then performed on the peak calls from the pseudo-replicates and an IDR cutoff of 0.1 was used. These comprised the ATAC-seq peaks from each cell state.

Association of ATAC-seq with H3K27ac

ATAC-seq peaks were associated with differential H3K27ac by checking for overlaps of the ATAC-seq peaks and H3K27ac peaks. We used a merged map of ATAC-seq peaks from both cell states and the merged map of H3K27ac peaks from both cell states. Those ATAC-seq peaks overlapping H3K27ac peaks stronger in ITGB4^hi or ITGB4^lo (FDR<0.05 and log₂ fold-change > 1) were labeled as ATAC-seq peaks with differential H3K27ac in either ITGB4^hi or ITGB4^lo. Background sets of ATAC-seq peaks for each state were determined using overlaps with H3K27ac peaks that were not differential for each state (either p-value>0.1 and log₂ fold-change < 0.5, or log₂ fold-change < 0).

Predicted TF binding sites by motifs and ATAC-seq

We determined predicted TF binding sites by associating a set of TF sequence motifs with the merged set of ATAC-seq peaks. Motifs were drawn from the following databases: motifs based on ENCODE data ⁸⁷ (http://compbio.mit.edu/encode-motifs), JASPAR ⁸⁸, uniPROBE ⁸⁹, and high-throughput SELEX ⁹⁰. Motifs mapping to the same TF were ranked according to the database that they originally came from (some sources of motifs contained motifs from multiple original databases): 1) uniPROBE, 2) JASPAR, 3) Jolma, 4) Transfac, 5) ENCODE. For each TF, motifs were chosen only from the database with the highest-ranking motif(s), creating the “Curated motif set.” Next, FIMO ⁹¹ was used to find instances of the motifs in the genome, with a p-value cutoff of 10⁻⁴, and using only the top 500k locations per motif.

Enrichment of predicted TF binding sites in differential H3K27ac

To find the TFs for which binding sites were associated with differential H3K27ac from ITGB4^hi, we determined which ATAC-seq peaks with differential H3K27ac in ITGB4^hi had a predicted binding site compared to which ATAC-seq peaks from the background set for ITGB4^hi had a predicted binding site. We then used a Fisher’s exact test to determine enrichment and adjusted the p-values using the Benjamini & Hochberg method to produce FDRs. If a TF had multiple motifs in the database, the one with the greatest log-odds ratio of enrichment was used. Motifs with an FDR<0.01 were considered enriched.

RNA-sequencing

Total RNA was extracted with TRIzol (Thermo Fisher) and purified using the RNeasy Plus Kit (Qiagen). Libraries were prepared for RNA-Seq using KAPA mRNA HyperPrep Kit (Roche), according to manufacturer’s directions. Briefly, total RNA was enriched for polyadenylated sequences using oligo-dT magnetic bead capture. The enriched mRNA fraction was then fragmented, and first-strand cDNA generated using random primers. Strand specificity is achieved during second-strand cDNA synthesis by replacing dTTP with dUTP. The resulting cDNA was A-Tailed and ligated with indexed adapters. Finally, the library was amplified using KAPA qPCR library quant kit (Roche) as per manufacturers protocol. Libraries were then pooled at equimolar concentration, for each lane, using qPCR concentrations. The pooled libraries were denatured using the Illumina protocol. The denatured libraries were sequenced on an Illumina HiSeq 2500 (40-nt long single end). Basecalls were performed using Illumina offline basecaller (OLB) and then demultiplexed.

Reads were mapped with STAR ⁹² to hg38 version of human genome using the “sjdbOverhang” parameter set to 39, and the annotation file from ENSEMBL GRCh38.90. At least 22 and 32 million of reads were uniquely mapped reads for the knockout and forced overexpression experiments, respectively. The samples were stranded, and counts for the 2nd read strand aligned with RNA were used for calculating the number of counts per gene. If none of the samples within comparisons had at least one read assigned to a gene, this gene was considered to be not expressed and removed from further analysis. Normalization and differential expression were done with DESeq2 (https://pubmed.ncbi.nlm.nih.gov/25516281/), which taking batch effect into design consideration. Genes with at least two-fold differences plus adjusted p-values no more than 0.01 were considered to be differentially expressed. To create the heatmap, the differentially expressed genes were used for hierarchical clustering with Cluster 3.0 ⁹³ based on uncentered correlation and average linkage.

Single-cell RNA-sequencing

Three days after tail-vein injection of ITGB4^hi NT or dKO cells, mice were euthanized and the lungs dissociated as described above. Flow cytometry was used to isolate live tdTom⁺ carcinoma cells, which were immediately processed for single-cell RNA-sequencing. Cells were processed using the 10X Genomics Chromium Controller with the Single Cell 3ʹ v3 Reagent Kit, according to manufacturer’s directions. Briefly, a target number of 1250 cells per library was roughly achieved by loading 1.6 times the target number of cells in suspension. The Chromium Controller processes individual cells so that each cell is marked with its unique barcode during reverse transcription. The end result is a single library representing one cell suspension, containing data for each individual cell. Libraries were amplified using KAPA qPCR library quant kit (Roche) as per manufacturers protocol. Libraries were then pooled at equimolar concentration, for each lane, using qPCR concentrations. The pooled libraries were denatured using the Illumina protocol. The denatured libraries were loaded onto an HiSeq 2500 (Illumina) and sequenced for 60×60 cycles. Fastq files were generated and demultiplexed with the bcl2fastq Conversion Software (Illumina).

Alignments, counts matrix and loupe browser files were generated using cellrangerv4 (10X Genomics) using mixed species reference genome (refdata-gex-GRCh38_and_mm10-2020-A) and human only reference genome (refdata-gex-GRCh38-2020-A). Reference genomes were downloaded from 10x Genomics website (https://support.10xgenomics.com/single-cell-gene-expression/software/downloads/latest).

Further downstream analysis was done using Seurat Bioconductor package (Lambert_scRNASeq_Analysis.R, https://github.com/sumeetg23/DifferentialExpression/blob/master/Lambert_scRNASeq_Analysis.R) to ensure any contaminating mouse cells were eliminated from the analysis. Specifically, from dual species analysis, the cell barcodes classified as human were identified from the gem_classification.csv file. We used the list of human cell barcodes to select for cells in the analysis done with human only reference genome, using cellranger reanalyze option. Read count matrix for human only dataset was then normalized and cluster specific markers for each cluster were identified using FindAllMarkers function (parameters - min.pct = 0.25, logfc.threshold = 0.25₎. The list of markers was further filtered for a fold-change of 1.5 and sorted by p-value to identify the top markers.

Gene Set Enrichment Analysis

Protein coding genes ranked by fold changes were used as input for GseaPreranked tool from the Broad Institute ⁹⁴. Hallmark annotation h.all.v7.2.symbols.gmt was downloaded from the Broad Institute. EGF response ⁹⁵ was from the GSEA result using the curated gene sets c2.all.v6.2.symbols.gmt as input. For epidermal wounding gene set, we downloaded the differentially expressed genes between basal cells from the unwounded (UW) and (wounded) WO mouse samples (Supplemental Table 4 from Haensel et al. ⁵⁶. To create the gene set of alveolar regeneration (Strunz et al. ⁵⁷), we downloaded Krt8_ADI_all genes.xlsx (http://146.107.176.18:3838/Bleo_webtool_v2/) selected mouse genes with at least two-fold up in Krt8+ ADI cells plus adjusted p-value less than 0.001. To get the gene set from revival intestinal stem cell (Ayyaz et al. ⁵⁸), we used the human homologs of genes from the revival stem cell signature ayyaz_ganesh.xls (downloaded from Ganesh et al. ⁶⁸, Supplementary Table 1). We analyzed the gene signature regulating wound healing in mouse tail epidermis (Aragona et al. ⁵⁹) using the limma package ⁹⁶. We fit a linear model to their GSE93638 microarray data, and used empirical bayes statistics for differential expression. Genes with adjusted p-value less than 0.01 and at least two-fold differences were selected to be in the gene sets. Since our ranked genes came from human cancer cells, all the genes from mouse gene sets were converted to human symbols using orthologs from ENSEMBL before running GSEA analysis.

Cell Proliferation Assay

Cell proliferation was measured using the CellTiter 96 Aqueous Cell Proliferation Assay according to the manufacturer’s protocol (Promega).

Tumor Sphere Culture

Tumorspheres were grown in Mammocult medium (STEMCELL Technologies) containing 1% methylcellulose and supplemented with 4 μg/mL heparin, 0.48 μg/mL hydrocortisone and penicillin/streptomycin. Between 500–1000 cells were seeded in 1mL of sphere media in 24-well ultra-low attachment plates (Corning). For erlotinib treatment experiments, a single dose of erlotinib (1μM) of DMSO was added at the time of cell seeding. Spheres were quantified after 7–10 days of growth.

Transendothelial Migration Assays

The transendothelial migration assay was performed using QCM Transendothelial Migration Assay according to the manufacturer’s instructions (Millipore Sigma).

TCGA and CTC Analysis

Normalized RNA-seq gene expression data for primary breast cancer profiles as part of The Cancer Genome Atlas (TCGA) were obtained using Firehose (http://firebrowse.org/?cohort=BRCA). EGFR ligand score was calculated as geometric mean of EREG, TGFA, AREG and EPGN genes. Correlation analysis was performed using PRISM 9 software. Gene expression data of circulating tumor cells from breast cancer patients was from the GSE111065 dataset ⁶¹.

QUANTIFICATION AND STATISTICAL ANALYSIS

All data are presented as the mean ± standard error of the mean (SEM) and statistical significance was assessed using two-tailed Student’s t-test unless specified otherwise. p < 0.05 was considered significant. Statistical analyses were performed using Prism software and detailed statistical information (including n of cell or animals) can be found in the figure legends.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
p63α (immunoblotting, ChIP, IHC/IF)	Cell Signaling Technology	Cat#13109
ΔNp63α (immunoblotting, IHC/IF)	Cell Signaling Technology	Cat#67825
p73 (immunoblotting, IHC/IF)	Cell Signaling Technology	Cat#14620
p73 (ChIP)	Bethyl Laboratories	Cat#A300-126a
H3K27ac (ChIP)	Abcam	Cat#ab4729
E-cadherin	Cell Signaling Technology	Cat#3195
N-cadherin	Cell Signaling Technology	Cat#13116
Fibronectin	BD Biosciences	Cat#610078
Zeb1	Cell Signaling Technology	Cat#3396
Snail	Cell Signaling Technology	Cat#3879
Slug	Cell Signaling Technology	Cat#9585
Twist1	Abcam	Cat#ab50887
phospho-ERK1/2 (T202/Y204)	Cell Signaling Technology	Cat#4370
ERK1/2	Cell Signaling Technology	Cat#4695
phospho-EGFR (Y1068)	Cell Signaling Technology	Cat#3777
EGFR	Cell Signaling Technology	Cat#4267
FLAG	Millipore Sigma	Cat#F1804
GAPDH	Cell Signaling Technology	Cat#2118
COX IV	Cell Signaling Technology	Cat#4850
β-tubulin	Cell Signaling Technology	Cat#2128
CD44-PeCy7	Thermo Fisher Scientific	Cat#25-0441-81
ITGB4 (CD104)-e660	Thermo Fisher Scientific	Cat#50-1049-82
EGFR-BV786	BD Biosciences	Cat#742606
Ki67	Cell Signaling Technology	Cat#9129
CD31-Alexa Fluor 647	BioLegend	Cat#102416
Cleaved caspase-3	Cell Signaling Technology	Cat#9661
Bacterial and virus strains
Stbl3™ E. coli	Thermo Fisher Scientific	Cat#C737303
DH5α E. coli	Thermo Fisher Scientific	Cat#18265017
Chemicals, peptides, and recombinant proteins
DMEM	Millipore Sigma	Cat#D6171
Ham’s F-12	Thermo Fisher Scientific	Cat#21700075
MammoCult	STEMCELL Technologies	Cat#05620
Polybrene	Millipore Sigma	Cat#TR-1003-G
X-tremeGENE 9	Roche	Cat#19128100
TRIzol	Thermo Fisher Scientific	Cat#15596018
Oligo-dT	Thermo Fisher Scientific	Cat#SO132
RNase inhibitor	Thermo Fisher Scientific	Cat#N8080119
RIPA	Millipore Sigma	Cat#R0278
Complete, EDTA-free protease inhibitor cocktail	Roche	Cat#11873580001
PhosSTOP phosphatase inhibitor cocktail	Roche	Cat#04906845
Protease inhibitor cocktail	Thermo Fisher Scientific	Cat#78430
Protein G Dynabeads	Thermo Fisher Scientific	Cat#10004D
NuPAGE MOPS SDS running buffer	Thermo Fisher Scientific	Cat#NP001
NuPAGE transfer buffer	Thermo Fisher Scientific	Cat#NP0006-1
Enhanced Plus-ECL solution	PerkinElmer	Cat#NEL104001EA
10% neutral buffered formalin	VWR	Cat# 89370-094
16% formaldehyde	Thermo Fisher Scientific	Cat#28908
Histo-Clear II	National Diagnostics	Cat#HS-202
Target retrieval solution	Agilent Dako	Cat#S1699
Horse serum	Vector Laboratories	Cat#S-2000
Hematoxylin QS	Vector Laboratories	Cat#H3404
ProLong glass antifade with NucBlue	Thermo Fisher Scientific	Cat#P36981
Tissue-Tek O.C.T compound	Sakura	Cat#4583
heparin	STEMCELL Technologies	Cat#07980
hydrocortisone	Millipore Sigma	Cat# H0396
DMSO	Millipore Sigma	Cat#D2650
erlotinib	Selleckchem	Cat#S1023
Collagenase	Millipore Sigma	Cat#11088793001
DNase	Worthington Biochemical	Cat#LS002007
RBC lysis buffer	BioLegend	Cat#420301
D-luciferin	PerkinElmer	Cat#122799
Critical commercial assays
DNA ligation kit, mighty mix	Takara	Cat#6023
Gateway LR clonase II	Thermo Fisher Scientific	Cat#11791020
High-fidelity 2x PCR master mix	New England Biolabs	Cat#M0541
Q5 site-directed mutagenesis kit	New England Biolabs	Cat#E0554S
Gel DNA recovery kit	Zymo Research	Cat#D4002
Zyppy plasmid minprep kit	Zymo Research	Cat#D4036
RNeasy mini kit	Qiagen	Cat#74104
MiniElute reaction cleanup kit	Qiagen	Cat#28206
High-capacity cDNA reverse transcription kit	Thermo Fisher Scientific	Cat#4368814
SYBR Green I master	Roche	Cat#14571520
ALDEFLUOR kit	STEMCELL Technologies	Cat#01700
VECTASTAIN Elite ABC peroxidase kit	Vector Laboratories	Cat#PK-6101
ImmPACT DAB peroxidase substrate kit	Vector Laboratories	Cat#SK-4100
TSA Plus cyanine 3 system	PerkinElmer	Cat#NEL744001KT
CellTiter 96 AQueous cell proliferation assay	Promega	Cat#G3582
QCM transendothelial migration assay	Millipore Sigma	Cat#ECM558
2S Plus DNA Library kit	Swift Biosciences	Cat#20024
KAPA HiFi amplification kit	Roche	Cat#KK2611
Illumina Tagment DNA TDE1 Enzyme and Buffer	Illumina	Cat# 20034198
Deposited data
H3K27ac ChIP-seq	This paper	GSE172407
ATAC-seq	This paper	GSE172407
p63/p73 ChIP-seq	This paper	GSE172407
RNA-sequencing of NT and p63/p73 KO lines	This paper	GSE171629
RNA-sequencing of control and p63/p73 OE lines	This paper	GSE171626
Single-cell RNA-sequencing of NT vs. dKO cells isolated from lungs	This paper	GSE171628
Experimental models: cell lines
Human 293T	Shibue, et al. 76	N/A
Human SUM159 ITGB4lo	Bierie et al. 28	N/A
Human SUM159 ITGB4hi	Bierie et al. 28	N/A
Human NAMEC8	Tam et al., 48	N/A
Human SUM159 ITGB4lo with ITGB4OE	Bierie et al. 28	N/A
Experimental models: organisms/strains
Mouse NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NSG)	The Jackson Laboratory	Cat#005557
Oligonucleotides
See Table S1
Recombinant DNA
lenti-CRISPR-V2-puro	Addgene	#52961
lenti-CRISPR-V2-GFP	Addgene	#82416
Luciferase-tdTomato	Frose et al. 77
ΔNp63alpha-FLAG	Addgene	#26979
pWZL-HRAS (G12V)-blast	Tam et al. 48
pLenti-CMV-rtta3-hygro	Addgene	#26730
doxycycline-inducible ΔNp63-FLAG	VectorBuilder
doxycycline-inducible TGFA	VectorBuilder
pENTR223-TAp73	Harvard PlasmID Database	Clone ID: HsCD003999879
pENTR1A	Thermo Fisher Scientific	Cat#A10462
pLenti6/V5-DEST	Thermo Fisher Scientific	Cat#V49610
pMD2.G	Addgene	#12259
psPax2	Addgene	#12260
Software and algorithms
FlowJo	BD Biosciences	https://www.flowjo.com/
Prism 8	GraphPad	https://www.graphpad.com/scientific-software/prism/
Living Image	PerkinElmer	https://www.perkinelmer.com/product/spectrum-200-living-image-v4series-1-128113
Zen	Zeiss	https://www.zeiss.com/microscopy/us/products/microscope-software/zen.html
Cell Ranger	10x Genomics	https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/what-is-cell-ranger
Loupe Browser	10x Genomics	https://support.10xgenomics.com/single-cell-gene-expression/software/visualization/latest/what-is-loupe-cell-browser
MACS2	Zhang et al. 78	https://pypi.org/project/MACS2/

Highlights.

• ΔNp63 and p73 maintain residence in the quasi-mesenchymal breast CSC state

• ΔNp63 and p73 are required for the post-extravasation proliferation of CSCs

• ΔNp63 controls distinct transcriptional programs in CSCs and normal basal SCs

• Quasi-mesenchymal CSCs use autocrine EGFR signaling to enable metastatic colonization