nMRTF BCC chromatin accessibility resembles HF matrix

We have previously shown that active nuclear MRTF (nMRTF) in BCCs confers Hh target gene transcription and persistent growth in the presence of SMO inhibitors (SMOⁱ) including vismodegib¹². However, the key driving factors that lead to active nMRTF remain unclear. To understand how MRTF interacts with the Hh pathway in the context of normal skin homeostasis, we examined MRTF cellular localization in mouse skin. Intriguingly, while BCCs are thought to derive from the bulge and outer root sheath (ORS)¹⁸, we found that nuclear MRTF is restricted specifically to the matrix transit-amplifying cells (TACs) of the HF and dermal papilla (DP) (Fig. 1a, Supplementary Fig. 1a), with other subdomains of the skin showing predominantly cytoplasmic localization. While both the ORS and TACs respond to Shh signaling, TACs are committed HF progenitors that undergo Shh-dependent expansion during anagen. Notably, the HF matrix contains cells that express the highest levels of Hh signaling¹⁹, and they are known to be highly responsive to TGFß and Wnt secreted from the DP^20,21 (Fig. 1a, Supplementary Fig. 1a).

a IF representative images of MRTF cellular localization in healthy mouse skin, in comparison to localization of HHIP (Hh-responsive matrix transit-amplifying cells), Keratin-14 (interfollicular epidermis and outer root sheath), and phospho-Smad2/3 (TGFß-responsive dermal papilla and matrix transit-amplifying cells). Inset areas (IFE top row, HF matrix bottom row) denoted by white dotted boxes. Single-color images are all derived from the same slide except for pSmad2/3, due to limitations of antibody co-staining. Images are representative of n > 50 hair follicles examined. Scale bar at low power = 50 μm, inset = 25 μm. b IF images of MRTF localization in clinical biopsies of basal cell carcinoma with nuclear or cytoplasmic MRTF (BCC-nMRTF or -cMRTF), pilomatricomas (PMX), epithelial inclusion cysts (EIC), and pilar cysts (PC). Images are representative of n = 3 patient tumors per subtype. Scale bar = 50 μm. c Principal component analysis (PCA) plot showing the relationship of ATAC-seq profiles between resistant BCC cell line ASZ001 (res-BCC), sensitive murine BCC (sens-BCC), basal transit-amplifying cells of hair follicle (basal TAC), bulge hair follicle stem cells (Bulge), hair germ (HG), interfollicular epidermal stem cells (IFE), and suprabasal transit-amplifying cells of the hair follicle (suprabasal TAC). See Supplementary Table 1 for data sources. d IF images of MRTF, Ki67, and Krt14 in mouse hair follicle skin explants treated with DMSO vehicle or 150 μM MRTFⁱ CCG-1423 for 72 h. Scale bar = 20 μm. e Quantification of Ki67+ cells per hair follicle in (d). n = 4 pairs of explants from three mice. ***p < 0.0001. f Representative H&E staining of DMSO or CCG-1423 treated mouse skin explants. Scale bar = 100 μm. g Quantification of hair follicle lengths in (f). Each point represents the length of one HF averaged over up to three separate measurements. **p = 0.0017. All error bars represent mean +/− SD. p-values were calculated using unpaired, two-tailed Student’s t-test.

To further characterize the differentiation state of nMRTF cells, we interrogated MRTF cellular localization in clinical samples of benign skin tumors with known cellular origins. Strong and uniform nuclear MRTF was observed in HF matrix-derived pilomatricomas (PMX), while epithelial inclusion cysts (EIC) derived from interfollicular epidermis (IFE) as well as trichilemmal/pilar cysts (PC) derived from ORS displayed cytoplasmic MRTF localization^22,23 (Fig. 1b). Furthermore, we found that out of 1201 MRTF target genes identified in resistant BCCs (Supplementary Data 1), 65% of those genes are upregulated in basal and/or suprabasal TACs²⁴. These observations reinforce the relation of MRTF activity with the hair matrix lineage.

In parallel, we took a global approach to analyze the chromatin landscape relationships between BCC and normal skin by comparing the Assay for Transposase Accessible Chromatin (ATAC-seq) profiles of resistant and sensitive BCCs with those of various epithelial cell types (Fig. 1c, Supplementary Fig. 1b, Supplementary Table 1). We compared open chromatin of murine nMRTF BCC cell line ASZ001²⁵, previously selected to be highly resistant to SMOⁱ-induced changes in cell growth (res-BCC)²⁶, with other published ATAC-seq datasets from sensitive BCC generated from K14-creER;Ptch1^fl/fl;Tp53^f/f mice (sens-BCC)²⁷, basal and suprabasal TACs (basal TAC and suprabasal TAC), bulge HF stem cells, hair germ (HG)²⁴, and interfollicular epidermal stem cells (IFE)²⁸. Principal component analysis (PCA) representation of these relationships reveals that the chromatin accessibility profiles of nMRTF resistant BCC cells cluster most closely with those of suprabasal and basal TAC, while sensitive BCC clusters with HG and bulge stem cells (Fig. 1c and Supplementary Fig. 1b).

We also compared nMRTF cells to residual BCC cells (resid-BCC), reversible Wnt-dependent, IFE-like tumor cells identified in a murine BCC model after vismodegib treatment²⁷. We see that multiple resistant nMRTF BCC cell lines (ASZ and BSZ²⁵) have quite disparate chromatin accessibility profiles from sensitive BCC as well as residual BCC (Supplementary Fig. 1d). Furthermore, the genes with increased chromatin accessibility in resistant nMRTF BCC comprise a uniquely pro-proliferative, pro-migratory program, suggesting they represent a distinct cell fate (Supplementary Fig. 1e). Significant phenotypic differences also exist between these BCC cell types, as nMRTF cells maintain high Hh signaling and proliferation rates when treated with vismodegib¹² (see below), while residual tumor cells resume growth only when treatment is discontinued²⁷. We conclude that SMOⁱ-resistant nMRTF BCCs possess a distinct cellular state within naive tumors most closely resembling HF matrix TACs.

The observation that MRTF-SRF-Gli chromatin occupancy confers SMO-independent enhancement of Gli activity suggests the intriguing hypothesis that nMRTF functions in the hair to extend hair matrix proliferation at a distance from the Shh signal localized to the lateral disc²⁹. Indeed, in mouse anagen HFs, proliferating matrix cells extend to the line of Auber on average 4–5 cell divisions and cease as nMRTF becomes cytoplasmic (Fig. 1a, d). We tested this hypothesis by treating mouse HF explants with CCG-1423, an MRTF-inhibitor (MRTFⁱ), and found markedly reduced proliferation in the matrix (Fig. 1d, e) as measured by Ki67 staining, and a concomitant decreased total HF length (Fig. 1f, g). Altogether, we conclude that MRTF nuclear localization is linked to a distinct cell state of differentiation, allowing it to extend the cellular response to Shh signaling.

LYPD3/TACSTD2/LY6D mark the nMRTF population in patient BCCs

We have shown previously that naive human BCCs, such as those excised through standard Mohs surgical techniques, contain a heterogenous number of cells displaying active nMRTF, which is predictive of the effectiveness of MRTFⁱ versus SMOⁱ for BCC treatments¹². Therefore, we took advantage of the naturally occurring heterogeneity and conducted scRNA-seq on four naive human BCC tumors to identify and segregate the MRTF-active vs. inactive subpopulations for further study (Supplementary Fig. 2a, b). From the 45,656 total cells analyzed, the epithelial tumor cells were segregated from fibroblast, endothelial, and immune populations through Krt14 expression (Supplementary Fig. 2c–g). Samples were batch-corrected using canonical correlation analysis (CCA) before dimensional reduction and visualization of clusters via tSNE (Fig. 2a).

a tSNE plot of unbiased clustering of tumor epithelia from 4 naive human BCC tumor scRNA-Seq datasets post multi-CCA alignment, filtered for positive Krt14 expression. b MRTF signature enrichment score (generated from genes in Supplementary Data 1) overlaid on tSNE clusters from (a). c Clusters ranked by mean MRTF signature enrichment score, then grouped as follows: clusters 1–3 = high MRTF, clusters 4–6 = med high, clusters 7–9 = med low, clusters 10–12 = low MRTF. d Heatmap of top 50 genes most enriched in high MRTF group (clusters 1–3) vs. low MRTF group (clusters 10–12). e Violin plots of gene expression levels per group for surface markers LYPD3, TACSTD2, and LY6D. f Representative IF images of surface marker expression in human and mouse MRTF-nuclear vs. MRTF-cytoplasmic BCC regions or HF matrix of normal mouse skin. scale bar = 50 μm. g Quantification of LYPD3, TACSTD2, and LY6D fluorescence intensity normalized to DAPI. Each point represents mean pixel intensity, normalized to mean DAPI intensity, averaged over at least three 100 × 100 μm microscopy fields. **p < 0.01, ***p < 0.001. h Quantification of LYPD3, TACSTD2, and LY6D fluorescence intensity normalized to DAPI of naive patient BCC explants treated with 10 μM MRTFⁱ CCG-1423 for 24 h. Each point represents mean pixel intensity, normalized to mean DAPI intensity, averaged over at least three 100 × 100 μm microscopy fields. *p = 0.0384, **p = 0.0072, ***p = 0.0002. All error bars represent mean  +/− SD. p-values calculated using unpaired, two-tailed Student’s t-test. i Relative reduction in Gli1 mRNA levels as measured by qRT-PCR vs. quantified fluorescence intensity of surface markers normalized to DAPI in naive human BCC (huBCC) explants treated with 1 μM vismodegib for 24 h. Respective linear regression r² values shown in matching colors. j Representative FACS plots showing distributions of ItgA6 (CD49f), LYPD3, TACSTD2 (TROP2), and LYPD3 protein expression in naive human BCC tumors. Final sorted populations taken for further analysis outlined in red and blue. k Gene Set Enrichment Analysis (GSEA) plot of MRTF signature gene list comparing RNA-seq of sorted surface marker positive (SM+) vs. negative (SM−) BCC cells. n = 4 replicates from two tumors. ES = enrichment score, p = nominal p-value⁶⁴. l Gene expression heatmap from GSEA in (k).

Cells with the highest MRTF activity were identified through an enrichment score based on the expression of a signature list of MRTF target genes (Fig. 2b). The list is derived from the intersection of SRF target genes identified by ChIP-seq, and MRTF-dependent genes identified by RNA-seq in resistant BCC cells treated with MRTF inhibitor CCG-1423¹² (Supplementary Fig. 2h, Supplementary Data 1). Clusters were then ranked based on their mean MRTF signature enrichment score, and the 12 clusters were separated into four groups: high MRTF, med-high, med-low, and low MRTF (Fig. 2c). Interestingly, the four tumors were not equally distributed among the clusters, as evidenced by the significant chi-squared statistic. The high MRTF clusters contained higher percentages of two of the four tumors (BCC8 and BCC3), consistent with our previous observation that at least 50% of naive clinical BCC samples contain nMRTF¹² (Supplementary Fig. 2i).

We then determined the most highly upregulated genes in the high MRTF group vs. low MRTF group (Fig. 2d). To separate MRTF-active BCC cells by fluorescence-activated cell sorting (FACS), we selected target genes of MRTF whose protein products are expressed on the cell surface. The top three most highly enriched surface markers in the MRTF high group were LYPD3, TACSTD2 (also known as TROP-2), and LY6D (Fig. 2e, Supplementary Fig. 2j–l). To validate the specificity of these surface markers, we immunostained human and mouse BCC tumors along with MRTF. We found the protein levels of all three markers was significantly higher in tumor regions with nMRTF (Fig. 2f, g). Interestingly, the surface markers are also expressed at high levels in the HF matrix in normal mouse skin, correlating with nuclear MRTF (Fig. 2f). We also treated naive patient BCC explants with MRTFⁱ CCG-1423 and observed significantly reduced protein levels of all three surface markers, further confirming that expression of these markers is dependent on MRTF activity (Fig. 2h). To correlate the expression of the surface markers with functional resistance, we treated naive patient BCC explants with SMOⁱ vismodegib, and found that protein expression of all three markers is significantly correlated with maintenance of Gli1 mRNA levels, indicating diminished response to SMO inhibition ex vivo (Fig. 2i). These data confirm that the three surface proteins LYPD3, TACSTD2, and LY6D are reliable indicators of nuclear MRTF activity as well as prognostic markers of SMOⁱ resistance in human BCC cells.

To further investigate the chromatin landscape and transcriptomic differences specific to resistant nMRTF BCC, we sorted fresh naive clinical BCC samples first based on epithelial ITGA6 expression, then on these three surface markers (Fig. 2j, Supplementary Fig. 2m). Interestingly, we see significant co-expression of LYPD3 and TACSTD2, while LY6D shows much scarcer expression levels, consistent with our scRNA-seq results (Fig. 2e, Supplementary Fig. 2j–l). For this reason, we sorted ITGA6+ LYPD3+ TROP2+ LY6D+/− cells as surface marker positive (SM+) population and ITGA6+ LYPD3⁻ TROP2⁻ LY6D⁻ cells as surface marker negative (SM−) and compared SM+ and SM− RNA-seq and ATAC-seq profiles (Supplementary Fig. 2p, q). Importantly, the expression of the MRTF signature gene list was significantly enriched by GSEA analysis in the SM+ population, further validating the three surface proteins as reliable markers of MRTF activity (Fig. 2k, l). As additional support of the specificity of these surface markers for resistant nMRTF BCC, we examined their expression in Gorlin syndrome patient BCCs, which are caused by inherited loss of ptch1 and respond to vismodegib with almost no resistance³⁰, as well as being consistently MRTF-inactive¹². Indeed, Gorlin’s syndrome BCC tumor cells lack expression of the three surface markers (Supplementary Fig. 2n). Chromatin accessibility profiles of Gorlin BCCs are more similar to sorted SM- than SM+ naive patient BCCs by PCA (Supplementary Fig. 2o), providing further support that the SM+ surface phenotype can prospectively enrich for BCC cells possessing the distinct resistant nMRTF cell fate, allowing further characterization from heterogeneous BCC populations.

BCC resistance requires coincident AP-1 and TGFß signaling

We used BETA analysis³¹ to integrate transcriptomic and chromatin accessibility data in order to identify differentially regulated genes and TF motifs between SM+ and SM− patient BCC cells (Supplementary Fig. 3a). Interestingly, the top Gene Ontology (GO) Biological Process terms³² enriched in the SM+ cells included terms associated with epidermal development and keratinocyte differentiation, underscoring the idea that resistant nMRTF BCC resemble a more differentiated cell state than sensitive BCC (Fig. 3a). To find driver TF pathways, we looked at enriched TF motifs as well as target gene expression from Chip-X Enrichment Analysis (ChEA)³³ (Fig. 3b, c). Notably, AP-1, Smad2/3, and p63 were among the top TFs acting in SM+ cells. As p63 is a known master regulator for epidermal development, we focused on AP-1 and Smad as potential activators of MRTF and non-canonical Hh signaling.

a Top Gene Onotology (GO) Biological Process terms enriched in upregulated genes of SM+ vs. SM− huBCC cells by Binding and Expression Target Analysis (BETA) integration of RNA-seq and ATAC-seq. p-values calculated using Fisher exact test. b Transcription factors with highest combined ChIP-X Enrichment Analysis (ChEA) score enriched in SM+ vs. SM− huBCC cells by BETA integration of RNA-seq and ATAC-seq. c Transcription factor motifs enriched in differentially open chromatin peaks in SM+ vs. SM− huBCC cells by BETA integration of RNA-seq and ATAC-seq. d Gli1 qRT-PCR in resistant BCC cell line ASZ001 treated with T5224 (AP-1 inhibitor) or SB431542 (ALK5 inhibitor) for 24 h, normalized to DMSO control. Vertical dotted line represents published IC50 in culture. e Cell viability of resistant BCC cell line ASZ001 treated with inhibitors for 72 h, measured by MTS assay. f Gli1 qRT-PCR of resistant BCC cell line ASZ001 transfected with siRNAs against AP-1 and TGFß family genes for 48 h, compared to universal negative control siRNA. Each pair of matching-colored bars represents two distinct siRNA oligos per target gene. **p < 0.01, ***p < 0.001. g Rho G-LISA assay with negative control cells serum starved for 24 h, and positive control cells stimulated with 10% FBS for 5 min. ***p < 0.001. Open circles on all bar graphs represent independent biological replicates. h IF staining of resistant BCC cell line ASZ001 treated with DMSO, 20 μM T-5224 (AP-1 inhibitor), or 10 µM SB431542 (ALK5 inhibitor). Scale bar = 25 μm. i Quantification of MRTF fluorescence intensity across cell radius in (h), measured as µm distance from cell center. Nuclear boundaries represented as DAPI intensity. Mean intensities of n > 50 cells shown as solid lines, with SEM as dotted lines. j MTS cell viability assay of resistant BCC cell line ASZ001 treated with increasing doses of vismodegib (SMOⁱ) only, or vismodegib + 10 μM SB431542 (ALK5 inhibitor) or 5 μM T5224 (AP-1 inhibitor). ***p < 0.001. All error bars represent mean +/− SD. p-values calculated using unpaired, two-tailed Student’s t-test.

First, we assessed the necessity of TGFß and AP-1 signaling in resistant BCC through genetic and pharmacologic perturbation. Treatment of BCC cells with various AP-1 small-molecule inhibitors (T5224³⁴ and SR11302³⁵) resulted in a dose-dependent decrease in expression of Gli1 measured by mRNA and protein, and cell viability (Fig. 3d, e and Supplementary Fig. 3b, c). We also observed elevated levels of phosphorylated JNK in active nMRTF BCCs in comparison to inactive cMRTF BCC (Supplementary Fig. 3g, h), and treatment of resistant BCC cells with inhibitors of JNK (SP600125³⁶ and JNK-IN-8³⁷) reduce Gli1 expression and cell viability (Supplementary Fig. 3d). By contrast, inhibitors of p38 or MEK fail to demonstrate similar inhibition¹², suggesting that AP1/JNK signaling mediates resistance. In parallel, a small-molecule inhibitor of TGFß signaling through ALK5 (SB435142³⁸) leads to a dose-dependent decrease in phosphorylated Smad3 levels (Supplementary Fig. 3e), as well as Gli1 expression and cell viability (Fig. 3d-e). Both AP-1 inhibitor and ALK5 inhibitor were preferentially toxic to BCC cells, as they affected cell viability of multiple BCC cell lines²⁵ to a significantly higher degree than noncancerous cell lines (Supplementary Fig. 3i), providing key pre-clinical data for a drug therapeutic window. Notably, the combination of AP-1 and ALK5 inhibitors on resistant BCC cells did not result in any additional effect on Gli1 expression or cell viability, suggesting that these pathways may be redundant or acting upstream of a shared pathway (Supplementary Fig. 3f).

We next wanted to determine the specific AP-1 and TGFß family members that operate in resistant BCCs. Although AP-1 consists of a family of Jun/Fos dimers all capable of binding to the TGA(C/G)TCA consensus sequence, different homo or heterodimer pairs can have drastically different transcriptional outputs³⁹. c-Jun, JunB, and JunD are all expressed robustly in BCC cells, while FosL2 is the only Fos family member expressed at significant levels (Supplementary Fig. 3k). Consistent with this observation, siRNA knockdowns of c-Jun, JunB, JunD, or FosL2 in BCC cells resulted in significant decrease in Gli1 expression, with JunD knockdown having the biggest effect, while knockdown of Fos or FosB did not change Gli1 expression (Fig. 3f, Supplementary Fig. 3l). In parallel, siRNA knockdown of TGFß family members TGFß1, TGFß3, Smad3, and Alk5 decreased Smad2/3 phosphorylation (Supplementary Fig. 3m) and reduced expression of Gli1 (Fig. 3f, Supplementary Fig. 3l, m).

nMRTF activity depends on RhoA activation¹², so we next examined whether AP-1 and TGFß cooperate to induce Rho signaling. Using a Rho G-LISA assay to specifically measure levels of active GTP-bound RhoA, we found that exogenous TGFß3 ligand was sufficient to activate Rho signaling, which could be attenuated by the addition of ALK5 or AP-1 inhibitor (Fig. 3g). In addition, we found that inhibition of TGFß or AP-1 signaling disrupted both actin polymerization and MRTF nuclear localization (Fig. 3h, i). Importantly, while TGFß signaling regulates Rho activity in both Smad-dependent and independent pathways⁴⁰, knockdown of Smad3 expression is sufficient to inhibit MRTF nuclear localization (Fig. 4g) and Gli1 expression (Fig. 3f). This suggests that in resistant BCCs, TGFß acts in a Smad3-dependent manner to promote non-canonical Hh signaling.

a Heatmap of differentially expressed genes as measured by RNA-seq of resistant ASZ001 cells treated with 10 μM SB431542 (ALK5 inhibitor) or 20 μM T5224 (AP-1 inhibitor). b Overlap between AP-1 and TGFß dependent genes by RNA-seq, defined as log₂ FC < −1 and p > 0.05 in inhibitor-treated cells. Genes listed in Supplementary Data 2. c GO Molecular function terms enriched in genes dependent on both AP-1 and TGFß signaling. P-values calculated by Fisher exact test. d Gli1 qRT-PCR of resistant BCC cell line ASZ001 transfected with siRNAs targeting selected GEFs for 48 h. Each pair of matching-colored bars represents two distinct siRNA oligos per target gene. *p < 0.05, **p < 0.01. e IF images of MRTF and Arhgef17 protein expression in murine sensitive (cMRTF) and resistant (nMRTF) BCCs¹². Scale bar = 50 μm. f Quantification of (e) as measured by Arhgef17 fluorescence intensity vs. DAPI. Each point represents an individual tumor, mean pixel intensity normalized to mean DAPI intensity quantified over at least three 100 × 100 μm microscopy fields. *p = 0.0107. g IF images of MRTF protein localization in resistant BCC cell line ASZ001 transfected with Cy-3 conjugated siRNAs. Scale bar = 25 μm. h Quantification of MRTF fluorescence intensity across cell radius in (g), measured as µm distance from cell center. Nuclear boundaries represented as DAPI intensity. Mean intensities of n > 50 cells shown as solid lines, with SEM as dotted lines. i Diagram describing putative signaling pathway and corresponding small-molecule inhibitors. j Epistatic studies measured by Gli1 qRT-PCR of resistant BCC cell line ASZ001 transfected with overexpression constructs and treated with inhibitors. ***p < 0.001. Open circles on all bar graphs represent independent biological replicates. All error bars represent mean +/− SD. P-values calculated using unpaired, two-tailed Student’s t-test.

If AP-1 and/or TGFß promote a non-canonical resistance mechanism to drive Hh signaling, then inhibiting these TF pathways should increase the sensitivity of cells to canonical SMOⁱ vismodegib. Indeed, we see that combining SMOⁱ treatment with AP-1 inhibition or ALK5 inhibition in resistant BCC cells leads to increased levels of cell death than with SMOⁱ alone (Fig. 3j). In contrast, neither AP-1 inhibition nor ALK5 inhibition have any additive effect on cell viability when combined with MRTFⁱ CCG-1423 (Supplementary Fig. 3j), suggesting that these pathways work upstream of MRTF. These initial experiments connect JNK-mediated signaling via Jun/FosL2 and TGFß signaling via Smad3 to maintain activation of Rho and MRTF, leading to non-canonical Hedgehog signaling in resistant BCC cells.

AP-1 and Smad3 induce transcription of Rho GEFs

To identify how AP-1 and Smad3 transcriptional targets activate Rho, we performed RNA-seq on resistant BCC cells treated with ALK5 or AP-1 inhibitors. Intriguingly, the top Molecular Function GO terms of genes dependent on both pathways included guanyl-nucleotide exchange factor (GEF) activity, proteins that directly activate Rho family GTPases (Fig. 4a–c, Supplementary Data 2). Previous studies have shown that TGFß regulates Rho GTPases through transcription of GEFs in contexts such as epithelial-mesenchymal transition, making these promising candidate genes^41,42. Targeted siRNA knockdown of the differentially expressed RhoGEFs identified the set of GEFs that work together to maintain Gli1 expression (Fig. 4d, Supplementary Fig. 4a, b). Although maximal Rho activity required several exchange factors, Arhgef17, also known as Tumor Endothelial Marker 4, was the most significantly enriched at the protein level in mouse resistant BCC tumors with nuclear MRTF, generated from transgenic mouse model Ptch1^+/−;K14-creER;p53 ^fl/fl43 (Fig. 4e, f). Arhgef17 facilitates GDP/GTP exchange for RhoA and is required for cell-cell adhesion⁴⁴. Knockdown of Arhgef17 as well as related GEF Arhgef18 using fluorescently labeled siRNAs phenocopied Smad3, JunD, and other AP-1 subunit knockdowns in abrogating MRTF nuclear localization (Fig. 4g, h and Supplementary Fig. 4c).

These findings suggest coincident TGFß and AP-1 signaling confers resistance by increasing transcription of Rho GEFs such as Arhgef17, which in turn activate RhoA and subsequent actin polymerization, leading to nuclear localization of MRTF and SRF, which as act non-canonical cofactors for Gli1 (Fig. 4i)¹². In order to confirm this resistance pathway, we conducted a series of epistatic studies using small-molecule inhibitors or siRNA knockdowns at various stages of the pathway (Fig. 4i) simultaneously paired with overexpression constructs, measuring Gli1 expression as the final output (Fig. 4j, Supplementary Fig. 4d, e). Treatment with SMO inhibitor vismodegib leads to a decrease in Hh signaling, which can be rescued by concurrent administration of TGFß3 ligand or transient overexpression of JunD, Arhgef17, RhoA or constitutively active MRTF (MRTF-N) (Fig. 4j). Importantly, the effects of ALK5 inhibition cannot be rescued by overexpression of JunD or other AP-1 subunits and conversely, AP-1 inhibition cannot be rescued by TGFß ligand administration (Fig. 4j, Supplementary Fig. 4e). These experiments together indicate that coincidental TGFß and AP-1 signaling are required for Arhgef17, RhoA, and MRTF activation that leads to non-canonical Hh signaling and tumor resistance.

JunD/AP-1, but not TGFß, is sufficient to drive nMRTF

To interrogate whether AP-1 or TGFß are sufficient to drive MRTF-mediated resistance, we used our NIH-3T3 model where overexpression of constitutively active MRTF (MRTF-N) in tandem with subthreshold Smoothened agonist (SAG) enhances Gli signaling¹². 3T3 cells are a useful proxy for sensitive BCC cells in vitro because they respond to canonical Hh signaling but do not intrinsically express the non-canonical MRTF-driven resistance pathway¹². We confirmed that the expression of MRTF-N, Arhgef17, Arhgef18, or RhoA is sufficient to amplify Gli1 mRNA expression (Fig. 5a). Interestingly, we find that overexpression of JunD as well as other Jun family members is sufficient to amplify Gli1, whereas increased TGFß signaling alone is not sufficient (Fig. 5a, Supplementary Fig. 4f). We find that this sufficiency is operating at the transcriptional level, as JunD but not TGFß overactivation is sufficient to increase the transcription of selected GEFs (Fig. 5b). We confirmed that a basal level of TGFß signaling exists and responds to ligand stimulation with significant elevation in phosphorylated Smad2/3 levels (Supplementary Fig. 4g) and expression of the canonical target gene Serpine1 (Fig. 5b).

a Enhancement of Hh signaling in NIH-3T3 cells measured by Gli1 qRT-PCR transiently transfected with overexpression constructs for 48 h and treated with 30 μM Smoothened agonist (SAG) for 24 h. **p < 0.01. b qRT-PCR of various target genes in 3T3s transiently transfected with GFP control vectorwith or without 5 ng/ml TGFß ligand supplementation, or JunD overexpression construct. ***p < 0.001. c Gli1 qRT-PCR in 3T3 cells transiently transfected with GFP control vector or JunD and treated with 30 μM Smoothened agonist (SAG) for 24 h. ***p < 0.0001. d RhoA activation quantified by G-LISA assay in 3T3 cells transiently transfected with GFP control or overexpression constructs for 48 h. **p < 0.01. Open circles on all bar graphs represent independent biological replicates. All error bars represent mean +/− SD. P-values calculated using unpaired, two-tailed Student’s t-test. e IF images of 3T3 cells transfected with HA-tagged MRTF construct with or without 5 ng/ml recombinant TGFß ligand supplementation, Arhgef17 or JunD overexpression construct. Scale bar = 25 μm. f Quantification of MRTF intensity across cell radius in (e), measured as µm distance from cell center. Nuclear boundaries represented as DAPI intensity. Mean intensities of n > 50 cells shown as solid lines, with SEM as dotted lines.

Consistent with a primary role in driving nMRTF activity, the effects of JunD overexpression can be attenuated with inhibition of either ALK5 activity, Arhgef17 expression, or MRTF activity (Fig. 5c). Overexpression of Arhgef17 or JunD is sufficient to increase RhoA activation (Fig. 5d) and MRTF nuclear translocation in 3T3 cells (Fig. 5e, f, Supplementary Fig. 4i, j). Importantly, AP-1 activity appears to have little effect on canonical Hh signaling driven by SMO, as its inhibition does not affect Gli1 expression levels in Hh-responsive 3T3 or C2C12 cells treated with SAG (Supplementary Fig. 4h). Therefore, we conclude that JunD/AP-1 is sufficient to amplify non-canonical Gli activity that depends on TGFß, RhoGEFs, RhoA, and MRTF signaling.

AP-1 establishes Smad3 DNA binding profile of resistant BCC

AP-1 has been shown to commission enhancers in conjunction with cell-type specific TFs⁴⁵. Due to JunD, but not Smad3 sufficiency in driving the AP-1/Smad/Rho/MRTF resistance pathway, we hypothesized that JunD/AP-1 regulates the unique chromatin accessibility profile that functions with TGFß signaling in resistant nMRTF BCC. Indeed, AP-1 binding motifs are highly enriched in the differential open chromatin of sorted SM+ naive human BCC (Fig. 3c), while few differences arose in canonical Smad3 binding motifs. This is consistent with previous findings that Smad proteins bind to their canonical DNA binding element (SBE) with 100-fold lower affinity than their interacting TF partners⁴⁶, and therefore need to cooperate with other TFs to influence transcription. To test this hypothesis, we conducted ATAC-seq and phospho-Smad3 chromatin immunoprecipitation sequencing (ChIP-seq) in resistant nMRTF BCC cells treated with AP-1 inhibitor. We found that AP-1 increases chromatin accessibility at Smad3 binding sites and modifies Smad3 DNA binding on a genome-wide level (Fig. 6a, Supplementary Fig. 5a–c, e). Interestingly, the intersection of genes dependent on AP-1 for maximal mRNA expression, chromatin accessibility, and Smad3 binding are responsible for interaction with the cytoskeleton and RhoGEF activity (Fig. 6b, c, Supplementary Data 3). These findings suggest that in resistant BCC, AP-1 shapes the chromatin accessibility landscape to open alternative Smad3 binding sites, allowing AP-1 and Smad3 to cooperatively induce expression of Rho regulators including GEFs.

a Heatmap and line graphs of ATAC-seq signal in resistant ASZ001 cells treated with and without 20 μM AP-1 inhibitor T5224 across pSmad3 ChIP binding sites. b Overlap of genes displaying loss of chromatin accessibility (ATAC-seq), Smad3 binding (ChIP-seq), and/or expression (RNA-seq) in response to AP-1 inhibitor treatment in resistant BCC cells, defined as log₂ FC < −1 and p > 0.05. Genes listed in Supplementary Data 3. c GO molecular function terms enriched in genes showing AP-1 dependence of chromatin accessibility, Smad3 binding, and mRNA expression levels. P-values calculated by Fisher exact test. d Visualization of ATAC and ChIP peaks at Arhgef17 regulatory locus. ATAC peak of interest has been highlighted, and scissors represent target sites for CRISPR guide RNAs. e List of RhoGEFs dependent on AP-1 by chromatin accessibility, Smad3 binding, and mRNA expression levels in resistant BCC cells. f Arhgef17 qRT-PCR in WT ASZ cells and Arhgef17 ATAC-peak deletion (Arhgef17^AD) cell line. ***p < 0.0001. g Gli1 qRT-PCR in WT and Arhgef17^AD ASZ cell lines, treated with 5 ng/ml of TGFß3 ligand for 24 h or 1 μg/ml of Rho activator II for 6 h. *p = 0.0344, ***p = 0.0003. h Gli1 qRT-PCR in WT and Arhgef17^AD ASZ cell lines transiently transfected with overexpression constructs. ***p = 0.0002. Open circles on all bar graphs represent independent biological replicates. All error bars represent mean +/− SD. P-values calculated using unpaired, two-tailed Student’s t-test. i Gli1 qRT-PCR in WT and Arhgef17^AD ASZ cell lines, treated with increasing dosages of vismodegib or CCG-1423. Calculated IC50 of ASZ WT shown in blue, ARHGEF17^AD in red. **p < 0.01, ***p < 0.001. j IF images of MRTF protein localization and actin polymerization in WT ASZ and Arhgef17^AD cell lines. k Quantification of MRTF fluorescence intensity across cell radius in (j), measured as µm distance from cell center. Nuclear boundaries represented as DAPI intensity. Mean intensities of n > 50 cells shown as solid lines, with SEM as dotted lines. l Correlation of relative reduction in Gli1 mRNA expression of human tumor explants treated with 40 μM AP-1 inhibitor T-5224 for 24 h, with their relative intensity of surface marker immunostaining. Tumors were categorized as nuclear MRTF if immunofluorescent staining of MRTF colocalizing with DAPI was present in at least one of four separate 200 μm × 200 μm microscopy fields, otherwise they were categorized as cytoplasmic MRTF.

To illustrate the AP-1/Smad3 chromatin cooperativity, we focused on accessibility changes in the regulatory regions of the Arhgef17 locus, although we observed similar patterns in the regulatory regions of other GEF loci (Fig. 6d, e, Supplementary Fig. 5d). We identified an AP-1 dependent ATAC peak in resistant BCC cells within the first intron, which is closed in SMOⁱ-sensitive BCCs²⁷, and showed that Smad3 binding within the ATAC peak disappears with AP-1 inhibition (Fig. 6d). To interrogate the function of this AP-1 dependent open chromatin region, we used the CRISPR-Cas9 system in resistant nMRTF BCC cells to delete the region of the ATAC peak (Supplementary Fig. 5f). Care was taken to ensure the neighboring exon was excluded from the targeted region (Fig. 6d) to avoid disrupting protein function. Arhgef17 ATAC-peak deletion (Arhgef17^AD) cells displayed diminished expression of Arhgef17 (Fig. 6f), and significant reduction of Gli1 levels, which could be rescued by overexpression of full-length Arhgef17, RhoA, or MRTF-N, but not TGFß3 ligand stimulation (Fig. 6g, h). Arhgef17^AD cells revealed diminished actin polymerization and decreased levels of nuclear MRTF (Fig. 6j, k), similarly to Arhgef17 siRNA knockdown (Fig. 4g, h). Furthermore, the Arhgef17^AD cells also display enhanced sensitivity to inhibition of Gli1 expression by SMOⁱ or MRTFⁱ (Fig. 6i). These genetic studies show that resistance-specific, AP-1-dependent accessible chromatin in regulatory regions of Arhgef17 and likely other Rho activator genes are crucial for optimal Smad3 binding and transcription.

Since AP-1 signaling drives SMOⁱ resistance through MRTF activation, we predicted that in treatment of patient tumors, AP-1 inhibitors, like MRTF inhibitors¹², would only display efficacy in BCC populations with nMRTF. Indeed, we see that in naive human BCC explants, tumors containing nuclear MRTF respond significantly to AP-1 inhibition with a decrease in Gli1 expression, while tumors with only cytoplasmic MRTF are less responsive. Furthermore, there is a significant correlation between the expression of surface markers LYPD3, TACSTD2, and LY6D and the responsiveness to AP-1 inhibition (Fig. 6l). We also see that patient BCC explant treatment with AP-1 inhibitors results in significantly reduced levels of nuclear MRTF (Supplementary Fig. 5g, h). We conclude that AP-1 drives the chromatin accessibility profile conducive to AP-1/Smad3-dependent nMRTF BCC resistance, and the identified surface proteins can act as prognostic markers for response rate to AP-1 inhibitors.

To further evaluate the clinical potential of these pathway inhibitors, we treated naive patient BCC explants with combinations of ALK5, AP-1, and/or SMO inhibitors. Similar to our findings in the mouse BCC cell line, SMO plus AP-1 inhibitors have an additive effect in reducing Gli1 expression, while AP-1 plus ALK5 inhibitors do not (Supplementary Fig. 5i). These findings support the potential efficacy of combination therapies targeting the canonical and non-canonical pathways simultaneously.