Main

Basal cell carcinoma is the most frequently occurring human cancer and results from aberrant activation of the Hedgehog (Hh) pathway4. Erivedge (vismodegib), a potent inhibitor of Hh signalling that acts at the level of Smoothened (SMO)5, has been approved for the treatment of locally advanced and metastatic BCC1,6,7, and is also effective for operable BCC2. However, the low rate of histological clearance observed raised concerns over whether residual BCCs can regrow once treatment is discontinued.

To determine how residual BCCs survive vismodegib treatment, we used a mouse model of BCC driven by inactivation of Ptch1 and Trp53 in skin basal cells3. Histological examination of skin from 8-week-old K14CreER;Ptch1fl/fl;Trp53fl/fl mice revealed superficial BCC, similar to the human condition (Fig. 1a, b). These tumours expressed high levels of the Hh target gene Gli1 and the BCC marker SOX98 (Fig. 1f, g). Ki67 staining revealed that roughly 50% of the tumour nests contained actively proliferating cells (Fig. 1h).

Fig. 1: BCCs persist in the absence of Hh signalling.
figure 1

a–d, Representative images of haematoxylin and eosin (H&E)-stained skin sections showing response of mouse (n = 6 per group) and human BCC (n = 24, cohort 1 of ref. 2) to vismodegib (vismo) treatment. a, Skin section from an 8-week-old BCC mouse. b, Skin biopsy from a patient with BCC (pre-Rx, pre-treatment). c, Residual disease in a BCC mouse. White dashed line denotes boundary between epidermis and dermis. d, Residual disease (outlined in red) in the patient shown in b. e, Histology scores (mean ± s.d.; circles show individual scores) of skin samples from BCC mice treated with vehicle (black) or vismodegib (grey; n = 4 per group). f–k, Representative images of Gli1, SOX9 and Ki67 expression in mouse BCC and residual disease (n = 4 per condition). f, The Hh pathway is active in mouse BCC. g, Mouse BCCs stain positive for SOX9. h, Half of the tumours stain positive for Ki67. i, The Hh pathway is blocked in residual BCCs during vismodegib treatment. j, Residual disease identified by Sox9. k, BCCs stop proliferating within 1 day of vismodegib treatment. All experiments were replicated at least twice. Scale bars: b, d, 90 μm; other panels, 100 μm. DAPI nuclear stain is in blue (g, h, j, k); n represents the number of either mice or patients.

Source data

Full size image

To test the effect of Hh pathway inhibition on established BCC, we treated 8 week-old mice with vismodegib for 28 days. Skin from treated mice frequently contained small residual tumours similar to the finger-like basaloid projections observed in patients2 (Fig. 1c, d). The absence of Gli1 expression in tumour nests indicated that vismodegib effectively blocked Hh signalling (Fig. 1i). Residual BCCs still expressed SOX9 (Fig. 1j) and stopped proliferating completely within 1 day of treatment initiation (Fig. 1k). Apoptosis was absent in untreated disease, but was observed in a small number of tumour cells early during treatment (Extended Data Fig. 1a–d). Residual tumours did not differentiate into normal epidermis (Extended Data Fig. 2) and remained present even after 56 days of drug treatment (Fig. 1e). Thus, residual BCCs in mice are quiescent and do not require active Hh signalling for their survival.

We next investigated whether residual lesions that survived drug treatment could reinitiate growth. Skin sections from BCC mice collected 0, 3, 6, or 12 days after the last dose revealed that superficial BCC returned quickly (Fig. 2a), with Gli1+ lesions resembling either residual tumours or newly formed tumour buds (Fig. 2b–d). Elevated Gli1 expression persisted as mice remained off drug for longer periods of time (Extended Data Fig. 1e, f). To distinguish growth of residual disease from de novo tumour formation due to inherent leakiness9 of the K14CreER, we labelled tumours with BrdU before treatment to mark residual disease (Fig. 2e, Extended Data Fig. 1g, h). Six days after cessation of vismodegib, labelled tumour nests stained positive for Ki67 and were diluting the incorporated BrdU, confirming that quiescent residual BCCs reinitiated tumour growth when drug treatment was discontinued (Fig. 2f, g).

Fig. 2: Residual BCCs resume growth when vismodegib is discontinued.
figure 2

a, Histology scores (mean ± s.d.) from BCC mice treated with vehicle or vismodegib for 28 days determined after the indicated number of days of recovery (rec) (n ≥ 3 per group). b–d, Representative images of Gli1 ISH on skin from BCC mice shown in a. b, Hh pathway is blocked in residual BCC immediately after treatment. c, d, A residual BCC (c, outlined) and a newly formed tumour bud (d, outlined), both with an active Hh pathway 3 days after treatment. e–g, Lineage tracing of residual BCCs and their subsequent growth after treatment. Representative images of residual BCCs stained for BrdU (green), Ki67 (red) and DAPI nuclear stain (blue) are shown (n = 4 per time point). e, BrdU labelling strategy of residual BCCs. d0, day 0 of recovery. f, BrdU+ residual tumour nests (outlined) are quiescent immediately after treatment. g, Residual BCCs have resumed growth 6 days after treatment. All experiments were replicated at least twice. Scale bars are 100 μm; n represents the number of mice.

Source data

Full size image

To uncover the mechanisms of BCC persistence, we performed RNA sequencing (RNA-seq) and compared the transcriptional profiles of untreated Ki67+ and Ki67− BCCs to those of residual tumours. Independent of proliferation status (Extended Data Fig. 3c), vismodegib affected the expression of known Hh targets (Fig. 3a) and a large number of other genes relative to untreated tumours (Extended Data Fig. 3a, b). Comparison with published skin compartment signatures (Extended Data Table 1) revealed that genes enriched in hair follicle bulge stem cells were downregulated, whereas genes enriched in basal cells from the interfollicular epidermis (IFE) or isthmus (ISTH) showed upregulation after treatment (Fig. 3b, Extended Data Fig. 3d–f). There was no significant overlap with hair germ, dermal papillae or outer root sheath (ORS) signatures (Fig. 3b, Extended Data Table 1). To confirm the shift from a hair follicle bulge to a mixed IFE/ISTH identity, we performed in situ hybridization (ISH) using key markers of these skin compartments. The transcription factor Lhx2, which has a key role in hair follicle bulge stem cells10 (Extended Data Fig. 3g), was abundant in untreated BCCs, but strongly depleted in residual disease (Fig. 3c, f, Extended Data Fig. 3l). On the other hand, IL-33 and Defb6, which are expressed in the normal IFE and ISTH, respectively (Extended Data Fig. 3h, i), were absent in established BCCs (Fig. 3d, Extended Data Fig. 3j), but induced in residual disease (Fig. 3g, Extended Data Fig. 3k), as was the ISTH marker MTS2411 (Extended Data Fig. 3l). Similar results for LHX2 and IL-33 were obtained in a lineage-tracing experiment with a Cre-inducible TdTomato reporter (Extended Data Fig. 4a–d), confirming that the cell identity shift occurs in residual tumour cells. Notably, the absence of IL-33 staining in untreated tumours and gradual appearance of this nuclear factor12 in residual disease (Extended Data Fig. 5) suggests that vismodegib induces a cell identity switch rather than selecting for pre-existing ISTH/IFE-like cells. The shift in cell identity was reversible upon cessation of drug treatment, as the proportion of BrdU+ residual tumour cells expressing LHX2 increased progressively in our BrdU labelling experiment (Extended Data Fig. 4e–i). Furthermore, GATA6 and KLF5 are normally restricted in skin to the ISTH and IFE, respectively13,14. Both transcription factors were undetectable in untreated BCCs, but were co-expressed in residual tumour cells (Fig. 3e, h), suggesting that an induced state of ‘lineage infidelity’14 may contribute to the development of residual disease.

Fig. 3: Residual BCCs adopt an ISTH/IFE fate upon vismodegib treatment.
figure 3

a, Heat-map of Hh target gene expression in untreated and residual BCCs (n = 5 per group; data are represented as mean-centred normalized reads per kilobase of transcript per million mapped reads (nRPKM)). b, Quantitative set analysis for gene expression (QuSAGE) of indicated signatures in residual versus Ki67+ untreated BCC (n = 5 per group). Coloured violins depict differential expression of entire gene set; grey dots represent the log2[fold change] in expression of individual genes. BU, hair follicle bulge. c, d, f, g, Representative images of Lhx2 (hair follicle bulge) and Il33 (IFE) expression in untreated and residual BCCs (n = 4 per condition). c, Untreated BCC with high levels of Lhx2 mRNA. d, Untreated BCC (outlined) lacking IL-33 staining. Red arrowheads indicate nuclear localization of IL-33 in the epidermis. f, Residual BCCs downregulate Lhx2. g, Residual BCCs express IL-33 (red arrowheads). e, h, Representative images of KLF5 (IFE) and GATA6 (ISTH) antibody staining in untreated and residual BCC (n = 4 per condition). Untreated BCCs (e) lack expression of these lineage-specific transcription factors, whereas residual tumour cells (h, outlined) co-express KLF5 (green) and GATA6 (red). i, Regulatory potential exerted by chromatin regions (opening, green; closing, blue; unchanged, grey) on indicated gene sets (n = 3 per group). Box plots show median, two hinges (25th and 75th percentile), two whiskers (1.5 × inter-quartile range (IQR)), and all outlying points individually. Orange bars indicate the regulatory potential when using random genes. j, Graph depicting the association between differentially accessible enhancers with indicated transcription factor binding motifs and differentially expressed genes. A one-tailed Kolmogorov–Smirnov test was used to determine significance (n = 3 per group). k, Chromatin traces showing mean ATAC peaks at the Lhx2 and Klf5 loci. U, untreated; V, vismodegib-treated. l, Expression of Lhx2 and Klf5 in FACS-sorted BCC cells (n = 3 per group; data represented as mean ± s.d. log2 normalized read counts per million (CPM)). All experiments were replicated at least twice. Scale bars, 100 μm; n represents the number of mice.

Source data

Full size image

To better define the mechanisms that drive this identity switch, we performed assay for transposase-accessible chromatin with high-throughput sequencing (ATAC–seq) coupled with RNA-seq on tumour cells isolated by fluorescence-activated cell sorting (FACS) before and after either 6 or 14 days of treatment (Extended Data Fig. 6a–f, i). Vismodegib had a relatively small effect on chromatin accessibility, as the majority of the open chromatin regions were shared between treatment groups (Extended Data Fig. 6g). Assessment of the regulatory potential15 exerted by differentially accessible chromatin on skin compartment signatures did not show enrichment for either opening or closing regions at these loci (Fig. 3i). Instead, these genes were more closely associated with constitutively open chromatin, indicating that in BCC, lineage-specific programs harbour a globally open chromatin structure that is poised for activation, similar to the small intestine16.

We next assessed whether chromatin accessibility correlated with general changes in gene expression. Closing regions were associated with vismodegib-downregulated genes, whereas opening chromatin was associated with upregulated genes (Extended Data Fig. 6j, k). Closing regions were enriched for GLI, the forkhead box (FOX) family, and LHX binding motifs (Extended Data Fig. 7a), consistent with inhibition of the Hh pathway and the BCC transcriptome resembling the hair follicle bulge compartment. Regions that opened during vismodegib treatment were enriched for binding motifs corresponding to SOX, NFAT, KLF and GATA transcription factors (Extended Data Fig. 7b), consistent with treated BCCs shifting towards an IFE and ISTH identity. Binding and expression target analysis (BETA) confirmed that peaks containing GLI, LHX2, SOX and KLF motifs were strongly associated with changes in gene expression (Fig. 3j, Extended Data Fig. 7c). Notably, TCF binding motifs were enriched in opening chromatin at the 6-day time point (Extended Data Fig. 7b), including the loci of the ISTH- and IFE-specific transcription factors Gata6 and Ahr13 (Extended Data Fig. 6h), suggesting that Wnt has an early role in mediating the identity switch.

Control of cell identity is thought to be regulated through binding and remodelling of chromatin regions termed super-enhancers, which control the expression of Lhx2 and Klf5 in hair follicle stem cells17. Strikingly, vismodegib-sensitive ATAC peaks were found in super-enhancers at both loci (Fig. 3k). The Lhx2 super-enhancer contains a GLI binding site that closed upon vismodegib treatment, whereas the Klf5 super-enhancer harbours two TCF binding sites that became more accessible. Accordingly, the expression of Lhx2 and Klf5 was modulated by vismodegib treatment (Fig. 3l). Collectively, our data support an escape mechanism of drug-induced cell identity switching in BCC that features a broadly open chromatin structure that remains sensitive to the activity of key transcription factors.

We next set out to identify environmental cues that drive cell identity switching in BCC and, on the basis of the above results, focused on Wnt. Consistent with published reports showing that Wnt is required during BCC formation18,19, initiating tumours expressed the universal Wnt target Axin220 (Fig. 4a). Unexpectedly, Axin2 was suppressed in established tumours, but reactivated in residual disease (Fig. 4a). Levels of Axin2 expression were inversely correlated with levels of Wif1, a potent secreted inhibitor of Wnt signalling21 (Fig. 4a, b), and were induced as early as 2 days after the initiation of treatment (Fig. 4b), well before significant tumour shrinkage occurred (Extended Data Fig. 8a, b). Expression of Wnt4 and Wnt10a, which have been implicated in the maintenance of normal basal epidermis22, followed the same pattern (Fig. 4b). Human tumours showed a similar pattern; AXIN2 mRNA was detectable in BCCs only after vismodegib treatment (Extended Data Fig. 8c–h). In addition to Lhx2, we also observed a return to pre-treatment levels of Axin2 and Wif1 in tumours that had been released from drug for 12 days (Extended Data Fig. 8i), again highlighting the plasticity of residual tumours.

Fig. 4: Wnt signalling is required for maintaining residual disease.
figure 4

a, Representative images of Axin2 and Wif1 ISH on mouse skin during initiating (3-week-old), established (8-week-old) and residual stages of BCC (n = 4 per stage). b, Relative expression levels (mean ± s.e.m.) of Axin2, Wif1, Wnt4, and Wnt10a in untreated BCCs and tumours treated for 2 or 22 days (n = 3 per group). c, Representative images of H&E-stained skin sections showing the amount of residual BCC after indicated treatments (n = 7 per group). d, Average residual tumour counts per length of skin in BCC mice from c. A two-tailed unpaired t-test was used to determine significance between groups (*P = 0.0328). All experiments were performed at least twice. Scale bars, 100 μm; n represents the number of mice.

Source data

Full size image

We next tested whether treatment of BCC with a combination of vismodegib and a Wnt pathway inhibitor would have an effect on residual disease. We used a function-blocking anti-LRP6 antibody23 (Extended Data Fig. 9a–c), as our RNA-seq analysis revealed that LRP6 was the predominant Wnt co-receptor in BCC. Treatment with vismodegib and anti-LRP6 led to a 33% decrease in the number of residual tumour nests compared to vismodegib monotherapy (Fig. 4c, d). Notably, the residual tumour burden correlated with the magnitude of Wnt pathway inhibition, as mice with fewer lesions experienced a greater decrease in Axin2 expression (Extended Data Fig. 9d). In addition, Oil Red O staining was enhanced within residual tumours after combination treatment (Extended Data Fig. 9e–h), consistent with attenuation of Wnt signalling being required for sebocyte differentiation24.

Finally, we investigated whether combined inhibition of Hh and Wnt signalling altered the regrowth of residual BCCs after treatment. Mice were dosed with BrdU to label tumour nests and were then treated for 28 days with either vismodegib alone or vismodegib and anti-LRP6. After discontinuation of treatment, tumour regrowth was delayed (but not abolished) in mice that were treated with combination therapy relative to those treated with vismodegib alone (Extended Data Fig. 9i–q), probably owing to incomplete Wnt inhibition (Extended Data Fig. 9d).

Together, our results show that despite the strong anti-tumour activity of vismodegib in BCC, residual tumour cells persist. Tumours display robust Hh pathway inhibition during treatment, indicating that the drug continues to block signalling. As such, residual tumour cells have not acquired drug resistance through de novo mutations, but have adopted an identity that no longer relies on Hh signalling. When treatment is discontinued, tumour cells reactivate the Hh pathway and resume growth. Although BCC patients experience clinical benefit from Hedgehog pathway inhibitors after a drug holiday2, this approach may lead to the development of de novo resistance if additional mutations in the Hh pathway are acquired25. It is therefore important to achieve complete elimination of all residual tumour cells. The Wnt pathway seems to be critical for cell identity switching in BCC through reprograming of super-enhancers that drive the expression of key transcription factors. However, tolerability of complete Wnt pathway inhibition remains a challenge26. The ability of residual tumour cells to adopt a state that allows them to survive while remaining fully quiescent may represent a more widespread mechanism of resistance to targeted therapies27. Strategies that block this process may provide an opportunity to increase the rate of complete responses.

Methods

BCC mice and dosing

All mouse experiments were performed according to the animal use guidelines of Genentech, a Member of the Roche Group, conforming to California state legal and ethical practices. BCC mice were generated using the following alleles28,29,30: K14CreER, Ptch1fl/fl (a generous gift from R. Toftgard and S. Teglund), and p53fl/fl. BCC mice were singly housed and monitored for signs of advanced tumour burden, such as a scruffy coat and ear thickening. Tumour formation occurred in the absence of tamoxifen, probably owing to inherent leakiness of the CreER recombinase9. Vismodegib was formulated as a suspension in 0.5% methyl-cellulose, 0.2% tween-80 (MCT). Minimal efficacious dosing of vismodegib by oral gavage was determined to be 75 mg/kg bodyweight twice a day, and this dosing schedule was used for all experiments. Anti-LRP6 bi-specific antibody23 was given once daily at 30 mg/kg bodyweight by intraperitoneal (IP) injection. For label-retaining experiments, mice were given BrdU injections (100 μl of 10mg/ml) for three consecutive days before the start of treatment to label all dividing cells in the skin. Subsequent treatment with vismodegib stopped proliferation and prevented the dilution of incorporated BrdU in tumour nests. Lineage tracing with the lox-stop-lox TdTomato reporter allele (Jackson strain code: 007914) was initiated in mice at weanling age with three consecutive daily doses of tamoxifen (100 μl of 20 mg/ml in sunflower oil).

BCC histology score

A pathologist determined the extent of BCC tumour burden using the following scoring system: 0, minimal basaloid nests involve <10% of linear extent of skin; 1, isolated basaloid nests are generally confined to superficial half of dermis (do not extend to base of hair bulb); 2, basaloid nests crowd superficial dermis and/or extend into deeper half of dermis, but do not expand to fill dermal space completely in areas of deeper penetration; 3, basal cells pack dermis in confluent masses in >50% of skin.

RNA sequencing

Hundreds of lesions were laser capture micro-dissected (LCM) from multiple skin dissections per animal. Adjacent skin sections were stained for Ki67 and counterstained with haematoxylin to determine the proliferation status of tumour nests from untreated BCC mice. Three groups, each consisting of five mice, were included: 1) Ki67+ tumours from untreated BCC mice; 2) Ki67– tumours from untreated BCC mice; and 3) residual tumours from BCC mice that were treated with vismodegib for 28 days. LCM samples were pooled on a per mouse basis and total RNA was extracted with the RNeasy kit (Qiagen). The concentration and integrity of the RNA were determined by NanoDrop 8000 (Thermo Scientific) and with a Bioanalyzer RNA 6000 Pico Kit (Agilent), respectively. cDNA was generated with the Ovation RNA-seq System V2 (Nugen) and sheared to 150–200 bp size using an LE220 focused ultrasonicator (Covaris). Fragment length was confirmed with the Bioanalyzer DNA 1000 Kit (Agilent) and samples were quantified with the Qubit dsDNA BR Assay (Life Technologies). A total of 1 μg of sheared cDNA and the TruSeq RNA Sample Preparation Kit v2 (Illumina) were used for the end repair step of each library. Library size was confirmed with High Sensitivity D1K screen tape on a 2200 TapeStation (Agilent Technologies), and concentration was determined by qPCR using the Library quantification kit (KAPA). Libraries were multiplexed and sequenced on a HiSeq 2500 (Illumina) to generate 30 million single-end 50-bp reads per library. Reads were first aligned to ribosomal RNA sequences to remove ribosomal reads. The remaining reads were aligned to the mouse reference genome (NCBI Build 38) using GSNAP31 version ‘2013-10-10’, allowing a maximum of two mismatches per 50 base pair sequence (parameters: ‘-M 2 -n 10 -B 2 -i 1 -N 1 -w 200000 -E 1–pairmax-rna = 200000–clip-overlap’). Transcript annotation was based on the RefSeq database (NCBI Annotation Release 104). To quantify gene expression, the number of reads mapped to the exons of each RefSeq gene was calculated. Read counts were scaled by library size, quantile normalized and precision weights calculated using the ‘voom’ R package32. Subsequently, differential expression analysis on the normalized count data was performed using the ‘limma’ R package33 by contrasting vismodegib-treated samples with untreated samples. Gene expression levels were considered significantly different across groups if we observed |log2[fold change]| ≥ 1 (estimated from the model coefficients) associated with a false discovery rate (FDR)-adjusted P ≤ 0.05. Gene expression was obtained in form of nRPKM as described previously34.

ATAC library preparation

ATAC–seq was performed as follows: 100,000 sorted cells were collected in 1 ml PBS + 3% FBS at 4 °C. Cells were centrifuged, then cell pellets were resuspended in 100 μl lysis buffer (Tris HCl 10 mM, NaCl 10 mM, MgCl2 3 mM, Igepal 0.1%) and centrifuged (500g) for 25 min at 4 °C. Supernatant was discarded and nuclei were resuspended in 50 μl reaction buffer (Tn5 transposase 2.5 μl, TD buffer 22.5 μl and 25 μl H2O – Nextera DNA sample preparation kit, Illumina). The transposase reaction was performed for 30 min at 37 °C and then blocked by addition of 5 μl clean up buffer (NaCl 900 mM, EDTA 300 mM). DNA was purified using the MinElute purification kit (QIAGEN). DNA libraries were PCR amplified (Nextera DNA Sample Preparation Kit, Illumina), and size selected for 200 to 800 bp (BluePippin, Sage Sciences) following the manufacturers’ protocols.

ATAC–seq

The libraries were sequenced on Illumina HiSeq 2500 sequencers. We obtained an average of 100 million paired-end reads (50 bp) per sample. Reads were aligned to the mouse reference genome (NCBI Build 38) using GSNAP31 version ‘2013-10-10’, allowing a maximum of two mismatches per read sequence (parameters: ‘-M 2 -n 10 -B 2 -i 1–pairmax-dna = 1000–terminal-threshold = 1000–gmap-mode = none–clip-overlap’). Reads aligning to locations in the mouse genome that contain substantial sequence homology to the MT chromosome or to blacklisted regions identified by the ENCODE consortium were omitted from downstream analyses. Properly paired reads derived from non-duplicate sequencing fragments were used to quantify chromatin accessibility according to the ENCODE pipeline standards with minor modifications as follows. Accessible genomic locations were identified by calling peaks with Macs235 using insertion-centred pseudo-fragments (73 bp; community standard) generated on the basis of the start positions of the mapped reads. Accessible peak locations were identified as described: in brief, we called peaks on a group-level pooled sample containing all pseudo-fragments observed in all samples within each group. Peaks in the pooled sample that were independently identified in two or more of the constituent biological replicates were retained for downstream analysis, using the union of all group-level reproducible peaks (https://www.encodeproject.org/atac-seq/#standards). We quantified the level of chromatin accessibility within each peak for each replicate as the number of pseudo-fragments that overlapped the peak in question and normalized these estimates using the TMM method36. To better understand the within- and between-group similarities in chromatin accessibility, we calculated the Pearson correlation coefficient for the top 5,000 most variable peaks and performed hierarchical clustering using the correlation measures.

We identified differentially accessible peaks between groups in the framework of a linear model implemented with the limma R package33 and incorporating precision weights calculated with the voom function in the limma R package32. Peaks that showed an increase in accessibility at either day 6 or day 14 of vismodegib treatment were called vismodegib peaks and control peaks were called untreated peaks. For subsequent analysis, peaks were divided into promoter regions (1 kb up- and 2 kb down-stream of transcription start sites) and enhancers (peaks outside of promoter regions).

We identified enriched transcription factor (TF) motifs using HOMER v4.737. To evaluate the significance of the TF enrichment we defined peaks as significantly differentially accessible based on a range of FDR adjusted P value thresholds between 1 and 0.01 and an |log2[fold change]| in accessibility ≥ 1 (estimated from the model coefficients). Given the strong enrichment of the top motifs across a wide range of P value cutoffs, we decided to consider peaks as different across groups for a |log2[fold change]| ≥ 1 and FDR adjusted P value ≤ 0.05 in subsequent analyses.

Chromatin accessibility was visualized as coverage tracks across genomic regions using the Integrated Genomics Viewer38 and as heatmaps using DeepTools v3.0.139.

Regulatory potential (BETA analysis)

We associated accessible chromatin regions with nearby genes using BETA15. The BETA minus mode was used to calculate the regulatory potential (determined through a distance-weighted measure) of specific sets of peaks within a certain distance to a target gene. The BETA basic mode allowed us to integrate differential expression with chromatin openness to evaluate whether the direct effect of changes in the chromatin landscape is promoting or repressing gene expression. In this mode all genes within 100 kb of a peak set are ranked (and listed along the x-axis) based on the regulatory potential using the ATAC–seq data. Subsequently, expression information is used to divide genes into downregulated by vismodegib (purple line), upregulated by vismodegib (red line) and transcriptionally unchanged (dashed line) genes. A one-tailed Kolmogorov–Smirnov test40 was used to determine whether the upregulated and downregulated groups differed significantly from the group of transcriptionally unchanged genes.

FACS sorting of tumour cells

Back skin was taken from BCC mice harbouring the Lgr5GFPDTR allele41 and processed into a single-cell suspension42. Cells were stained for CD34 and the live/dead marker SYTOX blue. GFP-only tumour cells were sorted on a BD Biosciences Influx ‘jet-in-air’ cell sorter equipped with a combination of 355-, 405-, 488-, 561-, and 640-nm lasers. Fluorescence emission was collected through 460/50 (Sytox Blue), 530/40 (GFP), and 670/30 (CD34-Alexa647) bandpass filters. Sorting was performed with an 86-μm nozzle, a pressure of 30 psi and a frequency of 48,610 Hz. Setup and alignment of the instrument were performed using 3 μm ultra rainbow polystyrene beads (URFP-30-20, Spherotech Inc.). See Extended Data Fig. 6 for gating strategies.

Skin epithelial signatures

We downloaded microarray studies that profiled gene expression across various skin epithelial compartments, including GSE1518543, GSE2156844, GSE4061219, GSE4170445, GSE2026946 and E-MTAB-160647. Data normalization was performed using either the affy or lumi R packages, and differential expression analysis was performed using the limma R package33. Genes were called significantly differentially expressed when |log2[fold change]| > 1 and adjusted P < 0.1. Genes were considered bulge-specific if they were significantly upregulated in either the Lgr5+ vs. Lgr6+ comparison (GSE20269) or the CD34+ vs. GFP– comparison (E-MTAB-1606). Genes were considered ISTH- or IFE-specific if they were significantly downregulated in the GSE20269 or E-MTAB-1606 studies, respectively. The ORS signature was obtained by downloading the gene count data from RNA-seq GEO study GSE9084748 and identifying genes that were significantly upregulated (log2[fold change] ≥ 1 and adjusted P < 0.01) in ORS as compared to hair follicle stem cells using the same voom and limma procedure as described above.

Gene set analysis

We performed QuSAGE49 to identify relevant biological processes associated with vismodegib treatment. For that purpose we compared vismodegib-treated samples with untreated samples. For each comparison we then calculated the gene set activity (the mean difference in log2 expression of the individual genes that comprise the set) for the four sets of bulge, isthmus, IFE and ORS marker genes identified from the above public microarray experiments. In addition, we assessed the up- or downregulation of bulge, isthmus, IFE and ORS genes using gene set enrichment analysis (GSEA)40. The significance of the enrichment (shown as FDR) was determined through 1,000 permutations of random gene sets.

Immunohistochemistry and Immunofluorescence staining

Primary antibodies and their dilutions used in this study: rabbit Sox9 1:300 (Millipore, AB5535), rabbit Keratin1 1:1,000 (Covance, PRB-165 [AF109]), chicken Keratin5 1:2,000 (Biolegend, 90501), rabbit Keratin10 1:1,000 (Biolegend, 90541), rabbit Loricrin 1:1,000 (Covance, PRB-145P), goat IL-33 1:500 (R&D Systems, AF3626), rabbit Ki67 1:500 (GeneTex, GTX16667 [SP6]), rat BrdU 1:400 (BioRad, MCA2060), rabbit CC3 1:400 (Cell Signaling, 9661), and human Ep-CAM undiluted (Ber-EP4; Ventana 760-4383). Dorsal skin was shaved, dissected, and fixed in either 4% paraformaldehyde (PFA) or 10% neutral buffered formalin overnight. Fixed skin was washed in PBS and 70% ethanol, processed and embedded into paraffin. Fresh frozen skin was cut into strips and embedded in OCT (Sakura). Skin sections were cut at 6 μm on either a Leica RM2255 microtome or a Leica CM 3050 S cryostat. Fresh frozen skin sections were fixed in 4% PFA for 10 min before staining. For IHC, slides were de-paraffinized in xylenes, re-hydrated, and boiled in Dako target retrieval buffer for 10 min. Samples were then blocked with Dako protein-free blocking solution for 10 min and primary antibodies were diluted in Dako antibody diluent and exposed to samples overnight at 4 °C. Secondary antibodies (Invitrogen, Molecular Probes) for immunofluorescence staining were also diluted in Dako antibody diluent. IHC was performed using a Dako Envision+ system-HRP polymer detection kit.

In situ hybridization

Traditional RNA ISH was performed on paraffin-embedded tissue sections using digoxigenin-labelled probes according to standard protocols50. Alkaline phosphatase activity was detected on tissue sections using BM Purple staining solution (Roche) after overnight incubation in alkaline phosphatase buffer.

Hybridizations using the RNAscope method were performed according to the manufacturer’s protocol (Advanced Cell Diagnostics) using the RNAscope 2.5 HD Reagent Kit-RED (322350). Probes used were MmLhx2 (485791), MmAxin2 (400331), HuAXIN2 (400241), MmDefB6 (430141), MmWif1 (412361), and MmGli1 (311001).

Taqman quantitative real-time PCR

Tumours were collected from fresh frozen skin sections by LCM. Total RNA was extracted using the RNeasy micro kit (Qiagen) and cDNA was prepared using the TaqMan RNA to Ct 1 Step kit (Applied Biosystems). TaqMan analysis was performed on an ABI7900HT (Applied Biosystems), data were analysed with SDS 2.3 software (Applied Biosystems) and normalized to Hprt transcript levels. The following Applied Biosystems Taqman assays were used: Axin2 (Mm00443610_m1), Wif1 (Mm00442355_m1), Wnt4 (Mm01194003_m1), Wnt10a (Mm00839783_m1), Lhx2 (Mm00839783_m1) and Plet1 (MTS24 antigen; Mm01170995_m1).

Quantification of Oil Red O+ and BrdU+ tumours

The number of residual tumours in a linear unit of backskin equal to four lengths of a 5× magnification field of view was counted. Oil Red O+ and BrdU+ residual tumours were expressed as a fraction of the total number of tumours. BCCs were considered Oil Red O+ when they contained at least one Oil Red O+ cell and were not associated with a hair follicle (to avoid contamination from sebaceous glands). BCCs were considered BrdU+ when they contained at least 20 intense BrdU+ nuclei and did not contain any cells with a more weak, diluted nuclear signal (suggesting proliferation).

Statistics and reproducibility

See individual Methods sections for specific statistical methods. Experiments were independently repeated at least twice leading to similar results. No statistical method was used to predetermine sample size, and no mice were excluded from the analysis. BCC mice were age matched and randomly assigned to control and treatment groups. The investigators were blinded during outcome assessment. The data meet the assumptions of the statistical tests used and are presented as mean ± s.e.m.; P ≤ 0.05 was considered statistically significant. For RNA-seq, differential expression analysis was performed using the ‘limma’ R package, which uses the moderated t-statistic for significance analysis. The adjusted P value was calculated using the Benjamini and Hochberg method to account for the false discovery rate.

Human subject data

The study (ClinicalTrials.gov identifier: NCT01201915) was conducted per FDA regulations, International Conference on Harmonization E6 Guideline for Good Clinical Practice, and applicable local, state, and federal laws. The protocol was approved by institutional review boards where applicable. Patients gave written informed consent. The study evaluated the activity of vismodegib in patients with smaller operable BCC by measuring the rate and durability of complete histologic clearance (CHC) of lesions. Patients with new, operable, nodular basal cell carcinoma received vismodegib (150 mg/d) followed by excision and Mohs micrographic surgery to ensure clear margins. Samples from cohort 1 were analysed for this study: patients received vismodegib for 12 weeks and target sites were immediately excised by standard means, followed by Mohs micrographic surgery to obtain clear margins. Biospecimen type: skin; anatomical site: scalp/head/neck or upper aspect of trunk, greater trunk; disease status of patients: new, nodular BCC, operable; clinical characteristics of patients: alive; clinical diagnosis of patients: nodular BCC by biopsy; pathology diagnosis: nodular BCC by biopsy; collection mechanism: excision, Mohs surgery; type of stabilization: formalin fixed; type of long-term preservation: formalin fixation; constitution of preservative: 10% neutral-buffered formalin.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.