Mutations in ACTRT1 and its enhancer RNA elements lead to aberrant activation of Hedgehog signaling in inherited and sporadic basal cell carcinomas

Basal cell carcinoma (BCC), the most common human cancer, results from aberrant activation of the Hedgehog signaling pathway. Although most cases of BCC are sporadic, some forms are inherited, such as Bazex–Dupré–Christol syndrome (BDCS)—a cancer-prone genodermatosis with an X-linked, dominant inheritance pattern. We have identified mutations in the ACTRT1 gene, which encodes actin-related protein T1 (ARP-T1), in two of the six families with BDCS that were examined in this study. High-throughput sequencing in the four remaining families identified germline mutations in noncoding sequences surrounding ACTRT1. These mutations were located in transcribed sequences encoding enhancer RNAs (eRNAs) and were shown to impair enhancer activity and ACTRT1 expression. ARP-T1 was found to directly bind to the GLI1 promoter, thus inhibiting GLI1 expression, and loss of ARP-T1 led to activation of the Hedgehog pathway in individuals with BDCS. Moreover, exogenous expression of ACTRT1 reduced the in vitro and in vivo proliferation rates of cell lines with aberrant activation of the Hedgehog signaling pathway. In summary, our study identifies a disease mechanism in BCC involving mutations in regulatory noncoding elements and uncovers the tumor-suppressor properties of ACTRT1.


2 2 6
VOLUME 23 | NUMBER 10 | OCTOBER 2017 nature medicine Basal cell carcinoma (BCC), the most common human cancer, results from aberrant activation of the Hedgehog signaling pathway 1 . Although most cases of BCC are sporadic, some forms are inherited, such as Bazex-Dupré-Christol syndrome (BDCS)-a cancer-prone genodermatosis with an X-linked, dominant inheritance pattern 2 . We have identified mutations in the ACTRT1 gene, which encodes actin-related protein T1 (ARP-T1), in two of the six families with BDCS that were examined in this study. High-throughput sequencing in the four remaining families identified germline mutations in noncoding sequences surrounding ACTRT1. These mutations were located in transcribed sequences encoding enhancer RNAs (eRNAs) [3][4][5] and were shown to impair enhancer activity and ACTRT1 expression. ARP-T1 was found to directly bind to the GLI1 promoter, thus inhibiting GLI1 expression, and loss of ARP-T1 led to activation of the Hedgehog pathway in individuals with BDCS. Moreover, exogenous expression of ACTRT1 reduced the in vitro and in vivo proliferation rates of cell lines with aberrant activation of the Hedgehog signaling pathway. In summary, our study identifies a disease mechanism in BCC involving mutations in regulatory noncoding elements and uncovers the tumor-suppressor properties of ACTRT1.
BDCS (MIM 301845) is an X-linked, dominantly inherited condition predisposing to BCC 2,6,7 (Supplementary Fig. 1a). By studying six families affected by BDCS (Fig. 1a), we mapped the gene associated with BDCS to a 7.5-Mb region at Xq25-q26.2 (Supplementary Fig. 1b,c) and identified an insertion in the ACTRT1 gene, encoding ARP-T1, that segregated with the disease in two of the six families (c.547_ 548insA, p.Met183Asnfs*17 in families C and D; Supplementary  Fig. 1d,e). The mutant cDNA encodes a 25-kDa truncated protein (Supplementary Fig. 1f). Yet, no mutations in the coding region of ACTRT1 were found in the other four families in which disease was linked to the same interval. No rearrangements in the candidate region were identified by high-density tiling-path array-based comparative genomic hybridization. Immunohistochemical, RT-PCR and western blot analyses of control skin samples taken from healthy individuals detected expression of ARP-T1 in epidermal layers and skin appendages involved in BDCS but not in dermal connective tissues ( Supplementary  Fig. 2a-d). Interestingly, immunohistochemical analyses failed to detect any specific ARP-T1 staining in BCC tumors from individuals with BDCS and detected only a weak signal in unaffected epidermis from these individuals (Supplementary Fig. 2e). Immunofluorescence (Fig. 1b,c) and qRT-PCR (Fig. 1d) analyses of the epidermis showed low ARP-T1 expression in all individuals with BDCS, regardless of the presence or absence of mutations in ACTRT1, suggesting that the remaining unexplained cases involved hitherto unknown ACTRT1 regulatory elements.
Conserved noncoding elements (CNEs), which are known to control expression of neighboring genes 8,9 , are concentrated in gene deserts 10 . Taking into consideration the fact that ACTRT1 is located in a 2.6-Mb gene desert, we examined 17 CNEs located either upstream or downstream of the ACTRT1 coding sequence (Supplementary Fig. 3a). Sanger sequencing detected a g.127372937A>T variation of CNE12 only in families E and F (CNE12, chr. X: 127,371,674-127,374,249; Supplementary Fig. 3b,c). Comparative genomic approaches aimed at predicting regulatory DNA sequences are known to have limitations, as regulatory elements are not necessarily conserved across species 11,12 . We therefore performed systematic array-based capture and high-throughput sequencing of the 7.5-Mb candidate region and used a specific genome browser to select candidate BDCSassociated variants (Supplementary Fig. 4a). We selected three additional candidate variants in family A (A1, g.125960325A>T; Mutations in ACTRT1 and its enhancer RNA elements lead to aberrant activation of Hedgehog signaling in inherited and sporadic basal cell carcinomas 1   The experiments were performed in triplicate. Results obtained from one control skin biopsy are shown but are representative of two independent experiments on two different control samples. Input 1%, sonicated but nonimmunoprecipitated DNA. The chromatin signature of the GAPDH promoter was assessed as a control. Note that the sequence of the GAPDH promoter was enriched in H3K4me3, a promoter-specific histone mark. The CNE12, A2 and B2 sequences were not tagged with H3K4me3, but they were enriched with the enhancer-specific markers H3K27ac and H3K4me1. Wild-type A1, A3 and B1 were not enriched in any specific histone marks. (f) An enhancer luciferase reporter assay based on the wild-type sequences surrounding candidate variants in HaCaT keratinocytes. Enhancer activity was observed with the A2, B2 and CNE12 sequences. No significant luciferase activity was observed with wild-type A1, A3 and B1 sequences in comparison to an empty-vector (EV) control. (g) Enhancer luciferase reporter assays demonstrating the impact of the variants in the A2 (g.125959394C>G), B2 (g.127968123T>C) and CNE12 (g.127372937A>T) sequences in HaCaT keratinocytes. In f and g, luciferase activity was normalized to that of Renilla luciferase; the experiments were performed in triplicate, and results in each panel are representative of three independent experiments. (h) Schematic representation of the target sites of sgRNAs around the A2, B2 and CNE12 sequences. sgRNAs that created indels in enhancer regions are shown in red, and sgRNAs that failed to delete enhancer regions are shown in black. The editing efficiency in genomic DNA (gDNA cut) was measured using tracking of indels by decomposition (TIDE) software (Supplementary Figs. [6][7][8]. The expression of ACTRT1 mRNA was measured in RT-PCR assays, and results were normalized to PGK1 mRNA levels. The experiment was performed in triplicate, and results are representative of two independent experiments. Data in c-h are expressed as the mean ± s.d. For box-and-whisker plots, midlines represent the median; upper and lower perimeters extend from the 25th to the 75th percentile; and whiskers extend from minimum to maximum. P values were calculated by one-way ANOVA with Bonferroni multiple comparisons for c and d (relative to controls), e (relative to IgG), f (relative to EV) and h (relative to control blank) and by t-test for g (wild-type sequence versus mutant sequence). *P < 0.05, **P < 0.01, ***P < 0.001.

2 2 8
VOLUME 23 | NUMBER 10 | OCTOBER 2017 nature medicine A2, g.125959394C>G; A3, g.126494053_126494054insT) and two variants in family B (B1, g.127061005G>C; B2, g.127968123T>C; Supplementary Fig. 4b,c). To identify putative disease-causing variants from among these candidates, we performed (i) a chromatin signature analysis to identify variants that mapped to active regulatory regions 13,14 , (ii) enhancer luciferase reporter assays to identify sequences capable of activating transcription 15 and (iii) in situ hybridization to identify regions that are transcribed in skin 16 .
Chromatin immunoprecipitation (ChIP) and targeted qPCR assays were performed on protein-DNA complexes extracted from normal human epidermis. Interestingly, an enhancer signature, as determined by enrichment of H3K27ac and H3K4me1 marks and absence of the H3K4me3 mark, was only found for the A2, B2 and CNE12 sequences (Fig. 1e). In these assays, a difference in amplification efficiency between primer sets constitutes a limitation for comparison of the data obtained using different primers. Luciferase reporter assays demonstrated that only the wild-type sequences encompassing the A2, B2 and CNE12 variant positions had enhancer activity in HaCaT keratinocytes (Fig. 1f). This activity was reduced when mutated constructs were used (Fig. 1g). Given that a subset of eRNAs are broadly transcribed 3,4 and that the A2, B2 and CNE12 sequences bound RNA polymerase II (RNA Pol II) 5 (Fig. 1e), we used RT-PCR and in situ hybridization to study transcription at the candidate loci. In agreement with the ACTRT1 expression pattern, the A2, B2 and CNE12 sequences were specifically expressed in the epidermis and its appendages ( Supplementary Fig. 5a-d). Interestingly, no staining for these sequences was detected when skin biopsies from the corresponding individuals were tested, demonstrating the dramatic impact of the A2, B2 and CNE12 variations on eRNA expression and/or stability ( Supplementary Fig. 5b).
Lastly, in order to provide conclusive in vitro evidence of a link between these enhancer sequences and ACTRT1 expression, we used CRISPR-Cas9 technology to disrupt the three enhancer regions 17 . We generated keratinocytes with insertions and/or deletions (indels) in the A2, B2 or CNE12 enhancer regions by independently introducing six single-guide RNAs (sgRNAs) specific for these enhancer regions (Supplementary Figs. 6-8). Disruption of these enhancers resulted in a decrease in ACTRT1 expression relative to cells without enhancer region editing, and decreased expression was correlated with editing efficiency (Fig. 1h and Supplementary Fig. 9). We also observed that enhancer mutagenesis increased the rate of keratinocyte proliferation (Supplementary Fig. 10). Taken together, our results suggest that BDCS is directly associated with loss-of-function mutations either altering the coding region of ACTRT1 or affecting specific enhancers located in transcribed noncoding surrounding regions.
In further immunohistological studies of sporadic BCC cases, no ARP-T1 signal was detected in BCC tissues from 51 of 60 unrelated sporadic cases (representative of one tumor from a single individual), whereas normal ARP-T1 staining was detected in BCC tissues from 5 individuals with xeroderma pigmentosum ( Supplementary  Fig. 11a). Further sequencing of a series of 20 BCC tissue samples identified three mutations in the coding sequence of ACTRT1 (Supplementary Fig. 11b) and one mutation in the B2 enhancer region (g.127968192G>A) located 69 bp downstream of the mutation found in BDCS family B (g.127968123T>C). These results emphasize the genetic heterogeneity of these skin tumors and suggest that both inherited and sporadic BCCs are associated with loss-of-function mutations in the ACTRT1 locus.
In order to elucidate the consequences of loss-of-function mutations in ACTRT1, we performed comparative transcriptomic analyses of skin samples from individuals with BDCS and identified 1,771 genes with deregulated expression relative to that of three skin samples from healthy individuals (for a description of the statistical analysis, see the Online Methods). Using Ingenuity Pathway Analysis (see URLs), we showed that these genes are mostly involved in regulation of cell-cycle progression, cell death and survival, and cell migration (Fig. 2a). As constitutive activation of the Hedgehog pathway is found in more than 70% of BCCs 1,18,19 , particular attention was given to transcription factors activated by this pathway. Indeed, among the genes with deregulated expression, 56 are directly controlled by the Hedgehog transcription factors GLI1 and GLI2 (Fig. 2b).    Flag-ACTRT1 547_548insA Flag-ACTRT1 547_548insA Flag-ACTRT1 547_548insA   Figure 3 Functional impact of ARP-T1 on the Sonic Hedgehog signaling pathway. (a) Luciferase activity of a GLI luciferase reporter (8×3′Gli-BS-delta51-LucII) in HaCaT keratinocytes transfected with wild-type (WT) or mutant (c.547_548insA) ACTRT1 construct after 24 h of stimulation with purmorphamine (3 µM). A luciferase reporter construct containing mutations in GLI-binding sites (8×3′Gli-BS-mutS4-delta51-LucII) was used as a negative control for activation of the Hedgehog pathway. Luciferase activity was normalized to that of Renilla luciferase. The experiment was performed in triplicate, and results are representative of three independent experiments. (b) qRT-PCR analysis of GLI1 mRNA levels normalized to PGK1 mRNA levels in primary keratinocytes after transfection with empty vector or with vector encoding wild-type or mutant ACTRT1 following 5 h of stimulation with purmorphamine (3 µM) or DMSO. qRT-PCR was performed in triplicate, and results are representative of two independent experiments. (c) Western blot analyses of cytoplasmic, membrane, soluble nuclear and chromatin-bound proteins from HEK293T cells transfected to express Flag-tagged wild-type or mutant ACTRT1 (following 24 of h stimulation of cells with DMSO or purmorphamine (3 µM)). The nuclear marker histone deacetylase 2 (HDAC2), the cytoplasmic marker heat-shock protein 90 (HSP90), the membrane marker calreticulin and the chromatin marker H3K27ac were used to assess protein loading and fraction purity. The plot to the right quantifies the fraction of Flag-tagged protein in each cellular compartment under the different stimulation conditions as determined by band intensity in the western blots (see Supplementary Data for full blots). (d) Schematic representation of the human GLI1 promoter region, showing the locations of the primers used in ChIP experiments (in f) relative to the transcriptional start site (TSS; indicated by the black arrow). The GLI1 promoter region includes a 5′ flanking sequence (5′ UTR), an untranslated exon (1) and part of the first intron (see Supplementary Fig. 12 for details on chromatin marks in the GLI1 promoter region in NHEK keratinocytes). Untranslated exon sequences are represented by a white square; translated exon sequences are depicted in gray. Numbering above the line indicates nucleotide locations relative to the position of the TSS. (e) Levels of GLI1 mRNA expression in primary keratinocytes over the course of stimulation with purmorphamine (3 µM) at the indicated times. Results were normalized to PGK1 mRNA levels. qPCR was performed in triplicate, and results are representative of two independent experiments. (f) ChIP-qPCR analysis was performed with primers binding to the locations depicted in d, showing the fold enrichment of RNA Pol II and ARP-T1 at the GLI1 promoter sites in primary keratinocytes at the times indicated on the x axis over the course of stimulation with purmorphamine (3 µM). Primers for the RPL10A promoter were used as a negative control for ARP-T1 binding. qPCR was performed in triplicate, and results are representative of two independent experiments. Data in a, b, e and f are expressed as the mean ± s.d. P values were calculated by one-way ANOVA with Bonferroni multiple comparisons for a (relative to EV), e (relative to stimulated cells at 0 h) and f (relative to IgG) and by t-test for b (unstimulated cells versus cells stimulated with purmorphamine). *P < 0.05, **P < 0.01, ***P < 0.001. NS, not significant. Consistent with these findings, qPCR showed that Hedgehog target genes were overexpressed in skin samples from individuals with BDCS carrying either a mutation in ACTRT1 (c.547_548insA) or the CNE12 variant (Fig. 2c). Transactivation assays in cells from the HaCaT keratinocyte cell line employing a reporter construct with GLI1-binding sites upstream of the luciferase gene showed    that expression of wild-type but not truncated ARP-T1 inhibited the Hedgehog pathway (Fig. 3a). Moreover, expression of wild-type but not truncated ARP-T1 in primary keratinocytes inhibited GLI1 expression after stimulation with the Smoothened (Smo) agonist purmorphamine (Fig. 3b). Taken together, these results strongly suggest that ARP-T1 has a role in regulating the activity of the Hedgehog signaling pathway. ARP-T1 belongs to the actin-related protein (ARP) family 20 . In the nucleus, ARP proteins are essential elements of the macromolecular machinery that controls nucleosome remodeling, histone acetylation, histone variant exchange, transcription and DNA repair 21,22 . Using ultrathin sections of normal epidermis processed for transmission electron microscopy, we detected ARP-T1 in both the nucleus and cytoplasm (Supplementary Fig. 12a). Western blot analyses of proteins from subcellular fractions of transfected HEK293T cells or control primary keratinocytes showed that ARP-T1 was mostly located in the nucleus and was bound to chromatin following stimulation of the Hedgehog pathway using purmorphamine ( Fig. 3c and Supplementary Fig. 12b). Conversely, the truncated ARP-T1 protein was absent from the chromatin-bound fraction of transfected HEK293T cells (Fig. 3c). It has been reported that GLI1 transcription and Hedgehog pathway activity are controlled by chromatin regulators, including the tumor-suppressors BRG1 (refs. 23,24) and SNF5 (ref. 25), two components of the mammalian SWI-SNF chromatin-remodeling complex that directly bind to GLI1 regulatory domains. To determine whether ARP-T1 binds to the GLI1 promoter ( Fig. 3d and Supplementary Fig. 13), we performed ChIP analysis with an anti-ARP-T1 antibody in control primary keratinocytes. An anti-RNA Pol II antibody was used as a positive control for transcriptional activation. Interestingly, stimulation of keratinocytes for 5 h with purmorphamine triggered an increase in GLI1 expression (Fig. 3e) correlating with enrichment of RNA Pol II (but not ARP-T1) binding in GLI1 promoter regions (Fig. 3f). Longer stimulation of keratinocytes with purmorphamine (25 h) resulted in enrichment of ARP-T1 binding in GLI1 promoter regions (Fig. 3f), which was concomitant with a decrease in GLI1 expression (Fig. 3e). Taken together, our results support the notion that ARP-T1 exerts negative control over GLI1 expression in a manner similar to that observed for SNF5 and BRG1.
In order to investigate both the tumor-suppressor activity of ACTRT1 in vivo and the impact of ACTRT1 mutations, we generated populations of cells that stably expressed wild-type or mutant ARP-T1 from the long-term human BCC cell line UW-BCC1-T2 (ref. 26 ; Fig. 4a), which is characterized by enhanced activation of the Hedgehog signaling pathway. Interestingly, expression of wild-type ARP-T1 significantly decreased the cell proliferation rate as compared to that of UW-BCC-T2 cells expressing the empty vector (blank) (Fig. 4b) concomitantly with a reduction in GLI1 and GLI2 expression (Fig. 4c). Conversely, mutant ARP-T1 (c.547_548insA) only partially inhibited cell proliferation and had a less pronounced effect in reducing GLI1 and GLI2 expression. Similarly, in tumor xenografts of these cells in AGR129 mice, wild-type ARP-T1, but not the truncated variant, attenuated tumor development (Fig. 4d-f) and Ki-67 (proliferation marker) expression (Supplementary Fig. 14). These results further confirm the tumor-suppressor role of ARP-T1 in limiting BCC development in vivo.
Interestingly, a rare deletion in the ACTRT1 locus has been reported in 17 of 63 familial cases of early-onset hereditary breast cancer that lack BRCA1 and BRCA2 mutations 27 . We hypothesized that ARP-T1 might have more general tumor-suppressor activity in addition to limiting BCC growth, so we selected two human cancer cell lines in which the Hedgehog signaling pathway is aberrantly active (the U2OS osteosarcoma cell line 28,29 and the MDA-MB-231 breast cancer cell line 30,31 ) for further study; notably, ARP-T1 is highly expressed in glandular breast lobules and weakly expressed in osteocytes (Supplementary Fig. 15), but its expression is absent in these two cancer cell lines (Fig. 4g,j). In addition, using the specific GLI inhibitor GANT61, we confirmed the GLI-dependent growth of U2OS and MDA-MB-231 cells 31 (Supplementary Fig. 16). Stable expression of ARP-T1 in these cell lines inhibited in vitro proliferation (Fig. 4g,j) and reduced GLI1 and GLI2 expression (Fig. 4h,k). Consistent with these findings, in tumor xenograft experiments in NMRI nude mice, ARP-T1 overexpression prevented in vivo growth of injected U2OS cells (Fig. 4i) and reduced the development of xenograft tumors from MDA-MB-231 cells (Fig. 4l). Previously, GLI1 inhibition has been shown to attenuate growth and migration of MDA-MB-231 cells by increasing apoptosis and decreasing cell proliferation 30 . In accordance with these observations, expression of exogenous wildtype ARP-T1 increased the number of apoptotic MDA-MB-231 cells, increased the expression of pro-caspase-3 and cleaved, active caspase-3 (Supplementary Fig. 17a-c), and decreased Ki-67 expression in MDA-MB-231 cells (Supplementary Fig. 18) in comparison to expression of the empty vector (blank). Expression of the wild-type protein also reduced the metastatic potential of these MDA-MB-231 cells in vitro and in vivo (Supplementary Fig. 19a-e).
Germline mutations in the ACTRT1 coding sequence and its surrounding noncoding elements constitute a hitherto unreported causative mechanism of inherited predisposition to BCC. Here, to our knowledge, we report for the first time the involvement of ACTRT1 in human disease. We suggest potential functions of ARP-T1 and mechanisms through which it could regulate GLI1 expression on the basis of our results and some models in the literature describing the ARP family. First of all, we demonstrated that wild-type ARP-T1, but not the truncated protein identified in BDCS families (C and D, carrying the c.547_548insA ACTRT1 mutation), binds to chromatin, suggesting that ARP-T1 has nuclear functions. Nuclear ARP proteins are components of the four main chromatin-remodeling complexes (CRCs): INO80, SRCAP, BAF (the human analog of the SWI-SNF complex) and TIP60/TRRAP 21,22 . ARP5 and ARP8 can bind to core histones to facilitate interaction of the CRCs with nucleosomes 21,22 . Mutations in the genes encoding the major mammalian SWI-SNF (BAF) CRC subunits are present in over 20% of human cancers 31 . SNF5 (the core component of the SWI-SNF CRC) and BRG1 (the ATPase subunit) are encoded by bona fide tumor-suppressors genes, in which mutations are responsible for various types of cancer [23][24][25] . Interestingly, both SNF5 and BRG1 act through direct inhibition of GLI genes 24,25 , which is also the mechanism of ARP-T1 action. Further studies are needed to determine whether ARP-T1 acts specifically at the level of regulatory elements through recognition of specific histone marks, repositioning of nucleosomes, histone exchange or binding to transcription factors. Also, we cannot exclude the possibility that ARP-T1 could interact with key proteins involved in various Hedgehog-interconnected pathways related to tumor proliferation and progression, such as the p16-RB, Wnt, and Polycomb pathways, as has been demonstrated for SNF5 (refs. 32,33).
The discovery of distal regulatory noncoding elements, known as enhancers, with critical functions in gene expression has added a new dimension to transcriptional regulation. eRNAs are potent transcription units, and their alteration can impact biological processes involved in human diseases, including cancer 34 . They cover a broad 1 2 3 2 VOLUME 23 | NUMBER 10 | OCTOBER 2017 nature medicine spectrum of molecular and cellular functions by implementing different modes of action. Studies have confirmed that long noncoding RNAs (lncRNAs) contribute to cancer initiation and progression by regulating gene transcription 35 . For instance, a link between lncRNAs and the SWI-SNF complexes has been reported in various tumoral conditions 36 . Another mechanism by which eRNAs act in gene regulation involves their interaction with cohesin and Mediator complexes to stabilize enhancer-promoter looping, causing chromatin stabilization and gene expression 37 . eRNAs are also involved in the recruitment of RNA Pol II to gene promoters and facilitate the access of specific transcription factors to enhancer sequences 38 . In addition to DNA methylation and histone modification, a role for noncoding RNA has recently emerged in epigenetic control. Alterations of eRNAs have been found to result in epigenomic reprogramming during tumor initiation and progression 39 . Further investigation is needed to place ARP-T1 and its noncoding regulatory elements in such complex mechanisms of gene regulation and cancer development. More broadly, our findings shed light on the functional relevance of genomic alterations in noncoding regions and their contribution to tumor development. Indeed, the clinical integration of noncoding RNAs as functionally relevant elements in conjunction with additional predictive biomarkers could improve the management of individuals with cancer.

MeTHoDS
Methods, including statements of data availability and any associated accession codes and references, are available in the online version of the paper.
Note: Any Supplementary Information and Source Data files are available in the online version of the paper.