1 MicroRNA-203 represses selection and expansion of oncogenic

2 Hras transformed tumor initiating cells

3

4 Kent Riemondy1, Xiao-jing Wang2, Enrique C. Torchia3, Dennis R. Roop3 and Rui Yi1,4

5

6 1Department of Molecular, Cellular, and Developmental Biology, University of Colorado,

7 Boulder, Boulder, CO 80309

8 2Department of Pathology, University of Colorado Denver Anschutz Medical Campus, Aurora,

9 CO 80045

10 3Department of Dermatology and Charles C. Gates Center for Regenerative Medicine and Stem

11 Cell Biology, University of Colorado Denver Anschutz Medical Campus, Aurora, CO 80045

12

13 4Correspondence:

14 Rui Yi

15 Department of Molecular, Cellular, and Developmental Biology

16 University of Colorado, Boulder

17 Boulder, CO USA 80303

18 Tel: 303-735-4886

19 Email: [email protected]

20

21 Competing Interests Statement

22 The authors declare no competing interests.

1 23 Abstract

24 In many mouse models of skin cancer, only a few tumors typically form even though many cells

25 competent for tumorigenesis receive the same oncogenic stimuli. These observations suggest an

26 active selection process for tumor-initiating cells. Here we use quantitative mRNA- and miR-Seq

27 to determine the impact of HrasG12V on the transcriptome of keratinocytes. We discover that

28 microRNA-203 is downregulated by HrasG12V. Using a knockout mouse model, we demonstrate

29 that loss of microRNA-203 promotes selection and expansion of tumor-initiating cells.

30 Conversely, restoration of microRNA-203 using an inducible model potently inhibits

31 proliferation of these cells. We comprehensively identify microRNA-203 targets required for

32 Hras-initiated tumorigenesis. These targets include critical regulators of the Ras pathway and

33 essential genes required for cell division. This study establishes a role for the loss of microRNA-

34 203 in promoting selection and expansion of Hras mutated cells and identifies a mechanism

35 through which microRNA-203 antagonizes Hras-mediated tumorigenesis.

36

37

38

39

40

41

42

43

44

45

46

2 47 Introduction

48 Recent efforts in comprehensively sequencing human cancer genomes have confirmed

49 ~140 protein-coding genes that, when mutated, can drive tumorigenesis [1]. When genome

50 sequencing data were utilized to construct the history of cancer cells in breast cancer, it was

51 revealed that a considerable amount of “molecular time” exists between the common ancestors

52 that harbor the great majority of driver mutations and the phenotypically identified cancer cells

53 that compose the bulk of the tumor [2]. In support of these observations, lineage tracing

54 experiments conducted in genetically engineered mouse models revealed only a few clones give

55 rise to tumors whereas a vast majority of mutated cells are unable to sustain tumorigenesis [3,4].

56 These results suggest that even after the acquisition of key driver mutations in the nascent cancer

57 cells, these cells must still undergo continuous evolution and likely clonal selection before

58 developing into clinically apparent tumors. To begin to understand the molecular basis

59 underlying such selection, we examined papilloma formation driven by oncogenic Hras in the

60 skin, a well characterized model where Hras has been shown to initiate the formation of tumors

61 that clonally evolve [3,5,6]. Oncogenic Ras mutations are some of the most frequently detected

62 driver mutations in human cancer. Among the three Ras genes (H-, K- and N-ras), Hras is

63 commonly mutated in tumors originated from stratified epithelial tissues including squamous cell

64 carcinoma in the skin, head and neck cancer as well as bladder cancer [7–9]. Experimental and

65 genomic sequencing studies have revealed that the vast majority of Ras mutations are missense,

66 point mutations at amino acid residues glycine 12 (G12), glycine 13 (G13) or glutamine 61

67 (Q61) [8]. Structural and biochemical studies have further confirmed that all of these mutations

68 generally interfere with the GTP binding pocket and compromise the GTPase activity of Ras

69 proteins. In turn, these mutations lead to uncontrolled activation of downstream effectors

3 70 including Raf/MEK/ERK and PI(3)K pathways, resulting in sustained cell survival and

71 proliferation observed in human cancers. Because of the prominent role of Ras mutations in

72 human cancer, extensive efforts have been devoted to uncover and subsequently target

73 downstream pathways that are regulated by Ras mutations. However, the immediate impact of

74 Ras mutations on the transcriptome, in particular with regards to microRNAs (miRNAs) remains

75 unclear.

76 miRNAs are a class of small, noncoding RNA species that are involved in virtually all

77 biological processes examined in mammals including mouse and human. These regulatory RNA

78 molecules function by repressing the protein producing ability of mRNA targets through

79 destabilization of mRNAs and inhibition of translation [10]. miRNAs typically target a large

80 number of mRNAs in a dosage- and cell context-dependent manner [11]. As prominent proto-

81 oncogenes, Ras mutations have long been recognized to interact with the miRNA pathway.

82 Indeed, Hras, Kras and Nras all harbor multiple binding sites for the let-7 miRNA, a founding

83 member of miRNAs, in their 3’UTRs [12]. Additionally, impaired miRNA biogenesis in the

84 form of Dicer1 disruption has been shown to be a tumor suppressing mechanism for the

85 development of Kras induced lung cancer in a mouse model [13]. A number of individual

86 miRNAs were also found to function as modifiers for Ras induced tumorigenesis that include

87 miR-21, -29 and miR-17~92 as tumor promoting miRNAs and miR-34, -15/16, and miR-143/145

88 as tumor suppressing miRNAs [14–16]. Collectively, these seminal studies demonstrate

89 unequivocally that the miRNA pathway and individual miRNAs play important roles in Ras-

90 induced tumorigenesis. However, it is unclear how Ras mutations, usually the tumor-initiating

91 drivers, directly alter the landscape of miRNA expression during tumorigenesis. Importantly, it is

92 also unknown whether the changes in miRNA expression play a role in the selection of

4 93 oncogenic Ras transformed cells during tumor initiation. Finally, the lack of a comprehensive

94 survey of high-confidence miRNA targets that may play a role downstream of Ras mutations

95 hinders our mechanistic understanding and limits the potential to develop miRNA based

96 therapeutics.

97 In this study, we utilized our recently improved quantitative miR-Seq techniques to

98 examine the impact of an oncogenic Hras mutation (HrasG12V) on both mRNA and miRNA

99 expression. We discovered that miR-203, the most highly expressed miRNA in the skin [17,18],

100 is downregulated by HrasG12V. Using both knockout (KO) and inducible models, we provide

101 evidence for an important role of miR-203 in restricting expansion of oncogenic Hras

102 transformed cells in vitro and in vivo. We comprehensively surveyed skin-specific targets of

103 miR-203 and identified a number of novel targets that have important implications for Hras-

104 mediated tumorigenesis. Our results suggest that miR-203 plays a tumor-suppressing role in

105 inhibiting selection and expansion of tumor-initiating cells early in tumor development.

106

107 Results:

108 HrasG12V profoundly deregulates the mRNA and miRNA transcriptome in the skin

109 Oncogenic mutation of the Hras gene is one of the initiating drivers in the development

110 of benign papillomas and malignant squamous cell carcinomas in murine skin chemical

111 carcinogenesis. However, the molecular consequences defining the cellular changes that

112 accompany expansion of oncogenic Hras-transformed keratinocytes to initiate papillomas

113 remain elusive. We first investigated the consequences of HrasG12V activation on the mRNA

114 transcriptome using a modified form of PolyA+ RNA-Seq, known as 3P-Seq, or 3seq (Figure

115 1A-C). Compared to traditional RNA-Seq, 3Seq allows both quantification of mRNA transcripts

5 116 and detection of changes in alternative 3’ UTR formation [19]. To examine the immediate

117 impact of HrasG12V on primary skin cells, we used primary keratinocytes isolated from newborn

118 skin and performed 3Seq after HrasG12V transduction. We did not observe widespread shortening

119 or alternative formation of 3’UTRs, which are often ascribed to oncogenic transformation when

120 comparing tumor cell lines to normal cells (data not shown). This is similar to our previous

121 observation that alternative 3’UTR usage is infrequent within the skin lineages [19]. Over 1,100

122 transcripts were differently expressed (two-fold change and FDR <0.05) in keratinocytes

123 expressing HrasG12V, compared to the control (Figure 1C and Figure 1-source data). Gene

124 ontology functional analysis revealed profound deregulation in three core processes by HrasG12V:

125 activation of cellular migration, upregulation of pro-angiogenic pathways, and suppression of the

126 terminal differentiation program (Figure 1D). All of these three processes are identified as

127 hallmarks of human cancer [20]. The observed widespread changes in the transcriptome also

128 endorse the driver role of HrasG12V in skin tumorigenesis. Importantly, transcripts upregulated by

129 HrasG12V in our primary keratinocytes strongly and significantly overlapped with the putative

130 cancer stem cell signatures obtained from murine squamous cell carcinoma (SCC) models [21].

131 In addition, transcripts upregulated by HrasG12V significantly overlapped with transcripts know to

132 be targets of the c-Fos transcription factor in a genetic model of SCC [22]. Furthermore, known

133 core components of the Hras signaling pathway were also among the differentially detected

134 genes [23] (Figure 1E). These transcriptome data indicate that we have captured the initiating

135 changes induced by oncogenic Hras in the keratinocytes.

136 To define the impact of the oncogenic Hras on the landscape of miRNA, we applied our

137 recently developed, quantitative miRNA-Seq [24] to HrasG12V transformed keratinocytes. Overall,

138 we detected 15 differentially expressed miRNAs upon HrasG12V expression (FDR <0.05, two-

6 139 fold change) (Figure 1F-H). Two key patterns emerged in these profiles. First, the epithelial

140 tissue-specific miRNAs, miR-203 and miR-205, which represent the most abundant miRNAs

141 expressed in murine skin and primary keratinocytes were strongly suppressed by HrasG12V.

142 Additionally, members of the abundantly expressed, miR-23/24/27 miRNA cluster were also

143 downregulated by HrasG12V. Secondly, miR-21 was strongly induced becoming the most highly

144 expressed miRNA, consistent with its direct activation by oncogenic Ras reported in other

145 systems [25]. The upregulation of miR-21 is also consistent with its well-appreciated oncogenic

146 function in skin cancer [26]. miR-146 was also induced by HrasG12V. However this miRNA is

147 expressed 2-orders of magnitude lower than miR-21, suggesting that its upregulation may have

148 limited contribution to Hras initiated tumorigenesis at this early stage.

149 We further measured mature miR-21, miR-203, and miR-205 RNAs by qPCR. In support

150 of the quantitative performance of our miR-Seq, the differential expression of all three miRNAs

151 measured by qPCR was nearly identical to the quantification by our miR-Seq (Figure 2A-B). To

152 initially probe the mechanism through which HRASG12V suppresses miR-203 expression, we

153 examined the level of miR-203 primary transcripts. We previously characterized the transcribed

154 genomic region of miR-203 including the promoter region and transcription start site [17].

155 Because the primary transcript of miR-203 harbors a polyadenylation [Poly(A)] signal and

156 generates a Poly(A) tail, we directly quantified the abundance of the primary transcripts by

157 counting the 3’end reads of the primary miRNA obtained by 3seq (Figure 2C). This result was

158 further confirmed by qPCR measurement specific to the pri-miR-203 (Figure 2D). The degree of

159 downregulation for both mature and primary miR-203 transcripts was similar as judged by these

160 two independent assays. Collectively, we conclude that the repression of miR-203 by HrasG12V is

7 161 most likely mediated by suppressing the production of primary miR-203 transcripts at an early

162 stage of oncogenic cellular transformation.

163

164 miR-203 silencing is an early event in mouse and human SCC

165 Our data revealed the early silencing of miR-203 by oncogenic Hras in an in vitro model.

166 We further investigated miR-203 expression with in situ hybridization during skin tumorigenesis

167 in vivo. In classic chemical carcinogenesis models initiated by DMBA/TPA treatment of mouse

168 skin, Hras is preferentially mutated, which compromises the GTPase activity and results in

169 constitutive Hras activation. Mutating Hras at Q61 leads to papilloma development and

170 infrequent malignant transformation to SCCs [27]. We first examined miR-203 expression in

171 benign papillomas. Consistent with our in vitro results, miR-203 expression was absent from

172 epithelial compartments adjacent to tumor stroma, the region where putative cancer stem cells

173 reside. Although we also observed moderately expressed miR-203 in tumor regions with

174 evidence of cellular differentiation (keratin pearls and large differentiated morphology), the

175 levels of miR-203 was considerably lower in these regions compared to the suprabasal cells of

176 adjacent normal skin, where miR-203 is normally expressed (Figure 2E). Overall, miR-203

177 expression levels were strongly reduced in tumor tissues compared to adjacent epidermal regions.

178 In an independent mouse model of SCC, we also found that miR-203 expression was gradually

179 lost in a KrasG12D/Smad4 cKO model where skin tumors progressed to invasive SCC through

180 serial passages of tumors [28] (Figure 2F). Taken together, these results demonstrate that miR-

181 203 is significantly downregulated at both early and late stages of papilloma and SCC formation

182 in mouse models.

8 183 To evaluate the relevance of the loss of miR-203 in human skin cancer, we examined 9

184 tumor samples obtained from patients with the early, middle and late stages of skin cancer. In

185 these human skin SCC samples, miR-203 was already downregulated at the onset of

186 tumorigenesis and progressively lost during the course of tumor progression, similar to the

187 pattern observed in the mouse models. In the poorly differentiated SCC samples, the miR-203

188 signal was completely absent, yet readily detectable in surrounding hyperplastic or normal

189 epidermal regions (Figure 2G-O). In addition, we mined publically available datasets from The

190 Cancer Genome Atlas (TCGA) and detected a significant reduction in miR-203 and an elevation

191 of miR-21 in head and neck SCC samples compared to patient matched normal tissues (Figure

192 2P). Altogether, these results corroborate our observations in mouse models and validate the

193 strong correlation between the loss of miR-203 and the development of skin cancer at multiple

194 stages of tumorigenesis. These expression analyses suggest that the loss of miR-203 coincides

195 with the tumor-initiating events and miR-203 might function as a tumor suppressing mechanism

196 in skin cancer.

197

198 Genetic deletion of miR-203 impacts early epidermal development in mouse

199 We then generated a conditional KO mouse model to assess the function of miR-203 in

200 murine skin development and carcinogenesis in vivo. We have characterized the genomic locus

201 of miR-203 and determined that miR-203 is located in an intergenic region, 3.3 kbp downstream

202 from the Asp gene and 15.3 kbp upstream of the Kif26a gene (Figure 3-figure supplement 1).

203 Two loxP sites were inserted to flank the miR-203 hairpin (Figure 3A). miR-203 was deleted by

204 first mating miR-203fl/fl mice with Actb-Flpe mice to remove the Neo cassette, followed by

205 breeding with EIIa-Cre or Krt14-Cre mice, resulting in complete miR-203 loss from all tissues or

9 206 only from skin tissues, respectively. In both cases, ablation of miR-203 was confirmed by qPCR

207 on isolated epidermis and in situ hybridization (Figure 3B-C). We previously demonstrated that

208 expression of miR-203 was largely restricted to stratified epithelial tissues. Within the skin, the

209 differentiated skin lineages express miR-203 ~10-fold higher than the stem/progenitor lineages.

210 Consistent with this observation, there are no discernible differences between constitutive miR-

211 203 null (miR-203-/-) mice and skin-specific cKO (Krt14-Cre/miR-203fl/fl) mice. Both strains

212 were born at the expected Mendelian ratio. They also showed no signs of gross developmental

213 defects as adults (Figure 3D). For all subsequent studies we used miR-203-/-, which was

214 maintained in the C57BL/6 background.

215 miR-203 is highly expressed primarily in stratified epithelial tissues, such as the

216 epidermis, tongue, esophagus, and cervix, yet poorly expressed or absent in other tissues such as

217 the small intestine, bladder, lung, kidney, liver, or brain (Figure 3-figure supplement 2). Because

218 miR-203 begins to be expressed by E13 when the epidermis begins to stratify, we examined the

219 proliferation rate and thickness of embryonic skin from E16 to P4. At E16, we observed a mild

220 increase in cell proliferation in the KO, as measured by BrdU incorporation (Figure 3E-F).

221 Although the difference in BrdU incorporation did not achieve statistical significance (p=0.07),

222 the thickness of the KO epidermis was significantly increased, compared to WT littermates

223 (p=0.03). Interestingly, the increase in epidermal thickness was most prominent at early stages

224 (E16 and E17) and waned as skin development progressed (P4) (Figure 3G). At P4, miR-203-/-

225 mice displayed normal histological development of the epidermis and hair follicles and the

226 difference in proliferation and epidermal thickness between KO and WT became

227 indistinguishable (Figure 3G). In addition, we found no evidence of perturbed differentiation in

228 miR-203-/- mice based upon analysis of early and late epidermal differentiation markers, Keratin

10 229 1 and Loricrin, for the spinous and granular layers, respectively (Figure 3H-I). Together, these

230 results provide evidence that miR-203 limits cell division when the proliferation rate is high

231 during early embryonic skin development but not at later stages when the proliferation rate

232 wanes.

233 We noted that the impact of miR-203 loss was correlated with the rate of cell proliferation.

234 To further test whether miR-203 functions to restrict the expansion of highly proliferative cells,

235 we investigated the roles for miR-203 in regulation of primary and established keratinocyte cell

236 lines, respectively. We observed ~2-fold higher colony forming capacity of miR-203-/-

237 keratinocytes, compared to the WT controls (Figure 3J). To further confirm the ability of miR-

238 203 to suppress cell proliferation cell-autonomously, we generated a miR-203fl/fl keratinocyte cell

239 line. By treating these cells with an adenoviral vector to express Cre (Ad-Cre), we determined

240 that within 48 h of Ad-Cre exposure miR-203 was completely depleted ( <1% remaining as

241 measured by qPCR). We again observed a similar, ~2-fold higher colony forming capacity by the

242 Ad-Cre treated miR-203fl/fl cells, compared to the Ad-GFP treated control cells (Figure 3K). In

243 both cases, although we detected some slightly larger clones with more cells formed by the miR-

244 203 null keratinocytes, the biggest differences were the significantly increased number of clones

245 formed by the miR-203 null cells. These results suggest that miR-203 limits the clonogenicity of

246 normal keratinocytes. When miR-203 is deleted, more cells are likely to become clonogenic.

247

248 Loss of miR-203 enhances chemical carcinogenesis in the skin

249 We determined that the loss of miR-203 is an early event in the initiation and

250 development of both mouse and human skin SCCs. Furthermore, genetic deletion of miR-203

251 confers ~2-fold higher colony forming ability on primary keratinocytes. Because oncogenic Hras

11 252 is a potent driver for tumorigenesis in the skin and our miR-seq data revealed a rapid and strong

253 repression of miR-203 by HrasG12V, these observations prompted us to investigate whether the

254 loss of miR-203 plays a role in skin carcinogenesis. We subjected WT and miR-203 null mice to

255 two-stage chemical DMBA/TPA carcinogenesis (Figure 4A). The chemical carcinogenesis

256 experiments were terminated at week 21 when tumor burden had reached a maximum. Because

257 our mice were generated in the C57BL/6 background and these mice are known to be resistant to

258 two-stage chemical carcinogenesis [27], all tumors generated in our mice were papillomas with

259 no evidence for malignant conversion to squamous cell carcinomas during the time frame of our

260 study. Nevertheless, we examined the temporal and numeric characteristics of tumor formation

261 in our mice for the role of miR-203 in tumor initiation. During the course of carcinogenesis, we

262 observed a slightly earlier tumor formation pattern on the backskin of miR-203 null animals

263 (Figure 4B). Furthermore, miR-203 null animals developed ~2.5-fold more tumors, when

264 compared to WT control animals (Figure 4B). Measurement of tumor sizes at week 17 showed

265 no statistically significant difference in tumor sizes between genotypes although miR-203 null

266 animals were more susceptible to tumor formation (Figure 4C). These results suggest that loss of

267 miR-203 increases the number of tumor-initiating cells but does not significantly alter the type of

268 tumors e.g. conferring tumors with more aggressive phenotypes.

269 Hematoxylin and Eosin staining revealed that the papillomas produced had similar

270 morphology, displaying exophytic lesions with evidence for squamous differentiation (Figure

271 4D-E). Assessment of proliferation (Ki67), differentiation markers (Krt1, Lor) revealed similar

272 dynamics between miR-203+/+ and miR-203-/- tumors (Figure 4F). To further probe the

273 mechanistic differences between miR-203 WT and KO tumors, we assessed the mutation spectra

274 of these tumors and found that they possess the canonical HrasQ61L mutation at similar

12 275 frequencies, 80% and 85.7% for WT and KO tumors, respectively (Figure 4-figure supplement

276 1). Taken together, these results provide direct evidence for miR-203’s tumor suppressing roles

277 at the stage of tumor initiation in the classic DMBA/TPA tumorigenic model.

278

279 miR-203 represses clonal selection of oncogenic Hras transformed cells in vitro

280 To further probe miR-203’s role in clonal selection, we infected miR-203 WT and null

281 primary keratinocytes with pBabe-HrasG12V. At the passage 1, the loss of miR-203 led to ~40%

282 increase in colony forming capacity immediately after the initial plating (Figure 5A). When we

283 passaged the HrasG12V transduced cells, we began to observe reduced numbers of colonies that

284 were formed by both the WT and null cells (Figure 5A). This was likely due to oncogenic stress

285 caused by HrasG12V induction. Strikingly, in contrast to the WT cells that generally formed

286 smaller colonies in subsequent passages, the miR-203 null cells generated more and bigger

287 clones compared to WT control cells at each passage (Figure 5A). Collectively, the serial

288 passage assay supported the enhanced selection of tumor-initiating cells in the absence of miR-

289 203 upon oncogenic Hras induction.

290 To further characterize the ability of miR-203 to suppress the growth of oncogenic Hras

291 transformed cells, we used our previously established Krt14-rtTA/pTre2-miR-203 inducible

292 model [17]. After infecting the inducible keratinocytes with either the pBabe vector control or

293 pBabe-HrasG12V, we treated the cells with 5ug/ml doxycycline to induce ~4-7-fold increase in

294 miR-203 expression, a physiologically relevant level of miR-203 typically observed during

295 epidermal differentiation (Figure 5B). Introduction of miR-203 resulted in suppression of

296 keratinocyte proliferation and colony formation ability (Figure 5C-D), as noted previously

297 [17,18,29]. Furthermore, whereas oncogenic Hras enhanced S-phase entry, short-term (~24 h)

13 298 induction of miR-203 completely abolished the gain of S-phase entry (Figure 5D). Over a longer

299 term, continuous induction of miR-203 severely compromised the colony forming capacity of the

300 transduced cells (Figure 5C). We did not detect any evidence for enhanced apoptosis caused by

301 miR-203, as measured by the absence of sub-G1 keratinocytes. Taken together, our results

302 suggest that the loss of miR-203 is critical for the initial selection and expansion of primary

303 keratinocytes harboring the oncogenic Hras mutation and the gain of miR-203 can effectively

304 suppress the growth of these cells.

305

306 Comprehensive identification of miR-203 targets reveals regulation of the Ras signaling

307 pathway

308 Our data so far have suggested a role of miR-203 in suppressing the expansion of tumor-

309 initiating cells driven by oncogenic Hras mutations. To decode the underlying mechanism, we

310 carried out comprehensive analyses to identify miR-203 targets in the skin. Recent studies

311 demonstrated that miRNAs’ impact on gene expression could be largely captured by measuring

312 the changes of mRNA levels upon manipulation of miRNA expression [30,31]. In parallel, the

313 HIgh-Throughput Sequencing of RNA isolated by CrossLinking ImmunoPrecipitation (HITS-

314 CLIP) approach directly crosslinks miRISC to their targets and identifies miRNA targets through

315 physical interaction [32]. Therefore, we took a combinatorial approach integrating multiple

316 datasets obtained from our KO and inducible mouse models with Ago2 HITS-CLIP analysis.

317 We applied several different profiling techniques including ribosome profiling, RNA-Seq

318 (3Seq) and microarray methods to determine upregulated genes in miR-203 KO samples and

319 downregulated genes in miR-203 induced samples. For ribosome profiling and 3Seq, we used

320 primary keratinocytes, which are amenable to effective cycloheximide treatment and therefore

14 321 also allowed us to investigate miR-203’s impact on translation efficiency (Figure 6A). For

322 microarray analysis, we used freshly isolated epidermis obtained from miR-203 WT and KO

323 animals at P4.5. To interrogate the impact of miR-203 gain-of-function, we used the Krt14-

324 rtTA/pTre2-miR-203 inducible mouse model that allows us to control the duration and dosage of

325 miR-203 expression [17]. We applied microarray profiling to determine downregulated genes in

326 FACS-purified neonatal epidermal and hair follicle progenitors with a short-term (24 hr)

327 induction of miR-203 (Figure 6A). Altogether, we accrued 20 genome-wide expression datasets

328 from miR-203 WT, KO and inducible samples. This comprehensive collection of functional

329 genomic data allowed us to characterize the action of miR-203 on the transcriptome.

330 We first analyzed the miR-203 overexpression samples. GO-analysis demonstrated that

331 genes involved in regulation of mitotic cell cycle, DNA synthesis and metabolism were

332 prominently downregulated upon miR-203 induction (Figure 6-source data 1). miR-203

333 upregulation in both epidermal and hair follicle progenitors resulted in strong suppression of

334 transcripts harboring perfect seed matches to miR-203 in their 3’UTRs (Figure 6-figure

335 supplement 1). Among them, transcripts containing the 8-mer matches were strongly suppressed,

336 compared to transcripts without a miR-203 seed match (p-value = 1.5x10-23 for epidermis and

337 2.6x10-25 for hair follicle, respectively) (Figure 6 - figure supplement 1A). In contrast to miR-203

338 overexpression, deletion of miR-203 did not perturb transcript levels genome-wide as strongly as

339 observed under the induced conditions, and offered little insight into miR-203’s function based

340 upon GO-analysis (Figure 6-source data 1). However, despite the relatively mild changes in the

341 transcript levels in the KO skin and primary keratinocytes, transcripts that contain 8-mer 3’UTR

342 matches were still more significantly upregulated than transcripts without any miR-203 seed

343 matches (p <0.05) (Figure 6-figure supplement-1B). Lastly, a ranked sum analysis that ranked

15 344 genes based on expression changes correlated with the levels of miR-203 showed that genes with

345 the 8-mer or 7-mer seed ranked significantly higher than those transcripts without seed matches

346 (Figure 6-figure supplement-1C).

347 By demanding all potential targets be upregulated in all the miR-203 KO samples and

348 downregulated in all the miR-203 induced samples, we identified 294 transcripts as our

349 candidates for miR-203 targets (Figure 6B). Among the top 20 most differentially expressed

350 transcripts when miR-203 was deleted or induced, 15 of them contain at least one 7-mer or 8-mer

351 match in their 3’UTRs (Figure 6C and Figure 6-source data 2). In support of this approach, an

352 unbiased de novo motif search using the 3’UTRs of these 294 transcripts revealed that the most

353 enriched motif (ACAUUUCA, p=1x10-13) perfectly matched to nucleotide position 2-9 of miR-

354 203, the 8-mer seed sequences of miR-203 (Figure 6D). Additional statistical analysis of the

355 number of genes containing perfect 3’UTR matches to the 7-mer or 8-mer seed among these

356 candidates revealed significant enrichment over the background distribution among mRNAs

357 expressed in the skin (Figure 6D). Therefore, we selected genes containing 7-mer or 8-mer seed

358 matched 3’UTR sites as miR-203 target candidates and obtained a collection of 100 potential

359 miR-203 targets (Figure 6B). .

360 We then applied HITS-CLIP to interrogate the direct interaction between miR-203 and its

361 mRNA targets. We generated four Ago2 HITS-CLIP libraries from primary WT keratinocytes,

362 which abundantly express miR-203 (Figure 1G). Ago2-RNA complexes were isolated from a

363 region extending from approximately 110kd – 130kd as expected (Figure 6-figure supplement

364 2A). Sequencing libraries from Ago2 HITS-CLIP samples were generated and analyzed using

365 previously described methods [32,33]. The HITS-CLIP faithfully captured miRNA species

366 expressed in keratinocytes, including miR-203 (Figure 6-figure supplement 2B and E). Overall

16 367 our HITS-CLIP reads and clusters alignment were similar to previous published results with a

368 significant portion aligned to 3’UTRs (Figure 6-figure supplement 2D). To further validate our

369 approach, we analyzed the positional enrichment of miRNA seed sequences within the 3’UTR

370 HITS-CLIP clusters and detected strong enrichment over dinucleotide shuffled cluster sequences

371 or randomly distributed 3’UTR regions (Figure 6-figure supplement 3A). Additionally, de novo

372 motif searching for 8mer motifs identified the seed sequences of miRNAs that are highly

373 expressed in keratinocytes (Figure 6-figure supplement 3B). The motif corresponding to miR-203

374 detected by the HITS-CLIP was also ACAUUUCA (p=1x10-19), identical to the motif detected

375 by our transcriptome analysis (Figure 6E and Figure 6-figure supplement 3B). A total of 113

376 mRNAs were detected to have miR-203 seed containing Ago2 binding sites. We next examined

377 the ranked sum expression of these targets and determined that transcripts with Ago2 HITS-CLIP

378 miR-203 binding sites were ranked significantly higher compared to non-targeted mRNAs,

379 indicating that many of these targets are functional (Figure 6-figure supplement 4). Finally, we

380 did not detect any evidence for global regulation at the translation level for miR-203 targets

381 based on our analysis of translational efficiency changes upon miR-203 deletion (Figure 6- figure

382 supplement 5). Together, these HITS-CLIP data independently validated that miR-203 uses its

383 seed sequences for mRNA targeting and predicted 113 miR-203 targets based on Ago2 binding.

384 By combining the targets detected by our differential expression datasets and by the

385 HITS-CLIP, we identified a list of high-confidence targets for miR-203 (Figure 6F). Importantly,

386 we found a number of key regulators in the Ras signaling pathway and important genes involved

387 in regulation of cell division including Hbegf, Ccnd1, Snai2, Met, and Pola1 in this list (Figure

388 6F). This collection of miR-203 targets suggests that miR-203 targets multiple pathways,

389 including several components of the Ras signaling pathway, to suppress cell proliferation.

17 390

391 miR-203 directly targets Hbegf and Pola1

392 Our genome-wide analyses identified a number of novel targets of miR-203. Given our

393 findings that the loss of miR-203 promotes the selection and expansion of oncogenic Hras

394 transformed cells both in vivo and in vitro, we were interested in understanding the underlying

395 mechanism. Overall, miR-203 targets identified in the meta-analysis were enriched in the

396 upregulated transcripts in HrasG12V transformed keratinocytes, compared to non-targeted

397 transcripts, consistent with the downregulation of miR-203 in these cells (Figure 7-figure

398 supplement 1A ). Among these targets, we were intrigued by the observation that multiple

399 regulators of the Ras signaling pathway and critical factors for DNA replication and cell cycle

400 progression were among our high confidence targets and additionally were among the most

401 upregulated miR-203 targets in HrasG12V transformed keratinocytes (Figure 7-figure supplement

402 1B ). To begin to validate the high-confidence set of miR-203 targets, we selected Pola1 and

403 Hbegf for further study. Pola1 is the catalytic subunit of the DNA-POL-alpha holoenzyme,

404 which is required in initiation of DNA replication during S-phase [34]. Hbegf is an Egf-like

405 ligand that activates MAPK signaling through activation of EGF-receptors, Erbb1 and Erbb4. In

406 keratinocytes, Hbegf is mitogenic and promotes keratinocyte migration [35]. In an epithelial

407 cancer cell line, Hbegf acts as an oncogene promoting cell proliferation [36]. Pola1 and Hbegf

408 contain 3’ UTR miR-203 target sites (9-mer and 8-mer respectively) that are targeted by miR-203,

409 validated by luciferase assay (Figure 7A). Furthermore, in miR-203 null epidermis, both Pola1

410 and Hbegf mRNAs were elevated (Figure 7B). In addition to mRNA levels, Pola1 protein levels

411 were also elevated in the absence of miR-203. It was further elevated in the presence of HrasG12V

412 and repressed by miR-203 induction (Figure 7C-D). We were unable to measure the protein level

18 413 of Hbegf due to poor antibody performance. Additionally, we observed that the expression of

414 Ccnd1, an essential cell cycle regulator that is often induced or amplified by oncogenic Ras

415 [9,37], showed a strong negative correlation to miR-203 (Figure 7C-D). This suggested that loss

416 of miR-203 increases the levels of Ccnd1 and contributes to the observed upregulation of this

417 critical gene.

418 To assess the functional consequences of Hbegf and Pola1 suppression on keratinocyte

419 proliferation, we knocked down these targets using three independent shRNAs. Suppression of

420 Hbegf and Pola1 strongly suppressed the growth potential of keratinocytes (Figure 7E).

421 Additionally, we knocked down Hbegf and Pola1 in established miR-203+/+ and miR-203-/-

422 keratinocytes transduced with HrasG12V and similarly observed potent suppression of

423 keratinocyte growth potential, demonstrating that these targets are also required in HrasG12V

424 transformed keratinocytes (Figure 7-figure supplement 2 ). Taken together these results validate

425 Hbegf and Pola1 as direct targets of miR-203. We hereby propose a model in which miR-203

426 restricts selection and expansion of Hras mutated cells by repressing multiple targets, a subset of

427 which are involved in the Ras signaling pathway (Figure 7F).

428

429 Discussion

430 Identification and experimental validation of driver gene mutations have been

431 instrumental for our understanding of cancer biology [1]. Although it is clear that not all cells

432 harboring driver gene mutations develop into tumors, the mechanism of clonal selection within

433 cells that have acquired driver gene mutations remains poorly understood. In this study, we used

434 an oncogenic Hras induced tumor model to determine the molecular consequences of this

435 oncogenic driver mutation and to study the selection process. We quantitatively measured the

19 436 impact of oncogenic HrasG12V on the landscape of mRNA and miRNA expression. Whereas our

437 mRNA-Seq data confirmed the profound ability of HrasG12V in promoting cellular transformation,

438 our miR-Seq data revealed new insights into the dynamic changes in highly expressed miRNAs

439 caused by HrasG12V (Figure 1). In particular, the downregulation of miR-203, one of the most

440 highly expressed miRNAs in the skin, likely occurs at the transcriptional level (Figure 2C-D). By

441 examining tumor samples obtained from both mouse and human at multiple stages, we showed

442 that the downregulation of miR-203 is associated with tumor initiation and progression (Figure

443 2E-P). To determine the role of miR-203 in this process, we generated a KO mouse model. Initial

444 insights into the function of miR-203 came from the analysis of miR-203 null epidermis during

445 skin development. Whereas we and others have shown the potent inhibition of epidermal

446 proliferation by miR-203 with gain-of-function approaches [17,18,29,38], it is unknown how the

447 loss of miR-203 affects the skin and mouse development in general. By analyzing the KO mouse,

448 we revealed an early, albeit mild, hyperproliferative phenotype in the embryonic but not

449 postnatal skin (Figure 3E-I). However, when we subjected miR-203 WT and null cells obtained

450 from postnatal skin for their colony forming ability, we observed an increased ability of the miR-

451 203 null cells to give rise to colony forming cells (Figure 3J-K). This observation was indicative

452 of a role for miR-203 in restricting the expansion of highly proliferative cells. Of note, we did not

453 observe any defects in the formation of the differentiated layers of the skin in the miR-203 KO

454 mice. This suggests that the primary function of miR-203 is to restrict cell proliferation but not to

455 promote epidermal differentiation per se. When we examined the role of miR-203 during

456 DMBA/TPA mediated chemical carcinogenesis, we observed strong increase in the number of

457 papillomas formed in the KO skin (Figure 4A-C). In vitro colony formation assays with serial

458 passages further illustrated the enhanced ability for selecting tumor-initiating cells with miR-203

20 459 KO cells when subjected to HrasG12V transformation (Figure 5A). However, due to tumor

460 resistance of our C57BL/6 strain, we were unable to determine if the loss of miR-203 also

461 promotes malignant transition. Of note, a recent study showed interesting results that restoration

462 of miR-203 suppresses human SCC metastasis [29]. Collectively, our results provide

463 experimental evidence for an important role of miR-203 in suppressing clonal selection and

464 expansion of Hras mutated nascent cancer cells.

465 A major challenge in understanding miRNA functions is to identify their high-confidence

466 targets globally. To this end, we employed two independent approaches: transcriptome analysis

467 with our KO and inducible models and direct miRNA target capture by Ago2 HITS-CLIP (Figure

468 6). Importantly, with our expansive dataset, we confirmed that miR-203 utilizes the 5’ seed

469 sequences to target perfectly matched sequences located on the 3’UTRs of its target genes.

470 Consistent with recent genome-wide studies for the impact of miRNA on mRNA levels and

471 translation efficiency in mammals [30,31], we found no evidence for global changes of

472 translational efficiency in the absence of miR-203, one of the most highly expressed miRNAs in

473 the skin. However, we also noted that Trp63, a well-established miR-203 target, showed little

474 change at the mRNA levels under all conditions (not shown). Our analysis with differential gene

475 expression did not identify Trp63 as a miR-203 target. In contrast, the recognition of the 3’UTR

476 site of Trp63 by miR-203 was robustly detected by the HITS-CLIP (Figure 6E). Thus, it is likely

477 that our identification of miR-203 targets based on the transcriptome analysis alone was

478 conservative. More studies will be required to further integrate direct miRNA target capture with

479 transcriptome analysis for miRNA target identification. Nevertheless, we still observed strong

480 enrichment of miR-203 targets in genes required for cell proliferation. These high-confidence

481 targets illustrate the broad targeting of cell proliferation by miR-203 and, suggest mechanisms by

21 482 which loss of miR-203 can facilitate the expansion of oncogenic Hras transformed cells. These

483 new insights point to the potential of utilizing miR-203 to simultaneously target multiple

484 effectors required for cell division in human cancers. We also note that miR-203 is known to

485 antagonize human papillomaviruses (HPV) [39]. Our identification of Pola1, a critical cellular

486 gene required for HPV genome replication, as a target of miR-203 may provide a potential

487 mechanism for miR-203-mediated antagonism against HPV infection. Finally, the relatively mild

488 requirement for miR-203 in slowly proliferating cells suggests that a reduction or increase in the

489 level of miR-203 may be well tolerated by most normal cells. This unexpected discovery could

490 be further explored to determine miR-203’s therapeutic potential in treating certain types of

491 epithelial cancers.

492

493

494

495

496 Materials and Methods

497 3seq library construction:

498 3’seq libraries were constructed using methods described previously [19]. Briefly, 500 ng

499 of total RNA isolated from primary keratinocytes was poly-A purified (Dynabead®Invitrogen)

500 and chemically fragmented by treatment 95°C for 8 minutes in Fragmentation Buffer (Ambion).

501 Fragmented RNAs were then oligoDT primed with a P7T20V oligo and reverse transcribed

502 using SuperScript® III (Invitrogen)(see Supplementary File 2 for Oligo Sequences). Following

503 ethanol precipitation, ligation competent cDNAs were generated via second-strand synthesis

504 using RNase H and DNA Pol I enzymes, end-repaired using a mix of T4 DNA polymerase,

22 505 Klenow DNA polymerase and T4 Polynucleotide Kinase in the presence of dNTPs (NEB), and

506 A-Tailed using Klenow 3’ to 5’ exo- in the presence of dATP (NEB). Ligation was then

507 performed using the P5 Adaptor with T4 DNA Ligase for 1 hour (NEB). cDNA inserts 80-100 nt

508 in length were isolated from 8% PAGE gels and subject to PCR amplification using RP1 and

509 RT-Primer oligos for 12-16 cycles (Phusion® NEB). PCR products were isolated from PAGE

510 gels and subject to 6 cycles of secondary PCR to introduce unique indices for multiplexed DNA

511 sequencing on a HiSeq 2000 (Illumina).

512 Bioinformatics processing of 3seq data was performed as described previously, including

513 adaptor trimming, read alignments, peak calling, peak filtering to remove internal polyA priming

514 events, and transcript quantification with minor modifications [19]. Following read alignment to

515 the mm10 genome, alignments from every library were grouped together for peak calling and

516 3’end filtering to create a master set of high confidence 3’ end peaks. From this 3’end dataset,

517 reads counts for each library were obtained (See Supplementary File 1 for mapping statistics). To

518 obtain transcript read counts, reads from all peaks that passed our internal priming filter and were

519 3’UTR localized were summed to obtain an overall transcript count for each transcript.

520 Normalized transcript counts were calculated as Reads per Million reads mapped (RPM). To

521 determine differential transcript expression, low-abundance transcripts with less than 10 reads in

522 two libraries were first excluded then the remaining transcripts were analyzed with EdgeR with

523 classical analysis parameters.

524 Small RNA-Seq:

525 Small RNA-Sequencing libraries were prepared using minor modifications of a

526 previously published protocol that reduces RNA-ligation biases and enables accurate miRNA

527 quantification [24]. Briefly, two micrograms of total RNA, was first ligated to a pre-adenylated 3’

23 528 linker (10 uM) using truncated T4 RNA Ligase 2 (20 Units NEB) in the presence of PEG-8000

529 (10% w/v) RNase-Out™ (20 Units Invitrogen) for 4 hours at room temperature. Ligation

530 reactions were then resolved on 7 M Urea-15% PAGE gel to isolate the miRNA-3’ adaptor

531 hybrids. Following overnight elution from the acrylamide gel in HSCB buffer (400 mM NaCl, 25

532 mM Tris-HCl pH 7.5, 0.1% SDS), ligation products were ethanol precipitated, resuspended in 5’

533 ligation reaction mixture containing a 5’ linker (10 uM), 1 mM ATP, Peg-8000 (20% w/v), 1x

534 T4 Rna Ligase buffer (NEB), denatured for 5 minutes at 70°C, after which RNase-Out™ (20

535 units) and T4 RNA Ligase 1 (10 units) were added. Following ligation for 37 °C for 2 hours,

536 cDNA was generated via reverse transcription using Superscript® III (Invitrogen) in the presence

537 of a 3’adaptor specific primer. cDNA products were then subjected to 10-14 PCR cycles and

538 resolved on 8% native PAGE gels. Libraries were then eluted as described above, precipitated,

539 and submitted for sequencing on a HiSeq 2000 (Illumina).

540 Small-RNA sequencing reads were analyzed following previous described methods with

541 the following modifications [24]. Adaptor Sequences were trimmed from reads using CutAdapt

542 using default parameters. Reads were then further trimmed to remove the randomized adaptor

543 dinucleotides on the 5’ and 3’ end. Read were next aligned to a database of mouse miRNA

544 sequences (miRbase) using blastn with the following settings (blastn -word_size 11 -outfmt 6 -

545 strand plus). Blast alignments were then parsed with custom python scripts to extract and count

546 the best miRNA alignment for each read with a minimum read alignment of 18 nucleotides.

547 miRNA counts for each library were then filtered to keep only miRNAs with a minimum count

548 of 50 reads in two libraries, and analyzed for differential expression using EdgeR, with classical

549 analysis parameters.

550 Affymetrix microarray analysis:

24 551 For the miR-203 epidermal loss-of-function microarray analysis, RNA was isolated from

552 total epidermal samples from two-pairs of miR-203+/+ and miR-203-/- animals at p4. For the

553 doxycycline-inducible miR-203 over-expression microarray analysis, RNA was isolated from

554 two pairs of doxycline induced or uninduced Krt14-rtTA/ pTre-miR-203 animals using FACS

555 sorting for Krt14-H2B-GFP+ cells as described previously [17]. The microarray analysis of miR-

556 203 overexpression in basal epidermis was previously published [17]. Subsequently total RNAs

557 (500 ng) were processed and hybridized to the Mouse Genome 430 2.0 array (Affymetrix, USA)

558 following the manufacturer’s instruction at the MCDB microarray facility. Microarray image

559 files were processed using the R Bioconductor suite and Mas5 normalization. Probesets were

560 then filtered to include only those probes with Present or Absent calls in at least two arrays.

561 Probesets were then collapsed using the probeset with the maximum averaged probeset intensity

562 to represent each GeneID. Log2 fold changes were then computed using the limma Bioconductor

563 package.

564 Ribosome Profiling:

565 Ribosome profiling was performed on primary keratinocyte lysates using the ARTSeq™

566 Ribosome profiling kit (Epicentre). Briefly, lysates from a 10 cm dish of primary keratinocytes

567 were isolated in the presence of cycloheximide (Sigma Aldrich, 50 μg/ml) and subject to limited

568 RNAse digestion (10 units) for 45 min at room temperature. RNase digestion was terminated by

569 addition of 15 μl of Super-RNase IN™ (Ambion) followed by ribosome isolation using illustra™

570 MicroSpin™ S-400 HR Columns (GE Healthcare). Following RNA extraction and precipitation,

571 rRNA was depleted using the Ribo-Zero Gold™ kit (Epicentre), with the remaining RNA then

572 fractionated through 18% PAGE gels. RNA species 28-32 nt were isolated for adaptor ligation,

25 573 reverse transcription, circularization, and PCR amplification following the manufacturer’s

574 protocol. PAGE gel isolated PCR products were then sequenced on a HiSeq 2000 (Illumina).

575 Raw reads were first trimmed to removed 3’ adaptors using cutAdapt with default

576 parameters. Read were aligned to mm10 rRNA, tRNA, and ncRNA (Ensemble annotation)

577 databases using Bowtie (default settings) to exclude reads aligning to abundant rRNA, tRNA,

578 and ncRNA sequences. Unaligned reads were then aligned via Tophat using default settings,

579 with a supplied .gtf annotation file containing combined Refseq and Ensembl gene annotations

580 (iGenomes Illumina downloaded 9/4/2013). Uniquely aligned read counts were quantified across

581 each CDS using HTSeq Count (settings: -s yes -m union -t CDS) using the above-mentioned

582 GTF annotation database. Transcripts with low reads counts were excluded by only keeping

583 transcripts with at least 50 reads in at least two libraries. Filtered transcript reads count data was

584 then analyzed for differential expression using EdgeR with classical analysis parameters. To

585 calculate translation efficiency for each transcript, Reads Per Million Mapped (RPM) values

586 from 3Seq were divided by Reads per Million Mapped to coding-sequence for the Ribo-Seq. The

587 change in translation efficiency was then computed as the ratio of translation efficiency in the

588 miR-203-/- and miR-203+/+ libraries.

589 Ago2 HITS-CLIP:

590 Ago2 hits-clip was performed as described with minor modifications [32]. 15 cm2 dishes

591 of primary keratinocytes were irradiated twice at 200 mJ/cm with 254 nm UVC light. Following

592 irradiation cell lysates were harvested by scraping and stored at -80°C. Following thawing,

593 lysates were further lysed by trituration 3 times through pre-chilled 25 and 30 gauge needles.

594 Lysates were then treated with 10 μl Turbo™ DNase (Promega) and 5 μl RNase-Out™

595 (Invitrogen) per ml of lysate. Limited RNase digestion was performed using 10 μl per ml of

26 596 lysate of a 1:20 dilution of an RnaseA/T1 mix (Sigma/Ambion 1x mix = 3.33 μl RnaseA (2

597 μg/μl) with 6.66 μl RnaseT1 (1U/μl)). Crosslinked Ago2 was recovered via

598 immunoprecipiatation for 2 hours at 4°C with 3 μg of a monoclonal anti-mouse Ago2 antibody

599 (Wako Chemicals USA clone 2D4) complexed with Protein-G Dynabeads™ (Invitrogen).

600 Immunoprecipitates were washed twice with High-Salt Buffer and PNK buffer, then end-labled

601 with 25 μCi 32P y-ATP using PNK 3’ phosphatase minus (NEB) for 5 minutes at 37°C. After

602 washing the beads as listed above, 5’ adaptor ligation was performed for two hours at room

603 temperate using T4 RNA Ligase 1 with 10 uM 5’ RNA Linker, 20% PEG-8000 (w/v final), 1

604 mM ATP, and RNaseOut™ (Invitrogen). Beads were again washed twice with PNK buffer then

605 resuspended in a phosphatase reaction with 5 μL FastAP (ThermoFisher) with RNaseOUT™.

606 Following washing twice with PNK buffer, Protein-RNA complexes were eluted from the beads

607 using 1x Novex® Loading buffer supplemented with 50 mM DTT at 70°C for 10 minutes.

608 Protein-RNA complexes were then resolved on an 8% Bis-Tris Gel and transferred to

609 nitrocellulose. Membranes were exposed to a phosphor screen for 1-2 hours to obtain an

610 autoradiograph. Subsequently, RNA-protein complexes migrating in the 110-130 kD range were

611 excised. RNA was recovered from the nitrocellulose using Proteinase K treatment followed by

612 phenol-chloroform extraction and ethanol precipitation. Isolated RNA was ligated to a 3’linker

613 using the same ligation reaction conditions as for 5’ ligation, and ligated RNA species were

614 fractionated away from adaptor-adaptor products on 10% UREA Page gels. The RNA was eluted

615 from the PAGE gel with HSCB buffer overnight at 4°C, then ethanol precipitated and

616 resuspended for reverse transcription with SuperScript™ III (Invitrogen). cDNA products were

617 then subjected to PCR amplification for 20-24 cycles and fractionated on an 8% native PAGE

27 618 gel. PCR products representing cDNA inserts of 20-50 nts were recovered and subject to

619 sequencing on a HiSeq 2000 (Illumina)

620 HITS-CLIP reads were analyzed as follows. First, reads were processed to identify

621 miRNA alignments using the same pipeline that we use for Small-RNA-Seq listed above. Reads

622 not mapping to miRNAs were next processed as follows. Reads were trimmed with Cutadapt to

623 remove adaptor sequences using default settings. To avoid PCR duplicates from biasing the

624 analysis, duplicate reads were then collapsed to a single read using Fastx_collapser (default

625 settings). The 5’ and 3’ adaptor sequences contain randomized dinucleotides on their 3’ and 5’

626 ends respectively, which were next trimmed from the reads. The reads were then aligned to the

627 mm10 genome assembly using NovoAlign requiring a minimum alignment length of 25

628 nucleotides (settings –s 1 –t 85 –l 25) (Novocraft). All unique alignments from each library were

629 then pooled to identify Ago2 HITS-CLIP clusters. Clusters were defined as two read alignments

630 that overlap by a minimum of one nucleotide. Clusters were next annotated to gene features in a

631 hierarchical manner in which clusters were annotated to protein coding RefSeq 3’ UTRs (with

632 5kbp extension allowed), RefSeq CDS regions, RefSeq 5’ UTRs, Ensemble ncRNA regions, and

633 RefSeq intron regions. Clusters not found in these regions were annotated as intergenic clusters.

634 For predicting miRNA target sites, 3’UTR cluster sequences were searched for 6mer seed-

635 sequence matches for miRNA species that accounted for 90% of miRNAs expressed in

636 epidermis based on small-RNA sequencing. From this dataset 117 binding sites in 113 mRNAs

637 were predicted be miR-203 target sites.

638 Meta-analysis of miR-203 targets:

639 The miR-203 overexpression datasets used in the meta-analysis included previously

640 published microarrays from sorted H2B-GFP+ epidermal cells with transient miR-203 induction

28 641 (GSE45121), and microarrays from sorted H2B-GFPhi+ hair follicle progenitor cells with

642 transient miR-203 induction described in this manuscript. The miR-203 knockout datasets used in

643 the meta-analysis included the microarrays from p4.5 total epidermal samples from miR-203+/+

644 and miR-203-/- animals, ribosome profiling of miR-203+/+ and miR-203-/- primary keratinocytes,

645 and 3seq of miR-203+/+ and miR-203-/- primary cultures. We also performed 3seq on miR-203-/-

646 cells transformed with HrasG12V, however there was no enrichment for miR-203 seed matches in

647 the upregulated transcripts based on CDF analysis, consistent with the low levels of miR-203 in

648 HrasG12V transformed keratinocytes, and therefore this dataset was excluded from the expression

649 meta-analysis (Data not shown). Gene Symbols were used to compare across microarray, 3Seq

650 and Ribo-Seq datasets. Log2 fold changes were used to assess differential gene expression in

651 each dataset. In total 6,407 genes were detectable in all the datasets, from which 294 satisfied the

652 criteria of being upregulated in all the miR-203 knockout datasets and downregulated in the miR-

653 203 overexpression datasets. Negative control datasets were constructed to analyze the

654 enrichment of genes containing miR-203 seed matches with the following criteria, randomly

655 selected genes from the list of detected transcripts in the meta-analysis, or genes that have a

656 positive correlation to miR-203 expression (downregulated in miR-203 knockout datasets and

657 upregulated in miR-203 overexpression datasets).

658 In order to compare gain-of-function and loss-of-function datasets in aggregate, a ranked

659 correlation to miR-203 metric was calculated. Transcripts were assigned a rank in each knockout

660 dataset with the most upregulated gene given a rank of 1. Transcripts were next assigned a rank

661 in each overexpression dataset with the most downregulated gene given a rank of 1. The ranked

662 values from each of the 5 datasets were then summed and ranked with the transcript most

29 663 upregulated upon miR-203 ablation and downregulated upon overexpression being assigned a

664 value of 1.

665 Hierarchical Clustering, GO-Term Analysis, GSEA, 3’UTR motif-searching:

666 Normalized transcript abundances (expressed as Reads Per Million) were used for mean-

667 centered unsupervised hierarchical clustering using Cluster 3.0 software. For 3seq data, only

668 transcripts with at least 2-fold change were selected and for miRNA data, only miRNAs with at

669 least an expression level of 1000 RPM and at least a 2-fold change were selected for

670 visualization in JavaTree View. GO-term enrichment analysis was performed by DAVID using

671 Gene-IDs as input, with analysis being performed using GO Biological processes datasets.

672 GSEA analysis was performed using ranked expression values for the displayed dataset, and

673 genesets selected from the referenced publications. De novo motif searching was performed

674 using the HOMER package to search for 7 or 8mer motifs in RefSeq 3’UTR sequences for

675 selected gene-sets. For genes with multiple 3’UTR isoforms, the longest 3’UTR was selected for

676 motif searching. miRNA seed-sequence searches in Ago2-HITS-CLIP clusters were performed

677 with a Python script using regular expressions.

678 Analysis of The Cancer Genome Atlas (TCGA) Data:

679 Head and Neck SCC miRNA-Seq data was obtained from the TCGA https://tcga-

680 data.nci.nih.gov/tcga/ (Download date 10/23/2014). Patient matched normal solid tissue and

681 tumor miRNA quantification records were identified using custom R scripts (Regular Expression

682 query for tumor and solid tissue normal samples respectively "TCGA-[0-9A-Z]{2}-[0-9A-

683 Z]{4}-0", "TCGA-[0-9A-Z]{2}-[0-9A-Z]{4}-11"). The normalized reads_per_million

684 quantification for miR-203 and miR-21 was then plotted to determine the relative expression in

685 normal and tumor tissue samples.

30 686 Mouse strains and generation of the conditional miR-203 knockout mouse:

687 A gene targeting vector was constructed that contained 11 kbp homologous region

688 surrounding the miR-203 locus (Figure 3-figure supplement 1). LoxP were inserted flanking the

689 pre-miR-203 sequences, with a neomycin selection cassette flanked by Frt sites. The construct

690 was electroporated into Cy2.4 ES cells (B6(Cg)-Tyr genetic background). Positive clones

691 were identified by Southern Blot analysis using a probe complementary to the 3’ end of the

692 targeted homologous region. ES cells were injected into blastocysts and chimeras were screened

693 based on white/black coat color selection. Upon obtaining germline transmission, the neo

694 cassette was excised by breeding the F1 progeny to an Actb-Flpe line maintained on a C57BL/6J

695 background (obtained from Jackson Labs). miR-203floxed animals were then bred to a EIIa:Cre

696 line maintained on a C57Bl/6 background (obtained from Jackson Laboratory) or a Krt14:Cre

697 line maintained on a mixed background (obtained from E. Fuchs Laboratory), to obtain germline

698 or conditional ablation of miR-203. The EIIa:Cre transgene was removed from the germline

699 miR-203 deleted line by backcrossing to a C57Bl/6 line and subsequently maintained on a

700 C57Bl/6 background. pTre2-miR-203/Krt14-rtTA mice were generated as described previously

701 and maintained on an FVB background [17]. Mice were bred and housed according to the

702 guidelines of IACUC at a pathogen-free facility at the University of Colorado (Boulder, CO,

703 USA).

704 Immunofluorescence, EdU detection, and miRNA in situ hybridization:

705 Frozen cryostat sections (8 μM) were fixed in 4% Paraformaldehyde for 10 minutes at

706 room temperature, wash thrice with PBS, and blocked for 10 minutes using Gelatin Block (0.1%

707 Triton X-100, 2% gelatin, 2.5% normal goat serum, 2.5% normal donkey serum, and 1% BSA in

708 PBS). Primary antibodies, diluted in gelatin block, were then incubated overnight (See

31 709 Supplementary File 2 for antibody references). Following three washes with PBS, sections were

710 incubated with appropriate Alexa-Fluor secondary antibodies (1:2,000) for 1-2 hours at room

711 temperature. Following three washes with PBS, sections were stained with Hoescht Dye, and

712 mounted in Anti-fade solution. miRNA in-situ hybridization for miR-203 was performed on

713 frozen sections as described previously [18]. EdU detection was performed following

714 manufacturer’s instructions, with the following parameters. P4 animals were IP injected with 50

715 μg/g EdU 4 hours prior to tissue harvest. Following EdU detection, the sections were blocked

716 and probed with antibodies as described above. BrdU detection was performed as previously

717 described, with the following parameters. Pregnant female mice were IP injected with 50 μg/g

718 BrdU for 2 hours prior to embryo harvest in OCT compound. miR-203 in situ hybridization on

719 FFPE mouse and human tumor samples was performed with the following modifications, after

720 deparaffinization in xylenes, the tissue was treated with proteinase K (20 μg/ml) for an extended

721 period of 20 min at an elevated temperature (37°C). Microscopy images were obtained using a

722 Leica DM5500B microscope with either a Leica camera (brightfield) or Hamamatsu C10600-

723 10B camera (fluorescence) and processed with the Leica image analysis suite, MetaMorph (MDS

724 Analytical Technologies) and Fiji software. BrdU or EdU image quantifications were performed

725 by counting the number of Krt5+/EdU or BrdU positive cells in randomly chosen microscopy

726 fields. The length of the basement membrane was used to represent the length of the epidermis

727 analyzed and was determined by tracing the basement membrane and calculating line length

728 using Fiji software. Epidermal thickness was assessed by tracing a line tangential to the

729 basement membrane and extending to the beginning of the stratum corneum and calculating the

730 line length.

731 qPCR and western blotting:

32 732 qPCR was performed using the Qiagen miR-script RT system and a BioRad CFX-384

733 machine. Fold-changes were computed using the ΔΔCt formula normalized to sno25 and Hprt

734 values. In all qPCR figures error bars denote standard errors of the normalized mean. Western

735 blotting was performed using 20-40 μg of protein lysate run on 8-12% SDS-PAGE gels. Proteins

736 were transferred to PVDF for detection of Pola1, β-tubulin (Tubb5), or Ccnd1. Primary

737 antibodies were incubated in 5% BSA overnight and detected using HRP-conjugated secondary

738 antibodies and Amersam ECL-Plus reagents (1:10,000). See Supplementary File 2 for antibody

739 descriptions and dilutions. X-ray films were scanned and processed with Fiji software to

740 calculate relative protein abundance.

741 Primary Keratinocyte harvesting and cell culture, viral infections, Edu Flow Cytometry, and

742 shRNA knockdown:

743 Primary keratinocytes were isolated from neonatal mice using previously described

744 methods with the following modifications [40]. Isolated skin was incubated on a solution of

745 Dispase overnight at 4°C to dissociate the epidermis from the dermis. The following day

746 epidermal sheets were incubated in 37°C Trypsin for 10 minutes to isolate keratinocytes.

747 Primary keratinocytes were then plated in 6 cm or 10 cm dishes with E-Low media

748 supplemented with 0.2 mM Calcium Chloride for the first 24 hours then switched to E-Low

749 media with 0.05 mM Ca++. Lentiviral particles were produced by transient transfection of pLKO-

750 shRNA constructs, PsPax.2, and pVSVG. 24 hours post transfection the media was changed to

751 E-Low calcium. Retroviral particles were produced by transient transfection of pBabe-vector-

752 puro , pBabe-HrasG12V-puro, pBabe-vector-neo, or pBabe-HrasG12V-neo, with PCL-Eco and

753 pAdvantage packaging plasmids. Viral supernatant was harvested every 12 hours for up to 96

754 hours, pooled and filtered with 0.45 μM filter. Ad-eGFP or Ad-CREeGFP adenoviruses were

33 755 obtained from the Iowa Gene Transfer Core and used at MOI of 50. Retroviral and Lentiviral

756 infections were performed 3-4 days after plating primary keratinocytes. Keratinocytes were

757 selected with 2 μg/ml puromycin for 48 hrs or 50 μg/ml neomycin for 7 days, at which time non-

758 infected cell cultures were non-viable. Spontaneously immortalized miR-203+/+ , miR-203-/- , and

759 miR-203fl/fl mouse keratinocyte cell lines were also generated via serial passage on mitomycin-C

760 treated NIH-3T3 feeder cell culture layer and utilized for assays as noted in the text. Flow

761 cytometry was performed as previously described, with minor modifications [17]. Cell cultures

762 derived from pTre2-miR-203/Krt14-rtTA animals were treated with 5μug/ml doxycycline for 24

763 hours to induce miR-203 expression, pulsed with EdU (10 μM) for 30 minutes, harvested and

764 analyzed according to the Click-iT® EdU Plus instructions on an BD Cyan™ flow cytometer

765 (Invitrogen). For colony formation assays, 2,000 cells were split into individual wells of 6-well

766 plates, cultured for 10-14 days, fixed with 4% PFA, and stained with 0.2% Crystal Violet in 70%

767 Ethanol. For induction of miR-203, 24 hours after plating, cells were supplied with fresh media

768 containing doxycycline at 5 μg/ml. Sigma-Aldrich ™ TRC lentiviral shRNAs against Hbegf and

769 Pola1 were obtained from the Functional Genomic Facility (University of Colorado at Boulder,

770 sequences listed in Supplementary File 2).

771 3’UTR Luciferase Assays:

772 3’UTR reporter constructs were generated by PCR amplification of 3’UTRs from cDNA

773 or gDNA and subcloning of the fragments into pGL3-Control (Promega) (primer sequences

774 listed in Supplementary File 2). 2 ng renilla luciferase control, 20 ng pGL3-3’UTR reporter, and

775 380 ng of Krt14 empty vector, or Krt14-miR-203 were transiently cotransfected into

776 keratinocytes in each well of a 12 well plate using Mirus Bio LT1 reagent (MirusBio). 24 hours

777 later cell lysates were collected and renilla and firefly luciferase activity were measured using

34 778 Dual-Glo Luciferase Assay system (Promega) as described previously [18]. Data are represented

779 as the ratio of firefly to renilla RFU values, normalized to Pgl3-Control values. Error bars

780 represent propagated standard deviations.

781 DMBA/TPA carcinogenesis:

782 DMBA/TPA carcinogenesis was performed as described previously [27]. The backskin

783 of 7-9 week old miR-203+/+and miR-203-/- mice was shaved. 48 hours later the backskin was

784 painted with a single dose of 25 μg of DMBA in 200 μl of Acetone. Two weeks following

785 DMBA treatment the mice began receiving bi-weekly treatments of 4 μg of TPA in Acetone. The

786 number of palpable tumors of at least 1 mm in diameter, persisting for at least two weeks, was

787 recorded weekly. Tumor diameters were measured using a digital caliper. Following 21 weeks of

788 TPA treatment, mice were euthanized and tumors were collected for HrasQ61l genotyping, OCT

789 embedding, and paraffin embedding.

790 HrasQ61L genotyping of DMBA/TPA tumors:

791 Tumor DNA was isolated by incubating the tissue in a DNA Lysis Buffer (400 mM NaCl,

792 0.1% SDS, 1 mM EDTA,1 μg/ml Proteinase K) at 55° C for 4 hours. Lysates were then vortexed

793 and lightly centrifuged to liberate DNA from the partially-digest tumor tissue. The supernatant

794 was then removed and subject to Phenol-Chloroform Extraction, followed by Isopropanol

795 precipitation. Isolated DNA pellets were then resuspended in TE buffer and quantified by UV

796 spectrophotometry (10 mM Tris pH 8.0, 1 mM EDTA). The Hras gene was PCR amplified using

797 primers that flank exon 2. Following amplification, the PCR reactions were digested with 5 units

798 of XbaI restriction enzyme at 37° C. The reaction products were then resolved and visualized on

799 a 3% agarose gel. DNA isolated from the tails of animals in the DMBA/TPA experiment was

800 treated in parallel as a negative control for detection of the HrasQ61L mutation.

35 801 Statistical Analysis:

802 Statistical analysis was performed using either R or Microsoft Excel. Statistical methods

803 employed are indicated in the figure legends. Unpaired two-sided student t-tests were used to

804 assess statistical significance unless indicated otherwise in the figure legends. For comparisons

805 with multiple categories, ANOVA was used with Tukey’s HSD post-hoc test. Non-parametric

806 Whitney-Mann U-Tests were used to assess significance for the tumor multiplicity

807 measurements [27]. The hypergeometric test was used to assess the enrichment of gene lists in

808 genome-wide studies. The Kolmogorov–Smirnov test was used to assess differences in

809 cumulative distributions functions.

810 Data access:

811 All sequencing and microarray data are deposited in the Gene Expression Omnibus (GSE66056).

812

813 Author contributions

814 R.Y. conceived and supervised the study. K.R. performed the experiments. X.J.W., E.C.T., and

815 D.R.R. provided critical advice for tumorigenesis experiments and important reagents for

816 analysis of miR-203 expression patterns in mouse and human tumor samples. K.R. and R.Y.

817 designed the experiments and wrote the paper.

818

819 Acknowledgements

820 We thank C. Yang, J. Gao and D. Feng for the generation of the miR-203 KO mouse. We thank

821 members of the Yi laboratory for their critical discussions. We also thank P. Muhlrad for critical

822 reading of the manuscript.

823

36 824 References

825 1. Vogelstein B, Papadopoulos N, Velculescu VE, Zhou S, Diaz LA, Kinzler KW. Cancer 826 Genome Landscapes. Science. 2013;339: 1546–1558. doi:10.1126/science.1235122

827 2. Nik-Zainal S, Van Loo P, Wedge DC, Alexandrov LB, Greenman CD, Lau KW, et al. The 828 life history of 21 breast cancers. Cell. 2012;149: 994–1007. doi:10.1016/j.cell.2012.04.023

829 3. Driessens G, Beck B, Caauwe A, Simons BD, Blanpain C. Defining the mode of tumour 830 growth by clonal analysis. Nature. 2012;488: 527–530. doi:10.1038/nature11344

831 4. Schepers AG, Snippert HJ, Stange DE, van den Born M, van Es JH, van de Wetering M, et 832 al. Lineage tracing reveals Lgr5+ stem cell activity in mouse intestinal adenomas. Science. 833 2012;337: 730–735. doi:10.1126/science.1224676

834 5. Beck B, Blanpain C. Unravelling cancer stem cell potential. Nat Rev Cancer. 2013;13: 727– 835 738. doi:10.1038/nrc3597

836 6. Brown K, Quintanilla M, Ramsden M, Kerr IB, Young S, Balmain A. v-ras genes from 837 Harvey and BALB murine sarcoma viruses can act as initiators of two-stage mouse skin 838 carcinogenesis. Cell. 1986;46: 447–456.

839 7. Agrawal N, Frederick MJ, Pickering CR, Bettegowda C, Chang K, Li RJ, et al. Exome 840 Sequencing of Head and Neck Squamous Cell Carcinoma Reveals Inactivating Mutations 841 in NOTCH1. Science. 2011;333: 1154–1157. doi:10.1126/science.1206923

842 8. Bos JL. ras oncogenes in human cancer: a review. Cancer Res. 1989;49: 4682–4689.

843 9. Stransky N, Egloff AM, Tward AD, Kostic AD, Cibulskis K, Sivachenko A, et al. The 844 Mutational Landscape of Head and Neck Squamous Cell Carcinoma. Science. 2011;333: 845 1157–1160. doi:10.1126/science.1208130

846 10. Bartel DP. MicroRNAs: target recognition and regulatory functions. Cell. 2009;136: 215– 847 33. doi:S0092-8674(09)00008-7 [pii] 10.1016/j.cell.2009.01.002

848 11. Mukherji S, Ebert MS, Zheng GXY, Tsang JS, Sharp PA, Oudenaarden A van. MicroRNAs 849 can generate thresholds in target gene expression. Nat Genet. 2011;43: 854–859. 850 doi:10.1038/ng.905

851 12. Johnson SM, Grosshans H, Shingara J, Byrom M, Jarvis R, Cheng A, et al. RAS is 852 regulated by the let-7 microRNA family. Cell. 2005;120: 635–647. 853 doi:10.1016/j.cell.2005.01.014

854 13. Kumar MS, Lu J, Mercer KL, Golub TR, Jacks T. Impaired microRNA processing 855 enhances cellular transformation and tumorigenesis. Nat Genet. 2007;39: 673–677. 856 doi:10.1038/ng2003

37 857 14. Iorio MV, Croce CM. MicroRNA dysregulation in cancer: diagnostics, monitoring and 858 therapeutics. A comprehensive review. EMBO Mol Med. 2012;4: 143–159. 859 doi:10.1002/emmm.201100209

860 15. Kasinski AL, Slack FJ. Potential microRNA therapies targeting Ras, NFkappaB and p53 861 signaling. Curr Opin Mol Ther. 2010;12: 147–157.

862 16. Mendell JT, Olson EN. MicroRNAs in stress signaling and human disease. Cell. 2012;148: 863 1172–1187. doi:10.1016/j.cell.2012.02.005

864 17. Jackson SJ, Zhang Z, Feng D, Flagg M, O’Loughlin E, Wang D, et al. Rapid and 865 widespread suppression of self-renewal by microRNA-203 during epidermal differentiation. 866 Development. 2013;140: 1882–91. doi:140/9/1882 [pii] 10.1242/dev.089649

867 18. Yi R, Poy MN, Stoffel M, Fuchs E. A skin microRNA promotes differentiation by 868 repressing “stemness.” Nature. 2008;452: 225–9. doi:nature06642 [pii] 869 10.1038/nature06642

870 19. Wang L, Dowell RD, Yi R. Genome-wide maps of polyadenylation reveal dynamic mRNA 871 3’-end formation in mammalian cell lineages. RNA. 2013;19: 413–25. doi:rna.035360.112 872 [pii] 10.1261/rna.035360.112

873 20. Hanahan D, Weinberg RA. Hallmarks of Cancer: The Next Generation. Cell. 2011;144: 874 646–674. doi:10.1016/j.cell.2011.02.013

875 21. Schober M, Fuchs E. Tumor-initiating stem cells of squamous cell carcinomas and their 876 control by TGF-β and integrin/focal adhesion kinase (FAK) signaling. Proc Natl Acad Sci 877 U S A. 2011;108: 10544–10549. doi:10.1073/pnas.1107807108

878 22. Durchdewald M, Guinea-Viniegra J, Haag D, Riehl A, Lichter P, Hahn M, et al. Podoplanin 879 is a novel fos target gene in skin carcinogenesis. Cancer Res. 2008;68: 6877–6883. 880 doi:10.1158/0008-5472.CAN-08-0299

881 23. Bild AH, Yao G, Chang JT, Wang Q, Potti A, Chasse D, et al. Oncogenic pathway 882 signatures in human cancers as a guide to targeted therapies. Nature. 2006;439: 353–357. 883 doi:10.1038/nature04296

884 24. Zhang Z, Lee JE, Riemonday K, Anderson EM, Yi R. High-efficiency RNA cloning 885 enables accurate quantification of miRNA expression by deep sequencing. Genome Biol. 886 2013;14: R109.

887 25. Talotta F, Cimmino A, Matarazzo MR, Casalino L, De Vita G, D’Esposito M, et al. An 888 autoregulatory loop mediated by miR-21 and PDCD4 controls the AP-1 activity in RAS 889 transformation. Oncogene. 2009;28: 73–84. doi:10.1038/onc.2008.370

890 26. Darido C, Georgy SR, Wilanowski T, Dworkin S, Auden A, Zhao Q, et al. Targeting of the 891 tumor suppressor GRHL3 by a miR-21-dependent proto-oncogenic network results in

38 892 PTEN loss and tumorigenesis. Cancer Cell. 2011;20: 635–648. 893 doi:10.1016/j.ccr.2011.10.014

894 27. Abel EL, Angel JM, Kiguchi K, DiGiovanni J. Multi-stage chemical carcinogenesis in 895 mouse skin: Fundamentals and applications. Nat Protoc. 2009;4: 1350–1362. 896 doi:10.1038/nprot.2009.120

897 28. White RA, Neiman JM, Reddi A, Han G, Birlea S, Mitra D, et al. Epithelial stem cell 898 mutations that promote squamous cell carcinoma metastasis. J Clin Invest. 2013;123: 4390– 899 4404. doi:10.1172/JCI65856

900 29. Benaich N, Woodhouse S, Goldie SJ, Mishra A, Quist SR, Watt FM. Rewiring of an 901 epithelial differentiation factor, miR-203, to inhibit human squamous cell carcinoma 902 metastasis. Cell Rep. 2014;9: 104–117. doi:10.1016/j.celrep.2014.08.062

903 30. Eichhorn SW, Guo H, McGeary SE, Rodriguez-Mias RA, Shin C, Baek D, et al. mRNA 904 destabilization is the dominant effect of mammalian microRNAs by the time substantial 905 repression ensues. Mol Cell. 2014;56: 104–115. doi:10.1016/j.molcel.2014.08.028

906 31. Guo H, Ingolia NT, Weissman JS, Bartel DP. Mammalian microRNAs predominantly act to 907 decrease target mRNA levels. Nature. 2010;466: 835–40. doi:nature09267 [pii] 908 10.1038/nature09267

909 32. Chi SW, Zang JB, Mele A, Darnell RB. Argonaute HITS-CLIP decodes microRNA-mRNA 910 interaction maps. Nature. 2009;460: 479–86. doi:nature08170 [pii] 10.1038/nature08170

911 33. Moore MJ, Zhang C, Gantman EC, Mele A, Darnell JC, Darnell RB. Mapping Argonaute 912 and conventional RNA-binding protein interactions with RNA at single-nucleotide 913 resolution using HITS-CLIP and CIMS analysis. Nat Protoc. 2014;9: 263–293. 914 doi:10.1038/nprot.2014.012

915 34. Lehman IR, Kaguni LS. DNA polymerase alpha. J Biol Chem. 1989;264: 4265–4268.

916 35. Stoll SW, Rittié L, Johnson JL, Elder JT. Heparin-binding EGF-like growth factor promotes 917 epithelial-mesenchymal transition in human keratinocytes. J Invest Dermatol. 2012;132: 918 2148–2157. doi:10.1038/jid.2012.78

919 36. Miyamoto S, Hirata M, Yamazaki A, Kageyama T, Hasuwa H, Mizushima H, et al. 920 Heparin-binding EGF-like growth factor is a promising target for ovarian cancer therapy. 921 Cancer Res. 2004;64: 5720–5727. doi:10.1158/0008-5472.CAN-04-0811

922 37. Downward J. Targeting RAS signalling pathways in cancer therapy. Nat Rev Cancer. 923 2003;3: 11–22. doi:10.1038/nrc969

924 38. Lena AM, Shalom-Feuerstein R, di Val Cervo PR, Aberdam D, Knight RA, Melino G, et al. 925 miR-203 represses “stemness” by repressing DeltaNp63. Cell Death Differ. 2008;15: 1187– 926 95. doi:cdd200869 [pii] 10.1038/cdd.2008.69

39 927 39. Melar-New M, Laimins LA. Human papillomaviruses modulate expression of microRNA 928 203 upon epithelial differentiation to control levels of p63 proteins. J Virol. 2010;84: 5212– 929 5221. doi:10.1128/JVI.00078-10

930 40. Lichti U, Anders J, Yuspa SH. Isolation and short-term culture of primary keratinocytes, 931 hair follicle populations and dermal cells from newborn mice and keratinocytes from adult 932 mice for in vitro analysis and for grafting to immunodeficient mice. Nat Protoc. 2008;3: 933 799–810. doi:10.1038/nprot.2008.50

934

935 Figure Legends

936 Figure 1. Genome-wide profiling of the oncogenic HrasG12V transformed miRNA and

937 mRNA transcriptome in primary keratinocytes. (A) Schematic of experimental approach to

938 identify deregulated mRNA and miRNA networks driven by oncogenic HrasG12V using small-

939 RNA Seq and 3Seq. The 3seq library preparation allows quantitative definition of poly-A+ RNA

940 3’ends and expression levels. (B) 3Seq reproducibly detects mRNA expression levels over 4

941 orders of magnitude. Pearson correlation coefficient displayed (C) Unsupervised hierarchical

942 clustering of log-transformed mean-centered mRNA expression levels for all transcripts

943 deregulated 2-fold by oncogenic HrasG12V (n=2 libraries per condition) (D) Gene Ontology

944 analysis of transcripts up and downregulated by HrasG12V (2-fold change FDR <0.05) indicates

945 enrichment for migratory and angiogenic processes, and suppression of keratinocyte

946 differentiation. (E) GSEA analysis of selected gene-sets relevant to skin carcinogenesis. (F)

947 Unsupervised hierarchical clustering of log-transformed mean-centered miRNA expression

948 levels for all transcripts deregulated 2-fold by oncogenic HrasG12V (n=2 libraries per condition)

949 (G-H) Abundant miRNAs such as miR-203, miR-205, and miR-21 are strongly deregulated by

950 oncogenic Ras.

951

40 952 Figure 2. miR-203 is strongly suppressed in mouse and human SCCs. (A-B) qPCR and

953 small-RNA-Seq independently validate downregulation of miR-203 and upregulation of miR-21

954 driven by oncogenic HrasG12V (n=3 biological rep. qPCR, n=2 small-RNA-Seq, mean +/- SEM

955 displayed, * = p-value <0.05, Students T-Test two-sided). (C) Gene track and quantification the

956 3’end of the miR-203 primary transcript based on 3Seq. (D) miR-203 primary transcript detection

957 by qPCR (n=3 biological replicates, Mean +/- SEM displayed, * = p-value < 0.05, ns = non-

958 significant, Students T-Test two-sided). (E) miR-203 is downregulated in DMBA/TPA produced

959 papillomas compared to normal adjacent tissue. Epi=epidermis, Der=dermis, T=tumor and

960 S=stroma. The black lines denote the epidermal/dermal and tumor/stroma boundary (F) miR-203

961 is downregulated in malignant SCCs derived from KrasG12D/Smad4cKO and passaged in

962 immunocompromised mice. (G-O) Reduced miR-203 expression is correlated with increasing

963 malignancy in human skin SCC cancers. Panels G,J,M were taken from regions with more

964 histologically normal regions to demonstrate successful miR-203 hybridization. (P) miRNA-Seq

965 quantification from patient matched normal and tumor tissue obtained from the TCGA

966 consortium data (bar indicates mean value, Students T-Test two-sided). Scale bar = 50 μm.

967

968 Figure 3. Loss of miR-203 modestly impairs embryonic epidermal development. (A)

969 Schematic of miR-203 conditional allele generation and knockout strategy. (B) Validation of

970 miR-203 ablation by qPCR from isolated epidermal samples (n=3 biological replicates, * = p-

971 value <0.05, ns = non-significant, Student T-Test two-sided). (C) Validation of miR-203 ablation

972 within the epidermis by in situ hybridization (Scale bar = 50 μm). (D) miR-203 knockout mice

973 are visibly indistinguishable from wild type counterparts. (E-G) miR-203 ablation results in mild

974 epidermal hyperplasia during embryonic development. (n=3 E16, n=4 E17, and n=3 p4 animals,

41 975 p-value provided in figure, Student T-test one-sided). (H) Representative Hematotoxylin and

976 Eosin image from p4.5 animals, demonstrating restored normal skin morphology in neonatal

977 animals. (Scale bars = 50 μm for inset and 100 μm for main images) (I) Epidermal differentiation

978 is not compromised by loss of miR-203. (Scale bars = 50 μm) (J) miR-203-/- Primary

979 keratinocytes are more clonogenic than wild type counterparts (representative results from 3

980 experiments, * = p-value <0.05, student t-test two-sided). (K) Conditional ablation of miR-203

981 from passaged miR-203fl/fl keratinocytes results in higher clonogenicity (representative results

982 from n=3 independent experiments, mean +/- standard deviation displayed, * = p-value <0.05,

983 student T-test, two-sided).

984

985 Figure 4. Loss of miR-203 sensitizes mice to DMBA/TPA skin carcinogenesis. (A)

986 Representative images of tumors that were formed in the skin of WT and miR-203 null mice

987 treated with DMBA/TPA. (B) miR-203-/- mice have a larger tumor burden than miR-203+/+

988 counterparts (n=6 and 7 miR-203+/+ and miR-203-/- animals respectively, mean +/- SEM

989 displayed, £ = p-value <0.05, Whitney-Mann U-test one-sided) (C) miR-203-/- tumor size

990 distribution is similar to wild type animals (ns = non-significant, student T-test two sided,

991 median displayed as bar). (D-E) miR-203+/+ and miR-203-/- papillomas display similar

992 morphologies and histology. (F) Proliferation and differentiation dynamics are similar between

993 miR-203+/+ and miR-203-/- tumors. (Scale bars = 50 μm).

994

995 Figure 5. miR-203 antagonizes HrasG12V driven keratinocyte proliferation. (A) HrasG12V

996 transduced miR-203-/- primary cultures are more colonogenic upon serial passage than wild-type

997 controls (representative of n=2 independent experiments, mean +/- standard deviation displayed.

42 998 * = p-value <0.05, , ns = non-significant, Student T-Test two-sided) (B) qPCR of miR-203

999 induction upon addition of doxycycline in vector and HrasG12V transduced cells. (mean +/- SEM

1000 displayed, n=3 biological replicates) (C) Restoration of miR-203 using a doxycycline inducible

1001 transgene, results in suppression of colony formation ability in HrasG12V transduced and control

1002 keratinocytes. miR-203 was induced with doxycycline (5 μg/ml) 24 hours after plating

1003 (representative of n=3 independent experiments, mean +/- standard deviation displayed, * = p-

1004 value = <0.05, , ns = non-significant). (D) miR-203 restoration suppresses HrasG12V driven S-

1005 Phase entry. miR-203 was induced for 24 hours prior to harvesting for flow cytometry. (n=3,

1006 mean +/- standard deviation displayed, * = p-value = <0.05).

1007

1008 Figure 6. Comprehensive identification of miR-203 targets using genome-wide expression

1009 analyses and Ago2 HITS-CLIP. (A) Schematic of genome-wide expression profiling datasets

1010 used in meta-analysis to identify bono-fide miR-203 targets. (B) Genes upregulated in all three

1011 miR-203 loss-of-function datasets (786 genes, no fold-change or p-value cut-off) were compared

1012 to genes downregulated in both miR-203 gain-of-function datasets (1704 genes, no fold-change

1013 or p-value cut-off) to identify a subset of genes with a strong inverse correlation to miR-203

1014 expression (294 genes), of which 100 genes contained miR-203 7mer or 8mer seed sequence

1015 matches in their 3’UTRs. (C) Table demonstrating top 20 genes identified in meta-analysis

1016 ranked by negative-correlation to miR-203 expression. Genes colored in red contain 3’ UTR

1017 miR-203 7 or 8mer seed matches. (D) De novo motif searching identified an 8mer miR-203 seed

1018 motif, complementary to the miR-203 seed sequence enriched in the 3’UTR of candidate miR-

1019 203 target genes identified in the meta-analysis (294 genes). Table demonstrating enrichment for

1020 7 or 8mer seed matches in the 3’UTR of candidate miR-203 target genes (294) over the

43 1021 background seed distribution in primary keratinocytes, which is not seen for randomly selected

1022 294 genes expressed in primary keratinocytes or a negative control gene set of genes upregulated

1023 in miR-203 gain-of-function and downregulated miR-203 loss-of-function (353 genes). (E)

1024 Schematic of Ago2 HITS-CLIP and the identified miR-203 seed motif. (F) Diagram of genes

1025 detected by expression meta-analysis and Ago2-HITS-CLIP. Table of 21 high confidence miR-

1026 203 targets identified through expression meta-analysis and that have Ago2-HITS-CLIP 3’UTR

1027 peaks with miR-203 seed matches.

1028

1029 Figure 7. Hbegf and Pola1 are direct miR-203 target genes critical for keratinocyte

1030 proliferation. (A) 3’UTR luciferase reporter assays demonstrate that miR-203 directly targets

1031 Trp63 (positive control), Pola1, and Hbegf in keratinocytes (representative of n=3 independent

1032 experiments, mean +/- propagated standard deviation displayed, * = p-value < 0.05, , ns = non-

1033 significant, student T-Test two-sided) (B) Hbegf and Pola1 are upregulated in miR-203-/- isolated

1034 epidermis (p4) (n=8 and n=10, miR-203+/+ and miR-203-/- animals respectively, mean +/- SEM

1035 displayed, * = p-value < 0.05). (C-D) Western blots from lysates with miR-203 overexpression

1036 (48 hours) or miR-203 ablation. (E) shRNA knockdown of Hbegf or Pola1 impairs keratinocyte

1037 colony formation ability (representative of n=3 independent experiments, * = p-value <0.05,

1038 mean +/- standard deviation displayed). (F) Model for the mechanism of miR-203 in restricting

1039 Hras-initiated tumorigenesis.

1040

1041 Supplementary Figure Legends

1042 Figure 3-figure supplement 1. Generation of a miR-203 conditional knockout mouse. (A)

1043 RNA-Seq and 3Seq detection of the miR-203 primary transcript (B) Schematic of the miR-203

44 1044 conditional allele. H3 = HindIII (C) Southern Blot confirmation of founder miR-203 conditional

1045 knockout mice (D) Small-RNA-seq confirmation of miR-203 loss. Blue lines indicate two-fold

1046 change (E) qPCR quantification of miR-203 and miR-205, demonstrating that the miR-203 floxed

1047 allele does not alter microRNA levels. Error bars represent S.E.M from reactions performed from

1048 n=2 mice in technical triplicate, ns = non-significant.

1049

1050 Figure 3-figure supplement 2. miR-203 expression in diverse mouse tissues.

1051 qPCR detection of mature miR-203 in various adult mouse organ tissues. Error bars represent

1052 S.E.M from reactions performed in technical duplicate.

1053

1054 Figure 4-figure supplement 1. The HrasQ61L mutation is common in both miR-203+/+ and miR-

1055 203-/- tumors. (A) Representative end-point PCR followed by XbaI digestion. The HrasQ61L allele

1056 produces 120bp and 87bp product, whereas wild-type produces a 207bp product (B) Table with

1057 quantification of genotyping results. non-significant p > 0.05, Chi-Squared Test.

1058

1059 Figure 6-figure supplement 1. Transcripts containing 3’UTR miR-203 seed matches are

1060 regulated by miR-203. (A) Genes containing miR-203 seed matches are more likely to be

1061 downregulated upon miR-203 overexpression, or (B) upregulated upon miR-203 ablation (Panel

1062 E) (p-value <0.05 for comparison of 8mer match to no-match, K-S Test, one-sided). (C)

1063 Transcripts are ranked based upon aggregate fold-changes consistent with miR-203 regulation to

1064 produce a ranked expression correlation metric. Transcripts with 7 or 8mer miR-203 seed

1065 matches are more likely to be regulated by miR-203 modulation. (p-value <0.05) (See methods)

1066

45 1067 Figure 6-figure supplement 2. Ago2-HITS-CLIP in primary keratinocytes.(A) Example

1068 autoradiogram of isolated Ago2-RNA complexes, red box indicates region excised for

1069 sequencing (B) Proportion of miRNAs detected by Ago2-HITS-CLIP (C) Number of 3’UTR

1070 Ago2 peak containing seed sequences from miRNA families accounting for 90% of all miRNAs

1071 in p4 epidermis. (D) Genome-wide distribution of Ago2 peaks and reads (E) Comparison of

1072 miRNAs detected by Ago2-HITS-CLIP and small-RNA-Seq.

1073

1074 Figure 6-figure supplement 3. Ago2 HITS-CLIP 3’UTR peaks are enriched in keratinocyte

1075 miRNA seed matches, including miR-203. (A) Position of miRNA seed matches for miRNAs

1076 highly expressed in total epidermal samples. The peak summit represents nucleotide position 0

1077 (C) De novo motif searching identifies most enriched 8mer motifs in 3’UTR peaks.

1078

1079 Figure 6-figure supplement 4. Predicted miR-203 targets based on HITS-CLIP are regulated

1080 by miR-203. (A) Cumulative distributions of miR-203 targets or not targeted transcripts based on

1081 HITS-CLIP in miR-203 overexpression datasets. (B) Cumulative distributions of miR-203 targets

1082 or not targeted transcripts based on HITS-CLIP in miR-203 knockout datasets (C) Ranked

1083 analysis of miR-203 targets detected by HITS-CLIP, based on 6,7,8mer seed only, or through

1084 meta-analysis (p value displayed on plots).

1085

1086 Figure 6-figure supplement 5. miR-203 targets do not display translation efficiency changes

1087 upon miR-203 ablation.(A) Comparison of translation efficiency for miR-203 targets (red) and

1088 non-targeted transcripts (blue) identified by expression meta-analysis. (B) Quantification of the

1089 change in translation efficiency in miR-203 KO samples for miR-203 targets identified by

46 1090 expression meta-analysis. (C) Comparison of translation efficiency for miR-203 targets identified

1091 by Ago2-HITS-CLIP (red) and non-targeted transcripts (blue). (D) Quantification of the change

1092 in translation efficiency in miR-203 KO samples for miR-203 targets based on Ago2-HITS-CLIP.

1093

1094 Figure 7-figure supplement 1. A subset of miR-203 targets are upregulated by HrasG12V. (A)

1095 miR-203 target genes identified through meta-analysis containing miR-203 seed matches, are

1096 more likely to be upregulated upon HrasG12V expression in primary keratinocytes than non-

1097 targeted transcripts (p-value = <0.05, K-S Test one-sided). (B) Top 20 miR-203 targets, based on

1098 expression meta-analysis, ranked by upregulation by HrasG12V, genes shown in red are also

1099 identified by Ago2-HITS-CLIP.

1100

1101 Figure 7-figure supplement 2. Hbegf and Pola1 are required for keratinocyte growth

1102 potential in HrasG12V transformed miR-203+/+ and miR-203-/- cultures. (A) Colony formation

1103 assays of established miR-203+/+ cultures stablely infected with pbabe-HrasG12V-neo, followed by

1104 Pola1 and Hbegf knockdown. (B) Colony formation assays of established miR-203+/+ cultures

1105 stablely infected with pbabe-HrasG12V-neo, followed by Pola1 and Hbegf knockdown. (p-value

1106 determined by ANOVA with Tukey HSD correction, n=3 wells per experiment).

1107

1108 Supplementary Files

1109

1110 Supplementary File 1. Sequencing Mapping Statistics.

1111 Supplementary File 2. Primers, Antibodies, and shRNAs used in this study.

1112

47 1113 Source Data Files

1114

1115 Figure 1-source data. Log2 Fold Changes for transcripts up or down regulated 2-fold with

1116 FDR <0.05 in HrasG12V transformed keratinocytes.

1117

1118 Figure 6-source data 1. GO-analysis of selected miR-203 datasets.

1119

1120 Figure 6-source data 2. Putative miR-203 targets detected in the Expression Meta-analysis.

48 Figure 1

A B Isolate primary 3seq 15 r =.989 ●● ●

mouse keratinocytes ● ● ● ● ●● ●● ●● ●● ● ● ● ●● ● ● ●●●● ● ●● ●●●● PolyA select RNA and Fragment ●●● ● ●● ●● ● ●● ● ●●●● ●●●●●● ●●●●●●● ● ●● ● ●●●●●●● ●●●●●● ● ●● ●●● ●●●●●●●●●● ● ●● ●● ● ●●● ● ●●●● ●●●●● ●●●● ● ●●●●●●● ●●●●●● ● ● ●●●● ● ●● ●●●● ●●● ●● G12V 10 ● ●●● ● ●●●●●●●●● ●●● ●●●●●●●● ● ●●●●●●●● or control ●●● ●●●●●●● Infect with Hras ● ● ●●●●●● OligodT primed cDNA Synthesis ●●●●●●●●●● ● ●●●● ●●●●● ● ●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●● ● ● ●●●●●●●●● ●● ● ●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●● virus ● ● ● ● ●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ●●●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● and select for resistance ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●● Second Strand Synthesis and End Repair ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● to puromycin ●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● 5 ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●●●●●● ●● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●● ●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● Size Select and PCR Enrich ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●● ● ● ● ●●●●● ●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ● ● ● ● ● ●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●● ● ● ● ● ● ● ● ●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●● ● ● ●●● ● ● ●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●● ● ● Harvest for ● ● ● ● ● ●●●●●●●●● ●●●●●●●●● ●●●●●●●●●●●●● ● ● ● ● ● ●●●● ● ●●●●●●●●●●●●●●●● ●●●●●●●●●●● ● ● ● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●● ● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ● ● ● ● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●● ● ● ●●●● ● ● ●●●●●●●●●●●●●●●●● ●●●●●●●●●● ●●● ● ● ● ● ● ● ●●●● ●●●●●●●●●●●● ● ● ●●● ● ● ● ● ● ● ●● ●●●● ●● ●●●● ●●● ● ● ● ● ●●●●●●●●●● ●●●●●●● ● ● ● ● ● ● ● ●●●●● ● ● ●● ●●●●● ●●●●●● ● ● small-RNA Seq 0 ● ● ● ● ●● ●●●● ●●●●● ●● ●●

Normalized Vector rep.2 3seq RPM rep.2 Normalized Vector ● ● ● ● ● ●●●●●●● ●● ●● ●● ● ● 3seq Tags ● ● ●●●●●●●●●●●● ●●● ● ●

2 ● ● ● ●●● ● ● ●●● ●●●● ● ● ● ● ● ● ●●●● ● ● ● ● ● and PolyA-Selected 3-Seq ● ● ● ● ●●●●●●●●● ●●● Dicer1 Log ● ● ● ● ● ● ● ●●● ● ● ●● ● ● ● ●● ● ● 1kb 0 5 10 15 Log2 Normalized Vector rep.1 3seq RPM

C 3seq D Upregulated Genes (687) Downregulated Genes (494) Vector HrasG12V biological adhesion epidermis development cell adhesion epidermal cell differentiation vasculature development ectoderm development cell migration keratinocyte differentiation blood vessel morphogenesis epithelial cell differentiation response to wounding epithelium development blood vessel development cell motion keratinization monosaccharide metabolic extracellular structure organization process 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 9 -log (corrected p-value) -log10(corrected p-value) 10

E SCC Cancer Stem Cell c-Fos Targets in Ras Model 1818 genes Hras Pathway Signature of Skin Carcinogensis

NES = 2.03 NES = 2.16 -3 -3 NES = 1.59 FDR < 1 x 10 FDR < 1 x 10 FDR < 1 x 10-3

F miRNA-seq G H miRNA Fold Change FDR Vector HrasG12V mmu-miR-99a-5p -9.83 3.60 x 10-06 20 mmu-miR-335-3p -7.86 3.00 x 10-06 -04 miR-21 ● mmu-miR-203-5p -5.87 2.95 x 10 mmu-miR-203-3p -5.22 3.71 x 10-03 RPM -03 ● miR-203 mmu-miR-335-5p -4.70 1.16 x 10 ● V12G ●● ● 15 ● miR-205 -02 ●● miR-29 ● ● mmu-miR-24-3p -4.20 1.70 x 10 ● ● ● -02 ● ●● ● ●●● ● ● ●● miR-24 mmu-miR-23b-3p -3.62 1.37 x 10 ● ● ● ● ● ● ● ● ● ● ● ● ● miR-27 -03 ● ● -2.89 8.31 x 10 ●● ●● mmu-miR-24-2-5p ● ●● ● ● ● ● ●● ●miR-23 miR-146 ● ● ● ●●● ● ● -02 ● ●● ●●● ● ●●● -2.83 1.70 x 10 ●● mmu-miR-205-3p ● ●● ● ●●● ● 10 ● ●●●● ● ● ● ● -02 50 miRNAs ●● ● ● ●● ● ● ●●● ● 2.68 1.70 x 10 ● ● ● ●● mmu-miR-7a-5p ● ● ● ● ● ●● ● ● ● -02 Normalized Hras ● ● ●● ● ● ●●● ●● ● ●

2 ● ● 3.31 1.55 x 10 ● ●● ● mmu-miR-322-5p ● ●●● ●●● ● ● ●●● ● ●●● ● ●● ● ●● ● ●●●●●●● ● -02 ● ●●●● ● ● ●● ● ● ● ● 3.72 3.73 x 10 ●● ●● mmu-miR-21a-5p Log ●● ● -04 5 mmu-miR-29b-3p 3.98 8.40 x 10 -04 5101520mmu-miR-29a-3p 5.68 6.12 x 10 -06 Log2 Normalized Vector miRNA RPM mmu-miR-146a-5p 13.71 3.00 x 10 Figure 2 A B E

* 5.0 Normal Skin DMBA/TPA Papilloma

Vector Cq) 4.5 HrasG12V Mature miRNA Fold Change 4.0 Epi ǻǻ small- 3.5 qPCR 3.0 RNA Seq 2.5 n.s. miR-203 -5.22 -8.46 Der 2.0 * miR-205 -2.83 -3.06 1.5 T 1.0 miR-21 3.72 4.04 0.5 0.0 miR-203 in situ S Rel. miRNA Levels 1.4 * Vector

miR-203 miR-205 miR-21 Cq) HrasG12V F D 1.2 SCC Passage 1 SCC Passage 2

C ǻǻ G12D 1.0 Kras Epi Smad4cKO miR-203 primary transcript (3Seq) 0.8 RPM Der 18 T 0.6 5’ 3’ Vector 0 18 0.4 Vector 0 18 0.2 T G12V Hras 0 18 0.0 G12V Hras pri-miRNA Levels 0 pri-miR-203 miR-203 100bp Primary miRNA Fold Change miR-203 in situ 3’Seq qPCR miR-203 -2.53 -2.72 Human SCC Human SCC Human SCC Well Moderately Poorly Differentiated (4/4) Differentiated (2/2) Differentiated (3/3) G J M P TCGA miRNA-Seq Head and Neck SCC

miR-203 miR-21

600,000 Sívalue= 600,000 Sívalue= 0.0288 Hí H K N 500,000 500,000

400,000 400,000

300,000 300,000

I L O 200,000 200,000 miRNA expression (RPM) miRNA expression miR-203 in situ hybridization 100,000 100,000

0 n=43 0 n=43 Normal Tumor Normal Tumor Figure 3 D A B Chr. 12 H I J Adult P4 miR-203 WT KO miR-203-/- miR-203+/+ P4.5 WT KO KO WT LoxP EIIa-Cre fl/fl X E P4 E17 E16 Frt LoxP

Number of Colonies miR-203 10 15 20 25 30 35 40 45 50 0 5

miR-203 +/+ +/+

* BrdU BrdU EdU Rel. miRNA Levels ǻǻCq) 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 miR-203 / / / Krt5 Krt5 Krt5 miR-203 miR-203 * -/- K

P4.5 KO WT miR-203 miR-203 miR-203 miR-205 Ad-CRE Ad-GFP ns Krt1/ -/-

miR-203 -/- +/- +/+ Krt5

F

BrdU or EdU+ cells / mm P8 C 10 20 30 40 50 60 70 80 WT 0 fl/fl KO WT

E16 E17 P4 E16 E17 p=.07 miR-203 miR-203 Lor Number of Colonies p=.13 100 120 / 20 40 60 80 -/- +/+ Krt5 0

Ad-GFP Ad-CRE p=.17 miR-203 miR-203 G * KO WT

Epidermal Thickness (μm) 10 20 30 40 50 60 70 0 fl/fl E16 E17 P4 E16 E17 p=.03 Trp63 miR-203 miR-203

0.0 Rel.0.2 miR-2030.4 0.6 Levels0.8 ǻǻ1.0 Cq)1.2 Ad-GFP Ad-CRE / p=.01 Krt5 miR-203 -/- +/+

*

p=.07

KO fl/fl Figure 4 A B C DMBA TTPAPA biweekly ns 16 8 £ = p-value <0.05 +/+ miR-203 £ 14 -/- -2 0 1 2 3 4 5 6 7 8 9 10 miR-203 £ £ £ 6 12 £ £ +/+ miR-203 miR-203-/- 10 4 8 6 2 4 Diameter (mm) Tumor 2 0 n=25 n=70

Number of Tumors per Mouse Tumors Number of 0 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 miR-203+/+miR-203-/- Weeks of TPA treatment D E

WT H and E KO WT H and E KOKO

F WT WT WT

Ki67/Krt5 Krt1/DAPI Lor/DAPI KO KO KO

D B A Figure 5 miR-203 Hras miR-203 Hras Rel. mRNA Levels ǻǻCq) Passage:123 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 HrasG12V Vector G12V G12V -/- +/+ Vector

S-Phase: 41.4%+/-1.236.7%1.8 S-Phase: 52.4%+/-1.735.1%1.0 10 10 10 10 10 EdU 10 10 10 10 10 0 * 0 1 3 2 4 0 1 3 2 4 0 40K 20K -Dox Hras + Dox - Dox * G12V

60K DNA

0 +DOX -DOX pTre-miR-203 /Krt14-rtTA C Vector Hras 0 40K 20K +Dox

60K G12V

Number of Colonies 10 20 30 40 50 60 % EdU+ Cells 0 10.0 20.0 30.0 40.0 50.0 60.0 0.0 miR-203 miR-203

10 Number20 of30 Colonies40 50 60 passage 1 0 ns Total 0.5-1.5 -/- +/+ Vector

*

+ Dox - Dox * * Number of Colonies 10 20 30 40 50 60 70 0 mm miR-203 miR-203 passage 2 * Hras 2 ns ns

* -/- +/+

G12V + DoxVector - DoxVector + DoxHras - DoxHras 1.5-3 mm *

Number of Colonies 2 4 0 1 2 3 5 6

* miR-203 miR-203 passage 3 >3 mm * * G12V G12V -/- +/+ *

2

Figure 6 +Dox vs.-DOX E C AB pTre miR-203 miR-203 Krt14-rtTA/ Trp63 500bp vs. Information 7meG  miR-203 3’UTR seedmatches (113) Genes containing miR-203

0.0 content1.0 2.0 32 RNase, Ago2IP, UV-Crosslink, Lyse Ago2 bindingsites Align andIdentify Sequence cDNA Į P Radiolabel

Ago2 HITS-CLIP miR-203 candidate targets

U C A +/+ -/-

80S identified in meta-analysis (294 genes) C Keratinocytes S Rapgef1 Tax1bp3 Slc12a4 Sh3bgrl Ppap2c Symbol Ndufb2 Pgam5 Rap1b Ribo-Seq Vopp1 Dhx36 Lasp1 Ctps2 Prmt6 h3pxd2b Gbe1 Pola1 Yars2 Tprkb Cav1 Lnx2 Gene Primary G A

80S U n=2 biologicalreplicatesperdataset 80S Microarray Progenitors Epidermal 14 18 .605 0.18 0.58 0.37 0.36 0.38 -1.83 0.51 -1.42 -1.34 -0.74 06 09 .902 0.20 0.24 0.08 0.29 0.69 0.28 -0.91 0.46 -0.67 0.97 -1.26 0.10 -0.65 -1.07 -0.95 05 07 .703 0.23 0.38 0.27 -0.79 -0.57 08 16 .209 0.57 0.91 0.42 -1.67 -0.83 11 13 .803 0.22 0.33 0.60 0.28 0.87 -1.31 0.49 -1.13 -1.29 -0.29 09 04 .105 .21.00 0.12 0.59 0.31 -0.43 -0.99 03 07 .702 .12.00 0.31 0.22 0.25 0.47 0.75 -0.79 0.45 -0.39 -0.29 -0.42 04 05 .304 0.31 0.44 0.43 -0.55 -0.43 08 06 .904 .60.00 0.26 0.50 0.44 0.11 0.09 0.32 -0.69 0.48 -0.84 -0.60 -0.61 06 06 .803 .31.50 0.13 0.33 0.38 -0.66 -0.64 02 04 .207 .5-2.00 0.15 -1.50 0.74 0.22 -1.00 0.42 0.51 0.15 -0.42 -0.50 0.22 -0.24 0.60 0.37 -0.58 0.58 -0.24 0.25 -0.19 0.33 -0.54 -0.38 -0.42

U Microarray Epidermal Purified

U Progenitors (+DOX)

Ago

Ago Microarray Hair Follicle Keratinocytes C Primary Progenitors (+DOX) 3’Seq miR-203 AAAAA G A Microarray Isolated Hair Follicle Epidermis (KO) Progenitors miR-203 p-value=

1 x10 Purified 3’seq Primary Keratinocytes (KO)

Ribo-Seq Primary Microarray F -19 Epidermis Targets Detected by Expression

Meta-analysis Keratinocytes (KO) Total

Log

2

FC

273 Genes containing D upregulated miR-203 KO 3’UTR sites 21 Genes 7 or8mer 8mer 7mer 6mer miR-203 Type Seed Targets Detected Ago2-HITS-CLIP

Information in

0.0 content1.0 2.0 92 miR-203 Targets by (294) U G A 999-17 .E0 64.2E-01 66 7.0E-13 4.2E-01 79 7.4E-06 48 9.9E-01 52 49 492 C De NovoMotifP-Value Value A 100 194 P- 294

miR-203 candidate targets identified by U both HITS-CLIP and Expression Meta-analysis 1410 Targets U (353) Non- AI837181 Slc39a10 Col17a1 792-12 5.5E-01 28 9.2E-01 27 Zmym4 Atp5g3 Symbol Naa50 Ccnd1 Snx18 Tnpo1 Hbegf .E0 64.8E-01 16 1.0E+00 8 Sept2 Bnip2 Snai2 Pola1 Gbe1 Cav1 Sidt2 Gene Avl9 Apc Met Cbl U A UTR sites 7 or8mer downregulated in C Genes without Value overexpression

P- 00 00 .702 0.46 0.28 0.67 -0.08 -0.09 06 12 .606 0.08 0.69 0.46 -1.26 -0.65 07 08 .501 0.11 0.15 0.05 -0.84 -0.78 02 02 .400 0.01 0.08 0.14 -0.26 -0.28 00 05 .705 0.10 0.53 0.07 -0.52 -0.01 09 10 .009 0.28 0.97 0.10 -1.07 -0.95 04 05 .402 0.06 0.25 0.14 -0.55 -0.42 01 01 .201 0.22 0.19 0.22 -0.17 -0.18 05 03 .500 0.04 0.08 0.15 -0.38 -0.50 02 02 .401 0.07 0.14 0.14 -0.27 -0.29 02 03 .101 0.16 0.15 0.31 -0.39 -0.24 14 18 .605 0.18 0.58 0.36 -1.83 -1.42 09 05 .802 0.03 0.22 0.18 -0.53 -0.99 00 06 .203 0.32 0.36 0.12 -0.61 -0.01 00 01 .400 0.19 0.01 0.00 0.24 0.29 -0.11 0.23 -0.06 -0.73 -0.71 05 05 .800 0.00 0.04 0.08 -0.59 -0.51 04 04 .001 0.08 0.11 0.10 -0.43 -0.43 04 04 .402 0.12 0.23 0.14 -0.44 -0.40 04 06 .402 0.05 0.20 0.04 -0.65 -0.49 03 05 .401 0.10 0.19 0.24 -0.57 -0.30 -2.00 miR-203 G A

Microarray Epidermal Genes

-1.50 Log Progenitors (+DOX) -1.00

Random

Targets

2

-0.50 Microarray Hair Follicle (294)

Fold Change 0.00 Progenitors (+DOX)

1x10

0.50 Microarray Isolated 1.00

Epidermis (KO) Value 1.50 P-

3’seq Primary -13 2.00 Keratinocytes (KO) Ribo-Seq Primary Keratinocytes (KO) Figure 7 ns ns A Vector 1.4 1.2 * 1.4 1.13 ns miR-203 1.00 Vector Vector * miR-203 0.96 1.00 1.00 1.2 1.04 0.90 1.2 miR-203 1.00 0.94 1.0 * 0.96 0.91 * 1.0 0.72 1.0 0.8 0.8 0.8 0.60 0.6 * 0.6 0.39 0.6 Normalized Normalized

Normalized * 0.4 0.33 0.4 0.4 0.25 0.26 Luciferase Activity Luciferase Activity Luciferase Activity 0.2 0.2 0.2 0.10

0.0 0.0 0.0

Į Į

pgl3-control pgl3-Trp63 pgl3-Pola1 pgl3-control pgl3-control pgl3-Hbegf pgl3-mut.Trp63 pgl3-mut. Pola1 pgl3-mut.Hbegf p4 Total Epidermis B C Vector HrasG12V Vector HrasG12V 1.8 D miR-203+/+ DOX: - + - + miR-203: WT KO WT KO * miR-203-/- Cq) 1.6 1.43 * 1.35 ns Pola1 1.4 Pola1 ǻǻ 1.00 1 0.04 1.56 0.15 1 7.9 3.5 22 1.2 1.00 1.00 1.01 1.0 ȕ-Tubulin ȕ-Tubulin 0.8 0.6 Ccnd1 0.4 Ccnd1 1 0.93 1.75 1.47 1 7.2 4.1 12 0.2 Rel. mRNA Levels 0.0 ȕ-Tubulin ȕ-Tubulin Hbegf Pola1 Krt10

E Scrambled shRNA-1 shRNA-2 shRNA-3 * * 140 * 140 * * * 120 120 100 100 Pola1 80 80 60

60 40 40 Number of Colonies 20 Number of Colonies 20 Hbegf 0 0

ScrambledPola1 sh1Pola1 sh2Pola1 sh3 ScrambledHbegf sh1Hbegf sh2Hbegf sh3 F

Pola1 Hbegf Snai2 PapillomaP Ccnd1 ǻNp63

Hras mutation miR-203

miR-203