Integrative analysis of transcriptomics and clinical data uncovers the tumor-suppressive activity of MITF in prostate cancer

The dysregulation of gene expression is an enabling hallmark of cancer. Computational analysis of transcriptomics data from human cancer specimens, complemented with exhaustive clinical annotation, provides an opportunity to identify core regulators of the tumorigenic process. Here we exploit well-annotated clinical datasets of prostate cancer for the discovery of transcriptional regulators relevant to prostate cancer. Following this rationale, we identify Microphthalmia-associated transcription factor (MITF) as a prostate tumor suppressor among a subset of transcription factors. Importantly, we further interrogate transcriptomics and clinical data to refine MITF perturbation-based empirical assays and unveil Crystallin Alpha B (CRYAB) as an unprecedented direct target of the transcription factor that is, at least in part, responsible for its tumor-suppressive activity in prostate cancer. This evidence was supported by the enhanced prognostic potential of a signature based on the concomitant alteration of MITF and CRYAB in prostate cancer patients. In sum, our study provides proof-of-concept evidence of the potential of the bioinformatics screen of publicly available cancer patient databases as discovery platforms, and demonstrates that the MITF-CRYAB axis controls prostate cancer biology.


Introduction 40
Balanced integration of intracellular circuits operates within a normal cell to sustain physiological 41 homeostasis. Alterations in some, if not all, of these circuits converge in changes on gene expression, 42 which will eventually enable the acquisition and sustenance of the hallmarks of cancer cells (1). This 43 event emphasizes the importance of maintaining the transcriptional homeostasis in normal cells and 44 places gene expression deregulation at the core of cancer research interests. 45 In the last decades, transcriptomics data derived from cancer specimens has become an important 46 resource for the classification, stratification and molecular driver identification in tumors. We and others 47 (Supplementary Fig. 2 A-C). MITFA was the isoform predominantly expressed in the three scenarios 113 analyzed, and we pursued the studies further with this isoform. Next, we aimed to analyze the biological 114  In order to decipher the molecular mechanism driving the tumor suppressive role of MITFA we 138 performed gene expression profiling of both doxycycline treated and control PC3 TRIPZ-MITFA cells and identified 101 probes that showed statistically differential signal between both conditions 140 (Supplementary table 2). We first performed a gene enrichment analysis with those genes which 141 displayed upregulated expression (76 genes) upon MITFA over-expression (Figure 3 A and  142   Supplementary table 3), as the number of downregulated genes (25) was no sufficient to obtain any 143 gen enrichment. Next, we aimed at identifying potential MITFA effectors of relevance in human PCa. 144 To this end, we established a threshold of 1.5 fold change over MITFA non-induced cells, which resulted 145 in 8 probes (corresponding to 6 annotated genes) upregulated upon the induction of the transcription 146 factor (Supplementary table 2; yellow bold highlighted). We next performed correlation analysis 147 between MITF and each of the 6 differentially expressed genes obtained from the microarray (Figure 3  Taken together, these data presented CRYAB as a direct target of MITFA and the best candidate to 166 mediate its tumor suppressive activity in PCa. 167

CRYAB mediates the tumor suppressive activity of MITF in PCa 169
We next studied the functional relevance of CRYAB for the tumor suppressive activity of MITFA in PCa. 170 Towards this aim, we constitutively silenced the expression of CRYAB by RNAi using two independent 171 short hairpin RNA (sh#1 and sh#2) in PC3 TRIPZ-MITFA cells. After validation that RNAi was achieved 172  We next asked whether the functional association between MITF and CRYAB could be employed to 182 identify PCa patients with high disease aggressiveness. We thus ascertained the stratification potential 183 of the MITF-CRYAB axis in PCa by means of consistency and robustness. We download the mRNA 184 expression raw data together with the clinical data (recurrence or not recurrence) from Taylor (11), 185 Glinsky (8)  in the way that we combat the disease. We are now able to deconstruct a tumor at a molecular level 199 using genomics, transcriptomics, proteomics and metabolomics. This, in turn, enables us to foresee, 200 identify and demonstrate the potential of patient stratification. Specifically, the transcriptomics 201 characterization of tumors is an invaluable strategy to identify clinically relevant genes that play key 202 roles in the progression of cancer, especially for those types with poorer prognosis (14). Thus, the 203 comprehensive and integrative analysis of gene expression changes and clinical parameters in cancer 204 has become a mainstream in cancer research. Mining cancer-associated transcriptome datasets is an 205 emerging approach used by top cancer research groups, but better tools are needed to increase its 206 power and user-friendliness. In order to face this challenge, new interfaces to exploit OMICs data, such 207 as cBioportal (40, 41) are being designed to help scientists interrogate, integrate and visualize large 208 amount of information contained on multiple credible and qualified cancer datasets. 209 In the present study, we exploited publicly available and well-annotated (transcriptomics and clinical 210 data) prostate cancer databases together with experimental assays to describe a novel tumor 211 suppressive activity of the transcription factor MITF in PCa, which is executed, at least in part, through 212 the direct regulation of the CRYAB expression. 213 The functional implication of MITF in cancer has been best defined in melanoma, in which the 214 expression of the transcription factor is heterogeneous. Although some controversy exists regarding its 215 oncogenic role in melanoma, MITF has been defined as a "lineage survival oncogene" with no data 216 pointing out at a tumor suppressive function (19,21,39,(42)(43)(44)(45). Even though the expression of MITF 217 has been detected in other cancer types (23,24,46), no data supporting a functional role of MITF 218 deregulation has been reported yet in a cancer scenario different from melanoma. 219 Here we show that MITF is downregulated in PCa when compared with normal specimens, in contrast 220 to the elevated expression reported in hepatocellular carcinoma (HCC) and chronic myeloid leukemia 221 (CML) (23, 46). Moreover, the novelty of our study relies on the observation and definition of the tumor 222 suppressive activity of MITF in PCa. In this context, MITFA upregulation was associated with a reduction 223 in cell proliferation and DNA replication. As occurs in melanoma, the modulation of MITF expression in 224 PCa cells induces the expression of the cell cycle inhibitor p21 but no changes in the cell cycle inhibitor 225 p16 were observed (data not shown). Thus, our results in PCa are in line with the canonical function of 226 MITF in cell cycle progression and proliferation in melanoma (39,44,45). 227 It's important to highlight that the tissue specific differences in MITF expression among different cancer 228 types suggest that in order to fully comprehend MITF's role in cancer, its expression and function has 229 to be analyzed in context of each particular cell and tissue type. 230 CRYAB is a member of the small heat shock protein family that functions as stress-induced molecular 231 chaperone. It inhibits the aggregation of denatured proteins, promotes cell survival and inhibits 232 apoptosis in the context of cancer (47). Paradoxically, CRYAB is highly expressed in some cancer types 233 but decreased in others and in both scenarios an association with cancer progression and prognosis 234 has been reported (25,26,(28)(29)(30)(31)(32)(48)(49)(50)(51)(52). In spite of the amount of information regarding the changes 235 in CRYAB expression in cancer, the transcriptional regulation of this chaperone has been poorly 236 explored (48). In this study, we described a novel direct transcriptional regulation of CRYAB by MITF. 237 Although there is no direct nor mechanistic evidence of the MITF-CRYAB transcriptional axis in other 238 cancer types, in melanoma both MITF and CRYAB expression are upregulated by BRAF/MEK-inhibitor 239 treatments (49, 52), suggesting that this regulation can go beyond both prostate cancer scenario. 240 Indeed, we observed that the correlation between MITF and CRYAB is also present in colorectal cancer, 241 but not in breast nor lung cancer (data not shown). 242

In our study, the MITF-CRYAB transcriptional axis is reduced and exerts tumor suppressive activity in 243
PCa. This is in agreement with the reduced expression of CRYAB observed in PCa patients and its 244 previous consideration as a protective gene against PCa (32). Yet, the exact molecular mechanism 245 underlying the tumor suppressive activity of CRYAB remains to be elucidated. 246 Importantly, the extensive interrogation of PCa transcriptomes and associated clinical data has led us 247 to propose the transcriptional axis MITF-CRYAB as a potential prognostic biomarker in PCa. The 248 individual expression of CRYAB and MITF has been previously associated with poor prognosis in 249 various tumor types (26, 29-31, 50, 51) and to therapy response in melanoma (53-55). However, our 250 data showing enhanced prognostic potential of the combined signature provides a new and exciting 251 perspective of the functional interaction of these genes in PCa. 252 Our study endorses the potential of transcriptional deregulation analysis, as either a cause or a 253 consequence of cancer, and its impact to support the discovery of novel cancer related genes and long-254 term development of novel cancer treatment strategies. 255

ACKNOWLEDGMENTS 256
Apologies to those whose related publications were not cited due to space limitations. The work of Doxycycline hyclate (Dox) and Puromycin were purchased from Sigma, and Hygromycin from 288

Xenotransplant assays 290
All mouse experiments were carried out following the ethical guidelines established by the Biosafety 291 and Welfare Committee at CIC bioGUNE. The procedures employed were carried out following the 292 recommendations from AAALAC. Xenograft experiments were performed as previously described (14), 293 injecting 10 6 cells per condition in two flanks per mouse (Nu/Nu immunodeficient males; 6-12 weeks of 294 age). PC3 TRIPZ-HA-MITFA cells alone or under CRYAB silencing were injected in each flank of nude 295 mice and 24 h post-injections mice were fed with chow or doxycycline diet (Research diets, 296 D12100402). 297

Patient samples 298
All samples were obtained from the Basque Biobank for research (BIOEF, Basurto University hospital) 299 upon informed consent and with evaluation and approval from the corresponding ethics committee 300 (CEIC code . 301

Molecular assays 302
Western blot was performed as previously described (14

Cellular assays 317
Cell number quantification with crystal violet was performed as referenced (14). 318 For starvation experiments 100,000 cells per well were seeded in a 6-well plate. Cells were initially 319 plated in 10% FBS media for 24 hours and then the media was changed to FBS free media and left 320

overnight. 321
Soft agar assays were performed as previously described (14), seeding 5,000 cells per well in 6-well 322

Chromatin Immunoprecipitation 329
Chromatin Immunoprecipitation (ChIP) was performed using the SimpleChIP ® Enzymatic Chromatin IP 330 Kit (Cat: 9003, Cell Signalling Technology, Inc). Four million PC3 cells were grown in 150 mm dishes 331 either with or without 0.5 µg mL -1 doxycycline during 3 days. Cells from three 150 mm dishes were 332 cross-linked with 35% formaldehyde for 10 min at room temperature. Glycine was added to dishes 333 during 5 min at room temperature. Cells were then washed twice with ice-cold PBS, and scraped into 334 PBS+PMSF. Pelleted cells were lysed and nuclei were harvested following manufacturer's instructions. 335 Nuclear lysates were digested with micrococcal nuclease for 20 min at 37°C and then sonicated in 336 500μl aliquots on ice for 6 pulses of 20 s using a Branson sonicator. Cells were held on ice for at least 337 1 min between sonications. Lysates were clarified at 11,000 × g for 10 min at 4°C, and chromatin was 338 in the different groups, and a p-value that estimates the statistical power of the differences observed. 366 Correlation analysis: Spearman correlation test was applied to analyse the relationship between paired 367 genes. From this analysis, Spearman coefficient (R) indicates the existing linear correlation 368 (dependence) between two variables X and Y, giving a value between +1 and −1 (both included), where 369 1 is total positive correlation, 0 is no correlation, and −1 is total negative correlation. The p-value 370 indicates the significance of this R coefficient. confirmed or assumed (for n<5) and Student T test was applied for two component comparisons. In the 381 statistical analyses involving fold changes, one sample t-test with a hypothetical value of 1 was 382 performed. The confidence level used for all the statistical analyses was of 95% (alpha value = 0.05). 383 Two-tail statistical analysis was applied for experimental design without predicted result, and one-tail 384 for validation or hypothesis-driven experiments. 385 Gene expression array data analysis: first, raw expression data were background-corrected, log2-386 transformed and quantile-normalized using the lumi R package7 387 , available through the Bioconductor repository. Probes with a "detection p-value" lower than 0.01 in at 388 least one sample were considered expressed. For the detection of differentially expressed genes, a 389 linear model was fitted to the probe data and empirical Bayes moderated t-statistics were calculated 390 using the limma package from Bioconductor. Only genes with differential fold-change (FC) >1.5 or <-391 1.5 and an adjusted p-value < 0.05 were considered as differentially expressed. 392

Accession numbers and datasets 393
Primary accessions: The transcriptomic data generated in this publication have been deposited in 394 NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number 395 GSE114345 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE114345).        Valcarcel-Jimenez at al. Figure 4