A prostate cancer susceptibility allele at 6q22 increases RFX6 expression by modulating HOXB13 chromatin binding

Transcription

1 Supplementary Information A prostate cancer susceptibility allele at 6q22 increases RFX6 expression by modulating HOXB13 chromatin binding Qilai Huang 1,2*, Thomas Whitington 3,4*, Ping Gao 1,2, Johan F. Lindberg 4, Yuehong Yang 1,2, Jielin Sun 5, Marja-Riitta Väisänen 6, Robert Szulkin 4, Matti Annala 7, Jian Yan 3, Lars A. Egevad 8, Kai Zhang 1,2, Ruizhu Lin 1,2, Arttu Jolma 3,9, Matti Nykter 7, Aki Manninen 1,2, Fredrik Wiklund 4, Markku H. Vaarala 6,10, Tapio Visakorpi 7, Jianfeng Xu 5, Jussi Taipale 3,9# & Gong-Hong Wei 1,2# *These authors contributed equally to this work. # To whom correspondence and request for materials should be addressed to G.H.W. (gonghong.wei@oulu.fi) or J. T. (jussi.taipale@ki.se). Affiliations: 1 Biocenter Oulu, University of Oulu, Oulu, Finland 2 Department of Medical Biochemistry and Molecular Biology, Institute of Biomedicine, University of Oulu, Oulu, Finland 3 Department of Biosciences and Nutrition, Karolinska Institutet, Stockholm, Sweden 4 Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden 5 Center for Cancer Genomics, Wake Forest School of Medicine, Winston-Salem, USA 6 Department of Pathology, Oulu University Hospital, Oulu, Finland 7 Prostate Cancer Research Center, Institute of Biomedical Technology and BioMediTech, University of Tampere and Tampere University Hospital, Finland 8 Department of Pathology and Cytology, Karolinska University Hospital, Stockholm, Sweden 9 Genome-Scale Biology Program, University of Helsinki, Helsinki, Finland 10 Medical Research Center Oulu, Oulu, Finland 1

2 This supplementary PDF file contains the following information: Supplementary Figures 1-14 Supplementary References Notes: Other Supplementary Materials for this manuscript includes the following: Supplementary Tables 1-12 (separately uploaded) Supplementary Table 1 GWAS vs ChIP-seq Supplementary Table 2 SNP-PWM match Supplementary Table 3 Tampere gene expression Supplementary Table 4 Swedish gene expression for rs eqtl analysis Supplementary Table 5 Clinical datasets of gene expression Supplementary Table 6 SNP genotyping at 6q22 Supplementary Table 7 Taylor clinical samples Supplementary Table 8 TCGA clinical samples Supplementary Table 9 Oligonucleotides used in this study Supplementary Table 10 Gene cloning and related primers Supplementary Table 11 Antibodies Supplementary Table 12 Target sequences of sirna and shrna Supplementary Data Set 1 (separately uploaded) Supplementary Data Set 1 Transcription factor position-weight matrix Inventory of Supplementary Figures and Tables Supplementary Figure 2 and Supplementary Table 1 relate to Figures 1 Supplementary Figure 3 and 4 relate to Figures 2 Supplementary Figure 5-9 and Supplementary Table 3-6 relate to Figures 3 Supplementary Figure 10 and 11 relate to Figures 4 Supplementary Figure 12-14, Supplementary Table 3-5, 7, 8 relate to Figures 5 2

3 3

4 Supplementary Figure 1: The homeodomain protein HOXB13 is a prostatelineage-specific transcription factor. (a) The tissue-specific pattern of HOXB13 mrna expression in a panel of 79 human tissues (GSE1133) 11,53. (b) HOXB13 mrna expression levels were determined in a set of 184 human primary tumor samples representing several distinct cancer types including prostate cancer 11,53. The expression levels of HOXB13 were determined by designed custom arrays including the vast majority of protein-coding genes in human genome 11. Note that HOXB13 is specifically and highly expressed in normal prostate gland and prostate tumors, in contrast to other normal human tissues (a) or tumors derived from different cancer types (b). 4

5 5

6 Supplementary Figure 2: The binding specificities of HOX transcription factors, AR and FOXA1 to the DNA sequences harboring the prostate cancer risk-associated SNP rs (a) Relative binding affinity of 19 HOX transcription factors to the DNA sequences harboring risk T allele of rs The binding affinities were measured using the luciferase-based DNA binding assay 21. Error bars, ± s.d. (n = 6 technical replicates). Note that, in contrast to other HOX members, HOXB13 indicate the strongest binding affinity to the DNA sequences harboring the rs T allele measured by Renilla luciferase activity. (b-d) EMSA for the binding activity of HOXB13 (b), AR (c) or FOXA1 (d) to the HOXB13 DNA-binding consensus sequences harboring the SNP rs Double-stranded FAM labeled HOXB13 consensus DNA was incubated with COS-1 cell nuclear extract with ectopically expressed protein as indicated. Note that rs T allele shows stronger binding affinity for HOXB13 than the C allele. In contrast, we observed no any binding activity of AR (c) or FOXA1 (d) to the HOXB13 DNA-binding consensus, as indicated by nonspecific bands in (c) or (d). The boiled/denatured COS-1 cell nuclear extract containing HOXB13 recombinant proteins does not show any binding activity to the HOXB13 DNA-binding consensus (lane 8 in b). Lane 1 represents negative control. The binding of HOXB13 consensus sequence to human recombinant HOXB13 protein (lane 2) is competed by a scrambled sequence (lane 3) and by the sequence containing the rs T allele (lane 4) and C allele (lane 5), the permutation of A base (lane 6) and G base (lane 7) at the same chromosome location as rs339331, and the lane 8 in b represents the results from the incubation of the HOXB13 DNA-binding consensus oligos with the boiled COS-1 cell nuclear extract containing ectopically expressed HOXB13 proteins. 6

7 Supplementary Figure 3: ChIP-qPCR confirmation of HOXB13 chromatin binding to the rs enhancer at the prostate cancer 6q22 risk locus. (a) ChIP-qPCR analysis confirmed the HOXB13 chromatin binding to an rs containing region and a positive 7

8 enhancer control in prostate cancer cell VCaP under different treatments. Error bars, ± s.d. (n = 3 technical replicates). ** P < 0.01, *** P < Eth means ethanol treatment for 24 hours; R1881, 1nM synthetic androgen methyltrienolone (R1881) treatment for 24 hours; and, DHT, 100nM dihydrotestosterone (DHT) treatment for 24 hours. (b) HOXB13 favors binding to the risk allele T at rs determined by ChIP followed by allele-specific quantitative PCR (ChIP-AS-qPCR; see also Fig. 1b). The qpcr primers were designed for allele-specific amplification of the rs region harboring T (red) or C allele (blue). Error bars, ± s.d. (n = 3 technical replicates). * P < 0.05, ** P < (c) ChIP-qPCR analysis of epigenetic signatures at a silent chromatin region. Note that the repressive histone mark H3K27me3 alone shows a strong enrichment at given silent chromatin region. Error bars, ± s.d. n = 3 technical replicates. *** P < compared to IgG control. In a-c, the P values were assessed using two-tailed Student s t-test. (d-e) Shown are genome browser representations of HOXB13, H3K27ac/me3 and H3K4me1 enrichment profiles determined by ChIP-seq in VCaP cells, indicating a positive enhancer region at the prostate specific antigen (PSA, also called KLK3) locus 54 (d), and a silent chromatin region with repressive mark H3K27me3 (e), a histone modification associated with polycomb silencing 55. Abbreviations: H3K4me1, the monomethylation of histone H3 at lysine 4; H3K27ac, the acetylation of histone H3 at lysine 27; and H3K27me3, the trimethylation of histone H3 at lysine 27. 8

9 Supplementary Figure 4: Cell type-specificity and allele-specific differences in chromatin state of rs enhancer at prostate cancer susceptibility 6q22 locus. (a) FAIRE-qPCR assays were performed in prostate cancer cell 22Rv1 to map open chromatin 9

10 state and nucleosome occupancy at rs centered 8kb region of 6q22 locus (indicated by chromosome coordinate, build, hg19). Fold enrichment was calculated as a ratio of the signal from the FAIRE sample relative to the signal from input control DNA by qpcr. The primer sets and amplicons are indicated as red arrows. FAIRE refers to formaldehyde-assisted isolation of regulatory elements 28,29. Note that the strongest FAIRE signals were observed at rs region. Error bars, ± s.d. n = 3 technical replicates. (b) Evidence from the ENCODE project indicating that rs is locating at a prostate cancer cell-type-specific open chromatin region. Top: Shown is DNase-seq signals in LNCaP cells and other 10 non-prostate cancer cell models from the ENCODE project 30,31. Bottom: Shown is FAIRE-seq signals in 5 different cell lines from the ENCODE project. The exact chromosome position of rs is highlighted by a vertical line. Both DNase-seq and FAIRE-seq can define regulatory elements with cell-type-specific open chromatin state 56. Note that a strong DNase-seq signal at rs region was specifically observed in prostate cancer cell line LNCaP, further suggesting this is a prostate cancer cell-type-specific enhancer. In addition, LNCaP DNaseseq result at rs is well consistent with our FAIRE-qPCR results (see panel a and Fig. 2g) and ChIP-seq profiles for HOXB13 and active enhancer marks H3K4me1 and H3K27ac 26,27,57 (see Fig. 1d and 2f). (c) Allele-specific differences in chromatin state of the rs enhancer. Sanger sequencing profiles of FAIRE and input DNA for the region around rs revealed that the T risk allele is enriched in the open chromatin fragments compared to input DNA in both VCaP and 22Rv1 cells. (d) ChIP-AS-qPCR reveals the preference of HOXB13 binding to the rs risk allele T in 22Rv1 cells (see also Fig. 2b and Supplementary Fig. 3b in VCaP cells). Note that the occupancy of HOXB13 to the rs region with T allele is significantly increased in response to DHT treatment. (e) Allele specificity of chromatin state at rs region. The FAIRE signals at rs region with T or C allele were examined using AS-qPCR. Note that the FAIRE signals are significantly higher at rs region with T allele in comparison with C allele. Overall, androgen treatment induces chromatin opening while the change at T allele is prominent (indicated by grey line). This analysis together with the observation in (c), suggest that the variation at rs may alter the chromatin opening states at this locus. Furthermore, the preference of HOXB13 occupancy at rs region harboring T risk allele (d) may contribute to enhanced chromatin opening at the same allele. In (d) and (e), Error bars, ± s.d. (n = 3 technical replicates). (f) Luciferase reporter assays showing increased enhancer activity of rs T over C alleles in VCaP. Note that androgen treatment stimulates rs T enhancer activity. The fragment of the enhancer used here is rs centered 600bp region (chr6: ). Error bars, ± s.d. (n = 4 technical replicates). In d-f, * P < 0.05, ** P < 0.01, ***P < were evaluated using two-tailed Student s t-test. 10

11 Supplementary Figure 5: The expression status of RFX6, GPRC6A and HOXB13 in human prostate tumors. (a) RFX6 expression levels in a clinical dataset of 57 human prostate tumors reported in Grasso et al. 35 The "Fold change" value pertains to the specified sample relative to a pooled benign prostate tissue sample (refer to GSE35988 for additional information) 35. (b) RFX6 expression levels in 35 human prostate carcinomas from the Tamura et al. dataset 36. (c-e) The expression levels of RFX6 (c), HOXB13 (d) or GPRC6A (e) in a cohort of 150 human prostate tumors (primary and metastatic) are from the Taylor et al. dataset 37. The relative expression level is displayed as Z-score versus matched normal control tissues. Positive Z-score indicates increased expression, and negative means decreased expression. Note that all the datasets in (a-c) show consistent RFX6 upregulation in a subset of human prostate tumors. HOXB13 are upregualted with higher frequency in the reported clinical tumor samples 37. In contrast, GPRC6A is less frequently expressed in the reported dataset

12 Supplementary Figure 6: shrna-mediated knockdown of HOXB13 confers reduced expression of RFX6 in prostate cancer cells. (a-c) Depletion of HOXB13 greatly decreases the transcriptional level of RFX6 gene in LNCaP (a), 22Rv1 (b), and VCaP (c). Error bars, ± s.d. (n = 3 technical replicates). Note that the results from this shrna-mediated knockdown assay are well consistent with our independent sirna-mediated depletion of HOXB13 in the given prostate cancer cell models (see also Fig. 3b-d). 12

13 13

14 Supplementary Figure 7: Expression correlation of RFX6 or GPRC6A with HOXB13 in human prostate tumor samples. (a,b) Scatter plot showing the positive correlation between RFX6 and HOXB13 expression (a), and the inverse correlation between GPRC6A and HOXB13 expression (b) in a subset of prostate tumors (n=103) with upregulated HOXB13 from the dataset 37. (c) Scatter plot showing the positive correlation between RFX6 and HOXB13 expression in a subset of prostate tumors (n=73) with upregulated RFX6 from the dataset 37. (d,e) Scatter plot showing a positive correlation between RFX6 and HOXB13 expression (d) and the inverse correlation between GPRC6A and HOXB13 (e) in an independent cohort of prostate tumors (n=176) from the TCGA dataset (queried from cbio Portal 40 ). (f,g) Scatter plot showing the positive correlations between RFX6 and HOXB13 expression in a Finnish cohort of prostate cancer (f), and a subset of commercialized prostate tumor samples (g). The statistical significance of the correlation was assessed by performing Pearson s correlation coefficient tests. In d-f, the statistical significance was also evaluated using Spearman s rank correlation coefficient. 14

15 Supplementary Figure 8: Depletion of HOXB13 confers reduced occupancy of transcription factors and chromatin openness at rs region. (a) shrnamediated depletion of HOXB13 reduces chromatin occupancy of AR and FOXA1 at rs containing region in prostate cancer cell LNCaP. Error bars, ± s.e.m. (n = 3 technical replicates). (b-d) shrna-mediated depletion of HOXB13 diminishes chromatin openness revealed by FAIRE-qPCR in prostate cancer cell line LNCaP (b), 22Rv1 (c), and VCaP (d). Fold enrichment was calculated as a ratio of the signal revealed by qpcr from FAIRE sample relative to the signal from input DNA. Error bars, ± s.d. (n = 3 technical replicates). *** P < by two-tailed Student s t-tests. 15

16 16

17 Supplementary Figure 9: The prostate cancer risk-associated SNP rs is involved in allelic imbalance of RFX6 expression modulated by HOXB13. The prostate cancer risk-associated SNP rs is involved in allelic imbalance of RFX6 expression modulated by HOXB13. (a) Genome browser track presents complete structure of exons and introns for RefSeq human gene RFX6. GWAS SNP rs is locating within intron 4 of RFX6. (b) A focused view of non-coding SNP rs centered 21bp in intron 4 and coding SNP rs centered 21bp in exon 14 of RFX6 gene, indicating each SNP reference allele is T base; the other allele is C base. (c) Sanger sequencing profiles indicate different signal density for the alleles of non-coding SNP rs and coding SNP rs in RFX6 gene. Sanger sequencing samples include genomic DNA and cdna reversely transcribed from RNA under different treatments. The first and second columns, respectively, show Sanger sequencing profiles surrounding rs site in 22Rv1 and VCaP cells. The third column shows Sanger sequencing signals for coding SNP rs in VCaP cells. Note that Sanger sequencing of both rs and rs regions result in similar chromatogram signals from VCaP cells under the same growth condition, further supporting a causal role of rs mediated disruption of the HOXB13 motif in modulating its chromatin occupancy, which in turn causes allelic imbalance in RFX6 expression (see also Fig. 3fh). (d) Western blot results indicate differential expression of HOXB13 at protein level across 6 prostate cancer cell lines (VCaP, LNCaP, DuCaP, PC3, DU145 and 22Rv1) and one breast cancer cell model MCF-7. Note that HOXB13 protein abundance in either VCaP or LNCaP is clearly higher than the level in 22Rv1 cells. (e) Ratio of T allele to C allele at coding SNP rs site shows allelic imbalance in RFX6 transcription in VCaP, consistent with the observations in Fig. 3f-h. (f) Depletion of HOXB13 reduces the chromatin occupancy of HOXB13 and AR preferentially at rs region with T allele. AS-PCR was applied to examine the ChIP signal specifically at rs region with T allele or C allele. Shown are the ChIP signals at T allele normalized to C allele. Error bars, ± s.d. (n = 3 technical replicates). *** P < by two-tailed Student s t-tests. (g) Depletion of HOXB13 reduces the chromatin open state specifically at rs region with T allele. AS-PCR was applied to examine the FAIRE signal specifically at rs region with T allele or C allele. Shown are the FAIRE signals at T allele normalized to C allele. Error bars, ± s.d. (n = 3 technical replicates). ** P < 0.01 by two-tailed Student s t-tests. (h) Luciferase reporter assay reveals increased enhancer activity of rs T over C alleles to the RFX6 promoter. The fragment of the enhancer used here is rs centered 600bp region (chr6: ), 300bp shorter than the one used in Fig. 3j. Error bars, ± s.d. (n = 5 technical replicates). The P values were evaluated using two-tailed Student s t-tests. 17

18 Supplementary Figure 10: The transcription factor gene RFX6 and HOXB13 influence cellular phenotypes related to tumor-associated properties in prostate cancer cells. (a,b) Efficiency of sirnas or shrna-mediated RFX6 knockdown was measured using quantitative real-time PCR. Error bars, ± s.d. (n = 3 technical replicates). Note that the most efficient sirna or shrna against RFX6 was selected for 18

19 subsequent assays to test functional roles of RFX6 in cell growth control and cellular transformation. (c) Cell proliferation was measured at the indicated time points by XTT colorimetric assay (absorbance at 450 nm; mean ± s.d. of three independent experiments). ** P < 0.01, *** P < 0.001, **** P < , using two-tailed Student s t-test. (d) Cell growth potential was real-time monitored and assessed by an Electric Cell-Substrate Impedance Sensing (ECIS) assay. Note that capacitance at high frequency exhibits significant correlation with cell free electrode area. When cells attached to the electrode or cell proliferation occurred, the capacitance will drop. (e) Cell growth rate was real-time monitored using an ECIS assay. Note that when cell proliferation occurs, the resistance will increase. (f) Representative images of cell migration assay of 22Rv1 cells transfected with control and the most efficient sirna against HOXB13. Scale bar, 100µm. (g) Quantitation of relative migration for the cells transfected with control and the most efficient sirna against HOXB13. (h) Representative images of cell invasion assay of 22Rv1 cells transfected with control and the most efficient sirna against HOXB13. Scale bar, 100µm. (i) Quantitation of relative invasion for the cells transfected with control and most efficient sirna against HOXB13. In g and i, error bars, ± s.d. n = 3 technical replicates, the P values were assessed using two-tailed Student s t-test. 19

20 Supplementary Figure 11: RFX6 and HOXB13 play a role in prostate cancer cell proliferation and invasion. (a) Representative images of invasion assay for control shrna lentivirus particle infected 22Rv1 cells transfected with empty, HOXB13 or RFX6 expression vector. (b) Representative images of invasion assay for HOXB13 shrna lentivirus particle infected 22Rv1 cells transfected with empty or RFX6 20

21 expression vector. (c) Representative images of invasion assay for RFX6 shrna lentivirus particle infected 22Rv1 cells transfected with empty or HOXB13 expression vector. In a-c, scale bar, 100µm. (d) Quantitation of relative invasion for control shrna lentivirus particle infected 22Rv1 cells transfected with empty, HOXB13 or RFX6 expression vector as shown in a. Note that ectopic overexpression of HOXB13 or RFX6 enhance 22Rv1 invasion, in agreement with our invasion assays with depletion of HOXB13 (Supplementary Fig. 10h,i) or RFX6 (Fig. 4e,f). (e) Quantitation of relative invasion for HOXB13 shrna lentivirus particle infected 22Rv1 cells transfected with empty or RFX6 expression vector. Note that ectopic overexpression of RFX6 significantly restored the decreased cell invasion caused by shrna-mediated HOXB13 knockdown. (f) Quantitation of relative invasion for RFX6 shrna lentivirus particle infected 22Rv1 cells transfected with empty or HOXB13 expression vector. In contrast to the functional rescue of RFX6 to HOXB13 depletion, to a lesser extent, ectopic overexpression of HOXB13 restored the decreased cell invasion caused by shrnamediated RFX6 knockdown. A possible explanation to this observation is that HOXB13 is an upstream regulator of RFX6 gene expression. The increased RFX6 transcripts may be simultaneously degraded by the shrnas against RFX6. In d-f, Error bars, ± s.e.m. (n = 3 technical replicates). The P values were assessed using two-tailed Student s t-test. 21

22 22

23 Supplementary Figure 12: RFX6 upregulation in human prostate cancer correlates with tumor progression and the risk of biochemical relapse. (a) RFX6 mrna expression is upregulated in a subset of prostate tumors in comparison with matched adjacent normal controls from the same prostate tissues. The box plot in the inset shows that RFX6 expression is significantly higher in the tumors than the matched normal controls. Note that RFX6 mrna expression levels were measured by quantitative PCR in a Finnish Oulu cohort of prostate cancer tissues from the patients who underwent radical retropubic prostatectomy. In contrast, there are no detectable expression levels of GPRC6A expression in most of the samples (data not shown). (b) RFX6 is highly expressed in metastatic prostate tumors from an independent clinical dataset 39. Normalisation procedure for RFX6 expression intensity is stated in GSE3325. (c) RFX6 mrna expression is strikingly upregulated in metastasis in the Taylor et al. dataset 37. By contrast, GPRC6A expression does not correlate with prostate tumor metastasis in the same dataset. Normalisation procedure for RFX6 expression intensity is stated in GSE (d) GPRC6A expression does not correlate with prostate tumor progression to high Gleason score in the ref. 37 dataset. In contrast, from the ref. 37 dataset, we observed a significant correlation of RFX6 upregulation with tumor progression to high Gleason score (see Fig. 5d). (e,f) RFX6 upregulation markedly correlates with elevated serum PSA levels (e), a diagnostic marker for prostate cancer, and biochemical recurrence (f) in a cohort of patients reported in the ref. 37. In contrast, GPRC6A expression does not show any apparent association with PSA levels (e) and recurrence (f). In a-f, the horizontal lines represent the median values. The significance of comparison between groups was assessed by Mann-Whitney U -tests. (g,h) Kaplan Meier plot of the risk for biochemical recurrence in two patient groups with higher (top 75% of GPRC6A upregulated tumors; n = 55) or lower (bottom 75% of GPRC6A downregulated tumors; n =42) expression of GPRC6A in a cohort of patients reported in the ref. 37 (g) (see Supplemental Table 7 for clinical variables), and with higher (top 75% of GPRC6A upregulated tumors; n = 22) or lower (bottom 75% of GPRC6A downregulated tumors; n =63) expression of GPRC6A in an independent cohort of patients from the TCGA datasets of prostate adenocarcinomas (TCGA Research Network; see also Supplemental Table 8). Overall, the patients with tumors expressing higher levels of GPRC6A shows no increased risk of biochemical relapse. The number of patients in each group at every 10-month interval was indicated. The P value was calculated by a Log-rank test. 23

24 Supplementary Figure 13: Prognostic potential of RFX6 or GPRC6A upregulation and the effect of RFX6 deletion on the risk of biochemical recurrence in prostate cancer. (a) Kaplan Meier plots of the risk of biochemical recurrence in prostate tumors with high (Z-score > 1.75) or low (Z-score <= 1.75) expression of RFX6 (left panel) or GPRC6A (right panel) in the ref. 37 dataset. (b) Kaplan Meier plots of biochemical recurrence in two patient groups with high or low RFX6 expressing tumors with the cases carrying RFX6 deletion excluded. (c) Kaplan Meier plots of biochemical recurrence in two patient groups carrying the tumors with an RFX6 deletion or without RFX6 deletion. Note that, the correlation between RFX6 expression and biochemical recurrence is not explained by RFX6 deletion status. The P values in a-c are calculated with a log-rank test. Numbers of patients in each category are shown for each ten-month interval. 24

25 Supplementary Figure 14: The correlation between RFX6 expression and tumor aggressiveness is not explained by tumor size. (a) Tumor size is a potential confounding variable that could explain the observed correlation between RFX6 expression and tumor aggressiveness. Hence, we have examined two hypotheses using the Taylor et al. dataset 37. H0: RFX6 expression does not actually correlate with tumor severity in pure tumor samples. Instead, the observed RFX6 expression/tumor severity correlation is due to small (and low severity) tumors being contaminated with normal cells, which results in small tumors yielding low RFX6 expression measurements. H1: 25

26 RFX6 expression does correlate with severity in pure tumor samples, and our observed RFX6/severity correlation is not due to contamination from normal cells in small tumors. This schematic summarizes the rationale for hypothesis H0. The remaining data in this figure provide evidence refuting this hypothesis. (b) Kaplan Meier plots of biochemical recurrence in two patient groups with small tumors (pathological stage T2A/T2B) or large tumors (pathological stage T2C or larger). Tumor size correlates with tumor severity (the rate of biochemical recurrence) as one would expect, since larger tumors are previously known to be more aggressive on average. This correlation suggests that the pathological staging metadata is reliable in this cohort. (c) Boxplot of RFX6 expression for primary tumors of increasing size. Contrary to hypothesis H0 defined above, RFX6 expression does not correlate with tumor size in the primary tumor samples. The whiskers in the boxplot indicate the range of the data and the horizontal bar shows the median. The P value was assessed by a Kruskal-Wallis test. (d) Kaplan- Meier plots of biochemical recurrence in two patient groups carrying the tumors with high or low levels of RFX6 expression (the small primary tumors (pathological stage T2A/T2B) excluded). This survival analysis still yields a significant difference in biochemical recurrence between high RFX6 and low RFX6 expression groups, even when small tumors (T2A and T2B) are excluded from the analysis. This suggests that normal tissue contamination in small tumors cannot account for the observed correlation. In b and d, numbers of patients in each category are shown for each tenmonth interval, and the P values were calculated with a log-rank test. (e) Quantilequantile plot comparing the distribution of RFX6 expression levels in metastatic tumors with that in primary tumors. In contrast to the analysis shown in Supplementary Fig. 12c, this analysis excluded the small primary tumors (pathological stage T2A/T2B). Therefore the dramatic difference between primary and metastatic RFX6 expression persists even after small primary tumors are excluded. Expression units are Z-score, defined as the number of standard deviations from the mean RFX6 expression as estimated from normal prostate tissue. 26

27 Supplementary References 53. Wu, C., Macleod, I. & Su, A. I. BioGPS and MyGene.info: organizing online, gene-centric information. Nucleic Acids Res. 41, D561-5 (2013). 54. Wang, Q., Carroll, J. S. & Brown, M. Spatial and temporal recruitment of androgen receptor and its coactivators involves chromosomal looping and polymerase tracking. Mol. Cell 19, (2005). 55. Kerppola, T. K. Polycomb group complexes--many combinations, many functions. Trends Cell Biol. 19, (2009). 56. Song, L. et al. Open chromatin defined by DNaseI and FAIRE identifies regulatory elements that shape cell-type identity. Genome Res. 21, (2011). 57. Rada-Iglesias, A. et al. A unique chromatin signature uncovers early developmental enhancers in humans. Nature 470, (2011). 27