bioRxiv preprint doi: https://doi.org/10.1101/220400; this version posted November 17, 2017. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY 4.0 International license. 1 Cholesterol Synthesis Pathway Genes in Prostate Cancer are consistently 2 downregulated when tissue confounding is minimized. 3 4 Morten Beck Rye*1,2, Helena Bertilsson3,1, Maria K. Andersen4, Kjersti Rise1, Tone F. 5 Bathen4, Finn Drabløs1, May-Britt Tessem2,4 6 7 1 Department of Clinical and Molecular Medicine, Norwegian University of Science and 8 Technology (NTNU), 7491 Trondheim, Norway 9 2 Clinic of Surgery, St. Olavs Hospital, Trondheim University Hospital, 7030 Trondheim, 10 Norway 11 3 Department of Urology, St. Olavs Hospital, Trondheim University Hospital, 7030 12 Trondheim, Norway 13 4 MI Lab, Department of Circulation and Medical Imaging, Norwegian University of Science 14 and Technology (NTNU), 7491 Trondheim, Norway 15 16 MBR – [email protected] 17 HB – [email protected] 18 MKA - [email protected] 19 KR – [email protected] 20 TFB – [email protected] 21 FD – [email protected] 22 MBT – [email protected] 23 24 *Corresponding author: 25 Morten Beck Rye 26 Department of Cancer Research and Molecular Medicine 27 Norwegian University of Science and Technology 28 P.O. Box 8905 29 NO-7491 Trondheim 30 Norway 31 Email: [email protected] 32 Phone: +47 97133155 33 34 35 36 37 38 39 40 41 1 bioRxiv preprint doi: https://doi.org/10.1101/220400; this version posted November 17, 2017. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY 4.0 International license. 1 Abstract 2 3 The relationship between cholesterol and prostate cancer has been extensively studied for 4 decades, where high levels of cellular cholesterol are generally associated with cancer 5 progression and less favorable outcomes. However, the role of in vivo cellular cholesterol 6 synthesis in this process is unclear, and data on the transcriptional activity of cholesterol 7 synthesis pathway genes in tissue from prostate cancer patients are inconsistent. A common 8 problem with cancer tissue data from patient cohorts is the presence of heterogeneous tissue 9 which confounds molecular analysis of the samples. In this study we present a method to 10 minimize systematic confounding from stroma tissue in seven patient cohorts consisting of 11 1713 prostate cancer and 230 normal tissue samples. When confounding was minimized, 12 differential gene expression analysis over all cohorts showed robust and consistent 13 downregulation of nearly all genes in the cholesterol synthesis pathway. Additional analysis 14 also identified cholesterol synthesis as the most significantly altered pathway in prostate 15 cancer. This surprising observation is important for our understanding of how prostate cancer 16 cells regulate cholesterol levels in vivo. Moreover, we show that tissue heterogeneity explains 17 the lack of consistency in previous expression analysis of cholesterol synthesis genes in 18 prostate cancer. 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 2 bioRxiv preprint doi: https://doi.org/10.1101/220400; this version posted November 17, 2017. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY 4.0 International license. 1 Introduction 2 Increased cholesterol levels in enlarged prostates and prostate cancer have been observed for 3 decades1-3, and extensive research has suggested that cholesterol have a role in prostate cancer 4 growth and progression3-5. Cholesterol homeostasis is important for cell viability, and is 5 dynamically regulated by a balance between synthesis, uptake, efflux and storage of 6 cholesterol4,6-9. For cellular cholesterol synthesis, the conversion of 3-hydroxy-3- 7 methylglutaryl coenzyme A (HMG CoA) to mevalonate is the first rate limiting step, which is 8 followed by over 20 flux controlling enzymatic reactions before cholesterol is synthesized as 9 the final product. In prostate cancer cell-lines, elevated activity of the cholesterol synthesis 10 pathway supports cancer growth and aggressiveness10-16. This has led to the general view that 11 increased cholesterol synthesis in prostate cancer cells contributes to cellular accumulation of 12 cholesterol and prostate cancer growth. A diet high in fat and cholesterol increase the risk of 13 prostate cancer, while statins directly targeting the cholesterol synthesis pathway are 14 associated with improved clinical outcome (reviewed in17). This is generally taken as support 15 for the relevance of increased cholesterol synthesis in vivo. This notion was also in line with a 16 recent study showing increased activity of the cholesterol synthesis enzyme squalene 17 monooxygenase (SQLE) in lethal prostate cancer18. Accordingly, one would expect that genes 18 in the cholesterol synthesis pathway are upregulated when prostate cancer is compared to 19 normal tissue. However, transcriptional changes in cholesterol genes are rarely highlighted 20 when such comparisons are performed in large patient cohorts. 21 22 We hypothesized that this is due to influence of confounding tissue components present in the 23 samples. Gene expression analysis in human tissue is challenged by the highly heterogeneous 24 tissue composition in each sample19,20. The standard way to account for such heterogeneity is 25 to incorporate tissue type percentages from histopathology during the analysis. Although 3 bioRxiv preprint doi: https://doi.org/10.1101/220400; this version posted November 17, 2017. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY 4.0 International license. 1 confounding due to tissue composition is generally acknowledged, data from histopathology 2 are missing in most publicly available patient cohorts, which may bias the molecular analyses. 3 In prostate cancer, the presence of stroma tissue is shown to hide underlying molecular 4 features in a differential analysis21,22. Prostate tissues are usually histopathologically divided 5 into benign epithelium, stroma tissue and prostate cancer. It is previously shown that the 6 different number of tissue types present in prostate cancer (three tissue types) and in normal 7 samples (two tissue types) leads to a systematic sampling bias of increased stroma content in 8 the normal samples23,24. This confounds differential analysis when cancer and normal samples 9 are compared, and controlling for these biases will potentiate the discovery of molecular 10 pathways and features otherwise hidden in the data. 11 12 To address this challenge we utilized two independent patient cohorts where the tissue 13 composition of prostate cancer and normal samples has been thoroughly assessed by 14 histopathology. Based on the gene expression analysis of stroma-enriched genes in these two 15 cohorts, we used Gene Set Enrichment Analysis (GSEA)25 to assess the stroma content in five 16 other patient cohorts where no histopathology is available. In total 1713 prostate cancer and 17 230 normal samples were assessed for their stroma content. To create datasets from all 18 cohorts where the confounding effect of stroma tissue is accounted for, we used our recently 19 published approach of balancing tissue composition23. When differential expression analysis 20 is performed on these datasets, consistent downregulation of genes in the cholesterol synthesis 21 pathway is highlighted as one of the most prominent features for primary prostate cancer. 22 23 24 4 bioRxiv preprint doi: https://doi.org/10.1101/220400; this version posted November 17, 2017. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY 4.0 International license. 1 Results and Discussion 2 3 Differentially expressed genes in seven publicly available prostate cancer cohorts 4 controlled for stroma tissue confounding. 5 We used seven publicly available cohorts of tissue samples from patients with prostate cancer 6 (Bertilsson, Chen, Taylor, TCGA, Prensner, Sboner and Erho, referred to as the seven-study- 7 cohort; N=1943 samples, 1713 prostate cancer and 230 normal, Table 1). Gene expression 8 measurements in the various cohorts had been generated using different microarray platforms 9 and RNA-sequencing. Of these seven cohorts, two cohorts (Bertilsson and Chen, referred to 10 as the histopathology cohorts, Table 1) contained detailed histopathology on prostate cancer, 11 stroma and benign epithelium in each sample. These two cohorts were used as a basis for 12 stromal assessment in all seven cohorts. A flow-chart of the different steps in this assessment 13 is provided in Figure 1, and a detailed description of each step is provided in the Methods 14 section. Of the seven cohorts, five cohorts contained measurements of both prostate cancer 15 and samples characterized as normal (Bertilsson, Chen, Taylor, TCGA and Prensner, referred 16 to as the five-study-cohort; 1117 samples, 887 prostate cancer and 230 normal). 17 18 Robust stroma assessment using Gene Set Enrichment Analysis (GSEA) with sets of 19 stroma-enriched genes 20 A key concept in this study is to utilize GSEA25 assessments of stroma content in the tissue 21 samples from the seven patient cohorts to create sub-datasets for differential expression 22 analysis where the confounding effect of stroma tissue is accounted for23.