Intratumoral Sterol-27-Hydroxylase (CYP27A1) Expression in Relation to Cholesterol Synthesis and Vitamin D Signaling and Its Association with Lethal Prostate Cancer

This article is featured in Highlights of This Issue, p. 1001

Prostate tissue has long been recognized to contain considerable amounts of cholesterol, particularly when undergoing carcinogenic transformation (1). More recently, several studies have suggested that higher serum cholesterol levels are associated with increased risk of advanced stage, higher grade, or fatal prostate cancer (2–4), while others reported null associations (5). Higher intratumoral synthesis of cholesterol, as assessed through expression of the second rate-limiting enzyme of cholesterol synthesis, squalene monooxygenase (SQLE), is associated with a higher risk of lethal prostate cancer (6).

A key metabolite of cholesterol is 27-hydroxycholesterol (Fig. 1). Intriguingly, high expression of the enzyme that synthesizes 27-hydroxycholesterol, 27-hydroxylase (CYP27A1), has been reported to be associated with higher tumor grade in breast cancer yet better prognosis (7, 8). CYP27A1 also catalyzes the 25-hydroxylation step of vitamin D, which might have a protective effect in various cancers including prostate cancer (9, 10). In prostate cancer, a recent study reported CYP27A1 expression to be strongly inversely related to Gleason grade (11), but the association with long-term clinical outcomes is unknown.

We hypothesized that low CYP27A1 expression, potentially resulting in cholesterol accumulation, occurs in prostate cancers that have higher expression of the cholesterol synthesis pathway. We also hypothesized that low CYP27A1 expression is associated with low vitamin D signaling. To test these hypotheses, we conducted a cross-sectional analysis of two large, well-characterized populations of patients with prostate cancer. In a longitudinal design, we tested our hypothesis that low CYP27A1 would be associated with a higher risk of lethal prostate cancer over long-term follow-up.

We studied patients who were diagnosed with prostate cancer during follow-up of two prospective cohort studies, the Health Professionals Follow-up Study (HPFS) and the Physicians’ Health Study (PHS).

The HPFS enrolled 51,529 male health professionals, aged 40 to 75 years, in 1986 (12). Participants have been followed through biannual questionnaires since. The PHS enrolled 29,071 male physicians, aged ≥40 years, in 1982, initially for randomized controlled trials of aspirin (13) and micronutrients (14). Blood samples were collected from cancer-free participants in 1982 (PHS) and in 1993–1995 (HPFS). Self-reported prostate cancer diagnoses in both cohorts were verified through review of medical records. Tissue from all patients included in this study also underwent centralized pathology review. Patients were followed prospectively for metastases and prostate cancer–specific death (lethal cancer). Adjudication of death causes was 98% complete in HPFS and 99% complete in PHS.

Within the prostate cancer biorepository from HPFS and PHS, we conducted a nested case–control study of whole-transcriptome profiling of the tumor tissue. It compared patients who developed lethal disease with those who remained metastasis-free for at least 8 years after cancer diagnosis (nonlethal disease) and oversampled patients with lethal outcome and those with available blood specimens from before cancer diagnosis (15).

Participants gave written informed consent by returning the baseline questionnaires. The research was approved by the institutional review boards at Harvard T.H. Chan School of Public Health and Partners Healthcare.

For all patients included in this study, we retrieved tumor specimens from cancer diagnosis from the treating hospital. Expert genitourinary pathologists performed centralized histologic rereview, including Gleason grading (16), and selected high-density tumor areas (>80% tumor cell density). Tumor tissue and, if available, adjacent noncancer prostate tissue, was measured on the Affymetrix GeneChip Human Gene 1.0 ST array (Gene Expression Omnibus: GSE62872), with postprocessing as described previously (17). Transcriptome profiling included mRNA expressions of CYP27A1, SQLE, and CYP24A1.

Plasma 25-hydroxy-vitamin D [25(OH)D] from blood samples before cancer diagnosis was measured as a part of a case–control study nested within HPFS. A radioimmunosorbent assay was used, as described previously (18).

To quantify vitamin D receptor (VDR) signaling (9), its expression in the cytoplasm and membrane was stained via IHC on tissue microarrays. Using a semiautomated quantitative image analysis system, the VDR score was generated as a combination of the relative area positively stained and the intensity of staining, as described previously (19). TMPRSS2:ERG status was determined using a genomically validated ERG IHC (20).

Our analysis plan had two main parts. First, we assessed cross-sectionally, how CYP27A1 expression was associated with measures of vitamin D signaling and intratumoral cholesterol synthesis. Second, in a longitudinal analysis, we assessed the association between CYP27A1 at cancer diagnosis and the risk of lethal disease over long-term follow-up. All tests were two-sided.

To assess the associations of 25(OH)D, VDR, SQLE, CYP24A1, ERG, and CYP27A1, we used linear regression. Values for 25(OH)D were adjusted for season and batch, as described previously (18); VDR scores were adjusted for differences in mean values between tissue microarrays (19). We modeled the predictor in categories and inspected plots to assess for potential nonlinear relationships, and we calculated tests for linear trend across quartiles by modeling the category medians (for 25(OH)D and VDR) or category indices (mRNA variables) as ordinal predictors. In a sensitivity analysis, we replaced SQLE as a proxy for cholesterol synthesis activity of the tumor by a summary score of all cholesterol synthesis genes (6). This summary score was the first principal component from principal components analysis of the cholesterol synthesis genes CYP51A1, DHCR24, DHCR7, EBP, FDFT1, FDPS, GGPS1, HMGCR, HMGCS1, HSD17B7, IDI1, IDI2, LBR, LSS, MVD, MVK, NSDHL, PMVK, SC5DL, SQLE, and TM7SF2. Higher levels indicated higher expression of the cholesterol synthesis pathway, as 20 of the 21 cholesterol synthesis genes were positively loaded on this principal component.

To assess the association of CYP27A1 expression (modeled in quartiles) and lethal disease, we used logistic regression to estimate ORs and 95% confidence intervals. Models were additionally adjusted for age (linear), year of diagnosis [categorical: pre-prostate–specific antigen (PSA) era, 1982–1988; peri-PSA era, 1989–1993; PSA era, 1994–2005), smoking status (binary: current smoker vs. never/prior smoking), family history of prostate cancer in father or brothers (binary: yes/no), body mass index (categorical: <25, 25–30, >30 kg/m²), and hyperlipidemia (binary: any self-report of hyperlipidemia by the health professionals on questionnaires before cancer diagnosis vs. no such report). In separate models, we adjusted for Gleason grade (categorical: 5–6, 3+4, 4+3, 8, 9–10), statin use at cancer diagnosis (binary: yes/no), and SQLE expression (categorical: quartiles). In an exploratory analysis, we assessed the association of CYP27A1 within low and high strata of cholesterol synthesis activity defined by SQLE and the cholesterol signature, and tested for statistical interaction using likelihood ratio tests. Given its strong association with lethal disease specifically in the highest quartile (21), SQLE in the fourth quartile was considered high; the upper half of the signature was considered high.

Characteristics of 254 patients from HPFS and 150 patients from PHS at the time of prostate cancer diagnosis are shown in Table 1. Fifty-nine percent of patients had pathologically organ-confined cancers (T1/T2 N0 M0), and 59% were diagnosed in the PSA screening era. Ninety-two percent of tumor samples were from radical prostatectomy. For 202 patients, adjacent noncancerous prostate tissue was assessed. Plasma concentrations of 25(OH)D before cancer diagnosis were available for a subset of 132 patients from HPFS. VDR protein expression had been quantified for 300 patients.

Notably, CYP27A1 expression was lower in higher grade, advanced stage, and ERG-positive cancers (Table 1). Compared with Gleason grade 5–6, tumors with Gleason grade 9–10 had on average 0.73 SD lower CYP27A1 expression (95% CI, 0.38–1.08 SD; P_trend < 0.001). ERG-positive tumors had 0.27 SD lower CYP27A1 expression (95% CI, 0.06–0.47) than ERG-negative tumors. CYP27A1 expression was lower by 0.42 SD in tumors with advanced stage (based on combined clinical and pathologic stage) compared with localized tumors (95% CI, 0.06–0.79).

We assessed the association of circulating and intratumoral indicators of vitamin D signaling and CYP27A1 mRNA expression. Circulating plasma 25(OH)D was not associated with CYP27A1; the difference in 25(OH)D expression between the lowest quartile of CYP27A1 and the highest quartile was −0.8 ng/mL (95% CI, −5.2 to 3.5; P_trend = 0.71; Fig. 2A). CYP27A1 expression was also not associated with VDR expression in the tumor; the difference in VDR expression score was 0.28 SD (95% CI, −0.06 to 0.61 SD) between the lowest and the highest quartile of CYP27A1 (P_trend = 0.09; Fig. 2B). In contrast, we observed a weak positive association between CYP27A1 and the expression of the VDR target gene CYP24A1, which had a 0.36 SD (95% CI, 0.09–0.63 SD) higher expression in the highest quartile of CYP27A1 expression compared with the lowest quartile (P_trend = 0.005; Fig. 2C).

To assess the association between intratumoral cholesterol synthesis and CYP27A1, we used the second rate-limiting enzyme of cholesterol synthesis, SQLE, and a score summarizing the mRNA expression of all cholesterol synthesis enzymes as proxies. CYP27A1 was lower in tumors with higher SQLE expression; the difference in CYP27A1 between lowest and highest quartile of SQLE was −0.42 SD (95% CI, −0.69 to −0.14; P_trend = 0.002 across quartiles of SQLE; Fig. 2D). Similar results were observed when we used the summary score instead of SQLE as a proxy for cholesterol synthesis in the tumor, observing a difference in CYP27A1 between lowest and highest quartile of the score of −0.49 SD (95% CI, −0.77 to −0.22; P_trend = 0.001).

In normal prostate tissue, we also did not observe associations between CYP27A1 expression and plasma 25(OH)D and VDR expression. In contrast to tumor tissue, CYP27A1 expression and CYP24A1 expression were not associated in normal prostate tissue (difference in CYP24A1 expression between lowest and highest quartile of CYP27A1, 0.20 SD; 95% CI, −0.20 to 0.59; P_trend = 0.33), and there was no statistically significant difference in CYP27A1 expression between the lowest and highest quartiles of SQLE (difference in CYP27A1, −0.15 SD; 95% CI, −0.54 to 0.25; P_trend = 0.44).

Patients were followed a median of 15.3 years for the development of metastases or death from prostate cancer (lethal disease). Lower intratumoral CYP27A1 mRNA expression was associated with a higher risk of lethal disease over long-term follow-up in both cohorts (Table 2). In HPFS, patients with CYP27A1 mRNA expression in the lowest quartile had a 2.64-fold higher odds of lethal disease (95% CI, 1.23–5.67), compared with patients with CYP27A1 in the highest quartile. In PHS, the OR was 4.65 (95% CI, 0.92–23.5). Combining both cohorts and adjusting for additional baseline characteristics, the OR was 3.04 (95% CI, 1.46–6.33; P_trend = 0.007 across quartiles of CYP27A1). The association of CYP27A1 and lethal disease was attenuated somewhat when additionally adjusting for SQLE (OR for lowest vs. highest quartile of CYP27A1, 2.64; 95% CI, 1.24–5.62). Results were similar when adjusting for the summary score of cholesterol synthesis or when additionally adjusting for statin use at cancer diagnosis. With additional adjustment for Gleason grade, the association of CYP27A1 and lethal disease was considerably attenuated and imprecisely estimated (OR, 1.76; 95% CI, 0.75–4.17).

As expected, CYP27A1 expression in tumor-adjacent noncancerous prostate tissue was not associated with lethal disease (OR for lowest vs. highest quartile, 1.51; 95% CI, 0.66–3.44; P_trend = 0.24).

Finally, we assessed whether the association of intratumoral CYP27A1 with lethal disease differed within levels of cholesterol synthesis. The association between CYP27A1 and lethal disease did not differ when stratifying by the summary score of cholesterol synthesis (Table 2; P_interaction = 0.88), although it appeared to be slightly stronger in patients with low SQLE (P_interaction = 0.12).

In this study, we assessed regulators of CYP27A1, which synthesizes 27-hydroxycholesterol from cholesterol, and associations of CYP27A1 expression with long-term prognosis in patients with primary prostate cancer. We found CYP27A1 expression to be low in tumors that had higher expression of cholesterol synthesis enzymes including SQLE. In contrast, we did not detect strong associations between several measures of vitamin D signaling and CYP27A1. Notably, low CYP27A1 expression was associated with a higher risk of lethal disease, beyond the elevated risk associated with higher expression of the cholesterol synthesis pathway.

We observed a strong inverse relationship between CYP27A1 expression and two different measures of intratumoral cholesterol synthesis, the expression of the second rate-limiting enzyme SQLE as well as a 21-gene signature of all enzymes in the cholesterol synthesis pathway. These observations suggest that in tumors with activated cholesterol synthesis, hydroxylation of cholesterol to 27-hydroxycholesterol is inhibited, perhaps giving these rapidly dividing cells a selective advantage when more cholesterol is available, for example, cell membrane formation. Concordantly, a preclinical study found the addition of 27-hydroxycholesterol to prostate cancer cell lines and xenografts attenuated their growth and decreased the expression of SREBP2, the main transcription factor regulating cholesterol synthesis (11). However, discordant experimental results, partially using the same cell line, have been reported as well (22).

Despite assessing multiple proxies of vitamin D signaling activity, including plasma 25(OH)D concentrations, VDR protein expression in tumor tissue, and mRNA expression of the VDR target gene CYP24A1, we did not find consistent evidence that showed CYP27A1 to be strongly related to vitamin D signaling. However, these measures may not have fully captured an effect of exogenous vitamin D on CYP27A1 expression, particularly if vitamin D is 25-hydroxylated directly within prostate cells without changing plasma 25(OH)D concentrations. This 25-hydroxylation step of vitamin D has indeed been observed in a nontumor prostate cell line, in which vitamin D also induced CYP27A1 expression (23). However, the 25-hydroxylase function of CYP27A1 may not be physiologically relevant in peripheral tissues as the prostate, but rather fulfilled by the microsomal 25-hydroxylase CYP2R1 (24). Moreover, blood draws for 25(OH)D measurements preceded cancer diagnosis up to 9.8 years (median, 3.1 years). The correlation between two repeated 25(OH)D samples from HPFS participants over 3 years was relatively high (r = 0.70; ref. 25). Nevertheless, possible nondifferential misclassification of 25(OH)D at cancer diagnosis by using prediagnostic 25(OH)D may have added some degree of bias to the null. Ultimately, our observations lend no additional support to CYP27A1 expression in prostate cancer tissue being tightly controlled by vitamin D signaling. We also assessed whether TMPRSS2:ERG status was associated with CYP27A1 expression. Bidirectional influences between vitamin D signaling, including VDR and CYP24A1, and TMPRSS2:ERG have been reported in prostate cancer cell lines (26, 27), and we previously observed that ERG-positive tumors have higher VDR expression (19). In this study, we only observed a modest association between ERG status and CYP27A1 expression, and CYP24A1 expression did not differ by ERG status (data not shown).

In breast cancer, several studies found CYP27A1 expression to be higher in high-grade compared with low-grade cancers (7, 8). In prostate cancer, a relatively strong, inverse relationship with Gleason grade has been reported previously (11) and was confirmed by our data. CYP27A1 expression has also been reported to be lower in castration-resistant cancer tissue compared with tissue from castration-sensitive tumors (28). How CYP27A1 expression would be associated with risk of clinically relevant outcomes, such as metastases or cancer-related death, was unknown. Our data indicated an approximately 2.6-fold higher odds of lethal cancer among the 25% patients with the lowest CYP27A1 expression (first quartile), compared with the 25% with the highest expression (fourth quartile; Table 2). Given the tight association of CYP27A1 and Gleason grade, it is unsurprising that these estimates were attenuated considerably when additionally adjusting for Gleason grade (OR, 1.76; 95% CI, 0.75–4.17). CYP27A1 does not appear to be well suited as a prognostic marker. Our results are supported by a previous study that included a larger set of patients with prostate cancer from HPFS and found single-nucleotide polymorphisms within CYP27A1 to be associated with the risk of lethal disease (18); however, we do not know if and how these single-nucleotide polymorphisms influence CYP27A1 mRNA expression. While statistical power for interaction testing was low, we did not find that the association of CYP27A1 expression and lethal disease differed across levels of cholesterol synthesis enzyme expression (Table 2).

Our results may be informative for mechanistic studies in both prostate and breast cancer. In preclinical breast cancer models, added 27-hydroxycholesterol stimulated tumor growth, acting partially as an endogenous selective estrogen receptor modulator (SERM; refs. 7, 29, 30). Upregulated cholesterol synthesis and production of 27-hydroxycholesterol in estrogen receptor–positive breast cancer under antiestrogen therapy has been suggested as a mechanism of therapy resistance (31, 32). Consequently, it has been suggested that SERM effects of 27-hydroxycholesterol are responsible for the association of lower CYP27A1 expression with worse prognosis in breast cancer (33). This association was most pronounced in premenopausal patients with estrogen receptor–positive tumors (8). In our study of prostate cancer, we also found a moderately strong association of lower CYP27A1 expression with higher risk of lethal disease. Besides cholesterol accumulation as one potential mechanism, additional SERM-related mechanisms could be contributing. Estrogen receptor beta is expressed in at least a subset of prostate tumors (34), and future studies might need to consider 27-hydroxycholesterol or CYP27A1 when studying estrogen receptor expression in prostate cancer.

It should be noted that we measured mRNA levels of CYP27A1 and not protein expression. We are unaware of a study directly comparing mRNA and protein levels for CYP27A1 within the same patients. In a small number of breast tissue samples, CYP27A1 protein appeared to show changes in the opposite direction than CYP27A1 mRNA (8); it is unclear how CYP27A1 mRNA and CYP27A1 protein were associated on a individual-patient level. In a similarly designed study of prostate cancer tissue, CYP27A1 protein expression was lost in the tumor epithelium in contrast to normal glands, consistent with observations on the mRNA level (11). In this study, for which the Gleason grade distribution was unknown, only about a quarter of tumors were found to express CYP27A1 protein (11). If one assumed that CYP27A1 protein expression was lost in three quarters of tumors even in our study population, this could explain why risk estimates are relatively similar across the lowest three quartiles of CYP27A1 mRNA expression. An additional limitation of our study is that only a relatively small subset of patients had prediagnostic plasma samples, which may have contributed to the null results for plasma 25(OH)D.

In summary, we found low intratumoral CYP27A1 mRNA expression to be associated with higher markers of intratumoral cholesterol synthesis, higher Gleason grade, and a higher risk of lethal disease over long-term follow-up. We did not find strong and consistent associations of CYP27A1 and circulating 25(OH)D or with two measures of intratumoral vitamin D signaling. Future studies should ideally attempt to directly measure intratumoral or circulating 27-hydroxycholesterol. Interestingly, serum 27-hydroxycholesterol concentrations were decreased by atorvastatin treatment and by vitamin D supplementation in two small-scale clinical trials among patients with breast cancer (8, 35). It remains to be seen how such interventions might affect intratumoral cholesterol and 27-hydroxycholesterol levels as well as clinical outcomes for patients with prostate cancer.

P.W. Kantoff is a board member at Context Therapeutics; has ownership interest (including stock, patents, etc.) at Context Therapeutics, Tarveda, Placon, Seer, and DRGT; and is a consultant/advisory board member for Bavarian Nordic, DRGT, New England Research Institutes, Sanofi, Thermo Fisher, OncoCellMDX, Progenity, Seer, Tarveda, GE, Janssen, Merck, and Genentech/Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: N.A. Khan, K.H. Stopsack, E.L. Giovannucci, L.A. Mucci, P.W. Kantoff

Development of methodology: N.A. Khan, P.W. Kantoff

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.L. Giovannucci, L.A. Mucci

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.H. Stopsack, T. Gerke, P.W. Kantoff

Writing, review, and/or revision of the manuscript: N.A. Khan, K.H. Stopsack, E.H. Allott, T. Gerke, E.L. Giovannucci, L.A. Mucci, P.W. Kantoff

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Gerke

Study supervision: P.W. Kantoff

We would like to thank the participants and staff of the Health Professionals Follow-up Study and the Physicians’ Health Study for their valuable contributions. In particular, we would like to recognize the contributions of Liza Gazeeva, Siobhan Saint-Surin, Robert Sheahan, and Betsy Frost-Hawes. We would like to thank the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WY. The authors assume full responsibility for analyses and interpretation of these data. The Health Professionals Follow-up Study is supported by the NIH (U01 CA167552). The Physicians’ Health Study was supported by the NIH (CA097193, CA34944, CA40360, HL26490, HL34595). The Department of Defense supported K.H. Stopsack (W81XWH-18-1-0330) and P.W. Kantoff (W81XWH-14-1-0515). This research was funded in part by the Dana-Farber/Harvard Cancer Center Specialized Programs of Research Excellence program in Prostate Cancer 5P50 CA090381 and the NIH/NCI Cancer Center Support Grants P30 CA008748 and P30 CA06516. E.H. Allot was supported by an Irish Cancer Society John Fitzpatrick fellowship. K.H. Stopsack and L.A. Mucci are Prostate Cancer Foundation Young Investigators.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.