Pre-diagnosis leukocyte telomere length and risk of ovarian cancer

Abstract

Background

The associations between telomere length and cancer risk are equivocal, and none have examined the association between pre-diagnosis leukocyte telomere length and the risk of developing ovarian cancer.

Methods

We prospectively measured leukocyte telomere length (LTL) collected from 442 ovarian cancer cases and 727 controls in the Nurses’ Health Studies and the Northern Sweden Health and Disease Study. Cases were matched to one or two controls on age, menopausal status, and date of blood collection. Odds ratios (OR) and 95% confidence intervals (CI) were estimated using conditional logistic regression.

Results

LTL was measured a median of 9.5 years before ovarian cancer diagnosis among cases. We observed a decreased risk of ovarian cancer with longer LTL. In multivariable models, women in the top quartile of LTL had an OR for ovarian cancer of 0.67 (95% CI=0.46–0.97) compared to those in the bottom quartile. Inverse associations were stronger for non-serous cases (OR_{quartile 4 vs. quartile 1 of LTL}=0.55, 95%CI=0.33–0.94) and rapidly fatal cases (i.e. cases who died within 3 years of diagnosis) (OR_{quartile 4 vs. quartile 1 of LTL}=0.55, 95%CI=0.32–0.95).

Conclusion

Our prospective findings suggest that longer circulating LTL may be associated with a lower ovarian cancer risk, especially for non-serous and rapidly fatal cases. The evaluation of LTL in relation to ovarian cancer risk by tumor subtypes is warranted in larger prospective studies.

Impact

Pre-diagnosis LTL may reflect an early event in the ovarian cancer development and could serve as a biomarker to predict future risk.

Introduction

Risk of ovarian cancer, the fifth leading cause of cancer death among U.S. women (1), is hypothesized to increase with greater number of lifetime ovulations. Ovulation-induced trauma to the ovarian surface epithelium generates reactive oxidants (2), a local inflammatory response, and stimulates the epithelium proliferation, leading to an accumulation of genetic errors that may augment ovarian cancer risk (3). Telomeres, which protect the physical integrity of linear chromosomes (4), are shortened with each cell division (5), a process that may be accelerated by damage incurred by oxidative stress (6). Tissue-based studies reveal patterns of telomere shortening, genomic instability, and upregulated telomerase expression for many tumor types, including ovarian cancer, as cells progress from noninvasive precursor lesions to cancer, implicating telomere shortening as a common event early in malignant transformation (7–10).

Three retrospective epidemiologic studies have explored the associations between telomere length and ovarian cancer risk and reported mixed results. In a small pilot study, Polish women in the shortest versus longest tertile of relative leukocyte telomere length (LTL) had a 3-fold increased risk of serous ovarian carcinoma compared to cancer-free controls (odds ratio [OR]=3.4, 95% confidence interval [CI] = 1.5–7.5), with the strongest association observed among women diagnosed with poorly differentiated tumors (11). A subsequent study observed a weaker, but significant association between shorter LTL and higher ovarian cancer risk (12). In contrast, the largest study to date conducted within the New England Case-Control Study, did not find evidence of an association between LTL and ovarian cancer risk overall or by histologic subtype; nevertheless, an association in the expected direction (inverse) emerged but was not statistically significant when cases who had recently been treated with chemotherapy were excluded (13). As blood samples were collected after ovarian cancer diagnosis in these retrospective studies, telomere shortening may have occurred as a result of the physiological changes stemming from the disease itself, cancer treatment, and/or the psychological impact of a cancer diagnosis. It remains inconclusive as to whether telomere shortening precedes development of ovarian cancer. Therefore, we prospectively investigated whether LTL from pre-diagnosis blood samples was associated with ovarian cancer risk using data from the Nurses’ Health Study (NHS), NHSII, and Northern Sweden Health and Disease Study (NSHDS).

Materials and Methods

Study population

NHS/NHSII

The NHS cohort was established in 1976 among 121,700 U.S. female registered nurses aged 30–55 years, and the NHSII began in 1989 among 116,430 female registered nurses aged 25–42 years. All women completed an initial questionnaire and have been followed biennially by questionnaires to update their demographics, lifestyle, and medical history. In 1989 to 1990, 32,826 NHS participants (aged 43–69 years) provided blood samples and completed a short questionnaire (14); follow-up was 93% in 2010. Between 1996–1999, 29,611 NHSII participants (aged 32–54 years) provided blood samples (18,521 eligible premenopausal women provided a sample timed in the luteal phase of the menstrual cycle) and completed a short questionnaire (15); follow-up was 95% in 2011. Plasma, buffy coat, and red blood cell aliquots have been stored in liquid nitrogen freezers since collection. These studies were approved by the Committee on the Use of Human Subjects in Research at the Brigham and Women’s Hospital.

Ovarian cancers were identified via self-report on questionnaires or from death certificates, and then confirmed by medical record review or linkage to the relevant tumor registry. Cases had no previous history of cancer, except non-melanoma skin cancer, before blood collection and were diagnosed with primary invasive epithelial ovarian or peritoneal cancer after blood draw and before June 1, 2012 (NHS) or June 1, 2011 (NHSII). Cases were matched to one or two controls, who were alive and had at least one intact ovary at the time of the case diagnosis, on menopausal status at baseline and diagnosis (premenopausal, post-menopausal, unknown), age (±1 year), month of blood collection (±1 month), time of day of blood draw (±2 hours), fasting status (>8 hrs, ≤8 hrs), and postmenopausal hormone use at blood draw (yes, no). For NHSII cases with timed samples, we additionally matched on day of the luteal blood draw (date of next menstrual cycle minus date of blood draw, ±1 day).

NSHDS

The NSHDS cohort consists of three sub-cohorts (the Västerbotten Intervention Programme (VIP) cohort, the Monitoring Trends and Determinants in Cardiovascular Disease (MONICA) cohort, and the mammary (mammography) screening cohort), and was established in 1985. At recruitment, participants provided venous blood samples that were stored at −80°C (16). Information on demographic factors, lifestyle factors (including exogenous hormone use and smoking), reproductive factors, and medical history were collected at the time of recruitment and/or through follow-up questionnaires.

Cases were identified through the cancer registry as women with primary invasive epithelial ovarian cancer diagnosed after blood donation, who had no preceding invasive cancer diagnosis (except non-melanoma skin cancer), did not use exogenous hormones at the time of blood donation and who were diagnosed before December 2013. A pathology report review was carried out by a gynecologic pathologist. One control who was alive and free of cancer, who did not report a bilateral oophorectomy, and did not use exogenous hormones at blood donation at case diagnosis was matched to each case on sub-cohort, menopausal status (premenopausal, postmenopausal, unknown), age (±6 months) and date of blood donation (±3 months).

Telomere assay

Genomic DNA was extracted from the buffy coat fraction of peripheral blood using QIAmp DNA blood kits (QIAGEN, Venlo, Netherlands). We used Real-Time Quantitative PCR (qPCR) to determine average relative LTL (17, 18). The assay determined the copy-number ratio between telomere repeats and a single-copy (36B4) reference gene (T/S Ratio, -dCt). Relative LTL was reported as the exponentiated T/S ratio corrected for a reference sample. A modified version of the qPCR telomere assay was performed in a 384-well format with a 7900HT PCR System (Life Technologies, Carlsbad, CA). Briefly, 5ng of buffy-coat derived genomic DNA was dried down in a 384-well plate and suspended in 10µL of either the telomere or 36B4 reaction mixture for 2 hours at 4°C. The telomere reaction mixture consisted of 1× Quantitect SYBR Green Master Mix (Qiagen, Venlo, Netherlands), 2.5mM of DTT, 270nM of Tel-1 primer-(GGTTTTTGAGGGTGAGGGTGAGGGTGAGGGTGAGGGT), and 900nM of Tel-2 primer-(TCCCGACTATCCCTATCCCTATCCCTATCCCTATCCCTA). The 36B4 reaction consisted of 1× Quantitect SYBR Green Master Mix, 300nM of 36B4U primer (CAGCAAGTGGGAA GGTGTAATCC), and 500nM of 36B4D primer-(CCCATTCTATCATCAACG GGTACAA). All samples for both the telomere and 36B4 reactions were performed in triplicate on different plates. Each 384-well plate contained a 6-point standard curve from 0.625ng to 20ng to assess PCR efficiency. A slope of −3.33+/−0.3 (>90% pcr efficiency) for the standard curve of both the telomere and 36B4 reactions was considered acceptable. Quality control samples was interspersed throughout the plates in order to assess inter-plate and intra-plate variability of Ct values. Mean coefficients of variation (CV) for the exponential T/S ratio of blinded QC samples ranged from 7.8–17.6% across batches. The correlation between T/S ratios and absolute telomere lengths determined by southern blot was 0.82 (P<0.001) (18). Samples with failed qPCR data (n=26) and those with a within-triplicate CVs greater than 20% (n=52) were removed for final analyses.

Statistical analysis

Relative leukocyte telomere length values were z-transformed to improve normality within batches. To control for variation across laboratory batches, we used the batch correction method proposed by Rosner et al. (19) adjusting for age, BMI, postmenopausal hormone use, smoking status and seasons of blood draw to obtain a batch-adjusted LTL in each cohort, with values in NHS as the reference batch. Cohort-specific quartiles of the batch-adjusted z-scores were determined based on the LTL distributions in the controls in each cohort. Demographics, reproductive factors, and clinical characteristics at blood draw were estimated across cohorts for cases and matched controls.

Conditional logistic regression was used to calculate ORs and 95% CIs across cohort-specific quartiles of LTL, with a higher quartile indicating longer LTL. In each cohort, LTL was inversely associated with ovarian cancer risk but the associations did not reach statistical significance (Supplementary table 1). Heterogeneity across cohorts was assessed using random effects meta-analysis techniques (20). There was little evidence for heterogeneity across cohorts (P for heterogeneity=0.78). Hence we pooled NHS, NHSII and NSHDS data, and re-determined batch-corrected LTL quartiles using the control distribution in the pooled study. Spearman correlation coefficients were performed between age at blood draw and LTL among controls. As expected, LTL was inversely correlated with age at blood draw (r_Spearman=−0.12, P=0.008) among control participants (Supplementary Figure 1). We used conditional logistic regression to estimate the ORs and 95%CIs across cohort-common quartiles of LTL conditioning on matching factors. In multivariable models, we adjusted for oral contraceptive use (ever vs. never), tubal ligation (yes vs. no), family history of ovarian or breast cancer (yes vs, no), parity (nulliparous, 1 child, 2 children, 3 children, 4+ children), smoking status (never smoker, former smoker, current smoker, missing), and BMI at blood draw (kg/m², continuous). We also modeled LTL as a continuous measure (per one standard deviation [s.d.]).

In secondary analyses, we evaluated whether associations were stronger for certain tumor subtypes, i.e., serous vs. non-serous cases, and tumors that were rapidly fatal (i.e., case died within 3 years of diagnosis) vs. less aggressive, using polytomous logistic regression. We assessed whether associations were modified by menopausal status (premenopausal vs. postmenopausal), age at blood draw (<55, 55–65, >65 years), smoking status (never smoking vs. ever smoking) and time interval between blood draw and diagnosis (<9.5 years vs. ≥9.5 years). Interaction terms were created by multiplying the variables above with indicators of quartiles of LTL. The statistical significance of the interaction was assessed using likelihood ratio tests. In addition, we also repeated all the analyses by applying the age-adjusted LTL using residual methods (21) given the inverse correlation between LTL and age at blood draw. All P-values were two sided and analyses were conducted using SAS release 9.4 (SAS Institute, Cary, NC, USA).

Results

The final samples size in the pooled analysis consisted of 442 cases and 727 controls. Demographic and reproductive factors were similar among cases and controls across three cohorts (Table 1). Among cases, the mean age at diagnosis was 68.4 years in NHS, 50.9 years in NHSII and 60.1 years in NSHDS. The mean time between blood collection and cancer diagnosis among cases was 11.6 years in NHS, 6.2 years in NHSII, and 7.7 years in NSHDS.

Table 1.

Characteristics of ovarian cancer cases and controls in NHS, NHSII and NSHDS at the time of blood collection^a

	NHS		NHSII		NSHDS
	Case (n=246)	Control (n=493)	Case (n=51)	Control (n=92)	Case (n=145)	Control (n=142)
Age at blood draw, y, mean (SD)	56.9 (6.5)	56.9 (6.5)	44.8 (4.6)	44.7 (4.8)	52.4 (9.2)	53.0 (9.3)
Age at diagnosis, y, mean (SD)	68.4 (8.3)	-	50.9 (5.7)	-	60.1 (9.3)	-
Time between blood draw and diagnosis, y, mean (SD)	11.6 (6.2)	-	6.2 (4.3)	-	7.7 (5.5)	-
BMI at blood draw, kg/m2, mean (SD)	28.1 (22.2)	28.7 (23.3)	32.9 (25.7)	28.4 (16.7)	25.9 (4.6)	24.7 (3.5)
Fasting status, %	63	67	63	72	63	57
Menopausal status at blood draw, %
Premenopausal	22	21	81	78	27	22
Postmenopausal	65	65	11	10	59	49
Missing	13	15	8	12	14	29
Use of post-menopausal hormones at blood draw, %	43	41	9	12	7	10
Oral contraceptive use, %	45	45	88	88	41	40
Number of children, %
Nulliparous	7	4	29	18	27	20
1 child	4	4	18	10	14	12
2 children	32	28	38	43	37	39
3 children	32	28	9	19	15	19
4+ children	26	37	6	10	7	10
Tubal ligation, %	15	17	16	30	5	3
Family history of ovarian or breast cancer, %	16	12	12	14	10	13
Smoking status at blood draw, %
Never	46	48	64	75	33	33
Former	42	39	31	15	20	21
Current	12	14	5	10	21	14
Missing	0	0	0	0	26	31
Histology subtypes, Nc
Serous/Non-serous	158/72	-	25/24	-	79/53	-
Rapidly fatal/non-fatal	86/103	-	7/32	-	38/93	-
Relative leukocyte telomere length, median [10th – 90th percentile]b
Values on original scale	0.47 [0.32–0.66]	0.48 [0.33–0.67]	0.55 [0.32–0.81]	0.57 [0.37–0.89]	0.46 [0.31–0.66]	0.46 [0.33–0.69]
Values after batch correction	0.47 [0.32–0.66]	0.48 [0.33–0.67]	0.49 [0.27–0.73]	0.50 [0.29–0.81]	0.48 [0.33–0.68]	0.48 [0.36–0.72]

^aValues are means (SD) or percentages and are standardized to the age distribution of the study population.

^bBatch correction method proposed by Rosner et al. (19) was used to control for variation across laboratory batches, adjusting for age, BMI, postmenopausal hormone use, smoking status and seasons at blood draw, with values in NHS as reference.

^cCases with undetermined histology subtypes were deleted for the subtype analyses in each cohort.

We observed an inverse association between LTL and ovarian cancer risk after pooling (Table 2). Women in the top quartile of LTL had an OR for ovarian cancer of 0.67 compared to those in the bottom quartile (95%CI= 0.46–0.97) with a borderline significant trend across quartiles (P for trend=0.07). When modeling LTL continuously, a one s.d. increase in LTL was significantly associated with 11% decreased risk of developing ovarian cancer (OR=0.89, 95%CI=0.78–1.01, P =0.04). Further adjusting for potential confounders did not considerably change the results.

Table 2.

Odds ratio and 95% confidence interval of ovarian cancer in three pooled nested-case control studies (NHSI, NHSII, NSHDS) according to common quartiles of circulating relative leukocyte telomere length^a

Relative leukocyte telomere length Quartile (median level)b	Case	Control	Model 1c	Model 2d
			OR (95%CI)	OR (95%CI)
Q1 (0.34)	122	179	1.00	1.00
Q2 (0.44)	111	184	0.83 (0.57, 1.22)	0.81 (0.55, 1.20)
Q3 (0.53)	114	183	0.83 (0.58, 1.18)	0.84 (0.58, 1.21)
Q4 (0.67)	95	181	0.68 (0.47, 0.97)	0.67 (0.46, 0.97)
P-trende			0.06	0.07
1 s.d. increase (0.14)f	442	727	0.88 (0.77, 0.99)	0.89 (0.78, 1.01)
P value			0.04	0.07

OR: odds ratio; CI: confidence interval

^aCommon quartiles of circulating relative leukocyte telomere length were obtained by using the control distribution in the pooled three cohorts. There was no heterogeneity across three studies so we pooled all data.

^bRelative leukocyte telomere length was z-scored and batch-corrected.

^cConditional logistic regression model conditioned on matching factors. Cases were matched to one or two controls on age, menopausal status, and date of blood collection.

^dConditional logistic model conditioned on matching factors and adjusted for oral contraceptive use (yes vs. no), tubal ligation (yes vs. no), family history of ovarian or breast cancer (yes vs. no), parity (nulliparous, 1 child, 2 children, 3 children, 4+ children), smoking status (never smoke, former smoker, current smoker, missing), and BMI at blood draw (kg/m², continuous).

^eP-trend was calculated by modeling the median of each category as a continuous term. All statistical tests were two-sided.

^fWe modeled relative leukocyte telomere length as a continuous variable (per one standard deviation [s.d.] increase). One s.d. of relative leukocyte telomere length =0.14.

When we further evaluated potential differences by ovarian cancer subtype, longer LTL was significantly related to lower risk of non-serous ovarian cancer and rapidly fatal cases that died within three years of diagnosis (Table 3). However, there was no significant heterogeneity by subtype. The ORs comparing extreme quartiles of LTL were 0.55 (95%CI: 0.33–0.94) for non-serous cases (P for trend =0.02) and 0.81 (95%CI: 0.53–1.24) for serous cases (P for trend=0.50) (P-heterogeneity=0.17). Additionally, the ORs were 0.55 (95%CI: 0.32–0.95) among rapidly fatal cases (P for trend=0.02) and 0.77 (95%CI: 0.49–1.23) among less aggressive ovarian cancer (P for trend=0.37) (P-heterogeneity = 0.18), comparing highest vs. lowest quartile of LTL. Furthermore, the inverse association between LTL and ovarian cancer was not significantly modified by menopausal status (P for interaction=0.27), age at diagnosis (P for interaction=0.20), smoking status (P for interaction=0.73), or time interval between blood draw and diagnosis (P for interaction=0.97). In addition, to more carefully account for the association of age with LTL, we obtained an age-adjusted LTL by residual method to model the association and observed similar results. The OR comparing extreme quartiles of LTL for ovarian cancer was 0.62 (95%CI: 0.43–0.90), similar to conventionally adjusted analyses.

Table 3.

Odds ratio and 95% confidence interval of ovarian cancer risk according to the common quartiles of circulating relative leukocyte telomere length in three pooled nested case-control studies by histologic subtype and rapidly fatal versus less aggressive disease ^a,^b

Relative leukocyte telomere length	Serous/poorly differentiated (n=262) OR (95%CI)	Non-serous (n=149) OR (95%CI)	P heterogeneityc
Quartile
Q1	1.00	1.00
Q2	0.90 (0.60, 1.36)	0.74 (0.45, 1.20)	0.52
Q3	1.14 (0.77, 1.69)	0.64 (0.38, 1.07)	0.08
Q4	0.81 (0.53, 1.24)	0.55 (0.33, 0.94)	0.27
P-trendd	0.50	0.02	0.17
1 s.d. increasee	0.98 (0.85, 1.13)	0.79 (0.65, 0.95)
P value	0.77	0.01	0.07
	Rapidly fatalf (n=131) OR (95%CI)	Less aggressive (n=228) OR (95%CI)
Quartile (median)
Q1 (0.34)	1.00	1.00
Q2 (0.44)	0.68 (0.41, 1.12)	1.01 (0.65, 1.56)	0.24
Q3 (0.53)	0.49 (0.28, 0.85)	1.22 (0.79, 1.86)	0.01
Q4 (0.68)	0.55 (0.32, 0.95)	0.77 (0.49, 1.23)	0.34
P-trendd	0.02	0.37	0.18
1 s.d. increasee	0.79 (0.64, 0.97)	0.93 (0.80, 1.09)
P value	0.02	0.37	0.20

OR: odds ratio; CI: confidence interval

^aCommon quartiles of circulating relative leukocyte telomere length were obtained by using the control distribution in the pooled three cohorts. There was no heterogeneity across three studies so we pooled all data. Relative leukocyte telomere length was z-scored and batch-corrected.

^bPolytomous logistic regression model conditioned on matching factors (age, menopausal status, and date of blood collection) and adjusted for oral contraceptive use (yes vs. no), tubal ligation (yes vs. no), family history of ovarian or breast cancer (yes vs. no), parity (nulliparous, 1 child, 2 children, 3 children, 4+ children), smoking status (never smoke, former smoker, current smoker, missing), and BMI at blood draw (kg/m², continuous).

^cP for heterogeneity was obtained by likelihood ratio test.

^dP-trend was calculated by modeling the median of each category as a continuous term. All statistical tests were two-sided.

^eWe modeled relative leukocyte telomere length as a continuous variable (per one standard deviation [s.d.] increase). One s.d. of relative leukocyte telomere length =0.14.

^fRapidly fatal cases are defined as those that died within three years of diagnosis.

Discussion

In this first prospective nested case-control study of ovarian cancer, longer LTL was associated with a lower risk of developing ovarian cancer. This association was not modified by menopausal status, age at diagnosis, smoking status or time interval between blood draw and diagnosis. Moreover, stronger inverse associations were observed for risk of non-serous and rapidly fatal ovarian cancer.

Inconsistent associations have been reported between telomere length and various cancers. For some studies, longer circulating telomere length was associated with lower risk of several cancers, including lung cancer (22–24), colorectal cancer (25), and breast cancer (26, 27), with stronger associations reported in retrospective studies with LTL measured after cancer onset. Other studies have documented an increased risk with longer telomere length for lung cancer (28, 29), melanoma (30), pancreatic cancer (31), breast cancer (32, 33), and prostate cancer (34), primarily among prospective studies. Null results have also been reported (35, 36). Greater cancer risk with shorter telomere length is biologically plausible, given evidence that shortened telomeres can play a causal role in carcinogenesis by instigating chromosomal instability, promoting genetic lesions, inactivating tumor suppressor checkpoints, and ultimately inducing cancer (10, 37, 38). However, there are also reasonable hypotheses for the alternate scenario, as cells with longer telomeres might be at higher risk of acquiring genetic abnormality because having longer telomeres may delay cellular senescence (39, 40). Furthermore, recent genome wide association studies have revealed bi-directional associations between genetic determinants of telomere length and different cancers (41, 42). Overall, the association between telomere length and cancer risk may vary by cancer site and may depend on other characteristics of the tumors at those sites (e.g., amount of genomic instability).

For ovarian cancer specifically, three retrospective case-control studies have examined circulating telomere length in relation to ovarian cancer risk, with two reporting an inverse association (11, 12) and one with null findings (13). However, the latter study did observe a suggestive association between genetic variation in the TERT gene and the risk (13). Due to their retrospective design, the estimates in these studies might be biased, because telomere shortening may occur after diagnosis, potentially due to treatment or disease processes. In contrast, our study associated risk of developing ovarian cancer with pre-diagnosis telomere length in blood collected years before diagnosis (mean 6.2–11.6 years across the studies), and observed an inverse association between LTL and overall ovarian cancer risk. These findings, with the advantage of prospective design, though generally consistent with previous retrospective investigations, may support the hypothesis that circulating telomere length can predict ovarian cancer risk. Tissue studies indicated that telomere shortening may be a critical early event in ovarian cancer development. Compared to normal tubal epithelium, progressively shorter telomeres have been observed in tubo-ovarian dysplasia (TOD) and serous tubal intraepithelial carcinoma (STIC), the putative precursor to high-grade serous carcinomas (HGSC) (43, 44). Furthermore, the number and size of chromosomal aberrations increased from TOD to STIC to HGSC, suggesting that genetic instability may be an early alteration in ovarian carcinogenesis (43, 44). Ovarian cancers, particularly HGSCs, are characterized by p53 mutations, a deregulated pRb pathway, and a high degree of genomic instability (45), features consistent with the telomere dysfunction hypothesis of carcinogenesis (46). Nevertheless, despite tissue evidence of shorter telomere length in serous tumors, we did not find a clear association between LTL and serous ovarian cancer risk; the association was stronger for non-serous tumors, although the number of cases with these tumors were limited. We also further conducted cross-classification with histology subtypes and found that (after excluding undetermined subtypes) among rapidly fatal cases there were 90 serous cases and 34 non-serous cases; while the numbers were 129 and 100 respectively among less aggressive cases. Moreover, we explored associations for high-grade vs. low-grade serous cases but did not find any associations. While these findings were limited by modest sample sizes of the various histologic types, telomerase reactivation and immortalization has been identified in high-grade serous ovarian tumor cells, in which longer and shorter telomeres coexisted in the same tumors (47). In addition, telomere length measured in leukocytes might not be the optimal surrogate for pre-diagnosis telomere length in ovarian and tubal tissue; nevertheless, telomere length does appear to be highly correlated across a variety of tissues within the same individual (48). Furthermore, although it is unclear whether our observed association among non-serous cases represents true biologic differences by tumor subtype, we recently reported that key risk factors exhibited significant heterogeneity by histology (49). Notably, most established ovarian cancer risk factors were more strongly associated with non-serous versus serous subtypes. Our observations that the associations of telomere length with ovarian cancer risk differ by histologic type, along with our prior publication on differences by traditional risk factors, add to the growing evidence that ovarian cancer is a highly heterogeneous disease and that evaluations by subtype are necessary for identifying novel risk factors. This is particularly important for developing risk prediction models, as, to date, such models have not taken differences by histologic type into account. As a result, future evaluations of telomere length by tumor subtypes in larger studies are warranted.

Interestingly, genetic research has been mixed with respect to variants in the TERT gene as well as variants associated with telomere length in relation to ovarian cancer risk. Several SNPs in TERT and its promoter have been associated with ovarian cancer risk, particularly with the serous subtype (13, 50–52), although one study noted that the SNPs associated with ovarian cancer risk were not associated with telomere length (52). Conversely, more recent studies using Mendelian randomization provided no evidence of a relationship between telomere-associated SNPs and overall ovarian cancer as well as three histologic subtypes (41, 42). Therefore, we cannot rule out the possibility that these associations in our study are due to chance or may reflect another exposure that can alter telomere length. For example, depression, which has been associated with reduced telomere length (53, 54), was recently associated with an increased risk of ovarian cancer (55). Strengths and limitations of this study are worth careful consideration. On one hand, the study used a prospective design, multiple independent cohorts, rigorous case-control matching, and rich covariate information to adjust for potential confounders. Nevertheless, as discussed above, peripheral blood leukocyte telomere length might not be an adequate surrogate for telomere length in ovaries or fallopian tubes given the dynamic immune system. We had a limited number of specific histologic subtypes, which reduced the precision of those relative risk estimates. The lack of racial/ethnic backgrounds might is another limitation.

In conclusion, our prospective findings indicate that longer LTL may be associated with a lower ovarian cancer risk, particularly for rapidly fatal disease and non-serous histology. These findings suggest that pre-diagnosis LTL may reflect an early event in the ovarian cancer development and could serve as a biomarker to predict future risk. Given the significant findings in this first prospective study, additional research to replicate these results is warranted, particularly to examine associations by tumor subtype. Confirmation of telomere length as a risk biomarker for ovarian cancer could have implications for improving identification of women at high risk of ovarian cancer.