TERC polymorphisms are associated both with susceptibility to colorectal cancer and with longer telomeres

Study participants

Three sets of samples were taken from the general population or from individuals without a cancer history³¹: (1) 2489 population controls (mean age 44 years, 49% men) from the Wellcome Trust Case–Control Consortium 2 National Blood Service (NBS) cohort (http://www.wtccc.org/; (2) 968 subjects (mean age 51 years, 52% men) from the Scottish population for the COGS study; and (3) 455 UK CORGI study controls (mean age 58 years, 41% men), these being individuals with a spouse or partner with CRC, but themselves unaffected by cancer and without a family history (to second degree relative level) of colorectal neoplasia.

CRC patients were from three sources:³¹ (1) 717 cases (mean age 67 years, 64% men) from the VICTOR study, a phase III, randomised controlled trial of rofecoxib after potentially curative therapy; (2) 888 early-onset (age ≤55 years) Scottish cases (mean age 49 years, 51% men) from COGS; and (3) 552 familial CRC cases (mean age 63 years, 48% men) from CORGI. For VICTOR, 51% of CRC were stage II, all the remainder being stage III; 62% of VICTOR cases received 5-fluorouracil-based chemotherapy and 8.3% had radiotherapy. Blood samples were taken a median of 127 weeks after surgery (interquartile range 100–170). For COGS, 19% of cancers were stage I, 31% stage II, 32% stage III and 18% stage IV; 28% of COGS patients received chemotherapy and 8% radiotherapy, and blood was sampled a median of 20 weeks after diagnosis (interquartile range 12–33). As CORGI samples were collected from cancer genetics clinics, they were highly variable in terms of their sampling dates post-diagnosis; data on cancer stage and chemotherapy were similarly variable in terms of their availability.

For studies of genetic susceptibility to CRC only, we additionally analysed 496 cases (mean age 62 years, 59% men) from the QUASAR2 study of adjuvant capecitabine with or without bevacizumab in stage II/III CRC. We excluded QUASAR2 samples from the telomere length studies, because blood was generally taken from these patients during the period of chemotherapy, potentially leading to considerable confounding effects from neutropaenia and other relative or absolute changes in the leucocyte count.

A total of 1569 individuals of white European ancestry with multiple (two to 10) or ‘significant’ (>1 cm diameter or villous histology) colorectal adenomas was identified from the Adenoma Prevention with Celecoxib Trial.

All DNA was extracted from whole peripheral blood. Single nucleotide polymorphism (SNP) genotyping, quality control and imputation were performed according to standard procedures outlined in the supplementary material, available online only. All studies were performed in accordance with local research ethics committee approval (Southampton and South-West Hampshire REC (A) 06/Q1702/99).

Telomere assays and quality control

Telomere length was measured using minor modifications of Cawthon's monochrome multiplex quantitative PCR method (details in Cawthon),³² which we found to provide far greater consistency than other previously published methods; 98 samples were assayed in duplicate and for 93 of these, the telomere length measurement lay within 15% of each other and only two lay outside 25%. All samples were analysed by study of origin in 96-well plate format together with five control dilutions of a single, anonymous standard DNA, which was used to fit linear regression lines for the log(telomere) and log(HBG control) Ct values and DNA concentration for each plate. Very good fits were observed (see supplementary figure 1, available online only). For each sample on that plate, the Ct value for the control probe was used to determine the amount of DNA in the reaction from the standard line. From this value, the expected Ct for the telomere was determined. The expected telomere Ct value was then subtracted from that observed for that sample according to the standard line, to give a relative telomere measurement (CorrDCt) for each sample on each plate. In summary, CorrDCt was the measured telomere length for each sample relative to that of the anonymous standard sample; the larger CorrDCt, the greater was the measured telomere length.

Assessment of functional effects of TERC variant

HCT116 CRC cells, a near-diploid line homozygous for the A allele at rs2293607, were used in this experiment. TERC messenger RNA was PCR-amplified using primers TCCCTTTATAAGCCGACTC and ATTCATTTTGGCCGACTT, so as to encompass TERC transcripts AI825849 and AW207347. This was cloned into pcDNA3.1 and clones of identical sequence, apart from the A and G alleles at rs2293607, were obtained by site-directed mutagenesis. HCT116 cells were transfected with this construct using Lipofectamine 2000 under standard conditions. Cells were co-transfected with pEGFP-C3 so as to measure transfection efficiency using GFP expression. This confirmed high (>90%) efficiency in all cases. Cells transfected with empty pcDNA3.1 and pEGFPc3 were used as controls. Based on pilot experiments that estimated the time of maximum TERC expression, we measured telomere lengths after 48 h. At the same time, we measured total TERC expression using Taqman real-time quantitative assay Hs03297287_s1 (Applied Biosystems), with SF3B1 expression as a control. All experiments were performed in triplicate.

Search for polymorphisms associated with telomere length in leucocytes: confirmation of SNP near TERC but not TERT

Using a candidate locus approach, we identified from public databases all SNP within 50 kb of a panel of genes proposed by Mirabello et al³³ as ‘telomere biology’ candidates (see supplementary table 2, available online only). One hundred and twenty-eight tagging SNP in these regions (see supplementary table 3, available online only) had been genotyped in our cases and controls in all samples. We used real-time quantitative PCR to determine the normalised telomere length in cases and controls from each of the following sample sets (CORGI, COGS, NBS, VICTOR). Data were stratified by case/control status, sample set, age and sex. Using a threshold for association of p=10⁻⁴, approximating to a false discovery rate of 0.01, we found five SNP (rs16854453, rs10936599, rs11709840, rs1920116, rs10936603) to be associated with telomere length (table 1). These SNP were all within a region of approximately 90 kb on chromosome 3q26.2 (170,974,795–171,062,665 bases), were in strong or moderate pairwise linkage disequilibrium (LD) (see supplementary figure 2, available online only), and lay within a haplotype block that encompassed the TERC locus (chr3: 170965092–170965542). We found that they were in strong LD with a SNP (rs12696304) proximal to TERC that had previously been reported to be associated with telomere length.¹³

No other ‘telomere biology’ SNP was associated with telomere length, the strongest association (p=2.0×10⁻³) being at a SNP, rs27061, distal to TERT. The failure to find an association between TERT SNP and telomere length confirms the data of other groups^{13 34} and is of note, because SNP at this site (rs401681, rs4635969, rs2736100, rs4975616, rs2736098) have previously been associated with cancer risk³⁵; each of these five SNP or a proxy (r²∼1.0) was typed in our samples.

TERC SNP are associated with increased risk of CRC

At the same time as our study of telomere length in CRC patients, we were undertaking a genome-wide association study of CRC. The latest stage of the genome-wide association study based on 11 case–control cohorts,³¹ including several that sampled incident cases, found that rs10936599 showed a significant association with CRC risk (p=2.55×10⁻⁸, OR_{per allele} 0.93, 95% CI 0.91 to 0.96), independent of age and sex. Table 2 shows rs10936599 genotype counts in the sample sets used in this study. As rs10936599 is a synonymous SNP within an exon of myoneurin (MYNN), we had reported MYNN as the gene close to rs10936599 most likely to harbour functional variation.³¹ However, our finding that rs10936599 was associated with telomere length prompted a more detailed investigation. We initially confirmed that, although the other four TERC SNP associated with telomere length were not genotyped in all 11 case–control cohorts, each showed evidence of association with CRC risk (see supplementary table 4, available online only). As the major allele at rs10936599 was associated with both longer telomeres and CRC risk, the genetic analysis strongly suggested that, if leucocyte telomeres are a valid proxy for those in colonocytes, longer telomeres were associated with an increased risk of CRC.

rs10936599 Genotype is associated with risk of colorectal adenomas

Telomerase activity has generally been associated with carcinomas rather than their benign precursors. However, we wondered whether the effect of rs10936599 might also be mediated in normal tissue or in early tumorigenesis (eg, by effects in normal stem cells). We therefore examined whether rs10936599 alleles were associated with disease in patients with colorectal adenomas who had not developed CRC. Cases were derived from the Adenoma Prevention with Celecoxib adenoma prevention trial;³⁶ controls were from the publicly available 1958 UK Birth Cohort.³⁷ After stratifying for a family history of CRC, the major allele at rs10936599 was associated with an increased risk of adenomas (table 2; OR 0.883, 95% CI 0.80 to 0.98; χ²₁=5.30, p=0.011). The variation close to TERC thus probably acts, at least partly, at an early stage in tumorigenesis.

Evidence that the variation tagged by rs10936599 acts through telomere length

Although we thought it highly plausible that rs10936599 alleles affected CRC risk through telomere length, it was possible that the functional variation associated with rs10936599 acted through another mechanism and that the association with telomere length was an epiphenomenon. Given the potential confounding effect on telomere length of patients having CRC or receiving treatment for their disease²⁶ and in the absence of prospective cohort data, we could not test this alternative hypothesis by directly examining rs10936599 genotypes and telomere lengths in our CRC cases and controls. We therefore undertook an indirect test. Assuming that telomere length in leucocytes is a proxy for that in colonocytes, we reasoned the following: if the influence of rs10936599 genotype on CRC risk is not mediated through telomere length, the effect size of rs10936599 should not vary with telomere length; any variation in the effect size of rs10936599 with telomere length would be most likely to result from a direct effect of longer telomeres on CRC risk, for example, by pushing those with generally long telomeres over some threshold at which cancer risk was particularly increased. In order to standardise telomere lengths among the case sample sets (CORGI, VICTOR, COGS), we ranked each individual according to the length of their telomeres in that sample set. We then used logistic regression analysis to determine the effect size of rs10936599 according to the rank of the individuals in each case data set compared with all controls, stratifying by sex, age and study of origin. We initially observed that when cases with relatively short telomeres (ie, in the first and/or second quartiles) were studied, the effect of the rs10936599 genotype was not significant, whereas a significant effect was seen in the third quartile and a highly significant effect (ln(OR) −0.43, p=5.04×10⁻⁸) was seen in the fourth quartile. More generally, we tested whether rs10936599 effect size was related to ranked telomere length by regressing ln(OR) on telomere decile; a significant linear relationship was found (ln(OR) (−0.0418×decile) +0.0561; R²=0.5882; t=−3.38, p=0.010; see supplementary figure 3, available online only). While the underlying origin of this association is unclear, its existence suggests that rs10936599 directly affects CRC risk through telomere length.

Annotation and fine mapping of the 3q26.2 region

From HapMap two data (http://www.hapmap.org/), we found that rs10936599 (chr3: 170,974,795) is in strong or moderate LD with SNP within the region chr3: 170,941,263–171,012,592 that includes the genes TERC, ARPM1, MYNN and LRRC34. Using 1000genomes project data (http://www.1000genomes.org/), 44 SNP within this region were found to be correlated with rs10936599 with r²>0.5 (see supplementary table 5, available online only).

The only SNP fulfilling these LD criteria that was both within or close to TERC and had promising functional annotation was rs2293607 (chr3: 170,965,029; r²=1.0 with rs10936599, http://www.biomart.org/). The most abundant TERC transcript maps to chr3: 170,965,092–170,965,542, but two other TERC transcripts contain rs2293607 (AI825849, chr3: 170,965,004–170,965,446 and AW207347, chr3: 170,965,002–170,965,373). rs2293607 was predicted by RNAfold (http://rna.tbi.univie.ac.at/cgi-bin/RNAfold.cgi) to alter the transcript's secondary mRNA structure (see supplementary figure 4, available online only). Moreover, data from the ENCODE project (http://genome.ucsc.edu/cgi-in/hgTrackUi?hgsid=174313571&c=chr15&g=wgEncodeReg) indicated two histone marks (H3K4Me1 and H3K4Me3) in the immediate vicinity of rs2293607. The same region was also DNaseI-hypersensitive and the predicted binding site of multiple transcription factors, including NFKB, PU.1, POU2F2 and MYC (http://genome.ucsc.edu/cgi-bin/hgTrackUi?hgsid=174313571&c=chr15&g=wgEncodeReg).

In order to refine the location of the causal variation in the region around TERC, we inspected association signals between SNP genotype and telomere length and CRC susceptibility, following imputation of SNP using the 1000genomes project as a reference. In general, the association plots for the two variables mirrored each other (see supplementary figure 5, available online only), and consequently the effect sizes were very strongly correlated (t=27.6, p<0.001). In both cases, the strongest signal was found at approximately 171.05 Mb, although the associations at that site were only marginally stronger than those present in a larger region that included rs2293607 between approximately 170.90 Mb and 171.05 Mb.

In-vitro analysis of the effects of rs2293607 genotypes in TERC transcripts

As a result of the possible effects of rs2293607 on TERC RNA structure or transcription, we performed functional assays on TERC transcripts that varied only at the A (major, CRC risk, longer telomere) and G (minor, CRC protection, shorter telomere) alleles at this site. In order to examine effects in cells of large bowel origin, we used the CRC cell line HCT116 as a model system. We expressed exogenous TERC transcripts that varied only at the A/G polymorphic site, performing each experiment in triplicate and controlling for potential differences in transfection efficiency between experiments. Initially, we quantified the expression of total (endogenous and exogenous) TERC mRNA for each of the two variants in comparison with empty vector and untransfected controls. We found that after 48 h, total TERC expression was higher in cells expressing the exogenous TERC, and that cells with the A allele construct expressed higher TERC than those with the G allele (figure 1).

We then examined whether the increased TERC expression associated with the A allele had functional consequences. Although no gross differences in HCT116 morphology or behaviour were observed at the 48 h time point, we did not necessarily expect these in a transformed cell line. We therefore assayed telomere length in cells with the A and G allele transcripts. This showed that, in concordance with our hypothesis, expression of the A allele TERC transcript resulted in longer telomeres in HCT116, whereas the G allele had little effect on telomeres compared with the control cells (figure 1).

Telomere length in CRC cases versus controls

Given the prediction from our work above that CRC risk is associated with longer telomeres, yet pre-existing evidence suggests that CRC cases have shorter telomeres than controls,²⁶ we compared telomere length in our three sets of CRC cases (VICTOR, CORGI and COGS) and in the three sets of controls (CORGI, COGS and NBS). Telomere length (CorrDCt) was approximately normally distributed (see supplementary figure 6, available online only) and decreased with age (see supplementary figure 7, available online only). Incorporating age, sex and study of origin as covariates, we performed logistic regression analysis to compare case and control telomere lengths. We used NBS controls as a comparison with VICTOR because of the absence of unaffected controls from the clinical trial sample set. Contrary to expectations from our genetic data, but consistent with previous retrospective studies, we found that cases had significantly shorter telomeres than controls (mean CorrDct_cases 1.04, mean CorrDCt_controls 1.11), although there was also considerable variation between sample series (figure 2). Overall, after stratifying for age, sex, sample set and rs10936599 genotype, telomeres were significantly shorter in cases than controls (OR 0.71, 95% CI 0.56 to 0.90, z=2.85, p=4.37×10⁻³).

The CORGI and VICTOR study cases were collected months, or occasionally years, after diagnosis and treatment. We therefore compared telomere lengths in these two sets of cases with those in the COGS cases, which were collected closer to diagnosis. After controlling for age, sex and rs10936599 genotype, the COGS samples had significantly longer telomeres than the other samples (p=4.22×10⁻⁶, logistic regression analysis). We then explored possible causes for this finding. We used multiple linear regression to search within VICTOR—as this was a high-quality clinical trial dataset with full clinical and follow-up data—for associations between telomere length and the following: (1) use of chemotherapy (all 5-fluorouracil-based) or rofecoxib; (2) use of radiotherapy; (3) blood sampling time after surgery and (4) CRC stage. After controlling for age and sex, we found no association between telomere length and chemotherapy (p=0.88), rofecoxib (p=0.87), radiotherapy (p=0.053), sampling time (p=0.18) or stage (p=0.42). There was, moreover, no association between recurrence-free survival and telomere length (HR 0.94, p=0.87, Cox proportional hazards model) after correcting for stage, radiotherapy, chemotherapy, age and sex (details not shown).