The current study provides evidence that susceptibility variants identified in published glioma GWAS differ among GBM, lower grade astrocytic tumors and oligodendrogliomas and mixed tumors. While these categories are broad, they define tumors with different clinical behavior and/or arising from cells with distinct functions in the CNS [9
] making distinguishing patterns in the genomic background plausible. Results have implications for interpreting published GWAS and for planning future GWAS which these data suggest should consider potential heterogeneity according to recognized histopathological subtypes of glioma.
In the present study, variants in CCDC26
were associated primarily with astrocytic and oligodendroglial tumors. CCDC26
rs4295627 provided the strongest signal in the Shete GWAS [4
]. Consistent with the GWAS of Wrensch et al. [3
] we did not detect association for CCDC26
rs4295627 or 4 other variants in CCDC26
also linked to risk in the Shete GWAS in a combined analysis. However, we were able to detect strong associations among astrocytic tumors and tumors with an oligodendroglial component. This finding helps reconcile results of the 2 GWAS as the study by Wrensch et al. was limited to high-grade astrocytic tumors (84% GBM and 16% anaplastic astrocytomas). The SNP rs498872 which maps to the 5′UTR of PHLDB1
provided the fifth-strongest signal in the Shete GWAS [4
]. This variant was also not detected by Wrensch et al. and in the current study, an association was limited to lower grade astrocytomas and oligodendrogliomas. Taken together, the findings suggest that both CCDC26
play a limited role in the genesis of primary GBM.
In these data, variants in TERT, RTEL
were primarily a feature of astrocytic tumors and GBM. TERT
rs4977756 and RTEL1
rs6010620 provided the second, third and fourth strongest signal, respectively, in the GWAS of Shete et al. [4
] and were concordantly identified in the GWAS of Wrensch et al. [3
] (although the TERT
SNP was detected only in the discovery (AGS) data in the latter GWAS). In the current study, TERT
rs2736100 was significantly associated only with GBM whereas RTEL1
rs6010620 was more strongly associated with the lower grade astrocytic tumors. Variants in CDKN2B
were most prominent in GBM although odds ratios were nonsignificantly elevated also for astrocytic tumors. The suggestion from these data that genetic variation in TERT, RTEL
play a more prominent role in tumors with astrocyte lineage should be explored in larger studies.
We examined association of the variants in these genes (listed in ) with mortality rates in our series of GBM patients and could demonstrate no significant associations. A recent study [10
] suggested that CCDC26
rs10464870 and rs891835 and RTEL1
rs2297440 and rs6010620 predict long-term survival (≥36 months) in GBM. We failed to detect overall association of these variants with GBM mortality. Too few deaths occurred more than 36 months after diagnosis (a total of 5 deaths among 9 patients which survived that duration) for a meaningful analysis according to survival time.
In addition to top-hit variants in these 5 genes, we detected associations in 2 intergenic SNPs identified the Shete GWAS (rs1384847 and rs7300686) [4
]. Both of these SNPs were significant in the discovery GWA but not the replication studies. Neither variant was in linkage dis-equilibrium with any implicated SNP on chromosomes 4 and 12, respectively, in the Wrensch GWAS (not shown) [3
]. Furthermore, there was no evidence in our data that associations involving these variants were confined to lower grade astrocytic tumors and/or oligodendrogliomas. Three variants identified uniquely in the Wrensch GWAS [3
] were not associated with risk in current study, overall, or among high-grade tumors. Further research is needed to establish the relevance of these SNPs in gliomagenesis.
We examined SNPs identified in GWAS of other cancers published up to Summer 2009 and found modest associations in several variants, although none remained significant after adjustment for multiple comparisons. An exception is TERT
rs2736100 which was identified in both glioma GWAS and has also been implicated in GWAS of testicular germ cell tumors [11
] and lung cancer [12
]. An intronic SNP (rs402710) located in CLPTM1L
, identified in lung cancer GWAS [14
], was marginally associated with risk in the current data (Ptrend
= 0.039) before adjustment; this SNP is uncorrelated with TERT
rs2736100 and TERT
rs2853676 (not shown). The CLPTM1L
SNP was marginally associated with GBM mortality in the current data (HR: 1.40; 95% CI 1.07–1.84; Ptrend
= 0.015) whereas the TERT
SNPs had no association with mortality. The SNP rs6983267 near POU5F1P1
on chromosome 8 is multicancer susceptibility marker linked to prostate [15
], colon [16
], and a range of other cancers [17
]. This variant showed no association with glioma in the present series. Two breast cancer variants in FGFR2
(rs1219648 and rs2981582) [18
] were each marginally associated with glioma risk before adjustment; that amplification of FGFR2
has been noted in GBM [19
] suggests these variants may have a causal role in glioma and should be examined in larger studies.
Strengths of the current study include the relatively large sample size for a study of a rare tumor such as glioma, pathologic confirmation of all cases, and the limited potential influence of survival bias given exceptionally rapid enrollment of cases. However, the study had several limitations. Our study had limited power to detect associations with rare variants and those with modest relative risks, and the study size only permitted us to examine associations according to broad histological strata of glioma. Our classification of tumors by histology was based on the diagnostic pathology reports and some misclassification was possible [20
]. We note, however, that our results are consistent with those in a recent report that showed distinct patterns of association for the five most prominent SNPs in the Shete GWAS according to WHO grade of glioma [21
]. As in the current study, risk alleles for TERT
rs2736100 were strongly correlated with the diagnosis of GBM whereas carrier frequencies of risk alleles of CCDC26
rs4295627 and PHLDB1
rs498872 were inversely correlated with GBM histology (RTEL1
rs6010620 had a stronger association with GBM in that study). In both series, CDKN2B
rs4977756 risk allele frequency did not vary strongly by histology in astrocyte lineage tumors. Concordance of results in the two studies suggests that classification in the current study was reasonably accurate at least among the astrocytic tumors. Finally, we could not evaluate effects of genotypes by race as ~98% of subjects enrolled in the study were Caucasian. Racial admixture among Caucasians in the study was not evaluated (approximately 93% of cases and 95% of controls reported European ancestry in one or both parents) and may have diluted some associations.
Rapid advances have recently been made in understanding genes and signaling networks involved in the incidence and pathogenesis of glioma [22
]. The current results suggest that stratification by broad histological subgroups has the potential to uncover novel risk loci and may shed additional light on genetic susceptibility to glioma in future studies.