Clinical Cancer Research CTRC-AACR San Antonio Breast Cancer Symposium
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
Cancer Research Clinical Cancer Research
Cancer Epidemiology Biomarkers & Prevention Molecular Cancer Therapeutics
Molecular Cancer Research Cancer Prevention Research
Cancer Prevention Journals Portal Cancer Reviews Online
Annual Meeting Education Book Meeting Abstracts Online
Clinical Cancer Research 13, 197-205, January 1, 2007. doi: 10.1158/1078-0432.CCR-06-1199
© 2007 American Association for Cancer Research
This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Supplementary Data
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Wrensch, M.
Right arrow Articles by Prados, M.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Wrensch, M.
Right arrow Articles by Prados, M.

Imaging, Diagnosis, Prognosis

Nonsynonymous Coding Single-Nucleotide Polymorphisms Spanning the Genome in Relation to Glioblastoma Survival and Age at Diagnosis

Margaret Wrensch1, Alex McMillan2, John Wiencke1, Joe Wiemels1, Karl Kelsey3, Joe Patoka1, Hywel Jones4, Victoria Carlton4, Rei Miike1, Jennette Sison1, Michelle Moghadassi1 and Michael Prados1

Authors' Affiliations: 1 Division of Neuroepidemiology, Department of Neurological Surgery, School of Medicine, University of California at San Francisco, San Francisco, California; 2 Division of Biostatistics, Department of Health Research and Policy, School of Medicine, Stanford University, Stanford, California; 3 Department of Cancer Cell Biology, School of Public Health, Harvard University, Boston, Massachusetts; and 4 Affymetrix, Santa Clara, California

Requests for reprints: Margaret Wrensch, Neuroepidemiology Division, Department of Neurological Surgery, University of California at San Francisco, 44 Page Street, Suite 503, San Francisco, CA 94102. Phone: 415-476-1979; Fax: 415-502-1787; E-mail: margaret.wrensch{at}ucsf.edu.


   Abstract
Top
 Abstract
Materials and Methods
 Results
 Discussion
 References
 
Purpose: Our aim was to discover possible inherited factors associated with glioblastoma age at diagnosis and survival. Although new genotyping technologies allow greatly expanded exploration of such factors, they pose many challenges.

Experimental Design: In this pilot study, we (a) genotyped 112 newly diagnosed glioblastoma patients ascertained through a population-based study (group 1) with the ParAllele assay panel of ~10,000 nonsynonymous coding single-nucleotide polymorphisms (SNP), (b) used several statistical and bioinformatic techniques to identify 17 SNPs potentially related to either glioblastoma age at diagnosis or survival, and (c) genotyped 16 of these SNPs using conventional PCR methods in an independent group of 195 glioblastoma patients (group 2).

Results: In group 2, only one of the 16 SNPs, rs8057643 (located on 16p13.2), was significantly associated with glioblastoma age at diagnosis (nominal P = 0.0017; Bonferroni corrected P = 0.054). Median ages at diagnosis for those with 0, 1, or 2 T alleles were 66, 57, and 59 years in group 1 and 64, 57, and 55 years in group 2 (combined P = 0.001). Furthermore, Cox regression analyses of time to death with number of T alleles adjusted for gender and patient group yielded a hazard ratio of 0.82 (95% confidence interval, 0.68-0.98; P = 0.03).

Conclusions: Although limited by a relatively small sample size, this pilot study, using well-characterized, unambiguous disease characteristics, illustrates the necessity of independent replication owing to the likelihood of false positives. Several other challenges are discussed, including attempts to incorporate information on the potential functional importance of SNPs in genome-disease association studies.

It is being increasingly shown that common gene polymorphisms affect response to cancer therapies or otherwise influence prognosis and survival (recently reviewed in refs. 1, 2). Glioma survival has been found to be associated with polymorphisms in epidermal growth factor, GSTP1 and GSTM1, HLA A*32 and HLA B*55, and GLTSCR1 S397S and ERCC2 D711D (3–6). Although none of these findings has yet been replicated and cautious interpretation is required, these studies have provided several potentially fruitful areas of discovery of genetic variation related to glioma survival (e.g., signaling pathways for growth factors, cell cycle regulators, modifiers of drug metabolism, and radiation and immune response). Potentially relevant associations of gene polymorphisms with treatment response or survival have been reported for other types of cancers (7–20). These reports substantiate the hypothesis that variants in cell cycle, DNA repair, detoxification, or immune response genes alter function and response of tumor cells to therapeutic agents (21); affect tumor aggressiveness; and/or provide variation in host defenses against the tumor. There is also evidence from the study of numerous cancers that inherited mutations or polymorphisms might influence cancer development, and such variants often lead to earlier age at diagnosis of the disease (22, 23).

Although modern genotyping technologies permit genotyping tens of thousands of polymorphisms at reasonable cost, there are significant challenges in interpreting the voluminous data to discover polymorphisms related to disease and disease outcomes (24). In this article, we report a pilot study using the ParAllele5 10,000 nonsynonymous coding single-nucleotide polymorphism (SNP) assay panel to genotype newly diagnosed glioblastoma patients ascertained through a population-based study. We used a variety of statistical analytic techniques to choose a small subset of SNP that were potentially related to either glioblastoma age at diagnosis or survival. We then genotyped an independent group of glioblastoma patients for the selected SNPs and found that one of the selected SNPs was significantly associated with glioblastoma age at diagnosis.


   Materials and Methods
Top
 Abstract
 Materials and Methods
Results
 Discussion
 References
 
Subjects. The University of California at San Francisco Committee for Human Research approved methods (including written consent forms) for both the initial subject recruitment and follow-up for survival. Individual hospital institutional review boards also approved methods for medical record abstraction.

Eligibility criteria, ascertainment, specimen collection, interview, medical record abstraction, and mortality follow-up methods for the glioma cases in group 1 have been previously described (25–27). Briefly, any adult (age >20 years) newly diagnosed with glioma (International Classification of Disease for Oncology, morphology codes 9380-9481) between August 1991 to April 1994 and May 1997 to August 1999 who resided in six San Francisco Bay Area counties at the time of diagnosis was eligible to participate. This study did not include patients with a previous glioma. Age at diagnosis was defined as the patient's age at histologic diagnosis. Potentially eligible cases were identified using a rapid case ascertainment program available through a Surveillance, Epidemiology, and End Results Program participating registry and the Northern California Cancer Center. The study group included consenting patients who were interviewed about a variety of factors, who gave written consent to obtain and review pathology specimens and records, and who were confirmed to have an eligible diagnosis by our pathology review. We have previously detailed methods for the pathology review (27, 28); briefly, tumors were obtained and classified by a specialist neuropathologist according to the WHO criteria described by Kleihues et al. (29, 30); glioblastoma corresponds to WHO grade 4. We have previously described follow-up methods for survival (27). All cases diagnosed with glioblastoma who had adequate blood specimen were eligible for genotyping with the ParAllele assay panel. We previously published information on the blood draw rates for these subjects (31). Median time from diagnosis to blood draw for genotyped subjects was 109 days.

For the replication set (group 2), glioblastoma cases were ascertained with similar methods but eligible diagnosis dates were from November 2001 to July 2003; newly diagnosed glioblastoma patients attending the University of California at San Francisco Neuro-oncology Clinic during this time period also were eligible regardless of residence in the San Francisco Bay Area counties. We followed patients by phone at 6-month intervals. The first 195 glioblastoma patients accrued during this period with adequate blood specimen were included in the replication set. Median time from diagnosis to blood draw was 80 days.

Because of the variation in genes by ancestry/ethnicity, we limited the genotyping for this pilot study to subjects who reported their ethnicity as White.

Genotyping. The ParAllele genotyping method has been previously described (32); the majority of validated nonsynonymous coding SNPs that were known at the time were included in the panel. A list of the 10,177 SNPs on the assay panel is available upon request by contacting authors H. Jones or M. Wrensch.

Of 17 SNPs selected for further study, we genotyped 16 SNPs using conventional PCR-based methods in the replication set. DNA was isolated from heparinized whole blood using Autogen DNA (AutoGen, Holliston, MA). Fourteen SNPs were genotyped using custom-designed Applied Biosystems Assays on Demand Taqman genotyping assays (ABI, Foster City, CA). Two SNPs were genotyped using single base pair extension. The 17th SNP was not genotyped because the SNP and surrounding sequence was nearly a perfect match with five other sites in the genome and attempts at specific amplification of the desired sequence failed. Amplification primers for each polymorphism are listed in Supplementary Table S1. Quality control measures include blinded analyses, replicating 10% of samples, and negative controls.

Statistical methods. For the ParAllele assay panel genotypes, we calculated two effect measures for each autosomal SNP with allele frequency >5%; these effect measures were hazard ratios (HR) from Cox regressions to time of death with number of variant alleles (0-2) adjusted for age at diagnosis and gender and ß coefficients from linear regressions of age at diagnosis with number of variant alleles adjusted for gender. We then used several analytic strategies to further examine the ParAllele genotyping data and these effect measures. These strategies included the following:

  1. Whole-genome visualization. We plotted effect measures by chromosome location and examined for areas in which there might be more associated SNPs than expected by chance. A list of the effect measures by rs numbers is available upon request by contacting author M. Wrensch.
  2. Focusing on SNPs in broadly defined chromosomal candidate regions (07p 09p 10p 10q 11p 13q 17p 19q 22q) that had previously been implicated in glioma pathogenesis to reduce the numbers of SNPs being analyzed.
  3. Analyzing all SNPs individually (using gene dose model) versus survival and age at diagnosis controlling for a false-discovery rate (FDR) of 5% with permutations. We developed code in house in R (33) adapting the approach of Tusher et al. (34) for categorical SNP array data. The approach consists of (a) focusing on identifying subsets of genes (SNPs) with suitably low false-positive rates (rather than meeting a single significance threshold); (b) defining a gene (SNP)–specific test statistic (e.g., Z scores based on HRs for survival adjusted for age and gender and ßs from linear regression for age at diagnosis adjusting for gender); and (c) estimating FDR by random permutations of the patient characteristics. The test compares the number of observed Z values of a given magnitude with the number expected in 100 random permutations of the patient characteristics, (i.e., survival time or age at diagnosis). This gives the SNPs whose associations with the characteristic under examination exceed that based on chance. In-house simulations showed that the code was accurate and the method was less conservative than Bonferroni corrections.
  4. Focusing on stop-codon SNPs because these are the most likely factors to affect protein structure and function. We chose to analyze these SNPs because stop codons should lead to truncated proteins, suggesting a functional effect. For example, some studies have found stop codon mutations to be involved in familial cancer syndromes (35–39). However, although truncation would, by definition, change the amino acid sequence, such changes would not necessarily cause measurable functional changes in protein function; it also is possible that stop codons reaching polymorphism frequencies (>5%) might have only minor functional effects or reside in superfluous proteins because these SNPs survived negative selection in the population.
  5. Conducting Random Forests (RF) (40, 41) analysis. RF analysis does not compute P values per se, but gives a measure of importance of each SNP in the classification of the association of interest. The RF method works on a bootstrap method that randomly selects a group of subjects for tree construction, then replaces them and randomly selects another group with the same numbers of subjects. In each set, there are, therefore, subjects who are "out of the bag" and the classification tree can be applied to them to estimate prediction error. For each subject, the "votes" for the prediction are counted across all trees for which that subject was out of the bag, and the subject is predicted to be in the class with the most votes. For each variable, the importance of the variable is related to how accurately it classifies subjects. Moreover, because the importance itself is a variable, it has a SD value. The relative importance of each variable is the importance score divided by the SD and is used to prioritize variables, similar to P values. We did RF analysis on effect measures from 71 autosomal stop codons. [Note that although the RF method uses linear regression on log(years survival) for the survival effect measure of association, rather than Cox regressions, this is not problematic in this sample as none of the patients was censored]. SNPs with relative importance >3 were then put in multivariate models (a Cox regression for survival and a linear regression for age at diagnosis) with other SNPs so selected and appropriate adjusting factors (age and gender for survival and gender for age at diagnosis).

After conducting these analyses, we selected 17 SNPs for evaluation in the replication set of subjects using the following criteria: (a) permutation FDR <0.10; (b) highest three HRs from survival analysis; (c) RF analysis selected SNP with importance >3; and (d) P value from subsequent Cox regression for survival or linear regression on age at diagnosis with RF selected SNPs was <0.05. For the genotyping data of the replication group, we computed effect measures in the same way as for the ParAllele SNPs and report nominal P values. Because 16 SNPs were genotyped in the replication group and we assessed two effect measures (HR and average age at diagnosis), the Bonferroni corrected P values for the replication group is 0.05/32 = 0.0016.


   Results
Top
 Abstract
 Materials and Methods
 Results
Discussion
 References
 
ParAllele genotyping. Of 164 White glioblastoma cases for whom we had blood specimens, 115 had sufficient specimen for this pilot study. Three of these 115 cases did not give satisfactory results on the ParAllele SNP assay panel, leaving 112 cases for analysis. Table 1 compares the patient characteristics (age, gender, and median survival) for the subjects who were genotyped with subjects who were part of the population-based series but not genotyped. As shown, the subjects for whom we have blood samples were generally younger and had better survival than subjects from whom blood could not be obtained. This is because, even with rapid case ascertainment, many glioblastoma patients died before they could be enrolled in the study.


View this table:
[in this window]
[in a new window]

  		Table 1. Survival estimates and patient characteristics from glioblastoma patients used in whole-genome association pilot study and replication set, San Francisco Bay Area Adult Glioma Study, 1991 to 2005

 
Among the 112 glioblastoma cases, 6,970 SNPs had allele frequencies >5%. Call rates (numbers of specimens for which the genotype for a particular SNP could be called) for the SNPs ranged from 74% to 100% with an overall average exceeding 98%. We had previously independently assessed two SNPs on the ParAllele assay panel in Dr. Kelsey's laboratory, GSTP C114T and GSTP A105G (42); results were completely concordant for GSTP C114T and one was discrepant for GSTP A105G ({kappa} = 0.99), indicating outstanding agreement.

Whole-genome visualization. Figures 1 and 2 show the whole-genome visualization for the association of the 6,970 SNPs with survival and age at diagnosis, respectively. As seen in Fig. 1, ~6.45% of SNPs had P ≤ 0.05, where 5% would be expected, indicating the possibility that the association of some SNPs with survival may not be due to chance. The graph also shows chromosomes that are likely to have SNPs influencing survival above that expected by chance. In contrast, 4.5% of 6,970 SNPs had P ≤ 0.05 for association with age at diagnosis, but the 95% confidence limit includes 5%. Please note that these analyses have not accounted for correlations among SNPs.


Figure 1
View larger version (56K):
[in this window]
[in a new window]

  		Fig. 1. Associations of SNPs with survival (adjusted for age and gender) on autosomes with allele frequencies >5%. Top, age- and gender-adjusted HRs for mortality among 112 White glioblastoma cases (group 1) for 6,970 SNPs on autosomes with allele frequencies >5% ordered by position on the chromosomes. Dark circles (n = 449), SNPs with P < 0.05; clear circles, SNPs with P > 0.05. Bottom, percentages and 95% confidence limits of SNPs with P < 0.05 overall and per chromosome; dotted line, 5% expected under the null hypothesis of no true positives overall; dashed line, overall observed 6.4% of SNPs with P < 0.05; small bracket farthest to the left, 95% confidence limit excludes 5%. Note that this analysis does not account for correlations among SNPs.

 

Figure 2
View larger version (72K):
[in this window]
[in a new window]

  		Fig. 2. Associations of SNPs with age at diagnosis (adjusted for gender) on autosomes with allele frequencies >5%. Top, gender-adjusted age at diagnosis differences among 112 White glioblastoma cases (group 1) for 6,970 SNPs on autosomes with allele frequencies >5% ordered by position on the chromosomes. Dark circles (n = 315), SNPs with P < 0.05; clear circles, SNPs with P > 0.05. Bottom, percentages and 95% confidence limits of SNPs with P < 0.05 overall and per chromosome; dotted line, 5% expected under the null hypothesis of no true positives overall; dashed line, overall observed 4.6% of SNPs with P < 0.05; small bracket farthest to the left, 95% confidence limit includes 5%. Note that this analysis does not account for correlations among SNPs.

 
Selecting SNPs for replication. Table 2 lists the National Center for Biotechnology Information rs numbers, target alleles, and gene names for the 17 SNPs selected for replication by the various methods presented below along with the effect measures and nominal P values.


View this table:
[in this window]
[in a new window]

  		Table 2. Description of 17 SNPs selected from ParAllele 10,000 nonsynonymous coding SNP assay panel as candidates for association with glioblastoma survival or age at diagnosis in 112 glioblastoma patients (group 1), San Francisco Bay Area Adult Glioma Study, 1991 to 2005

 
Permutation FDR on all 6,970 SNPs with allele frequencies >5% and on 1,473 SNPs in broadly defined chromosomal regions. Using these two sets of SNPs for the two outcomes, case-survival and age at diagnosis, gave four analytic sets. In this small pilot study, none of these sets revealed SNP disease characteristic associations with FDRs <5%. However, one such analysis yielded a FDR of 10% for three SNPs; these three SNPs were selected for replication. Figure 3 shows the permutation results of observed versus expected Z scores based on regression coefficients of age at diagnosis by gene dose, controlling for gender for SNPs within the broadly defined candidate chromosomal regions (07p 09p 10p 10q 11p 13q 17p 19q 22q).


Figure 3
View larger version (10K):
[in this window]
[in a new window]

  		Fig. 3. Quantiles of observed versus expected Z scores from linear regression of age at case diagnosis with 1,473 SNPs in broadly defined chromosomal candidate regions; the top three SNPs have a FDR of 0.10.

 
RF analyses using 71 stop codon SNPs. For 71 stop codons, we compared the performance of RF with the permutation FDR method described above for detecting potential associations with case survival and age at diagnosis and survival. To summarize the results, the smallest FDRs detected for the 71 stop codons were 0.12 for survival and 0.49 for age at diagnosis. For the RF method, 2,000 trees were estimated and SNPs that were selected had a relative importance >3 (corresponding to an importance >3 SDs greater than the average importance scores for all SNPs). Seven SNPs were found to have relative importance >3 for RF on log(survival years) and another seven SNPs had relative importance >3 for RF on age at diagnosis. All seven SNPs that RF selected as related to glioblastoma multiforme survival were nominally statistically significantly associated (P ≤ 0.05) with survival in multivariate Cox regression. Four of seven SNPs that RF selected as related to glioblastoma age at diagnosis were nominally statistically significantly associated with age at diagnosis in multivariate linear regression, P < 0.05. Thus, the RF analyses followed by either multivariate Cox regression on survival and linear regression on age at diagnosis yielded 11 SNPs for further study.

Because the whole-genome visualization analysis suggested that the associations of some of the SNPs with survival might have exceeded chance expectation, we also included the three SNPs with the highest HRs for survival in the replication study.

Replication set. Table 1 shows the patient characteristics for the 195 glioblastoma multiforme patients used for replication in relation to the population- and clinic-based groups from which they were chosen. We chose the first consecutive 195 subjects with blood specimens for the genotyping. As shown in Table 3 , the genotype frequencies for all 16 SNPs for the 112 glioblastoma multiforme patients genotyped with the ParAllele assay panel and for the 195 glioblastoma multiforme patients genotyped using Taqman or single base extension methods were very similar.


View this table:
[in this window]
[in a new window]

  		Table 3. Genotype frequencies (and percentages) for 17 SNPs in two independent sets of glioblastoma cases, the San Francisco Bay Area Adult Glioma Study (1991-2005)

 
None of the SNPs selected were associated with survival (all P values >0.10), but one SNP, rs8057643, was associated with age at diagnosis (P = 0.0017), giving a Bonferroni-corrected P value of 0.05 (0.0017x32). (See Supplementary Tables S2 and S3 for comparisons of medians of survival and ages at diagnosis for individuals by genotype of each of the 16 SNPs for the two sets of glioblastoma subjects.) The median ages at diagnosis by number of variant alleles in rs8057643 were similar for the two groups of subjects. For 0, 1, or 2 T alleles, respectively, the median ages at diagnosis were 66, 57, and 59 years for the 112 patients genotyped with the ParAllele platform and were 64, 57, and 55 years for the 195 patients genotyped with Taqman. Given this similarity, we did a combined analysis adjusted for gender and group, and found P = 0.001 for association of age at diagnosis with number of variant alleles in rs8057643. The same P values were obtained for the comparison of age at diagnosis for those with 1 or 2 versus no T alleles. Also, note that the direction of the association of median survival times with number of T alleles for the two sets of glioblastoma patients is similar (Supplementary Table S3); the combined HR with left truncation for number of T alleles adjusted for age, gender, and group (ParAllele set or replication set) is 0.82 (0.68-0.98; P = 0.03).


   Discussion
Top
 Abstract
 Materials and Methods
 Results
 Discussion
References
 
We studied a very large number of nonsynonymous coding SNPs in this pilot study to test the feasibility of this approach to understand the relationship of inherited genetic variation to glioma survival and age at diagnosis. This pilot study was conducted on a limited number of patients (group 1) and resulted in a large number of potentially associated SNPs. However, only a small subset survived a vetting procedure based on FDRs, RF analyses, and analyses of genomically focused regional SNP sets. When this subset of 16 SNPs was genotyped in the replication set (group 2), only a single one was significantly associated with age at diagnosis and none was associated with survival using stringent criteria for multiple comparisons. This SNP was selected for replication because of an association with survival in group 1; the association of this SNP with age at diagnosis was not significant in group 1. However, the observed trend of the association of variant alleles in this SNP with age at diagnosis was somewhat similar in the two groups of patients studied.

This pilot study points out many of the issues encountered in studying large numbers of SNPs in relation to disease characteristics. An essential problem is that of false-positive associations and the corollary of how to discover, among the many positive results, which are likely to be real; or as some put it, how to discover the signal over the noise. Replication in an independent set of subjects is necessary to help decide which associations are unlikely to be due to chance (24, 43, 44).

The choice of test and replication sets is likely to be critical. Clearly, it would be desirable to have had a larger set of subjects for the initial large-scale genotyping. Also, it is difficult to come up with a truly identical population for replication. In our sample, the replication set had somewhat better survival owing to clinic- and population-based versus population-based recruitment only; there also was a shorter interval between diagnosis and recruitment in the replication set. However, even with the modest sample of subjects available and the dissimilarities in the two sets of glioblastoma patients, all 16 identified SNPs had nearly identical genotype frequencies. Despite having some limitations, this pilot study is strengthened by having well-characterized subjects with clear criteria for study inclusion, including a well-defined disease entity (newly diagnosed, histologically confirmed glioblastoma) and unambiguous disease characteristics, survival, and age at diagnosis. Furthermore, a SNP was identified in the replication group with somewhat similar results between both groups for associations with age at diagnosis and survival.

This SNP, rs8057643, was included in the assay panel because it was thought to be a nonsynonymous coding SNP in the hypothetical gene LOC400496 located on chromosome 16p13.2 with existence of the gene supported by the existence of mRNA AL162011. The T allele was thought to code for termination (a stop codon) and the C allele represents a glutamine (45). Information on this hypothetical gene and SNPs within it has recently been discontinued on the National Center for Biotechnology Information Web site. Now, the SNP is described only as being in the 5' end of the A2BP1 (ataxin-2–binding protein 1) gene, which exists in four alternative isoforms. For three of these isoforms, rs8057643 is located before the transcription start site and may affect expression. For the fourth isoform, rs8057643 is located in intron 3, which is before the first coding exon (exon 4). The possibility exists that this SNP is in strong linkage disequilibrium with other functional SNPs, although there are no confirmed nonsynonymous coding SNPs within A2BP1 (45). A2BP1 has an RNP motif that is highly conserved among RNA-binding proteins. This protein binds to the COOH terminus of ataxin-2 and may contribute to the restricted pathology of spinocerebellar ataxia type 2 (SCA2). Ataxin-2 is the gene product of the SCA2 gene, which causes familial neurodegenerative diseases. Ataxin-2–binding protein 1 and ataxin-2 are both localized in the trans-Golgi network. There were two other SNPs in the general vicinity of rs8057643 (at position 6,900,691; HR, 0.71; P = 0.03; age at diagnosis ß = –2.51; P = 0.18) on the ParAllele assay panel; these were rs11077108 at position 6,900,674 (HR, 1.38; P = 0.03; age at diagnosis ß = 2.63; P = 0.16) and rs8056547 at position 6,900,777 (HR, 1.03; P = 0.79; age at diagnosis ß = 2.28; P = 0.16; the next nearest SNPs were at positions 5,355,972 and 8,747,455). Thus, the coefficients for association with age at glioblastoma diagnosis for the two nearby SNPs were very similar, but were in the opposite direction from rs8057643. The association with glioblastoma survival was very similar for rs11077108 (but in the opposite direction), but not for rs8056547. Rs11077108 and rs8056547 also were included on the ParAllele assay panel because they were thought to be nonsynonymous coding SNPs in LOC400496 with the rs11077108 A allele giving a histidine and the G allele giving an arginine, and rs8056547 with the C allele giving a histidine and the G allele giving a glutamine. It is important to note that these SNPs may not be functionally related to glioma age at diagnosis or survival; however, if the associations are replicated in additional studies, it may indicate linkage to a SNP associated with these disease characteristics.

Although we used multiple strategies to attempt to identify important SNPs, each strategy has limitations. For example, with respect to the whole-genome visualization, it is important to point out that multiple SNPs showing importance in a single chromosomal region does not necessarily imply multiple important SNPs in that region because SNPs near each other may be in strong linkage disequilibrium, creating the appearance of runs of associated SNPs. Some important SNPs will not be in linkage disequilibrium with nearby SNPs and will not appear as runs. In addition, there may be important SNPs within the large gaps between the nonsynonymous coding SNPs assessed in this preliminary analysis that we have missed.

There also are some limitations in the choice of genotyping panel for this pilot study. The nonsynonymous coding SNP assay panel lacked promoter or gene expression SNPs. Furthermore, the state of knowledge of the human genome is incomplete and rapidly changing. Many more nonsynonymous coding SNPs have been identified since the assay panel used in this study was created, and there are no tools yet to unambiguously categorize or quantify the functional significance of SNPs so that they could be weighted in statistical analyses. Furthermore, there is variability in the genotype call rates for SNPs on this panel. The lower the call rate, the more likely subjects would be to be misclassified. The effect of such misclassification, assuming no systematic bias with disease characteristics, would be to reduce the effect measures toward the null. Moreover, as noted above, the one SNP that showed possibly statistically significant associations with survival in one patient group and age at diagnosis in the other is no longer classified as a nonsynonymous coding SNP.

Other SNPs on the ParAllele assay panel may be truly associated with glioma survival and age at diagnosis. These SNPs may be discovered through generation of additional candidates by computational methods that take into consideration the functional effect of individual SNPs based on known functional assays or computational prediction methods such as GenMAPP (46–48), LS-SNP (University of California at San Francisco (46–50), PolyPhen (51), and SIFT (52, 53) followed by validation on an independent set of subjects. We also believe it to be advantageous to more clearly define candidate SNPs and regions before instituting large-scale genotyping interrogations to limit the number of false positives. Thus, we plan to develop a nonsynonymous SNP assay panel enriched for regions and genes thought to be associated with gliomagenesis, and then genotype these on a larger number of subjects initially. This would be followed by genotyping a replication set that includes a larger number of candidate SNPs than were possible in this pilot study.


   Footnotes
 
Grant support: Accelerate Brain Cancer Cure; the National Brain Tumor Foundation; NIH grants CA52689, CA097257, CA89032, ES06717, and ES04705; and Robert J. and Helen H. Glaser Family Foundation.

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.

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

5 http://www.affymetrix.com/corporate/parallele.affx. Back

Received 5/17/06; revised 9/14/06; accepted 10/ 6/06.


   References
Top
 Abstract
 Materials and Methods
 Results
 Discussion
 References
 

  1. Loktionov A. Common gene polymorphisms, cancer progression and prognosis. Cancer Lett 2004;208:1–33.[CrossRef][Medline]
  2. Nagasubramanian R, Innocenti F, Ratain MJ. Pharmacogenetics in cancer treatment. Annu Rev Med 2003;54:437–52. Epub 2001 December 3.[CrossRef][Medline]
  3. Bhowmick DA, Zhuang Z, Wait SD, Weil RJ. A functional polymorphism in the EGF gene is found with increased frequency in glioblastoma multiforme patients and is associated with more aggressive disease. Cancer Res 2004;64:1220–3.[Abstract/Free Full Text]
  4. Okcu MF, Selvan M, Wang LE, et al. Glutathione S-transferase polymorphisms and survival in primary malignant glioma. Clin Cancer Res 2004;10:2618–25.[Abstract/Free Full Text]
  5. Tang J, Shao W, Dorak MT, et al. Positive and negative associations of human leukocyte antigen variants with the onset and prognosis of adult glioblastoma multiforme. Cancer Epidemiol Biomarkers Prev 2005;14:2040–4.[Abstract/Free Full Text]
  6. Yang P, Kollmeyer TM, Buckner K, Bamlet W, Ballman KV, Jenkins RB. Polymorphisms in GLTSCR1 and ERCC2 are associated with the development of oligodendrogliomas. Cancer 2005;103:2363–72.[CrossRef][Medline]
  7. Angele S, Romestaing P, Moullan N, et al. ATM haplotypes and cellular response to DNA damage: association with breast cancer risk and clinical radiosensitivity. Cancer Res 2003;63:8717–25.[Abstract/Free Full Text]
  8. DeMichele A, Martin AM, Mick R, et al. Interleukin-6 –174G->C polymorphism is associated with improved outcome in high-risk breast cancer. Cancer Res 2003;63:8051–6.[Abstract/Free Full Text]
  9. Iacopetta B, Grieu F, Joseph D. The –174 G/C gene polymorphism in interleukin-6 is associated with an aggressive breast cancer phenotype. Br J Cancer 2004;90:419–22.[CrossRef][Medline]
  10. Hefler LA, Grimm C, Ackermann S, et al. An interleukin-6 gene promoter polymorphism influences the biological phenotype of ovarian cancer. Cancer Res 2003;63:3066–8.[Abstract/Free Full Text]
  11. Shu XO, Gao YT, Cai Q, et al. Genetic polymorphisms in the TGF-ß1 gene and breast cancer survival: a report from the Shanghai Breast Cancer Study. Cancer Res 2004;64:836–9.[Abstract/Free Full Text]
  12. Neben K, Mytilineos J, Moehler TM, et al. Polymorphisms of the tumor necrosis factor-{alpha} gene promoter predict for outcome after thalidomide therapy in relapsed and refractory multiple myeloma. Blood 2002;100:2263–5.[Abstract/Free Full Text]
  13. Dasgupta RK, Adamson PJ, Davies FE, et al. Polymorphic variation in GSTP1 modulates outcome following therapy for multiple myeloma. Blood 2003;102:2345–50. Epub 003 Jun 5.[Abstract/Free Full Text]
  14. Illmer T, Schuler US, Thiede C, et al. MDR1 gene polymorphisms affect therapy outcome in acute myeloid leukemia patients. Cancer Res 2002;62:4955–62.[Abstract/Free Full Text]
  15. Morimoto Y, Ozaki T, Ouchida M, et al. Single nucleotide polymorphism in fibroblast growth factor receptor 4 at codon 388 is associated with prognosis in high-grade soft tissue sarcoma. Cancer 2003;98:2245–50.[CrossRef][Medline]
  16. Chen J, Hunter DJ, Stampfer MJ, et al. Polymorphism in the thymidylate synthase promoter enhancer region modifies the risk and survival of colorectal cancer. Cancer Epidemiol Biomarkers Prev 2003;12:958–62.[Abstract/Free Full Text]
  17. Sakano S, Berggren P, Kumar R, et al. Clinical course of bladder neoplasms and single nucleotide polymorphisms in the CDKN2A gene. Int J Cancer 2003;104:98–103.[CrossRef][Medline]
  18. Wang YC, Chen CY, Chen SK, Chang YY, Lin P. p53 codon 72 polymorphism in Taiwanese lung cancer patients: association with lung cancer susceptibility and prognosis. Clin Cancer Res 1999;5:129–34.[Abstract/Free Full Text]
  19. Wang YC, Lee HS, Chen SK, Chang YY, Chen CY. Prognostic significance of p53 codon 72 polymorphism in lung carcinomas. Eur J Cancer 1999;35:226–30.[CrossRef][Medline]
  20. Hsia TC, Chiang HC, Chiang D, Hang LW, Tsai FJ, Chen WC. Prediction of survival in surgical unresectable lung cancer by artificial neural networks including genetic polymorphisms and clinical parameters. J Clin Lab Anal 2003;17:229–34.[CrossRef][Medline]
  21. Sullivan A, Syed N, Gasco M, et al. Polymorphism in wild-type p53 modulates response to chemotherapy in vitro and in vivo. Oncogene 2004;23:3328–37.[CrossRef][Medline]
  22. Strong L, Amos C. Inherited susceptibility. In: Schottenfeld D, Fraumeni JF, editors. Cancer epidemiology and prevention. 2nd ed. New York: Oxford University Press; 1996. p. 559–83.
  23. Bale AE, Brown SJ. Etiology of cancer: cancer genetics. In: DeVita VT, Hellman S, Rosenberg S, editors. Cancer principles and practice of oncology. 6th ed. Philadelphia: Lippincott; 2001. p. 207–17.
  24. Botstein D, Risch N. Discovering genotypes underlying human phenotypes: past successes for mendelian disease, future approaches for complex disease. Nat Genet 2003;33:228–37.
  25. Wiemels JL, Wiencke JK, Sison JD, Miike R, McMillan A, Wrensch M. History of allergies among adults with glioma and controls. Int J Cancer 2002;98:609–15.[CrossRef][Medline]
  26. Wrensch M, Lee M, Miike R, et al. Familial and personal medical history of cancer and nervous system conditions among adults with glioma and controls. Am J Epidemiol 1997;145:581–93.[Abstract/Free Full Text]
  27. Wrensch M, Rice T, Miike R, et al. Diagnostic, treatment and demographic factors influencing survival in population based glioma patients in the San Francisco Bay Area. Neuro-Oncology 2006;8:12–26.[Abstract/Free Full Text]
  28. Aldape K, Simmons ML, Davis RL, et al. Discrepancies in diagnoses of neuroepithelial neoplasms: The San Francisco Bay Area Adult Glioma Study. Cancer 2000;88:2342–9.[CrossRef][Medline]
  29. Kleihues P, Burger PC, Scheithauer BW. Histological typing of tumors of the central nervous system. 2nd ed. In: Springer Verlag, editors. World Health Organization international histological classification of tumors. Berlin: Springer Verlag; 1993.
  30. Kleihues P, Burger PC, Scheithauer BW. The new WHO classification of brain tumours. Brain Pathol 1993;3:255–68.[Medline]
  31. Wrensch M, Wiencke JK, Wiemels J, et al. Serum IgE, tumor epidermal growth factor receptor expression, and inherited polymorphisms associated with glioma survival. Cancer Res 2006;66:4531–41.[Abstract/Free Full Text]
  32. Hardenbol P, Yu F, Belmont J, et al. Highly multiplexed molecular inversion probe genotyping: over 10,000 targeted SNPs genotyped in a single tube assay. Genome Res 2005;15:269–75.[Abstract/Free Full Text]
  33. R Development Core Team. R: a language and environment for statistical computing; 2005 [cited]. Available from: http://www.r-project.org/.
  34. Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci U S A 2001;98:5116–21.[Abstract/Free Full Text]
  35. Luce MC, Marra G, Chauhan DP, et al. In vitro transcription/translation assay for the screening of hMLH1 and hMSH2 mutations in familial colon cancer. Gastroenterology 1995;109:1368–74.[CrossRef][Medline]
  36. Lin X, Choi JH, Lynch P, Xi L, Wu E, Frazier ML. Reduction in hMSH2 mRNA levels by premature translation termination: implications for mutation screening in hereditary nonpolyposis colorectal cancer. Dig Dis Sci 1999;44:553–9.[CrossRef][Medline]
  37. Ballhausen WG. Genetic testing for familial adenomatous polyposis. Ann N Y Acad Sci 2000;910:36–47; discussion 49.[Abstract/Free Full Text]
  38. Origone P, Bellini C, Sambarino D, et al. Neurofibromatosis type 1 (NF1): identification of eight unreported mutations in NF1 gene in Italian patients [corrected]. Hum Mutat 2003;22:179–80.[Medline]
  39. Upadhyaya M, Osborn M, Cooper DN. Detection of NF1 mutations utilizing the protein truncation test (PTT). Methods Mol Biol 2003;217:315–27.[Medline]
  40. Breiman L, Cutler A. Random Forests; 2005. Available from: http://www.stat.berkeley.edu/users/breiman/RandomForests/.
  41. Lunetta KL, Hayward LB, Segal J, Van Eerdewegh P. Screening large-scale association study data: exploiting interactions using random forests. BMC Genet 2004;5:32.[CrossRef][Medline]
  42. Wrensch M, Kelsey KT, Liu M, et al. Glutathione S-transferase variants and adult glioma. Cancer Epidemiol Biomarkers Prev 2004;13:461–7.[Abstract/Free Full Text]
  43. Au WW, Salama SA. Use of biomarkers to elucidate genetic susceptibility to cancer. Environ Mol Mutagen 2005;45:222–8.[CrossRef][Medline]
  44. Pharoah PD, Dunning AM, Ponder BA, Easton DF. Association studies for finding cancer-susceptibility genetic variants. Nat Rev Cancer 2004;4:850–60.[CrossRef][Medline]
  45. National Center for Biotechnology Information. Entrez Gene. c2005 and 2006. Available from: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?CMD=search&DB=gene).
  46. Gladstone Institutes. GenMAPP: gene map annotator and pathway profiler. San Francisco (CA): Gladstone Institutes; 2005. Available from: http://www.GenMapp.org.
  47. Doniger SW, Salomonis N, Dahlquist KD, Vranizan K, Lawlor SC, Conklin BR. MAPPFinder: using Gene Ontology and GenMAPP to create a global gene-expression profile from microarray data. Genome Biol 2003;4:R7. Epub 2003 Jan 6.[CrossRef][Medline]
  48. Dahlquist KD, Salomonis N, Vranizan K, Lawlor SC, Conklin BR. GenMAPP, a new tool for viewing and analyzing microarray data on biological pathways. Nat Genet 2002;31:19–20.[CrossRef][Medline]
  49. Karchin R, Diekhans M, Kelly L, et al. LS-SNP: large-scale annotation of coding non-synonymous SNPs based on multiple information sources. Bioinformatics 2005;12:12.
  50. Karchin R, Kelly L, Sali A. Improving functional annotation of non-synonomous SNPs with information theory. Pac Symp Biocomput 2005:397–408.
  51. Bork P. PolyPhen: prediction of functional effect of human nsSNPs. 2005 [cited]. Available from: http://tux.embl-heidelberg.de/ramensky/.
  52. Ng PC, Henikoff S. Sorting intolerant from tolerant; 2005. Available from: http://blocks.fhcrc.org/sift/SIFT.html.
  53. Ng PC, Henikoff S. Accounting for human polymorphisms predicted to affect protein function. Genome Res 2002;12:436–46.[Abstract/Free Full Text]



This article has been cited by other articles:


Home page
 Cancer Epidemiol. Biomarkers Prev.Home page
J. S. Chang, R.-F. Yeh, J. K. Wiencke, J. L. Wiemels, I. Smirnov, A. R. Pico, T. Tihan, J. Patoka, R. Miike, J. D. Sison, et al.
Pathway Analysis of Single-Nucleotide Polymorphisms Potentially Associated with Glioblastoma Multiforme Susceptibility Using Random Forests
Cancer Epidemiol. Biomarkers Prev., June 1, 2008; 17(6): 1368 - 1373.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Supplementary Data
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Wrensch, M.
Right arrow Articles by Prados, M.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Wrensch, M.
Right arrow Articles by Prados, M.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
Cancer Research Clinical Cancer Research
Cancer Epidemiology Biomarkers & Prevention Molecular Cancer Therapeutics
Molecular Cancer Research Cancer Prevention Research
Cancer Prevention Journals Portal Cancer Reviews Online
Annual Meeting Education Book Meeting Abstracts Online