With an average 3.7
years of follow-up, 494 recurrences or deaths were observed among 1,056 successfully genotyped invasive epithelial ovarian cancer cases with follow-up for disease outcome. As shown in Table , 62% of cases were of serous histology, 71% were diagnosed at advanced stage, and the majority of subjects were enrolled within 4
months of diagnosis.
Clinical characteristics of invasive epithelial ovarian cancer cases.
Analysis of individual normalized marker intensities followed by combination of results across multiple consecutive markers is the most powerful approach for the detection of associations with small, common CNVs. Two regions showed suggestive association with ovarian cancer survival at multiple markers [smoothed −log10
) >2]. On 14q31.3 (379
kb, 63 markers) a smoothed p
0.001 was observed suggesting a modest regional association with survival. As shown in Figure A, normalized intensities at rs2274736, a non-synonymous SNP in PTPN21
, alone appeared to be driving the regional association. In fact, this SNP was the most significant single marker in genome-wide analysis (p
); note, however that it did not reach traditional genome-wide significance. Genomic segmentation in 14q31.3 was then done to identify specific gains or loss among study participants; however, only two samples were detected with gains and 18 samples with losses. Due to the lack of called CNVs in this region, further analysis was not carried out in 14q31.3.
Figure 1 Association between ovarian cancer survival and normalized intensities at individual markers (black dots) and smoothed regional association (red line) in (A) 14q31 and (B) 22q13; Genomic Build 36. Analysis adjusted for study site, age at diagnosis, and (more ...)
A second region with suggestive association with cancer survival at multiple markers was on 22q13.31 centered at rs2285164 (smoothed p
kb, 160 markers; Figure B). Genomic segmentation of 22q13 was then computed to identify specific gains or loss among study participants (Inc, 2008
), where 35 showed gain and 154 showed loss. Association testing of gain (N
35), normal (N
867), or loss (N
154) with survival did not reveal association (p
0.29 for two degrees of freedom test; treating CNV as a categorical variable; p
0.67 for one degree of freedom trend test treating CVN as a continuous variable). Thus, even though a signal was observed for association from the single marker analysis on 22q13, CNV calling, and subsequent analysis of this region showed no association between CNV and overall survival (Figure ).
Figure 2 Kaplan–Meier plot of the 22q13 CNV association with ovarian cancer survival. The different lines represent the survival curve for subjects with a “loss,” “gain,” or “normal” for the genomic segment. (more ...)
An inverse approach analyzing pre-defined regions of CNV change is most powerful for the detection of associations with large, rare CNVs. Genome-wide CBS identified 564 regions with variable copy number among the study population, including 78 regions with gain and 486 regions with loss (available upon request). Association testing of these regions revealed 14 regions with p
-values <0.05, including one region of gain and 13 regions of loss (Table ). Results at the most statistically significant regions (p
0.002) suggested that loss of a region on 9p23 was associated with poorer survival (HR 1.44, 95% CI 1.14–1.81) as was gain of a region on 15q22.31 (HR 1.34, 95% CI 1.11–1.61). However, no region was statistically significant after correction for multiple testing using a Bonferroni procedure (564 tests).
Because the overall amount of variation from normal copy number across the genome (CNV burden) may contribute to disease, CNV burden for each case was estimated as summarized in Table . There was no association between survival and number of gains (p
0.42), number of losses (p
0.94), or total number of gains and losses (p
Mean genome-wide CNV burden by vital status and association with overall survival.