Measurement methods and accuracy in copy number variation: failure to replicate associations of beta-defensin copy number with Crohn's disease

Abstract

The copy number variation in beta-defensin genes on human chromosome 8 has been proposed to underlie susceptibility to inflammatory disorders, but presents considerable challenges for accurate typing on the scale required for adequately powered case-control studies. In this work, we have used accurate methods of copy number typing based on the paralogue ratio test (PRT) to assess beta-defensin copy number in more than 1500 UK DNA samples including more than 1000 cases of Crohn's disease. A subset of 625 samples was typed using both PRT-based methods and standard real-time PCR methods, from which direct comparisons highlight potentially serious shortcomings of a real-time PCR assay for typing this variant. Comparing our PRT-based results with two previous studies based only on real-time PCR, we find no evidence to support the reported association of Crohn's disease with either low or high beta-defensin copy number; furthermore, it is noteworthy that there are disagreements between different studies on the observed frequency distribution of copy number states among European controls. We suggest safeguards to be adopted in assessing and reporting the accuracy of copy number measurement, with particular emphasis on integer clustering of results, to avoid reporting of spurious associations in future case-control studies.

Figure 1.

The copy variable beta-defensin genomic region showing the locations in the repeat sampled by the PRT-based ‘triplex’ test. The approximate genomic extent of the seven copy variable beta-defensin genes DEFB4, DEFB103–DEFB107 and SPAG11 is shown, without the detail of intron–exon boundaries. Below this is shown an example result from ‘triplex’ PRT-based typing of copy number for beta-defensins. PRT assays were designed to specifically amplify DNA sequences that occur within the beta-defensin (DEFB) CNV repeat unit but have additional paralogues elsewhere in the genome. The primers are designed to amplify specifically from exactly two loci of the repeat sequence (one of which is at DEFB), with corresponding products distinguished by length. For example, HSPD21-PRT primers specifically amplify only the chr8 (DEFB) and chr21 copies of a repeat element, with products differing by 8 bp. The third assay (rs5889219) examines a triallelic indel polymorphism (123/126/128 bp) that can have different sizes in individual DEFB repeats. This trace shows a sample with a copy number of 4, reflected in the ratios between the ‘test’ (=DEFB) and ‘reference’ peaks in the two PRT systems HSPD21 and PRT107A, and in the approximate 1:1:2 ratios of alleles (consistent with a total of four copies) for the multi-allelic indel rs5889219. Each analysis is in turn performed in two parallel but independent replicate amplifications using different fluorescent labels which are combined before capillary electrophoresis. Data from the two different fluorescent labellings for each system are combined before further analysis. This combined analysis therefore returns six results useful in measurement of beta-defensin copy number: duplicates of two independent PRT-based estimates of copy number and duplicate ratios of allelic products for the multi-allelic indel.

Figure 2.

Comparison of unrounded copy number measurements based on the HSPD21 and PRT107A PRT tests for (A) 779 cases and controls from the London collection and (B) 659 cases and controls from the Edinburgh collection, showing predominant clustering of measurements—most frequently centred around integer values 2, 3, 4, 5, 6 or 7. See also ‘Materials and Methods’ for details of procedures and data analysis.

Figure 3.

Distribution of PRT-based results for (A) 833 London and (B) 672 Edinburgh cases and controls, showing the mean of HSPD21 and PRT107A measurements, or (if one PRT failed) the single PRT recorded, with clear peaks around integer values. (C) Real-time PCR data from 821 Edinburgh samples analysed by a ΔΔCT method shows no such clustering, suggesting a wider range of measurement error.

Figure 4.

Real-time PCR data (A) comparing two different methods of analysing real-time PCR data, indicating no serious discrepancy between analysis methods but no obvious clustering around integer values. (B) Real-time and PRT-based measurements on the same samples are compared. The PRT-based data show grouping around the integer values, but are spread vertically in this presentation by the much greater variation in measurements resulting from real-time PCR.

Figure 5.

Frequency distributions of integer copy number classes for (A) 648 cases and 185 controls from the London collection using the PRT-based triplex test (P = 0.381), (B) 358 cases and 314 controls from the Edinburgh collection by the triplex test (P = 0.066) and (C) 407 cases and 411 controls from the Edinburgh collection using real-time PCR (analysed by a ΔΔCT method: P = 0.094).