The results of this study demonstrate that applying a minimum threshold for the ratio of double-stranded to total DNA quantity can improve the likelihood of successful genotyping for buccal DNA samples. Furthermore, applying this minimal level of quality control results in buccal genotyping results that are only slightly lower than the gold standard of blood DNA. This study also suggests specific additional considerations that may further improve results from buccal samples. Overall, we have shown that buccal samples not only can be successfully genotyped, but also provide high quality analytic data suitable for large-scale or high-throughput genetic studies.
The major finding of this study is that it is important to monitor ds/total DNA ratio to identify buccal samples at high likelihood of failure. In this study, low ds/total DNA ratio was a significant independent predictor of pre-genotyping failure, assuming acceptable 260/280 and 260/230 ratios. In this study, buccal samples drawn more recently were also more likely to fail, contrary to expectations, but time since sample collection did not alter the effects of ds/total DNA ratio in our analysis. Reasons for this finding are unclear, but may be driven statistically by a gap of over a year in which recruitment was dormant (e.g., artificially making the differences between the medians larger), as study protocols and staffing were consistent throughout the study. These results, however, suggest that longer cryogenic storage times do not negatively impact buccal DNA quality. The threshold of 34% ds/total DNA was selected to provide good discriminative value and low failure rate, as samples not meeting this threshold were 18 times more likely to fail than those meeting it. Application of this QC standard thus prevented the unnecessary running of samples on expensive chips if their likelihood of failure was significant, while retaining maximal sample size. This threshold may not be universally applicable to all studies depending on measurement of DNA concentration and laboratory techniques, but should be considered for studies conducting high-throughput genotyping using buccal-derived DNA.
In this large study, buccal samples had similar overall pre-genotyping failure rate, but lower cQC and lower DM call rates than blood samples, suggesting a shift toward lower overall quality. The similarity in overall failure rate may be somewhat optimistic, as buccal sample selection in the last 70% of this study was intentionally shifted toward those most likely to pass QC (e.g., ≥34% ds/total DNA); extracted samples below this threshold were not included in genotyping runs after the 3rd plate. The current analysis suggests that buccal samples may require higher QC thresholds to ensure that cQC and DM quality metrics are comparable to blood samples, particularly for the DM call rate, which was lower for both passed and failed buccal samples compared with the blood samples. There is precedent for subjecting different “populations” of samples or loci to different thresholds to improve performance, which has shown promise in previous situations . However, establishing higher thresholds would result in the rejection of more buccal samples in the QC phase, limiting sample size.
Birdseed genotyping results from passed samples were comparable for buccal and blood samples. Although all results were statistically better among blood samples, the magnitude of the differences are frequently nominal. This is consistent with a previous analysis of paired buccal and blood samples in this study that revealed similar call rates and 98.8% concordance of genotyping calls between buccal and blood samples, indicating high fidelity of genotyping across sample type . Median call rates were high for both sample types (>99%) and deviations from HapMap minor allele frequencies for Caucasian populations were similarly minor. Analysis of MAF deviations showed similar patterns, even for rare SNPs that are frequently ignored in genetic studies for concern over accurate genotype calling. However, all MAFs estimated for this study deviated positively from HapMap estimates, which may be due to regional differences in population compared with the CEU HapMap. The overestimation of MAF may also suggest the presence of missing call bias where no-calls are more likely in one genotype than another , although this effect appears similar for buccal and blood samples. Visual inspection of cluster plots revealed no obvious issues, with distinct clusters and minimal no-calls for both sample types, even for SNPs of low MAF. Furthermore, the median test revealed no batch effects on estimated MAF by locus across plates (data not shown). However, the average raw intensities for buccal samples were significantly lower than the blood samples (data not shown), which suggests that buccal samples and blood samples should be clustered separately for accurate genotyping.
These similarities between buccal and blood samples suggest that both sample types can be used in high-throughput genetic studies with only minor caveats. However, with the increasing use of “out of study” controls in large-scale genetic studies, these minor differences may be magnified. In particular, when cases and controls are derived from different sample types (e.g., cases are buccal DNA, while controls are blood DNA), differences in sample failure rates, call rates and MAF estimation between buccal and blood samples may result in spurious false-positive associations with disease status . It is thus imperative that sample types match between cases and controls, or are at least represent a balance of sample types in both groups.
There are several important strengths of this study. A large number of samples were evaluated in the context of an active genome-wide association study, and processed by the same laboratory personnel. Plates were constructed to incorporate both cases and matched controls of the same sample type on the same plate, and collection of buccal samples was conducted by trained study nurses using standardized protocols. Despite these significant strengths, this study also has certain limitations. The present analysis does not include paired buccal and blood samples from the same individuals, so we cannot determine concordance in genotyping between sample types. However, our previous study in the same population suggested buccal DNA provide reasonable concordance with paired blood samples . Also, identification of the ds/total DNA ratio threshold was conducted in the course of the genotyping, which changed the decision about which samples would be genotyped in the remainder of study. Thus, it was not possible to fully evaluate whether the 34% threshold is optimal, or whether a different threshold would have been better if all samples had been run blinded to ds/total DNA ratio.