As NGS technologies advance into clinical settings, it is critical to establish quality control metrics that can guide reliable sequencing results. To this end, entities such as the Next-generation Sequencing Standardization of Clinical Testing (Nex-StoCT) workgroup (coordinated by the Centers for Disease Control), and the College of American Pathologists have proposed criteria for assuring quality NGS data and interpretations. For example, Nex-StoCT recommended a series of post-analytical quality control metrics relevant to NGS, including depth and uniformity of coverage, transition/transversion ratio, base call quality score, mapping quality, and others . Pre-analytical quality control metrics, such as determining the minimum DNA requirements needed to perform the test, are also critical. Although DNA characterization using spectrophotometry is appropriate for many molecular tests and specimen types, FFPE DNA samples pose unique challenges, particularly for amplification-based assays. Targeted detection of FFPE DNA analytes by NGS, a method that offers true digital quantification, demands a careful consideration of template library complexity to achieve reliable and accurate results.
In this study, we compared three assays for assessing FFPE DNA inputs into targeted NGS. The first was spectrophotometry, a method that only reports the ‘bulk’ DNA concentration. Compared to the other quantification methods, this approach overestimated the ‘functional’ DNA concentration by approximately 15-fold across 165 FFPE DNA samples. As a result, we can conclude that spectrophotometry is inappropriate to determine FFPE DNA inputs into PCR-based NGS enrichment since it provides no information to ensure accurate results with both low and high quality DNA samples.
The second comparator method was a fluorescent dye-binding assay (Qubit). This assay is widely used, and offers simplicity, sensitivity, high throughput, and tolerance to various contaminants. Interestingly, we find that the fluorescent dye in this assay is affected by DNA modifications introduced by the fixation/embedding process and thus behaves as a ‘poor man’s’ structural probe that can segregate the highest and lowest quality DNA samples relative to PCR amplification (Figure 2). However, this assay is not capable of differentiating among lower quality FFPE DNA templates (that is, QFI <3 to 6%), nor can it prescribe specific adjustments in DNA input that may help offset the deleterious effects of poor functional quality. In fact, the number of variants called in AmpliSeq NGS was inflated more than 7-fold for the 5 lowest quality FFPE DNA samples in our 44 sample subset when stratified by QFI (median 222 variants) compared to Qubit (median 31 variants). Thus, QFI-PCR, but not Qubit, identified the lowest quality samples at the level of NGS variant calls. Moreover, correlation between QFI-PCR and Qubit by template ‘functionality’ for these samples was poor (R = 0.30). Although a comprehensive understanding of the molecular features of the binding of this particular dye to DNA is lacking, the binding site for similar dyes is only two to four nucleotides . One explanation for this observation is that the lowest quality FFPE samples contain DNA fragments that are receptive to dye binding and fluorescent signal enhancement, but that these samples are poorly amplified by PCR. This is a critical insight that must be accommodated when the downstream assay is based on PCR enrichment, as it is for many targeted NGS assays.
In contrast, we find that the third method, QFI-PCR, is well suited to profile FFPE DNA intended for targeted amplification prior to NGS. First, it is logical to design a quality control with the same methodology that is used for targeted enrichment. For this reason, we designed the target amplification region of QFI-PCR to match the median amplicon size produced by the AmpliSeq Cancer Panel multiplex PCR. Second, QFI-PCR offers absolute quantification that can be used independent of other methods to calculate a minimum copy number input to satisfy downstream assay requirements. Third, the assay is sensitive to PCR inhibitors, and thus can predict potentially poor performance in library enrichment due to extraction contaminants. QFI-PCR respects that high template quality is not the sole sample-level variable that drives successful library preparation. Lastly, the assay is cost-effective, high throughput, and leverages a ubiquitous install base of real-time thermal cyclers that can facilitate adoption by research and clinical laboratories.
It is important to note that the utility of QFI-PCR depends on the use of a genomic locus that is unaffected by tumor ploidy such that the measured copy number in FFPE DNA reflects the functional quality of the DNA specimen and not a separate process. We selected a region in the TBP gene since this gene was reported to be unchanged in copy number in >98% of TCGA samples from the cancers investigated in this study. Caution is warranted for use of this locus in neoplasms such as adenoid cystic carcinomas that have a high rate of TBP amplification or deletion. In these cases, other genes such as FTH1 may be targeted instead. Alternatively, a single multiplex assay that includes both TBP and FTH1 loci may be useful.
Importantly, the results of QFI-PCR can be used to calculate the minimum amount of sample input for targeted PCR enrichment by measuring the percentage of DNA templates that are competent for PCR amplification. This insight can reduce the risk of false positives and false negatives in variant calling using both laboratory-developed and commercially available procedures for enrichment and subsequent NGS. This conclusion is evinced in Figures 3 and 4, and Table 1, which demonstrate that: i) false negatives and inaccurate mutation fractions can be rescued by increases in DNA input that are guided by the QFI; and ii) a commonly used commercial method for the multiplexed PCR enrichment of cancer genes can produce an overwhelming number of false positives if the ‘functional’ DNA copy number is unacceptably low. As a result, the integration of a pre-analytical step based on QFI-PCR offers a much improved approach to ensure accuracy in NGS data interpretations. This advance is particularly timely since ‘benchtop’ NGS instrument placements now number in the thousands and many solid tumor tests that are currently performed can (and are) being rapidly supplanted by PCR-based targeted NGS assays.
Our results have important implications not only for the evaluation of FFPE DNA prior to NGS, but also for other assays that rely on PCR amplification. Rigorous and quantitative characterization of DNA-poor samples is essential to ensure that results are generated from sufficient copies of functional DNA templates, interpreted with consideration of DNA quality, and can support reliable mutation calls. The consequences of a misguided diagnostic decision based on sequencing results from inadequate amplification of DNA template are serious and could lead to inappropriate patient treatment by failing to identify an actionable mutation or prescribing the wrong treatment based on a false positive result. Such errors may also undermine retrospective biomarker association studies relevant to cancer drug development. To this point, we observed that 75% of the false positives reported in Table 1 were C>T or G>A transition mutations. Repair of FFPE DNA with uracil DNA glycosylase has recently been reported to reduce the incidence of such artifactual mutations [25, 26] and other potentially restorative methods for treating FFPE DNA  have been described. These approaches may be particularly beneficial for low quality FFPE DNA, which will require very high analytical sensitivity for variant detection to capture a broader group of clinically relevant mutations, such as those encountered in early stage drug resistance [28, 29]. In addition, we note that quantitative sample characterization is required for the accurate determination of the mutational load in a tumor, which may have therapeutic value . It is not surprising, then, that confirmation testing is an indispensable component of existing clinical NGS recommendations to guard the accuracy of variant calls . We suggest that quantitative FFPE sample characterization using threshold-based metrics such as the QFI can improve the cost, efficiency, and accuracy of confirmation, and help de-risk the final clinical report. Although FFPE DNA quality can vary considerably across cohorts, the concerns are most acute with the lowest quality specimens. In this study, the median QFI was less than 7% for each of the three collection cohorts, and across all 6 of the tissue types tested (colon, lung, skin, ovary, breast, and thyroid). In fact, nearly half (79/165, 48%) of the 165 FFPE samples in our cohort had a QFI of <3%.