Concern regarding classification of germline TP53 variants as likely pathogenic

To the Editor, We read with some concern the recent article in Human Mutation on the frequency of potentially actionable germline variants in the TP53 gene (de Andrade et al., 2019). A prevalence estimate, including TP53 variants widely accepted as pathogenic/ likely pathogenic (LP), was reported as one carrier in 3,555‐5,476 individuals. However, an additional estimate of >1 in 500 was also proffered in the abstract, generated from the inclusion of several variants labelled as “likely pathogenic” using an arguably spurious classification system devised by and unique to the author, albeit that their classification as “likely pathogenic” for three common variants (p.N235S, p.V31I, and p.R290H) were juxtaposed with the term “controvertible.” We acknowledge this approach is distinct from and less stringent than that would be appropriate for clinical use. However, it is of great concern that readers interpreting TP53 variants, often found on panel testing in individuals with no personal history of cancer, may use the authorʼs classification and counsel as having Li Fraumeni syndrome individuals carrying these de Andrade et al. (2019) so‐ called “likely pathogenic” variants. We have used four sources of data to review the pathogenicity of variants present in gnomAD at a frequency of n ≥ 6, and assessed by de Andrade et al. (2019) using 2,3, or 4 classifications as being LP/P (Table 1). First the authorʼs own table on likely frequency in the Cancer Genome Atlas (TCGA) to develop a “case control” assessment, second to use the proportions of the variants as a percentage of all gnomAD variants compared with the proportions of the variant in the IARC TP53 database. Germline variants reported in IARC are largely the result of testing of families with Li Fraumeni criteria and of sentinel Li Fraumeni tumours at young ages. As such, this provides evidence of the specific pattern of tumors linked to germline TP53 pathogenic variants and is potentially stronger evidence than a simple case control analysis of all cancer. If all the variants reported by de Andrade as LP/P had the same effect the ratio would be close to 1.0 for all variants. If a common variant in Gnomad was only rarely associated with a typical Li Fraumeni pattern, perhaps as a chance finding, then the ratio would drop well below 1.0 providing evidence against causality. For the TCGA data, ratios of TCGA to gnomAD of less than 2.5‐fold were considered against pathogenicity (in terms of high risk) and in IARC a ratio of below 0.5 was considered against. In reverse, ratios of case control from TCGA above 2‐fold and above 1.5‐fold in the IARC ratios were considered supportive. Third, for core binding domain missense variants a recent functional classification has been made available (Kotler et al., 2018). We also obtained computational analyses of the pathogenic likelihood using the PolyPhen‐2, SIFT, and Align GVGD algorithms. In addition, on the basis of the data outputs available, we also classified the variants using the recent ACMG guidelines. Examples of variants with gnomAD frequencies < n = 6 were also included to act as controls for our analyses. The results of these analyses indicate that for the four variants, (c.848G>A; p.R283H, c.374C>T; p.T125M, c.655C>T; p.P219S, c.784G>A; p.G262S), that were classified as pathogenic by de Andrade et al. (2019), there is no case control data to support this conclusion with c.784G>A; p.G262S having three entries in gnomAD and 0 in IARC, insufficient to prove pathogenicity (see Table 1). In total, 15 of the LP and these four labeled pathogenic variants would be reclassified below LP on the basis of the Kotler data. For those 10 variants classified by de Andrade et al. (2019) as likely pathogenic and having a frequency in gnomAD n ≥ 6, we show: (a) for eight of these variants the case control data is against pathogenicity (b) there is no support for pathogenicity for the 7/8 variants for which functional data are available and (c) the formal ACMG classification is either benign, likely benign or VUS. For c.847C>T; p.R283C, the data outputs are conflicting but its gnomAD frequency (n = 24) and lack of representation in IARC would suggest that even if there is pathogenicity (which is not supported by the functional studies) at most this can only be of low‐moderate penetrance. Similarly, c.523C>T; p.R175C gives conflicting results and, therefore, the ACMG classification as a VUS is appropriate. Of interest of the three “controvertible” variants, there is extremely strong evidence against pathogenicity for p.N235S and p.R290H with all four assessments being in favor of a benign classification. For p.N235S the frequency of one in 2,347 for non‐ TCGA gnomAD alone would count strongly against pathogenicity. There are nonetheless a small group of LP variants that had support for pathogenicity with c.527G>T p.C176F being supported by Kotler and c.1,010G>A p.R337H (both TCGA proportion/gnomAD non TCGA proportion and IARC proportion/gnomAD only proportion > 10), c.970G>C p.D324H (TCGA proportion/gnomAD non TCGA proportion = 9.12) from TCGA and IARC ratios. These three only contribute three individuals to the overall total of variants where there is good support for pathogenicity. As such, the conservative estimate for the frequency of TP53 LP and pathogenic variants in non‐TCGA gnomAD drops to only one in 6,258