Training set of pathogenic and benign ED variants to define the usability and strength of the ACMG/AMP criteria

Literature and database searches were performed using PubMed, Mastermind, gnomAD, and ClinVar (accessed February 2021), and variants with strong evidence of pathogenicity or benignity/neutrality were considered to define a training set of variants used to help in the definition of the specifications of the ACMG/AMP guidelines. Variant selection was based on a simplistic model where strong pieces of evidence in favor of pathogenicity or neutrality were considered (Table 1): 17 variants (13 in POLE and 4 in POLD1) were considered pathogenic, and 5 (3 in POLE and 2 in POLD1) benign (Table 2; Additional file 3: Table S3).

Table 1 Evidence scoring system to select the pathogenic and benign ED variants that were used to define the usability and strength of the ACMG/AMP criteriaTable 2 Germline POLE and POLD1 ED variants reported in the literature with strong evidence to be considered (likely) pathogenic or (likely) benign. The criteria considered for their selection as pathogenic or benign (criteria in Table 1) are highlighted in bold. Details and references are shown in Table S3 POLE and POLD1 ED-specific variant curation criteria

POLE and POLD1 specifications to the ACMG/AMP criteria are shown in Table 3. Of the 28 original criteria, 8 were excluded (PVS1, PM3, PM4, PP2, PP5, BP1, BP3, and BP6). Rules were modified by detailing the content and/or changing the strength level of the original recommendations.

Table 3 POLE and POLD1 ED-specific ACMG/AMP recommendations. In blue, population data; in green, segregation and phenotypic data; in grey, variant nature, location and in silico predictive data; and in yellow, functional data. In orange italics, criteria combinations that are not allowed or criteria that need modification when co-used with another onePopulation data

BA1 and BS1 are criteria against pathogenicity based on the frequency of the variant in general population. To calculate the allele frequency threshold, the prevalence and penetrance of germline pathogenic variants in POLE and POLD1 ED should be considered. Available data indicate that PPAP is a rare syndrome with very low population prevalence: Only one of 17 POLE ED pathogenic variants considered (Table 2) was detected in gnomAD non-cancer individuals (POLD1:c.946G>C; p.Asp316His: 1 in ~ 230,000 alleles). Although accurate unbiased penetrance estimates are still unavailable, available data [5, 6] suggests that the penetrance for POLE and POLD1 ED pathogenic variants might be close to other autosomal dominant cancer syndromes caused by DNA repair defects, such as Lynch syndrome (MLH1, MSH2), with an estimated average CRC risk of ~ 40%—50% by age 70 [60]. By using the Whiffin/Ware calculator [61] (http://cardiodb.org/allelefrequencyapp/), the inferred allele frequency threshold (AFT) (95% CI) obtained for BA1, with allele heterogeneity set at 1, was 0.002%, and for BS1, with allele heterogeneity set at 0.1, 0.0002% (Additional file 1: Supplementary Results). Due to the scarcity of available data, the rough estimation of the syndrome penetrance, and the fact that the number of pathogenic variants is likely underestimated (missense variants are harder to classify than loss-of-function variants), we recommend applying higher AFTs: BA1 to variants with a population allele frequency ≥ 0.02%, and BS1 to variants with a population allele frequency ≥ 0.002%. Data may be obtained from gnomAD (non-cancer), or from any outbred (non-founder) population groups in that repository (non-Finnish European, African/African American, Latino/admixed American, South Asian, or East Asian). The variant must be present in at least 5 alleles.

PM2 uses absence in controls for autosomal dominant diseases. Based on the incomplete penetrance and/or late disease onset, we recommend using PM2 with a supporting level of strength for variants absent, or present in ≤ 1 in 200,000 alleles (≤ 0.0005%) in gnomAD non-cancer dataset (all individuals) (coverage of variant position > 30X) [59]. Supportive of this threshold is the fact that POLE p.Leu424Val, the most recurrent known pathogenic germline variant, is not present in non-cancer gnomAD individuals (~ 200,000 in gnomAD v.2.1.1 and v.3.1.1).

BS2 uses the presence of the variant in healthy adult individuals when full penetrance is expected at an early age. We specified the code to account for the reduced penetrance and later age of PPAP onset. Also, heterozygotes identified among non-cancer individuals could have polyps that have not been detected or reported. Considering the families with the 17 pathogenic variants listed in Table 2, among the 169 carriers reported (POLE n = 128 and POLD1 n = 41), there are 47 cancer-free individuals. Of them, 12 carriers had no polyp information and/or had not undergone colonoscopy screening. Of the cancer-free carriers with polyp information (35/47), 97% (34/35) had polyps (any number). Of these, detailed information on polyp number was specified for 25 individuals: 60% of them (15/25) had been diagnosed with ≥ 10 polyps (median age at diagnosis: 35; age range: 15–53) (Additional file 4: Table S4). Based on the available data, and on the extremely low prevalence of PPAP-associated recurrent pathogenic variants, we recommend using BS2, with a supporting level of strength, for variants that have been identified in ≥ 5 cancer-free individuals aged > 60. If BA1/BS1 is applied and gnomAD data is used for BS2_supp, apply only BA1/BS1. A strong level of strength may be applied if the variant is identified in ≥ 10 cancer-free and adenoma-free individuals aged > 60. To apply this level of strength, the cancer-free individuals must have been subjected to colonoscopy screening (not applicable for gnomAD individuals).

No biallelic germline ED pathogenic variants have been identified in humans. It has been speculated that those could likely be embryonic lethal [3]. Interestingly, depending on the nature of the pathogenic variant, biallelic mutant mice may be viable [62, 63]. While PoleP286R/P286R mice showed embryonic lethality, homozygotes for other ED pathogenic variants survived into adulthood but developed cancer very early in life. In that same line, PoleP286R/+ mice develop more severe phenotypes than heterozygotes for other ED pathogenic variants, which may even be indistinguishable from wildtype animals, suggesting a more severe effect in humans than in mice [62,63,64]. Pold1 homozygous mutant mice die of cancer at extremely early ages [63]. Based on the mice findings, even in the hypothetical case that biallelic ED-mutated humans were identified (viable), we would expect extremely aggressive tumor phenotypes, probably with very early age of onset. Therefore, we propose to apply BS2 to variants identified in homozygous state in one cancer- and adenoma-free adult individual if his/her homozygosity status has been confirmed by genotyping the parents. BS2_supporting may be applied when two homozygous adult cases are identified without available parental confirmation and/or polyp information (e.g. gnomAD non-cancer dataset).

PS4 is based on the statistically significant higher frequency of the variant in patients compared to controls. We recommend applying the case–control criterion, considering PPAP-associated phenotypes (Table 4), when the resulting p-value is ≤ 0.05 and OR ≥ 2 or the lower 95% CI is ≥ 1.5 [43]. Also, we recommend applying PS4, with supporting level of strength, when a CMMRD-like phenotype [65] in absence of germline biallelic MMR (likely) pathogenic variants or VUSs is identified in one proband, and with moderate level of strength, when the CMMRD-like phenotype is identified in ≥ 2 probands. No other PPAP-associated phenotypes are considered due to their non-specificity. See Table 3 for permitted co-usages.

Table 4 Clinical phenotypes of PPAP considering the 169 carriers (122 cancer-affected) reported in the literature with any of the 17 pathogenic variants listed in Table 2. Columns 2–4 consider individual cancers (if one person was diagnosed with several primary tumors, they are individually accounted for). Columns 5–7 consider the number of carriers with a specific phenotype, See Table S4 for details Variant nature and location, and in silico predictions

Evidence suggests that loss-of-function and outside-ED POLE and POLD1 variants are nonpathogenic for PPAP, and only missense and in-frame indel variants within the ED should be considered as potential cause of PPAP and as predictive biomarkers in oncology [3, 5, 66, 67]. Therefore, PVS1, PM4, BP1, and BP3 are not considered due to their irrelevance to the syndrome and its mechanism of pathogenicity.

The PM1 criterion is given to mutational hotspots and/or critical well-established functional domains without benign variation. We recommend applying PM1 (moderate) for: (i) somatic mutational hotspots (observed in ≥ 10 tumors), which currently include: POLE P286R, S297F, V411L, A456P and S459F (somatic hotspot information obtained from TCGA and COSMIC tumors: Additional file 5: Table S5); and (ii) variants affecting the exonuclease catalytic sites POLE D275 and E277, and POLD1 D316 and E318 [4], when the resulting amino acid shows a negative BLOSUM62 when compared to the wildtype residue. Available data indicate that variants affecting the binding of the exonuclease with the DNA, and/or located within the Exo motifs are likely to be pathogenic [68]. In fact, 11 of the 13 non-catalytic pathogenic variants, and none of the benign variants, affect residues of Exo motifs and/or are in contact (distance < 6 Å) with the DNA when the polymerase is in proofreading position (Table 2). We recommend applying PM1 with a supporting level of strength to any variant fulfilling either one of these two conditions, when PM1 (moderate) has not been applied. ED amino acids at < 6 Å from the DNA are listed in Additional file 4: Table S2. See Table 3 for permitted co-usages of PM1 with other criteria.

PP3 and BP4 are related to in silico pathogenicity predictions. Following recent ClinGen indications [69], PP3 should be applied for variants with REVEL scores ≥ 0.644, which occurs for 11 of the 17 pathogenic variants and none of the benign variants, and BP4 for silent (synonymous) variants or intronic variants without predicted splicing effect, and for missense ED variants with a REVEL score ≤ 0.290. Sensitivity/specificity analysis should be performed to set gene-specific cutoff values for POLE and POLD1 ED variants, when enough pathogenic and benign variants are identified to be used for the analysis. Predictions of loss of function or splicing impact (unless it causes an in-frame splicing defect that affects the ED) should not be considered as supporting evidence of pathogenicity or benignity.

BP7 is applied for any synonymous or intronic variant at or beyond +7/-21 for which splicing prediction algorithms predict no impact to the splice consensus sequence nor creation of a new splice site, regardless of nucleotide conservation.

PS1 considers any missense nucleotide change that translates into an amino acid change that has been previously established as (likely) pathogenic with a different nucleotide change (i.e., different nucleotide variant, same amino acid change). Strong level of evidence is recommended for pathogenic variants, and moderate, for likely pathogenic variants. Likewise, PM5 relates to a missense variant at a residue where a different pathogenic missense variant caused a change to a different amino acid. In this case, we recommend using PM5 only when the resulting amino acid shows equal or lower BLOSUM62 score (i.e., equally or more damaging) than the previously classified pathogenic (PM5) or likely pathogenic (PM5_supporting) amino acid [70].

Segregation and phenotypic data

PP1 original criterion uses cosegregation of the variant with the disease in multiple family members affected with the associated phenotype as evidence for pathogenicity. The main PPAP-associated tumor types are colorectal, endometrial, ovarian, breast, brain, and upper gastrointestinal cancers, as well as polyposis (> 10 adenomas), all with prevalence values > 10% among cancer-affected carriers (Table 4; Additional file 4: Table S4). Nevertheless, due to the broad phenotypic spectrum and the relative high population frequency of most PPAP-associated tumor types, which may lead to phenocopies, we recommend considering only the three most prevalent PPAP-associated phenotypes, i.e. adenomatous polyposis (> 10 adenomas), CRC and endometrial cancer, unless tumor mutational data indicate that other tumor types are hyper/ultra-mutated and harbor the gene-specific mutational signature(s).

Based on the gradations considered by ClinGen variant curation expert panels [39, 40, 42, 71], we recommend the system that considers the number of meiosis across one or more families [72]: strong level of evidence when co-segregation is observed in ≥ 7 meiosis in ≥ 2 families; moderate level of evidence when cosegregation is observed in ≥ 5 meioses in ≥ 1 family; and supporting level when cosegregation is observed in 3–4 meioses in ≥ 1 family. The meiosis counting-based system may not be optimal for cosegregation analyses in cancer-related genes [72], particularly when there are variable ages at onset, high probability of phenocopies, and/or incomplete penetrance, as happens for PPAP. When more accurate data on the syndrome are available, this rule code will likely implement a Bayes factor-based approach, which measures the likelihood that cosegregation patterns represent a gene-disease penetrance model [72].

BS4 is used when there is lack of segregation. Due to existence of de novo cases, the wide tumor spectrum observed in PPAP, the expected incomplete penetrance and the -often- late onset of cancer, we recommend considering only non-carrier family members affected with > 10 adenomas, or CRC, or endometrial, or any other hyper- or ultra-mutated tumor (≥ 10 mut/Mb) with the mutational signature(s) associated with the corresponding polymerase proofreading deficiency. BS4 should be applied, with a supporting level of strength, when there is ≥ 1 family with ≥ 2 meiosis with a genotype-negative phenotype-positive situation, in absence of pathogenic or likely pathogenic variants or variants of unknown significance in other known hereditary cancer or polyposis genes that could explain the phenotype. As for PP1, this criterion will likely implement a Bayes factor–based approach [72] in the future.

PS2 and PM6 contemplate the presence of de novo variants. We recommend applying the point-based criteria based on phenotypes indicated in Table 3 to determine the levels of strength. Points are additive per each de novo case.

We recommend applying BP2 when the variant is observed in trans with another (likely) pathogenic ED variant in the same gene in a tumor-free (cancer- and adenoma-free) adult (see comment in “Population data” section; BS2 criterion) or when the variant is identified ≥ 3 times with additional ED (likely) pathogenic variants in the same gene with unknown phase. The other observed ED variant must have been classified as (likely) pathogenic using the herein defined recommendations.

Tumor data: mutational burden and signatures

To evaluate the specificity of the proofreading-associated mutational signatures, we analyzed 134 tumor samples (different tumor types) including: i) 50 MMR proficient (pMMR) and 20 dMMR TCGA tumors without ED variants, and ii) 50 pMMR, 12 dMMR tumors and 2 tumors without available MMR status information with somatic pathogenic ED variants (62 tumors with POLE and 2 with POLD1 ED mutations) that represent 9 of the 17 pathogenic variants listed in Table 2 (data source: 59 TCGA tumors and 5 COSMIC tumors with available exome sequencing data). The results are represented as a Heatmap in Fig. 1 (details in Additional file 6: Table S6). SBS10a, SBS10b, SBS28 and SBS14 were highly specific of Polε proofreading deficiency; no trace of those signatures was detected among the tumors without ED variants. SBS14 was mostly, although not exclusively, found among dMMR tumors.

Fig. 1

Heatmap showing the clustering of tumors based on the contribution of tumor mutational signatures SBS10a, SBS10b, SBS10d, SBS28, SBS14, SBS20 and “other signatures”. Analysis was performed with 64 tumor samples with somatic pathogenic variants in POLE and POLD1 EDs, 3 tumors belonging to three probands with germline pathogenic variants in POLD1, and 70 TCGA tumor samples without polymerase exonuclease domain variants

Only two POLD1 ED-mutated tumors, both dMMR, could be included in the analysis: one tumor had 10% SBS14 contribution and no trace of Polδ proofreading-deficient signatures (SBS10d or SBS20), and the other had 83% SBS20 contribution. Unlike the other polymerase proofreading-associated signatures, SBS20 was also observed in a subgroup of dMMR tumors (n = 15) without ED variants, at contributions ranging from 18 to 40%. Due to its non-specificity, we recommend not using SBS20 for variant classification. Due to the lack of pMMR, Polδ proofreading-deficient sporadic tumors, we re-analyzed exome/genome sequencing data obtained from three additional proofreading-deficient tumors (two CRCs and one adenoma), developed by heterozygous carriers of germline POLD1 p.Leu474Pro, p.Asp316His, and p.Ser478Asn [54, 73]. All three samples were hypermutated (59, 114 and 36 mut/Mb respectively) and had 34%-68% contribution of SBS10d, highly specific of Polδ proofreading deficiency in tumors (Fig. 1). Moreover, all three tumors had copy-neutral loss of heterozygosity (cnLOH) in the POLD1 region that caused the loss of the wildtype allele [73].

The 50 pMMR Polε proofreading-deficient cancers had an average of 144 mut/Mb (range: 2.6—325), and the 10 dMMR Polε proofreading-deficient cancers, 255 mut/Mb (range: 109 – 531). Only 2 samples, both harboring POLE p.Leu424Val had TMBs < 25 mut/Mb (2.6 and 4.4 mut/Mb). All 62 POLE ED-mutated tumors, regardless of their MMR status, had > 5% contribution of signatures SBS10a and/or 10b (median: 65%; range: 6%– 87%). When considering all Polε proofreading-deficient signatures combined, i.e. SBS10a, SBS10b, SBS28 and SBS14, 100% of samples reached > 20% contribution.

In the generic ACMG/AMP guidelines, PP4 corresponds to highly specific phenotypes or family history of a disease with a single genetic etiology, and BP5, to variants found in cases with an alternate molecular disease basis. We propose to adapt these criteria to the presence or absence of the proofreading deficiency-specific mutational signatures and high TMB. To consider PP4, no other (somatic) ED missense variant classified as (likely) pathogenic or of unknown significance in the same gene (POLE or POLD1) should occur in the tumor, and at least PM2_supporting must be fulfilled. We recommend performing the mutational signature analysis when the tumors are hypermutated (> 10 mut/Mb) or have at least a total of 80 somatic SNVs, to minimize the detection of false (artifact) signatures generated from an extremely small number of variants. Optimally, the use of exome or genome sequencing data is recommended, although the use of sequencing data obtained from panels that include a relevant number of genes may also be used.

We recommend using PP4 with a strong level of strength: For POLE ED variants, when at least two tumors have SBS10a, SBS10b, SBS28, and/or SBS14; and for POLD1 ED variants, when at least two tumors have SBS10d or when one tumor has SBS10d and loss of heterozygosity (LOH) that causes the loss of the wildtype allele. PP4_moderate may be applied for POLE ED variants when one tumor has SBS10a, SBS10b, SBS28, and/or SBS14; and for POLD1 ED variants when there is one tumor with SBS10d (no available 2nd hit information or no LOH). These recommendations are based on the data obtained from fresh/frozen tumor samples. To minimize the potential effect of FFPE sequencing artifacts, a ≥ 5% contribution of the gene-specific signatures will be considered to apply PP4 strong and moderate criteria.

We recommend using BP5 when two or more tumors with the ED variant have ≤ 1 mut/Mb. For POLE variants, BP5 should be used when two or more tumors harboring the variant, with > 1 mut/Mb or at least > 80 total single nucleotide variants, have neither SBS10a, nor SBS10b, nor SBS28, nor SBS14; or when one tumor has ≤ 1 mut/Mb and another one, with > 1 mut/Mb or > 80 single nucleotide variants, has neither SBS10a, nor SBS10b, nor SBS28, nor SBS14. For POLD1 variants, use BP5 when two or more pMMR tumors harboring the variant, with > 1 mut/Mb or at least > 80 total single nucleotide variants, do not have SBS10d; or when one tumor has ≤ 1 mut/Mb and one pMMR tumor, with > 1 mut/Mb or at least > 80 total single nucleotide variants, has no SBS10d. In all instances, at least two tumors are required to minimize the possible analysis of phenocopies and the effect of FFPE-derived sequencing artifacts.

Functional data

Available in vitro assays to test the functionality of POLE and POLD1 ED variants assess the proofreading ability of the polymerases in absence and presence of the variant. The studies reported to date rely mostly on yeast-based assays, although cell-free assays, in vitro human or murine cell line experiments, and in vivo mouse models, have also been used (Additional file 3: Table S3).

PS3 and BS3 rely on well-established in vitro or in vivo functional studies supporting or discarding a damaging effect of the variant. Based on available data and the fact that the performance of the functional studies published so far has not been evaluated, we recommend using PS3_moderate when results from at least 2 independent experiments (at least one in a non-yeast model) that assess, with proper positive and negative controls, the proofreading function of the corresponding polymerase in presence and absence of the variant, show defects and are concordant. If only results from one experiment are available, or the results, even from multiple experiments, are produced exclusively in yeast-based systems [5], we recommend applying a supporting level of strength. We propose to decrease the level of strength for yeast-based evidence because published results show high variability among replicates and experiments (publications in Additional file 3: Table S3), and some concerns have been raised regarding the assessment of variants affecting the DNA binding, which might show an effect in yeast even when the variant is non-pathogenic [5, 10, 68]. We currently recommend using BS3_supporting, when at least two independent experiments (≥ 1 in a non-yeast model) show no proofreading defect. For both PS3 and BS3 criteria, the assayed amino acid change must be the same as the one identified in the patient.

The ClinGen Sequence Variant Interpretation Committee recommends assessing the performance of any functional assay using variants classified as pathogenic or benign according to clinical parameters (cross validation) [74], which has not been done for any of the POLE/POLD1 functional assays reported to date. Calibration according to the cross-validation results is recommended to correctly apply the PS3 and BS3 rules, providing the correct level of strength, or a calibrated quantitative value if Bayesian transformation of the ED-specific ACMG/AMP guidelines is applied.

Classification of reported variants

The defined classification recommendations (Table 3) were applied to 128 variants reported in the literature (reviewed: March 2023) and ClinVar (access date: July 2021), including the 23 variants used for the definition of the guidelines. Of the 128 variants considered, 7 were classified as pathogenic, 10 as likely pathogenic, 7 as benign, and 10 as likely benign. Of the 17 a priori pathogenic variants included in Table 2, all but POLE:c.824A>T; p.(Asp275Val) and POLE:c.830A>G; p.(Glu277Gly), now classified as variants of unknown significance, were classified as P (n = 7) or LP (n = 8). Moreover, two additional variants were classified as likely pathogenic: POLE:c.857C>T; p.(Pro286Leu) and POLE:c.1373A>T; p.(Tyr458Phe). Additional file 3: Table S3 shows the classification of all 128 variants taking into consideration the data available.

Clinical features of reported individuals with constitutional POLE or POLD1 ED pathogenic or likely pathogenic variants

To date, literature reports include 205 individuals heterozygous for the 17 POLE or POLD1 variants classified as pathogenic or likely pathogenic following the defined recommendations. Of the 205 heterozygotes, 149 (73%) were diagnosed with cancer: 120 (58% of the 205 carriers) with CRC (mean age at diagnosis: 41; range: 13–80), 21 (22% of 95 female carriers) with endometrial cancer (age: 50; range: 31–58), 11 (12% of female carriers) with breast cancer (age: 55; range: 38–65); 8 (8% of female carriers) with ovarian cancer (age: 42; range: 33–50), 19 (9%) with extracolonic gastrointestinal cancers (age: 45; range: 35–78), 18 (9%) with brain cancer (age: 28; range: 4–66), and 9 (4%) with other cancer types. The majority of heterozygotes (88%) had reports of cancer, and/or preneoplastic lesions, and/or non-tumoral extracolonic manifestations (e.g. café-au-lait macules). Sixty-four percent of those with polyp information (70/108) were reported to have > 10 gastrointestinal polyps (detailed phenotypes in Additional file 4: Table S4).

While these phenotypes should currently guide clinical surveillance in carriers, future prospective collaborative efforts will provide more accurate (unbiased) estimates of cancer risk and penetrance. Furthermore, oncologic therapeutic decisions in the context of the hereditary cancer syndrome, and for cancers with somatic pathogenic or likely pathogenic POLE or POLD1 exonuclease variants, should consider the good prognosis and response to immune checkpoint inhibitors of polymerase proofreading deficient tumors [75,76,77].