Gene-specific criteria for PTEN variant curation: Recommendations from the ClinGen PTEN Expert Panel

3.2.1 Population data (BA1, BS1, PM2, and PS4)

PHTS is considered a rare disorder. A study within the Dutch population reported CS as occurring in approximately one in 250,000 individuals (Nelen et al., 1999); however, at the time CS was the only clinical diagnosis associated with PTEN, thus this study did not include individuals with other PHTS phenotypes and is likely an underestimate of true disease prevalence, which is to date unknown. PTEN pathogenic variants have been identified in 1% or fewer of unselected individuals with breast, thyroid, or endometrial cancers, and higher rates (up to 17% in one study) among patients with autism and macrocephaly (Butler et al., 2005; Nagy et al., 2011; Ring et al., 2016; Tung et al., 2016; Wong et al., 2016). To consider the maximum possible frequency of PHTS in the population, the group considered the population incidence and percent causation owed to PHTS for these four PHTS-component features (Supporting Information, Table S1). The highest estimates of disease incidence and attribution to PHTS were used to calculate conservative estimates of allele frequency attributable to PHTS. PTEN is the only gene associated with PHTS, and while a few recurrent variants have been reported, pathogenic variants of diverse types (missense, truncating, splicing, and large rearrangement) have been reported across the gene (Mester & Eng, 2013), making further adjustments for genetic heterogeneity unnecessary. Additionally, no founder effect exists and de novo variants often occur (Mester & Eng, 2012).

Summing the estimates gave a total allele frequency of 0.000961 (0.0961%). Rounding up, the EP accepted an allele frequency ≥0.001 (≥0.1%) as the cutoff for application of BS1. Based on these data, the EP also felt comfortable lowering the allele frequency for BA1 set by ACMG/AMP from ≥0.05 (≥5%) to ≥0.01 (≥1%). These values are purposefully conservative to account for the unknown population prevalence of PHTS and penetrance of several component features, sub-clinical disease presentations, and unknown population prevalence and percent of disease attributable to PHTS for many other component phenotypes. Additionally, per the ExAC database, the missense constraint (z = 3.71) and probability of loss of function (LOF) intolerance (pLI = 0.98) scores for PTEN suggest high constraint, making it unlikely that a pathogenic missense or LOF variant would rise to high allele frequency. The EP also adopted the recommendation by the SVI that the variant be present in at least five alleles with a minimum number of 2,000 alleles analyzed within the population of interest to minimize the risk of sequencing error or chance inclusion of an affected individual.

With respect to PM2 (variant is absent within population databases), the group initially tested their criteria with this rule as written, requiring that a variant be completely absent in order to apply this criterion. During the rule testing process, it was noted that several LOF alleles were present at ultra-rare frequency (1/240,000+ alleles) in the gnomAD cohort (Lek et al., 2016), including PTEN c.50_51delAA, listed as pathogenic per multiple ClinVar submitters and selected for the PATH/LPATH test set. While LOF is an established disease mechanism for PHTS, and variants leading to LOF would be expected to be PATH/LPATH, the ACMG/AMP guidelines require additional pieces of evidence in addition to PVS1 in order to achieve a PATH or LPATH classification. Although the gnomAD cohort excludes individuals with severe pediatric-onset disease (Lek et al., 2016), some of the cohorts, such as the Swedish Schizophrenia & Bipolar Studies cohort (dbGaP accession phs000473.vs.p2) contributing to the dataset include individuals as young as 18 years of age. Additionally, some individuals also came from The Cancer Genome Atlas cohort (https://cancergenome.nih.gov/) and are positive for cancer phenotypes. Cumulative cancer risk in PHTS is estimated to be less than 50% by age 40 (Bubien et al., 2013; Nieuwenhuis et al., 2014), and the benign features associated with this condition are often under-recognized (Eng, 2003), making it possible that affected individuals could be included in a general population cohort. Thus the group chose to modify PM2 to permit use if present at <0.00001 (0.001%) allele frequency in the gnomAD dataset or another large-sequenced population, with allele frequency <0.00002 (0.002%) in an ancestry-specific subpopulation if multiple alleles are present.

As per the ACMG/AMP criteria, PS4 may be applied in one of two ways: (1) significant case-control data or (2) counting multiple unrelated patients with the same phenotype. Not surprisingly given the rarity of PHTS, the PTEN EP could not identify an existing case-control study for which application of PS4 in the first manner would be appropriate. Should such a study come to light, PS4 may be applied to a case-control study finding an odds ratio >2 with P <0.05 and 95% confidence interval with lower limit >1.5. The EP has adapted the second application for use of PS4, counting multiple unrelated probands, to incorporate phenotype specificity (PP4) as described in section 3.2.6 (phenotype).

3.2.2 Computational and predictive data (BP1, BP3, BP4, BP7, PP2, PP3, PM1, PM4, PM5, PS1, and PVS1)

As a gene with several pathogenic missense variants, BP1 (missense variant in a gene for which primarily truncating variants cause disease) does not apply to PTEN. Likewise, PTEN does not contain a repetitive region without known function as would be used to apply the ACMG/AMP definition of BP3 (in-frame deletions/insertions in a repetitive region without known function), leading the EP to remove that criterion as well. However, PTEN is a highly conserved protein, with numerous pathogenic missense variants causing disease (Mester & Eng, 2013) and as of March 2018, no missense variants classified as BEN/LBEN by any ClinVar submitter. Thus the EP opted to maintain PP2 as written to be applied to PTEN. The EP also decided to adopt PM5 (novel missense change at an amino acid residue where a different missense change determined to be pathogenic has been seen before), adding that the missense variant in question need not be novel, but should contain a BLOSUM62 (Henikoff & Henikoff, 1992) value equal to or less than the known variant. The EP also decided that this rule may be applied when the known variant is likely pathogenic unless applying would lead to a higher (pathogenic) classification for the variant being assessed.

Again given the lack of BEN/LBEN PTEN missense variants, the EP was left without a set of established BEN/LBEN missense variants to be used to test the accuracy of in silico predictors to be used as evidence to apply BP4 or PP3 to PTEN missense variants. While investigating potential in silico tools, the EP also came to find that some algorithm predictions were highly sensitive to sequence alignment, further limiting confidence in these tools. Should the EP classify several missense variants as BEN/LBEN, another attempt will be made to validate in silico tools to apply PP3/BP4 for missense variants.

The EP did maintain PP3/BP4 use for intronic or synonymous variants which may impact splicing. HSF version 3.1 (https://www.umd.be/HSF3/index.html; Desmet et al., 2009), NNSplice 0.9 version (https://www.fruitfly.org/seq_tools/splice.html; Reese, Eeckman, Kulp, & Haussler, 1997), and MaxEntScan (https://genes.mit.edu/burgelab/maxent/Xmaxentscan_scoreseq_acc.html; Yeo & Burge, 2004) scores correlated well with published data, providing concordant predictions for 23/23 (100%) variants functionally proven to impact splicing and 8/11 (72.7%) with no splicing impact or that were putatively benign based on gnomAD population frequency data (Lek et al., 2016; Supporting Information, Table S2). At least two of these three predictors were capable of detecting the native splice sites for every splice donor/acceptor with the exception of the intron 1 splice donor, and the intron 6 splice acceptor could not be predicted by NNSplice and was weakly predicted by MaxEntScan. The PTEN intron 1 splice donor site has a noncanonical sequence (CCTgtatcc—single nucleotide code) that leaves existing in silico tools unable to predict its presence or interrogate the impact of potential splicing variants in this region. The intron 6 splice acceptor has nearby potential acceptor sites of comparable strength that may compete with the native site. Additionally, only one variant near the intron 1 splice acceptor, PTEN c.80-3C>G, had undergone splicing analysis with no impact observed (Chen, Romigh, Sesock, & Eng, 2017), and in silico predictors were discordant, predicting a splicing impact. Therefore, the EP will not consider applying PP3/BP4 for variants which may impact the intron 1 splice donor or acceptor sites, and will approach variants around the intron 6 splice acceptor with caution.

The group agreed to adopt BP7 (synonymous variant for which splicing prediction algorithms predict no impact) as defined by ACMG/AMP, and extended this criteria to apply to intronic variants at or beyond the +7 or –21 positions. For intronic variants, it was further stipulated that the nucleotide position should not be conserved; conserved positions were defined as having a PhastCons score = 1 or a PhyloP score > 0.1 (Pollard, Hubisz, Rosenbloom, & Siepel, 2010; Siepel et al., 2005).

The EP also maintained PVS1 (null variant in a gene where LOF is a disease mechanism) as defined by ACMG/AMP, applying it to nonsense or frameshift variants predicted to cause nonsense-mediated decay (NMD), canonical splice site variants, and single- or multi-exon deletions. Additionally, internal data from EP laboratory members as well as published literature report de novo variants in individuals with a strong PHTS-related phenotype as 3′ as c.1120_1121dupCC (p.D375fs; Vanderver et al., 2014). Although this variant occurs within exon 9, PTEN’s final exon, and NMD is not predicted to occur, it is predicted to result in the disruption of the C-terminal domain which includes PEST motifs, residues that undergo phosphorylation, and a PDZ domain-binding motif (Wang & Jiang, 2008), leading the EP to set this position as the 3′ boundary for use of PVS1. For protein truncating variants causing disruption 3′ of c.1121 or protein extension, the group agreed to apply PM4 (protein length changes) as currently defined. The group also agreed to apply this criterion to small in-frame deletions or duplications disrupting one of PTEN's catalytic motifs, which include the WPD loop (residues 90–94), P-loop (also described as phosphatase core, residues 123–130), and the TI-loop (residues 166–168) (NP_ 000305.3) (Lee et al., 1999). These three regions also comprise the residues for which PM1 (mutational hot spot and/or critical and well-established functional domain) may be applied for missense variants.

The EP applied PS1 (same amino acid change as a previously established pathogenic variant) as currently defined by ACMG/AMP, and expanded it to include a different nucleotide substitution for an intronic splice site variant if the predicted impact is equal to or greater than the known pathogenic variant per in silico splicing tools. A caveat that caution should be used when applying this criteria to exonic variants causing aberrant splicing was included.

3.2.3 Functional data (BS3 and PS3)

As part of its role as a tumor suppressor, PTEN functions as a phosphatase, catalyzing the conversion of phosphatidylinositol triphosphate (PI(3,4,5)P₃) to PI(4,5)P₂ and leading to inhibition of the AKT pathway. When this phosphatase activity is diminished, over-activation of AKT occurs, driving tumor development (Myers et al., 1998; Stambolic et al., 1998). Numerous PTEN missense variants have been identified which alter the structure of the active site pocket such that phosphatase activity is impacted, and several independent researchers have used this assay as a means of measuring PTEN dysfunction (Han et al., 2000; Mighell, Evans-Dutson, & O'Roak, 2018; Rodríguez-Escudero et al., 2011). The EP therefore felt PS3 (well-established in vitro or in vivo functional studies supportive of a damaging effect) would be suitable for a missense variant found to cause over a 50% reduction in lipid phosphatase activity or any variant causing aberrant splicing on functional interrogation. Decreased PTEN or increased pAKT expression, although not as direct a measure as phosphatase activity, have been significantly associated with presence of pathogenic PTEN variation (Spinelli, Black, Berg, Eickholt, & Leslie, 2015; Tan et al., 2011). In addition to phosphatase activity, pathogenic PTEN variants have been found to disrupt protein cellular localization (Gil et al., 2015; He et al., 2012; Lobo et al., 2009) and lead to aberrant cellular phenotypes, including defective cell migration, proliferation, and invasion (Costa et al., 2015; Malek et al., 2017). Knock-in and knock-out Pten mouse models have also been developed, with mice harboring pathogenic variants or haploinsufficient Pten demonstrating PHTS-related phenotypes and increased tumor burden (Di Cristofano, Pesce, Cordon-Cardo, & Pandolfi, 1998; Mester, Tilot, Rybicki, Frazier, & Eng, 2011; Tilot et al., 2014). The EP elected to apply a supporting level of evidence (PS3_Supporting) for in vitro cellular assays not meeting PS3 and transgenic model organism studies.

Although normal lipid phosphatase activity may suggest retained function, PTEN missense variants have been described which retain lipid phosphatase activity, but lead to dysfunction via loss of protein phosphatase activity, protein instability, abnormal cellular localization, or other mechanisms unrelated to lipid phosphatase activity (Davidson et al., 2010; Gil et al., 2015; Yang et al., 2017). The EP would therefore consider applying BS3_Supporting to variants with one functional study appropriate to the protein domain demonstrating results comparable to wild type. BS3 (well-established in vitro or in vivo functional studies show no damaging effect) may be applied at the strong evidence level for variants with both lipid phosphatase activity and results from a second assay appropriate to the protein domain demonstrating no statistically significant difference from wild type. BS3 may also be applied for intronic or synonymous variants demonstrated to not cause a splicing impact via RNA, mini-gene, or other splicing assays. Phosphatase assays for which BS3/PS3 may be applied must include a catalytic dead control, such as p.C124S, and at least three biological replicates would be required to apply criteria to results of any of these or similar readouts of PTEN function (Costa et al., 2015; Malek et al., 2017).

3.2.4 Segregation data (BS4 and PP1)

The EP adopted the approach being taken by other ClinGen EPs (Gelb et al., 2018; Kelly et al., 2018) and supported by the SVI and other work (Jarvik & Browning, 2016) that three or four meioses be required for application of PP1, with additional meioses meriting higher evidence levels (five or six meioses for moderate and seven or more meioses for strong) based on estimated LOD scores of 0.9, 1.5, and 2.1, respectively (Kelly et al., 2018). The EP further stipulated that cosegregation with disease must be observed across at least two families for application of the criteria at the strong evidence level, to avoid the possibility that a true undiscovered disease allele is present in linkage disequilibrium with the variant under interrogation.

The EP defined that BS4 can be used when a PHTS phenotype is present in several individuals from one side of a family, and the variant in question is found to be inherited from the opposite side. For application of BS4 at its pre-defined strong evidence level, the EP specified that such lack of segregation with disease must be seen in two or more families. A supporting level of evidence (BS4_Supporting) may be used when one such family is identified.

3.2.5 De novo data (PM6 and PS2)

The EP agreed to adopt the definitions of PM6 and PS2 as defined by ACMG/AMP, and decided to follow the SVI-approved approach to apply higher levels of evidence with increasing numbers of de novo probands or to lower the evidence level when phenotypic specificity is lacking (https://www.clinicalgenome.org/site/assets/files/8490/recommendation_ps2_and_pm6_acmgamp_critiera_version_1_0.pdf; Table 1 and Supporting Information, Table S3). In accordance with ACMG/AMP guidance, family history must also be consistent with a de novo event, and the patient's phenotype should be a match for PHTS, but they need not meet the strict criteria for PP4 outlined in the following section.

3.2.6 Phenotype (PS4/PP4)

Given the extreme population rarity of any one specific pathogenic PTEN variant, it is unlikely that a case-control study will find statistically significant results for a specific variant, leading the EP to adopt the second use of PS4 (multiple unrelated individuals with the same phenotype). PHTS causes increased risk for a diverse combination of phenotypes, several of which are specific to this disorder and uncommon in the general population. Phenotype among affected individuals is also highly variable, even within the same family, and is also age- and gender-dependent (Lachlan, Lucassen, Bunyan, & Temple, 2007), necessitating a careful approach whereby PP4 (phenotype specificity) is wrapped into the PS4 criteria. For adults, a scoring system called the Cleveland Clinic (CC) score predicts a priori PTEN mutation risk based on personal history, with findings highly specific to PHTS, such as Lhermitte–Duclos disease or hamartomatous polyps, given higher weight (Tan et al., 2011; Supporting Information, Table S4). The EP decided that either one adult with phenotypic features specific enough to result in a CC score ≥30 (≥80% mutation risk), or two with a score of 25–29 (62–80% mutation risk) would qualify for PS4_Supporting. In other words, an adult with a CC score of 35 would merit 1 phenotype specificity point; an adult with a CC score of 28 would merit 0.5 points.

For children, no similar tool existed, and the CC scoring tool could not be applied given the age-dependent penetrance of many PHTS phenotypes. Therefore, creation of a separate scoring system for children was required. EP members from two U.S.-based testing laboratories and a clinician in the United Kingdom who used a national referral lab for testing contributed de-identified phenotype data from children who had undergone PTEN mutation analysis. Age- and gender-matched pediatric patients were selected with either positive (PATH) or negative (no variants identified) results. The EP phenotype workgroup drafted scores for findings observed in children with PHTS, assigning higher scores to features more specific to PHTS and lower ones to those observed more often in children referred for genetics evaluation and/or testing (Table 2). Using this scoring system, paired t-test identified significantly higher scores for children with PATH variants compared to those with negative PTEN testing (4.45 vs. 2.95, P = 0.009, Supporting Information, Table S5). When a threshold of 5 was applied, 100% specificity was achieved for both boys and girls, leading the EP to adopt this score as the cutoff for application of PS4_Supporting (1 phenotype specificity point) for pediatric cases. Using a cutoff of 4, specificity dropped to 50% due to the high prevalence of autism/developmental delay (DD)/intellectual disability (ID) diagnoses across the cohort, but sensitivity increased to 80% (Supporting Information, Table S6). Thus, similar to the approach accepted for adults, a score of 4 was accepted as appropriate for 0.5 phenotype specificity points, with the caveat that autism/DD/ID may not contribute to the score given the common nature of this phenotype for pediatric genetics referral (Woodward, Alves, & Butler, 1993).

Similar to de novo evidence, EP members chose to apply PS4_Supporting at higher evidence levels when multiple probands meeting these phenotype criteria were identified. Given the strict criteria developed for applying phenotype specificity points, the duplicative approach to reach higher evidence levels for de novo criteria was employed, but with 16 or more specificity points required for a Very Strong evidence level to more closely match the exponential increase in odds for pathogenicity between the Strong and Very Strong evidence levels supported by the SVI (Tavtigian et al., 2018). Table 3 provides a summary of the phenotype specificity scores required for increased levels of evidence with examples for application. The EP also specified that this criterion not be applied when BS1 is met to avoid coincidental accumulation of proband specificity points for variants with high allele frequency.

3.2.7 Allelic data (BS2, BP2, and BP5)

Murine Pten homozygous knock-out or knock-in models demonstrate embryonic lethality (Di Cristofano et al., 1998; Wang et al., 2010). To date, we are not aware of a human individual with homozygosity or compound heterozygosity for pathogenic PTEN variants. Only one homozygous potentially hypomorphic missense variant, p.L182S, has been reported in siblings with a PHTS phenotype by exome analysis whose heterozygous parents were phenotypically normal (Schwerd et al., 2016). However, this variant has not been reported elsewhere. Individuals have been reported with one germline pathogenic PTEN variant and a somatic “second hit,” with affected tissues demonstrating overgrowth, vascular malformations, and lipomatosis reminiscent of Proteus syndrome (Caux et al., 2007). Together these data suggest that germline homozygosity or compound heterozygosity for pathogenic PTEN variants in humans may not be compatible with life or would result in a severe PHTS phenotype. Thus the PTEN EP wished to apply BS2 (observed in a healthy adult individual) when a variant is present in the homozygous state in at least one individual whose homozygous status was confirmed via parental testing or in two individuals without such confirmation, so long as the individual(s) are confirmed as healthy/unaffected persons or do not exhibit features that would be expected for PHTS. A supporting level of evidence may be used when at least two homozygous observations exist in the absence of clinical data, or the criteria for BS2 is met but BS1 (allele frequency greater than expected for disorder) has also been applied. The caveat regarding BS1 was added to ensure a variant could not reach benign status (BS1 + BS2) driven mainly by homozygous occurrences due to population allele frequency lower than the threshold set for application of BA1. BP2 may be applied when a variant is observed in trans with a known PTEN pathogenic variant, or observed in cis or with unknown phase with three or more different pathogenic PTEN variants.

The PTEN EP also agreed to adopt BP5 (variant found in a case with an alternate molecular cause for disease) for two or more co-occurrences with pathogenic variants in a different gene that fully explained the patient's phenotype, but specific circumstances would need to be met in order for a case to be considered for inclusion. First, the variant in the other gene must be considered highly penetrant, with both the individual's age and gender taken into consideration. Additionally, the patient's personal and family history (including up to 2nd degree relatives) should not overlap with features seen in PHTS. As an example, an individual with a personal and family history of breast cancer who harbored a PTEN variant in addition to a pathogenic BRCA2 variant would not apply, because breast cancer is a known risk in PHTS and the PTEN variant might have contributed to this individual's breast cancer risk. However, an individual with a personal and family history of ovarian and pancreatic cancer who was positive for a PTEN variant as well as a pathogenic BRCA2 variant would be considered for BP5 application, as neither ovarian nor pancreatic cancer is associated with PHTS.