Expert specification of the ACMG/AMP variant interpretation guidelines for genetic hearing loss

2.1 Hearing Loss Clinical Domain Working Groups

As a ClinGen Clinical Domain Working Group (CDWG), the Hearing Loss CDWG aims to create a comprehensive, standardized knowledge base of genes and variants relevant to syndromic and nonsyndromic HL. Members were identified and recruited based on their expertise in HL, and are representative of diverse institutions worldwide, spanning Asia, Australia, Europe, and North America. Members include otolaryngologists, clinical geneticists, molecular geneticists, ClinGen biocurators, clinical researchers, and genetic counselors from over 15 institutions. The Hearing Loss CDWG has so far formed two major efforts defined by ClinGen as a Gene Curation Expert Panel, and a Variant Curation Expert Panel (HL-EP).

2.2 Specifications of the ACMG/AMP guidelines

A smaller task team utilized biweekly conference calls in addition to email correspondence to review, specify, and reach consensus for each of the rules within the ACMG/AMP guidelines. Typically, 15–20 members from the Laboratory for Molecular Medicine (LMM), Children's Hospital of Philadelphia (CHOP), ARUP Laboratories, Molecular Otolaryngology and Renal Research Laboratories (MORL), and Mayo Clinic participated on each call. Members with specific areas of expertise (e.g., Pendred syndrome) were also contacted for relevant topics. The chairs, ClinGen biocurators, and coordinator of the HL-EP also utilized weekly conference calls for additional planning and preparation. For the larger CDWG, a quarterly conference call was used to present specifications finalized by the smaller group and to obtain final approval from all members. Upon specifying the ACMG/AMP rules, conference calls and email correspondence were used for the variant pilot using a set of 51 variants from nine common HL genes that represent the varied inheritance patterns, evidence types, and phenotypic spectrum seen across HL cohorts (see Variant Pilot methods below). The final set of specified rules is shown in Table 1.

2.3 Pathogenic allele frequency for recessive HL

The frequencies of the most common pathogenic variants in HL were determined using a cohort of 3,673 patients tested at the LMM. The three most common sequence variants were NM_004004.5:c.35delG (p.Gly12Valfs), NM_004004.5:c.109G>A (p.Val37Ile), and NM_004004.5:c.101T>C (p.Met34Thr) in GJB2, while the fourth variant was NM_206933.2:c.2299delG (p.Glu767Serfs) in the USH2A gene (Table 2). Test-based cohorts for the GJB2 (n = 1,375 probands) or USH2A (n = 1,887 probands) genes were used to calculate the disease population allele frequency of each one of those variants.

The allele frequencies were confirmed using another HL cohort (n = 2,066) tested at the MORL (Iowa City, IA). Although their frequencies were slightly different, the rankings of the four most common pathogenic variants were recapitulated in this cohort (Table 2). The frequency differences can be attributed to the fact that the MORL cohort excluded patients with positive findings upon GJB2 prescreening. Therefore, the higher allele frequency (from the LMM) for each variant was conservatively used in calculating BA1 and BS1 for recessive HL. Our calculations were made under the assumption of Hardy–Weinberg equilibrium and can be obtained using a recently published web application (https://www.cardiodb.org/allelefrequencyapp/) (Whiffin et al., 2017).

2.4 Performance analysis of functional studies

To assess the performance characteristics (sensitivity, specificity, positive predictive value, and negative predictive value) of the functional assays in each of GJB2, SLC26A4, and COCH, the number of true and false variant calls was determined as follows. Variants classified in ClinVar and included in functional assays were reviewed for classification accuracy and then used for informing performance. Pathogenic and likely pathogenic variants that had abnormal assay readout, defined as statistically significant deviation from wild-type variant(s), were labeled as true positives. Normal assay readouts for pathogenic or likely pathogenic variants were considered false negatives. Similarly, benign and likely benign variants with normal assay readouts (i.e., does not significantly deviate from wild-type variants) were counted as true negatives. False positives were benign or likely benign variants with abnormal assay readouts. A summary of true and false variant calls by five functional assays in the above three genes is shown in Supporting Information, Table S1.

2.5 Variant pilot

Following the specification of the ACMG/AMP guidelines, the rules were refined by interpreting a set of 51 variants in GJB2, SLC26A4, USH2A, MYO7A, CDH23, COCH, KCNQ4, MYO6, and TECTA. Each variant was assessed independently by two variant curators, and the variant classification and rules applied were reviewed on conference calls to resolve discrepancies and reach consensus. Curators utilized ClinGen's Variant Curation Interface (https://curation.clinicalgenome.org/) to assess and document the applicable rules for each variant.

3.1 Summary of specifications

The HL-EP recommended specifications for 21 ACMG/AMP rules (Table 1). Four rules had general recommendations on the application of the rule (PS1, PP3, BS4, BP4, and BP5). Seven rules had gene or disease based specifications (PS3, PM1, PM2, PP4, BA1, BS4, and BP2). Seven rules had strength-level specifications (PVS1, PS2, PM3, PM5, PM6, PP1, and BS3). Three rules had both gene and disease based specifications and strength-level specifications (PS4, BS1, and BS2). No changes were recommended for three rules (PM4, BP3, and BP7), and four rules were considered not applicable (PP2, PP5, BP1, and BP6).

Two changes to the ACMG/AMP recommendations were made on how to combine criteria to classify sequence variants. The first is that a likely pathogenic classification can be reached if a variant in a gene associated with autosomal recessive HL meets PVS1 and PM2_Supporting. The second change was that a variant can be classified as likely benign if a variant meets BS1 without valid conflicting evidence that would suggest a pathogenic role.

It should be noted that we do not discuss the technical challenges associated with sequencing some of HL genes, such as homology, GC-rich, and/or repetitive regions, which may affect variant calling and interpretation. False positive variants in these regions can exist in population or disease databases and may result in erroneous classifications. Laboratories should carefully assess the technical quality of any variant before interpretation. A full list of technically challenging regions in 109 HL genes was recently characterized (DiStefano et al., 2018).

3.2 Population data (BA1, BS1, and PM2)

Recessive and dominant forms of HL were distinguished with respect to the interpretation of variant frequency data from the general population. This is mainly due to differences in prevalence, penetrance, and gene contribution. Evidence-based estimates of these attributes are critical for establishing recessive and dominant allele frequency thresholds, at which a benign or pathogenic criterion might be assigned for a given variant. The minor allele frequencies required to apply these criteria are shown in Figure 1. For this work, the ExAC and gnomAD population databases were used, though other databases with a minimum of 2,000 alleles are also sufficient. Caution should be used if the cutoffs specified below are applied on an uncharacterized population or a population with minimal representation in ExAC or gnomAD. Furthermore, although we used ExAC which was the main general population dataset at the time of this study, we expect these databases to continue to grow (as is the case already in gnomAD), and therefore we encourage curators to use the largest datasets in their filtering frequency calculations.

3.4 PS4, PS4_Moderate, and PS4_Supporting: Prevalence in affected individuals statistically increased over controls

Published case-control studies demonstrating a statistically increased presence of a variant in affected individuals compared to race and ancestry-matched controls can be used as strong evidence toward pathogenicity, as outlined in the ACMG/AMP guidelines. For many variants, published case-control studies may not exist, but a case-control comparison to ancestry-matched controls in gnomAD can still provide critical evidence. This is particularly true for recessive variants with relatively high frequency in the general population. The following recommendations were made for variants in which a published case-control study does not exist.

A Chi-squared or Fisher's Exact test using a 2 × 2 contingency table can be performed to determine if the MAF of a variant is statistically higher in cases compared to the general population. This analysis can be used as strong evidence if the MAF of the variant is significantly higher (i.e., the P-value of the Chi-squared or Fisher's Exact test is ≤0.05) in cases than in the general population. It is important to note that to do this comparison, the case and population alleles should be adequately ancestry-matched, the total numbers of positive and negative alleles should be known for both the case and the population cohorts, and the relevant gene must be definitively or strongly associated with HL per the ClinGen Gene Curation framework (Strande et al., 2017).

Large general population databases, such as ExAC or gnomAD, are not true control cohorts and likely contain individuals with HL, given that there is no exclusion for HL. As such, absence of statistical significance should not be used as evidence against pathogenicity.

The ACMG/AMP guidelines allow for the counting of probands as evidence for PS4 in the absence of case-control studies. For autosomal recessive HL, we recommend counting probands through PM3 (see below) where allelic data can be taken into account. For autosomal dominant HL, we adopted the specifications previously published by the ClinGen Cardiomyopathy Expert Panel (CMP-EP) given that dominant HL has similar attributes to cardiomyopathy in terms of penetrance and phenocopies (Kelly et al., 2018). These specifications state that if a variant is absent or extremely rare in the general population (i.e., PM2 is met), identification of ≥ 2, ≥ 6, or ≥ 15 probands with the same variant would reach supporting (PS4_Supporting), moderate (PS4_Moderate), or strong (PS4) evidence, respectively. The rule PM2 must be met to use these PS4 specifications. As noted above, PM2 for autosomal dominant HL was set at ≤ 0.002%, which is slightly lower than the PM2 value of ≤ 0.004% set by the CMP-EP.

4.1 Loss of function variants (PVS1, PVS1_Strong, PVS1_Moderate, and PVS1_Supporting)

The HL-EP and the ClinGen Sequence Variant Interpretation Working Group (SVI) jointly refined this rule to be used for all recessive and haploinsufficient genes across most diseases including HL. This refinement included a detailed guideline for interpreting the PVS1 criteria across all loss of function (LOF) variant types (nonsense, frameshift, canonical splice site variants [+/–1,2], deletions, duplications, and initiation codon variants), while accounting for variant, exon and transcript-specific considerations, and the varied evidence weights. This guideline is available in an accompanying paper in this issue (Abou Tayoun et al., this issue) and will not be discussed further here.

4.2 Variants affecting the same amino acid residue (PS1, PM5, and PM5_Strong)

The ACMG/AMP guideline considers the presence of a known pathogenic variant at the same amino acid residue as either strong or moderate evidence of pathogenicity. This PS1 rule is applied if the variant is a novel nucleotide change but results in the same amino acid change as the known pathogenic variant, and the PM5 rule is applied if the variant is a novel missense change at the same amino acid residue as a known pathogenic missense variant. It should be noted that the ACMG/AMP guidelines recommend assessing whether the variants in question could have an impact at the DNA level, such as through an impact on splicing, before applying PS1 or PM5. The HL-EP added a strength modification to the PM5 rule (PM5_Strong) so that a strong level of evidence can be applied if a novel missense variant is located at an amino acid residue with at least two other pathogenic variants at the same residue.

The HL-EP also specified that PS1 can be applied for DNA variants located at the same nucleotide in the splice consensus sequence as a known pathogenic variant. Based on the most conserved regions known to impact splicing, we defined the splice consensus sequence here as the first base and last three bases of the exon, the –12 through –1 positions in the intron, and the +1 through +6 positions in the intron. The +/–1,2 positions were excluded, given that these positions should be evaluated using PVS1 rules. In addition, splice prediction algorithms must predict a similar or greater effect than the comparator known pathogenic variant.

4.3 Computational predictive tools (PP3, BP4, and BP7)

Computational tools are commonly used to inform variant interpretation. For missense variants, the HL-EP recommends using REVEL, a computational tool that combines predictions from commonly used algorithms for missense variants (Ioannidis et al., 2016). REVEL was selected based on recently published work from members of ClinGen as well as evaluation within the HL-EP and the SVI (Ghosh, Oak, & Plon, 2017). Furthermore, we recommend specific REVEL scores are reached to apply supportive evidence of pathogenicity (PP3) or supportive evidence of benign (BP4). PP3 can be applied with a REVEL score of ≥0.7, given that 95% of benign variants are excluded at this score (Ioannidis et al., 2016). BP4 can be applied with a REVEL score of ≤0.15, given that 95% of pathogenic variants are excluded at this score (Ioannidis et al., 2016). Variants with a REVEL score between 0.15–0.7 will not receive either rule (PP3 or BP4).

The HL-EP recommends using MaxEntScan when assessing potential splicing impacts at the DNA level for missense, silent, and splice consensus sequence variants outside of the canonical ±1 or 2 sites. For variants at the canonical splice sites, PP3 cannot be applied if PVS1 is used (Abou Tayoun et al., this issue).

4.4 Computational and predictive data rules with no changes (PM4, BP3, and BP7)

The HL-EP did not further specify the rules pertaining to protein length changing variants (PM4 and BP3). However, the PVS1 flowchart mentioned above should be carefully followed for large in-frame deletions or insertions, particularly those affecting one or more exons. Silent variants with no predicted splicing impact can have BP7 applied, with no specified changes.

5.1 Functional studies (PS3, PS3_Supporting, and BS4_Supporting)

Based on systematic assessment of the literature, the HL-EP determined that no functional assays met a strong level of evidence for or against pathogenicity, except for a variant-specific knock-in mouse model. This is mainly due to the lack of well-established functional assays with enough data to assess their validity in predicting variant pathogenicity in most HL genes. Except for the functional assays in the COCH, GJB2, and SLC26A4 genes described below, a PS3_Supporting or BS3_Supporting can be appropriately used for evaluating variants in other genes not specified here if the assay is well-validated with positive and negative controls and the results are consistent with a known protein function.

5.2 GJB2

This gene, which encodes the gap junction protein, connexin 26, is the most common cause of genetic HL and has been well studied. Several research groups have measured the impact of genetic variants on its function primarily using electrical coupling or dye diffusion assays in Xenopus oocytes or mammalian cell systems (Bicego et al., 2006; Choung, Moon, & Park, 2002; D’Andrea et al., 2002; Lee, Derosa, & White, 2009; Mani et al., 2009). In both settings, a mutant and wild-type control cDNA are first transfected into separate oocytes or cell lines. In the electrical coupling assay, connexin 26 (encoded by GJB2) protein function is examined by altering the voltage in one cell and then measuring a current change in a neighboring cell through patch clamping or ion injection and dye sequestering (Ambrosi et al., 2013; Bruzzone et al., 2003; Haack et al., 2006; Mese, Londin, Mui, Brink, & White, 2004; Zhang et al., 2005). In the dye diffusion assay, movement of fluorescent dyes between neighboring cells is quantified as a measure of connexin 26 protein function. Variants disrupting connexin 26 function are expected to significantly reduce electrical coupling or dye transfer compared to control cDNA.

We searched the literature for variants that were assessed by these two assays, and that were validly classified as pathogenic, likely pathogenic, benign, or likely benign. We then determined the performance characteristics for each assay as described in the methods section and shown in Supporting Information, Table S7. Because both assays demonstrated a high sensitivity and positive predictive value, the HL-EP recommends using PS3_Moderate for any novel GJB2 variant resulting in a statistically significant decrease or absence of current or dye transfer compared to wild-type run under the same experimental conditions. On the other hand, given the limited number of benign variants used as negative controls (Supporting Information, Table S1), the HL-EP was less confident regarding the ability of either assay to accurately predict benign effects, and in such cases, using BS3_Supporting is more appropriate.

It should be noted that in dye diffusion assays, a negative control (e.g., nontransfected or water injected control) should not display dye diffusion. In electrical coupling assays, water injected controls should display negligible currents at all potentials.

5.3 SLC26A4

This gene encodes an anion transporter, and its function has been assayed by measuring transport of radioactive anion isotopes (e.g., iodide or chloride) in cells expressing normal or mutant protein subunits (Bizhanova, Chew, Khuon, & Kopp, 2011; Choi, Stewart, et al., 2009; Dossena, Bizhanova, et al., 2011; Dossena, Vezzoli, et al., 2006; Gillam et al., 2004; Ishihara et al., 2010; Palos et al., 2008; Reimold et al., 2011; Scott et al., 2000; Yuan et al., 2012). Alternatively, other groups have used halide-sensitive fluorescent probes to measure anion transport (Cirello et al., 2012; Dossena, Bizhanova, et al., 2011; Dossena, Nofziger, et al., 2011; Dossena, Rodighiero, et al., 2006; Fugazzola et al., 2007; Jang et al., 2014; Pera et al., 2008). An assay readout is considered abnormal if it is statistically different than the wild-type vector in the same experiment where a negative control (empty vector) shows absent immunofluorescent staining compared to wild type. In both radioactive and fluorescence based assays, several cell lines usually including Xenopus oocytes, mammalian, or human cells can be used to express relevant SLC26A4 variants. As with GJB2, we determined performance characteristics of both assays by reviewing the results from valid pathogenic/likely pathogenic and benign/likely benign variants in ClinVar (Supporting Information, Tables S1 and S7). Again, there was a limited number of benign variants to assess the ability of both assays to predict benign effects (Supporting Information, Table S1). Furthermore, compared to GJB2, the two assays had lower positive predictive values (Supporting Information, Table S7). Therefore, we recommend using PS3_Supporting or BS3_Supporting if a pathogenic or a benign effect, respectively, was suggested by either assay.

5.4 COCH

This gene encodes cochlin, which is normally secreted into the extracellular matrix of the inner ear (Fransen et al., 1999; Robertson et al., 2006; Robertson et al., 1998). Immunofluorescence and western blotting techniques have been used to assay cochlin secretion, localization, and posttranslational modifications subsequent to expression in bacterial cells, such as Escherichia coli or mammalian cell lines (Bae et al., 2014; Cho et al., 2012; Grabski et al., 2003; Jung, Kim, Lee, Yang, & Choi, 2015). Compared to wild-type cochlin, abnormal patterns include absence of extracellular deposition (i.e., no secretion), intracellular aggregation (i.e., in the endoplasmic reticulum and Golgi), and stable dimerization due to inappropriate disulfide bond formation interfering with protein secretion and/or function (Cho et al., 2012; Nagy et al., 2004; Street et al., 2005; Yao, Py, Zhu, Bao, & Yuan, 2010). Based on the high sensitivity and positive predictive value (Supporting Information, Tables S1 and S7) of these protein assays, we recommend using PS3_Moderate if a variant exhibits an abnormal secretion, dimerization, or localization pattern compared to wild-type protein. On the other hand, due to a dearth of benign variants tested by these assays, we recommend using only BS3_Supporting if a variant shows a comparable pattern to wild-type cochlin using these protein assays. No criteria should be applied if multiple assay results do not agree.

5.5 Mutational hot spots or functional domains (PM1)

The HL-EP identified one gene, KCNQ4, in the variant pilot set that harbored a functional domain for which PM1 can be applied. Missense variants located in the pore forming region (amino acids 270–297) of KCNQ4 are eligible for PM1, based on this region's extreme intolerance to variation in the general population and significant enrichment of pathogenic missense variants observed in individuals with dominant sensorineural HL (Naito et al., 2013).

However, a systematic review of mutational hot spots or functional domains for all genes associated with HL was not performed by the HL-EP. We acknowledge that PM1 can likely be applied to other regions, such as variants impacting glycine residues in the Gly-X-Y motifs of the collagen genes (COL11A2, COL4A3, COL4A4, and COL4A5), which are essential for the triple-helical structure formation (Chakchouk et al., 2015; Naito, Kawai, Nomura, Sado, & Osawa, 1996; Savige et al., 2016; Tryggvason, Zhou, Hostikka, & Shows, 1993).

8.1 PM3, PM3_Strong, and PM3_VeryStrong: In trans with a pathogenic variant (recessive disorders)

For recessive disorders, identifying a variant in trans with a pathogenic variant on the second allele in an affected individual is considered evidence toward pathogenicity. This was previously defined in the ACMG/AMP guidelines as moderate evidence (PM3; Richards et al., 2015). However, if the variant under consideration is found in multiple probands with pathogenic variants on the other allele, the strength of that evidence should be considered greater. Therefore, in conjunction with the SVI, we developed a scoring system to account for the number of compound heterozygous and homozygous probands identified with a variant, as shown in Tables 8 and 9. Each instance that a variant is found in a proband, either in homozygosity or compound heterozygosity with a second variant, is given a specified number of points, and the total points corresponds to the strength of PM3 that can be applied. Other scenarios, such as homozygous variants in consanguineous families, and variants of unknown phase, were also taken into account.

Furthermore, if a variant has been observed in many probands, each with a different pathogenic variant on the other allele, the likelihood that the variants are in trans is higher even if phasing has not been performed in every case, assuming the gene has been fully sequenced to detect all variants. In these cases, point values can be adjusted based on confidence of in trans occurrence.

8.2 BS2: Observation in controls inconsistent with disease penetrance

BS2 can be utilized when a variant is observed in an adult with normal hearing tests when the HL is expected to be fully penetrant from birth or prelingually. For recessive HL, this would be applicable if the variant is observed in trans with a known pathogenic variant or if the variant occurs in the homozygous state. However, we do not recommend using BS2 if the known pathogenic variant observed in the unaffected individual is a hypomorphic or low penetrance allele, such as the NM_004004.5:p.(Val37Ile) or NM_004004.5:p.(Met34Thr) variants in GJB2. By definition, these hypomorphic or low penetrance alleles could result in normal or low normal hearing thresholds.

For autosomal dominant HL, the variant would be heterozygous. However, BS2 should not be implemented when variable onset and severity of HL has been associated to the gene in which the variant occurs.

8.3 BP2: Observation in trans with a pathogenic variant for dominant disorders or observation of a variant in cis with a pathogenic variant

The identification of a variant in cis with a pathogenic variant is supportive evidence for a benign interpretation. However, caution is recommended for the application of this rule for a case with an in trans observation of a dominant pathogenic variant. Careful assessment should include whether an earlier onset or more severe phenotype is possible, which could be consistent with an in trans pathogenic variant.