Standards and Guidelines for the Interpretation of Sequence Variants: A Joint Consensus Recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology

Terminology

A mutation is defined as a permanent change in the nucleotide sequence, while a polymorphism is defined as a variant with a frequency above 1%. However, the terms “mutation” and “polymorphism”, which have been used widely, often lead to confusion due to incorrect assumptions of pathogenic and benign effects respectively. Thus, it is recommended that both terms be replaced by the term “variant” with the following modifiers: (1) pathogenic, (2) likely pathogenic, (3) uncertain significance, (4) likely benign, or (5) benign. While these modifiers may not address all human phenotypes, they comprise a five-tier system of classification for variants relevant to Mendelian disease as addressed in this guidance. It is recommended that all assertions of pathogenicity (including "likely pathogenic") be reported with respect to a condition and inheritance pattern (e.g. c.1521_1523delCTT (p.Phe508del), pathogenic, cystic fibrosis, autosomal recessive).

It should be noted that some laboratories may choose to have additional tiers (e.g. sub-classification of variants of uncertain significance (VUSs), particularly for internal use) and this practice is not considered inconsistent with these recommendations. It should also be noted that the terms recommended here differ somewhat from the current recommendations for classifying copy number variants (CNVs) detected by cytogenetic microarray.⁶ The schema recommended for CNVs, while also 5-tiers, uses "uncertain clinical significance - likely pathogenic” and “uncertain clinical significance - likely benign”. The majority of the workgroup was not supportive of using “uncertain significance” to modify the terms “likely pathogenic” or “likely benign" given that it was felt that the criteria presented here to classify variants into the “likely” categories included stronger evidence than outlined in the CNV guideline and combining these two categories would create confusion for the healthcare providers and individuals receiving clinical reports. However, it was felt that the use of the term "likely" should be restricted to variants where the data supports a high likelihood that it is pathogenic or a high likelihood that it is benign. Although there is no quantitative definition of the term "likely", guidance has been proposed in certain variant classification settings. However, a survey of the community during an ACMG open forum suggested a much wider range of uses of the term ‘likely’. Recognizing this, we propose that the terms ‘likely pathogenic’ and ‘likely benign’ be used to mean greater than 90% certainty of a variant either being disease-causing or benign in order to provide laboratories with a common, albeit arbitrary, definition. Similarly, the International Agency for Research on Cancer (IARC) guideline² supports a 95% level of certainty of pathogenicity, but the workgroup (confirmed by feedback in the ACMG open forum) felt that clinicians and patients were willing to tolerate a slightly higher chance of error leading to the 90% decision. It should also be noted that at present most variants do not have data to support a quantitative assignment of variant certainty to any of the five categories given the heterogeneous nature of most diseases. It is hoped that over time experimental and statistical approaches will be developed to objectively assign pathogenicity confidence to variants and that more rigorous approaches to defining what the clinical community desires in terms of confidence will more fully inform terminologies and likelihoods.

The use of new terminologies may require education of the community. Professional societies are encouraged to engage in educating all laboratories as well as healthcare providers on the use of these terms and laboratories are also encouraged to directly educate their ordering physicians.

Nomenclature

A uniform nomenclature, informed by a set of standardized criteria, is recommended to ensure the unambiguous designation of a variant and enable effective sharing and downstream use of genomic information. A standard gene variant nomenclature (http://www.hgvs.org/mutnomen) is maintained and versioned by the Human Genome Variation Society (HGVS)⁷ and its use is recommended as the primary guideline for determining variant nomenclature except as noted.⁶ Laboratories should note the version being used in their test methods. Tools are available to provide correct HGVS nomenclature for describing variants (https://mutalyzer.nl).⁸ Clinical reports should include sequence reference(s) to ensure unambiguous naming of the variant at the DNA level as well as provide coding and protein nomenclature to assist in functional interpretations (e.g., “g.” for genomic sequence, “c.” for coding DNA sequence, “p.” for protein, “m.” for mitochondria, etc.). The coding nomenclature should be described using the “A” of the ATG translation initiation codon as position number 1. Where historical alternate nomenclature has been used, current nomenclature should be used with an additional notation of the historical naming. The reference sequence should be complete and derived from either the NCBI RefSeq database (http://www.ncbi.nlm.nih.gov/RefSeq/)⁹ with the version number or the Locus Reference Genomic (LRG) database (http://www.lrg-sequence.org).¹⁰ Genomic coordinates should be used and defined according to a standard genome build (e.g. hg19) or a genomic reference sequence that covers the entire gene (including the 5' and 3' untranslated regions (UTRs) and promoter). A reference transcript for each gene should be used and provided in the report when describing coding variants. The transcript should either represent the longest known transcript and/or most clinically relevant transcript. Community-supported reference transcripts can often be identified through LRG¹⁰, CCDS Database¹¹, Human Gene Mutation Database (http://www.hgmd.cf.ac.uk), ClinVar (http://www.ncbi.nlm.nih.gov/clinvar) or a locus-specific database. However, laboratories should evaluate the impact of the variant on all clinically relevant transcripts including alternate transcripts that contain additional exons or extended untranslated regions when there are known variants in these regions that are clinically interpretable.

Not all types of variants (e.g., complex variants) are covered by the HGVS recommendations, but possible descriptions for complex variants have been reported.^7,12 In addition, this ACMG recommendation supports three specific exceptions to the HGVS nomenclature rules: 1) "X" is still considered acceptable for use in reporting nonsense variants in addition to the current HGVS recommendation of "*" and "Ter"; 2) it is recommended that exons be numbered according to the chosen reference transcript used to designate the variant; and 3) the term “pathogenic” is recommended instead of “affects function” as clinical interpretation is typically evaluating pathogenicity directly.

Literature and Database Use

A large number of databases contain a growing number of variants that are continuously being discovered in the human genome. When classifying and reporting a variant, clinical laboratories may find valuable information in databases, as well as in published literature. As noted above, sequence databases can also be used to identify appropriate reference sequences. Databases can be useful for gathering information, but should be used with caution.

Population databases (Table 1) are useful in obtaining the frequencies of variants in large populations. Population databases cannot be assumed to include only healthy individuals, and are known to contain pathogenic variants. These population databases do not contain extensive information regarding the functional effect of these variants or any possible associated phenotypes. When using population databases, one must determine whether healthy or disease cohorts were used, and if possible, whether more than one individual in a family was included, and the age range of the subjects.

Disease databases (Table 1) primarily contain variants found in patients with disease and assessment of the variants’ pathogenicity. Disease and gene-specific databases often contain variants that are incorrectly classified including incorrect claims published in peer-reviewed literature because many databases do not perform primary review of evidence. When using disease databases, it is important to consider how patients were ascertained, as described below.

When using databases, clinical laboratories should: (1) determine how frequently the database is updated, whether data curation is supported and what methods were used for curation; (2) confirm the use of HGVS nomenclature and determine the genome build and transcript references used for naming variants; (3) determine the degree to which data is validated for analytical accuracy (e.g. low pass next generation sequencing versus Sanger-validated variants) and evaluate any quality metrics that are provided to assess data accuracy, which may require reading associated publications; and (4) determine the source and independence of the observations listed.

Variant assessment also includes searching the scientific and medical literature. Literature using older nomenclature and classification or based on a single observation should be used with caution. In identifying individuals and families with a variant, along with associated phenotypes, it is important to consider how patients were ascertained. This caveat is important when assessing data from publications as affected individuals and related individuals are often reported multiple times depending on the context and size of the study. This may be due to authorship overlap, inter-laboratory collaborations, or a proband and family members being followed across different clinical systems. This may mistakenly lead to duplicate counting of affected patients and a false increase in variant frequency. Overlapping authorship or institutions is the first clue to the potential for overlapping datasets.

Clinical laboratories should implement an internal system to track all sequence variants identified in each gene and clinical assertions when reported. This is important for tracking genotype-phenotype correlations and the frequency of variants in affected and normal populations. Clinical laboratories are encouraged to contribute to variant databases, such as ClinVar, including clinical assertions and evidence used for the variant classification, to aid in the continued understanding of the impact of human variation. Clinical information should be provided whenever possible, following Health Insurance Portability and Accountability Act (HIPAA) regulations for privacy. Clinical laboratories are encouraged to form collaborations with clinicians to provide clinical information to better understand how genotype influences clinical phenotype and resolve differences in variant interpretation between laboratories. Because of the great potential to aid clinical laboratory practice, efforts are underway for clinical variant databases to be expanded and standardized. Standardization will provide easier access to updated information as well as facilitating submission from the clinical laboratory. For example, the ClinVar database allows for the deposition of variants with clinical observations and assertions, with review status tracked to enable a more transparent view of the levels of quality of the curation.

Computational (In Silico) Predictive Programs

A variety of in-silico tools, both publicly and commercially available, can aid in the interpretation of sequence variants. The algorithms used by each tool may differ, but can include determination of the effect of the sequence variant at the nucleotide and amino acid level including determination of the effect of the variant on the primary and alternative gene transcripts, other genomic elements, as well as the potential impact of the variant on the protein. The two main categories of such tools include those that predict whether a missense change is damaging to the resultant protein function or structure, and those that predict if there is an effect on splicing (Table 2). Newer tools are beginning to address additional non-coding sequences.¹³

The impact of a missense change depends on criteria such as the evolutionary conservation of an amino acid or nucleotide, the location and context within the protein sequence, and the biochemical consequence of the amino acid substitution. The measurement of one, or a combination, of these criteria is used in various in-silico algorithms that assess the predicted impact of a missense change. Several efforts have evaluated the performance of available prediction software, to compare them to each other and to assess their ability to predict “known” disease-causing variants.^14–17 In general, most algorithms for missense variant prediction are 65–80% accurate when examining known disease variants.¹⁶ Most tools also tend to have low specificity, resulting in over-prediction of missense changes as deleterious and are not as reliable at predicting missense variants with a milder effect,¹⁸ Some of the more commonly used in-silico tools for missense variant interpretation in clinical laboratories include PolyPhen2¹⁹, SIFT²⁰ and MutationTaster²¹. A list of missense variant prediction in-silico tools can be found in Table 2.

Multiple software programs have been developed to predict splicing as it relates to the creation or loss of splice sites at the exonic or intronic level.²² In general, splice site prediction tools have increased sensitivity (~90–100%) relative to specificity (~60–80%) in predicting splice site abnormalities.^23,24 A list of some of the commonly used in-silico tools for splice site variant interpretation is provided in Table 2.

While many of the different software programs use different algorithms for their predictions, they have similarities in their underlying basis; therefore, the combination of predictions from different in-silico tools are considered as a single piece of evidence in sequence interpretation as opposed to independent pieces of evidence. The use of multiple software programs for sequence variant interpretation is also recommended because the different programs each have their own strengths and weaknesses depending on the algorithm and in many cases performance can vary by the gene and protein sequence. However these are only predictions, and their use in sequence variant interpretation should be used carefully. It is not recommended that these predictions be used as the sole source of evidence to make a clinical assertion.

PVS1 Null variants

Certain types of variants (e.g. nonsense, frameshift, canonical +/−1 or 2 splice sites, initiation codon, single exon or multi-exon deletion) can often be assumed to disrupt gene function by leading to complete absence of the gene product by lack of transcription or nonsense-mediated decay of an altered transcript. However, one must exercise caution in classifying these variants as pathogenic by considering the following principles:

1. When classifying such variants as pathogenic, one must ensure that null variants are a known mechanism of pathogenicity consistent with the established inheritance pattern for the disease. For example, there are genes for which only heterozygous missense variants cause disease and null variants are benign in a heterozygous state (e.g. many hypertrophic cardiomyopathy genes). A novel heterozygous nonsense variant in the MYH7 gene would not be considered pathogenic for dominant hypertrophic cardiomyopathy based solely on this evidence whereas a novel heterozygous nonsense variant in the CFTR gene would likely be considered a recessive pathogenic variant.

2. One must also be cautious when interpreting truncating variants downstream of the most 3’ truncating variant established as pathogenic in the literature. This is especially true if the variant occurs in the last exon or in the last 50 bps of the penultimate exon such that nonsense-mediated decay²⁶ would not be predicted and there is a higher likelihood of an expressed protein. However, the length of the predicted truncated protein would also factor into the pathogenicity assignment and such variants cannot be interpreted without a functional assay.

3. For splice site variants, the variant may lead to exon skipping, shortening or inclusion of intronic material, due to alternative donor/acceptor site usage or creation of new sites. Although splice site variants are predicted to lead to a null effect, confirmation of impact requires functional analysis by either RNA or protein analysis. One must also consider the possibility of an in-frame deletion/insertion which could retain the critical domains of the protein, and hence lead to either a mild or neutral effect with a minor length change (PM4) or a gain-of-function effect.

4. It is important to consider the presence of alternate gene transcripts and understand which are biologically relevant, and in which tissues the products are expressed. If a truncating variant is confined to only one or not all transcripts, one must be cautious about over-interpreting variant impact given the presence of the other protein isoforms.

5. One must also be cautious in assuming that a null variant will lead to disease if found in an exon where no other pathogenic variants have been described given the possibility that the exon may be alternatively spliced. This is particularly true if the predicted truncating variant is identified as an incidental finding (unrelated to the indication for testing) given the low prior likelihood of finding a pathogenic variant in that setting.

PS1 Same amino acid change

In most cases, when one missense variant is known to be pathogenic, a different nucleotide change that results in the same amino acid [e.g. c.34G>C (p.Val12Leu) and c.34G>T (p.Val12Leu)] can also be assumed to be pathogenic, particularly if the mechanism of pathogenicity is through altered protein function. However, it is important to assess the possibility that the variant may act directly through the specific DNA change (e.g. through splicing disruption as assessed through at least computational analysis) instead of through the amino acid change, in which case the assumption of pathogenicity may no longer be valid.

PS2 PM6 De novo variants

A variant observed to have arisen de novo (parental samples test negative), is considered strong support for pathogenicity if the following conditions are met:

1. Both parental samples were shown through identity testing to be the biological parents of the patient. Note: PM6 applies if identity is assumed but not confirmed.

2. The patient has a family history of disease that is consistent with de novo inheritance (e.g. unaffected parents for a dominant disorder). However, it is possible that more than one sibling may be affected due to germline mosaicism.

3. The phenotype in the patient matches the gene’s disease association with reasonable specificity. For example, this argument is strong for a patient with a de novo variant in the NIPBL gene who has distinctive facial features, hirsutism and upper limb defects (i.e. Cornelia de Lange syndrome) whereas it would be weaker for a de novo variant found in exome sequencing in a child with non-specific features such as developmental delay.

PS3 BS3 Functional studies

Functional studies can be a powerful tool in support of pathogenicity; however, not all functional studies are effective in predicting an impact on a gene or protein function. For example, certain enzymatic assays offer well-established approaches to assess the impact of a missense variant on enzymatic function in a metabolic pathway (e.g. α-galactosidase enzyme function). On the other hand, some functional assays may be less consistent predictors of the effect of variants on protein function. To assess the validity of a functional assay, one must consider how closely the functional assay reflects the biological environment. For example, assaying enzymatic function directly from biopsied tissue from the patient or an animal model provides stronger evidence than expressing the protein in vitro. Likewise, evidence is stronger if the assay reflects the full biological function of the protein (e.g. substrate breakdown by an enzyme) as compared to one component of function (e.g. ATP hydrolysis for a protein with additional binding properties). Validation, reproducibility and robustness data that assess the analytical performance of the assay and account for specimen integrity, which can be affected by the method and time of acquisition, as well as storage and transport, are important factors to consider. These factors are mitigated in the case of an assay in a CLIA laboratory developed test or commercially available kit. Assays that assess the impact of variants at the mRNA level can be highly informative when evaluating the effects of variants at splice junctions and within coding sequences, untranslated regions, as well as deeper intronic regions (e.g. mRNA stability, processing or translation). Technical approaches include direct analysis of RNA and/or cDNA derivatives and in vitro minigene splicing assays.

PS4 PM2 BA1 BS1 BS2 Variant Frequency and Use of Control Populations

Assessing the frequency of a variant in a control or general population is useful in assessing its potential pathogenicity. This can be accomplished by searching publicly available population databases (e.g. 1000 Genomes, NHLBI Exome Sequencing Project (ESP) Exome Variant Server, Exome Aggregation Consortium (ExAC); see Table 1) as well as using race-matched control data that are often published in the literature. The ESP dataset is useful for Caucasian and African American populations and has coverage data to determine if a variant is absent. Although the 1000 Genomes data cannot be used to assess the absence of a variant, it has a broader representation of different racial populations. More recently ExAC released allele frequency data from >60,000 exomes from a diverse set of populations which includes approximately two-thirds of the ESP data. In general, an allele frequency in a control population that is greater than expected for the disorder (Table 6) is considered strong support for a benign interpretation for a rare Mendelian disorder (BS1) or if over 5% then it is considered as stand-alone support (BA1). Furthermore, if the disease under investigation is fully penetrant at an early age and the variant is observed in a well-documented healthy adult individual for a recessive (homozygous), dominant (heterozygous), or X-linked (hemizygous) condition then this is considered strong evidence for a benign interpretation (BS2). If the variant is absent, one should confirm that the read depth in the database is sufficient for an accurate call at the variant site. If a variant is absent from (or below the expected carrier frequency if recessive) a large general population or a control cohort (>1000 individuals) and the population is race-matched to the patient harboring the identified variant, then this observation can be considered a moderate piece of evidence for pathogenicity (PM2). However, many benign variants are "private," (unique to individuals or families) and therefore absence in a race-matched population is not considered sufficient or even strong evidence for pathogenicity.

The use of population data for case-control comparisons is most useful when the populations are well-phenotyped, have large frequency differences and the Mendelian disease under study is early-onset. Patients referred to a clinical laboratory for testing are likely to include individuals sent to “rule-out” a disorder, and thus may not qualify as well-phenotyped cases. When using a general population as a control cohort, the presence of individuals with subclinical disease is always a possibility. However, in both of these scenarios, a case control comparison will be underpowered with respect to detecting a difference and as such, showing a statistically significant difference can still be assumed to provide supportive evidence for pathogenicity as noted above. In contrast, the absence of a statistical difference, particularly with extremely rare variants and less penetrant phenotypes, should be interpreted cautiously.

Odds Ratios (OR) or Relative Risk (RR) is a measure of association between a genotype (i.e. the variant is present in the genome) and a phenotype (i.e. affected with the disease/outcome) and can be used for either Mendelian diseases or complex traits. In this guideline we are only addressing its use in Mendelian disease. While RR is different from OR, RR asymptotically approaches OR for small probabilities. OR of 1.0 means that the variant does not affect the odds of having the disease, values above 1.0 mean there is an association between the variant and the risk of disease, and those below 1.0 mean there is a negative association between the variant and the risk of disease. In general, variants with a modest Mendelian effect size will have an OR of 3 or greater, while highly penetrant variants will have very high ORs (example: APOE E4/E4 homozygotes compared to E3/E3 homozygotes have an OR of 13 (www.tgen.org/data/neurogenomics). However, the confidence interval (CI) around the OR is as important as the measure of association itself. If the CI includes 1.0 (e.g. OR= 2.5, CI = 0.9–7.4), there is little confidence in the assertion of association. In the above APOE example the CI was approximately 10–16. Very simple OR calculators are available on the web (e.g. http://www.hutchon.net/ConfidOR.htm and http://easycalculation.com/statistics/odds-ratio.php).^27,28

PM1 Mutational hot spot and/or critical and well-established functional domain

Certain protein domains are known to be critical to protein function and all missense variants identified to date in these domains have been shown to be pathogenic. These domains must also lack benign variants. In addition, mutational hotspots in less well characterized regions of genes are reported in which pathogenic variants in one or several nearby residues have been observed with greater frequency. Either evidence can be considered moderate evidence of pathogenicity.

PM3 BP2 Cis/trans testing

Testing parental samples to determine whether the variant occurs in cis (the same copy of the gene) or in trans (different copies of the gene) can be important for assessing pathogenicity. For example, when two heterozygous variants are identified in a gene for a recessive disorder, if one variant is known to be pathogenic, then determining that the other variant is in trans can be considered moderate evidence for pathogenicity of the latter variant (PM3). In addition, this evidence could be upgraded to strong if there are multiple observations of the variant in trans with other pathogenic variants. However, if the variant is present in the general population, a statistical approach would be needed to control for random co-occurrence. In contrast, finding the second variant in cis would be supporting evidence, though not definitive, for a benign role (BP2). In the case of uncertain pathogenicity of two heterozygous variants identified in a recessive gene, then the determination of the cis versus trans nature of the variants does not necessarily provide additional information with regard to pathogenicity of either variant. However, the likelihood that both copies of the gene are impacted is reduced if the variants are found in cis.

In the context of dominant disorders, the detection of a variant in trans with a pathogenic variant can be considered supporting evidence for a benign impact (BP2) or in certain well-developed disease models, may even be considered stand-alone evidence as has been validated for use in assessing CFTR variants.³

PM4 BP3 Protein length changes due to in-frame deletions/insertions and stop losses

The deletion or insertion of one or more amino acids as well as the extension of a protein by changing the stop codon to an amino acid codon (e.g. a stop loss variant) is more likely to disrupt protein function as compared to a missense change alone due to length changes in the protein. Therefore, in-frame deletions/insertions and stop losses are considered moderate evidence of pathogenicity. The larger the deletion, insertion or extension, and the more conserved the amino acids are in a deleted region, then the more substantial is the evidence to support pathogenicity. In contrast, small in-frame deletions/insertions in repetitive regions, or regions that are not well conserved in evolution, are less likely to be pathogenic.

PM5 Novel missense at the same position

A novel missense amino acid change occurring at the same position as another pathogenic missense change (e.g. Trp38Ser and Trp38Leu) is considered moderate evidence but cannot be assumed to be pathogenic. This is especially true if the novel change is more conservative compared to the established pathogenic missense variant. Also, the different amino acid change could lead to a different phenotype. For example, different substitutions of the Lys650 residue of the FGFR3 gene are associated with a wide range of clinical phenotypes: p.Lys650Gln or p.Lys650Asn causes mild hypochondroplasia, p.Lys650Met causes severe achondroplasia with developmental delay and acanthosis nigricans, and thanatophoric dysplasia type 2, a lethal skeletal dysplasia, arises from p.Lys650Glu.

PP1 BS4 Segregation analysis

Care must be taken in using segregation of a variant in a family as evidence for pathogenicity. In fact, segregation of a particular variant with a phenotype in a family is evidence for linkage of the locus to the disorder, though not evidence of pathogenicity of the variant itself. A statistical approach has been published^29,30 with the caveat that the identified variant may be in linkage disequilibrium with the true pathogenic variant in that family. Statistical modeling takes into account age-related penetrance and phenocopy rates, with advanced methods also incorporating in-silico predictions and co-occurrence with a known pathogenic variant into a single quantitative measure of pathogenicity.³¹ Distant relatives are important to include, because they are less likely to have both the disease and the variant by chance than members within a nuclear family. Full gene sequencing (including entire introns and 5’ and 3’ UTRs) may provide greater evidence that another variant is not involved or identify additional variants to consider as possibly causative. Unless the genetic locus is evaluated very carefully, one risks misclassifying a non-pathogenic variant as pathogenic.

When a specific variant in the target gene is observed to segregate with a phenotype or disease in multiple affected family members and multiple families from diverse ethnic backgrounds, linkage disequilibrium and ascertainment bias are less likely to confound the evidence for pathogenicity. In this case, this criterion may be taken as Moderate or Strong evidence, depending on the extent of segregation, rather than Supporting evidence.

On the other hand, lack of segregation of a variant with a phenotype provides strong evidence against pathogenicity. Careful clinical evaluation is needed to rule out mild symptoms of reportedly unaffected individuals as well as possible phenocopies (affected individuals with disease due to a non-genetic or different genetic cause). Also biological family relationships need to be confirmed to rule out adoption, non-paternity, sperm and egg donation and other non-biological relationships. Decreased and age-dependent penetrance also must be considered to ensure that asymptomatic family members are truly unaffected.

Statistical evaluation of co-segregation may be difficult in the clinical laboratory setting. If appropriate families are identified, clinical laboratories are encouraged to work with experts in statistical or population genetics to ensure proper modeling and to avoid incorrect conclusions of the relevance of the variant to the disease.

PP2 BP1 Variant spectrum

Many genes have a defined spectrum of pathogenic and benign variation. For genes in which missense variation is a common cause of disease and there is very little benign variation observed in the gene, a novel missense variant can be considered supporting evidence for pathogenicity (PP2). In contrast, for genes in which truncating variants are the only known mechanism of variant pathogenicity, missense variants can be considered supporting evidence for a benign impact in these genes (BP1). For example, truncating variants in ASPM are the primary type of pathogenic variant in this gene that causes autosomal recessive primary microcephaly and the gene has a high rate of missense polymorphic variants. Therefore missense variants in ASPM can be considered to have this line of supporting evidence for a benign impact.

PP3 BP4 Computational (in silico) data

It is important not to overestimate computational evidence, particularly given that different algorithms may rely on the same (or similar) data to support predictions and most algorithms have not been validated against well-established pathogenic variants. In addition, the algorithms can have vastly different predictive capabilities for different genes. If all of the in silico programs tested agree on the prediction, then this evidence can be counted as supporting. However, if in silico predictions disagree, then this evidence should not be used in classifying a variant. The observation of the variant amino acid change being present in multiple non-human mammalian species in an otherwise well-conserved region suggesting the amino acid change would not compromise function can be considered strong evidence for a benign interpretation. However, one must be cautious about assuming a benign impact in a non-conserved region if the gene has recently evolved in humans (e.g. genes involved in immune function).

PP4 Using phenotype to support variant claims

In general, the fact that a patient has a phenotype that matches the known spectrum of clinical features for a gene, is not considered evidence for pathogenicity given that nearly all patients tested in disease-targeted tests have the phenotype in question. However, if the following criteria are met, the patient’s phenotype can be considered supporting evidence: (1) the clinical sensitivity of testing is high with most patients testing positive for a pathogenic variant in that gene; (2) the patient has a well-defined syndrome with little overlap with other clinical presentations (e.g. Gorlin syndrome including basal cell carcinoma, palmoplantar pits, odontogenic keratocysts); (3) the gene is not subject to substantial benign variation which can be determined through large general population cohorts (e.g. ESP); and (4) family history is consistent with the mode of inheritance of the disorder.

PP5 BP6 Reputable source

There are increasing examples where pathogenicity classifications from a reputable source (e.g. clinical laboratory with long-standing expertise in the disease area) have been shared into databases, yet the evidence that formed the basis for classification was not provided and may not be easily obtainable. In this case, the classification, if recently submitted, can be used as a single piece of supporting evidence. However, laboratories are encouraged to share the basis for classification as well as communicate with submitters to enable the underlying evidence to be evaluated and built upon. If the evidence is available, this criterion should not be used and instead, the criteria relevant to the evidence should be used.

BP5 Alternate locus observations

When a variant is observed in a case with a clear alternate genetic cause of disease, this is generally considered supporting evidence to classify the variant as benign. However, there are exceptions. An individual can be a carrier of an unrelated pathogenic variant for a recessive disorder; therefore, this evidence is much stronger for supporting a likely benign variant classification in a gene for a dominant disorder as compared to a gene for a recessive disorder. In addition, there are disorders where having multiple variants can contribute to more severe disease. For example, two variants, one pathogenic and one novel, are identified in a patient with a severe presentation of a dominant disease. A parent also has mild disease. In this case, one must consider the possibility that the novel variant could also be pathogenic and contributing to the increased severity of disease in the proband. In this clinical scenario, observing the novel variant as the second variant would not support a benign classification of the novel variant (though it is also not considered support for a pathogenic classification without further evidence). And finally, there are certain diseases where multigenic inheritance is known to occur, such as Bardet-Beidel syndrome, in which case the additional variant in the second locus may also be pathogenic but should be reported with caution.

BP7 Synonymous variants

There is increasing recognition that splicing defects, beyond disruption of the splice consensus sequence, can be an important mechanism of pathogenicity, particularly for genes in which loss-of-function is a common mechanism of disease. Therefore, one should be cautious in assuming that a synonymous nucleotide change will have no effect. However, if the nucleotide position is not conserved over evolution and splicing assessment algorithms neither predict an impact to a splice consensus sequence nor the creation of a new alternate splice consensus sequence, then a splicing impact is less likely. Therefore, if supported by computational evidence (BP4), one can classify novel synonymous variants as Likely Benign. However, if computational evidence suggests a possible impact to splicing, or there is raised suspicion for an impact (e.g. the variant occurs in trans with a known pathogenic variant in a gene for a recessive disorder) then the variant should be classified as uncertain significance until a functional evaluation can provide a more definitive assessment of impact or other evidence is provided to rule out a pathogenic role.

Results

The results section should list variants using HGVS nomenclature (see Nomenclature section). Given the increasing number of variants found in genetic tests, presenting the variants in tabular form with essential components may best convey the information. These components include: nomenclature at both the nucleotide (genomic and cDNA) and protein level, gene name, disease, inheritance, exon, zygosity and variant classification. Parental origin may also be included if known. In addition, if specific variants are analyzed in a genotyping test, the laboratory should specifically note the variants interrogated with their full description and historical nomenclature if it exists. Furthermore, when reporting results from exome or genome sequencing, or occasionally very large disease-targeted panels, grouping variants into categories such as “Variants in Disease Genes with an Established Association with the Reported Phenotype”, “Variants in Disease Genes with a Likely Association with the Reported Phenotype” and (where appropriate) “Incidental (Secondary) Findings” may be beneficial.

Interpretation

The interpretation should contain the evidence supporting the variant classification including its predicted effect on the resultant protein and whether any variants identified are likely to fully or partially explain the patient's indication for testing. The report should also include any recommendations to the clinician for supplemental clinical testing such as enzymatic/functional testing of the patient’s cells and variant testing of family members to further inform variant interpretation. The interpretation section should address all variants described in the results section but may contain additional information. It should be noted if the variant has been reported previously in the literature or in disease or control databases. The references, if any, that contributed to the classification should be cited where discussed and listed at the end of the report. The additional information described in the interpretation section may include a summarized conclusion of the results of in-silico analyses and evolutionary conservation analyses. However, individual computational predictions (e.g. scores or terms such as “damaging”, etc.) should be avoided given the high likelihood of misinterpretation by health care providers who may be unfamiliar with the limitations of predictive algorithms (see In Silico Predictive Programs section above). A discussion of decreased penetrance and variable expressivity of the disorder, if relevant, should be included in the final report.

Methodology

The methods and types of variants detected by the assay and those refractory to detection should be provided on the report. Limitations of the assay used to detect the variants should also be reported. Methods should include those used to obtain nucleic acids (e.g. PCR, capture, whole genome amplification) as well as those to analyze the nucleic acids (e.g. bi-directional Sanger sequencing, next generation sequencing, chromosomal microarray, genotyping technologies, etc.) as this may provide the healthcare provider with the necessary information to decide if additional testing is required to follow up on the results. The methodology section should also give the official gene names approved by the Human Genome Organization Gene Nomenclature Committee (HGNC), RefSeq accession numbers for transcripts and genome build including versions. For large panels, gene level information may be posted and referenced by URL. The laboratory may choose to add a disclaimer which addresses general pitfalls in laboratory testing such as sample quality and sample mix up.

Access to patient advocacy groups, clinical trials and research

Although specific clinical guidance for a patient is not recommended for laboratory reports, provision of general information for categories of results (e.g. all positives) is appropriate and helpful. A large number of patient advocacy groups and clinical trials are now available for support and treatment of many diseases. Laboratories may choose to add this information to the body of the report or attach the information so it is sent to the healthcare provider along with the report. Laboratories may make an effort to connect the healthcare provider to research groups working on specific diseases when a variant’s effect is classified as “uncertain” as long as HIPAA patient privacy requirements are followed.

Variant re-analysis

As evidence on variants evolves, previous classifications may later require modification. For example, the availability of large population variant frequency data has led many “uncertain significance” variants to be re-classified as benign, and testing additional family members may result in the re-classification of variants.

As the content of sequencing tests expands and the number of variants identified grows, expanding to thousands and millions of variants from exome and genome sequencing, the ability for laboratories to update reports as variant knowledge changes will be untenable without appropriate mechanisms and resources to sustain those updates. In order to set appropriate expectations with healthcare providers and patients, laboratories should provide clear policies on the reanalysis of data from genetic testing and whether additional charges may apply for reanalysis. Laboratories are encouraged to explore innovative approaches to give patients and providers more efficient access to updated information.^37,38

For reports containing VUSs in genes related to the primary indication and in the absence of updates that may be proactively provided by the laboratory, it is recommended that laboratories suggest periodic inquiry by healthcare providers to determine if knowledge has changed on any VUSs including variants reported as “Likely Pathogenic”. In contrast, laboratories are encouraged to consider proactive amendment of cases when a variant reported with a near definitive classification (Pathogenic or Benign) must be reclassified. Please see ACMG guidelines on duty to re-contact regarding physician responsibility.³⁹

Confirmation of findings

Recommendations for the confirmation of reported variants is addressed elsewhere.^35,36 Except as noted, confirmation studies using an orthogonal method are recommended for all sequence variants that are considered to be pathogenic or likely pathogenic for a Mendelian disorder. These methods may include, but are not limited to: re-extraction of the sample and testing, testing of parents, restriction enzyme digestion, sequencing the area of interest a second time or using an alternate genotyping technology.

Evaluating and Reporting Variants in Genes of Uncertain Significance (GUS) based on the Indication for Testing

Genome and exome sequencing are identifying new genotype-phenotype connections. When the laboratory finds a variant in a gene without a validated association to the patient's phenotype, this is a gene of uncertain significance (GUS). This can occur when a gene has never been associated with any patient phenotype or when the gene has been associated with a different phenotype from that under consideration. Special care must be taken when applying the recommended guidelines to a GUS. In such situations, utilizing variant classification rules developed for recognized genotype-phenotype associations is not appropriate. For example, when looking across the exome or genome, a de novo observation is no longer strong evidence for pathogenicity given that all individuals are expected to have approximately one de novo variant in their exome or 100 in their genome. Likewise, thousands of variants across a genome could segregate with a significant LOD score. Furthermore, many deleterious variants may be detected that are clearly disruptive to a gene or its resultant protein (nonsense, frameshift, canonical +/−1,2 splice site, exon-level deletion); however, this is insufficient evidence for a causative role in any given disease presentation.

Variants found in a GUS may be considered as candidates and reported as ‘Variants in a gene of uncertain significance’. These variants, if reported, should always be classified as “Uncertain Significance”. Additional evidence would be required to support the gene’s association to disease before any variant in the gene itself can be considered pathogenic for that disease.⁵ For example, additional cases with matching rare phenotypes and deleterious variants in the same gene would enable the individual variants to be classified according to the recommendations presented here.

Evaluating Variants in Healthy Individuals or as Incidental Findings

Caution must be exercised when using these guidelines to evaluate variants in healthy or asymptomatic individuals or to interpret incidental findings unrelated to the primary indication for testing. In these cases, the likelihood of any identified variant being pathogenic may be far less than when performing disease-targeted testing. As such, the required evidence to call a variant pathogenic should be higher and extra caution should be exercised. In addition, the predicted penetrance of pathogenic variants found in the absence of a phenotype or family history may be far less than predicted based upon historical data from patients ascertained with disease.

Mitochondrial variants

The interpretation of mitochondrial variants other than well-established pathogenic variants is complex and remains challenging and has several special considerations which are addressed here.

The nomenclature differs from standard nomenclature for nuclear genes, using gene name and m. numbering (e.g. m.8993T>C) and p. numbering, but not the standard c. numbering (see also nomenclature). The current accepted reference sequence is the Revised Cambridge Reference Sequence (rCRS) of the Human Mitochondrial DNA: GenBank sequence NC_012920 gi:251831106.^40,41

Heteroplasmy or homoplasmy should be reported along with an estimate of heteroplasmy of the variant if the test has been validated to determine heteroplasmy levels. Heteroplasmy percentages may vary in different tissue types from the sample tested; therefore, interpreting low heteroplasmic levels also must be done in the context of the tissue tested, and may be meaningful only in the affected tissue such as muscle. Over 275 mtDNA variants relating to disease have been recorded (http://mitomap.org/bin/view.pl/MITOMAP/WebHome).⁴² MitoMap is considered the main source of information related to mitochondrial variants as well as haplotypes. Other resources, such as frequency information (http://www.mtdb.igp.uu.se/),⁴³ secondary structures, sequences, and alignment of mitochondrial tRNAs (http://mamit-trna.u-strasbg.fr/),⁴⁴ mitochondrial haplogroups (http://www.phylotree.org/)⁴⁵and other information (http://www.mtdnacommunity.org/default.aspx),⁴⁶ may prove useful in interpreting mitochondrial variants.

Given the difficulty in assessing mitochondrial variants, a separate evidence checklist has not been included. However, any evidence needs to be applied with additional caution (for review, see Wong, 2007).⁴⁷ The genes in the mitochondrial genome encode for tRNA as well as for protein; therefore, evaluating amino acid changes is only relevant for genes encoding proteins. Similarly, since many mitochondrial variants are missense variants, evidence criteria for truncating variants will likely not be helpful. Because truncating variants do not fit the known variant spectrum in most mitochondrial genes, their significance may be uncertain. Although mitochondrial variants are typically maternally inherited, they can be sporadic, yet de novo variants are difficult to assess due to heteroplasmy that may be below an assay’s detection level, or different between tissues. The level of heteroplasmy may contribute to the variable expression and reduced penetrance seen within families. Nevertheless there still remains a lack of correlation between the percent heteroplasmy and disease severity.⁴⁷ Muscle, liver or urine may be additional specimen types useful for clinical evaluation. Undetected heteroplasmy may also affect outcomes of case, case/control, and familial concordance studies. In addition, functional studies are not readily available, although evaluating muscle morphology may be helpful (i.e. presence of ragged red fibers). Frequency data and published studies demonstrating causality may often be the only assessable criteria on the checklist. An additional tool for mitochondrial diseases may be haplogroup analysis, but may not represent a routine method that clinical laboratories have employed and the clinical correlation is not easy to interpret.

Consideration should be given to testing nuclear genes associated with mitochondrial disorders, as variants in nuclear genes could be causative of oxidative disorders or modulating the mitochondrial variants.

Pharmacogenomics

Establishing the effects of variants in genes involved with drug metabolism is challenging, in part because a phenotype is only apparent upon exposure to a drug. Still, variants in genes related to drug efficacy and risk for adverse events have been described and are increasingly used in clinical care. Gene summaries and clinically relevant variants can be found at the Pharmacogenomics Knowledge Base (PharmGKB; http://www.pharmgkb.org/).⁴⁸ Alleles and nomenclature for the cytochrome P450 gene family is available at http://www.cypalleles.ki.se/.⁴⁹ Although the interpretation of pharmacogenomic (PGx) variants is beyond the scope of this document, we include a discussion of the challenges and distinctions associated with the interpretation and reporting of PGx results.

The traditional nomenclature of PGx alleles uses star (*) alleles, which often represent haplotypes, or a combination of variants on the same allele. Traditional nucleotide numbering using outdated reference sequences is still in practice. Converting traditional nomenclature to standardized nomenclature using current reference sequences is an arduous task, but is necessary for informatics applications with next generation sequencing.

Many types of variants have been identified in PGx genes, such as truncating, missense, deletions, duplications (of functional as well as non-functional alleles) and gene conversions, resulting in functional, partially functional (decreased or reduced function) and non-functional (null) alleles. Interpreting sequence variants often requires haplotype determination from a combination of variants detected. Haplotypes are typically presumed based on population frequencies and known variant associations rather than testing directly for chromosomal phase (molecular haplotyping).

In addition, for many PGx genes (particularly variants in genes coding for enzymes) the overall phenotype is derived from a diplotype, which is the combination of variants or haplotypes on both alleles. Because PGx variants do not directly cause disease, it may be more appropriate to use terms related to metabolism (rapid, intermediate, poor), efficacy (resistant, responsive, sensitive) or “risk”, rather than pathogenic. Further nomenclature and interpretation guidelines are needed to establish consistency in this field.

Common Complex Disorders

Unlike Mendelian diseases, the identification of common, complex disease genes, such as those contributing to type 2 diabetes, coronary artery disease, and hypertension has largely relied on population-based approaches (e.g. genome wide association studies or GWAS) rather than family-based studies.^50,51 Currently, numerous GWAS reports have resulted in the cataloguing of over 1200 risk alleles for common, complex diseases and traits. However, most of these variants are in non-genic regions and additional studies will be required to determine if any of the variants are directly causal through effects on regulatory elements, for example, or are in linkage disequilibrium with causal variants.⁵²

Common, complex risk alleles typically confer low relative risk and are meager in their predictive power.⁵³ To date, the utility of common, complex risk allele testing for patient care⁵⁴ has been unclear and models to combine multiple markers into a cumulative risk score are often flawed and are usually no better than traditional risk factors such as family history, demographics and non-genetic clinical phenotypes.^55,56 Moreover, in almost all of the common diseases, the risk alleles can only explain up to 10% of the variance in the population even when the disease has high heritability. Given the complexity of issues, this recommendation does not address the interpretation and reporting of complex trait alleles. However, we recognize that some of these alleles are identified during the course of sequencing Mendelian genes and therefore guidance on how to report such alleles when found incidentally is needed. The terms “pathogenic”, and “likely pathogenic” are not appropriate in this context even when the association is statistically valid. An interim solution, until better guidance is developed, is to report these variants as “risk alleles” or under a separate "other reportable" category in the diagnostic report. The evidence for the risk, as identified in the case-control/GWAS studies can be expressed by modifying the terms such as “established risk allele”, “likely risk allele”, or “uncertain risk allele”, if desired.

Somatic Variants

The description of somatic variants, primarily those observed in cancer cells, includes complexities not encountered with constitutional variants, because the allele ratios are highly variable and tumor heterogeneity can cause sampling differences. Interpretation helps select therapy, predicts treatment response or prognosis of overall survival or tumor progression free survival, further complicating variant classification. For the interpretation of negative results, understanding the limit of detection of the sequencing assay (at what allele frequency can the variant be detected by the assay) is important, and requires specific knowledge of the tumor content of the sample. Variant classification categories are also different with somatic variants as compared to germline, with terms such as “responsive”, “resistant”, “driver”, and “passenger” often used. Confirmation of whether a variant is truly somatic is done by sequence analysis of the patient’s germline DNA. A different set of interpretation guidelines is needed for somatic variants, with tumor-specific databases used for reference in addition to databases used for constitutional findings. To address this, a workgroup has recently been formed by the Association for Molecular Pathology.