The Alzforum Mutations database aggregates information about genetic variants reported in the literature. This database focuses on a subset of genes implicated in Alzheimer’s disease. We started out by curating variants in genes associated with autosomal-dominant AD, and then broadened the scope to include variants in genes linked to AD in other ways, such as through association studies or pathology. We take a “gene-centric” approach, with the goal of cataloguing all reported variants identified in patients, families, or controls, regardless of whether the variant is thought to be pathogenic, not pathogenic, or somewhere in between, i.e., modifying a person’s risk of developing AD.

Inclusion Criteria

All variants are eligible for inclusion, whether they were reported in a peer-reviewed publication, a meeting abstract, a review, or by personal communication.

Data Sources

Information is curated from the published literature, primarily peer-reviewed scientific articles. Review articles and book chapters are also sources. When no peer-reviewed publication is available, we occasionally include findings reported in conference abstracts. In addition, authors provide information to Alzforum via personal communication. Within the database, some papers are identified as “primary papers”; these are generally the first paper(s) to identify a variant.

Curation Strategy

We monitor the literature weekly for new variants and new information about existing variants. While we aim to include all eligible information, we prioritize novel mutations over newly identified carriers of previously identified variants. We also prioritize variants identified as pathogenic, risk modifiers, or possible risk modifiers, over nonpathogenic or unclear variants. Our variant descriptions summarize the evidence for genetic association with disease, clinical findings, neuropathology/imaging, and reported biological effects. We extract this information from the primary papers, follow-up papers, and other sources, consulting with authors for clarification as needed. 

Pathogenicity Classification
Determining which genetic variants play a role in disease and which are benign is a big challenge in neurogenetics. There are efforts in the field to devise systematic approaches to assess the pathogenicity of AD-associated variants in known disease-causing genes, allowing for a more standardized, objective method of classifying variants based on the available evidence (e.g., Guerreiro et al., 2010 , Hsu et al., 2018; Koriath et al., 2018; Hsu et al., 2020 ). However, to date, there are no universally agreed-upon criteria for classifying pathogenicity of AD associated variants. Moreover, relevant data are not always publicly available, resulting in classification differences between groups and confusion in the field.

Several factors complicate classification. Some authors have erred on the side of reporting novel variants as pathogenic, and a critical classification scheme might cast doubt on these initial classifications. Also, as additional information becomes available, classifications require updating. Moreover, the prevalence of variants differs between populations, and genetic backgrounds can influence pathogenicity (Dehghani et al., 2021). We include information on the country of origin and/or the ancestry of the individual or cohort in the database when available.

Variants in the database are classified according to the curator’s synthesis of the available evidence, as described below. As new information emerges, variants may be reclassified. Alzforum welcomes submissions of additional information to improve classifications.

1. APP, PSEN1, and PSEN2 variants
To classify the pathogenicity of variants in APP, PSEN1, and PSEN2, Alzforum applies guidelines developed by the American College of Medical Genetics and Genomics and Association of Molecular Pathology (ACMG-AMP) for the classification of genes that cause Mendelian disorders (Richards et al., 2015). In most cases, familial Alzheimer’s disease (AD) and a few related cerebrovascular diseases caused by mutations in APP, PSEN1, or PSEN2 behave in a Mendelian manner, following an autosomal dominant pattern of inheritance. For cases in which a deviation from this Mendelian pattern is suspected (for example, when a variant appears to act as a risk factor), a note is added below the variant’s classification. Pathogenicity related to other, non-AD-related diseases is not assessed within the current scope of this database.

The ACMG-AMP guidelines classify variants as Pathogenic, Likely Pathogenic, Benign, Likely Benign, or Uncertain Significance. Two sets of criteria are used to make these assignments: one that assesses pathogenic features and another that assesses benign features (see table below). Each pathogenic criterion is weighted as very strong (PVS), strong (PS); moderate (PM), or supporting (PP), and each benign criterion is weighted as stand-alone (BA), strong (BS) or supporting (BP). Each criterion also has an identification number whose sole purpose is to distinguish it from other criteria; the identification number does not convey any information about its weighting.

To classify a given variant in these three genes, the Alzforum curator selects the criteria that apply based on a critical review of the available data, and then combines their corresponding weights following the rules described in (Richards et al., 2015). The software tool InterVar is used to automate this process. (One exception to this classification scheme was made with the APP A673T (Icelandic). It was classified as Protective, a term outside the scope of the 2015 guidelines (Richards et al., 2015).

ACMG-AMP 2015 Classification Criteria

Pathogenic
PVS1 Null variant (nonsense, frameshift, canonical +/−1 or 2 splice sites, initiation codon, single or multi-exon deletion) in a gene where loss of function (LOF) is a known mechanism of disease.

PS1 Same amino acid change as a previously established pathogenic variant regardless of nucleotide change.

PS2 De novo (both maternity and paternity confirmed) in a patient with the disease and no family history.

PS3 Well-established in vitro or in vivo functional studies supportive of a damaging effect on the gene or gene product.

PS4 The prevalence of the variant in affected individuals is significantly increased compared to the prevalence in controls.

PM1 Located in a mutational hot spot and/or critical and well-established functional domain (e.g. active site of an enzyme) without benign variation.

PM2 Absent from controls (or at extremely low frequency if recessive) in Exome Sequencing Project, 1000 Genomes Project, or Exome Aggregation Consortium. *Alzforum uses the gnomAD variant database.

PM3 For recessive disorders, detected in trans with a pathogenic variant PM4 Protein length changes due to in-frame deletions/insertions in a non-repeat region or stop-loss variants.

PM5 Novel missense change at an amino acid residue where a different missense change determined to be pathogenic has been seen before.

PM6 Assumed de novo, but without confirmation of paternity and maternity.

PP1 Co-segregation with disease in multiple affected family members in a gene definitively known to cause the disease: *Alzforum requires at least one affected carrier and one unaffected non-carrier from the same family to fulfill this criterion.

PP2 Missense variant in a gene that has a low rate of benign missense variation and where missense variants are a common mechanism of disease.

PP3 Multiple lines of computational evidence support a deleterious effect on the gene or gene product (conservation, evolutionary, splicing impact, etc.). *In most cases, Alzforum applies this criterion when the variant’s PHRED-scaled CADD score is greater than or equal to 20.

PP4 Patient’s phenotype or family history is highly specific for a disease with a single genetic etiology.

PP5 Reputable source recently reports variant as pathogenic but the evidence is not available to the laboratory to perform an independent evaluation.

Benign
BA1 Allele frequency is >5% in Exome Sequencing Project, 1000 Genomes Project, or Exome Aggregation Consortium. *Alzforum uses the gnomAD variant database.

BS1 Allele frequency is greater than expected for disorder. *Alzforum uses the gnomAD variant database. BS2 Observed in a healthy adult individual for a recessive (homozygous), dominant (heterozygous), or X-linked (hemizygous) disorder with full penetrance expected at an early age.

BS3 Well-established in vitro or in vivo functional studies shows no damaging effect on protein function or splicing.

BS4 Lack of segregation in affected members of a family.

BP1 Missense variant in a gene for which primarily truncating variants are known to cause disease.

BP2 Observed in trans with a pathogenic variant for a fully penetrant dominant gene/disorder; or observed in cis with a pathogenic variant in any inheritance pattern.

BP3 In-frame deletions/insertions in a repetitive region without a known function.

BP4 Multiple lines of computational evidence suggest no impact on gene or gene product (conservation, evolutionary, splicing impact, etc). *In most cases, Alzforum applies this criterion when the variant’s PHRED-scaled CADD score is less than 20.

BP5 Variant found in a case with an alternate molecular basis for disease.

BP6 Reputable source recently reports variant as benign but the evidence is not available to the laboratory to perform an independent evaluation.

BP7 A synonymous (silent) variant for which splicing prediction algorithms predict no impact to the splice consensus sequence nor the creation of a new splice site AND the nucleotide is not highly conserved.

From Richards et al., 2015 . *Notes on Alzforum’s application of criteria. For further explanation see below.

Application and tuning of ACMG-AMP guidelines
In order to put a variant through this classification scheme, Alzforum requires that a minimum amount of information be available. We do not classify variants that are very rare or absent from the gnomAD database, and for which a single affected carrier has been reported without evidence of cosegregation (see below Segregation of variant with disease). These variants are marked as “Not Classified”. Moreover, variants that are absent, or very rare, in the gnomAD database and found in multiple members of a single affected family, without evidence of cosegregation, are not classified, unless there is evidence of a functional effect (Nykamp et al., 2018).

The ACMG-AMP guidelines allow for tuning of the “weights” of different criteria to accommodate variations in the types of data available. Increasing the weight of a pathogenic criterion, for example, adds to a variant’s overall pathogenic score, making it more likely to qualify for a Pathogenic or Likely Pathogenic classification. Conversely, reducing the weight of a pathogenic criterion shifts the score towards Benign classification. Alzforum represents this tuning with a dash followed by the letter S (strong), M (moderate), or P (supporting) after the name of each criterion. For example, the PM1 criterion, which considers whether a variant is located in a mutational hotspot or a functional region, is denoted PM1-M when applied at its default, moderate setting. When a variant resides in a region that is both a mutational hot spot and of likely functional importance, however, PM1 is upgraded to strong (PM1-S), increasing the likelihood of a pathogenic classification. Conversely, when a variant is at the edge of a hot spot, without clear evidence of functional importance, PM1 is downgraded to supporting (PM1-P) (see below Location of variant (PM1 criterion)).

Functional evidence (PS3 and BS3 criteria): Most functional evidence reported for variants in APP, PSEN1 and PSEN2 refers to alterations in the production or secretion of Aβ peptides, with most assays focusing on Aβ40 and Aβ42. However, other APP peptides (e.g., Aβ43, Aβ38, Aβ37, β-CTFs) may also contribute to pathogenicity, and other functions beyond APP processing may affect disease. Given this, Alzforum endeavors to assess all available functional data for evaluating the functional effects of a given variant. If different groups report conflicting observations—for example, if one group reports an increase in the Aβ42/40 ratio in cells expressing a putative pathogenic variant in APP and another does not—neither PS3 (evidence of a damaging effect) nor BS3 (evidence of no damaging effect) are applied. If one set of results is judged to be more robust than another, the relevant criterion is applied but downgraded to either moderate or supporting and a note is addedto the relevant criterion in the pathogenicity section of the variant’s webpage.

To qualify as functional evidence, observations must derive from experimental assays in which variants are tested directly, isolated from the whole-body disease process, usually using cultured cells or isolated proteins. Measurements of Aβ peptides in the fluids or tissues of patients alone are not used for pathogenicity classification.

Prevalence of variant in affected individuals (PS4 criterion): This criterion is applied in downgraded form (PS4-M), to rare variants for which there are no case-control studies, but three or more unrelated, affected carriers have been reported, and the variant is absent from controls.

Location of variant (PM1 criterion): This criterion is upgraded to strong (PM1-S) if the variant is located both at a mutational hot spot and at a position predicted to be critical for activity relevant to AD (e.g., γ-secretase cleavage of APP). The criterion is downgraded to supporting (PM1-P) if the variant is at the edge of a mutational cluster.

Allele counts in variant databases (PM2 and BS1 criteria): In general, PM2 is applied if the allele count in the gnomAD variant database (v2.1.1) is two or less, and BS1 is applied if it is greater than four. These thresholds are based on calculating the maximum tolerated reference allele count using the CardioDB web tool. Inputs are as follows: prevalence of early onset AD = 30 per 100,000 (Koriath et al., 2018), genetic heterogeneity = 1, allelic heterogeneity = 0.01 (conservative estimate based on Alzforum’s compiled number of variants with strong evidence of pathogenicity across all three PSEN1, PSEN2, and APP genes, assuming equal prevalence of each variant), penetrance = 0.9 (since most variants are inherited in an autosomal dominant manner), confidence = 0.95 (default), reference population size = 250,000 (common allele number for variants in these genes in gnomAD v2.1.1). When reference allele numbers differ by more than a factor of two (e.g., when using other versions of gnomAD), the maximum tolerated reference allele count is re-calculated. Of note, this approach yields thresholds similar to those used by Celeste Karch and colleagues to classify APP, PSEN1, and PSEN2 mutations (Hsu et al., 2020). 

The prevalence of variants often varies between populations of different ancestries. This is noted for variants that fulfill the BS1 criterion.

Variant affects same amino acid as a known pathogenic mutation (PM5 criterion): To avoid circular reasoning, the known pathogenic mutation must fulfill the requirements for Pathogenic classification without including PM5 in its own classification.

Segregation of variant with disease (PP1 criterion): At least one affected carrier and one unaffected non-carrier from the same family are required to fulfill this criterion. The healthy non-carriers’ age(s) must be at least two standard deviations greater than the family’s mean age at onset. The criterion is upgraded from supporting (PP1-P) to moderate (PP1-M) if two affected carriers are reported, and to strong (PP1-S) if three or more affected carriers are reported. Also, if segregation is seen in more than one family, regardless of carrier numbers, PP1 is upgraded to strong (PP1-S).

Missense variants as a common mechanism of disease (PP2 criterion): This criterion is applied to all missense variants in PSEN1 and to those in APP  located in exons 16 or 17.

Computational evidence (PP3 and BP4 criteria): For all single nucleotide variants (SNVs), Alzforum uses the Combined Annotation Dependent Depletion (CADD) tool to predict deleteriousness. CADD integrates variant information including sequence context, gene model annotations, evolutionary constraints, epigenetic measurements, and functional predictions into a single score using a machine learning model. To put this score in context, it is often transformed into a log-based score that ranks deleteriousness based on the genome-wide distribution of scores for all potential SNVs (PHRED-scaled). For example, a scaled score of 10 or greater indicates the variant has a predicted deleteriousness in the top 10 percent of all SNVs, and a score of 20 or greater indicates it is in the top 1 percent. Alzforum applies PP3 if a SNV’s PHRED-scaled CADD score is greater than or equal to 20, and BP4 if it is less than 20. The same guidelines are followed for insertion/deletion variants if the PHRED-scaled CADD scores are available. If not, the combined results of other predictive algorithms (e.g., SIFT, PolyPhen) are used for classification: if multiple algorithms consistently predict a damaging effect, then PP3 is applied. Conversely, if multiple algorithms consistently predict the variant is not damaging, then BP4 is applied.

Unavailable evidence (PP5 and BP6 criteria): These criteria are not applied to Alzforum’s classifications. Observations used for classification derive mostly from peer-reviewed publications, and occasionally from meeting abstracts, reviews, or personal communications. Sources are cited in the detail page text.

2. APOE
Variants in APOE are classified as studied in a neurological context, studied in a non-neurological context, or studied in both neurological and non-neurological contexts. Highlighted variants (in dark blue) have been examined in three or more publications and/or were deemed to be of particular interest by Alzforum curators.

ApoE domains and regions depicted in the diagram were selected based on the following publications:

N-terminal domain (1-185): Chen et al., 2011, Chen et al., 2020, Martens et al., 2022. Note different boundaries in Wilson et al., 1991, Frieden et al., 2017, Henry et al., 2018.

C-terminal domain (225-317): The boundary between this domain and the hinge region varies between publications: Chen et al., 2011, Frieden and Garai, 2012, Frieden et al., 2017, Henry et al., 2018, Chen et al., 2020, Martens et al., 2022.

Hinge domain (186-224): Boundaries vary between publications: Chen et al., 2011, Frieden et al., 2017, Chetty et al., 2017, Henry et al., 2018, Chen et al., 2020, Martens et al., 2022.

Receptor-binding region (154-168): Chen et al., 2020, Martens et al., 2022.  Note different amino acid boundaries in Frieden et al., 2017, Peters-Libeu et al., 2006, Mahley et al., 2008.

Lipid-binding region (262-290): Yamazaki et al., 2019. Note different amino acid boundaries in Dong et al., 1994, Frieden and Garai, 2012, Frieden et al., 2017, Martens et al., 2022.

Major HSPG binding region (162-165): Cardin et al., 1986, Libeu et al., 2001. Note that the labeled amino acids are reported as crucial, but additional amino acids have been implicated in HSPG binding Weisgraber et al., 1986, Libeu et al., 2001, Martens et al., 2022.

Homo-oligomerization region (248-261 and 280-288): Huang et al., 2011. Note that the region in between the two sites may also be involved in oligomerization (Frieden and Garai 2012).

N-terminal α-helices: Although not labeled, the four parallel segments in the N-terminus of the diagram, after the peptide sequence, roughly match the helix bundle described in structural studies (Wilson et al., 1991 and Chen et al., 2011).  Alpha-helices are also likely to exist in the C-terminus.

Also of note, long-range interactions are predicted to occur between ApoE domains that differ between the three common ApoE isoforms (ApoE2, ApoE3, and ApoE4). Moreover, ApoE is a dynamic protein adopting different conformations depending on its lipid-binding status, homo-oligomerization, interactions with other proteins, micro-environment, and source of production (e.g., Chen et al., 2020, Stuchell-Brereton et al., 2023, Nemergut et al., 2023).

3. MAPT and TREM2
Variants in MAPT and TREM2 are classified as Pathogenic, Unclear, or Benign. Some variants, primarily those within TREM2, are classified as Risk Modifiers or Possible Risk Modifiers. In addition, a given variant may be classified as a risk modifier for one disease and a possible risk modifier for other diseases. For example, TREM2 R47H is classified as a risk modifier for AD and a possible risk modifier for PD, FTD, and ALS.

Pathogenicity Terms

Pathogenic
A variant is classified as pathogenic for a disease when it segregates with the disease phenotype within a family. However, evidence of segregation is not always available, for example when the genotype of unaffected family members is unknown. In this case, the curator evaluates the supporting evidence for pathogenicity. Such evidence may include: multiple affected variant carriers, a clear family history of disease in an affected carrier, absence of the variant in healthy controls (in diseases exhibiting a recessive pattern of inheritance, controls may be heterozygous), in silico predictions of protein function, and experimental evidence of a pathobiological effect. In some cases, variants may be classified as pathogenic, even if one or more carriers are healthy. If the preponderance of the evidence suggests a variant is pathogenic, an unaffected carrier may be an example of incomplete penetrance or be considered an "escapee," i.e., a variant carrier who remained healthy beyond the family’s average age of onset.

Risk Modifier
A variant is classified as a risk modifier when a preponderance of evidence supports an association to the disease. For example, the R62H variant in TREM2 is classified as a risk modifier based largely upon the strength of a single study involving a large number of individuals, which reported a genome-wide significant association. In cases where a large, definitive study is unavailable, a variant may be classified as a risk modifier when the number of studies reporting an association exceeds the number that do not.

Possible Risk Modifier
A variant is classified as a possible risk modifier when one or more studies support an association, but at least an equal number do not. The “possible” designation does not imply a specific degree of confidence, and could range from unlikely to probable.

Benign
A variant is classified as benign if it does not appear to cause disease; in other words, it does not segregate with disease or meet other criteria for a pathogenic, risk modifier, or possible risk-modifier classification (see above).

Unclear
A variant is classified as unclear when, in the judgment of the curator, conflicting evidence exists regarding the role of the variant in disease, preventing classification into one of the above categories. A variant may also be classified as unclear in cases in which there is insufficient data to make a classification.

Reference Isoforms

By convention, variants are named according to their amino acid position in a reference isoform. The reference isoform is not necessarily the canonical sequence. In cases where the reference isoform does not include the variant, such as variants that are in exons excluded from the reference isoform by alternative splicing, the variant is named according to its position in an isoform in which it is found. For example, TREM2 W191X is found only in the shortest TREM2 transcript; the “191” in its name refers to amino acid 191 encoded by that transcript, not the 191st amino acid in the reference isoform. When a variant is reported in the literature by its position in a non-reference isoform, but is present in the reference isoform, Alzforum standardizes its name to the reference isoform.

Acknowledgement
Alzforum is indebted to Jose Brás and Rita Guerreiro for their expert advice.

NEXT: Glossary