Genetic Variants: SNPs, Indels, CNVs & Structural Variation

A genetic variant is any position at which one person's DNA sequence differs from a reference or from another person's, and the study of variation is the foundation of nearly everything else in genetics. Variants are the raw material on which inheritance, evolution, and disease all operate, because without differences in sequence there would be nothing to inherit, to select, or to diagnose. They sit at the base of the hierarchy this site explores, beneath genes, pathways, and physiology, because a variant is the smallest unit of difference from which all genetic effects ultimately flow. The scale of variation is large but finite: a typical human genome differs from the reference at roughly four to five million sites, and deep population surveys have together catalogued well over a hundred million variant sites across humanity. The overwhelming majority of these differences are common, ancient, and harmless, while a small minority are rare and powerful enough to cause disease on their own. The central intellectual task of modern genetics is to distinguish those two groups, separating the benign background from the rare signal that matters for health. This matters for medicine because every genetic diagnosis, carrier test, and risk score rests on correctly reading what a variant is and what it does. It matters for longevity because the same variation that causes overt disease also tunes the slower processes of aging, and because a handful of rare protective variants reveal how the body can be unusually shielded from age-related disease.

Variants are organized first by their size and structure. The smallest is the single-nucleotide variant, a change of one DNA letter, which when common is also called a single-nucleotide polymorphism and which makes up the great bulk of human variation. Next come insertions and deletions, or indels, the gain or loss of a small number of bases, which can shift a gene's reading frame if the number involved is not a multiple of three. Larger still are copy-number variants and the broader category of structural variants, defined as changes of fifty bases or more, which include deletions, duplications, inversions that flip a segment end to end, and translocations that relocate one. A separate class is the short tandem repeat, in which a small motif is repeated a variable number of times and can expand pathologically. A second axis sorts variants by how often they occur, from common variants carried by more than one person in twenty, through low-frequency variants, to rare variants seen in fewer than one in two hundred. A third axis is functional consequence, which depends entirely on where a variant lands: a change in a protein-coding exon may be silent, may swap one amino acid for another, or may truncate the protein, while a change in a splice site, promoter, or enhancer can disrupt how or how much a gene is used. These three axes together, size, frequency, and consequence, form the coordinate system within which any variant is interpreted.

The modern picture of human variation rests on two great cataloguing efforts. The first, the 1000 Genomes Project, sequenced 2,504 people from 26 populations and reported in 2015 a reference set of more than 88 million variants, establishing that a typical genome carries between 4.1 and 5.0 million differences from the reference and that most common variation is shared across all human populations. It showed that each person also carries thousands of structural variants which, though far fewer than single-base changes, rearrange a larger total amount of DNA. The second effort, the Genome Aggregation Database, pushed the scale far higher: its 2020 release aggregated 141,456 individuals, and its 2024 successor reached 807,162, turning population allele frequency into a precise filter for variant interpretation. By measuring where expected variation is missing, gnomAD ranked every gene by how strongly it resists damaging mutation, distinguishing the constrained genes that cause severe dominant disease from the majority that tolerate some loss. Together these resources changed the practice of genetics from arguing about individual variants in isolation to checking each one against a population baseline of millions of people. The simple question of how often a variant appears in a large, diverse reference set is now the single most informative piece of evidence about whether it could cause a severe rare disease. This shift, from expert opinion to population data, is the quiet revolution behind modern variant interpretation.

Translating a catalogue of variants into a clinical decision requires a shared language for how confident anyone can be that a given change causes disease. The ACMG and AMP framework, published in 2015, supplies that language, sorting each variant into pathogenic, likely pathogenic, uncertain, likely benign, or benign by weighing population frequency, computational prediction, functional studies, and inheritance within families. Laboratories deposit their classifications into shared databases such as ClinVar, which lets the field detect and resolve the disagreements that were once common when each laboratory worked alone. The framework's hardest problem is the variant of uncertain significance, a verdict of insufficient evidence that accounts for a large share of findings and that is more common in people of non-European ancestry because reference databases under-represent them. This ancestry imbalance is one of the most consequential equity problems in genomics, since a variant that would be confidently classified in a European-ancestry patient may remain uncertain in a patient from an under-studied population. For longevity, the same framework that flags disease-causing variants also identifies the rare protective ones, such as the loss-of-function changes in PCSK9 that confer lifelong low cholesterol, which serve as natural models for prevention. The most common failure in translation is to treat a variant as a verdict rather than as evidence, when in practice a variant is a probability statement whose meaning depends on the gene, the person, and the strength of the supporting data.

Core Concepts

Single Nucleotide Variants (SNVs and SNPs)

The simplest and most common kind of genetic variant is a change at a single position in the DNA, where one base is replaced by another. This is called a single-nucleotide variant, and when the same change is common enough in a population to be a stable feature of it, the older term single-nucleotide polymorphism, or SNP, is often used. These two terms describe the same physical change and differ only in connotation, with polymorphism implying commonness. The 1000 Genomes Project catalogued roughly 84.7 million such variants, making them by far the largest class of human variation. Substitutions come in two flavors, transitions between similar bases and transversions between dissimilar ones, and transitions are about twice as common across the genome. The great majority of single-base variants fall in the vast non-coding portion of the genome and have no detectable effect on health. When one does land in a protein-coding region, its consequence is read from the genetic code: a synonymous change leaves the encoded amino acid unaltered, a missense change swaps one amino acid for another, and a nonsense change converts an amino-acid codon into a stop signal that truncates the protein. The classic example is the sickle cell variant in HBB, a single base change that replaces glutamate with valine and was the first disease ever traced to a defined sequence difference, by Vernon Ingram in 1957. Other single-base variants, such as the HFE C282Y change behind hereditary hemochromatosis and the LRRK2 G2019S change in Parkinson disease, show that the effect of a one-letter change ranges from negligible to severe depending entirely on where it falls and what it does.

Insertions and Deletions (Indels)

The next class of variant by size is the insertion or deletion of a small number of bases, collectively shortened to indel and conventionally defined as a change of fewer than fifty bases. Indels are the second most common type of variant after single-base changes, with the 1000 Genomes catalogue listing several million. Their consequence depends critically on whether they occur in a coding region and, if so, on how many bases are involved. Because the genetic code is read in three-base groups, an insertion or deletion whose length is not a multiple of three shifts the reading frame downstream, scrambling every subsequent codon and usually introducing a premature stop. Such frameshift variants are among the most damaging, because they typically abolish the protein. An indel whose length is a multiple of three, by contrast, adds or removes whole amino acids while leaving the rest of the reading frame intact, which is generally less catastrophic but can still be severe. The textbook example is the F508del variant in CFTR, an in-frame deletion of three bases that removes a single phenylalanine and causes the protein to misfold and be destroyed, accounting for the majority of cystic fibrosis alleles. Indels in non-coding regions, like non-coding single-base changes, are usually harmless. The practical difficulty with indels is that they are harder to detect and align accurately than single-base changes, especially within repetitive sequence, so they are a more error-prone class to call.

Copy Number Variants (CNVs)

When the gain or loss of DNA grows beyond the scale of an indel, into thousands or millions of bases, the variant is called a copy-number variant. These are deletions that remove a stretch of the genome or duplications that add extra copies, and they can encompass parts of a gene, whole genes, or many genes at once. Conrad and colleagues mapped these variants genome-wide in 2010 and showed that they are a common and functionally important form of human variation, not a rare aberration. Many recurrent copy-number variants arise through a specific mechanism: blocks of near-identical sequence scattered through the genome occasionally misalign during the formation of eggs and sperm, and recombination between the misaligned copies deletes or duplicates the DNA in between. Because a copy-number variant changes the dosage of the genes it spans, its effect depends on how sensitive those genes are to dose. For some genes the loss of a single copy is enough to cause disease, a state called haploinsufficiency, while others are unaffected until both copies are lost. Duchenne muscular dystrophy is the classic copy-number disorder, with about two-thirds of cases caused by deletion of one or more exons of the DMD gene, and spinal muscular atrophy results from loss of the SMN1 gene with the severity tuned by the copy number of its backup, SMN2. Copy-number variants are detected by chromosomal microarray and by sequencing read-depth analysis, and they are now a first-line consideration in unexplained developmental disorders. They occupy the middle ground between single-base changes and the whole-chromosome abnormalities covered on the chromosomes page.

Structural Variation (Inversions, Translocations, Mobile Elements)

Copy-number variants are part of a broader category, structural variation, conventionally defined as any change of fifty bases or more. Alongside the deletions and duplications that change copy number, structural variation includes events that rearrange DNA without necessarily changing its total amount. An inversion flips a segment of a chromosome end to end, leaving the same genes present but in reversed order. A translocation moves a segment from one location to another, sometimes exchanging material between two chromosomes. Insertions of mobile genetic elements, sequences that copy themselves and paste the copy elsewhere in the genome, are another recurrent source of structural change and occasionally land inside a gene and disrupt it. Sudmant and colleagues mapped structural variation across the 1000 Genomes cohort in 2015, and Collins and colleagues built a reference of 433,371 such variants from nearly 15,000 genomes for gnomAD in 2020, showing that a typical genome carries thousands of them. A crucial lesson from this work is that even a balanced rearrangement, which deletes no DNA, can cause disease, either by breaking a gene at the point where the rearrangement joins two sequences or by separating a gene from the regulatory elements it needs. Structural variants are the hardest class to detect with the short DNA reads that dominate sequencing, because a rearrangement is easiest to see when a single read spans the junction, so they remained the most underascertained form of human variation until long-read sequencing matured. They are also the variant class that connects most directly to the chromosome rearrangements and enhancer-hijacking phenomena described on the genome-organization page.

Short Tandem Repeats and Repeat Expansions

A distinctive class of variation arises not from changing the bases themselves but from changing how many times a short motif is repeated. The genome is dotted with short tandem repeats, stretches in which a unit of one to a few bases is repeated many times in a row, and the number of repeats varies naturally between people. Most of this variation is harmless, and short tandem repeats are useful precisely because their high variability makes them excellent genetic fingerprints. In a subset of locations, however, the repeat can expand beyond a threshold and cause disease, producing the repeat-expansion disorders, of which more than fifty are now known. The archetype is Huntington disease, in which a CAG triplet in the HTT gene that is normally repeated fewer than 27 times expands beyond 36, and the longer the expansion the earlier the disease begins. Fragile X syndrome, the most common inherited cause of intellectual disability, arises when a CGG repeat in the FMR1 gene expands so far that it silences the gene. A characteristic feature of these disorders is anticipation, in which the repeat tends to grow longer as it passes from one generation to the next, so the disease appears earlier and more severely in successive generations. Because an expanded repeat is a change in length rather than in sequence, the standard pipelines that call single-base variants do not detect it, and dedicated tests are required. This blind spot is a recognized cause of diagnostic delay for these conditions.

Allele Frequency and the Variant Spectrum

The size of a variant is only half of what defines it; the other half is how often it occurs. Geneticists describe this with the minor allele frequency, the proportion of chromosomes in a population that carry the less common version of a variant. By convention, a variant carried by more than five percent of chromosomes is called common, one between roughly half a percent and five percent is low-frequency, and one below half a percent is rare. This frequency spectrum is shaped by the history of human populations and by natural selection. Most common variants are ancient, arose long ago in human history, and are shared across all populations, whereas most rare variants are recent and tend to be more specific to particular populations. Selection sculpts the spectrum by removing strongly harmful variants before they can become common, which is why severe disease-causing variants are almost always rare. The reverse logic is the single most useful filter in variant interpretation: if a variant is common in any well-sampled population, it is very unlikely to cause a severe, rare, dominant disease, because selection would not have let it become common. A small number of variants defy the simple rule through balancing selection, in which an allele harmful in two copies is maintained because it is beneficial in one, the sickle cell variant in HBB being the textbook case because a single copy protects against malaria. Understanding where a variant sits on the frequency spectrum is therefore inseparable from understanding what it might do.

How a Variant Is Interpreted

From Reference Genome to Variant Call

A variant only has meaning relative to a reference, and the modern reference is the human genome sequence first drafted in 2001 and completed without gaps in 2022. To find a person’s variants, their DNA is sequenced and the resulting reads are aligned to this reference, and every position where the person’s sequence differs is recorded as a variant call. This process is straightforward for single-base changes in unique sequence but becomes progressively harder for indels, for variants in repetitive regions, and for the large structural variants whose detection depends on reads that span a rearrangement junction. It is important to recognize that the reference genome is not a standard of health but an arbitrary composite assembled from a few individuals, so at many positions the so-called reference allele is itself the minor or even the higher-risk version. Calling variants accurately also depends on sequencing enough copies of each region, because a true variant must be distinguished from the errors that any sequencing technology produces. The output of this stage is a list of differences from the reference, often numbering in the millions for a whole genome, which then must be filtered and interpreted. Everything that follows in variant interpretation is an effort to reduce that long list to the few changes that actually matter.

Functional Annotation and Consequence Prediction

The first step in making sense of a list of variants is annotation, the process of asking where each variant falls and what it is predicted to do. Tools such as the Ensembl Variant Effect Predictor, described by McLaren and colleagues in 2016, map each variant onto the genes and transcripts it overlaps and label its likely consequence, distinguishing a synonymous change from a missense change, a frameshift, a splice-site disruption, or a purely non-coding location. For variants that change an amino acid, a second layer of computational predictors estimates how damaging the change is likely to be, including CADD, introduced by Kircher and colleagues in 2014, and the ensemble predictor REVEL from Ioannidis and colleagues in 2016. Splice-altering variants, including some far from any coding sequence, can be flagged by the deep-learning tool SpliceAI from Jaganathan and colleagues in 2019, and the more recent AlphaMissense has extended such predictions across tens of millions of possible missense changes. These tools are genuinely useful for triage, ranking thousands of variants so that attention falls on the most promising candidates. They are explicitly classified as supporting evidence only, however, never sufficient on their own to declare a variant pathogenic, because they are imperfect, frequently disagree with one another, and perform worst in the non-coding genome where the rules linking sequence to function are least understood. Annotation narrows the field, but it does not deliver a verdict.

The ACMG and AMP Classification Framework

Turning annotation into a clinical conclusion requires a shared standard for weighing evidence, which the ACMG and AMP framework provided in 2015. The framework, set out by Richards and colleagues, sorts every variant into one of five categories: pathogenic, likely pathogenic, uncertain significance, likely benign, or benign. It reaches that conclusion by combining distinct lines of evidence, each with a defined weight, including how often the variant appears in the population, what computational tools predict, whether functional experiments show an effect, and how the variant tracks with disease through a family. A variant absent from large reference databases and predicted damaging and shown to disrupt the protein and inherited with disease across a family accumulates strong evidence for pathogenicity, whereas a variant common in the population accumulates evidence for benignity. Before this framework, laboratories often reached different conclusions about the same variant, and its adoption substantially improved consistency. The category that remains most troublesome is the variant of uncertain significance, which is not a mild form of pathogenic but an explicit statement that the evidence cannot yet decide. Because this verdict is so common and so easily misread, understanding that uncertain means unresolved, not dangerous, is one of the most important pieces of variant literacy.

Population Databases as a Filter

The single most powerful tool in modern variant interpretation is the large population reference database, of which gnomAD is the leading example. By aggregating the sequences of hundreds of thousands of people, gnomAD reports how often any given variant appears across diverse populations, and that frequency is decisive evidence. A variant seen at appreciable frequency in a reference set of healthy adults cannot, by the logic of selection, be a common cause of severe early-onset dominant disease. The same databases do more than report frequency: by measuring where expected variation is conspicuously missing, gnomAD ranks each gene by how strongly it resists damaging mutation, captured in the LOEUF metric introduced by Karczewski and colleagues in 2020. A loss-of-function variant in a gene that normally never loses function is far more suspicious than the same kind of variant in a gene that tolerates loss in healthy people. The 2024 expansion of gnomAD to 807,162 individuals deepened this filter further, particularly for rarer variants that smaller databases could not resolve. The central limitation is that these databases over-represent people of European ancestry, which can make a variant appear rarer, and therefore more suspicious, in an under-sampled population than it truly is. Reading a variant against a population baseline has nonetheless become the foundation on which the rest of interpretation is built.

Clinical & Longevity Relevance

Mendelian Disease and Diagnostic Sequencing

The most established clinical use of variant interpretation is the diagnosis of Mendelian disorders, the conditions caused by one or a few high-effect variants in a single gene. When a child presents with an unexplained severe condition, sequencing the protein-coding genes or the whole genome, often together with both parents, can identify the responsible variant and end a long diagnostic odyssey. The diagnostic yield of such sequencing in undiagnosed disorders is substantial, frequently identifying a causal variant in a large minority of cases, and it rises as more variant classes are interrogated, since adding detection of copy-number variants and repeat expansions catches cases that single-base analysis alone would miss. A confirmed diagnosis can change management, guide reproductive decisions, and connect families to others with the same condition. The interpretation rests on the same logic throughout: a candidate variant must be rare in the population, fall in a gene consistent with the phenotype, and carry sufficient evidence of a damaging effect. The growth of shared databases such as ClinVar has accelerated this process, because a variant already classified by other laboratories need not be evaluated from scratch. Diagnostic sequencing is where the abstract framework of variant classification meets an individual patient most directly.

Carrier Status and Reproductive Decisions

A second major clinical use is carrier screening, which identifies healthy people who carry a single copy of a recessive disease variant. Because recessive conditions appear only when a child inherits a damaging variant from both parents, two unaffected carriers of variants in the same gene face a one-in-four risk in each pregnancy, and identifying such couples before or early in pregnancy allows informed reproductive choices. The variant classes that matter here span the full spectrum, since recessive disorders can be caused by single-base changes, by indels such as the CFTR F508del variant, and by copy-number changes such as the SMN1 deletion behind spinal muscular atrophy. This is why modern carrier screening for conditions like spinal muscular atrophy must specifically test copy number rather than rely on single-base analysis alone. The clinical value of carrier screening depends on the same interpretive discipline as diagnosis, because a variant of uncertain significance found during screening provides little actionable information. Carrier screening illustrates how variant interpretation supports decisions made by healthy people, not only the diagnosis of those already ill.

Variants of Uncertain Significance

Perhaps the most clinically consequential feature of variant interpretation is the sheer prevalence of uncertainty. As sequencing has expanded, the fastest-growing category of result has become the variant of uncertain significance, a finding for which the evidence is genuinely insufficient to call the variant harmful or harmless. These uncertain results create a real clinical burden, because they can generate anxiety, prompt unnecessary surveillance, and occasionally drive irreversible decisions on the basis of a finding that later proves benign. The recommended posture is patience: most uncertain variants are eventually reclassified as benign as population data accumulate, and periodic reanalysis of an uncertain result is now considered good practice because a variant called uncertain today may be resolved within a few years. The uncertainty is not evenly distributed, falling much more heavily on people of non-European ancestry whose variants are under-represented in reference databases, an inequity addressed in its own section below. Managing uncertain results well, neither ignoring them nor over-reacting to them, is one of the central skills of clinical genetics.

Longevity-Specific Considerations

For a longevity-oriented reader, genetic variation matters in two opposite directions. The first is the rare high-effect variant that accelerates an age-related disease, such as the LDLR variants behind familial hypercholesterolemia that drive early heart disease, or the LMNA variant that causes the accelerated-aging syndrome progeria described on the chromosomes page. Identifying such a variant can move prevention decades earlier than it would otherwise begin. The second and more distinctive direction is the rare protective variant, which reveals how the body can be unusually shielded from age-related disease. The clearest example is the loss-of-function variants in PCSK9, which lower LDL cholesterol for life and were associated with a roughly 88 percent reduction in coronary heart disease in the people who carry them, an observation that directly inspired a class of cholesterol-lowering drugs. Studying naturally protected individuals, rather than only sick ones, has become a recognized strategy for finding targets that mimic lifelong protection. For the slower, polygenic component of aging, the relevant variation is the large set of common, low-effect variants whose combined influence is the subject of the polygenic risk scores page. The longevity lesson is that the same catalogue of variation that explains disease also contains the natural experiments that point toward prevention.

Equity and Ancestry Considerations

The reference databases and clinical evidence that underpin variant interpretation were built disproportionately from people of European ancestry, and this imbalance has direct consequences for equity. When a person from an under-represented population carries a variant, that variant is less likely to appear in reference databases, which can make it look artificially rare and therefore more suspicious, and it is less likely to have an existing interpretation in shared archives such as ClinVar. The practical result is that variants of uncertain significance are returned far more often to patients of non-European ancestry, meaning the people already least served by genomic medicine receive the least actionable results. This is not a property of the variants themselves but of the data available to interpret them, and it narrows only as reference cohorts grow more diverse. Efforts to expand sequencing in under-represented populations are therefore not merely a matter of fairness but a scientific necessity for accurate interpretation. Until those databases are balanced, every variant result in an under-sampled population should be read with the awareness that uncertainty may reflect missing data rather than true risk. This ancestry imbalance recurs across genetic testing and polygenic risk scores and is one of the defining open problems of the field.

Limitations and Open Questions

Several limitations temper the clinical picture of variant interpretation. The largest is uncertainty itself, since the rate of variant discovery vastly outpaces the rate at which variants can be tested in the laboratory, leaving most rare variants without direct functional evidence. Computational predictors help triage this backlog but are imperfect, disagree with one another, and perform poorly in the non-coding genome that contains most common disease-associated variation. Whole classes of variant, particularly structural variants and repeat expansions, are detected unreliably by the short-read sequencing that dominates practice, so they remain systematically underascertained without long-read or specialized methods. Penetrance, the probability that a variant actually causes disease in a carrier, is frequently overestimated because early estimates came from families ascertained precisely because they were severely affected. And the ancestry imbalance in reference data means interpretation is most reliable for the populations already best studied. None of these limitations undermines the established principles of variant interpretation, but each marks a frontier where current practice is incomplete and where the answers a test can give remain provisional.

Practical Application

Reading a Variant in a Genetic Report

The first step in reading a variant is to understand the standardized name it is given, written in the nomenclature defined by the Human Genome Variation Society. Such a name identifies the gene, the precise position of the change, and its effect at both the DNA and protein levels, so that the cystic fibrosis variant, for instance, is written as c.1521_1523del at the DNA level and p.Phe508del at the protein level, capturing that three bases and one phenylalanine are deleted. The report should also state the variant’s classification on the five-tier scale, from pathogenic to benign, and the zygosity, meaning whether one or both copies of the gene are affected. Equally important is knowing which method produced the result, because a panel that reads single bases cannot detect a large deletion, a repeat expansion, or a deep-intronic change, so a normal result from one method does not exclude a variant of a different class. Reading these elements together, the name, the classification, the zygosity, and the method, converts a line in a report into a meaningful statement about a person’s genome. Anything ambiguous in that statement is a reason to seek expert interpretation rather than to guess.

Tools and Databases

A small set of public resources operationalizes the principles of variant interpretation, and knowing what each one offers is the core of practical variant literacy. The Genome Aggregation Database, gnomAD, reports how often a variant appears across large and diverse populations and ranks how strongly each gene resists damaging mutation, providing the frequency evidence that anchors interpretation. ClinVar aggregates the clinical interpretations that laboratories have assigned to variants, letting a reader see whether a variant has been classified before, by how many submitters, and whether they agree. Annotation tools such as the Variant Effect Predictor describe where a variant falls and its likely functional consequence, while predictors including CADD, REVEL, SpliceAI, and AlphaMissense estimate how damaging coding and splice-altering changes are likely to be. As with all such tools, the outputs are evidence to be weighed, not verdicts to be obeyed, and they describe populations and predictions that may not fit an individual case. Using them well means matching the resource to the question and respecting the limits of each.

When to Involve a Specialist

Although the concepts of genetic variation can be learned by any motivated reader, interpreting a specific clinical result calls for professional judgment. A variant reported as pathogenic or likely pathogenic, an uncertain result with potential medical implications, or any finding that might inform a reproductive or treatment decision should be evaluated by a clinical geneticist or genetic counselor. These specialists can apply the full ACMG and AMP framework, weigh the variant against the patient’s personal and family history, arrange confirmatory or additional testing for variant classes the original method could not detect, and explain what a result does and does not mean. Findings from direct-to-consumer genetic tests in particular warrant confirmation in an accredited clinical laboratory before any medical action, because such tests interrogate only a fraction of variants and are not designed for clinical decision-making. The recurring principle is that learning what variants are builds the literacy to ask good questions, while turning a specific variant into a sound decision remains a specialized skill. This page is a foundation for understanding, not a substitute for individualized evaluation.