Incomplete Penetrance and Variable Expressivity: From Clinical Studies to Population Cohorts

1Approximately 72% of all rare diseases are genetic in origin, and most of these are thought to be monogenic in nature. Rare, deleterious variants are known to cause thousands of different genetic disorders in humans, and while the molecular basis of over 6,000 monogenic diseases has been uncovered, with more than 200,000 pathogenic variants described, the genetic basis of most rare disorders remains to be determined. With advances in next-generation sequencing (NGS) and the increasing availability of whole exome/genome sequencing (WES/WGS), the study of genotype–phenotype relationships has become more widespread as determining how the genotype causes a phenotype is a fundamental step toward understanding disease pathology. Protein-coding variants that are associated with disease phenotypes directly link DNA variation to altered protein function or dosage and to the phenotypic outcome, and so much of what we know about the genotype–phenotype relationship is based on the study of rare variants that cause monogenic disease. Monogenic genotypes can be highly predictive for specific individual disorders, but sometimes this relationship can be complicated, with some damaging dominant monogenic variants not following the expected Mendelian inheritance patterns. Individuals with the same genotype can display distinctly different clinical phenotypes, including being clinically asymptomatic. Currently, there are gaps in translating how the individual genomic variation affects phenotypic presentation and how genetic variants exert their functional impact to cause disease.

2The study of genetic disease has often been divided into rare monogenic forms of disease and more common polygenic complex disorders. Current evidence suggests that these groups may be more overlapping than previously thought as the genetic variation present across the genome highlights the complexity underlying the phenotypic presentation. There are both rare variants in individual genes that cause monogenic forms of complex disease and common variants that affect the severity of monogenic disease. Such complexity makes investigating the genotype–phenotype relationships more complicated, which is only exacerbated by erroneous variant associations due to study design problems. Human genetic diversity displays considerable variability, with individual genomes differing from the reference at 4.1–5 million sites. Although most variation is common and predicted to be functionally neutral, each individual has on average 85 heterozygous and 35 homozygous protein-truncating variants (PTVs). Population cohort studies have shown that the average genome contains around 200 very rare variants per person and 54 variants previously reported as disease-causing, including 7.6 rare non-synonymous coding variants in monogenic disease genes. Variant interpretation is an ongoing challenge within diagnostic medicine, making understanding the phenotypic consequences of underlying genetic variation a key aim of genomics research.

3A deleterious genotype should be no more prevalent in the population than the disease that it causes. However, the same genetic variant can result in different disease presentations in different people, from clinically asymptomatic to severely affected, even among members of the same family. The proportion of individuals who possess a particular genotype and exhibit the expected clinical symptoms is defined as the penetrance of that genotype. If everyone with the genotype presents with clinical symptoms by a particular age, then it is said to be fully penetrant, whereas if it falls below this, it is said to exhibit reduced or incomplete penetrance. Genotypes can also display variable expressivity, where the severity of the phenotype caused by the genotype can vary among affected individuals (Table 1); this differs from pleiotropy, where different variants in the same gene can cause different, potentially unrelated phenotypes that may even be categorized as different diseases (Figure 1). Although penetrance, expressivity, and pleiotropy are three distinct concepts, biological reality means that their overall effects often overlap, especially in population cohorts where it is difficult to identify the cause of the phenotypic diversity. Multiple distinct phenotypes, in aggregate, could either be classified as a single more severe phenotype or different disease subtypes. As these three are likely to be caused by overlapping or similar mechanisms, especially in genetically heterogenous conditions, we will discuss them together in this review.

4Incomplete penetrance can be observed in both dominant and recessive conditions. However, the cause of variability in genotype–phenotype correlations can be difficult to elucidate; phenotypic variation has been observed in mice with identical environmental and genetic backgrounds, including variability in lethality for gene knockouts despite the introduction of identical variants. Establishing that a identified variant is the sole (or primary) cause of an individual’s clinical phenotype can be difficult, which is an important concern when it comes to diagnosis and providing accurate genetic counseling, and such difficulties can lead to incorrect or delayed diagnosis. The widespread presence of incomplete penetrance and variable expressivity through many overlapping mechanisms (Figure 2) can explain why apparently unaffected parents can pass on pathogenic variants to affected offspring and why seemingly healthy individuals’ genomes can contain a large number of putatively damaging variants and yet not suffer any obvious adverse effects.

5Although databases of clinically identified variants in affected individuals are useful for assessing pathogenicity, population-based datasets that include WES/WGS alongside phenotypic and medical information are increasingly important for investigating the penetrance and expressivity of these variants. Large population cohort studies have shown the occurrence of apparently pathogenic variants is much higher than previously estimated through small clinical or familial cohort studies, and their frequency highlights either the incomplete penetrance, variable expressivity, or misclassification of such variants. The existence of PTVs in dosage-sensitive genes in healthy individuals also remains problematic when it comes to determining pathogenicity. The potential for genomic technologies and WGS to detect individuals at risk of genetic disease is enormous, but incomplete penetrance and variable expressivity present a challenge for clinicians, especially when an incidental finding occurs without any prior clinical indication, leading to uncertainty over whether a clinical phenotype will develop, and if so, when. This problem is highlighted when testing unselected population cohorts, who may or may not have phenotypes of relevance to genomic findings at the point of testing. To understand how genetic disorders develop, we need to consider how deleterious variants interact with the rest of the variation in the genome and how variation can affect phenotypic presentation. This may also identify targets that help prevent disease progression. The presence of putatively pathogenic variants in asymptomatic adults also highlights the possibility that there are disease resistance mechanisms we can identify through the sequencing of general population cohorts.

6Traditionally, rare pathogenic variants were identified in small phenotypically enriched clinical cohorts of individuals and families with similar monogenic disease. Population cohorts allow us to utilize the information from small clinical studies to investigate the penetrance of variants in the general “healthy” population, where such severe monogenic phenotypes are likely to be depleted, and the potential to identify the causes of clinical heterogeneity. Ascertainment bias can occur with any study design, with volunteer population cohorts tending to be healthier than the average individual and clinical cohorts tending to have more severe phenotypes. Estimates of the maximum and minimum variant effect sizes across different ascertainment contexts are needed to avoid falsely predicting that a significant proportion of the healthy population is at risk for a monogenic condition. The proportion of individuals affected and the average age of onset (i.e., age-dependent penetrance) can vary depending on the ascertainment context (Figure 3). For example, individuals with putatively pathogenic variants in HNF1A and HNF4A, known for causing maturity-onset diabetes of the young (MODY), develop diabetes significantly later or not at all when tested outside of the context of clinical referrals for the suspected MODY.

7For almost all human genetic diseases, individual variability in the phenotype is influenced by background variation in the genome. As genetic testing has become more widely available, both through healthcare systems, direct-to-consumer testing, our understanding of how genomic variation affects disease progression and prevalence has become significantly more important, both for clinical utility and for our functional understanding of the disease. Variation in the genome can predispose individuals to disease through traditional monogenic variants that disrupt physiological pathways and exert a large effect on the phenotype, or through the accumulation of polygenic effects that involve many variants of small effect sizes in different pathways, or as is increasingly becoming clear, through their combined effect.

8Within population cohorts, penetrance estimates for monogenic variant carriers average 60% or lower for most conditions, illustrating that many individuals have highly penetrant, pathogenic variants in known monogenic disease-causing genes who never develop the corresponding phenotype. Generally, 70% of the “Wellderly” healthy aging cohort, all of whom reached 80 without any chronic diseases, had one heterozygous deleterious variant in genes listed in the American College of Medical Genetics and Genomics (ACMG) secondary findings. Similarly, one in 75 (1.3%) of healthy elderly individuals in the APSREE trial carried a previously identified pathogenic variant, including in Lynch syndrome and familial hypercholesterolemia genes, without having the associated phenotype. These cases demonstrate that carrying such pathogenic variants does not always cause the associated disease and that other mechanisms may contribute to the protection of human health, including genetic modifiers that ‘rescue’ individuals from a disease phenotype.

9For genetically heterogenous monogenic diseases, the penetrance and expressivity can vary between different genes or variants, with the same phenotype potentially caused by numerous different variants across multiple genes. Even within the same gene, some deleterious variants in known monogenic disease genes may exhibit complete penetrance, while others show incomplete or low penetrance. Variation can be due to functional redundancy of genes, or the location and type of variant, with missense and PTVs in the same gene often causing different phenotypes. For example, hereditary angioedema can show great phenotypic diversity, even among members of the same family, and individuals with missense variants in SERPING1 typically display a milder and later onset of disease than patients with PTVs. In contrast, missense variants in BMPR2 cause earlier and more severe pulmonary hypertension than PTVs in the same gene.

10Pathogenic PTVs typically cause disease through loss of function (LoF) due to degradation of the RNA by nonsense-mediated decay (NMD). NMD is an mRNA surveillance pathway that recognizes and degrades damaged mRNA transcripts that would produce misfolded or shortened proteins that can accumulate in the cell and initiate the endoplasmic reticulum (ER) stress response. However, the production of a variant protein can either exacerbate disease severity through the accumulation of toxic proteins in the cell or alleviate it through providing a residual function that protects against haploinsufficiency-mediated disease in the heterozygous state, meaning the occurrence of NMD can affect phenotypic severity depending on the mechanism of disease. PTVs may also cause LoF through aberrant splicing, which is also regulated by NMD. In some cases, the location of NMD boundaries at the 5′ and 3′ ends of genes containing causal variants can explain phenotypic variation between individuals with different PTVs in the same gene. For example, PTVs located outside of the region that triggers NMD in SOX10 escape NMD and produce proteins that have dominant-negative activity, causing the severe complex neurological disorder PCWH, whereas PTVs located within the NMD region produce transcripts that are recognized by NMD and removed, causing the relatively milder WS4 syndrome via haploinsufficiency. This variability in penetrance or expressivity could potentially be classed as distinct subtypes of disease, with different variants causing disease through different mechanisms and producing distinct syndromes. Pathogenic variants in KAT6B show a similar disease manifestation, with two distinct syndromes depending on whether NMD is triggered or not. Variants in KAT6A cause severe intellectual disability (ID) and neurodevelopmental disorders (NDD), with late PTVs more likely to cause a severe phenotype, compared to 60% of early PTVs which conferred a mild phenotype, potentially due to whether NMD is activated or not. The position of the PTV within the gene has also been seen to modulate the severity of clinical phenotypes in Marfan syndrome and Charcot-Marie-Tooth disease. Disease due to SFTPB variants typically presents in neonates as respiratory distress syndrome, resulting in death within the first few months; variants that allow partial production of the SP-B protein confer longer survival times and later onset of disease, whereas the variants that cause complete deficiency of SP-B due to NMD cause fatal neonatal respiratory distress syndrome.

11Missense variants can also result in LoF due to substantially reduced protein function or stability. Although many missense variants have little or no effect, they can result in conformational changes, increased protein misfolding, and aberrant protein trafficking, which can lead to intracellular retention or accumulation, increased ER stress, activation of the unfolded protein response, or increased pro-apoptotic signaling and apoptosis. Some missense variants, small insertions/deletions, and gene duplications can also result in gain of function (GoF) effects due to increased activity, increased protein production, or via protein products that gain a new damaging function. Some GoF variants can exhibit a more severe phenotype than LoF variants in the same gene; for example, GoF variants in KCNA2 were associated with more severe epilepsy phenotypes than LoF variants. Where in a gene a variant is located can affect the mechanism of disease, as well as penetrance and expressivity through molecular subregional effects; the impact of a variant depends on whether it is located at sites that undergo post-translational modification, within sites that are critical for tertiary and quaternary structure, at protein–protein interaction interfaces or ligand binding sites, or inside versus outside of functional domains. For example, missense variants in GRIN2A located in transmembrane or linker domains were more frequently associated with severe developmental phenotypes than those located elsewhere, such as within amino-terminal or ligand-binding domains, with a wide range of phenotypes observed from normal to mild epilepsy, to severe developmental phenotypes and epileptic encephalopathy; similarly, GoF variants in highly conserved regions of the potassium channel of KCNA2 were associated with more severe epileptic encephalopathy than variants located elsewhere. An improved understanding of the protein structure and the functionality of interacting domains will help elucidate specific variant effects on the resulting phenotypic presentation.

12Finally, there are a small but increasing number of pathogenic non-coding variants that have been identified as causes of monogenic diseases. These variants can operate either through LoF or GoF mechanisms by altering the gene or isoform expression. For example, biallelic variants in the PTF1A enhancer are a well-established cause of recessive pancreatic agenesis through tissue-specific LoF; de novo LoF variants in the 5′ untranslated region (UTR) of MEF2C have been shown to account for around a quarter of developmental disorder diagnoses in this gene; and a single GoF variant that creates a novel promoter has been shown to cause α-thalassemia. However, establishing the pathogenicity of non-coding variants is often much more challenging than coding variants, and thus, studies of penetrance and expressivity of these variants are likely to lag behind.

13Repeat expansion disorders are caused by genomic expansions of short tandem repeat (STR) sequences that either affect the gene expression or protein sequence, with the penetrance and expressivity affected by the number of repeats (Table 2). Anticipation is often observed in families due to molecular instability around the repeats; in each generation, the repeat length can increase, resulting in the earlier onset of disease and increased severity. For example, Fragile X syndrome is caused by the expansion of over 200 repeats in the CGG motif in the 5′UTR of FMR1 on the X chromosome, resulting in hypermethylation of the promoter, silencing the gene. Fragile X exhibits incomplete penetrance and reduced expressivity, with 100% of males and 60% of females presenting with ID and 50–60% of males and 20% of females diagnosed with autism spectrum disorder (ASD). Wild type (WT) alleles contain <44 CGG repeats, while full mutations in affected individuals typically have >200 repeats. Those with premutation alleles of 55–200 repeats have milder phenotypes than full mutation carriers, although they have an increased risk of Fragile X-associated tremor/ataxia syndrome and primary ovarian insufficiency prior to age 40 compared to WT. Monotonic dystrophy shows a similar mechanism, with unaffected individuals having 5–37 CTG repeats in the 3′UTR of DMPK and fully affected individuals having >80 repeats (although repeats of >1,000 have been seen in congenitally affected children), with an number of repeats correlating with the earlier age of onset.

14Although the number of repeats accounts for a large proportion of variable expressivity, there are still missing genetic factors accounting for differences in the age of onset. For example, in Huntington’s disease, a lower number of N-terminal CAG repeats in HTT is associated with reduction in penetrance and later onset of clinical symptoms, but while the number of repeats is inversely correlated with the age of onset of motor symptoms, they only account for 70% of the variability. The remaining unexplained variance displays a high degree of heritability, suggesting further genetic modifiers. Additional genetic variants in the DNA mismatch repair pathway have been linked with anticipation and overall severity of disease, and functional studies showing the knockout of base-excision repair or transcription-coupled repair pathways in animal and cellular models of nucleotide repeat disorders can inhibit the expansion and reduce the phenotypic severity. Variants in the DNA repair gene MSH3 have also been linked with differences in disease severity through somatic instability. As non-penetrant individuals will not necessarily come to clinical attention and large triplet repeats are hard to genotype accurately using NGS, it is suspected that individuals with fewer than 41 CAG repeats in HTT may exist at a higher frequency than previously expected in the general asymptomatic population.

15It has been hypothesized that the differential expression of alternative alleles in the gene containing causal variants could affect the presentation of phenotypic traits in individuals with identical genotypes. This mechanism has been proposed primarily for dominantly inherited conditions where haploinsufficiency is the cause of the disease, including Lynch syndrome and hypertrophic cardiomyopathy (HCM), where an allelic imbalance could cause either higher expression of the WT allele, thus compensating for the haploinsufficiency and resulting in reduced penetrance, or lower expression of the WT allele, thus exacerbating the haploinsufficiency and resulting in higher penetrance. Significant allelic imbalance has been observed in up to 88% of genes in human tissues, potentially caused by genetic modifiers or stochastic factors, and has been identified as both tissue-specific and genome-wide in mouse models. Structural variants such as duplications that are in trans with a pathogenic LoF variant can alleviate the potential clinical phenotype when disease would be caused by haploinsufficiency, by providing an additional WT copy of a gene, thus resulting in a normal level of gene expression, as has been observed in DiGeorge syndrome. Additional variants in the untranslated regions of mRNA can also affect the translational efficiency and gene expression can also vary widely across tissues, highlighting the importance of sequencing disease-relevant tissue in the interpretation of genetic variation. Compared to synonymous variants, rare missense variants show a significant reduction in allelic expression across many tissues in proportion to their predicted pathogenicity, suggesting deleterious variants are depleted from highly expressed haplotypes. Some highly differentially expressed genes have been shown to contain fewer disease-associated variants, which are less likely to accumulate on haplotypes that are highly expressed, or in high-penetrance combinations. For example, genetically heterogenous monogenic eye disorders display both incomplete penetrance and variable expressivity and also display significant variability in gene expression levels throughout the population. The differential expression of alleles has also been shown to play a role in the variable expressivity of Marfan’s syndrome.

16The differential expression of alleles can also potentially cause recessive conditions to present in a dominant fashion. For example, Zellweger spectrum disorder (ZSD) is an autosomal recessive disorder caused by deleterious variants in any of 13 PEX genes, with the most common cause being variants in PEX1 or PEX6. Affected heterozygous carriers have been identified with ZSD despite lacking a second pathogenic allele, with all affected heterozygotes presenting with the allelic overexpression of the variant allele compared to WT, and a common polymorphism has been linked to this allelic overexpression. In HCM, the proportion of sarcomeric proteins produced by variant alleles can vary with the allelic expression, and 30–80% of the sarcomere structure can be made up of proteins with reduced function, causing variation in overall phenotypic severity.

17Stochastic variation within normal cellular and developmental processes can potentially be amplified by disease-causing variants and thus play a role in incomplete penetrance and variable expressivity. Random monoallelic expression (RME) is the transcription of only one allele from a homologous pair and can be constitutive, with all cells expressing the same allele throughout (as seen in imprinted genes), or somatic, with individual cells showing variation in expression levels. Overall levels of RNA in cell populations tend to be stable, but dynamic allelic fluctuation through RME can present variability in the gene expression. Genes that show little RME are mostly housekeeping genes that have higher expression levels. Although no variation in the disease trait has yet been definitively linked to somatic RME, conceptually it could explain the phenotypic variation either through alteration of gene dosage or the higher expression of a variant allele. RME during embryonic development has been tentatively linked with variation in developmental disorders such as Holt-Oram syndrome. Model organism research has suggested stochastic variation in the gene expression can affect the expressivity of variant genotypes, with 20% of genes causing variation in phenotypes in two different isolates with defined genetic backgrounds in C. elegans. Phenotypic variability has also been observed in inbred mice with a defined genetic background, as well as in monozygotic (MZ) twins, suggesting the influence of stochastic molecular events in variable expressivity.

18Production of different transcripts of genes may also lead to the differential expression of traits and explain why potentially deleterious variants in haploinsufficient genes are found in population cohorts. Annotations based on transcription levels of different isoforms in haploinsufficient genes identified that 23% of LoF variants are in under-expressed exons and had similar effect sizes to synonymous variants. In monogenic cardiomyopathies caused by LoF variants in the giant muscle protein titin, studies of TTN expression levels indicate that LoF variants found in unaffected population cohorts occur predominantly in exons that are absent from the most highly expressed transcripts and thus do not cause the phenotypic effect associated with deleterious variants. Similarly, haploinsufficiency of TCF4 causes the highly penetrant Pitt-Hopkins syndrome, PTVs identified in these gene in unaffected individuals were all found to be located in minimally expressed exons, suggesting that functional protein can be made in the presence of these variants. The expression of tissue-specific isoforms can also affect the penetrance of a genotype, potentially resulting in distinct disease subtypes. For example, CACNA1C has two clinically important isoforms with mutually exclusive exons that explain two different forms of Timothy syndrome; pathogenic variants across the widely expressed transcript produce a multi-system disorder (type 1), while pathogenic variants in the alternative exon of a transcript predominantly expressed in the heart are much rarer and result in more severe cardiac-specific defects and fewer syndromic phenotypes (type 2). Further examples are likely to be uncovered through large-scale analysis of isoform expression in different tissues and at different times.

19Variants in regulatory regions can affect the phenotypic presentation of disease by altering the gene expression and through modulation of deleterious genetic variants found in associated protein-coding regions, potentially affecting the penetrance and expressivity of the monogenic variant. Cis-acting elements are DNA sequences located on the same haplotype as the gene they affect, whereas trans-acting factors are proteins or elements that bind to the cis-acting sequences to affect the gene expression. Variants in these non-coding regions can have multiple downstream effects, through interactions with other genetic features or through effects on monogenic variants. Small changes within transcription factor binding or expression can lead to dysregulation that affects multiple genes within the same regulatory network and therefore could potentially alter the final phenotypic presentation. Cis-regulatory variants have been identified that modify the penetrance of coding variants and therefore contribute to disease risk or presentation. Pathogenic coding variants are depleted from higher-expressed haplotypes with cis-regulatory variants in the general population, suggesting that individuals who present with a disease phenotype may have an enrichment of cis-regulatory variants that increase the expression of the pathogenic allele, compared to individuals who are asymptomatic who have an enrichment of ‘protective’ regulatory variants that decrease the expression and, therefore, penetrance of the pathogenic allele.

20Upstream open reading frames (uORFs) are tissue-specific cis-regulators of protein translation found in the 5′UTR region of protein-coding genes, and variants that alter uORFs can affect whether a deleterious protein-coding variant causes a disease phenotype or not and may alter the phenotypic presentation of the disease. Active translation of a uORF can reduce downstream protein levels by up to 80% via several mechanisms, including the production of a peptide that stalls the translating ribosome and termination at a uORF stop codon that can trigger NMD. Variation that either introduces or removes uORF start or stop codons can, therefore, affect the phenotypic presentation, and uORF variants may also have a role in disease pathology. Variants in the downstream 3′UTR region may also play a role in regulation of the gene expression through altering the mRNA stability or translational efficiency. For example, a common single nucleotide polymorphism (SNP) downstream of GATA6 has been shown to reduce its expression, potentially resulting in a more severe pancreatic agenesis phenotype when found in trans with a LoF variant in the same gene. Similarly, polymorphisms in the 3′UTR region of KCNQ1 have been suggested to alter the expression of the cis allele, either increasing the severity of the disease or reducing it through an uneven expression of WT or variant alleles. However, an attempt to replicate this in a diverse group of population cohorts found no association between the identified polymorphisms and the severity of disease, highlighting the difficulties with trying to identify non-coding modifiers of rare disease, both in clinical cohorts and population studies.

21Approximately 400,000 candidate enhancer regions have been identified in the human genome, with an average of around 20 enhancers per gene. Non-coding variants within enhancer regions can be a cause of phenotypic diversity through alterations in gene expression, therefore affecting overall disease phenotype presentation. Although identifying non-coding variants that affect disease presentation can be very difficult, there are some notable examples. A large study identified an SNP in an intronic enhancer of RET that appeared to increase the penetrance of Hirschsprung disease in patients with rare RET/coding variants. Intronic variants have also been suggested to affect the penetrance of coding variants in patients with Stargardt disease, where a deep intronic variant has been shown to be a major cis-acting modifier of the most common pathogenic variant in ABCA4. A small study also suggested that SNPs in promoter regions affect the severity of arrhythmias among individuals with LoF variants in SCN5A. Variants that create novel binding sites for transcription factors have been implicated in affecting penetrance through altering the gene expression, including a common non-coding polymorphism that alters the hepatic expression of SORT1, contributing to myocardial infarction. Further WGS research is needed to identify non-coding variants that affect gene expression levels.

22Genes are often associated with multiple cis-regulatory elements through topologically associated domains (TADs). These domains are thought to affect the gene expression and mediate the effects of cis- and trans-regulatory factors through the 3D conformation of chromatin, and therefore, variants in these domains can affect penetrance and expressivity of genotypes. Although the expression of some genes has been shown to be unaffected by changes in TADs, the creation of new TADs has been implicated in the pathogenicity of rare duplications. Alterations to the 3D chromatin structure within and between TADs can lead to mis-alignment of genes, enhancers, and silencers, affecting transcriptional control of the gene expression. Variants in TAD loops may have no effect on healthy individuals but could affect disease presentation in those with an underlying monogenic variant. Common genetic variants in cis-regulatory domains can affect the gene expression, and rare variants have been identified that disrupt the structure of the domain, and both could contribute to varying phenotypic expressivity of identical protein-coding sequences by causing changes in upstream mechanisms of gene regulation. Structural changes that affect transcription factor binding can lead to functional gene expression changes, as seen in the EPHA4 locus, where deletions or duplications that overlap the TAD boundary can cause severe limb malformations, while deletion of the entire locus does not, which is thought to be due to differential gene enhancer associations.

23Postzygotic de novo mutations that occur during cell division can result in somatic genetic variation that differs between cells, leading to mosaicism. Monogenic disease is usually less severe in mosaic individuals than those who have the same variant expressed constitutively and depending upon which cells or tissues contain the pathogenic variant, mosaicism can result in non-penetrance or reduced expressivity. Somatic mosaicism is suspected to be more widespread than is usually detected, especially when testing only a single tissue sample that may or may not contain the clinically relevant variant(s), although NGS is making it easier to identify lower-level genetic changes.

24Mosaic somatic variants have been suggested to be more representative than germline variants of the true diversity and range of potential variation in human disease as genotypes that are lethal in the constitutive form can be identified when present as mosaic. These include variants that cause osteogenesis imperfecta, where a mosaic father presented with mild symptoms, but the constitutive form was incompatible with life, Proteus syndrome and CLOVES syndrome, two overgrowth disorders that are lethal in the constitutive form, and various mosaic aneuploidies. Alternatively, mosaic individuals can display different or milder phenotypes than those with germline variants in the same gene. For example, mosaic individuals with a variant in HRAS present with benign keratinocytic epidermal nevi (“woolly hair”), whereas those with the same constitutive variant have the more severe Costello syndrome. Other diseases that have been demonstrated to show a milder phenotype when caused by somatic mosaicism include telangiectasis and polycystic kidney disease. Mosaic genotypes can also display varying phenotypes that include segmental forms of the constitutive disease, such as segmental neurofibromatosis type 1, where clinical manifestations are only shown in certain parts of the body. In addition to presenting with variable expressivity, mosaic variants can also be incompletely penetrant. In individuals with primary immunodeficiencies, 80% of mosaic individuals were clinically asymptomatic, with the remaining 20% exhibiting partial clinical symptoms. Similarly, mosaic chromosomal aneuploidy has been shown to be incompletely penetrant in population cohorts, with women who had 45,X/46,XX mosaicism presenting with normal reproductive lifespan and birth-rate and no cardiovascular complications, compared to those with the non-mosaic genotype. Unaffected parents with mosaic pathogenic variants can pass their genotype onto their offspring as a constitutive germline variant, so an incompletely penetrant or milder disease in one generation can cause a completely penetrant disease in the next.

25Somatic mosaicism can also rescue an individual from disease, through cellular reversion that reduces the expressivity of a phenotype. For example, somatic reversions have been observed in several cell lineages from individuals with immunodeficiency caused by biallelic variants in DOCK8, including variants that correct or remove germline PTVs, and recombination events that attenuate or remove the deleterious variant from one allele. These somatic reversions improve overall survival time, but they are unable to completely eliminate the disease phenotype. Somatic reversion has been observed in other primary immunodeficiencies and may partially explain incomplete penetrance. Reversion of the clinical phenotype in individuals with recessive dystrophic epidermolysis and Fanconi anemia has also been identified. Remarkably, long-term remission from WHIM syndrome, caused by GoF variants in CXCR4, was seen in an adult who had undergone chromothripsis of chromosome 2 resulting in deletion of the disease allele in a single hematopoietic stem cell, leading to the repopulation of the bone marrow with the haploinsufficient CXCR4 cells.

26Epigenetic modifications are molecularly heritable changes that alter gene expression without altering the DNA sequence itself, including DNA methylation, histone modifications, and microRNA (miRNA) expression. Differential epigenetic modifications between individuals carrying the same pathogenic genotype can potentially account for incomplete penetrance and variable expressivity of the phenotype. DNA methylation is important in the control of tissue-specific gene expression, alternative splicing, prevention of cryptic initiation of transcription from alternative promoters, and X chromosome inactivation, all of which have been shown to affect the progression of disease. Studies of MZ twins that are discordant for disease phenotypes have highlighted how epigenetic mechanisms could affect the penetrance or expressivity of disease. For example, MZ twins with neurofibromatosis, caused by variants in NF1, showed significant discordance in the presence of tumors and severity of scoliosis, suggesting that additional non-hereditary factors were modifying their phenotypes. Similarly, one MZ twin with a pathogenic homozygous variant in GBA was diagnosed with Gaucher disease, while the other was clinically asymptomatic, and differences in their epigenome were posited as a mechanism to explain this discordance. However, epigenetic studies are generally more challenging than genetic studies as variation may be both tissue and time-specific, making it harder to elucidate how epigenetic mechanisms affect the penetrance of such genotypes. One suggested mechanism is that epigenetics may compensate for the presence of a deleterious variant, and segregate through several generations without any ill effects until the epigenetic modifications are no longer functional. This has been seen in Xq24 microdeletions that are inherited from mothers with extremely skewed X-chromosome inactivation, which modifies the penetrance. Skewed X inactivation is also suggested to be a cause behind the clinical heterogeneity in Klinefelter syndrome. Epigenetic mechanisms have also been suggested to partially compensate for deletions in healthy carriers of IMMP2L deletions, which cause ID and NDD, as reduced DNA methylation levels were seen in healthy carriers but not in affected offspring.

27Another mechanism by which epigenetic changes may affect the penetrance of monogenic diseases is via miRNAs, small non-coding RNAs that regulate the gene expression. One miRNA can influence multiple genes, and a gene can be affected by several miRNAs, potentially highlighting how variants in one miRNA may lead to multiple downstream phenotypic effects. Differential miRNA expressions can be caused by genetic variation, and variants within miRNA could, thus, affect the allelic expression and modify the penetrance or expressivity of monogenic diseases. The expression of numerous miRNAs may affect the penetrance and expressivity in hereditary breast and ovarian cancer (HBOC); incomplete and age-dependent penetrance is common in carriers of pathogenic variants in BRCA1 and BRCA2, and variation in several miRNAs that bind the 3′UTRs and downregulate the expression of both genes have been linked with an increased risk of earlier onset cancer.

28There may be a threshold that has to be met for the manifestation of a clinical disease phenotype, and genetic and other factors may vary in their relative contribution to meeting this threshold for different diseases and in different individuals (Figure 4). Some highly penetrant monogenic disease variants may always be sufficient to push the genetic burden above the threshold of the disease, although secondary variants may still contribute to severity. For example, Dravet syndrome (DS) is a highly penetrant and devastating form of childhood epilepsy caused by de novo LoF variants in SCN1A. Although DS displays considerable clinical heterogeneity within families and severity may relate to background genetic variation, there are no known modifiers that protect against the effects of the primary causal variant; the LoF variant alone is sufficient to push the individual above the threshold for disease and other variants can only change the severity of the phenotype above this point. Individuals with monogenic variants that are causative of disease alone and, thus, are already above the threshold for disease can be further modulated by secondary monogenic variants in related genes that also cause the same phenotype, and the accumulation of these PTVs is associated with a more severe phenotype as the burden is pushed way beyond the threshold. For example, in monogenic polycystic kidney disease, individuals with PTVs in each of the causative genes, PKD1 and PKD2, present with a much more severe disease than those with just one PTV. Many monogenic disease-causing variants have been found to have secondary genes or loci that affect the severity of their related clinical phenotypes (Table 3).

29In contrast, some monogenic disease-causing variants may be partially tolerated and transmitted through unaffected generations unnoticed, until they surpass the threshold for causing disease in the presence of other contributory factors. For example, large copy number variants (CNVs) are well-known causes of NDDs, but some—such as recurrent 16p12.1 deletions—have been widely observed to be inherited from unaffected parents. In this case, the penetrance of a phenotype that is severe enough to present clinically requires an additional variant that modulates the primary genetic variant supporting a “two-hit” model of NDDs. Similarly, deleterious variants in CNTNAP2 and LRRC4C are insufficient to cause the disease alone but together may impair the development and function of synapses, suggesting a possible digenic mechanism for modulation of phenotypes. In many cases, however, there are likely to be numerous factors that affect whether an individual lies above or below the disease threshold, including the overall deleteriousness of the primary causal variant(s), the level of expression of the causal gene or isoform, and other genetic and non-genetic modifiers (Figure 4). Global modifiers that might affect penetrance and expressivity include polygenic risk, genetic compensation, variation in the NMD efficiency, family history, age, sex, and environmental factors.

30The penetrance and expressivity of genotypes can be altered through the accumulated impact of many common genetic variants throughout the genome. The “omnigenic” model proposes that due to their interconnected nature, variants in gene-regulatory networks that are expressed in disease-relevant cells or tissues may affect the functioning of “core” disease-related genes due to effects on genes outside of the core pathways, suggesting that many unrelated variants contribute to the presentation of a phenotype. Proposed as a factor in the inheritance of complex traits, this polygenic architecture could potentially also affect the presentation of monogenic conditions in a similar way, through non-coding variation that affects overall gene regulation, and many loci have been shown to additively affect expressivity and penetrance of monogenic variants in model organisms.

31Genome-wide association studies (GWAS) have uncovered thousands of susceptibility loci for hundreds of diseases, suggesting that the polygenic background can either predispose or protect individuals from diseases. Polygenic background can be quantified into a polygenic risk score (PRS) and potentially used as a tool for the prediction of the overall disease risk in both monogenic and polygenic disorders. PRS associations highlight the additional risk of polygenic components in affecting the severity of monogenic disease, with the polygenic risk being shared across monogenic variant carriers and the general population. The effect of PRS has been widely explored to improve clinical interpretation of the penetrance of pathogenic variants across a range of monogenic conditions, including numerous familial cancer syndromes. The penetrance estimates for individuals with a pathogenic BRCA1 or BRCA2 variant range from 45 to 85% for breast cancer and from 10 to 65% for ovarian cancer, some of which can be explained by a polygenic background. Using a PRS generated from breast cancer GWAS, it has been shown that individual carriers of monogenic variants have risk differences of over 10% between the top and bottom PRS deciles. Interestingly, the majority of the SNPs identified as polygenic risk variants in breast cancer are common non-coding variants within regulatory regions, the target genes of which overlap with other known somatic cancer driver genes. Polygenic risk can also have a large effect on phenotypic diversity, even within individuals who have a known monogenic variant, illustrating that the genetic architecture for many diseases can be viewed as a spectrum rather than a binary classification of clinically symptomatic versus asymptomatic. Although the overall polygenic contribution to the disease phenotype can be weaker in individuals with a monogenic variant, it can be useful in predicting overall penetrance and risk stratification.

32The phenomenon of genetic compensation (or genetic buffering), where another gene or genes in a network can functionally compensate for LoF variants, has been shown in model organisms and hypothesized to play a role in incomplete penetrance in humans. The upregulation of related genes or pathways or the differential expression of compensating alleles can help suppress a disease phenotype, either through a small number of compensatory mechanisms or via a global shift in the gene expression. The functional redundancy of genes and rewiring of affected genetic networks may affect the penetrance and expressivity of corresponding phenotypes, and the consequence of a pathogenic variant may be influenced by variation across the genome and explain why certain LoF variants are tolerated by some individuals but not others. Haploinsufficiency can influence the expression of other genes in the same network, to maintain homeostasis or suppression of disease phenotypes. The functional loss of one gene can be compensated for through functional redundancy. Genes that contain high numbers of PTVs in general population cohorts and thus are less likely to cause adverse phenotypes were found to belong to larger gene families than genes that contain known pathogenic PTVs, suggesting functional redundancy as a mechanism affecting penetrance. Further research is needed to find robust evidence of this mechanism in humans.

33The efficiency of NMD varies between individuals, which could act as a potential modifier of penetrance and expressivity of PTVs targeted by NMD, irrespective of the specific causal variant(s). The variation in the NMD efficiency across codons, genes, cells, and tissues can affect disease pathology. In studies of model organisms, the variant alleles that caused milder phenotypes were those that exhibited more NMD, with reduction in NMD being correlated with a more severe phenotype. In this case, NMD could either help trigger a compensatory response, or haploinsufficiency could produce a milder phenotype than accumulation of truncated proteins. Variants in genes that encode the NMD machinery, or that either downregulate or remove NMD activity, have been linked to several NDD and ID syndromes, including variants in UPF2, UPF3A, EIF4A3, SMG8, and RNPS1, highlighting its importance in development and phenotypic expression. Common polymorphisms within the NMD pathway have been suggested to cause differences in NMD efficiency, which could help explain differences in the expressivity of diseases caused by haploinsufficiency, with severity linked to whether they trigger NMD or not. Interindividual variability in NMD efficiency has the ability to alter the expressivity of genetic variants, by converting the cause of the disease phenotype from dominant-negative to haploinsufficiency, or vice versa. For example, two patients with the same PTV in the DMD gene displayed different clinical phenotypes, with one diagnosed with Duchenne muscular dystrophy, and the other with the milder Becker muscular dystrophy; here, the difference in the phenotype was suspected to be caused by weaker NMD efficiency in the less severely affected patient, which resulted in the production of the damaged but still partially functional DMD protein.

34Family history can be seen as a crude but effective proxy for the combined effect of many shared genetic and environmental modifiers of disease phenotypes. In many cases, the pathogenicity and penetrance of variants in monogenic diseases have only been determined through studies of large families with multiple affected individuals, which can make it difficult to disentangle the relative contribution of different modifiers. Family history is a well-known major risk factor for hereditary cancer syndromes, and the number of affected relatives increases the risk of a pathogenic variant carrier developing cancer. Although the evidence base for estimating penetrance in individuals without a family history is currently very limited, individuals identified with a pathogenic variant for a heritable monogenic disease but without a family history of that disease may have a lower penetrance than those with a family history.

35Evaluating genetic differences between affected and unaffected carriers in the same family—such as de novo variants or unique combinations of modifiers—can be informative for understanding penetrance. It has been shown that children with monogenic NDDs have an excess of other damaging genetic variants compared to their either mildly clinically affected or asymptomatic carrier parents, with the extra genetic burden being enriched in genes that are highly expressed within the brain and in neurodevelopmental pathways. Similarly, children with 22q11.2 deletion syndrome display a wide variability in IQ scores that is highly correlated with the scores of their immediate relatives. The IQ of individuals affected by 22q11.2 deletion syndrome follows a normal distribution curve, similar to that of the general population, only 30 points lower. The significant association seen between parental and proband IQ suggests that inherited genetic variants associated with intelligence may alleviate some of the deleterious impact of the 22q11.2 deletion on phenotypic presentation. The heritability of intelligence may be driven either by the cumulative effect of many common small-effect variants, similar to the heritability within population cohorts, or by a small number of rare high-effect variants. Similarly, individuals carrying 16p11.2 deletions present with variable phenotypic diversities and are frequently present in “healthy” general population cohorts, albeit with a range of cognitive and neuropsychiatric difficulties despite none of them reaching traditional clinical diagnosis threshold levels. Within these carrier individuals, the best overall predictor of the phenotype was that of the average of their parental phenotype for the traits of interest, with individuals displaying deleterious effects relative to their phenotypic family background.

36It can be argued that penetrance is an almost meaningless concept without specifying an age threshold as many diseases do not present until later in life. As we age, gene expression and chromatin structure across the genome change, which can increase the penetrance or expressivity of disease. Expression of certain genes can cause change in a predictable way throughout life, with some only being expressed in the foetus or during early childhood, and others only after this developmental period. For example, the relative proportion of two protein subunits in the NMDA receptor alters with age due to the varying expression levels of the two genes, GRIN2A and GRIN2B, which can alter phenotypic expression of deleterious variants in these genes; prenatally expressed GRIN2B is linked with severe cognitive defects from birth, while postnatally expressed GRIN2A is linked with epilepsies in childhood and schizophrenia in adults. Studies of individuals who are below the age-penetrant threshold for known age-dependent diseases could explain why some pathogenic variants are found in apparently asymptomatic population cohorts. Classical examples of conditions where penetrance increases with age include cancer predisposition syndromes such as Li-Fraumeni, Lynch Syndrome, and HBOC, where penetrance is affected by the accumulation of DNA damage over time. Meta-analysis studies have shown that the cumulative breast cancer risks for BRCA1 and BRCA2 pathogenic variant carriers by age 70 are 57–65% and 45–49%, respectively, highlighting the difficulties with predicting the course of disease even in known pathogenic variant carriers and the importance of considering family history as well as other genetic and environmental factors. Age-dependent penetrance of cognitive phenotypes is also seen in diseases caused by the slow accumulation of aberrant proteins, where variation can affect the rate at which the protein accumulates. For example, retinitis pigmentosa (RP) has been suggested to be caused by retention of misfolded proteins, which leads to upregulation of genes that encode for proapoptotic machinery, and leads to apoptosis of photoreceptor cells, accumulating damage over time and eventually reaching the disease threshold and causing penetrant disease. Age-dependent penetrance may also be caused through gradual loss of neurons, causing the associated disease phenotype when the number of surviving cells drops below a certain threshold or overcomes brain plasticity. For example, progressive and late occurring neurological manifestations in patients with DNMT1 variants may originate from the gradual loss of DNA methylation over time, affecting adult neurogenesis.

37The penetrance of age-dependent variants, present a diagnostic and prognostic challenge for individuals with such genotypes. Previously, testing for many conditions early in life was not possible, and so little is known about long-term effects of mildly deleterious variants. Variants in HFE cause hereditary hemochromatosis, which can lead to iron overload in adulthood, and were previously thought to be an adult-onset condition. However, healthy cohort studies of children have shown that the effects of homozygous variants in HFE can be seen in childhood and that the cumulative effect of excess iron over a lifetime may affect the penetrance of numerous iron-related diseases. Recent population studies of adults have also shown substantially higher morbidity in homozygous HFE variant carriers with increasing age. In this case early identification of individuals at risk can help with monitoring disease progression and introducing timely interventions (such as blood donation).

38Sex can affect the penetrance and expressivity of some genetic disorders, most obviously when deleterious genetic variants occur on the X chromosome, with hemizygous males more phenotypically affected than heterozygous females. Although differences in the penetrance of inherited variants based on sex have been reported in a variety of disorders, mechanisms behind sex-dependent penetrance outside those that occur on the X chromosome are mostly unknown. However, there are widespread sex-biased differences in gene expression, so differences in penetrance of phenotypes are also likely to be common. Females are less likely to be diagnosed with neurodevelopmental disorders than males, with a fourfold increase in the number of males diagnosed with autism spectrum disorders (ASD) compared to females, suggesting that there may be a female protective effect that affects the penetrance of such conditions. Girls diagnosed with ASD have an increased number of CNVs compared to boys with the same diagnosis, and asymptomatic mothers with children diagnosed with NDDs or ASD had a higher genetic burden of deleterious variants than fathers, suggesting there may be some other cause for the incomplete penetrance and variable expressivity in females compared to males. However, females are ascertained at a closer frequency to males when they are more severely affected, suggesting some bias in clinical ascertainment due to differing phenotypic presentations between the sexes, supported by the fact that males were more likely to be referred for genetic testing than females carrying the same autosomal variant.

39The environment can affect disease penetrance or expressivity in both a negative and positive manner and includes diet, drugs, alcohol intake, physical activity, ultraviolet light, in utero exposures, education, and socio-economic status, among many others factors. Epigenetic factors can provide a mechanistic link between the environment and gene expression, and studies of the human microbiome can also explain some extreme variability in genotype–phenotype presentation. However, although gene–environment interactions are likely to be widespread, they are often extremely hard to prove as the complete and systematic collection of an individual’s environment is almost impossible, and detailed relevant exposure data are rarely available alongside genetic data.

40Inborn errors of metabolism perhaps provide the simplest examples of monogenic diseases where both a pathogenic genotype and an environmental exposure are required to cause disease. A clear example of the dietary impact on phenotypic variation is phenylketonuria, a rare autosomal recessive disease that is usually detected through newborn screening, whereby individuals who have damaging biallelic variants in PAH can be put on a low phenylalanine diet to avoid serious disease progression. Later onset monogenic disease penetrance can also be affected by the environment, as seen in several cancer syndromes, including colorectal cancer, where inherited genetic variants interact with dietary variables and BMI to confer the overall risk. Cancer susceptibility can also be altered through gene–environment interactions such as smoking or sunburn, which can accelerate the accumulation of somatic variants that contribute toward tumorigenesis. Similarly, environmental exposure to cigarette smoke, air pollution, and other airborne toxins can cause accumulation of unfolded or misfolded proteins and therefore affect the penetrance or expressivity of chronic lung disease. Individuals who carry a damaging monogenic variant may also be more susceptible to some environmental exposures, which can affect phenotypic severity. For example, cystic fibrosis is characterized by progressive damage to the lungs, and non-genetic factors may account for up to 50% of the clinical variation seen. Environmental factors such as smoking, air pollutants, temperature, and high-fat diets have all been shown to affect the severity and progression of disease, and the specific CFTR variant can also modulate how much environmental impact has on disease severity. Environmental factors can also affect the presentation of disease in primary atopic disorders, commonly seen as monogenic allergic disorders, where diet, microbiome at the epithelial-environment interface, presence/extent of infection, and psychological stress can all affect the penetrance or expressivity of the related phenotype.

41Determining the penetrance and expressivity of a variant can be difficult because it is sensitive to ascertainment context, and many studies are designed to enable the discovery of causative pathogenic variants in clinically affected individuals rather than to analyze effect sizes in populations. This has been demonstrated in recent studies that stress the importance of cohort background for the determination of penetrance. Investigating clinically classified pathogenic variants in large population cohorts can provide additional information about penetrance and expressivity, or determine whether variants or genes have been misclassified. However, finding low penetrance pathogenic variants in large numbers of asymptomatic individuals challenges the concept of pathogenicity, particularly in the absence of known modifiers. What does it mean to describe a genotype as pathogenic if it is frequently found in individuals without disease and no explanation as to why? Reclassification of previously reported pathogenic variants occurs frequently, with variants first classified prior to the release of large population datasets showing a higher rate of reclassification. A study reappraising pathogenic variants in Brugada syndrome showed that only one gene (SCN5A) out of 21 could be definitively identified as causal, and another study has raised doubt over the involvement of 11/58 genes thought to cause inherited monogenic retinal disease. Variants that show low penetrance or a wide range of expressivity can also be potentially classified as risk alleles rather than causative variants. Some CFTR variants have been classified this way, with variations in cystic fibrosis phenotypes from very mild to very severe, and over 1900 different genotypes have been reported. Many genotype–phenotype associations are only reported once, or they are reported several times but with inconsistent results due to differences in data collection, differences in methods, or differences in cohort ascertainment. Associations can also differ due to poor annotation of coding genes, lack of relevant functional information for non-coding regions, sequencing and annotation errors, and varying penetrance and expressivity, making a simple binary classification of many genetic variants very difficult.

42An overlapping genetic basis between complex traits and monogenic conditions is becoming increasingly apparent across the genome. Deleterious variants in genes causative of monogenic disease can be further dysregulated by non-coding variants that are associated with common traits, and monogenic forms of numerous common complex diseases have been identified. This overlap can cause considerable complexity when it comes to determining genotype–phenotype relationships. The prevalence of incomplete penetrance and variable expressivity raises questions as to what constitutes a disease state as opposed to extremes of normal phenotypic variation, especially within conditions that show significant clinical heterogeneity, with many traits that constitute a clinical phenotype being the extreme end of either side of the bell curve of continuous distribution in the general population. Therefore, defining the penetrance of a genotype can be difficult, especially when there is ambiguity as to what defines the “disease state,” particularly for disorders where clinical features are only identified when they reach above a certain threshold.

43Relatively few studies have investigated low penetrant rare variants in detail or identified why such variants cause disease in one individual and not another. Despite increasing numbers of sequenced individuals, identification of genetic modifiers for monogenic conditions remains challenging. By definition, carriers of rare variants that cause monogenic conditions will be rare, with even fewer individuals having identical genetic modifiers that explain incomplete penetrance or variable expression. NGS approaches involving bioinformatic algorithms, including pathogenicity score-based prioritizations, can produce conflicting results and often need manual curation to identify candidate variants. A computational approach that could comprehensively analyze and prioritize candidate variants and potential modifiers would be a great advantage. Even in large population cohorts genome-wide analysis of genetic interactions lacks statistical power and can be easily affected by confounders. Many genetic modifiers are likely to be located in non-coding regions, making it challenging to determine their direct functional effect on the gene expression, especially as much of the genome is found to be bound by at least one transcription factor, many of which have no known function yet. Improved computational approaches to identify candidate modifier gene interactions across the genome are needed, as well as identification of functional non-coding regions and the genes that they affect, and machine learning approaches such as DeepSEA and Enformer could improve annotation of these regions.

44Participants in population studies are usually investigated in a research-based environment rather than a clinical context, and despite rigorous phenotypic collection in some population studies, individuals involved may have subclinical manifestations of disease phenotypes that were unnoticed at the time of recruitment, or were not recorded in their medical histories. Lack of comprehensive phenotypic data can make using population cohorts to calculate the penetrance of genotypes very difficult but can at least provide a lower boundary of penetrance, with small clinical studies providing the upper boundary. Variant interpretation guidelines suggest that the penetrance of pathogenic variants in general population cohorts should be taken into account when calculating the overall penetrance of such variants; however, even within healthy population cohorts there have been individuals identified with the associated phenotype but who have previously been described as unaffected, as well as individuals who display symptoms but are below the clinical threshold for classification. This is further complicated by conditions that are late-onset. In addition, genetic studies of human disease currently fail to capture the diversity that exists across the world, with most studies involving individuals of European descent. This issue directly affects penetrance estimates, particularly as GWAS results and PRS may not be transferrable across diverse populations due to differing allele frequencies. Many deleterious variants may not be sufficient alone to cause disease, and therefore, estimates of penetrance need to consider the presence of other genetic variants and potential environmental effects (Figure 4). Calculating the etiological fraction of rare variants in specific conditions may provide a useful way to evaluate the probability that a variant detected in an individual with disease is causative, and disease-specific variant classifiers may also be of use.

45As WGS becomes more common, individuals at risk of genetic disease will be identified earlier in life, potentially even from birth and often prior to the appearance of relevant phenotypes. This can have a positive impact on overall health, with individuals who have no family history but a previously unknown high risk of disease being identified, enabling preventative screening or early treatment interventions. However, it can also cause harm through overdiagnosis; As seen across a number of population cohort studies, healthy individuals can harbor many potentially deleterious variants without ever developing any clinical symptoms. The effective use of genomic data requires a comprehensive understanding of functional genotype–phenotype correlations, which goes beyond that of Mendelian inheritance patterns. The increased sequencing of unselected populations, linked with electronic health records or other longitudinal phenotypic data, gives us an unprecedented ability to identify and reclassify rare variants and calculate penetrance estimates for a wide range of diseases and genotypes. These large-scale studies are crucial to inform the development of genomic screening programs and the management of incidental or secondary findings. Discovery estimates of secondary findings vary from 1–3% of the population, with the majority of identified variants being those that confer susceptibility to cancer. Incidental findings are predicted to be detectable at an appreciable level in individuals in the general population, many of whom may never develop the corresponding disease, suggesting that more robust determinations of pathogenicity are needed, including penetrance estimates for those without a family history of the disease.

46Incomplete penetrance and variable expressivity are a significant concern for the correct interpretation of genetic variation and of diagnosing genetic disease. Correctly estimating penetrance and expressivity is challenging, with clinical cohorts and population studies both offering a different insight into its quantification. Although many monogenic disease-causing variants are fully penetrant, many are not, and improving our knowledge will involve WGS of population cohorts of increasing size and diversity, as well as functional studies of individual patients with specific clinical phenotypes. Achieving a mechanistic understanding of how incomplete penetrance and variable expressivity occur will help inform diagnostic and prognostic testing, clinical management, and accurate genetic counseling. To improve diagnostics and clinical interpretation of incompletely penetrant genotypes, a more sophisticated approach to disease genetics may be needed that integrates disease mechanism and specific variants with variation in levels of gene and tissue-specific isoform expressions and other genetic and non-genetic modifiers. Improving our knowledge of how variants exert their effects on genes, cellular pathways, and overall phenotypes will improve our understanding of disease and facilitate the development of new therapeutic interventions.