https://orcid.org/0000-0002-9729-6519
Figures
Abstract
Estimating heritability has been fundamental in understanding the genetic contributions to complex disorders like late-onset Alzheimer’s disease (LOAD) and provides a rationale for identifying genetic factors associated with disease susceptibility. While numerous studies have established substantial genetic contribution for LOAD, the interpretation of heritability estimates remains challenging. These challenges are further complicated by the binary nature of LOAD status, where estimation and interpretation require additional considerations. Through a systematic review, we identified LOAD heritability estimates from 6 twin studies and 17 genome-wide association studies, all conducted in populations of European ancestry. We demonstrate that these heritability estimates for LOAD vary considerably. The variation reflects not only differences in study design and methodological approaches but also the underlying study population characteristics. Our findings indicate that commonly cited heritability estimates, often treated as universal values, should be interpreted within specific population contexts and methodological frameworks.
Authors summary
Understanding the relative role that genetic factors play in the development of late-onset Alzheimer disease (LOAD) is important for designing studies that identify specific genetic risk factors. Multiple studies have estimated this heritability, but no comprehensive assessment has been performed to compare them. Here we analyze results from 23 papers that include both family-based analyses (six twin studies) as well as those based on genetic variation among unrelated cases and controls (17 studies based on single nucleotide polymorphisms). The proportion of the phenotype explained by genetics varies substantially among these papers, ranging from 3% to 79% with twin studies generally estimating larger genetic components. Each of the presented studies makes different assumptions regarding analytical strategies and source of data, although all samples were of European descent. These results emphasize the need to critically examine heritability estimates in the existing literature and incorporate their potential biases when interpreting and applying them to future research. This is true for LOAD as well as virtually all other complex diseases.
Citation: Liu S, Bush WS, Akinyemi RO, Byrd GS, Caban-Holt AM, Rajabli F, et al. (2025) Alzheimer disease is (sometimes) highly heritable: Drivers of variation in heritability estimates for binary traits, a systematic review. PLoS Genet 21(9): e1011701. https://doi.org/10.1371/journal.pgen.1011701
Editor: Renato Polimanti, Yale University, UNITED STATES OF AMERICA
Received: April 25, 2025; Accepted: August 25, 2025; Published: September 4, 2025
Copyright: © 2025 Liu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: As this is a systematic review, all data are available from the cited literature. Data were extracted from references 22, 25, 26, 36-41, 72, 74-79 and 87-93.
Funding: The following grants from the US National Institutes of Health supported this work: R01 AG072547 (M.P.-V., R.A., G.B., A.C.-H., J.H., F.R., C.R., S.W.), U19 AG074865 (M.P.-V., G.B., W.B., J.H., B.K., C.R., G.T., J.V.), and U01 AG058654 (J.H. W.B., M.P.-V.). All authors received salary support from the NIH. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Introduction
Heritability is commonly defined as the proportion of phenotypic variation attributable to genetic variation within a specific population [1], a concept first described mathematically more than a century ago [2]. Early estimates of heritability for human traits were determined using family study design, while more contemporary estimates leverage population-scale studies with genome-wide genotyping or sequencing data [1,3,4]. Heritability estimates from the 1980s through the early 2000s were initially used to demonstrate that a trait of interest has a measurable genetic component [5], justifying the search for specific genes using either genetic linkage analyses or genetic association [6]. Almost 20 years into the era of genome-wide association studies (GWAS) and genome sequencing, estimates of heritability can now be used to estimate how much of a trait’s genetic component is yet to be discovered [7,8], thereby justifying additional genetic studies [9,10].
Both how heritability is estimated and how it is used have evolved over the past few decades. The literature on heritability study designs, statistical methods, and applications is deep, highlighting general strengths, weaknesses, and misinterpretations [3,4,11–13]. The continued discussions underscore the importance of accurate estimations of heritability given the estimate provides insights into the genetic architecture of complex traits, further informing the relative contribution of genetic relative to environmental effects and thus guiding the downstream research priorities. One major misinterpretation is that heritability is a fixed quantity [3,4]. In reality, heritability estimates are complex and dynamic, changing as a result of variable environmental effects and their interactions with genetic factors [14–17]. Study designs and statistical methodology also impact resulting heritability estimates. As an example, the heritability of the quantitative trait human height has been studied extensively for over a century. Classical family studies have consistently estimated the heritability of height at ~ 80% [18], while the large-scale genetic studies of unrelated participants estimated that imputed common genetic variation accounts for 56% of the observed phenotypic variance [19], illustrating the ceiling and floor of heritability estimates possible when different study designs and methodologies are deployed on different datasets from different contexts and samples.
Methods to estimate heritability were originally developed for quantitative traits. The concept was then extended and applied to binary traits such as late-onset Alzheimer’s disease (LOAD), a complex neurodegenerative disorder characterized by memory loss and dementia among adults ≥65 years of age [20,21]. One of the most highly cited estimates presents a very high heritability for LOAD (79%) [22], and while the identification of LOAD-associated genetic variants and genes has been successful [23–27], the range of LOAD heritability across a variety of contexts and methodologies and its implications for the discovery of new associating genes are not yet fully appreciated. It is worth highlighting that the methods for binary outcomes, such as LOAD, require additional steps for heritability estimation (i.e., assuming a liability threshold model) and warrant cautious interpretation as compared to those for quantitative traits. This limitation is particularly true in the context of LOAD given its age-dependent nature. This latter point we argue is not thoroughly described in the literature. In this systematic review, we examine LOAD heritability estimates, illustrate differences across study populations and study designs, and highlight differences based on assumptions and methodologies employed. Overall, we found that published LOAD heritability estimates indicate that generalizability is limited. In addition, the variation in reported LOAD heritability reveals population or cohort-specific variability that may be important in both identifying LOAD loci and designing interventions among populations.
Methods
In this systematic review, we focus on narrow sense (h2) heritability of LOAD as this is generally the measure presented in the literature. Heritability can also be estimated as broad-sense (H2) heritability. H2 captures the total genetic variance, which includes additive (A) and dominance (D) effects, as well as the epistatic effect (I). In contrast, h2 only defines the additive genetic component. While closely related, these two metrics describe distinct but overlapping aspects of genetic influence on phenotypic variation within a population. Distinguishing these forms is important, as each has different implications and requires flexible interpretation in research contexts [11,28].
Data collection
Data were collected from the primary literature and included original, peer-reviewed studies of LOAD narrow sense heritability. Our primary literature search was conducted using PubMed, the National Library of Medicine’s free resource for the search and retrieval of citations and abstracts from MEDbroadLINE and PubMed Central (PMC) databases [29]. We formulated the search query as “Alzheimer’s disease” AND “heritability,” tailored to PubMed’s search functions, to identify articles that mentioned heritability values for LOAD. This initial search, conducted in April 2024, yielded 624 articles, including both research and review papers.
In addition to the initial search, we explored other resources to ensure that our review was comprehensive and not limited by PubMed or the corresponding search terms we used. These included relevant publications identified from citations of published twin registries focusing on LOAD, along with citations within and publications citing the recent large LOAD GWAS captured in the initial search results. This “backward searching” added 52 articles, and these articles were combined with our initial search results and screened using the approaches as described below.
To finalize the pool of articles to include in this review, we applied a multi-step filtering process to identify relevant studies from the initial search, excluding studies that only mention or restate published heritability values. Manual screening of the titles and abstracts of all 676 papers was first performed to exclude articles that were irrelevant, such as those not focusing on LOAD. This initial screening eliminated 64 papers, leaving 612 papers to consider. Of these 612, 23 were excluded because they could not be retrieved through institutional access or were non-English publications. For the remaining 589 articles, we conducted a comprehensive screening of both abstracts and full texts to identify studies that involved original research on the heritability of LOAD. We first removed studies that only referenced the heritability value(s) of LOAD without calculating their own estimates (n = 512; Fig 1). We then performed an in-depth full-text examination of the remaining 77 papers to determine if they met our criteria for conducting original research on heritability estimation for LOAD. We defined original research as studies with detailed methods for computing heritability, including details of the study design, modeling process, and study population. Manuscripts without sufficient detail were excluded from further consideration. For example, one retrieved twin study lacked explicit details on how the LOAD heritability estimates were derived [30]. A parent-offspring study was also excluded due to inaccessible supplementary materials that described both the study population and the detailed methodology [31]. In total, 54 manuscripts were excluded at this step (Fig 1), leaving 23 peer-reviewed manuscripts with sufficient detail for this review.
Characteristics of the literature
We categorized the 23 eligible studies based on the study design and methods employed. The evaluation included the data sources, the study design employed in the original studies, the assumptions embedded in each approach, and the populations to which these estimates are intended to be generalized [8,19,28,32–35]. Heritability estimation methods can be categorized into two primary groups: 1) family-based methods, rooted in studies of related individuals, such as twins and extended pedigrees, without necessarily requiring genomic data, and 2) single nucleotide polymorphism (SNP)-based methods, consisting of population-based methods leveraging genome-wide data from large samples of often unrelated individuals. Based on this, we examined the articles, focusing on their methodologies, and stratified them into twin-based and SNP-based heritability estimates. Among the 23 articles passing the criteria, seven were based on twin studies, while 16 were SNP based studies.
Results
We identified a total of 23 relevant studies published between 1997 and 2023, including both twin-based and SNP-based studies. In addition to the heritability estimate and standard error, for each study the following essential information was collected for all studies: study population characteristics, sample size, case criteria, and specific estimation methods (Tables 1–3). For SNP-based studies, we also collected number of SNPs, included covariates, summary statistics dataset reference, and prevalence used for liability transformation (Tables 2 and 3). The heritability estimates for LOAD across different study designs and populations ranged from 3.1% [26] to 79% [22]. These studies primarily focused on participants of European ancestry, with a mean age mostly greater than 70 years, except for one study leveraging the National Research Council Registry of Aging Twin Veterans that has an average age of 63.1 years [36]. The sample sizes varied considerably, ranging from 38 twin pairs [37] to 761,704 participants in a SNP-based study [26]. Compared with SNP-based estimates, heritability estimates from family-based studies generally provided higher values, representing the likely ceiling of trait heritability [8]. SNP-based estimates correspond to the proportion of genetic variance explained by SNPs assayed, imputed, or tagged and therefore are unlikely to capture the totality of genetic effects. Given the fundamental differences of what is being measured, we present our results in two main categories: twin-based heritability estimates and SNP-based heritability estimates. We then address details within each category by examining specific methodologies and their impact on the resulting heritability values, providing a comprehensive view of LOAD heritability across diverse study designs and populations (Box 1).
Box 1. Methodological considerations in binary trait heritability estimation
- Challenges for binary traits: As opposed to continuous trait, estimating heritability for binary traits like LOAD generally requires additional steps, thus warranting special consideration. The process typically involves the liability threshold model that assumes an underlying continuous liability distribution with a threshold inferred from population prevalence, above which individuals would be considered cases and below which as controls. In theory, the liability scale heritability could be compared across studies. In practice, however, the appropriate prevalence is hard to determine and also affects the resulting estimates.
- Age-dependent onset: Age-dependent complex diseases pose another challenge since the phenotyping is hard to determine as the symptoms might not present at the time of ascertainment or participants have not yet reached the age of typical onset, leading to misclassification as unaffected. This also makes obtaining appropriate prevalence estimates difficult due to varying prevalence across age groups. Prospective studies with adequate follow-up or survival analysis approaches could potential help address the issue.
- ACE modeling: Heritability estimation leveraging twin studies typically involves the ACE models that partition phenotypic variance into additive genetic (A), shared environmental (C), and unique environmental (E) components, with model selection (full and reduced models) based on a combination of different criteria to determine the influences of the specific components and can significant impact the heritability estimate obtained.
- GCTA vs LDSC: Both genome-wide complex trait analysis (GCTA) and linkage disequilibrium score regression (LDSC) are powerful methods to estimate heritability based on genome-wide SNP data, primarily for unrelated individuals, with the former requiring individual-level data while the latter only requiring summary statistics from genome-wide association studies. These methods rely on different assumptions and may yield different estimates, with LDSC generally leading to more conservative values.
Twin-based heritability
Seven studies using the family-based approach met our inclusion criteria, all of which were based on twin studies. Despite similar study designs and similar estimation approaches, LOAD heritability estimates based on twin studies vary widely, ranging from 37% to 79% (Table 1 and Fig 2). We further investigated the potential reasons that could lead to the differences among the estimated values, including variations in twin registries, study population characteristics, and LOAD phenotyping.
The figure presents the heritability estimates (bars, left y-axis) and corresponding sample sizes (line with points, right y-axis) for studies arranged chronologically along the x-axis. Each bar represents a single study, with colors denoting different twin registries utilized. Sample sizes are not presented for two studies (no points/line) since their sample sizes do not follow the definition on the y-axis.
Twin registries.
The majority of the identified studies used the Swedish Twin Registry (STR) [42], with two exceptions: one study leveraged the Norwegian Twin Registry (NTR) [43] and another the National Research Council Registry of Aging Twin Veterans (NAS-NRC), from the US [44]. The NTR study had the smallest sample size (n = 38 twin pairs with AD diagnosis) [37], a size expected given the study formulated the target probands via cross-referencing the cognitively impaired elderly born between 1895–1925 within the twin registry. The NAS-NRC panel, established in the mid-1950s, focused on white male twins [45]. The registry’s focus on white males and the high non-participation rate (24%) of the LOAD study [36] potentially makes its heritability estimates less representative compared with other estimates [44].
The STR-based LOAD heritability studies were more complex than those from the two registries discussed above, involving longitudinal efforts and variability within sub-studies. Three studies leveraged samples from the Study of Dementia in Swedish Twins [38] based on the Swedish Adoption/Twin Study of Aging (SATSA), a subset of the population-based STR with detailed enrollment criteria [46]. These studies included twins drawn from STR with variations in the inclusion criteria across studies, resulting in varying numbers of twin pairs in the analysis. In addition to the SATSA panel, Pedersen et al (2004) also leveraged the Origins of Variance in the Oldest Old: Octogenarian Twins (OCTO-Twin) [47] that incorporated longitudinal observations into the study, resulting in 662 twin pairs without symptoms with follow-ups spanning five years on average [40]. The sample size was further enriched with the introduction of the Study of Dementia in Swedish Twins (HARMONY) initiated in the year 1998 [48]. Prior to the HARMONY study, STR assessments were not comprehensive for dementia, limiting sample sizes. A larger sample size was achieved in the Gatz et al (2006) study (n = 392 pairs with at least 1 AD case) that employed a two-phase phenotype scheme where all participants were initially screened for cognitive dysfunction and suspected AD cases were followed up with complete clinical diagnostic evaluations. This STR LOAD heritability study also included opposite-sex twin pairs [22]. More recent efforts, as demonstrated in Karlsson et al. (2022), have further expanded the use of STR data by incorporating four STR sub-studies (SATSA, OCTO-Twin, HARMONY, and Aging in Women and Men) [41]. The inclusion of Aging in Women and Men (also known as GENDER) added value by introducing opposite-sex twin pairs to LOAD heritability research [49].
Other twin registries exist that are not included in this review. For example, the Finnish Twin Registry study by Räihä et al. (1996) [50] was also mentioned by Pederson et al (2001) as part of their exploration of LOAD heritability using a single-threshold model, although it was not the primary focus of their research [39]. The original study included same-sex twin pairs from Finland born before the year 1958, with disease status identified through the linkage to the Hospital Discharge Register, leading to a total of 94 AD affected out of 178 twin individuals [50]. Incomplete record linkage could potentially explain the lower heritability estimate of 63% reported by Pedersen et al. (2001) for the Finnish Twin Registry study [39].
Twin registry characteristics.
The study populations differed in numerous characteristics that may influence measures of heritability. Notable variation in participants’ ages, a demographic strongly correlated with LOAD prevalence [51], were observed across studies. The twin veterans study had an average age of 63.1 years (SD: 5.4) [36]. This younger sample might not have fully manifested LOAD symptoms at the time of assessment, potentially contributing to a relatively lower heritability estimate (h2 = 37%). In contrast, Pedersen et al. (2004) observed a much later average age of 83.9 years (SD: 6.3) [40] and higher heritability (48%; Table 1) that could be attributed to their longitudinal study design, where twin pairs aged 52–98 were followed for approximately 5 years. The Karlsson study in 2022, leveraging enriched samples from the STR, benefited from a longer follow-up (through 2016) resulting in an average age of 85.28 years (SD: 7.0) and higher heritability (71%) compared with both Meyer and Breitner (1998) and Pederson et al (2004) [41]. These age variations highlight the importance of considering age, and when possible, birth cohort, of the participants in LOAD studies to capture a more comprehensive picture of its heritability.
Several studies also explicitly indicated the inclusion of different sex twin pairs. LOAD disproportionally affects females, with higher lifetime risk and sex-specific risk factors observed in previous studies [52,53]. Moreover, prior research indicates that genetic components may manifest differently in males and females [54]. Bergem et al. (1997) was the first study we identified that included twins of differing sex; however, it was limited by a small proportion of male-female twins (33% among the dizygotic twin pairs). Later studies leveraging HARMONY, as described above, achieved a more balanced composition of like- and unlike-sex twin pairs; the former study did not identify differences in heritability across sexes, while the latter also incorporated additional opposite-sex twin pairs drawn from GENDER, representing 21% of the total sample.
One study stands out for its focus on incident LOAD cases rather than prevalent cases [40]. While most studies examine prevalent cases to estimate heritability, this study incorporating two longitudinal twin studies investigated the relative importance of genetic and environmental impact in disease development across different age of onset groups. The unique study design yielded an overall heritability estimate of 48%, with age-stratified estimates of 59% for onset before age 80 and 40% for onset after age 80. These estimates are generally lower than those obtained from studies of prevalent cases (Table 1), revealing the potential impact of study design on heritability estimates.
Twin study modeling strategies
A straightforward method for heritability estimation is Falconer’s formula [], which relies on the phenotypic correlation between twin pairs [55]. Bergem et al. (1997) applied Falconer’s formula to estimate the heritability of AD, incorporating a series of probability percentages for positive cases, and obtained heritability estimates ranging from 55% to 61% [37]. An advantage of phenotypic correlation is its ease in interpretability. However, this approach relies heavily on the assumption of equal shared environmental variance and lacks the ability to address more complex aspects of genetic analysis, such as model performance assessment, incorporation of extended family data, and investigation of gene-by-environment interactions, potentially leading to oversimplified estimates that do not fully capture the genetic architecture of complex traits like LOAD.
With advances in analytical techniques and software for effective data handling, more sophisticated modeling approaches have been developed to estimate variance components using information from twin studies. Among these methods, structural equation modeling (SEM) is now becoming frequently employed, with model fitting relying on maximum likelihood estimation. SEM offers greater flexibility in determining the contributions of additive genetic (A), shared environmental (C), and unique environmental (E) components, collectively forming the ACE model, while also allowing for the incorporation of covariates into the analysis, providing a more comprehensive understanding of the factors influencing heritability [11,56]. Typically, these studies involve testing both the full ACE model and reduced models, successively dropping either A or C, to identify the best-fitting model along with the corresponding parameter estimations [22,36,38–41]. Model performance is generally assessed using indicators such as the Akaike Information Criterion (AIC) and chi-square test of difference, while some studies also incorporate critical information, such as consistency with observed correlations and concordance estimates, into the determination of the best-fitting model [38,40].
The majority of the heritability estimates presented in LOAD twin studies (Table 1) originated from the AE model that excludes shared environmental variance, assuming that additive genetic components and unique environmental impact are the primary contributors to LOAD risk. Exceptions to this approach include Gatz et al. (1997) who reported that the full ACE model performed best, estimating heritability at 74% and attributing 24% to the shared environmental effects [38]. Meyer and Breitner (1998) found neither the AE nor CE models could be rejected compared to the full model, yielding heritability estimates of 74% in the AE model and 37% in the full ACE model [36], the latter of which is presented in Table 1. Choice of the ACE model in combination with their relatively younger cohort (mean age = 63.1 years) likely contributes to the lower estimate of Meyer and Breitner (1998) as compared to the other studies. Collectively, these studies demonstrated the importance of considering multiple models and interpreting results cautiously. The preference for the AE model in the LOAD twin studies does not necessarily negate the relevance of shared environmental factors. Instead, it indicates their contribution may be minor or more difficult to detect given current methodologies, emphasizing the need for larger studies and sophisticated modeling approaches to fully elucidate the complex interplay of factors contributing to disease risks.
In heritability studies of LOAD, analyzing binary traits (affected vs. unaffected) presents unique challenges compared to quantitative traits [57,58]. Dichotomous outcomes necessitate a specialized approach to estimating variance component partitions, typically via the assumption of a latent, normally distributed liability for the trait. Disease status is then determined by a threshold, often corresponding to the population disease prevalence, with individuals above this threshold considered affected, while those below as unaffected. This liability threshold model is necessary for incorporating binary outcomes into a framework suitable for heritability analysis. Thus, it is important to use the appropriate threshold(s) in the study, not only to address the ascertainment bias embedded in the twin study design but also to accurately reflect the population characteristics for age-dependent diseases, like LOAD.
The complexity of this process is evident in the approaches adopted in different studies. Some studies leveraged previously published epidemiological data to inform the liability threshold. Bergem et al. (1997) utilized the published population prevalence of LOAD, setting the prevalence at 10% in their analysis based on previous studies with similar age distributions [59,60]. In contrast, several other studies [22,36,38–40] relied on the estimated prevalence or incidence rates derived from their study samples. While this method potentially offers specificity to the study population, it makes it difficult to compare heritability estimates across studies.
Given the challenges of using prevalence or incidence, some studies focused on a threshold modeling process for LOAD to better capture the potential right-censoring in the data, along with the accurate representation of study samples. This methodology moves beyond the single threshold model used in earlier studies [37,38] to more sophisticated approaches that address the relationship between age of onset and disease liability, such as the implementation of multiple threshold models. Meyer and Breitner (1998) and Pedersen et al. (2001) incorporated the five-year age groups for monozygotic and dizygotic twin groups, assigning multiple thresholds corresponding to the specific prevalences estimated in these stratified samples [36,39]. Pedersen et al. (2001) further refined this approach by incorporating an additional age group or bin to distinguish censored observations and by testing models with both population-based and sample-specific prevalence estimates [39]. Pedersen et al. (2004) also incorporated differing incidence rates within two age groups (below and above 80 years old) in their heritability analysis. While this approach has the potential to offer a more nuanced understanding of age-related genetic influences on disease onset, no differences were observed in estimates for twins <80 years of age compared with twins ≥80 years of age [40].
One study not only employed biometrical analysis as discussed above but also incorporated a sex-limitation twin model, leveraging data from a reasonable number of different-sex twin pairs [22]. Unlike previous studies that had a single major stratum based on twin similarity (e.g., monozygotic versus dizygotic), Gatz et al. allowed for estimated thresholds to differ both by sex and age of twins. This more complex modeling offered more insights into variance partitioning, enabling the estimation of sex-specific effects and demonstrating the lack of differences in LOAD heritability between men and women.
Twin study meta-analyses.
In addition to the seven twin-based heritability studies detailed here, two meta-analyses based on family data from individual studies exist in the literature. In one meta-analysis, a weighted mean heritability was estimated based on five previously published twin studies [61]. Among the studies meeting their inclusion criteria, four [22,36–38] are detailed in Table 1 whereas the additional included twin study is from the Finnish twin registry [50]. By weighing the number of twin pairs where at least one is affected by the disorder, the analysis yielded a heritability estimate of 75%. Another meta-analysis of twin correlations from seven studies is available in the literature [62] that includes all of the twin studies described in Table 1 except Karlsson et al. 2022, which was published later. Overall, monozygotic twins for dementia in Alzheimer’s disease had a higher correlation (0.86) compared with dizygotic twins (0.50) (MaTCH) [62]. Using ACE models, the meta-analysis yielded a heritability of 63% for all twins and 59% for the same sex twins [62]. Applying a different method, least squares models based on Falconer’s formula, to the same data led to a higher heritability estimate of 71% for the same-sex twins [62].
SNP-based heritability
Compared with study designs requiring related individuals, population-based approaches for heritability estimation have gained popularity in recent years largely due to the widespread availability of genome-wide SNP data from large case-control studies as well as the advancements in statistical methodologies [35,63–66]. Unlike family-based studies, the population-based approaches utilize the realized genetic matrices computed from the genetic data of large cohorts, circumventing the need for specific family designs and thus broadening the scope for heritability estimation [67]. SNP-based heritability refers to the estimated heritability that is attributable to the assayed or imputed SNPs associated with a complex trait or outcome. Multiple methods have been developed to estimate SNP-based heritability. Some require individual-level data while others require only summary statistics [68].
GCTA.
Genome-wide complex trait analysis (GCTA) is one of the most commonly used methods that leverages individual-level data while relying on different algorithms [69,70]. GCTA employs linear mixed models (LMM) with a normality assumption for the residuals and partitions the phenotypic variance into variance components, leveraging the genetic relatedness matrix (GRM) constructed from the genetic data. The heritability can then be estimated through the genome-based restricted maximum likelihood (GREML) procedure. We identified seven LOAD heritability studies that employed this method (Table 2).
The first LOAD GCTA heritability estimates were published in 2013 (Table 2 and Fig 3). The resulting LOAD heritability estimate of 24% was based on a sample of 7,139 participants (3,290 cases, 3,849 controls) that included both elderly screened and population controls [71,72]. Concurrently, Ridge et al. (2013) leveraged the Alzheimer’s Disease Genetics Consortium (ADGC) dataset [73], including 10,922 individuals (5,708 cases, 5,214 controls), and estimated a higher heritability estimate of 33% [74]. The ADGC dataset has since continued to expand, facilitating more comprehensive analyses of LOAD. A 2016 update included data from 30 studies within the ADGC, and the heritability estimate increased to 53%. This study included only 9,699 individuals (3,877 cases, 5,822 controls) [75]. This sample size was smaller in this latter study due to the requirement for non-missing data across 21 known AD genes, resulting in a higher LOAD heritability estimate compared with the 2013 analysis and highlighting the impact of the quality control (QC) process and demand for complete data at known LOAD risk loci. Inclusion of more known LOAD genes increased the heritability, indicating that these genes have substantial impact on overall heritability estimates.
The figure illustrates SNP-based heritability estimates and sample sizes for different studies, ordered by year of publication and author. Bars represent heritability estimates obtained from each study, with different colors indicating the methods used to estimate heritability (GCTA or LDSC). The solid line demonstrates the number of cases in each study, while the dashed line represents the number of controls.
Later and larger two-phase ADGC datasets focusing on age- and sex-specific heritability utilized samples of up to 17,896 participants [76,78]. Unlike the smaller twin study of Pederson et al (2004) that found no difference in LOAD heritability by age, Lo et al (2019) with a much larger sample size of unrelated individuals estimated a higher heritability for participants >80 years of age (n = 5,198; h2 = 24.1%) compared with those aged 60–69 years of age (n = 12,698; h2 = 16.9%). Differences by sex were subtler with a slightly higher LOAD heritability among women (21.5%) compared with men (19.5%) [78]. Overall, variation in study design, case definition, and sample characteristics can greatly impact heritability estimates, necessitating a more cautious interpretation of these findings.
The evolution of LOAD genetic studies has been marked not only by increasing sample sizes but also by a substantial enhancement in marker density. This improvement can be attributed to advancements in both genotyping technologies and imputation techniques, significantly impacting the comprehensiveness of SNP-based heritability estimates. The relatively lower heritability estimation (h2 = 24%) obtained by Lee et al (2013) can be partially attributed to the less dense genotype data, utilizing only 499,757 SNPs [72]. This limitation in marker density likely constrained its ability to fully capture SNP-based heritability, which was also supported by the observed decrease in heritability estimates when more stringent QC processes were applied. As opposed to being solely dependent on the directly genotyped data, subsequent studies have leveraged the power of imputation, hugely increasing the number of genetic variants analyzed. In the same year, Ridge et al. (2013) exemplified this leap forward by employing data imputed against the HapMap phase II (release 22) reference panel [80], resulting in 2,042,116 SNPs after quality control, a substantial increase compared to non-imputed datasets that all had fewer than 500,000 SNPs [74]. The trend towards higher marker density has been further accelerated by advances in imputation reference panels. The adoption of the 1000 Genomes Project reference panel [81], offering more comprehensive coverage, has enabled even denser imputation. Ridge et al. (2016) utilized 8,712,879 SNPs in their analysis, while for more recent studies leveraging the two-phase ADGC data, up to 38 million SNPs were incorporated in the analysis [76,78]. The substantial increase in marker density, from ~500,000–38 million SNPs, represents a significant methodological advancement in LOAD genetic studies albeit with new analytical challenges, including more stringent requirements for QC processes and increased computational demands. While the differences in marker density could explain some of the variations in heritability estimates across studies, they alone do not explain the observation that the less dense dataset of Ridge et al (2016) has a much higher heritability (53%) compared with either Lo et al (19%; 2019) or Wang et al (21%; 2021).
Analyses with fewer samples and SNPs provided higher estimates, but the analyses differed in ways that may inform our understanding of heritability estimates. For example, inclusion of specific covariates and variable assumed prevalences, among other variables such as marker density (Table 2), complicate comparisons across studies and heritability expectations. As noted above, the GCTA approach that uses LMM allows for the inclusion of covariates as fixed effects, enabling more accurate heritability estimates for LOAD. Age and sex, well-established factors affecting LOAD risk, have been consistently included as covariates in most studies, with the notable exception of Lee et al. (2013) [72]. Principal components (PCs) are universally recognized as crucial for adjusting population structure in genetic studies [82]; however, there is considerable variation in both the number of PCs included in the analyses and how the PCs themselves are calculated. While most studies incorporate 10 PCs [72,74–76,78], Nazarian and Kulminski (2019) used only the first five PCs [77], and Baker et al (2023) employed a flexible approach where the number of PCs was cohort specific [79]. PCs can be calculated within individual cohorts, as demonstrated by Lo et al. (2019), or using the entire combined dataset as did Ridge et al [75]. These methodological heterogeneities may also contribute to differences observed across heritability estimates.
Methodological heterogeneity when using large collaborative studies to estimate SNP-based heritability with GCTA is not limited to PCs. In practice, large collaborative studies are made up of different cohorts that have undergone varied ascertainment processes and genotyping strategies. While increased sample size and diversity are valuable, they come at the cost of potential heterogeneity across cohorts. Lo et al. (2019) made a methodological advance by incorporating cohort indicators that correspond to the specific cohort from which individuals were recruited, into their heritability estimation to account for potential heterogeneity across different cohorts [76]. This approach yielded a lower estimate of 19% for LOAD compared to previous studies. Their sensitivity analysis showed an increase to 32% when cohort indicators were removed, revealing a substantial cohort effect. Wang et al. (2021) similarly incorporated cohort indicators in their analysis of 2-phase ADGC data, obtaining an expected comparable heritability estimate of 21% [78]. While GCTA-based approaches have significantly advanced our understanding of LOAD heritability, they underscore the need for cautious interpretation of estimates due to methodological variation, particularly in covariate handling, highlighting the importance of continued refinement of analytical strategies. We also cannot ignore the possibility that cohorts within a single study actually have variation in true heritability due to unmeasured environmental parameters.
LOAD phenotyping approaches across studies used to estimate heritability are highly heterogeneous. The majority of heritability estimates, particularly those utilizing data from ADGC, rely on a combination of clinical diagnosis, histopathologic findings, and in some cases, biomarkers. Clinical diagnosis have inherent misdiagnosis rates [83] that can affect heritability estimates. Baker et al (2023) illustrated this by demonstrating that the histopathologically confirmed LOAD yielded higher heritability estimates (31% - 57%) compared to clinically diagnosed LOAD (12% - 32%) when applying a consistent model with a 5% liability threshold to five independent cohorts [79]. Notably, within the Amsterdam Dementia Cohort, using the amyloid-confirmed cases showed a heritability estimate of 57%, while the clinical diagnosed cases from the same population yielded an estimate of 25%. This marked difference emphasizes the significant impact of diagnostic criteria on heritability estimation, potentially helping explain some of the observed variability in heritability across studies.
SNP-based heritability estimates of LOAD must address APOE, the major genetic risk factor for LOAD [23]. Compared with other complex trait-associated genetic variation, APOE has an outsized effect and alone contributes substantially to LOAD heritability, accounting for an estimated 4% to 13.42% of the total phenotypic variance (Fig 3) [72,74–76,78]. The lowest heritability estimate (4%) was obtained using proxy SNPs for APOE [72], and although not significantly different from the proxy-based estimate, two higher heritability estimates for APOE were obtained when the APOE ε2 and ε4 alleles were directly genotyped [74,75]. In contrast to the SNP partitioning method, some studies employed an approach leveraging the best linear unbiased prediction (BLUP) and involved regressing on the number of APOE ε4 alleles to estimate heritability attributable specifically to the APOE ε4 alleles [76,78]. Regardless of the variability, published estimates confirm that APOE explains a substantial portion of the genetic risk for LOAD.
LDSC.
With the proliferation of large-scale GWAS, and their accompanying summary statistics, methods to estimate heritability such as linkage disequilibrium (LD) score regression (LDSC) have become popular. LDSC only requires summary-level data and involves the process of LD score calculations using the appropriate reference panel for each variant. These calculations can then be incorporated into a regression model where the observed GWAS summary statistics are regressed against the pre-computed LD scores to compute the heritability estimation [84,85]. While convenient and in some cases the only option for difficult-to-access datasets, such methods typically lead to an underestimation of SNP-based heritability compared with those derived from individual-level data [35,86] as evident for LOAD (Tables 2 and 3). We identified nine LOAD heritability studies that employed LDSC (Table 3).
Four major GWAS of LOAD have been pivotal in heritability estimation using LDSC (Table 3 and Fig 3), all involving direct phenotyping of cases [24–26,90]. As discussed earlier, multi-consortia collaborative efforts revolutionized large-scale genetic studies of LOAD but also introduced heterogeneity into both study design and included populations, leading to variability in heritability estimates. Beginning with the LOAD GWAS published by Lambert et al. (2013), its stage 1 summary statistics formed the basis for three subsequent LOAD heritability estimates [87–89]. This foundational study involved a meta-analysis of four consortia (the ADGC, the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium, the European Alzheimer’s Disease Initiative (EADI), and the Genetic and Environmental Risk in Alzheimer’s Disease (GERAD) Consortium) under the International Genomics of Alzheimer’s Project (IGAP), with a total of 54,162 participants in the analysis. Building on these efforts, Kunkle et al. (2019) leveraged the expanded IGAP study to become one of the largest GWAS of LOAD to date [25]. This comprehensive study incorporated 46 cohorts from the IGAP, including 17 new ones, increasing the sample to 63,926 individuals. Two LOAD heritability estimates were based solely on the stage 1 discovery sample [25,92]. Jansen et al. (2019) performed GWAS utilizing not only IGAP data but also incorporating two additional independent consortia: the Psychiatric Genomics Consortium (PGC-ALZ) and the Alzheimer’s Disease Sequencing Project (ADSP), boosting the total sample size to 79,145 [90,91]. Wightman et al. (2021) further expanded the scope of LOAD genetic research by including additional cohorts from Europe and the US not previously considered in Jansen et al. (2019), resulting in a sample size of 761,702 (43,725 cases, 717,979 controls) and its own heritability estimate based on this LOAD GWAS [26]. All of these large GWAS included cases and controls directly phenotyped for LOAD status.
In contrast to the direct LOAD phenotyping approaches described above, Marioni et al. (2018) contributed a study using proxy-phenotypes based on family history (GWAX), leveraging data from large electronic health records available for UK Biobank participants ages 40–69 [94]. Using an innovative approach, de la Fuente et al. (2022) broadened the heritability estimation of LOAD by integrating both GWAS and GWAX while addressing the attenuated heritability estimates when directly combining the summary statistics from the two [93].
Beyond the heterogeneity across cohorts, variations in heritability estimates are observed even within studies using the same LOAD summary statistics. This variability could stem from the unique challenges in estimating heritability for binary traits like LOAD, as compared to quantitative traits. The process of estimating heritability for binary traits involves an initial estimation made on the observed scale, using the sample prevalence as the threshold followed by a transformation to the liability scale utilizing the population prevalence [95,96]. This approach results in a mix of observed- and liability-scale heritability estimates across studies, even when leveraging the same summary statistics, contributing significantly to the diverse heritability values reported in the literature. The impact of this methodological variation on the heritability estimates can be seen in studies leveraging summary statistics [24]. Zheng et al. (2017) and Chen et al. (2021) reported similar observed-scale heritability estimates of 6.88% and 6.80%, respectively [87,89]. In contrast, the Brainstorm Consortium et al. (2018) provided a markedly different liability-scale heritability estimate of 13%, using a population prevalence of 17% for the transformation [88]. Liability scale heritability is critical in adjusting for the ascertainment of the binary traits and making the heritability estimates comparable across studies using appropriate population prevalence for transformation. The challenge of interpreting and comparing these mixed-scale estimates is further compounded by the wide range of population prevalence figures used in liability-scale transformations. Across various studies, these prevalence estimates range from 4.3% to 17% [26,88,90,93]. This variability in prevalence estimates can significantly impact heritability figures, making direct comparisons between studies difficult.
SNP-based heritability re-evaluation.
In addition to the 16 SNP-based studies (Tables 2 and 3) included in this review that met our criteria (Fig 1), we identified several “re-evaluations” of LOAD SNP-based heritability where the heritability estimates were primarily generated in the context of gene discoveries. With variation in the data included, the methodology considered, and assumptions made, these re-evaluations based on previous studies resulted in a range of incomparable estimates. At least three SNP-based heritability re-evaluations studies have been published: one utilizing both GCTA and LDSC [97], one using only LDSC [98], and a third one employing LDAK [99], a method that accounts for minor allele frequency and LD when estimating a SNP’s influence on a trait’s heritability [100,101]. The number of datasets or studies included ranged from four [97] to ten [98] that resulted in a large range of heritability estimates with a consistent population prevalence of 5%: the LOAD heritability estimates using LDSC ranged from 9% to 17% [97] and 3% to 42% [98]. As expected, the GCTA estimates were higher at 25% to 31% [97]. LDAK estimates generally aligned more closely with GCTA, and estimates derived from LDAK were generally higher than LDSC. As an example, heritability estimates using data from Kunkle et al (2019) increased from 7% with LDSC (Table 3) to 21% using LDAK [99].
Conclusion and discussion
In this review we evaluated multiple measures of heritability for LOAD and showed that direct comparisons between heritability studies are difficult due to study population variability, methodological differences, and other factors outlined in this review. Understanding and critically evaluating heritability estimates require careful consideration of underlying assumptions, methodological nuances, and the challenges inherent in quantifying genetic contributions to complex diseases like LOAD. Another significant limitation of current LOAD heritability research is its predominant focus on European descent populations. This lack of diversity, well-documented in genetic studies [102], substantially impacts the generalizability and broader applicability of obtained estimates. However, it is of note that all of the studies we reviewed were of European ancestry and the variation in heritability estimates was still huge. Also of note: the heritability estimates discussed throughout this paper are confined to narrow sense, which, while commonly used, captures only additive genetic effects and further limits the ability to disentangle the influence of gene-gene and gene-environmental interactions, both of which likely contribute to phenotypic variance as demonstrated by prior studies in model organisms but remain challenging to model for human studies [103–108]. These non-additive effects might also very across populations [109,110], yet their extent remains under characterized, highlighting the needs for more search both to better reflect the full spectrum of genetic effects and to obtain population-specific estimates.
Moving forward, the recognition that there is no one real measure of heritability should be promoted, reinforcing that each estimate captures nested components of genetic architecture within the specific methodological and population context from which they were obtained. Clearly, future studies would benefit from increased population diversity, standardized phenotyping approaches, and careful consideration of both genetic and environmental factors to provide a more comprehensive understanding of LOAD heritability across diverse populations. In addition, for late onset complex diseases, such as LOAD, we need to recognize that it may be necessary to build new analytical methods for the estimation of heritability that allow us to address time-to-event (survival) outcomes in a competing risk setting to mitigate the age dependent complexities. If done properly such methods may allow some meta-analyses that estimate heritability to be performed. These types of analyses may be able to leverage already existing data such as the UK Biobank that includes not only phenotypes and genotypes but also a large number of twins.
Despite the limitations in estimating LOAD heritability, doing so is still important as LOAD is a neurodegenerative disorder with significant public health implications, and knowledge of heritability is a gateway to better understanding its genetic architecture and guiding future research efforts [3,66,67,111]. In this systematic review, we identified 23 studies providing LOAD heritability estimates, employing either twin-based (n = 7) or SNP-based (n = 16) approaches. The included studies demonstrated a range of heritability estimates that varied considerably (Fig 4), reflecting the diversity in study designs, estimation methodologies, and the underlying heterogeneity of LOAD. Family-based approaches that leverage the genetic similarities between relatives to estimate heritability, generally yielded higher estimates, ranging from 37% to 79% based on twin studies. The higher estimates for twin studies are expected as these implicitly include all genetic variants and their possible non-additive effects, while also potentially including shared environmental effects that are difficult to dissect. This potentially leads to an overestimation of heritability. In contrast, SNP-based approaches that compute heritability estimates based on measured genetic variants across the genome, provided relatively lower estimates (3.1% to 53%), as they are limited by what has been genotyped or imputed [58]. Among them, studies using summary statistics provided relatively lower estimates (LDSC-based; 3.1% to 13%) compared with those using genotyped or imputed data (GCTA-based; 19% to 53%). As we recognize that LDSC-based heritability estimates tend to be downwardly biased [86,101,112,113], in contrast to twin studies that can capture gene-environment interactions, along with other factors, that potentially inflate heritability estimates [105,114,115]. Acknowledging that heritability of LOAD is variable to some extent [79] and subject to variations introduced by a variety of factors as discussed, we emphasize the need to critically examine heritability estimates in the existing literature and incorporate their potential biases when interpreting and applying them in future research.
Plotted are the heritability estimates for LOAD from all studies identified, published between 1997 and 2022. Within the figure, each data point represents an individual estimate corresponding to the value on the y-axis, with shapes distinguishing between family-based (circle) and SNP-based (triangle) approaches, while colors indicating specific methods: purple for twin studies, gray for GCTA, and orange for LDSC.
References
- 1.
Lynch M, Walsh B. Genetics and analysis of quantitative traits. 1998.
- 2. Fisher RA. XV—The Correlation between Relatives on the Supposition of Mendelian Inheritance. Trans R Soc Edinb. 1919;52(2):399–433.
- 3. Visscher PM, Hill WG, Wray NR. Heritability in the genomics era--concepts and misconceptions. Nat Rev Genet. 2008;9(4):255–66. pmid:18319743
- 4. Robette N, Génin E, Clerget-Darpoux F. Heritability: What’s the point? What is it not for? A human genetics perspective. Genetica. 2022;150(3–4):199–208. pmid:35092541
- 5. Silventoinen K, Kaprio J, Lahelma E, Koskenvuo M. Relative effect of genetic and environmental factors on body height: differences across birth cohorts among Finnish men and women. Am J Public Health. 2000;90(4):627–30. pmid:10754982
- 6. Perola M, Ohman M, Hiekkalinna T, Leppävuori J, Pajukanta P, Wessman M, et al. Quantitative-trait-locus analysis of body-mass index and of stature, by combined analysis of genome scans of five Finnish study groups. Am J Hum Genet. 2001;69(1):117–23. pmid:11410840
- 7. Maher B. Personal genomes: The case of the missing heritability. Nature. 2008;456(7218):18–21. pmid:18987709
- 8. Manolio TA, Collins FS, Cox NJ, Goldstein DB, Hindorff LA, Hunter DJ, et al. Finding the missing heritability of complex diseases. Nature. 2009;461(7265):747–53. pmid:19812666
- 9. Yengo L, Vedantam S, Marouli E, Sidorenko J, Bartell E, Sakaue S, et al. A saturated map of common genetic variants associated with human height. Nature. 2022;610(7933):704–12. pmid:36224396
- 10. Wainschtein P, Jain D, Zheng Z, Cupples LA, et al.; TOPMed Anthropometry Working Group, NHLBI Trans-Omics for Precision Medicine (TOPMed) Consortium. Assessing the contribution of rare variants to complex trait heritability from whole-genome sequence data. Nat Genet. 2022;54(3):263–73. pmid:35256806
- 11. Tenesa A, Haley CS. The heritability of human disease: estimation, uses and abuses. Nat Rev Genet. 2013;14(2):139–49. pmid:23329114
- 12. Srivastava AK, Williams SM, Zhang G. Heritability Estimation Approaches Utilizing Genome-Wide Data. Curr Protoc. 2023;3(4):e734. pmid:37068172
- 13. Friedman NP, Banich MT, Keller MC. Twin studies to GWAS: there and back again. Trends Cogn Sci. 2021;25(10):855–69. pmid:34312064
- 14. Hoffmann A, Merilä J. Heritable variation and evolution under favourable and unfavourable conditions. Trends Ecol Evol. 1999;14(3):96–101. pmid:10322508
- 15. Charmantier A, Garant D. Environmental quality and evolutionary potential: lessons from wild populations. Proc Biol Sci. 2005;272(1571):1415–25. pmid:16011915
- 16. Kruuk LEB, Hadfield JD. How to separate genetic and environmental causes of similarity between relatives. J Evol Biol. 2007;20(5):1890–903. pmid:17714306
- 17. Wilson AJ, Réale D, Clements MN, Morrissey MM, Postma E, Walling CA, et al. An ecologist’s guide to the animal model. J Anim Ecol. 2010;79(1):13–26. pmid:20409158
- 18. Silventoinen K, Sammalisto S, Perola M, Boomsma DI, Cornes BK, Davis C, et al. Heritability of adult body height: A comparative study of twin cohorts in eight countries. Twin Res. 2003;6(05):399–408.
- 19. Yang J, Bakshi A, Zhu Z, Hemani G, Vinkhuyzen AAE, Lee SH, et al. Genetic variance estimation with imputed variants finds negligible missing heritability for human height and body mass index. Nat Genet. 2015;47(10):1114–20. pmid:26323059
- 20. Reitz C, Brayne C, Mayeux R. Epidemiology of Alzheimer disease. Nat Rev Neurol. 2011;7(3):137–52. pmid:21304480
- 21. 2024 Alzheimer’s disease facts and figures. Alzheimers Dement. 2024;20(5):3708–821. pmid:38689398
- 22. Gatz M, Reynolds CA, Fratiglioni L, Johansson B, Mortimer JA, Berg S, et al. Role of genes and environments for explaining Alzheimer disease. Arch Gen Psychiatry. 2006;63(2):168–74. pmid:16461860
- 23. Corder EH, Saunders AM, Strittmatter WJ, Schmechel DE, Gaskell PC, Small GW, et al. Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science. 1993;261(5123):921–3. pmid:8346443
- 24. Lambert JC, Ibrahim-Verbaas CA, Harold D, Naj AC, Sims R, Bellenguez C, et al. Meta-analysis of 74,046 individuals identifies 11 new susceptibility loci for Alzheimer’s disease. Nat Genet. 2013;45(12):1452–8. pmid:24162737
- 25. Kunkle BW, Grenier-Boley B, Sims R, Bis JC, Damotte V, Naj AC, et al. Genetic meta-analysis of diagnosed Alzheimer’s disease identifies new risk loci and implicates Aβ, tau, immunity and lipid processing. Nat Genet. 2019;51(3):414–30. pmid:30820047
- 26. Wightman DP, Jansen IE, Savage JE, Shadrin AA, Bahrami S, Holland D, et al. A genome-wide association study with 1,126,563 individuals identifies new risk loci for Alzheimer’s disease. Nat Genet. 2021;53(9):1276–82. pmid:34493870
- 27. Bellenguez C, Küçükali F, Jansen IE, Kleineidam L, Moreno-Grau S, Amin N, et al. New insights into the genetic etiology of Alzheimer’s disease and related dementias. Nat Genet. 2022;54(4):412–36. pmid:35379992
- 28.
Falconer DS, Mackay FCT. Introduction to Quantitative Genetics. 1996. p. 464.
- 29. Ossom Williamson P, Minter CIJ. Exploring PubMed as a reliable resource for scholarly communications services. J Med Libr Assoc. 2019;107(1):16–29. pmid:30598645
- 30. Plassman BL, Breitner JC. The genetics of dementia in late life. Psychiatr Clin North Am. 1997;20(1):59–76. pmid:9139296
- 31. Wingo TS, Lah JJ, Levey AI, Cutler DJ. Autosomal recessive causes likely in early-onset Alzheimer disease. Arch Neurol. 2012;69(1):59–64. pmid:21911656
- 32. Sabatti C, Service SK, Hartikainen A-L, Pouta A, Ripatti S, Brodsky J, et al. Genome-wide association analysis of metabolic traits in a birth cohort from a founder population. Nat Genet. 2009;41(1):35–46. pmid:19060910
- 33. Speed D, Hemani G, Johnson MR, Balding DJ. Improved heritability estimation from genome-wide SNPs. Am J Hum Genet. 2012;91(6):1011–21. pmid:23217325
- 34. Gusev A, Bhatia G, Zaitlen N, Vilhjalmsson BJ, Diogo D, Stahl EA, et al. Quantifying missing heritability at known GWAS loci. PLoS Genet. 2013;9(12):e1003993. pmid:24385918
- 35. Yang J, Zeng J, Goddard ME, Wray NR, Visscher PM. Concepts, estimation and interpretation of SNP-based heritability. Nat Genet. 2017;49(9):1304–10. pmid:28854176
- 36. Meyer JM, Breitner JC. Multiple threshold model for the onset of Alzheimer’s disease in the NAS-NRC twin panel. Am J Med Genet. 1998;81(1):92–7. pmid:9514594
- 37. Bergem AL, Engedal K, Kringlen E. The role of heredity in late-onset Alzheimer disease and vascular dementia. A twin study. Arch Gen Psychiatry. 1997;54(3):264–70. pmid:9075467
- 38. Gatz M, Pedersen NL, Berg S, Johansson B, Johansson K, Mortimer JA, et al. Heritability for Alzheimer’s disease: the study of dementia in Swedish twins. J Gerontol A Biol Sci Med Sci. 1997;52(2):M117-25. pmid:9060980
- 39. Pedersen NL, Posner SF, Gatz M. Multiple-threshold models for genetic influences on age of onset for Alzheimer disease: findings in Swedish twins. Am J Med Genet. 2001;105(8):724–8. pmid:11803520
- 40. Pedersen NL, Gatz M, Berg S, Johansson B. How heritable is Alzheimer’s disease late in life? Findings from Swedish twins. Ann Neurol. 2004;55(2):180–5. pmid:14755721
- 41. Karlsson IK, Escott-Price V, Gatz M, Hardy J, Pedersen NL, Shoai M, et al. Measuring heritable contributions to Alzheimer’s disease: polygenic risk score analysis with twins. Brain Commun. 2022;4(1):fcab308. pmid:35169705
- 42. Cederlöf R, Lorich U. The Swedish Twin Registry. Prog Clin Biol Res. 1978;24(Pt B):189–95. pmid:569308
- 43. Kringlen E. Norwegian twin registers. Prog Clin Biol Res. 1978;24(Pt B):185–7. pmid:569307
- 44. Breitner JC, Welsh KA, Gau BA, McDonald WM, Steffens DC, Saunders AM, et al. Alzheimer’s disease in the National Academy of Sciences-National Research Council Registry of Aging Twin Veterans. III. Detection of cases, longitudinal results, and observations on twin concordance. Arch Neurol. 1995;52(8):763–71. pmid:7639628
- 45. Page WF. The NAS-NRC Twin Registry of WWII military veteran twins. National Academy of Sciences-National Research Council. Twin Res. 2002;5(5):493–6. pmid:12537883
- 46. Pedersen NL, McClearn GE, Plomin R, Nesselroade JR, Berg S, DeFaire U. The Swedish Adoption Twin Study of Aging: an update. Acta Genet Med Gemellol (Roma). 1991;40(1):7–20. pmid:1950353
- 47. McClearn GE, Johansson B, Berg S, Pedersen NL, Ahern F, Petrill SA, et al. Substantial genetic influence on cognitive abilities in twins 80 or more years old. Science. 1997;276(5318):1560–3. pmid:9171059
- 48. Gatz M, Fratiglioni L, Johansson B, Berg S, Mortimer JA, Reynolds CA, et al. Complete ascertainment of dementia in the Swedish Twin Registry: the HARMONY study. Neurobiol Aging. 2005;26(4):439–47. pmid:15653172
- 49. Gold CH, Malmberg B, McClearn GE, Pedersen NL, Berg S. Gender and health: a study of older unlike-sex twins. J Gerontol B Psychol Sci Soc Sci. 2002;57(3):S168-76. pmid:11983743
- 50. Raiha I, Kaprio J, Koskenvuo M, Rajala T, Sourander L. Alzheimer’s disease in Finnish twins. Lancet. 1996;347(9001):573–8. pmid:8596319
- 51. Rajan KB, Weuve J, Barnes LL, McAninch EA, Wilson RS, Evans DA. Population estimate of people with clinical Alzheimer’s disease and mild cognitive impairment in the United States (2020-2060). Alzheimers Dement. 2021;17(12):1966–75. pmid:34043283
- 52. Plassman BL, Langa KM, Fisher GG, Heeringa SG, Weir DR, Ofstedal MB, et al. Prevalence of dementia in the United States: the aging, demographics, and memory study. Neuroepidemiology. 2007;29(1–2):125–32. pmid:17975326
- 53. Mielke MM. Sex and Gender Differences in Alzheimer’s Disease Dementia. Psychiatr Times. 2018;35(11):14–7. pmid:30820070
- 54. Farrer LA, Cupples LA, Haines JL, Hyman B, Kukull WA, Mayeux R, et al. Effects of age, sex, and ethnicity on the association between apolipoprotein E genotype and Alzheimer disease. A meta-analysis. APOE and Alzheimer Disease Meta Analysis Consortium. JAMA. 1997;278(16):1349–56. pmid:9343467
- 55.
Falconer DS, MacKay TFC. Introduction to Quantitative Genetics. 4th ed. London, England: Longman; 1995. p. 480.
- 56. Rijsdijk FV, Sham PC. Analytic approaches to twin data using structural equation models. Brief Bioinform. 2002;3(2):119–33. pmid:12139432
- 57. Falconer DS. The inheritance of liability to certain diseases, estimated from the incidence among relatives. Ann Human Genet. 1965;29(1):51–76.
- 58. Witte JS, Visscher PM, Wray NR. The contribution of genetic variants to disease depends on the ruler. Nat Rev Genet. 2014;15(11):765–76. pmid:25223781
- 59. Rocca WA, Hofman A, Brayne C, Breteler MM, Clarke M, Copeland JR, et al. Frequency and distribution of Alzheimer’s disease in Europe: a collaborative study of 1980-1990 prevalence findings. The EURODEM-Prevalence Research Group. Ann Neurol. 1991;30(3):381–90. pmid:1952826
- 60. Engedal K, Haugen PK. The prevalence of dementia in a sample of elderly Norwegians. Int J Geriat Psychiatry. 1993;8(7):565–70.
- 61. Bienvenu OJ, Davydow DS, Kendler KS. Psychiatric “diseases” versus behavioral disorders and degree of genetic influence. Psychol Med. 2011;41(1):33–40. pmid:20459884
- 62. Polderman TJC, Benyamin B, de Leeuw CA, Sullivan PF, van Bochoven A, Visscher PM, et al. Meta-analysis of the heritability of human traits based on fifty years of twin studies. Nat Genet. 2015;47(7):702–9. pmid:25985137
- 63. Lee SH, Goddard ME, Visscher PM, van der Werf JH. Using the realized relationship matrix to disentangle confounding factors for the estimation of genetic variance components of complex traits. Genet Sel Evol. 2010;42(1):22. pmid:20546624
- 64. Browning SR, Browning BL. Identity-by-descent-based heritability analysis in the Northern Finland Birth Cohort. Hum Genet. 2013;132(2):129–38. pmid:23052944
- 65. Yang J, Zaitlen NA, Goddard ME, Visscher PM, Price AL. Advantages and pitfalls in the application of mixed-model association methods. Nat Genet. 2014;46(2):100–6. pmid:24473328
- 66. Zhu H, Zhou X. Statistical methods for SNP heritability estimation and partition: A review. Comput Struct Biotechnol J. 2020;18:1557–68. pmid:32637052
- 67. Mayhew AJ, Meyre D. Assessing the Heritability of Complex Traits in Humans: Methodological Challenges and Opportunities. Curr Genomics. 2017;18(4):332–40. pmid:29081689
- 68. Barry C-JS, Walker VM, Cheesman R, Davey Smith G, Morris TT, Davies NM. How to estimate heritability: a guide for genetic epidemiologists. Int J Epidemiol. 2023;52(2):624–32. pmid:36427280
- 69. Yang J, Benyamin B, McEvoy BP, Gordon S, Henders AK, Nyholt DR, et al. Common SNPs explain a large proportion of the heritability for human height. Nat Genet. 2010;42(7):565–9. pmid:20562875
- 70. Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome-wide complex trait analysis. Am J Hum Genet. 2011;88(1):76–82. pmid:21167468
- 71. Harold D, Abraham R, Hollingworth P, Sims R, Gerrish A, Hamshere ML, et al. Genome-wide association study identifies variants at CLU and PICALM associated with Alzheimer’s disease. Nat Genet. 2009;41(10):1088–93. pmid:19734902
- 72. Lee SH, Harold D, Nyholt DR, ANZGene Consortium, International Endogene Consortium, Genetic and Environmental Risk for Alzheimer’s disease Consortium, et al. Estimation and partitioning of polygenic variation captured by common SNPs for Alzheimer’s disease, multiple sclerosis and endometriosis. Hum Mol Genet. 2013;22(4):832–41. pmid:23193196
- 73. Naj AC, Jun G, Beecham GW, Wang L-S, Vardarajan BN, Buros J, et al. Common variants at MS4A4/MS4A6E, CD2AP, CD33 and EPHA1 are associated with late-onset Alzheimer’s disease. Nat Genet. 2011;43(5):436–41. pmid:21460841
- 74. Ridge PG, Mukherjee S, Crane PK, Kauwe JSK, Alzheimer’s Disease Genetics Consortium. Alzheimer’s disease: analyzing the missing heritability. PLoS One. 2013;8(11):e79771. pmid:24244562
- 75. Ridge PG, Hoyt KB, Boehme K, Mukherjee S, Crane PK, Haines JL, et al. Assessment of the genetic variance of late-onset Alzheimer’s disease. Neurobiol Aging. 2016;41:200.e13-200.e20. pmid:27036079
- 76. Lo M-T, Kauppi K, Fan C-C, Sanyal N, Reas ET, Sundar VS, et al. Identification of genetic heterogeneity of Alzheimer’s disease across age. Neurobiol Aging. 2019;84:243.e1-243.e9. pmid:30979435
- 77. Nazarian A, Kulminski AM. Evaluation of the Genetic Variance of Alzheimer’s Disease Explained by the Disease-Associated Chromosomal Regions. J Alzheimers Dis. 2019;70(3):907–15. pmid:31282417
- 78. Wang H, Lo M-T, Rosenthal SB, Makowski C, Andreassen OA, Salem RM, et al. Similar Genetic Architecture of Alzheimer’s Disease and Differential APOE Effect Between Sexes. Front Aging Neurosci. 2021;13:674318. pmid:34122051
- 79. Baker E, Leonenko G, Schmidt KM, Hill M, Myers AJ, Shoai M, et al. What does heritability of Alzheimer’s disease represent? PLoS One. 2023;18(4):e0281440. pmid:37115753
- 80. International HapMap Consortium. The International HapMap Project. Nature. 2003;426(6968):789–96. pmid:14685227
- 81. 1000 Genomes Project Consortium, Abecasis GR, Auton A, Brooks LD, DePristo MA, Durbin RM, et al. An integrated map of genetic variation from 1,092 human genomes. Nature. 2012;491(7422):56–65. pmid:23128226
- 82. Abegaz F, Chaichoompu K, Génin E, Fardo DW, König IR, Mahachie John JM, et al. Principals about principal components in statistical genetics. Brief Bioinform. 2019;20(6):2200–16. pmid:30219892
- 83. Beach TG, Monsell SE, Phillips LE, Kukull W. Accuracy of the clinical diagnosis of Alzheimer disease at National Institute on Aging Alzheimer Disease Centers, 2005-2010. J Neuropathol Exp Neurol. 2012;71(4):266–73. pmid:22437338
- 84. Bulik-Sullivan B, Finucane HK, Anttila V, Gusev A, Day FR, Loh P-R, et al. An atlas of genetic correlations across human diseases and traits. Nat Genet. 2015;47(11):1236–41. pmid:26414676
- 85. Bulik-Sullivan BK, Loh P-R, Finucane HK, Ripke S, Yang J, Schizophrenia Working Group of the Psychiatric Genomics Consortium, et al. LD Score regression distinguishes confounding from polygenicity in genome-wide association studies. Nat Genet. 2015;47(3):291–5. pmid:25642630
- 86. Evans LM, Tahmasbi R, Vrieze SI, Abecasis GR, Das S, Gazal S, et al. Comparison of methods that use whole genome data to estimate the heritability and genetic architecture of complex traits. Nat Genet. 2018;50(5):737–45. pmid:29700474
- 87. Zheng J, Erzurumluoglu AM, Elsworth BL, Kemp JP, Howe L, Haycock PC, et al. LD Hub: a centralized database and web interface to perform LD score regression that maximizes the potential of summary level GWAS data for SNP heritability and genetic correlation analysis. Bioinformatics. 2017;33(2):272–9. pmid:27663502
- 88. Brainstorm Consortium, Anttila V, Bulik-Sullivan B, Finucane HK, Walters RK, Bras J, et al. Analysis of shared heritability in common disorders of the brain. Science. 2018;360(6395):eaap8757. pmid:29930110
- 89. Chen H-H, Petty LE, Sha J, Zhao Y, Kuzma A, Valladares O, et al. Genetically regulated expression in late-onset Alzheimer’s disease implicates risk genes within known and novel loci. Transl Psychiatry. 2021;11(1):618. pmid:34873149
- 90. Jansen IE, Savage JE, Watanabe K, Bryois J, Williams DM, Steinberg S, et al. Genome-wide meta-analysis identifies new loci and functional pathways influencing Alzheimer’s disease risk. Nat Genet. 2019;51(3):404–13. pmid:30617256
- 91. Monereo-Sánchez J, Schram MT, Frei O, O’Connell K, Shadrin AA, Smeland OB, et al. Genetic Overlap Between Alzheimer’s Disease and Depression Mapped Onto the Brain. Front Neurosci. 2021;15:653130. pmid:34290577
- 92. Wu P, Du B, Wang B, Yin R, Lv X, Dai Y, et al. Joint Analysis of Genome-Wide Association Data Reveals No Genetic Correlations Between Low Back Pain and Neurodegenerative Diseases. Front Genet. 2021;12:744299. pmid:34630533
- 93. de la Fuente J, Grotzinger AD, Marioni RE, Nivard MG, Tucker-Drob EM. Integrated analysis of direct and proxy genome wide association studies highlights polygenicity of Alzheimer’s disease outside of the APOE region. PLoS Genet. 2022;18(6):e1010208. pmid:35658006
- 94. Marioni RE, Harris SE, Zhang Q, McRae AF, Hagenaars SP, Hill WD, et al. GWAS on family history of Alzheimer’s disease. Transl Psychiatry. 2018;8(1):99. pmid:29777097
- 95. Robertson A, Lerner IM. The Heritability of All-or-None Traits: Viability of Poultry. Genetics. 1949;34(4):395–411. pmid:17247323
- 96. Lee SH, Wray NR, Goddard ME, Visscher PM. Estimating missing heritability for disease from genome-wide association studies. Am J Hum Genet. 2011;88(3):294–305. pmid:21376301
- 97. Zhang Q, Sidorenko J, Couvy-Duchesne B, Marioni RE, Wright MJ, Goate AM, et al. Risk prediction of late-onset Alzheimer’s disease implies an oligogenic architecture. Nat Commun. 2020;11(1):4799. pmid:32968074
- 98. Escott-Price V, Hardy J. Genome-wide association studies for Alzheimer’s disease: bigger is not always better. Brain Commun. 2022;4(3):fcac125. pmid:35663382
- 99. Lai D, Zhang M, Li R, Zhang C, Zhang P, Liu Y, et al. Identifying Genes Associated with Alzheimer’s Disease Using Gene-Based Polygenic Risk Score. J Alzheimers Dis. 2023;96(4):1639–49. pmid:38007651
- 100. Speed D, Cai N, Johnson MR, Nejentsev S, Balding DJ; UCLEB Consortium. Reevaluation of SNP heritability in complex human traits. Nat Genet. 2017;49(7):986–92. pmid:28530675
- 101. Speed D, Balding DJ. SumHer better estimates the SNP heritability of complex traits from summary statistics. Nat Genet. 2019;51(2):277–84. pmid:30510236
- 102. Sirugo G, Williams SM, Tishkoff SA. The Missing Diversity in Human Genetic Studies. Cell. 2019;177(4):1080. pmid:31051100
- 103. Carlborg O, Haley CS. Epistasis: too often neglected in complex trait studies? Nat Rev Genet. 2004;5(8):618–25. pmid:15266344
- 104. Su G, Christensen OF, Ostersen T, Henryon M, Lund MS. Estimating additive and non-additive genetic variances and predicting genetic merits using genome-wide dense single nucleotide polymorphism markers. PLoS One. 2012;7(9):e45293. pmid:23028912
- 105. Zuk O, Hechter E, Sunyaev SR, Lander ES. The mystery of missing heritability: Genetic interactions create phantom heritability. Proc Natl Acad Sci U S A. 2012;109(4):1193–8. pmid:22223662
- 106. Bloom JS, Kotenko I, Sadhu MJ, Treusch S, Albert FW, Kruglyak L. Genetic interactions contribute less than additive effects to quantitative trait variation in yeast. Nat Commun. 2015;6:8712. pmid:26537231
- 107. Miller AK, Pan C, Bartlett J, Lusis AJ, Crawford DC, Williams SM, et al. Concordance between male- and female-specific GWAS results helps define underlying genetic architecture of complex traits. Genetics. 2022.
- 108. Matsui T, Mullis MN, Roy KR, Hale JJ, Schell R, Levy SF, et al. The interplay of additivity, dominance, and epistasis on fitness in a diploid yeast cross. Nat Commun. 2022;13(1):1463. pmid:35304450
- 109. Mackay TFC. Epistasis and quantitative traits: using model organisms to study gene-gene interactions. Nat Rev Genet. 2014;15(1):22–33. pmid:24296533
- 110. Tsouris A, Brach G, Schacherer J, Hou J. Non-additive genetic components contribute significantly to population-wide gene expression variation. Cell Genom. 2024;4(1):100459. pmid:38190102
- 111. Eichler EE, Flint J, Gibson G, Kong A, Leal SM, Moore JH, et al. Missing heritability and strategies for finding the underlying causes of complex disease. Nat Rev Genet. 2010;11(6):446–50. pmid:20479774
- 112. Yang J, Bakshi A, Zhu Z, Hemani G, Vinkhuyzen AAE, Nolte IM, et al. Genome-wide genetic homogeneity between sexes and populations for human height and body mass index. Hum Mol Genet. 2015;24(25):7445–9. pmid:26494901
- 113. Luo Y, Li X, Wang X, Gazal S, Mercader JM, 23 and Me Research Team, et al. Estimating heritability and its enrichment in tissue-specific gene sets in admixed populations. Hum Mol Genet. 2021;30(16):1521–34. pmid:33987664
- 114. Purcell S. Variance components models for gene–environment interaction in twin analysis. Twin Res. 2002;5(06):554–71.
- 115. Felson J. What can we learn from twin studies? A comprehensive evaluation of the equal environments assumption. Soc Sci Res. 2014;43:184–99. pmid:24267761
- 116. Page MJ, McKenzie JE, Bossuyt PM, Boutron I, Hoffmann TC, Mulrow CD, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. Syst Rev. 2021;10(1):89. pmid:33781348