IGEMS Studies

From an original consortium of eight twin studies (Pedersen et al., Reference Pedersen, Christensen, Dahl, Finkel, Franz, Gatz, Horwitz, Johansson, Johnson, Kremen, Lyons, Malmberg, McGue, Neiderhiser, Petersen and Reynolds2013), IGEMS has expanded to include 21 studies from five countries, representing many of the most significant available longitudinal twin studies of adulthood and aging in the world. The total sample size is now 145,011, including members of 23,811 monozygotic (MZ) pairs and 41,063 dizygotic (DZ) pairs. The summary below outlines the sampling principles for each study. Numbers of pairs and age ranges at intake are provided in Table 1, as well as the number of waves and length of follow-up, where appropriate. Total Ns refer to individuals and include members of incomplete pairs. Intake age indicates the age at which twins were first recruited into a project. For several studies, additional data are available from the originating registry prior to recruitment, or from registry linkages established before the study began. Instances of this are indicated in the descriptions below. Note that updates for several of the individual studies in IGEMS are separately included in this issue.

Table 1.

Description of IGEMS study cohorts

Note: MZ, monozygotic, DZ, dizygotic, OSDZ, opposite sex dizygotic pairs. Total N refers to individuals from both complete and incomplete pairs. Some individuals may have participated in more than one study; e.g., in A50 and OATS. The totals in the bottom row count each pair or individual once.

IGEMS Measures

Measures used in IGEMS analyses include aging-relevant outcomes in three broad domains: physical health and functional ability (e.g., self-reported diseases, subjective health, body mass index, grip strength, motor function, activities of daily living), psychological wellbeing (e.g., depressive symptoms, anxiety symptoms, subjective wellbeing, loneliness), and cognitive health (i.e., scores on cognitive tests; dementia). Predictors and covariates include health behaviors (e.g., smoking, alcohol, physical activity, cognitively engaging leisure activity), social resources, and indicators of SES. Table 2 presents a list of some of the primary phenotypes assessed and the number of IGEMS studies that include each variable.

Table 2.

Number of IGEMS studies with key variables

Because participating studies differed in how similar constructs were assessed, IGEMS places emphasis on harmonizing relevant phenotypes and outcomes. Creating scores that are common across studies enables pooling data across samples, in order to increase statistical power. Score harmonization requires overlapping item content across studies as well as across time for longitudinal hypotheses. For some measures, it has been straightforward to create a common metric; for example, BMI, lung function, and blood pressure. For harmonizing education and occupation, we have recoded all studies to the International Standard Classification of Education (ISCED; UNESCO Institute for Statistics, 2012), and the International Standard Classification of Occupations (ISCO; Ganzeboom et al., Reference Ganzeboom, De Graaf and Treiman1992) as an international standard. Where a common metric was not already available, overlapping item content and response formats were identified and item response theory (IRT) or factor-analytic techniques were implemented to create harmonized scores across studies. When no common items were available across studies, IGEMS collected separate samples administered with the different scales used to measure a given construct, and used those results to establish ‘crosswalks’ between the scales (Gatz, Reynolds et al., Reference Gatz, Reynolds, Finkel, Hahn, Zhou and Zavala2015).

Fifteen Years of IGEMS

IGEMS has greatly expanded over the past 15 years (see Figure 1), and we continue to onboard new studies and grow research collaborations. Phase 1 of IGEMS, described by N. L. Pedersen et al. (Reference Pedersen, Christensen, Dahl, Finkel, Franz, Gatz, Horwitz, Johansson, Johnson, Kremen, Lyons, Malmberg, McGue, Neiderhiser, Petersen and Reynolds2013) and proposed in response to an RFA and funded by R01 AG037985 (Pedersen), focused on establishing a collaboration among existing twin studies of aging to provide the rich phenotyping, genotyping, and high power necessary to investigate models of gene–environment interplay of social contexts and aging-related outcomes (N. L. Pedersen et al., Reference Pedersen, Christensen, Dahl, Finkel, Franz, Gatz, Horwitz, Johansson, Johnson, Kremen, Lyons, Malmberg, McGue, Neiderhiser, Petersen and Reynolds2013). IGEMS began with five twin cohorts and quickly grew to nine during this initial phase. Data harmonization protocols were tested and established (Gatz, Reynolds et al., Reference Gatz, Reynolds, Finkel, Hahn, Zhou and Zavala2015), and procedures for requesting and sharing data were instituted. Foundational analyses of harmonized variables were completed (Franz et al., Reference Franz, Finkel, Chase and Panizzon2013; Mosing et al., Reference Mosing, Cnattingius, Gatz, Neiderhiser and Pedersen2016; Pahlen et al., Reference Pahlen, Hamdi, Dahl Aslan, Horwitz, Panizzon, Petersen, Zavala, Christensen, Finkel, Franz, Gatz, Johnson, Kremen, Krueger, Neiderhiser, Reynolds, Pedersen and McGue2018) and multiple methods for leveraging twin data and genotyping to test models of GE-interplay were implemented (Finkel et al., Reference Finkel, Franz, Horwitz, Christensen, Gatz, Johnson, Kaprio, Korhonen, Neiderhiser, Petersen, Rose and Silventoinen2016; Petersen et al., Reference Petersen, Pedersen, Rantanen, Kremen, Franz, Johnson, Panizzon, Christiansen, McGue, Christensen, Krueger, Hamdi and Reynolds2016; Petkus et al., Reference Petkus, Beam, Johnson, Kaprio, Korhonen, McGue, Neiderhiser, Pedersen, Reynolds and Gatz2017; Reynolds et al., Reference Reynolds, Gatz, Christensen, Kaprio, Korhonen, Kremen, Krueger, McGue and Pedersen2016). For example, Reynolds and colleagues (Reference Reynolds, Gatz, Christensen, Kaprio, Korhonen, Kremen, Krueger, McGue and Pedersen2016) not only used a within-MZ-pair approach to investigate G × E interaction for body mass index, depressive symptoms, and cognitive measures, but they also incorporated the Alzheimer’s genetic marker APOE in the models as a potential source of the ‘G’ in G × E. They found that G × E interaction effects were evident in all three countries represented in the dataset. Moreover, they found evidence that APOE may represent a variability gene for depressive symptoms and spatial reasoning, because it was associated not just with trait mean but also with trait variability. In addition to launching the IGEMS consortium, Phase 1 resulted in 36 peer-reviewed publications.

Figure 1.

Phases of the IGEMS Consortium.

Phase 2 of IGEMS focused on growth of the consortium and expansion of the scientific aims as described by Finch et al. (Reference Finch, Pedersen and Gatz2019) and N. L. Pedersen et al. (Reference Pedersen, Larsen, Nygaard, Mengel-From, McGue, Dalgård, Hvidberg, Hjelmborg, Skytthe, Holm and Christensen2019). Twelve additional twin cohorts of aging joined IGEMS, adding one country (Australia) and increasing the total IGEMS sample by more than 700%. Although funding and scientific aims were divided into two projects, centralized data management and harmonization efforts, biweekly zoom meetings of the full IGEMS team, and annual in-person meetings continued to nurture the collaboration. In addition, we began using the powerful combination of twin methodology with molecular genetics in the form of polygenic scores. One project, led by Pedersen, Finch, and Gatz (R01 AG059329), focused on the frequently cited health ‘gradient’ for SES — the continuous, monotonic association between SES and health across the full spectrum of SES — which cannot be explained solely by poorer health among the most disadvantaged (Adler et al., Reference Adler, Boyce, Chesney, Cohen, Folkman, Kahn and Syme1994). IGEMS is unique in integrating individual- and country-level contributors to health gradients in twin studies to understand GE-interplay. Analysis results indicate moderation of genetic and environmental influences on health, cognition, frailty, and mortality by subjective and objective measures of SES (Ericsson et al., Reference Ericsson, Lundholm, Fors, Dahl Aslan, Zavala, Reynolds and Pedersen2017; Finkel et al., Reference Finkel, Zavala, Franz, Pahlen, Gatz, Pedersen, Finch, Dahl Aslan, Catts and Ericsson2022; Zavala et al., Reference Zavala, Beam, Finch, Gatz, Johnson, Kremen, Neiderhiser, Pedersen and Reynolds2018). Sleep, loneliness, and smoking have been investigated as mechanisms linking the SES-health gradient to physical and cognitive health (Pahlen et al., Reference Pahlen, Franz, Kremen, Aslan, Lee, Sachdev, Catts and Reynolds2021; Phillips et al., Reference Phillips, Finkel, Petkus, Muñoz, Pahlen, Johnson, Reynolds and Pedersen2023; Vo et al., Reference Vo, Pahlen, Kremen, McGue, Dahl Aslan, Nygaard, Christensen and Reynolds2022), while country-level processes have also been explored (Finch, Reference Finch2021; Finch et al., Reference Finch, Thomas, Gatz, Pedersen, Ericsson, Mosing and Finkel2022; Gatz et al., Reference Gatz, Finch, Beam and Thomas2018). The second Phase 2 project, led by Gatz and Pedersen (R01 AG060470), examined midlife and later-life risk and protective factors for AD/ADRD. In addition to harmonizing memory and verbal fluency measures that can indicate early cognitive decline (Gustavson et al., Reference Gustavson, Panizzon, Kremen, Reynolds, Pahlen, Nygaard, Wod, Catts, Lee and Gatz2021; Luczak et al., Reference Luczak, Beam, Pahlen, Lynch, Pilgrim, Reynolds, Panizzon, Catts, Christensen and Finkel2023), the team derived and validated a latent dementia index using cognitive tests and measures of instrumental activities of daily living (Beam et al., Reference Beam, Luczak, Panizzon, Reynolds, Christensen, Dahl Aslan, Elman, Franz, Kremen and Lee2022). Having this index allows inclusion of studies without clinical dementia diagnoses in analyses of risk and protective factors. Researchers investigated mechanisms of the relationship between BMI and AD/ADRD (Karlsson et al., Reference Karlsson, Lehto, Gatz, Reynolds and Dahl Aslan2020; Karlsson et al., Reference Karlsson, Gatz, Arpawong, Dahl Aslan and Reynolds2021; Karlsson et al., Reference Karlsson, Escott-Price, Gatz, Hardy, Pedersen, Shoai and Reynolds2022). Investigations of sex differences in AD/ADRD mechanisms (Beam et al., Reference Beam, Kaneshiro, Jang, Reynolds, Pedersen and Gatz2020; Karlsson et al., Reference Karlsson, Gatz, Arpawong, Dahl Aslan and Reynolds2021; Luo et al., Reference Luo, Beam, Karlsson, Pike, Reynolds and Gatz2020) and the role of polygenic scores and DNA methylation (Karlsson et al., Reference Karlsson, Escott-Price, Gatz, Hardy, Pedersen, Shoai and Reynolds2022; Karlsson et al., Reference Karlsson, Ploner, Wang, Gatz, Pedersen and Hägg2023; Reynolds et al., Reference Reynolds, Tan, Munoz, Jylhävä, Hjelmborg, Christiansen, Hägg and Pedersen2020) were also completed. A third project, led by Panizzon (R21 AG074212) focused on the physiological mechanisms of sex differences in ADRD. Identification of a U-shaped association between age at natural menopause and dementia risk challenged simplistic interpretations of the role of estrogen (Saelzler et al., Reference Saelzler, Sundermann, Foret, Gatz, Karlsson and Panizzon2024). To date, Phase 2 has produced over 100 peer-reviewed publications.

IGEMS infrastructure includes a data management team that conducts and documents data harmonization and maintains documentation of studies that are members of IGEMS, as well as a project manager to coordinate IGEMS researchers and projects. The primary goals of Phase 3 are to scale up the infrastructure of the consortium to support expansion of (a) the database by increasing the number of study cohorts, the number of harmonized variables, and the diversity of the IGEMS sample, and (b) the research team by inviting and supporting more researchers to join the IGEMS consortium. These developments will allow IGEMS to substantially extend the scientific focus beyond ADRD and to explore the effect of lifecourse exposures on health and functioning in multiple domains (Wagner et al., Reference Wagner, Carmeli, Jackisch, Kivimäki, van der Linden, Cullati and Chiolero2024), with particular interest in international comparisons of health disparities and differences in contexts that may drive cross-national variations in health outcomes (National Institutes of Health [NIH], 2024).

Interest in the effects of life-course exposures on the experience of aging has been growing over the past 20 years (Arcaya et al., Reference Arcaya, Tucker-Seeley, Kim, Schnake-Mahl, So and Subramanian2016; Fratiglioni et al., Reference Fratiglioni, Paillard-Borg and Winblad2004; Wagner et al., Reference Wagner, Carmeli, Jackisch, Kivimäki, van der Linden, Cullati and Chiolero2024), motivated by recognition that the genome alone cannot explain all of the variance in aging outcomes (Argentieri et al., Reference Argentieri, Amin, Nevado-Holgado, Sproviero, Collister, Keestra, Doherty, Hunter, Alvergne and van Duijn2023; Wild, Reference Wild2005; Zhang et al., Reference Zhang, Carlsten, Chaleckis, Hanhineva, Huang, Isobe, Koistinen, Meister, Papazian and Sdougkou2021). Studies of risk and protective factors stemming from the environment have made substantial contributions to our understanding of aging processes (Darin-Mattsson et al., Reference Darin-Mattsson, Fors and Kåreholt2017; Wolfova et al., Reference Wolfova, Csajbok, Kagstrom, Kåreholt and Cermakova2021). One of the greatest challenges in epidemiology is drawing causal inferences from observational data. Although twin designs do not solve this problem, they strengthen the inferences that can be made by allowing the partition of risk into genetic, shared environment, and non-shared environment components (Duncan et al., Reference Duncan, Mills, Strachan, Hurvitz, Huang, Moudon and Turkheimer2014; McGue et al., Reference McGue, Osler and Christensen2010). Twin designs can also identify relationships driven by familial confounding, providing strong tests of causality.

Moreover, exposures will clearly differ across age, cohorts, and countries. Use of twin samples from multiple countries allows IGEMS to explore the impact of exposures and inequalities using different birth cohorts of twins compared at different ages within and between countries. This approach is important for several reasons, including: (a) heritability varies by social context (Uchiyama et al., Reference Uchiyama, Spicer and Muthukrishna2022), and country-level differences in heritability are becoming more apparent (Branigan et al., Reference Branigan, McCallum and Freese2013; Cesarini & Visscher, Reference Cesarini and Visscher2016; Min et al., Reference Min, Chiu and Wang2013); (b) the exposure-outcomes relationships (e.g., education-ADRD) vary by context, making country and birth cohort differences suitable for comparing contexts (Beckfield et al., Reference Beckfield, Olafsdottir and Bakhtiari2013; Chmielewski & Reardon, Reference Chmielewski and Reardon2016; Mackenbach et al., Reference Mackenbach, Stirbu, Roskam, Schaap, Menvielle, Leinsalu and Kunst2008; Olafsdottir, Reference Olafsdottir2007; Präg et al., Reference Präg, Mills and Wittek2016); (c) as implied by cross-national variation in heritability, G × E relationships may vary by country (Colodro-Conde et al., Reference Colodro-Conde, Rijsdijk, Tornero-Gomez, Sanchez-Romera and Ordonana2015) — for example, evidence suggests that gene by SES effects on intelligence vary widely by country with more egalitarian countries showing a reversal or zeroing out of these effects (Tucker-Drob & Bates, Reference Tucker-Drob and Bates2016); and (d) if exposures are based more on social class than on cognitive functioning or principles of equal access, then the exposure-outcome link may appear weakened (Sharp & Gatz, Reference Sharp and Gatz2011), suggesting that the underlying cause may be genetically influenced functioning, not exposure per se.

All IGEMS cohorts include country and birth cohort identifiers, enabling the linking of macro-level economic data to examine direct effects as well as gene–environment and environment–environment interactions. For example, IGEMS has merged data from the Global Database on Intergenerational Mobility (GDIM, 2023), which provides country- and cohort-specific correlations in educational attainment between parents and offspring, and between mothers and daughters, by year and country. In addition, IGEMS has incorporated data from the World Inequality Database (van der Weide et al., Reference van der Weide, Lakner, Mahler, Narayan and Gupta2024) —a publicly available resource that documents global income, education, and wealth disparities, including top income and wealth shares, measures of educational inequality, GINI coefficients, percentile distributions of income, and long-term trends in inequality both within and across countries. IGEMS researchers can specify the most relevant year of data based on hypothesized mechanisms; for example, income inequality at age 20 may be most pertinent to job market outcomes or income, whereas educational inequality at age 10 may be most relevant for educational attainment.

Recently funded IGEMS projects support the infrastructure and substantive goals of Phase 3. The aims of R01AG081248 (Finch/Finkel/Gatz) are to investigate mechanisms of educational influences on cognitive functioning and ADRD risk at multiple levels: genetic (polygenic score), individual, intergenerational (parental education), and environmental (GINI score for education), while investigating the impact of women’s differential access to educational and occupational opportunities across cohorts and countries. A recent expansion of the project focuses on increasing the number of African American twins in the IGEMS database to explore whether increasing access to education for women and African Americans influences dementia incidence as well whether the relative contribution of genetic influences may vary by gender or race/ethnicity. Responding directly to the Lancet report on modifiable risk factors for ADRD (Livingston et al., Reference Livingston, Huntley, Liu, Costafreda, Selbæk, Alladi, Ames, Banerjee, Burns and Brayne2024; Livingston et al., Reference Livingston, Huntley, Sommerlad, Ames, Ballard, Banerjee, Brayne, Burns, Cohen-Mansfield and Cooper2020), R01AG089666 (Reynolds/Neiderhiser) will apply twin models to strengthen or refute causal hypotheses and test gene–environment interplay among modifiable factors for ADRD, considering risk and resilience within and across developmental periods of the lifespan. The project will evaluate the role of early-, mid-, and late-life physical health, health behaviors, and socioemotional factors, with a particular focus on how early life risk factors work together with mid- and late-life health and socioemotional factors to influence ADRD risk and resilience. This project will also begin to take advantage of sibling and adoption designs that are embedded in several of the twin studies now being onboarded as well as in CAATSA and PTTR.

Finally, R21AG087486 (Luczak) examines alcohol consumption as a modifiable risk factor for ADRD. This exploratory/developmental research study will have the power to model alcohol risk for ADRD in nuanced ways and broaden our understanding of how and why alcohol may affect ADRD risk. This project leverages 50 years of data to understand differences in these relationships between men and women and how they intersect with the APOE gene, the strongest genetic risk factor for ADRD. These studies comprising IGEMS Phase 3 will enhance understanding of the interplay between genetic and environmental factors in ADRD risk and protection and will identify potential mechanisms essential for developing effective interventions to reduce the burden of ADRD.

Challenges

Building, maintaining, and growing an international consortium of twin cohorts is not an easy task, and the IGEMS team has faced and continues to face many challenges. Logistical challenges include coordination of data and researchers across multiple countries. Many of the twin cohorts in IGEMS are in EU countries and thus subject to General Data Protection Regulations (GDPR). As a result, although pooled analyses are the goal of most IGEMS projects, in some cases data-sharing protocols will only support parallel or meta-analyses. Careful development and implementation of data-use agreements are vital to support coordination across sites. Analytical challenges arise from the variety in birth cohorts, ages of intake, year of intake, length of follow-up, and sample sizes across studies. For example, multiple methods for data harmonization have been assessed and our efforts indicate that no one method is appropriate for all variables (Gatz, Reynolds et al., Reference Gatz, Reynolds, Finkel, Hahn, Zhou and Zavala2015). Differences in sample sizes and composition require exploration of weighting schemes or replication of results using ‘leave one out’ approaches to investigate the impact of any given twin cohort. Harmonizing PGS scores across studies with different genotyping platforms and methods requires careful coordination. Finally, issues that any twin study, longitudinal study, or study of aging may face (e.g., censoring, selective survival) can be multiplied in a consortium of multiple twin cohorts with different recruitment procedures, age cohorts, and follow-up periods.

Summary

The IGEMS consortium harnesses a combination of twin designs and multiple studies representing different cohorts and contexts. The accomplishments of the consortium demonstrate the feasibility of this type of collaboration in addressing G × E interplay with respect to important age-related outcomes. In addition to increasing diversity through onboarding, we can increase the historical coverage of birth cohorts around the world and increase our sample sizes with more recent studies that include genotyping. Researchers interested in information about IGEMS variables, IGEMS harmonization methods, access to analytical scripts, proposing an analysis using IGEMS data, or joining the IGEMS research consortium are encouraged to visit https://dornsife.usc.edu/cesr/igems. In consultation with the IGEMS data management team, individual researchers can develop an abstract of intent describing a proposed project, which is presented to the IGEMS team for approval. Representatives of twin cohorts interested in joining IGEMS are encouraged to contact the IGEMS leadership team at IGEMS.Consortium@gmail.com.

IGEMS consortium includes

Duke University: Brenda Plassman; Johns Hopkins University: Keith Whitfield; National Academies of Sciences, Engineering and Medicine: David Butler; University of Jyväskylä: Elina Sillanpää; Karolinska Institutet: Nancy Pedersen, Miriam Mosing, Malin Ericsson, Ida Karlsson; The Pennsylvania State University: Jenae Neiderhiser; QIMR Berghofer: Nicholas G. Martin, Michelle Lupton; University of California, San Diego: William Kremen, Carol Franz, Matthew Panizzon, Jeremy Elman; University of Colorado, Boulder: Chandra Reynolds, Daniel Gustavson; University of Helsinki: Eero Vuoksimaa; University of Minnesota: Matt McGue, Robert Krueger; University of New South Wales: Perminder Sachdev, Vibeke Catts, Teresa Lee, Karen Mather, Anbu Thalamuthu; University of Southern Denmark: Kaare Christensen, Marianne Nygaard; University of Southern California: Margaret Gatz, Brian Finch, Deborah Finkel, Christopher Beam, Susan Luczak, Thalida Em Arpawong, Andrew Petkus, Orla Hayden, Ellen Walters; Virginia Commonwealth University: Elizabeth Prom-Worley.