class: center, middle, inverse, title-slide .title[ # Temporal Changes in the Heritability of Longevity over the last 300 years <br> 👪 ] .subtitle[ ## Preliminary Evidence from Extended Pedigrees with 176 Million Kinship Pairs ] .author[ ### S. Mason Garrison <br> Wake Forest University ] --- layout: true <div class="my-footer"> <span> <a href="https://DataScience4Psych.github.io/DataScience4Psych/" target="_blank">S. Mason Garrison</a> </span> </div> --- <!-- Studies 1 and 2 evaluated temporal influences on heritability estimates from meta-analytic and raw twin data. Despite employing different methods in non-overlapping samples with different types of twin data, both studies identified temporal changes in genetic estimates of at least .03 per decade. Although our confidence is buoyed by this constructive replication, the decline in heritability estimates could be specific to twin designs or to the most recent historical epoch. For Study 3, we thus sought to clarify whether the pattern of changing heritability estimates might extend to an extended pedigree sample with 5 generations of kinship pairs dating back three centuries: the Utah Population Database (UPDB). The UPDB includes 1,018,929 deceased individuals born between 1700 and 1925, with 176,348,110 unique kinship links amongst them. Because formal models for temporal moderation of heritability in pedigree data do not yet exist, we conducted an indirect test by evaluating whether kinship similarity varied by century of birth. The similarity of first-degree relatives varied minimally (but significantly, a consequence of the very large N) depending on their century of birth. In contrast, second- through fifth-degree relatives were substantially more similar when both were born in the 20th century, with correlations approximately double those of pairs born prior to the 20th century. Preliminary Falconer-inspired analyses showed the same pattern, estimating heritability at 22% in the 20th century but 27% prior to the 20th century. Such results offer strong additional support for temporal changes in heritability. --> <!-- This is work with Michael Hunter,Ken Smith, and Alex Burt, and it was supported by the National Institute on Aging [RF1-AG073189].--> # Temporal Changes in the Heritability of Longevity ### Evidence from the Utah Population Database - S. Mason Garrison (Wake Forest University) - Michael D. Hunter (Penn State University) - Ken R. Smith (University of Utah) - S. Alexandra Burt (Michigan State University) **NIA R01-AG073189** <!-- Slide 1: Title (30 sec) --> <!-- Script: Good [morning/afternoon/evening]. I'm Mason Garrison from Wake Forest University. Today I’m going to talk about whether the heritability of longevity should be treated as stable across historical time. We usually estimate heritability as though it describes a trait in a relatively fixed way, but longevity is measured across people who lived under very different mortality environments. I've got 300 years of pedigree data and 176 million kinship pairs, so let's find out. --> <div class="figure" style="text-align: center"> <img src="data:image/png;base64,#d00_slide_files/figure-html/unnamed-chunk-2-1.png" alt="QR code for these slides" width="20%" /> <p class="caption">QR code for these slides</p> </div> .footnote[.center[ [r-computing-lab.github.io/slides/00_bga_2026/d00_slide](https://r-computing-lab.github.io/slides/00_bga_2026/d00_slide.html#1) ] ] --- <!-- Slide 2: Road map (30 sec) --> <!-- Script: I’ll organize the talk in five parts. First, I’ll explain why there are theoretical reasons to expect the heritability of longevity to change across historical time. Second, I’ll explain why longevity is a strong case for asking this question, because the mortality environment has changed dramatically across historical time. Third, I’ll introduce the Utah Population Database and explain what its pedigree structure lets us compare. Fourth, I’ll describe the analytic strategy, which uses kinship similarity across birth cohorts as an indirect test of temporal change. Finally, I’ll close with what these preliminary results imply for the next methodological step: formal pedigree models that can estimate temporal moderation directly. --> # Road map - Motivation: Why historical time may matter for h² - Data: The Utah Population Database - Method: Kinship similarity as an indirect test - Results by degree of relatedness - Next step: What formal modeling needs to do next --- <!-- Slide 3: Background and question (1 min) --> <!-- Script: We already know that longevity is heritable. Twin studies typically put h² around twenty to twenty-five percent, and pedigree-based estimates are in a similar range, with additional evidence for maternal-lineage or mitochondrial effects. More recent work suggests that estimates may be higher when extrinsic causes of death are accounted for. But these estimates usually share a deeper assumption: they treat heritability as if it is stable across historical time. For longevity, that is a strong assumption. The question here is not simply whether longevity is heritable. The question is whether the genetic contribution to longevity has remained stable across changing mortality environments. --> # The Usual Assumption: h² Is Stable .pull-left[ - Longevity is heritable - Twin estimates: h² ≈ 20–25% .small[(e.g., Herskind et al., 1996; McGue et al., 1993)] - Pedigree estimates: h² ≈ 25% + mt² ≈ 5% .small[(Burt, Garrison, et al 2025)] - Correcting for extrinsic mortality: h² ≈ 50% .small[(Shenhar et al., 2026)] ] .pull-right[ - But most estimates treat h² as time-invariant - That assumption is strong for a trait measured across historical cohorts - The key question is not whether longevity is heritable - The question is whether h² has been stable across time ] --- # Why Historical Time Should Matter <!-- Slide 4: Theoretical motivation (1 min) --> <!-- Script: Heritability is not a fixed biological constant. It is a variance ratio estimated in a particular population under a particular set of environmental conditions. That matters for longevity because the causes and timing of death changed substantially across the historical period represented in these pedigrees. The environment of 1750 was radically different from 1950. Life expectancy has climbed from around 32 to over 73 years. The dominant causes of death have shifted from infectious disease to chronic, degenerative conditions. If genetic and environmental contributions to longevity are sensitive to those historical shifts — as theory predicts — then heritability should not be constant across time. --> .pull-left[ - The environment has changed dramatically: - Life expectancy: ~32 → ~73 years - Dominant cause of death: infectious → chronic/degenerative - Antibiotic era, sanitation, nutrition — all compress extrinsic mortality - **If environment shapes how genetic differences express, h² should shift too** ] -- .pull-right[ - h² is a population-specific variance ratio - It depends on the environmental context in which the trait is measured - If environmental risk changes, genetic variance may become more or less visible - Longevity should therefore be historically contingent ] --- # Longevity Has Been Going Up <!-- Slide 5: Motivation figure (1 min) --> <!-- Script: Here are several over-engineered figures that I made to illustrate what the longevity landscape has looked like across the last three centuries. On the left, I'm showing data about the state of utah from the Social Security Administration, which has detailed records of all deaths in the state from 1941 to 2023. On the right, I'm showing data about Sweden from the Human Mortality Database, which has detailed records of all deaths in Sweden from 1750 to 2025. In both cases, panel A shows survival curves shifting right and compressing over time. Panel B shows the death-age density narrowing. Panel D shows between-population spread in median age at death, with historical events marked. WWI, the 1918 flu pandemic, and WWII all spike the between-population variance — catastrophic environmental shocks that overwhelmed genetic differences, temporarily suppressing h². Then the antibiotic era permanently compressed that variability. If the mortality environment changes, then the relative contributions of genetic and environmental variance may change as well. --> .pull-left[ .small[Utah demographics from the Social Security Administration (1941 - 2023)] <img src="data:image/png;base64,#img/panel_UT_CI_continent_min_1900_croped.png" alt="" width="60%" height="88%" style="display: block; margin: auto 0 auto auto;" /> ] .pull-right[ .small[Swedish Demographics from Human Mortality Database (1750 - 2025)] <img src="data:image/png;base64,#img/panel_SWE_CI_continent_min_1750_croped.png" alt="" width="60%" height="88%" style="display: block; margin: auto auto auto 0;" /> ] --- # Why Heritability Could Change Over Time <!-- Slide 6: Theoretical motivation (1.5 min) --> <!-- Script: Let's zoom in on panel C, which I actually cut out of the previous figure... This shows the age-by-time shifts in mortality. Gains have accumulated unevenly across the lifespan. The most dramatic improvements have been in infancy and old age, with more modest gains in midlife. This is important because it means that the environmental context for genetic expression has changed differently at different ages. --> ## SSA data for the entire USA: (1900-2019) .pull-left-wide[ <img src="data:image/png;base64,#img/highlight_ages_usa.gif" alt="" width="90%" style="display: block; margin: auto;" /> ] -- - Interpretation: Relative to a person born in 1900, what proportion of deaths were postponed? --- background-image: url(data:image/png;base64,#img/Slide9_cropped.PNG) background-size: 98% background-position: center background-repeat: no-repeat class: middle <!-- Slide 7: UPDB (15 sec) --> <!-- Script: The Utah Population Database is one of the world's deepest genealogical resources, containing over a million deceased individuals born between 1700 and 1925, linked into multigenerational pedigrees via birth records, death records, and historical census data (Skolnick et al., 1979). That gives us 176 million unique kinship pairs spanning multiple birth centuries. And critically — unlike our AD work last year — everyone has a death record. Phenotype coverage is complete. --> # The --- background-image: url(data:image/png;base64,#img/Slide10_cropped.PNG) background-size: 94% background-position: center background-repeat: no-repeat # The Data: What We Have <!-- Slide 8: UPDB data structure (45 sec) --> --- background-image: url(data:image/png;base64,#img/Slide11_cropped.PNG) background-size: 93% background-position: center background-repeat: no-repeat <!-- Slide 9: Sample structure (45 sec) --> # The Data: Who We Have --- background-image: url(data:image/png;base64,#img/Slide12_cropped.PNG) background-size: 93% background-position: center background-repeat: no-repeat <!-- Slide 10: Sample structure (45 sec) --> # The Data: Who We Have --- <!-- Slide 11: Analytic sample (1 min) --> # The Data: What We Used <!-- Script: We used the Utah Population Database—one of the world’s largest and deepest genealogical datasets, with over 11 million individuals linked across multigenerational pedigrees. For this project, we extracted a subset of 4.8 million individuals embedded in multigenerational family trees, anchored around AD cases and matched controls. These pedigrees span up to 17 generations, seeded with founders from 19th-century Utah (Skolnick et al., 1979; O’Brien et al., 1994). --> .pull-left[ - For this project, we were able to use a more manageable subset - **1,018,929** deceased individuals - Born 1700–1925 - Pedigrees up to 17 generations deep - Linked via genealogical and vital records .small[(Skolnick et al., 1979)] ] .pull-right[ - **176,348,110** unique kinship pairs - 1st through 5th-degree and beyond - Multiple birth centuries represented - Same dataset as Burt, Garrison et al. (2025) ] --- <!-- Slide 12: Pedigree reconstruction (1 min) --> <!-- Script: The pedigrees were reconstructed using BGmisc, our R package for extended behavior genetic analysis, in combination with graph-theoretic path-based relatedness estimation. For each dyad, we traced nuclear relatedness, maternal and paternal lineage, mitochondrial relatedness, and indicators of potential shared environment. The point is that we are not only counting relatives. We are using the pedigree graph to characterize what each pair shares. --> # The Data: What We Made .pull-left[ - We reconstructed extended pedigrees: - using **BGmisc**, our custom R package for extended behavior genetic analysis .small[(Garrison, Hunter, Lyu, Trattner, & Burt, 2024)], - in combination with graph theory, - and computed path-based relatedness estimates .small[(Hunter, Garrison et al., 2026)]. - For each dyad, we traced: - nuclear relatedness, - maternal vs. paternal lineage, - mtDNA, and potential shared environment. ] -- .pull-right[ - Go read the 35-page appendix of Burt, Garrison et al if you want more specifics... <img src="data:image/png;base64,#img/mtdna.png" alt="" width="55%" style="display: block; margin: auto;" /> ] --- # The Analytic Strategy <!-- Slide 8: Methods (1.5 min) --> <!-- Script: Because formal models for temporal moderation of heritability in pedigree data don't yet exist, what we have is an indirect test. We split kinship pairs by whether both members were born in the 20th century (1900 or later) versus both born prior, then ask: does within-pair similarity differ by birth century, and does that pattern vary by degree of relatedness in the way we'd expect if heritability itself were changing? --> .pull-left[ - **The question**: - does within-pair similarity in longevity differ by birth century? - Birth-period contrast: - both members born in the 20th century - both members born before the 20th century ] .pull-right[ - Kinship contrast: - 1st-degree relatives share A and shared context (C etc) - 2nd–5th degree relatives share A, not shared household - Key test: - If h² changes over time, distant kin should show it most clearly ] --- # The Phenotype and Similarity Metric <!-- Slide 14: Outcome and similarity metric (45 sec) --> <!-- Script: The phenotype is age at death, recorded for every individual in the analytic sample. For these preliminary analyses, we use a top-ten-percent longevity indicator and estimate tetrachoric correlations within kinship pairs. Those correlations are stratified by degree of relatedness and by historical birth-period pairing. This lets us ask whether relatives are more similar in longevity in one historical period than another, and whether that pattern depends on how closely they are related. --> .pull-left[ - Phenotype: - age at death - recorded for every individual - Longevity threshold: - top 10% survival indicator ] .pull-right[ - Similarity : - tetrachoric correlations .small[(Olsson, 1979; Fox, 2015)] - stratified by degree of relatedness - estimated separately by birth-period pairing ] --- # First-Degree Relatives Are Not the Cleanest Test <!-- Slide 9: 1st degree results (1 min) --> <!-- Script: For first-degree relatives, we found minimal variation by birth century. The difference was statistically significant — but this is a consequence of the enormous sample size, not of substantive effect size. The practical pattern is flat. This is exactly what our theoretical model predicts: if A rises as C falls, the two changes partially offset each other in first-degree pairs. The signal is not here. --> .pull-left-narrow.medi[ - Within-pair similarity for 1st-degree relatives (Burt, Garrison et al., 2025, Table S7): - **Minimal** variation across birth centuries - Practically trivial - Prediction: - If A↑ as C↓ over time, net change in 1st-degree similarity ≈ 0 - first-degree similarity may therefore look relatively flat ] -- .pull-right-wide.medi[ <div class="datatables html-widget html-fill-item" id="htmlwidget-1b4ff99564eb6e8884a5" style="width:100%;height:auto;"></div> <script type="application/json" data-for="htmlwidget-1b4ff99564eb6e8884a5">{"x":{"filter":"none","vertical":false,"data":[["20th","Pre-20th","20th","Pre-20th","20th","Pre-20th","20th","Pre-20th","20th","Pre-20th","20th","Pre-20th","20th","Pre-20th"],[1,1,0.5,0.5,0.25,0.25,0.125,0.125,0.0625,0.0625,0.03125,0.03125,0.015625,0.015625],[327830,535831,580305,2520353,493935,8052111,3451434,18127495,5553659,36892638,19175756,50524112,29857152,54773096],[0.9999,0.9999,0.1235078047474133,0.1357027123194933,0.05158316111279208,0.03819206895110792,0.03105964121523105,0.002476782920107961,0.01687337462233144,-0.01528398991022814,0.01106304437777373,-0.01864622327317465,0.006162812107978925,-0.01674921293631701],[0.001746536395900366,0.001366115119469938,0.001312729398994774,0.0006298977967916087,0.001422878238106818,0.0003524076157827023,0.0005382705149895762,0.0002348719615646699,0.0004243367341126916,0.0001646380372955125,0.0002283619908345794,0.0001406859349804157,0.0001830104353334728,0.0001351189864696294],[64.3328736614089,59.7332165180323,63.60525340153492,60.7095724160673,62.96404400218871,59.18040396307772,64.1190407068943,58.45305731121724,63.67033999168741,59.0357794725955,64.14532160073105,59.37378574853397,64.25575624328623,59.90649939775407],[0.339831806614223,0.202590031814946,0.3387370667910346,0.2414319759344405,0.3260147456561721,0.2431064679777758,0.3430613071000568,0.2416766645955414,0.3371849735003302,0.2560436998665237,0.3422237977150716,0.2666347264807573,0.3430776730732427,0.2780832815466849],[1,1,0.9891505329094183,0.7138265949253934,0.5861601222832964,0.4124445129978958,0.2520595787142388,0.2069488917249736,0.1209379257891059,0.1050292472986074,0.05733552304274209,0.05502426247491495,0.02856468024813619,0.0306796789431074]],"container":"<table class=\"display\">\n <thead>\n <tr>\n <th>cohort<\/th>\n <th>R<\/th>\n <th>n_pairs<\/th>\n <th>rho_w<\/th>\n <th>se_w<\/th>\n <th>age_w<\/th>\n <th>top10_w<\/th>\n <th>mtdna_w<\/th>\n <\/tr>\n <\/thead>\n<\/table>","options":{"pageLength":10,"scrollX":true,"autoWidth":true,"columnDefs":[{"targets":5,"render":"function(data, type, row, meta) {\n return type !== 'display' ? data : DTWidget.formatRound(data, 1, 3, \",\", \".\", null);\n }"},{"targets":7,"render":"function(data, type, row, meta) {\n return type !== 'display' ? data : DTWidget.formatRound(data, 1, 3, \",\", \".\", null);\n }"},{"targets":3,"render":"function(data, type, row, meta) {\n return type !== 'display' ? data : DTWidget.formatRound(data, 3, 3, \",\", \".\", null);\n }"},{"targets":4,"render":"function(data, type, row, meta) {\n return type !== 'display' ? data : DTWidget.formatRound(data, 3, 3, \",\", \".\", null);\n }"},{"targets":6,"render":"function(data, type, row, meta) {\n return type !== 'display' ? data : DTWidget.formatRound(data, 3, 3, \",\", \".\", null);\n }"},{"targets":1,"render":"function(data, type, row, meta) {\n return type !== 'display' ? data : DTWidget.formatRound(data, 3, 3, \",\", \".\", null);\n }"},{"className":"dt-right","targets":[1,2,3,4,5,6,7]},{"name":"cohort","targets":0},{"name":"R","targets":1},{"name":"n_pairs","targets":2},{"name":"rho_w","targets":3},{"name":"se_w","targets":4},{"name":"age_w","targets":5},{"name":"top10_w","targets":6},{"name":"mtdna_w","targets":7}],"order":[],"orderClasses":false}},"evals":["options.columnDefs.0.render","options.columnDefs.1.render","options.columnDefs.2.render","options.columnDefs.3.render","options.columnDefs.4.render","options.columnDefs.5.render"],"jsHooks":[]}</script> ] --- # Distant Relatives Carry the Cleaner Signal <!-- Slide 10: KEY RESULT (2 min) --> <!-- Script: Here's where the story is. For 2nd through 5th-degree relatives — cousins and beyond — pairs born in the 20th century were substantially more similar in longevity than pairs born before the 20th century. The correlations were approximately double. These pairs share genetics but not shared household. So this pattern specifically implicates increased genetic influence on longevity in the modern era. The effect is consistent across all distant kin classes — it's not driven by any one degree. And it is consistent with the theoretical prediction: as extrinsic environmental mortality compressed in the antibiotic era, genetic differences between families account for a progressively larger share of longevity variation. --> <!-- [FIGURE: Add results figure for 2nd-5th degree correlations by birth century] --> .pull-left[ - *2nd–5th degree relative pairs** (Burt, Garrison et al., 2025, Table S7): - Born in the **20th century**: correlations **~2× higher** than - pairs born **prior to the 20th century** ] -- .pull-right[ <img src="data:image/png;base64,#img\cohortpis_all.png" alt="" width="90%" style="display: block; margin: auto;" /> ] --- # Why Pooled h² Is Not the Main Result <!-- Slide 11: Falconer estimates (1 min) --> <!-- Script: We also ran preliminary Falconer-inspired analyses. These gave h² of 22% for 20th-century pairs and 27% for pre-20th-century pairs — which seems to contradict the distant-relative result. But this is an artifact of the aggregate estimate being dominated by 1st-degree pairs, which are the most numerous and show minimal change. The flat first-degree pattern drowns out the distant-kin signal in the pooled estimate. Formal models that can decompose these effects by degree and birth period are the obvious next step — and they don't exist yet. --> .pull-left[ - Preliminary Falconer-inspired h²: - **20th century**: h² ≈ **22%** - **Pre-20th century**: h² ≈ **27%** - At first glance, this appears to contradict the distant-relative pattern ] -- .pull-right[ - It doesn't — the aggregate estimate is dominated by 1st-degree pairs - Most numerous, minimal change → swamps the distant-kin signal - Formal temporal moderation models for pedigrees **don't yet exist** - These estimates are preliminary and indirect - Decomposing properly is the next methodological step ] --- # What We Learned <!-- Slide 12: Discussion (1 min) --> <!-- Script: The central finding is that temporal change in kinship similarity extends beyond twins and beyond the most recent historical period. In these extended pedigrees, the clearest signal appears among distant relatives, where genetic relatedness is preserved but shared household context is much weaker. That pattern is consistent with the idea that the genetic contribution to longevity changed across historical mortality regimes. At the same time, the results are still preliminary because the current analysis is indirect. The field needs formal models that can estimate temporal moderation of genetic and environmental variance directly in extended pedigrees. --> .pull-left[ - Temporal change in kinship similarity seen in **extended pedigree data** going back **300 years** - Signal is in **distant kin** — specifically implicates genetic influence - This pattern is consistent with changing genetic influence across mortality regimes ] -- .pull-right[ - First-degree relatives are not expected to carry the cleanest signal - Pooled h² estimates are not sufficient for this question - Formal temporal moderation models for pedigrees are needed ] --- <!-- Slide 19: Next steps (1 min) --> <!-- Script: The next step is methodological. We need a formal multigenerational pedigree model that can estimate temporal moderation directly, rather than relying on degree-specific correlations as an indirect test. That model needs to allow genetic and environmental variance components to change over historical time, including both smooth trajectories and discrete shifts around historically meaningful events. We also need simulation work to establish when the model can and cannot detect these effects. Once that framework is validated, the same approach can be applied to the full UPDB and extended to other aging-related outcomes, including Type II diabetes, Alzheimer’s disease, and Parkinson’s disease. --> # Next Steps - **Formal modeling**: - multigenerational pedigree model for temporal moderation of A, D, C, and E - Linear, quadratic, and cubic forms - **Validate by simulation**... - **Apply in the full UPDB**... - **Extend to related aging outcomes**... --- # Acknowledgements - **Utah Population Database** (University of Utah / Huntsman Cancer Institute) - **Co-authors**: Michael D. Hunter, Ken R. Smith, S. Alexandra Burt - **R packages**: `BGmisc` (Garrison, Hunter, Lyu, Trattner, & Burt, 2024), `ggpedigree` (Garrison, 2025) - **Human Mortality Database** for theoretical illustration figures --- ## Any Questions? Feel free to ask now, or reach out: _garrissm@wfu.edu_ | _github.com/smasongarrison_ <img src="data:image/png;base64,#d00_slide_files/figure-html/qr_bga2026-1.png" alt="" width="30%" style="display: block; margin: auto;" /> .footnote[.center[ [r-computing-lab.github.io/slides/00_bga_2026/d00_slide.html](https://r-computing-lab.github.io/slides/00_bga_2026/d00_slide.html) ] ]