{"paper_id":"0e87470b-cd5a-44b7-a749-e8e15b8ae8b7","body_text":"Subcontinental Genetic Diversity in the All of Us Research Program: Implications for Biomedical \nResearch \n \nMateus H. Gouveia1‡, Karlijn A. C. Meeks1,2, Victor Borda3,4, Thiago P. Leal5, Fernanda S. G. Kehdy6, \nReagan Mogire1, Ayo P. Doumatey1, Eduardo Tarazona-Santos7, Rick A. Kittles8, Ignacio F. Mata5, \nTimothy D. O’Connor3,4, Adebowale A. Adeyemo1‡*, Daniel Shriner1‡* and Charles N. Rotimi1‡*. \n \n1Center for Research on Genomics and Global Health, National Human Genome Research Institute, \nBethesda, Maryland, USA. \n2Division of Endocrinology, Diabetes and Nutrition, Department of Medicine, University of Maryland \nSchool of Medicine, Baltimore, MD, USA. \n3The University of Maryland-Institute for Health Computing, University of Maryland School of Medicine, \nNorth Bethesda, Maryland, USA. \n4Institute for Genome Sciences, University of Maryland School of Medicine, Baltimore, Maryland, USA. \n5Lerner Research Institute, Genomic Medicine, Cleveland Clinic, Cleveland, Ohio, USA. \n6Laboratório de Hanseníase, Instituto Oswaldo Cruz, Fundação Oswaldo Cruz (FIOCRUZ), Rio de Janeiro, \nBrazil. \n7Department of Genetics, Ecology, and Evolution. Instituto de Ciências Biológicas, Universidade Federal \nde Minas Gerais, Belo Horizonte, Brazil. \n8Morehouse School of Medicine, Atlanta, GA, USA \n‡Corresponding authors: \nMateus H. Gouveia: mateus.gouveia@nih.gov, mateushgbio@gmail.com \nAdebowale A. Adeyemo: adeyemoa@mail.nih.gov \nDaniel Shriner: shrinerda@mail.nih.gov \nCharles N. Rotimi: rotimic@mail.nih.gov \n* Joint senior authors \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nABSTRACT \nThe All of Us  Research Program ( All of Us ) seeks to accelerate biomedical research and address the \nunderrepresentation of minorities by recruiting over one million ethnically diverse participants across the \nUnited States.  A key question is how self -identification with discrete, predefined race a nd ethnicity \ncategories compares to genetic diversity at continental and subcontinental levels. To contextualize the \ngenetic diversity in All of Us , we analyzed ~2 million common variants from 230,016 unrelated whole \ngenomes using classical population genetics methods, alongside reference panels such as the 1000 \nGenomes Project, Human Genome Diversity Project, and Simons Genome Diversity Project. Our analysis \nreveals that participants within self -identified race and ethnicity groups exhibit a gradient of genetic \ndiversity rather than discrete clusters. The distributions of continental and subcontinental ancestries show \nconsiderable variation within race and ethnicity, both nationally and across states, reflecting the historical \nimpacts of U.S. colonization, the transatlantic slave trade, and recent migrations. All of Us samples filled \nmost gaps along the top five principal components of genetic diversity in current global reference panels. \nNotably, “Hispanic or Latino” participants spanned much of the three-way (African, Native American, and \nEuropean) admixture spectrum. Ancestry was significantly associated with body mass index (BMI) and \nheight, even after adjusting for socio-environmental covariates. In particular, West -Central and East \nAfrican ancestries showed opposite associations with BMI. This study emphasizes the importance of \nassessing subcontinental ancestries, as the continental approach is insufficient to control for confounding \nin genetic association studies.  \n \n \n \n \n \n \n \n \n \n \n \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nINTRODUCTION \nUnderrepresentation of racial and ethnic minorities in the United States (U.S.) has significantly limited \nbiomedical studies, reflecting a broader global issue of underrepresentation of non -European ancestry \npopulations. This lack of diversity hampers progress in developing precision medicine that effectively \nserves the nation's highly diverse population 1. Additionally, many genomic and biomedical studies in the \nU.S. are constrained by small sample sizes, lack of comprehensive phenotype data, and limited funding \nwhich impedes the ability to support long-term longitudinal research efforts. The  All of Us  Research \nProgram (All of Us) aims to accelerate biomedical research and address minority underrepresentation by \ncollecting comprehensive longitudinal health data on at least one million samples reflecting diversity \nacross the U.S.\n2.  \n \nRecently, All of Us  investigators published an analysis of the current release ( version 7)  of short -read \nwhole-genome sequencing (WGS) data from 245,388 participants3. This study and others3,4 reported that \nUniform Manifold Approximation and Projection (UMAP) 5 recapitulated known patterns of population \nstructure and that participants may cluster within discrete race and ethnicity categories, sparking debate \non the validity and interpretability of these results\n6,7. UMAP 8 is a nonlinear technique to reduce the \ncomplexity of multidimensional data and facilitate visualization in low -dimensional plots. This method is \ncommonly applied in single-cell analysis and is designed to preserve the local structure of the data at the \nexpense of the global structure. Consequently, differences can be exaggerated, and patterns that do not \nexist in the original data can be introduced.  Since admixture generates populations with allele frequencies \nthat are linear combinations of those in the parental populations, admixture is fundamentally a linear \nadditive process.  \n \nThere has been extensive discussion and recent recommendations regarding using race, ethnicity, and \nancestry in scientific research\n9,10. Ancestry is defined as the population origin of an individual's alleles at \npolymorphic sites 11, which can be summarized as an average across the genome. In this study, we \nconsistently adopt the genome-wide definition of ancestry. In contrast, race and ethnicity are socio -\ncultural or geopolitical constructs, with race associated with abstract belief s about shared genetic \nancestry or biological traits, and ethnicity connected to abstract perceived cultural practices 12. In the \nquestionnaire used by All of Us, racial categories primarily reflect politically recognized definitions of race \nwithin the country used by the U.S. Census, while ethnicity refers solely to whether a person is Hispanic \nor Latin American13. The definitions and use of race and ethnicity vary significantly across countries14, with \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nexamples like the different classifications in Brazil 15, where race is often considered fluid, meaning it is a \nsubjective perception of physical appearance and individuals may identify differently depending on social \ncontext, in contrast to the more rigid racial categories commonly used in the U.S. Similar variations can \nbe observed across other countries, such as in Colombia and Mexico, where mestizo (mixed) identity is a \ndominant classification, despite the complex interplay of genetic and cultural diversity\n15. In the \nNetherlands, the country of birth indicator is widely accepted as a criterion for identifying ethnic groups16, \nwhile the concept of race is not utilized. Another complexity is that ethnic groups can be racialized, as \nexemplified by Hispanics/Latinos in the U.S., while racial groups can be ethnicized, as seen with Asians \nthrough shared cultural characteristics\n10. Despite the correlation between self -identified race/ethnicity \nand genetic ancestries, including in the U.S. 17,18 and Latin America19, race and ethnicity are poor proxies \nfor genetic ancestry9. \n \nBased on classical population genetics methods, we assess the genetic diversity, admixture, and ancestry \nof 230,016 unrelated All of Us participants in the context of a newly compiled reference panel of genetic \ndiversity. Our analyses address four key questions: (1) What does the population structure in All of Us look \nlike across U.S. Census race and ethnicity population descriptors? (2) Does the genetic diversity in All of \nUs mirror the diversity seen in global diversity panels? (3) Is there variability in continental and \nsubcontinental ancestries within racial and ethnic groups, and does this variability differ across U.S. \ngeographical regions and states? (4) Are genetic ancestries associated with biological traits after \naccounting for major socio-environmental factors? \n \n \nRESULTS \nGenetic Diversity in All of Us \nTo assess the genetic diversity in All of Us , we analyzed the WGS data of 230,016 unrelated participants. \nOur preliminary analysis using all genetic variants in the WGS dataset proved computationally prohibitive, \nconsistent with previous All of Us population genetics analyses\n3. Therefore, our analyses were based on a \nsubset of variants from the WGS dataset; however, this subset included ~2 million high -quality genome-\nwide variants across the 22 autosomes compared with ~200,000 variants across two chromosomes used \nin a previous study\n3. Our analyses of genetic diversity rely on classical methods in population genetics, \nincluding principal component analysis (PCA)20, ADMIXTURE21, and F statistics22. We acknowledge that the \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\ngenetic variation, population structure, and sample composition of the dataset influence unsupervised \nclustering analyses (e.g., PCA and ADMIXTURE). \n \nUnsupervised PCA of All of Us  participants identified five statistically significant ( p < 0.05) principal \ncomponents (PCs) with a genome-wide contribution of SNPs to the variance explained by each PC (SNP \nloading) and large eigenvalues (Figs. 1 and 2 and Fig. S1). These five PCs were used to interpret population \nstructure within All of Us. PCs beyond PC 5 primarily reflected genetic differentiation due to variation in \nspecific loci ( e.g., the MHC locus in PC 6 and both lactase [LCT ] and MHC in PC 9), rather  than broad \ngenome-wide differentiation (Fig. S1). Participants were mapped with gradients of genetic diversity along \nPCs 1 and 2 (Fig. 1A and 1B), consistent with findings from biobank -level genomic studies of ancestrally \ndiverse U.S. samples in New York City\n23. Interestingly, being born in the U.S. or elsewhere did not influence \nthe overall pattern of population structure (Fig. S2). This suggests that recent migrants reflect the overall \ngenetic diversity of All of Us participants born in the U.S.  “Black or African American” participants \nrepresent a wide range of diversity along PC1 (Fig. 1A), while “Hispanic or Latino” participants are \ndistributed across PCs 1 and 2 (Fig. 1B), encompassing nearly the whole triangle of space expected in \nthree-way admixture.  \n \nMost (92%) individuals who did not self -identify within any U.S. race category (Fig. 1A) self -identified \nethnically as “Hispanic or Latino” (Fig. 1B), suggesting that the existing U.S. racial categories do not reflect \nthe identity of this group. We observed varying degrees of cross -classification by race among those \nparticipants who self -identified ethnically as “Hispanic or Latino”: “White” (3,682), “Black or African \nAmerican” (842), “Asian” (225), “Native Hawaiian or Other Pacific Islander” (43), and “Mor e than one \nrace” (490). This result reflects the complexity of how race and ethnicity categories are perceived in the \nU.S. by “Hispanic or Latino” individuals. The observation that “Hispanic or Latino” ethnicity includes \nindividuals from all predefined categories of race reinforces how relying on these categories may provide \ninsufficient adjustment for population structure in association studies. \n \nBased on classical measures of genetic differentiation (F-statistics or fixation indices) between participants \nwithin race and ethnicity categories, we observed that most of the genetic variance is within race and \nethnicity groups (1-F\nIT = 93.4%) rather than between groups (average FST = 0.042). This result is consistent \nwith studies of human genetic diversity24–26 showing that within-population differences account for most \n(84-95%) of the genetic variation, while differences among major geographic populations account for up \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nto 5%. The average genetic differentiation between race and ethnicity groups ( FST = 0.042) is about half \nthe average ( FST ~ 0.1) among populations across different continents 27,28, suggesting that on average \nindividuals assigned to worldwide populations will be more genetically distinct than individuals assigned \nto race or ethnicity. \n \nMulticontinental Genetic Diversity Represented by All of Us \nTo better explore the relationship between self-identification and gradients of genetic variation, we \nmerged race/ethnicity categories into the same PCA representation (Fig. 2). We identified genetic \ndiversity within All of Us  by projecting samples from widely used reference panels (the 1000 Genomes \nProject\n29 and the Human Genome Diversity Project [HGDP] 30, combined referred to as the \n“multicontinental diversity panel”) onto All of Us samples (Fig. 2).  We observed that the genetic diversity \nin All of Us  not only recapitulated but also exceeded the spectrum of diversity reflected in our \nmulticontinental diversity panel across PCs 1 to 5 (Fig. 2).  \n \n“Black or African American” participants were distributed along PC 1, representing a gradient between \nAfrican and European genetic diversity. “Hispanic or Latino” participants were spread across the top two \nPCs, reflecting the same gradient of African and European genetic diversity along PC 1 in addition to a \ngradient of European and Native American genetic diversity along PC 2. Also, we observed clustering of \n“Hispanic and Latino” participants distributed along a gradient of Native American genetic diversi ty \ndefined by PCs 3 and 4. Importantly, All of Us samples project to spaces not covered by the 1000 Genomes \nProject or HGDP, including a broader range of three-way admixture patterns not observed in the Admixed \nAmerican or Native American reference samples (PCs 1 and 2). Conversely, some genetic diversity within \nHGDP is not represented in All of Us  (PCs 3 and 4), such as Central Asians (Hazara and Uygur) and \nOceanians (Bougainville, Papuan Sepik, and Papuan Highlands). \n \nAll of Us Coverage of Multicontinental Genetic Diversity  \nTo evaluate the coverage of genetic diversity in All of Us  in the context of multi -continental genetic \ndiversity, we created a comprehensive panel capturing the genetic diversity, including 162 populations \nacross all continents from different population genomics studies referred as to “global diversity panel”: \nthe harmonized 1000 Genomes Project and the Human Genome Diversity Project (HGDP)\n31, and the \nSimons Diversity Project (SGDP) 32, which provides a much broader globally sampling of indigenous \npopulations32. PCA analysis of our global diversity panel showed different gradients of genetic diversity \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n(Fig. 3): African vs. European (PC 1), European vs. Native American (PC 2), Asian vs. Native American (PC \n3), Eastern vs. Southern Asian (PC 4), South African hunter -gatherer vs. other African populations (PC 5), \nand Northern European vs. Southern European and Middle Eastern (PC 6).  \n \nProjection of the All of Us  samples onto our global diversity panel, along with the calculation of convex \nhull PC areas covered by All of Us, revealed substantial coverage of multi-continental diversity across five \nPCs (PCs 1 –4 and 6; Fig. 3). Similar to our findings based on unsupervised PCA (Figs. 1 and 2), All of Us  \ndiversity not only covered most of PCs 1 and 2 but also extended to fill nearly the entire triangular space \nof three-way admixture. For PCs 3 and 4, All of Us  samples covered broad genetic di versity, including a \nwider set of admixture combinations than typical two-way or three-way ancestries. Despite this extensive \ncoverage, All of Us did not overlap with some populations, such as specific Native American populations \n(e.g., the Karitiana in Brazil), specific European populations ( e.g., Sardinians and Basques, PC 3 vs PC 4), \nMiddle Eastern populations ( e.g., Bedouin, PC 6), and Southern Asian populations ( e.g., Indian Telugu in \nthe UK, PC 6). We confirmed that “White” participants covered most North -to-South European genetic \ndiversity (~90%) of our panel of European subcontinental diversity (Fig. S3)\n33. Consistent with previous \nreports on the genetics of the African diaspora 34,35, the genetic diversity in All of Us  did not capture the \ndiversity specific to African hunter-gatherer populations (PC 5). Our analysis suggests that All of Us  \ncaptures much of the worldwide genetic diversity in humans, while covering a broader range of admixture \nthan is represented in current reference panels.  \n \nAncestry and Admixture in All of Us \nTo assess admixture in All of Us, we first performed unsupervised ADMIXTURE\n21 analysis using our global \ndiversity panel.  This analysis identified the most likely number of ancestral clusters as 13 (Fig. S4 and \nTable S1). Next, we projected All of Us samples onto our global diversity panel (supervised ADMIXTURE \nanalysis) to estimate individual ancestry proportions across these 13 ancestry clusters (Fig. 4A and B). \nDendrogram analysis of genetic differentiation, based on estimates of F\nST among the ancestry clusters, \nrevealed six clades of ancestries that we contextualize in terms of present -day geographic distributions \n(Fig. 4C): 1) West Central, West, South, and East African, 2) North European, South European, and Middle \nEastern, 3) Indi an and South Asian, 4) North Asian and Southeast Asian, 5) Native American, and 6) \nOceanian. These clades of ancestries are consistent with estimated ancestry clusters reflecting major \ngeographic regions observed in classical studies of human genetic diversity\n26,24. \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nWe observed large variability in individual ancestry proportions within race and ethnicity categories (Figs. \n4 and 5, Tables S3 and S4). “Hispanic or Latino” participants had mean ancestral proportions consistent \nwith three-way admixture (Table S2): total European (49.95%, SD = 19.15), Native American (31.35%, SD \n= 22.59), and African (13.47%, SD = 17.30, Table S2). Most “Hispanic or Latino” participants (75.1%) had \nat least 50% total African, European, or Native American (Table S3). “Black or African Ameri can” \nparticipants had mean ancestral proportions mostly consistent with two -way admixture (Table S2): \n82.72% (SD = 10.87) African and 14.31% (SD = 9.35) European. However, we observed “Black or African \nAmerican” participants with ≥ 50% total European (n = 483, 1.06%) ancestry as well as other ancestries \n(Table S3).  \n \n“White” participants primarily had European ancestries (90%, SD = 5.58), with a minor proportion of South \nAsian ancestry (8.38%, SD = 2.85, Table S2). South Asian ancestry likely reflects admixture from ancient \nmigrations of southern steppe peoples\n36. Additionally, some “White” participants were found to have ≥ \n50% African (n = 29, 0.23%), Native American (n = 103, 0.08%), and other ancestries (Table S3). “Middle \nEastern or North African” participants were primarily mixtures of European (65.59%, SD = 14.62) and \nSouth Asian (26.49%, SD = 15.30) ancestries (Table S2).  “Asian” participants predominantly had East Asian \n(68.35%, SD = 42.76) and South Asian (24.73%, SD = 38.00) ancestries. “Native Hawaiian or Other Pacific \nIslander” participants exhibited a highly admixed ancestry profile, with a notable proportion of East Asian \nancestry (37.14%, SD = 32.74, Table S2). We also observed a wide distribution of subcontinental ancestries \nwithin race and ethnicity groups. For example, while the majority of 'White' participants (n = 120,673; \n97%) had predominantly Southern European ancestry, a subset (n = 3,418; 3%) exhibited predominantly \nNorthern European ancestry, which is strongly associated with Finnish and Estonian European \npopulations. \n \nRegional Variability in the All of Us  \nWhen assessing the distribution of ancestries across U.S. states, we observed regional variation. Among \n“Black or African American” participants (Fig. 6), mean total African ancestries was higher in southern \nstates, such as South Carolina (85.6%) and Florida (85.3%) compared to western states like California \n(77.3%) and Arizona (79.0%). African subcontinental ancestries largely reflected the overall pattern of \ntotal African ancestries, with a notable exception in South Carolina, where West African ancestry was \nslightly elevated. East African ancestry, however, was more uniformly distributed across U.S. states. This \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nalignment between total African ancestries and subcontinental patterns is likely influenced by the \nhistorical transatlantic slave trade, which predominantly involved individuals from West African regions37. \n \nFor “Hispanic or Latino” participants (Fig. 7), ancestry patterns were highly diverse across U.S. states. The \nhighest mean Native American ancestry proportions were observed in southwestern and western states, \nincluding California (45.6%), Texas (39.4%), a nd Arizona (38.0%), while certain southeastern states, such \nas South Carolina (50.1%) and Tennessee (45.0%), also showed elevated levels. In contrast, “Hispanic or \nLatino” participants had lower Native American ancestry proportions in northeastern states such as \nPennsylvania (16.3%), New York (16.6%), and New Jersey (18.1%), but also in Florida (17.4%). “Hispanic or \nLatino” participants with the highest total European ancestry proportions were found in Florida (64.3%) \nand Pennsylvania (63.0%), followed by New Mexico (58.3%) and New Jersey (55.3%). Total African \nancestry in “Hispanic or Latino” participants exhibited an increasing gradient from west to east, with the \nhighest proportions observed in New York (31.5%) and the lowest in New Mexico (3.9%), California (5.4%), \nand Arizona (5.6%). \n \nThe mean total European ancestry of “White” participants (Fig. 8) was higher in northeastern states, while \nlower proportions were found along the Mid -Atlantic and southern states. Subcontinental European \nancestry also varied by state, with North European ancestry more prevalent in Midwestern states and \nSouth European ancestry more prevalent in the South. Notably, Middle Eastern ancestry showed a sharp \ncontrast, with mean proportions significantly higher in New York and New Jersey (~15%) compared to \nstates like Alabama and Arkansas (~3%). \n \nAll of Us Participants in Race/Ethnicity and Predicted Ancestry Groups \nGenetic epidemiology studies in the U.S. often conduct association analyses, including genome-wide \nassociation studies (GWAS), by stratifying participants according to self -identified race and ethnicity\n38–40. \nHowever, in the All of Us  Research Workbench, which includes GWAS data on approximately 3,400 \nphenotypes in the \"All by All tables\", participants are stratified by categorical ancestry groups (based on \nancestry estimation) rather than self-identified race and ethnicity. As GWAS findings in the All by All tables \nhave been used to replicate previous associations from studies designed based on self -identified \nrace/ethnicity, we examined the correspondence between participants stratified by race/ethnicity and \ngenetic ancestry.  \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nWe compared the ancestry distributions of “Black or African American” and “White” participants with \npredicted African and European ancestry participants, respectively (Fig. 9A and B). We noticed that \nindividuals stratified based on predicted categorical ancestry groups had more homogeneous ancestry \ndistributions, with fewer participants with outlier profiles of ancestry compared with participants \nstratified by race (Fig. 9A and B). Despite similar ancestry distributions of participants stratified based on \nthese approaches, we observed that participants stratified by African (Fig. 9C) and European (Fig. 9D) \ngenetic ancestry are distributed across multiple race/ethnicity groups. For example, participants in the \npredicted African category include individuals self-identifying as “Hispanic or Latino” (n = 2,981) or as \n“more than one race” (n = 1,400). Similarly, participants in the predicted European category include \nindividuals self-identifying as “Hispanic or Latino”, “More than one race” (n = 1,400), or “Middle Eastern \nor North African” (n = 510).  \nOverall, subsets of participants classified by race/ethnicity and ancestry prediction are not entirely \nconcordant. Association studies, including GWAS and GWAS meta -analyses, should carefully consider \nthese differences, especially since race/ethnicity may capture important socio -cultural or environmental \neffects. \nGenetic Ancestries and Association with Biological Traits in All of Us \nTo evaluate the impact of ancestry on biological traits, we assessed associations between ancestry clusters \nin All of Us individuals and body mass index (BMI) and height. Given the substantial variability in genetic \nancestry (both nationally and at the state level) within race/ethnicity and the inadequacy of these \ncategories as proxies for genetic ancestry, we did not stratify individuals by race/ethnicity in our \nassociation analyses. Instead, we included race/ethnicity alongside comprehensive social and \nenvironmental covariates to adjust the association models. To mitigate the effects of data sparsity, we \nperformed the association analyses including individuals with at least 10% of the corresponding \ncontinental ancestry (or the combined subcontinental ancestries within that continent; Table S4 and S5).  \n \nFirst, we evaluated the association between genetic ancestry and height (Table S4), a trait that varies \nsignificantly across populations and has a large narrow-sense heritability (> 80%)\n41. In Europe, for example, \nheight follows a pronounced south -to-north gradient, with taller statures more commonly observed in \nnorthern regions41. Consistent with previous studies33,42, North European ancestry was strongly associated \nwith greater height (β = 0.107, SE = 0.004, p-value = 4.35 × 10–204 ) whereas both South European Ancestry \nand Middle Eastern ancestry were associated with lower height (β = -0.028, SE = 0.004, p-value = 2.21 × 10–\n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n14 ). West-Central African ancestry was associated with greater height (β = 0.045, SE = 0.004, p-value = \n4.35 × 10–29 ) while Hunter-Gatherer African ancestry was associated with lower height (β = -0.196, SE = \n0.043, p-value = 5.22 × 10 –6 ). West and East African ancestries were not significantly associated with \nheight. Both Indian, South Asian, and Southeast Asian ancestries were significantly associated with lower \nheight whereas North Asian ancestry was associated with increasing height. In line with previous reports43, \nNative American ancestry was strongly associated with lower height (β = -0.088, SE = 0.002, p-value = \n1.40 × 10–274 ).  \n \nNext, we evaluated the association between ancestry and BMI (Fig. 10 and Tables S5). We found that \nAsian and European ancestry clusters were significantly associated with lower BMI, whereas Native \nAmerican ancestry was associated with higher larger BMI (β = 0.009, SE = 0.003, p-value = 1.56 × 10\n–3). \nNotably, West-Central African ancestry was associated with higher BMI (β = 0.019, SE = 0.005, p-value = \n1.80 × 10–05) while East African ancestry was associated with lower BMI (β = -0.041, SE = 0.014, p-value = \n3.18 × 10–3). Consistently, BMI polygenic scores (PRS), adjusted to account for differences due to ancestral \nbackground, revealed a positive correlation between West-Central African ancestry and PRS estimates for \nBMI (r² = 0.078, p = 0.0001). In contrast, East African ancestry showed a negative correlation with BMI PRS \nestimates (r² = -0.033, p = 0.10). These results indicate that subcontinental ancestries can have opposite \neffects on biological traits and diseases. BMI polygenic scores (PRS) \n \nOverall, associations between ancestry and traits were attenuated upon the inclusion of socio-cultural or \nenvironmental factors (Fig. 10 and Tables S5 and S7). Furthermore, after adjusting for a set of covariates \nincluding 3 -digit zip codes, the inclusion of both race/ethnicity and country of birth systematically \nimproved model fit. This result suggests that while race and ethnicity should not be used as proxies for \ngenetic population structure, they may capture additional environmental effects not typicall y accounted \nfor by standard covariates in association models. Although race/ethnicity may serve as a proxy for \nenvironmental effects, directly adjusting for more specific environmental factors affecting the outcome is \npreferable when such data are availabl e\n44. Additionally, the country of birth may account for the effects \nof recent migration patterns to the U.S. and should be considered in association models using the All of \nUs dataset. \n \nAs a sensitivity analysis, we tested the associations between ancestry and BMI and height among \nparticipants with at least 50% of the corresponding continental ancestry (or combined subcontinental \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nancestries within that continent; Tables S6 and S7). Most results were consistent with the 10% threshold \napproach. However, including race/ethnicity as a covariate led to a poorer model fit for participants with \nat least 50% Native American ancestry, as the vast majority (99.6%) self-identified as 'Hispanic or Latino'. \n \n \nDISCUSSION \nBy analyzing whole-genome sequencing data on 230,016 unrelated All of Us participants with a new panel \nof global genetic diversity, we conducted the largest population genomics analysis of U.S. samples that \nreflect the nation's genetic diversity.  \n \nTo assess genetic diversity in All of Us  in the context of multicontinental populations, we created two \npanels of genetic diversity:  the multicontinental diversity (1kGP, HGDP) and the global diversity (1kGP, \nHGDP, and SDGP). We employed two complementary approaches: first, an unsupervised PCA of the All of \nUs data, onto which the multicontinental diversity panel was projected; and second, an unsupervised PCA \nof the global diversity panel, onto which the All of Us data were projected. Our findings demonstrate that \nAll of Us not only captures much of the diversity represented in existing reference panels but also bridges \ngaps along the first five axes of genetic diversity within these panels. This is consistent with the fact that \n~50 million people living in the U.S. are migrants from all continents and nearly every country in the world \n45. Our results align with findings from the Bio Me Biobank, a diverse multi -ethnic cohort from New York \nCity (NYC)23, that identified that NYC individuals partially reflect the genetic diversity in populations from \nmultiple countries around the world. This diversity offers important opportunities to improve the \nrepresentation of populations previously excluded from genomics research. \n \nUnlike previous analyses UMAP analyses, which suggested that participants in race/ethnicity categories \nmay be distributed in discrete clusters\n3,4, we found gradients of genetic diversity that cut across those \ncategories, consistent with findings from Bio Me23.  Notably, All of Us “Hispanic or Latino” participants \nspanned nearly the entire spectrum of genetic diversity along the first two principal components. Most \n“Hispanic or Latino” (92%) did not self-identify within any specific race category, and those (8%) who did \nwere represented across all possible pre -defined racial categories. Latin America has a recent history of \nadmixture among multiple Native American, African, and European populations\n17,34,46,47. Furthermore, \nracial descriptors in Latin America are more fluid, whereas the U.S. has a more binary classification of \n\"white\" vs. \"non-white”15,48.  \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n \nOur results build upon findings from previous studies using smaller datasets 17,23,49,50 that African, Native \nAmerican, and European ancestries vary among individuals in the U.S. We further revealed significant \nvariation in subcontinental ancestries, including pronounced regional differences across the country.\n \nThese results highlight the complexity of genetic backgrounds and admixture within these groups and \ndemonstrate that social constructs of race and ethnicity do not accurately reflect underlying genetic \nancestry. Therefore, we do not recommend using race and ethnicity as proxies for ancestry in genetic \nstudies, including association models. Rather, we support the use of race and ethnicity as markers of \nsocial, environmental, and historical factors that influence health outcomes\n9, when more direct \nmeasurements of these markers are not available44. \n \nWe observed that ancestral proportions within race/ethnicity categories vary by state. Most of the \nobserved geographical variation in ancestries within racial and ethnic categories might be attributed to \nthe history of colonization, the transatlantic slave trade, and recent migration patterns in the U.S. Among \nBlack or African American participants, African subcontinental ancestries mirrored the total African \nancestry, especially for West -Central African ancestry, which is the predominant African ancestry i n the \nAmericas\n34. This alignment between total African ancestry and subcontinental patterns is likely shaped by \nthe U.S. slave trade, in which ancestry is traced to West -Central and West African countries such as \nNigeria, Ghana, Benin, Ivory Coast, The Gambia, and Senegal 37. This contrasts with the African ancestry \nprofile in Brazil, where a more diverse array of African ancestries can be traced to regions including West, \nSouth, and East Africa (e.g., Nigeria, Ghana, Angola, and Mozambique)34. An exception to this pattern was \nobserved in South Carolina, where West African ancestry was elevated compared to other states. This \nobservation is consistent with previous reports of a significant Grain Coast ancestry among African \nAmerican men in South Carolina51, and aligns with the historical narrative of rice plantation farmers in \nSouth Carolina preferentially obtaining enslaved Africans from western Africa’s “Grain Coast” ( e.g., \nSenegal and Sierra Leone) 52. “Hispanic or Latino” participants from U.S. Western states exhibited the \nhighest proportion of Native American ancestry compared to other U.S. regions. This pattern aligns with \nthe geographic proximity to Mexico, the historical context of western states (Ca lifornia, Texas, and \nArizona) being part of Mexico until the mid-19th century, and recent migration patterns from Latin \nAmerica. In contrast, “Hispanic or Latino” participants in New York had the highest proportion of African \nancestry, consistent with the recent migration of Caribbeanns\n53, including Puerto Ricans to New York54. In \nagreement, a recent report highlighted differences in continental ancestries among Latin American \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nimmigrant groups in the U.S. 55; for instance, Mexicans exhibited higher proportions of Native American \nancestry, whereas Puerto Ricans showed greater levels of African ancestry. Among All of Us  “White” \nparticipants (Fig. 6), the mean European ancestry was highest in the Northeastern U.S., while lower \nproportions were observed along the East Coast and in Southern states. This pattern is consistent with \nthe settlement preferences of European migrants during the colonial period, influenced by favorable \nclimates more similar to those of their European countries of origin\n56. \nWe demonstrated that classifying All of Us  participants by race/ethnicity or genetic ancestry yields \ngroupings that are not fully concordant. Consequently, using summary statistics from thousands of \nprecomputed GWASs in the All of Us  Researcher Workbench (\"All by All tables\") for replication or meta -\nanalyses in studies designed around race/ethnicity requires careful evaluation. One approach to address \nthis issue is to adopt the same stratification strategy for All of Us  participants as employed in the study  \nbeing replicated or meta-analyzed.\n \nWe found that BMI and height are associated with genetic ancestries, even after adjusting models for key \nsocio-cultural or environmental covariates such as age, sex, race, ethnicity, education, income, ZIP code, \nand country of birth . These results warrant new studies based on ancestry across the thousands of \nphenotypes available in All of Us, including admixture mapping\n57, a powerful gene mapping technique \nthat leverages locus -specific ancestry to identify loci associated with differential risk or trait values by \nancestry. Notably, our findings reveal that West -Central and East African ancestries are associated with \nBMI, but in opposite directions. Differential associations across African ancestries have also been shown \nfor other health-related traits\n58,59. For instance, Meeks et al.60 demonstrated that triglycerides do not exert \nan equally strong association with other established risk factors in West and East Africans, underscoring \nthe complexity of genetic and environmental factors influencing metabolic health. Our results and \nprevious reports highlight the importance of avoiding using African continental ancestry as a single entity. \nGiven this consideration and the well-documented subcontinental ancestries within Africa\n34,61, the \nAmericas34,55, Asia62, and Europe33, along with their substantial regional variation within continents and in \nthe U.S., we recommend treating continental ancestry not as a singular entity but as a composite of \ndiverse subcontinental ancestries. Researchers should use plural forms, such as African, Native American, \nAsian, or European ancestries, and adjust association models to account for genetic diversity components \nthat reflect subcontinental ancestries. \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nIn conclusion, our analyses of the largest and most diverse U.S. biobank demonstrate that genetic diversity \nis characterized by distribution along a small number of gradients. Ancestry varies widely within race and \nethnicity groups, both nationally and at the state level, emphasizing the inadequacy of such descriptors \nfor defining genetically or biologically distinct populations. However, race and ethnicity may serve as \nproxies for capturing socio -cultural or environmental factors ( e.g., racism and social i nequality) that are \nnot typically accounted for by standard covariates in association models. The genetic diversity captured \nin All of Us reflects most of the diversity represented in public genetic reference panels, filling in gaps in \nexisting panels. Furthermore, our findings highlight the importance of accurately characterizing \nsubcontinental ancestries, as reliance on the continental an cestry approach is insufficient to control for \nconfounding in association studies. \nMETHODS \nSamples \nWe analyzed data from 230,016 unrelated All of Us  participants with short-read whole -genome \nsequencing (WGS) available through the Curated Data Repository (CDR, Control Tier Dataset v7) in the All \nof Us  Research Workbench. To establish a global multi -continental genetic diversity reference, we \ncompiled genome-wide data from three multi-continental studies: the harmonized set of  genomes\n31 from \nthe 1000 Genomes Project (1kGP)28 and the Human Genome Diversity Project (HGDP)30 and genomes from \nthe Simons Genome Diversity Project 32 (Fig. 1A and Supplementary Data 1). The SGDP was designed to \nsample a broader range of globally diverse populations, with genomes sequenced from a few individuals \n(2-4) from previously underrepresented populations. We generated two reference datasets: The  first \ncomprising 3,433 unrelated individuals from 80 populations in the most used  genetic diversity panels: \n1kGP and the HGDP datasets (referred to as “multicontinental diversity panel”). The second, more \ncomprehensive dataset combines 1kGP, HGDP, and SGDP, including 3,667 unrelated individuals \nrepresenting 162 populations worldwide (referred to as “global diversity panel”). \n \n \nData Curation \nWe used the Allele Count/Allele Frequency (ACAF) threshold files available in the All of Us Research \nWorkbench. This call set includes approximately 99 million variants that are common in All of Us ancestry \nsubpopulations, with population -specific allele frequencies (maf > 1%) or allele counts > 100. Our \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\npreliminary analysis revealed that current methods for assessing population structure 20,63–65 using the \nentire All of Us set of variants and individuals were computationally inefficient or infeasible. Based on \nguidance from previous studies21,34, we thinned the dataset to approximately two million common SNPs. \nWe performed data cleaning within and between datasets using PLINK 1.966. Variants were filtered based \non minor allele frequency ( --maf 0.01), genotype missingness per variant ( --geno 0.05), and genotype \nmissingness per individual (--mind 0.05). Additionally, single nucleotide polymorphisms (SNPs) in linkage \ndisequilibrium (LD) were pruned (--indep-pairwise 50 10 0.2). \nRelatedness \nFrom 245,388 All of Us WGS participants, we generated a set of 230,016 unrelated samples by pruning a \nset of relatives (available in the All of Us  Workbench67) optimized to maximize the set of unrelated \nsamples. Briefly, relatedness was assessed using the pc_relate function implemented in Hail68 to estimate \nsamples’ pairwise kinship coefficient ( Φij). Pairs with Φij estimates above 0.1 were linked in a network \nframework to identify the largest set of unrelated individuals, minimizing the number of samples requiring \npruning. Similarly, using our global multi -continental reference panel, we applied the KING method 69 to \nestimate Φij.  Genetic relationships were modeled as networks70, where individuals were linked if their Φij \nexceeded a threshold of 0.0884 (indicating first- and second-degree relatives69). Related individuals were \nthen excluded using a maximum clique approach to minimize sample loss70. \n \nPopulation Structure and Genetic Ancestry \nWe performed supervised and unsupervised principal components analysis (PCA) using GCTA\n64 in the \nunrelated All of Us participants following two different approaches. First, to assess population structure \nin All of Us  and identify known reference populations represented by All of Us  diversity, we performed \nunsupervised PCA of All of Us followed by projection of our multi-continental diversity panel. We used ~2 \nmillion SNPs not pruned by LD for this analysis. Second, to assess the coverage of All of Us participants of \nglobal genetic diversity, we performed unsupervised PCA using 535,933 independent variants in our global \ndiversity panel followed by projection of the All of Us samples. We calculated the convex hull areas71 based \non principal components to assess All of Us  coverage of genetic diversity relative to our global diversity \npanel. For computational efficiency, unsupervised PCA on All of Us  participants was performed using 10 \nrandom subsets of the entire dataset.  \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nGenetic ancestry scores currently available in the All of Us  workbench were determined using a trained \nclassifier based on 16 principal components derived from gnomAD reference samples, based on 151,159 \nautosomal SNPs distributed on only two chromosomes. In our analysis to assess ancestry and admixture \nin All of Us, we used 535,933 independent variants distributed across the entire autosome. We performed \nunsupervised ADMIXTURE21 analysis using our global diversity panel to identify the most likely number of \nancestral clusters in our panel. To evaluate the genetic relationship of the inferred ancestry clusters, we \nperformed dendrogram clustering analysis of genetic relationships a mong the estimated ancestry \nclusters, based on estimates of F ST. Using the allele frequencies of the inferred ancestral clusters \n(ADMIXTURE projection analysis21), we projected All of Us  participants onto our global diversity panel to \nestimate their ancestry proportions.  \n \n \nAssociation Analysis \nTo assess the association of ancestry with BMI and height, we included all subcontinental ancestries within \na continent (based on dendrogram F\nST analysis) in the model simultaneously to evaluate the independent \neffect of each ancestry on the trait while controlling for the effect of the others. We performed the \nassociation analyses using four different models. Model 1 tested the association between the biological \ntrait and subcontinental ancestries: trait ~ ancA\n1 + ancA 2 + ... + ancA n, without adjusting for  possible \nconfounders. Model 2 included socio-economic covariates: trait ~ ancA1 + ancA2 + ... + ancAn + age + sex \n+ education + income + zip code . In model 3, considering that All of Us  includes participants not born in \nthe US, we added a covariate for recent immigration: trait ~ ancA 1 + ancA2 + ... + ancA n + age + sex + \neducation + income + zip code + country of birth. In model 4, to account for socio -environmental effects \ncaptured by the U.S. Census categories, we added race and ethnicity: trait ~ ancA 1 + ancA2 + ... + ancAn + \nage + sex + education + income + zip code + country of birth + race + ethnicit y.  To mitigate the effects of \ndata sparsity, we performed the association analyses including individuals with at least 10% of the \ncorresponding continental ancestry (or the combined subcontinental ancestries within that continent). \nFor sensitivity analysis, we repeated the associations analyses, but using participants with at least 50% of \nthe corresponding continental ancestry (or combined subcontinental ancestries within that continent). \n \nPolygenic risk scores (PRS) estimates for BMI \nPolygenic risk scores (PRS) for BMI were calculated for a subset of genetically diverse participants in the \nAll of Us program (n = 13,475), who were previously selected to calibrate PRS estimates within the \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nElectronic Medical Records and Genomics (eMERGE) network72. Briefly, the raw PRS was computed for all \nparticipants by summing the weighted risk alleles derived from previously published GWASs. Genetic \nancestry was inferred based on the projection of participants into PC space. To mitigate biases introduced \nby var iation in allele frequencies and linkage disequilibrium patterns across populations, a calibration \nmodel was developed using genetically diverse data from 13,475 All of Us  participants. This model \nnormalized the mean and variance of the PRS within ancestry groups, enabling the derivation of adjusted \nPRS values. Calibration ensured that PRS distributions were standardized, facilitating direct comparability \nacross individuals irrespective of their continental genetic ancestry. \n \nACKNOWLEDGMENTS \nThis work utilized the computational resources of the NIH HPC Biowulf cluster ( https://hpc.nih.gov). The \ncontents of this publication are solely the responsibility of the authors and do not necessarily represent \nthe official views of the National Institutes of Health. The authors express their gratitude to the staff and \nparticipants of the All of Us Re search Program. We thank Christopher Kachulis  for insightful discussions \non PRS estimates using eMERGE PRS pipeline. \n \nFUNDING \nM.H.G. is supported by the National Human Genome Research Institute K99/R00 Pathway to \nIndependence Award (1K99HG012211-01A1) and K.A.C.M. by the National Institute of Diabetes and \nDigestive Kidney Diseases K99/R00 Pathway to Independence Award (DK131018).  The CRGGH is \nsupported by the National Human Genome Research Institute, the National Institute of Diabetes and \nDigestive and Kidney Diseases, and the Office of the Director at the National Institutes of Health \n(1ZIAHG200362). E.T.-S. received funding from  CNPq-Brazil (Conselho Nacional de Desenvolvimento \nCientífico e Tecnológico) and FAPEMIG (Minas Gerais State Research Agency). \n \nAUTHOR CONTRIBUTIONS  \nThe project was conceived by M.H.G., D.S., C.N.R., and A.A.A. M.H.G. and D.S. assembled datasets. M.H.G. \nand D.S. analyzed genetic data. M.H.G., K.A.C.M., V.B., T.P.L., F.S.G.K., R.M., A.P.D., E.T.S., R.A.K., I.F.M., \nT.D.O’C., D.S., A.A.A., D.S., and C.N.R. contributed to data interpretation. M.H.G., K.A.C.M., A.A.A., D.S., \nand C.N.R. wrote the manuscript. All authors read the manuscripts and contributed with suggestions. \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nREFERENCES \n1. US Census Bureau (2024). Racial and Ethnic Diversity in the United States: 2010 Census and 2020 \nCensus. \n2. All of Us Research Program Investigators, Denny, J.C., Rutter, J.L., Goldstein, D.B., Philippakis, A., \nSmoller, J.W., Jenkins, G., and Dishman, E. (2019). The “All of Us” Research Program. N Engl J Med 381, \n668–676. \n3. All of Us Research Program Genomics Investigators (2024). Genomic data in the All of Us Research \nProgram. Nature 627, 340–346. \n4. Karczewski, K.J., Francioli, L.C., Tiao, G., Cummings, B.B., Alföldi, J., Wang, Q., Collins, R.L., Laricchia, \nK.M., Ganna, A., Birnbaum, D.P., et al. (2020). The mutational constraint spectrum quantified from \nvariation in 141,456 humans. \n5. McInnes, L., Healy, J., Saul, N., and Großberger, L. (2018). UMAP: Uniform Manifold Approximation \nand Projection. J. Open Source Softw. 3, 861. \n6. Kaiser, J. (2024). Huge genome study confronted by concerns over race analysis. Science. \n7. Kozlov, M. (2024). “All of Us” genetics chart stirs unease over controversial depiction of race. Nature. \n8. Becht, E., McInnes, L., Healy, J., Dutertre, C.-A., Kwok, I.W.H., Ng, L.G., Ginhoux, F., and Newell, E.W. \n(2018). Dimensionality reduction for visualizing single-cell data using UMAP. Nat Biotechnol. \n9. National Academies of Sciences, Engineering, and Medicine (2024). Rethinking Race and Ethnicity in \nBiomedical Research. \n10. Read “Using Population Descriptors in Genetics and Genomics Research: A New Framework for an \nEvolving Field” at NAP.edu. \n11. McKeigue, P.M. (1997). Mapping genes underlying ethnic differences in disease risk by linkage \ndisequilibrium in recently admixed populations. Am. J. Hum. Genet. 60, 188–196. \n12. Cornell, S., and Hartmann, D. (2006). Ethnicity and Race: Making Identities in a Changing World \n(SAGE Publications). \n13. US Census Bureau (2024). About the Topic of Race. \n14. Central Intelligence Agency (2023). The CIA world factbook 2023-2024 (New York, NY: Skyhorse \nPublishing). \n15. Telles, E.E. (2014). Race in Another America: The Significance of Skin Color in Brazil (Princeton \nUniversity Press). \n16. Stronks, K., Kulu-Glasgow, I., and Agyemang, C. (2009). The utility of “country of birth” for the \nclassification of ethnic groups in health research: the Dutch experience. Ethn Health 14, 255–269. \n17. Bryc, K., Durand, E.Y., Macpherson, J.M., Reich, D., and Mountain, J.L. (2015). The genetic ancestry of \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nAfrican Americans, Latinos, and European Americans across the United States. Am J Hum Genet 96, 37–\n53. \n18. Tang, H., Quertermous, T., Rodriguez, B., Kardia, S.L.R., Zhu, X., Brown, A., Pankow, J.S., Province, \nM.A., Hunt, S.C., Boerwinkle, E., et al. (2005). Genetic structure, self-identified race/ethnicity, and \nconfounding in case-control association studies. Am J Hum Genet 76, 268–275. \n19. Lima-Costa, M.F., Rodrigues, L.C., Barreto, M.L., Gouveia, M., Horta, B.L., Mambrini, J., Kehdy, F.S.G., \nPereira, A., Rodrigues-Soares, F., Victora, C.G., et al. (2015). Genomic ancestry and ethnoracial self-\nclassification based on 5,871 community-dwelling Brazilians (The Epigen Initiative). Sci. Rep. 5, 9812. \n20. Price, A.L., Patterson, N.J., Plenge, R.M., Weinblatt, M.E., Shadick, N.A., and Reich, D. (2006). \nPrincipal components analysis corrects for stratification in genome-wide association studies. Nat. Genet. \n38, 904–909. \n21. Alexander, D.H., Novembre, J., and Lange, K. (2009). Fast model-based estimation of ancestry in \nunrelated individuals. Genome Res. 19, 1655–1664. \n22. Wright, S. (1931). Evolution in Mendelian Populations. Genetics 16, 97–159. \n23. Belbin, G.M., Cullina, S., Wenric, S., Soper, E.R., Glicksberg, B.S., Torre, D., Moscati, A., Wojcik, G.L., \nShemirani, R., Beckmann, N.D., et al. (2021). Toward a fine-scale population health monitoring system. \nCell 184, 2068–2083.e11. \n24. Bastos-Rodrigues, L., Pimenta, J.R., and Pena, S.D.J. (2006). The genetic structure of human \npopulations studied through short insertion-deletion polymorphisms. Ann. Hum. Genet. 70, 658–665. \n25. Barbujani, G., Magagni, A., Minch, E., and Cavalli-Sforza, L.L. (1997). An apportionment of human \nDNA diversity. Proc. Natl. Acad. Sci. U. S. A. 94, 4516–4519. \n26. Rosenberg, N.A., Pritchard, J.K., Weber, J.L., Cann, H.M., Kidd, K.K., Zhivotovsky, L.A., and Feldman, \nM.W. (2002). Genetic structure of human populations. Science 298, 2381–2385. \n27. Bhatia, G., Patterson, N., Sankararaman, S., and Price, A.L. (2013). Estimating and interpreting FST: \nthe impact of rare variants. Genome Res. 23, 1514–1521. \n28. 1000 Genomes Project Consortium, Auton, A., Brooks, L.D., Durbin, R.M., Garrison, E.P., Kang, H.M., \nKorbel, J.O., Marchini, J.L., McCarthy, S., McVean, G.A., et al. (2015). A global reference for human \ngenetic variation. Nature 526, 68–74. \n29. Byrska-Bishop, M., Evani, U.S., Zhao, X., Basile, A.O., Abel, H.J., Regier, A.A., Corvelo, A., Clarke, W.E., \nMusunuri, R., Nagulapalli, K., et al. (2022). High-coverage whole-genome sequencing of the expanded \n1000 Genomes Project cohort including 602 trios. Cell 185, 3426–3440.e19. \n30. Bergström, A., McCarthy, S.A., Hui, R., Almarri, M.A., Ayub, Q., Danecek, P., Chen, Y., Felkel, S., \nHallast, P., Kamm, J., et al. (2020). Insights into human genetic variation and population history from 929 \ndiverse genomes. Science 367,. \n31. Koenig, Z., Yohannes, M.T., Nkambule, L.L., Zhao, X., Goodrich, J.K., Kim, H.A., Wilson, M.W., Tiao, G., \nHao, S.P., Sahakian, N., et al. (2024). A harmonized public resource of deeply sequenced diverse human \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\ngenomes. bioRxiv. \n32. Mallick, S., Li, H., Lipson, M., Mathieson, I., Gymrek, M., Racimo, F., Zhao, M., Chennagiri, N., \nNordenfelt, S., Tandon, A., et al. (2016). The Simons Genome Diversity Project: 300 genomes from 142 \ndiverse populations. Nature 538, 201–206. \n33. Gouveia, M.H., Bentley, A.R., Leal, T.P., Tarazona-Santos, E., Bustamante, C.D., Adeyemo, A.A., \nRotimi, C.N., and Shriner, D. (2023). Unappreciated subcontinental admixture in Europeans and \nEuropean Americans and implications for genetic epidemiology studies. Nat. Commun. 14, 6802. \n34. Gouveia, M.H., Borda, V., Leal, T.P., Moreira, R.G., Bergen, A.W., Kehdy, F.S.G., Alvim, I., Aquino, \nM.M., Araujo, G.S., Araujo, N.M., et al. (2020). Origins, admixture dynamics and homogenization of the \nAfrican gene pool in the Americas. Mol. Biol. Evol. \n35. Micheletti, S.J., Bryc, K., Ancona Esselmann, S.G., Freyman, W.A., Moreno, M.E., Poznik, G.D., Shastri, \nA.J., 23andMe Research Team, Beleza, S., and Mountain, J.L. (2020). Genetic Consequences of the \nTransatlantic Slave Trade in the Americas. Am J Hum Genet 107, 265–277. \n36. Shriner, D. (2018). Re-analysis of Whole Genome Sequence Data From 279 Ancient Eurasians Reveals \nSubstantial Ancestral Heterogeneity. Front. Genet. 9, 268. \n37. Slave Voyages. https://www.slavevoyages.org/ \n38. (1989). The Atherosclerosis Risk in Communities (ARIC) Study: design and objectives. The ARIC \ninvestigators. Am. J. Epidemiol. 129, 687–702. \n39. Tsao, C.W., and Vasan, R.S. (2015). Cohort Profile: The Framingham Heart Study (FHS): overview of \nmilestones in cardiovascular epidemiology. Int J Epidemiol 44, 1800–1813. \n40. Bild, D.E., Bluemke, D.A., Burke, G.L., Detrano, R., Diez Roux, A.V., Folsom, A.R., Greenland, P., Jacob, \nD.R., Jr, Kronmal, R., Liu, K., et al. (2002). Multi-Ethnic Study of Atherosclerosis: objectives and design. \nAm. J. Epidemiol. 156, 871–881. \n41. Silventoinen, K., Sammalisto, S., Perola, M., Boomsma, D.I., Cornes, B.K., Davis, C., Dunkel, L., De \nLange, M., Harris, J.R., Hjelmborg, J.V.B., et al. (2003). Heritability of adult body height: a comparative \nstudy of twin cohorts in eight countries. Twin Res. 6, 399–408. \n42. Campbell, C.D., Ogburn, E.L., Lunetta, K.L., Lyon, H.N., Freedman, M.L., Groop, L.C., Altshuler, D., \nArdlie, K.G., and Hirschhorn, J.N. (2005). Demonstrating stratification in a European American \npopulation. Nat. Genet. 37, 868–872. \n43. Asgari, S., Luo, Y., Akbari, A., Belbin, G.M., Li, X., Harris, D.N., Selig, M., Bartell, E., Calderon, R., \nSlowikowski, K., et al. (2020). A positively selected FBN1 missense variant reduces height in Peruvian \nindividuals. Nature 582, 234–239. \n44. Rotimi, C.N. (2023). 2022 ASHG presidential address-One human race: Billions of genomes. Am J \nHum Genet 110, 398–401. \n45. Geiger, A. (2024). What the data says about immigrants in the U.S. \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n46. Kehdy, F.S.G., Gouveia, M.H., Machado, M., Magalhães, W.C.S., Horimoto, A.R., Horta, B.L., Moreira, \nR.G., Leal, T.P., Scliar, M.O., Soares-Souza, G.B., et al. (2015). Origin and dynamics of admixture in \nBrazilians and its effect on the pattern of deleterious mutations. Proceedings of the National Academy \nof Sciences 112, 8696–8701. \n47. Moreno-Estrada, A., Gignoux, C.R., Fernández-López, J.C., Zakharia, F., Sikora, M., Contreras, A.V., \nAcuña-Alonzo, V., Sandoval, K., Eng, C., Romero-Hidalgo, S., et al. (2014). Human genetics. The genetics \nof Mexico recapitulates Native American substructure and affects biomedical traits. Science 344, 1280–\n1285. \n48. Winant, H. (2001). The world is a ghetto : race and democracy since World War II (1st ed. New York : \nBasic Books, 2001.). \n49. Parra, E.J., Marcini, A., Akey, J., Martinson, J., Batzer, M.A., Cooper, R., Forrester, T., Allison, D.B., \nDeka, R., Ferrell, R.E., et al. (1998). Estimating African American admixture proportions by use of \npopulation-specific alleles. Am J Hum Genet 63, 1839–1851. \n50. Bryc, K., Auton, A., Nelson, M.R., Oksenberg, J.R., Hauser, S.L., Williams, S., Froment, A., Bodo, J.-M., \nWambebe, C., Tishkoff, S.A., et al. (2010). Genome-wide patterns of population structure and admixture \nin West Africans and African Americans. Proc. Natl. Acad. Sci. U. S. A. 107, 786–791. \n51. Torres, J.B., Doura, M.B., Keita, S.O.Y., and Kittles, R.A. (2012). Y chromosome lineages in men of \nwest African descent. PLoS One 7, e29687. \n52. Carney, J.A. (2009). Black Rice: The African Origins of Rice Cultivation in the Americas (Harvard \nUniversity Press). \n53. Young Scholars of PS Caribbean Immigrants in Brooklyn. \n54. Acosta-Belén, E., and Santiago, C.E. (2018). Puerto Ricans in the United States: A Contemporary \nPortrait (Lynne Rienner Publishers). \n55. Borda, V., Loesch, D.P., Guo, B., Laboulaye, R., Veliz-Otani, D., French, J.N., Leal, T.P., Gogarten, S.M., \nIkpe, S., Gouveia, M.H., et al. (2024). Genetics of Latin American Diversity Project: Insights into \npopulation genetics and association studies in admixed groups in the Americas. Cell Genom 4, 100692. \n56. Meinig, D.W. (1986). The Shaping of America: A Geographical Perspective on 500 Years of History: \nVolume 1: Atlantic America 1492-1800 (Yale University Press). \n57. Shriner, D. (2017). Overview of Admixture Mapping. Curr. Protoc. Hum. Genet. 94, 1.23.1–1.23.8. \n58. Rotimi, C.N., and Jorde, L.B. (2010). Ancestry and disease in the age of genomic medicine. N. Engl. J. \nMed. 363, 1551–1558. \n59. Nolan, D., Gaudieri, S., and Mallal, S. (2003). Pharmacogenetics: a practical role in predicting \nantiretroviral drug toxicity? J. HIV Ther. 8, 36–41. \n60. Meeks, K.A.C., Bentley, A.R., Agyemang, C., Galenkamp, H., van den Born, B.-J.H., Hanssen, N.M.J., \nDoumatey, A.P., Adeyemo, A.A., and Rotimi, C.N. (2023). Ancestral and environmental patterns in the \nassociation between triglycerides and other cardiometabolic risk factors. EBioMedicine 91, 104548. \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n61. Tishkoff, S.A., Reed, F.A., Friedlaender, F.R., Ehret, C., Ranciaro, A., Froment, A., Hirbo, J.B., \nAwomoyi, A.A., Bodo, J.-M., Doumbo, O., et al. (2009). The genetic structure and history of Africans and \nAfrican Americans. Science 324, 1035–1044. \n62. Baker, J.L., Rotimi, C.N., and Shriner, D. (2017). Human ancestry correlates with language and reveals \nthat race is not an objective genomic classifier. Sci. Rep. 7, 1572. \n63. Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M.A.R., Bender, D., Maller, J., Sklar, P., de \nBakker, P.I.W., Daly, M.J., et al. (2007). PLINK: a tool set for whole-genome association and population-\nbased linkage analyses. Am. J. Hum. Genet. 81, 559–575. \n64. Yang, J., Lee, S.H., Goddard, M.E., and Visscher, P.M. (2011). GCTA: a tool for genome-wide complex \ntrait analysis. Am. J. Hum. Genet. 88, 76–82. \n65. Alexander, D.H., and Lange, K. (2011). Enhancements to the ADMIXTURE algorithm for individual \nancestry estimation. BMC Bioinformatics 12, 246. \n66. Purcell, S., Todd-Brown, N.B., Thomas, K., Ferreira, L., Bender, M., Maller, D., De Bakker, J., Daly, P., \nand Sham, M.J. PLINK: a toolset for whole-genome association and population-based linkage analysis. \nAmerican Journal of Human Genetics. \n67. All of US Researcher Workbench. https://www.researchallofus.org/data-tools/workbench/ \n68. Hail Team. Hail 0.2. https://github.com/hail-is/hail. \n69. Manichaikul, A., Mychaleckyj, J.C., Rich, S.S., Daly, K., Sale, M., and Chen, W.-M. (2010). Robust \nrelationship inference in genome-wide association studies. Bioinformatics 26, 2867–2873. \n70. Leal, T.P., Furlan, V.C., Gouveia, M.H., Duarte, J.M.S., Fonseca, P.A.S., Tou, R., de Oliveira Scliar, M., \nde Araujo, G.S., Zolini, C., Peixoto, M.G.C., et al. (2021). NAToRA, a relatedness-pruning method to \nminimize the loss of dataset size in genetic and omics analyses. \n71. Bivand, R.S., Pebesma, E., and Gómez-Rubio, V. (2013). Applied Spatial Data Analysis with R (Springer \nScience & Business Media). \n72. Lennon, N.J., Kottyan, L.C., Kachulis, C., Abul-Husn, N.S., Arias, J., Belbin, G., Below, J.E., Berndt, S.I., \nChung, W.K., Cimino, J.J., et al. (2024). Selection, optimization and validation of ten chronic disease \npolygenic risk scores for clinical implementation in diverse US populations. Nat. Med. 30, 480– 487. \n \n \n \n \n \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nFigures \n \nFig. 1. Genetic diversity in All of Us and the overlap between self-identified U.S. race and \nethnicity. A, B) The first two principal components (PCs 1 and 2) of All of Us participants with \nrace (A) and ethnicity (B) categories. C) Venn diagram showing the overlap between All of Us \nparticipants who self-identified with one race and as “Hispanic or Latino”.  \n \n \n \n \n \n \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n \nFig. 2. The breadth of genetic diversity in All of Us in the context of global genetic diversity. \nA, B) The first two principal components (PCs 1 and 2) in All of Us  participants alongside our \nreference genetic diversity panel. C, D) The third and fourth components (PCs 3 and 4). E, F) The \nfifth and sixth components (PCs 5 and 6). The inset highlights that SNP loadings on PC 6 reflect \ngenetic differentiation driven by the MHC locus, rather than genome-wide differentiation. 1kGP = \n1000 Genomes Project, HGDP = Human Genome Diversity Project. Race/ethnicity categories are \nintegrated into the same PCA representation for visualization. The black circles represent genetic \ndiversity in All of Us that is not captured by the reference panels, while the red circles represent \ngenetic diversity within the reference panels that is absent in All of Us. \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\nFig. 3. Coverage of multicontinental genetic diversity by All of Us.  A) Population labels and \nB) the first two principal components (PCs 1 and 2) in our panel of multicontinental genetic \ndiversity. C, D) The first two principal components (PCs 1 and 2) in All of Us participants alongside \nour reference genetic diversity panel. E, F) The third and fourth components (PCs 3 and 4). G, H) \nThe fifth and sixth components (PCs 5 and 6). Convex hull areas in the PC plots represent genetic \ndiversity coverage, with race/ethnicity categories integrated into the same plot. Panels C, E, and \nG highlight All of Us data whereas B, D, F, and H highlight our global diversity panel. \n \n \n \n \n \n \nFig. 4. Continental and subcontinental ancestries in All of Us. A)  Color legend representing \nthe 13 most likely ancestry clusters identified from a global genetic diversity reference panel. B) \nDendrogram of genetic relationships among the 13 ancestry clusters, based on estimates of F\nST. \nC) Bar plot displaying individual ancestry proportions in All of Us participants with self-identified \nrace and ethnicity categories, inferred from ADMIXTURE projection analysis. *More = All of Us  \nparticipants self-identified in more than one race. \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n \nFig. 5. Distribution of continental ancestry proportions in All of Us participants within race \nand ethnicity categories. A)  Asian, B) Black or African American, C)  Hispanic or Latino, D)  \n*Middle Eastern = Middle Eastern or North African, E) *Native Hawaiian = Native Hawaiian or \nPacific Islander, and F) White. AFR = African, EUR = European, E_AS = East Asian, S_AS = \nSouth Asian, NAT = Native American, OCE = Oceanian. \n \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n \nFig. 6. The distributions of continental and subcontinental ancestries in self -identified \n“Black or African American” participants by state. A) Total African, B) West-Central African, \nC) West African, D)  East African, E) European, and F)  Native American ancestry proportions. \nStates with less than 50 individuals are excluded in gray. \n \n \n \n \n \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n \n \n \n \n \nFig. 7. The distributions of continental ancestries in self -identified “Hispanic or Latino” \nparticipants by state. A)  European, B) Native American, and C)  African ancestry proportions. \nStates with less than 50 individuals are excluded in gray. \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n \nFig. 8. The distributions of European and Middle Eastern or North African continental and \nsubcontinental ancestries in self -identified “White” participants by state. A)  Total \nEuropean, B) North European, C)  South European, and D)  Middle Eastern or North African \nancestry proportions. States with less than 50 individuals are excluded in gray. \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n \nFig. 9. Distribution of All of Us  participants by race/ethnicity and predicted genetic \nancestry, highlighting overlaps between self-identified race/ethnicity and genetic ancestry \npredictions. A) Violin and box plots comparing African ancestry distribution among participants \nself-identifying as “Black or African American” with those stratified by predicted African genetic \nancestry. B) Violin and box plots comparing European ancestry distribution among participants \nself-identifying as “White” with those stratified by predicted European genetic ancestry. C) Venn \ndiagram illustrating the overlap between predicted African genetic ancestry and self-identified \nrace/ethnicity categories. D) Venn diagram illustrating the overlap between predicted European \ngenetic ancestry and self-identified race/ethnicity categories. \n \n \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint \n\n \nFig. 10. Forest plots showing the association between BMI and multiple traits, accounting \nfor different levels of control of confounders across four different models. WC_AFR = West \nCentral African, W_AFR = West African, E_AFR = East African, S_AFR = South African. All of Us \nparticipants with at least 10% of the combined subcontinental African ancestries were included. \n \n \n \n105 and is also made available for use under a CC0 license. \n(which was not certified by peer review) is the author/funder. This article is a US Government work. It is not subject to copyright under 17 USC \nThe copyright holder for this preprintthis version posted January 10, 2025. ; https://doi.org/10.1101/2025.01.09.632250doi: bioRxiv preprint","source_license":"Public-Domain","license_restricted":false}