Characterizing Race/Ethnicity and Genetic Ancestry for 100,000 Subjects in the Genetic Epidemiology Research on Adult Health and Aging (GERA) Cohort

Abstract

Using genome-wide genotypes, we characterized the genetic structure of 103,006 participants in the Kaiser Permanente Northern California multi-ethnic Genetic Epidemiology Research on Adult Health and Aging Cohort and analyzed the relationship to self-reported race/ethnicity. Participants endorsed any of 23 race/ethnicity/nationality categories, which were collapsed into seven major race/ethnicity groups. By self-report the cohort is 80.8% white and 19.2% minority; 93.8% endorsed a single race/ethnicity group, while 6.2% endorsed two or more. Principal component (PC) and admixture analyses were generally consistent with prior studies. Approximately 17% of subjects had genetic ancestry from more than one continent, and 12% were genetically admixed, considering only nonadjacent geographical origins. Self-reported whites were spread on a continuum along the first two PCs, indicating extensive mixing among European nationalities. Self-identified East Asian nationalities correlated with genetic clustering, consistent with extensive endogamy. Individuals of mixed East Asian–European genetic ancestry were easily identified; we also observed a modest amount of European genetic ancestry in individuals self-identified as Filipinos. Self-reported African Americans and Latinos showed extensive European and African genetic ancestry, and Native American genetic ancestry for the latter. Among 3741 genetically identified parent–child pairs, 93% were concordant for self-reported race/ethnicity; among 2018 genetically identified full-sib pairs, 96% were concordant; the lower rate for parent–child pairs was largely due to intermarriage. The parent–child pairs revealed a trend toward increasing exogamy over time; the presence in the cohort of individuals endorsing multiple race/ethnicity categories creates interesting challenges and future opportunities for genetic epidemiologic studies.

POPULATION genetic structure analyses have recently increased in number due to improvements in capabilities to perform large-scale genomic investigations. Technological developments have improved our ability to address questions associated with phenotypic variation (Wellcome Trust Case Consortium 2007), human genetic variation (Jakobsson et al. 2008; Li et al. 2008), and evolution (Lohmueller et al. 2008). These studies play an important role in a variety of applied settings, including genome-wide association studies (GWAS), admixture analyses, and dissection of traits associated with ancestry. For example, in association studies, error rates due to confounding by ancestry can be improved when population structure is taken into account (Tian et al. 2008a). At the same time, the relationship between self-identified race/ethnicity/nationality and genetic ancestry based on genetic marker data has become a topic of great interest (Risch et al. 2002; Burchard et al. 2003; Cooper et al. 2003).

Studies of human evolution have typically focused on indigenous population samples broadly distributed geographically across the globe. One such resource that has been highly exploited for this purpose is the Human Genome Diversity Project panel of 55 indigenous populations (Jakobsson et al. 2008; Li et al. 2008). On the other hand, GWAS utilizing U.S.-based samples often include more heterogeneous populations in terms of ancestry, although the number of ethnic groups included is typically limited.

In the present study we utilize the large, ethnically diverse Kaiser Permanente (KP) Research Program on Genes, Environment, and Health (RPGEH) Genetic Epidemiology Research on Adult Health and Aging (GERA) cohort to examine the question of genetic ancestry in a representative northern California population and how it relates to racial/ethnic self-identification. The cohort consists of 103,006 adult members of Kaiser Permanente Northern California (KPNC), ranging in age from 18 to 100 years at enrollment. The cohort was created to enable studies of genetic and environmental influences on many different health conditions and traits by linking high-density genome-wide SNP data with comprehensive longitudinal clinical information from electronic health records (EHR) as well as self-reported data on demographic factors and health behaviors from a structured survey. The GERA cohort is one of the first very large multi-ethnic cohorts created for GWAS for a wide variety of health conditions. The cohort was genotyped using custom ancestry-specific SNP arrays to better capture rare variants specific to different ethnic groups and provide better genome-wide coverage, thus permitting investigation of potential associations that may differ between groups. Understanding and characterizing the genetic diversity within a sample is essential to GWAS, since population structure both within and between groups can lead to artifactual associations. The multi-ethnic GERA cohort thus provides an unprecedented opportunity to understand human genetic diversity in a U.S. population sample. This article presents the results of analyses of population genetic structure, confirming previous observations, but also adding further understanding of mixed genetic ancestry, including the extent of distant vs. recent admixture. We also provide estimates of principal components needed for adjustment of population structure in GWAS and examine the self-reported race/ethnicity distribution of first-degree relative (parent–child and full sib) and monozygotic (MZ) twin pairs. Finally, we examine how the identified genetic structure correlates with participants’ self-identification in terms of race/ethnicity/nationality.

Materials and Methods

Participants

Individuals composing the GERA cohort are participants in the KPNC RPGEH. KPNC is an integrated health care delivery system with >3 million members in northern California. The membership is representative of the general population with respect to race/ethnicity and socioeconomic status, although extremes of income are under-represented (Krieger et al. 1993). The RPGEH was established as a resource for research on genetic and environmental influences on health and disease. The development of the RPGEH and GERA cohort are described elsewhere (dbGaP phs000674.v1.p1). Briefly, adult members of KPNC were asked to complete a mailed survey; survey respondents then completed a broad written consent and provided a saliva sample for extraction of DNA. Participants self-reported their race, ethnicity, and nationality on the survey by endorsing as many of 23 race, ethnicity, and nationality categories as applied (Table 1 provides a list of the choices). Participants were asked their religion, and this question in conjunction with a race/ethnicity/nationality question was used to identify Ashkenazi individuals (those who responded “Ashkenazi Jewish” to the nationality question or “Jewish” to the religion question).

Table 1. Distribution of responses to survey question on race/ethnicity/nationality along with proportion female and average ages.

	Category	No.	% female	Mean age (SE)
1	African American	3,117	0.57	60.66 (0.24)
2	African	129	0.43	52.90 (1.43)
3	Afro-Caribbean	119	0.68	56.24 (1.30)
4	Mexican	4,613	0.56	56.67 (0.22)
5	Central-South American	1,034	0.70	55.34 (0.46)
6	Puerto Rican	322	0.69	56.68 (0.83)
7	Cuban	106	0.71	55.41 (1.42)
8	Other Latino/Hispanic	1,545	0.70	57.41 (0.38)
9	South Asian–Indian/Pakistani	575	0.42	54.58 (0.60)
10	Chinese	3,433	0.58	56.75 (0.25)
11	Japanese	1,739	0.61	61.56 (0.34)
12	Korean	234	0.66	53.83 (1.04)
13	Filipino	1,708	0.59	55.59 (0.37)
14	Vietnamese	317	0.50	53.23 (0.82)
15	Other Southeast Asia	176	0.64	51.85 (1.10)
16	Native Hawaiian	144	0.65	58.41 (1.23)
17	Samoan	14	0.64	59.36 (3.44)
18	Other Pacific Islander	132	0.57	53.88 (1.35)
19	Native American Indian/Alaska Native	3,884	0.66	61.20 (0.22)
20	White European American	80,079	0.59	63.27 (0.05)
21	Middle Easterner	914	0.43	62.18 (0.48)
22	Ashkenazi Jewish	2,399	0.66	62.49 (0.28)
23	Other ethnicity	75	0.73	56.53 (1.64)

To maximize the diversity of the sample, the GERA cohort was formed by including all racial and ethnic minority participants with saliva samples (19% of the total); the remaining participants were drawn randomly from white non-Hispanic participants (81% of the total). Among cohort members, the average length of KPNC health plan membership was >23 years, providing extensive longitudinal data on diagnoses and procedures, laboratory test results, pharmaceutical prescriptions, radiological findings, and other clinical information from EHR for use in GWAS of health conditions and traits.

The Human Genome Diversity Project (HGDP) (Cavalli-Sforza 2005; Li et al. 2008) subjects were used to facilitate geographic interpretation of the GERA principal components.

Self-reported race/ethnicity

Self-reported race/ethnicity for each individual was derived from responses to the survey question on race/ethnicity/nationality (Table 1). Nationalities within a single race/ethnicity group were collapsed. Specifically, all East Asian nationalities (codes 10–15) were collapsed into a single East Asian group; all Pacific Islander nationalities (codes 16–18) were collapsed into a single Pacific Islander group; all Latino nationalities (codes 4–8) were collapsed into a single Latino category; all African descent populations (codes 1–3) were collapsed into a single group; all white-European ethnicities (codes 20–22) were collapsed into a single category; the single categories of South Asians and Native Americans remained as such. A small number of individuals (<1%) had implausible race/ethnicity responses from the survey (e.g., checked off every category) or specified “other.” For these individuals, we used KPNC administrative databases to assign race/ethnicity. For other individuals, a discrepancy was observed between their original and scanned survey responses. These subjects were also adjudicated to their original form results as described in SI Methods (File S1).

Genotyping and array assignment

To maximize genome-wide coverage of common and less common variants, four custom Affymetrix Axiom arrays (Hoffmann et al. 2011a,b) were designed for individuals of non-Hispanic white (EUR), East Asian (EAS), African American (AFR), and Latino (LAT) race/ethnicity. The number of SNPs varied among arrays, ranging from 674,518 on the EUR array to 893,631 on the AFR array (Hoffmann et al. 2011b). A total of 254,438 SNPs were common to all four arrays. Genotyping was performed at the University of California, San Francisco and is described elsewhere (Kvale et al. 2015).

The assignment of subjects to arrays was based on the race/ethnicity categories formed as described above (Table S2 and Table S3). Assignments were hierarchical to accommodate individuals reporting multiple racial/ethnic categories. Specifically, individuals reporting any Latino or Native American race/ethnicity/nationality (possibly in combination with other races/ethnicities/nationalities) were assigned to the LAT array, with the exception of individuals who reported African/African American race/ethnicity and Native American race/ethnicity, who were assigned to the AFR array, and individuals reporting East Asian race/ethnicity and Native American race/ethnicity, who were assigned to the EAS array. All other individuals reporting any African, African American, or Afro-Caribbean race/ethnicity but no Latino race/ethnicity were assigned to the AFR array. All those reporting any East Asian but not African, African American, Afro-Caribbean, or Latino race/ethnicity were assigned to the EAS array. Subjects reporting white-European American, South Asian, Middle Eastern, or Ashkenazi race/ethnicity, but none of the previously mentioned races/ethnicities, were assigned to the EUR array. Therefore, for example, individuals with European and East Asian race/ethnicity were assigned to the EAS array; individuals with African American and East Asian race/ethnicity/nationality were analyzed on the AFR array. The various arrays were designed to allow for the relevant admixture (Hoffmann et al. 2011b).

Quality control

High-quality genotype data for the GERA cohort was obtained by systematic examination and removal of SNP genotypes according to a specific protocol, as described in detail elsewhere (Kvale et al. 2015). For the genetic structure analyses, only SNPs that were common across all four arrays and that had a call rate >99.5% were considered. This set also excluded SNPs that showed extreme deviation from Hardy–Weinberg equilibrium (P < 10⁻⁵). This resulted in a set of 144,799 high-performing SNPs used in further analyses of population structure and admixture.

Principal components analysis

Filtering:

Principal components analysis (PCA) was performed using the smartpca program, which is part of the EIGENSOFT4.2 software package (Patterson et al. 2006). The initial PCA runs were performed separately for individuals genotyped on different arrays. The initial set of 144,799 high-performing SNPs (described above) that were common across all four array types was used in the preliminary analyses. When the HGDP samples were included in subsequent runs and projected onto the GERA principal components (PCs) to facilitate geographic interpretation, 43,988 high-performing SNPs were used. Initial analyses revealed that a number of individuals appeared to be discordant between their genetic ancestry and the array to which they were assigned, and the PCA was re-run after reclassifying these individuals (see SI Methods, File S1).

PC projection approach:

PCA requires the inversion of a data matrix, which for very large data sets may be computationally challenging. For the East Asian, African American, and Latino subgroups in the GERA data set, the sample sizes were small enough so that all subjects within each subgroup were run together. For example, all 7520 East Asian subjects were run together in one PCA. The white-European American sample, however, is very large and required inverting an ∼80,000 by 80,000 (6.4 billion elements) matrix. Furthermore, the version of the Smartpca program used at the time of analyses was not able to analyze the entire European ancestry sample of >83,000 individuals. Therefore, our approach was to select a large but manageable number of subjects on which to perform an initial PCA and then use the resulting SNP loadings to project the remaining subjects.

Because we planned to select a random subset of 20,000 individuals for the initial PCA on which the remaining subjects would be projected, we examined the effect of using different subsets by calculating the correlations of the SNP loadings for three different random subsets (Supporting Information, Table S1). The numbers of subjects in the three subsets were the following: 18,677 for set 1; 20,121 for set 2; and 17,691 for set 3. For the first six PCs there was very good correlation of the SNP loadings for all three pairs of subsets, also suggesting that most of the signal regarding genetic structure is derived from the first six PCs. Given these results, we selected a random set of 20,000 European ancestry subjects and projected the remaining subjects onto the PCs obtained.

Since the SNPs used for the PCA and admixture estimation were common among all four genotyping arrays it was possible to produce “global” PCA scores for the GERA subjects. Subsets of individuals from the EUR (15,500) AFR (3100), EAS (5600), and LAT (3000) arrays were used for the initial PCA, and the remaining subjects were projected onto these PCs to obtain PC scores for each individual. Table S4 shows the number of SNPs remaining after LD and structural variation loci pruning for each of the eight different PCA runs (File S1).

Genetic ancestry/admixture estimation

To determine individual ancestral admixture proportions in admixed subjects such as African Americans and Latinos (and others), the full maximum-likelihood software package frappe (Tang et al. 2005) was used. In this analysis, individual ancestry proportions are estimated by calculating the probability of a set of genome-wide genotypes in an individual as a weighted average of allele frequencies of putative ancestors, where the weights represent the admixture proportions. In general, the same HGDP population samples described above were used to derive allele frequencies for the ancestral groups.

Relationship determination

Relationships were determined using the software KING_v1.4 (Manichaikul et al. 2010) with the robust version that allows for population substructure. KING provides standard thresholds for characterizing monozygotic twin, parent–child, and sibling relationships, which we followed. In our data, these relationships were clearly separated into distinct clusters. All subjects were included irrespective of the array type used for their analysis. This analysis was based on the 144,799 high-performing SNPs common across the four arrays described above.

Results

Distribution of race/ethnicity/nationality categories reported

This multi-ethnic cohort includes representation from a broad distribution of races/ethnicities/nationalities (Table 1). For individuals who reported more than one category, all categories are included; hence, the numbers in Table 1 sum to greater than 103,006, the total cohort size. All of the major continents are represented and many nationalities/ethnicities. Collapsing the selections into race/ethnicity categories (see Materials and Methods), of the 106,733 total selections, 3365 (3.2%) include an African/African American race/ethnicity; 7620 (7.1%) include a Latino race/ethnicity; 575 (0.5%) include South Asian race/ethnicity; 7607 (7.1%) include an East Asian race/ethnicity; 290 include a Pacific Islander race/ethnicity (0.3%); 3884 (3.6%) include Native American race/ethnicity; and 83,392 (78.1%) include a white-European race/ethnicity. The majority of those endorsing a Latino race/ethnicity are Mexican and Central American, while the largest groups endorsing an East Asian race/ethnicity are Chinese, Japanese, and Filipino. We also examined the sex and age distributions across the different categories (Table 1). Compared to those reporting white-European race/ethnicity, those endorsing African/Afro-Caribbean, Latino, East Asian, and Pacific Islander race/ethnicity are younger; with the exception of those reporting Mexican nationality, the Latino groups tend to have a higher proportion of females, as do those reporting Ashkenazi Jewish ethnicity; those reporting South Asian and Middle Eastern nationalities have a lower proportion of females.

Structure of individuals run on the EUR array

Individuals who self-reported Ashkenazi, Middle Eastern, and non-Hispanic white or European race/ethnicity but no other ethnicities were run on the EUR array and analyzed together. The initial analysis showed, as expected, a clear Ashkenazi cluster and a larger cluster depicting the northwest–southeast European cline (Price et al. 2008; Tian et al. 2008c). Figure S1A shows those who self-reported a single ethnicity/nationality, while Figure S1B shows individuals who self-reported more than one. It is evident that endorsement of more than one ethnicity can imply mixed genetic ancestry but not automatically. Comparing Figure S1, A and B, we observe a higher proportion of individuals with mixed genetic ancestry among those who endorsed both Ashkenazi and European or Middle Eastern ethnicity; however, we still observe a large proportion of nonadmixed individuals, suggesting that endorsement of Ashkenazi and European may reflect a joint perception of ethnicity and continent of origin. By contrast, in Figure S1A we observe a substantial number of individuals who appear to have Ashkenazi and European admixture but self-reported a single category only (most often European).

A similar observation can be made about those endorsing Middle Eastern ethnicity, where those endorsing that as a sole response appear to have more Middle Eastern genetic ancestry, while those endorsing Middle Eastern and European ethnicity show more evidence of European genetic ancestry. However, in Figure S1A we also observe substantial numbers of individuals reporting only European ethnicity whose genetic ancestry appears to be Middle Eastern and vice versa. Again, these reports may reflect recent geographic origin as well as nationality/ethnicity.

We also repeated the PC analysis after removing the Ashkenazi and part-Ashkenazi subjects. The PC scores for the Ashkenazi subjects were then derived by projecting their genotypes onto the resulting PCs. Individuals reporting a single ethnicity/nationality are depicted in Figure S2A, while those endorsing more than one are displayed in Figure S2B. The first PC corresponds to a northwest–southeast cline through Europe and the Middle East and the second PC corresponds to a southwest–northeast cline within Europe, as has been observed in numerous previous studies (Menozzi et al. 1978; Sokal et al. 1991; Cavalli-Sforza et al. 1993; Cavalli-Sforza et al. 1996; Barbujani and Bertorelle 2001; Belle et al. 2006; Seldin et al. 2006; Bauchet et al. 2007; Novembre et al. 2008; Price et al. 2008; Tian et al. 2008c). The first and second PCs account for 31.9 and 13.4% of the total variance of the first 10 PCs, respectively.

Subjects who self-identified as South Asian were also run on the EUR array and subjected to a separate PCA. For these subjects, to characterize the observed PCs and the relationship to geographic ancestry, we employed onomastics. In particular, we analyzed surnames to characterize individuals based on surname geographic region of origin. These subjects are mainly of Indian origin, and the clusters formed in the PCA depict subgroups from different regions of India (Figure S3). The first PC accounts for 19.1% of the total variance of the first 10 PCs and the second PC accounts for 10.0%. The analysis also shows that northern Indians are genetically closer to Europeans (Reich et al. 2009) and eastern Indians are genetically more similar to East Asian populations. As expected, those reporting European as well as South Asian ethnicity are positioned closer in the diagram to the HGDP Europeans.

Structure of individuals run on the EAS array

Individuals run on the EAS array included subjects self-reporting European and East Asian race/ethnicity and those reporting solely East Asian race/ethnicity. The first PC for these individuals (Figure S4, A and B) is responsible for clustering of individuals with different East Asian–European ancestry proportions (mostly 50 or 75% European). Those with genetic ancestry that is both East Asian and European are most clearly observed in Figure S4B, among those self-reporting both races/ethnicities, and there are very few GERA individuals in this figure who do not have mixed genetic ancestry. Among individuals reporting only an East Asian nationality (Figure S4A), the large majority have only East Asian genetic ancestry; however, there are also individuals who appear to have mixed East Asian–European genetic ancestry who self-reported only their East Asian nationality. Of particular interest is the continuous nature of a modest amount of European genetic ancestry in self-identified Filipinos, consistent with older European admixture. The second PC corresponds to the north-to-south cline in East Asia (Su et al. 1999; Tian et al. 2008b; Hugo Pan-Asian Snp Consortium 2009), and the distinct clusters observed that represent different East Asian nationalities are consistent with extensive endogamy in these groups. The first and second PCs account for 59.71 and 20.39% of the total variance of the first 10 PCs, respectively.

Individuals endorsing a Pacific Islander ethnicity are displayed in Figure S5. Those also reporting an East Asian ethnicity appear to cluster more closely to the HGDP East Asians, while those also reporting European ethnicity appear to cluster more closely to the HGDP Europeans. While those reporting Hawaiian and Samoan ethnicity are reasonably well separated from both the HGDP Europeans and East Asians, some individuals who identified as “other Pacific Islander” appear to overlap quite closely with the HGDP East Asians. Also of interest, another subgroup of “other Pacific Islanders” appears to form its own cluster at the bottom of Figure S5. We note that a number of these individuals self-reported both Pacific Islander and South Asian ethnicity. Based on onomastics, these individuals have Indian surnames and are likely to be Indo Fijians. Approximately 37.5% of the population of Fiji is of Indian origin, according to the 2007 census (http://www.statsfiji.gov.fj). The observation that some Pacific Islanders cluster near to the East Asians is also an indication that clear separation of genetic ancestry for these groups is likely to be challenging.

Structure of individuals run on the AFR array

Subjects run on the AFR array revealed, as expected, extensive African and European genetic ancestry (Figure S6, A and B) (Parra et al. 1998; Fernandez et al. 2003; Tang et al. 2006; Tishkoff et al. 2009; Zakharia et al. 2009). The first PC, which accounts for 63.8% of the total variance of the first 10 PCs, reflects African vs. European genetic ancestry, while the second PC denotes East Asian and/or Native American genetic ancestry. This is consistent with the array assignments, whereby individuals reporting both African/African American race/ethnicity and East Asian or Native American race/ethnicity were assigned to the AFR array. Individuals who self-reported African ancestry only were also subject to onomastics to determine likely countries of origin. We were able to identify subjects of Ethiopian, Eritrean, and Kenyan nationality. For the Kenyans, Figure S6A indicates a location consistent with 100% African genetic ancestry. By contrast, the Ethiopian/Eritrean subjects occupy an intermediate position on the PC1 axis, suggesting proximity to European/Middle Eastern populations. Also of note is the modest variation in their PC1 scores. This is likely due to ancient admixture with Middle Eastern populations (Hodgson et al. 2014). These results confirm that Ethiopians have a unique genetic structure among African populations.

Individuals self-reporting mixed African and East Asian race/ethnicity generally reflect that admixture from the genetic perspective as well (Figure S6B); however, a number of individuals who reported only African American ethnicity also appear to have similar levels of East Asian admixture (Figure S6A). Those reporting both African American and European ethnicity generally occupy a position on the PC1 axis closer to Europeans than those who do not (Figure S6B).

The mean African ancestry proportion in this sample is 73.6 ± 17.4%. There is a reasonably high level of variation in the African genetic ancestry proportion, ranging from 10.6 to 100%.

Structure of individuals run on the LAT array

Latinos may have ancestry deriving from multiple continents, including Europe, Africa, Asia, and the Americas (Bonilla et al. 2004; Tang et al. 2006, 2007). Figure S7A provides the PCA results for all those who endorsed Latino or Native American as their sole race/ethnicity. PC1 represents the European vs. Native American axis of genetic variation, and PC2 represents the African axis of genetic variation. PC1 and PC2 account for 70.95 and 11.57% of the total variance of the first 10 PCs, respectively. Nearly all Latinos show evidence of European/West Asian genetic ancestry, and a substantial subset also show evidence of African genetic ancestry. Similarly, all individuals self-reporting Native American race/ethnicity show some degree of European/West Asian genetic ancestry. Latinos of different nationalities exhibit varying proportions of European, African, and Native American ancestries (Figure S7B). Those reporting Mexican and Central-South American nationality have genetic ancestry that is primarily European and Native American, with slight but varying amounts of African ancestry. Those reporting Cuban nationality have primarily European genetic ancestry with a small number of individuals having primarily African genetic ancestry. Those reporting Puerto Rican nationality show some Native American genetic ancestry but are primarily admixed between European and African genetic ancestry. Individual ancestral admixture proportions were determined for these subjects and are provided in Table S5.

The LAT array also included a variety of individuals who self-reported more than one race/ethnicity. These individuals are represented in Figure S7C. Individuals who reported European as well as Latino race/ethnicity tend to have slightly more European genetic ancestry than those who did not; similarly, a number of individuals who reported African/African American race/ethnicity in addition to Latino race/ethnicity have substantial African genetic ancestry; however, many such individuals also appear to have the same modest degree of African genetic ancestry as those who reported only a Latino race/ethnicity. Those who reported Native American race/ethnicity in addition to Latino race/ethnicity also appear to have slightly increased Native American genetic ancestry. Those who reported European and Native American race/ethnicity appear to be similar to those who solely reported Native American race/ethnicity; all have European/West Asian genetic ancestry, and while some show evidence of Native American genetic ancestry, European/West Asian is the sole or primary genetic ancestry for the majority. For those with 100% European genetic ancestry and who self-reported only European and Native American race/ethnicity (n = 2155), we also calculated European PCs. Finally, those who reported East Asian in addition to Latino race/ethnicity generally have evidence of East Asian genetic ancestry (as observed in Figure S7C by proximity to the HGDP East Asians) ranging from 25 to 50 and 100%.

Global PCA for GERA subjects

Figure S8 shows that the first PC mainly separates Europeans from East Asians (and Native Americans) and PC2 separates Africans from all the other groups; PC3 seems to separate Native Americans from the other groups and PC4 also separates Native Americans from the other groups but also shows some separation among the Europeans; PC5 separates the different East Asian groups (mainly north vs. south) and also East Asians from Oceania, and PC6 separates Central–South Asians from the other groups; PC7 again separates the various East Asian regions, and PC8 separates the European groups (mainly north to south); PC9 and PC10 separate East Asians from Oceania, but also the Russians (not labeled) are separated from the other European groups.

Relationship between self-reported race/ethnicity and genetic ancestry

Table S6 displays the full relationship of self-reported race/ethnicity to genetic ancestry for the six continental genetic ancestries of Europe/West Asia, Africa, East Asia, Pacific Islands, South Asia, and the Americas. A genetic continental ancestry was assigned to an individual if her/his estimate for that ancestry was at least 5%. A total of 91,502 individuals (93.9%) reported a single race/ethnicity; 5475 individuals reported two races/ethnicities (5.9%); and 512 individuals (0.5%) reported three races/ethnicities (Table 2). As expected, all individuals who self-identified as European/West Asian had evidence of European/West Asian genetic ancestry. The next largest genetic ancestry component in this group was South Asian (4.3%), primarily attributable to individuals of West Asian ethnicity. Because there is a continuum of genetic ancestry from Europe to West Asia, Central-South Asia to East Asia, genetic overlap exists for individuals whose national origins are geographically between these divisions (Li et al. 2008). Nearly 1% of this group also had evidence of Native American genetic ancestry, while a smaller fraction had evidence of African or East Asian genetic ancestry (0.3 and 0.4%, respectively). Nearly all individuals (99.7%) self-reporting African/African American race/ethnicity had evidence of African genetic ancestry; 91% also had evidence of European genetic ancestry, consistent with broad European admixture among African Americans. Native American and East Asian genetic ancestry occurred in this group at a similar low level as observed in the Europeans/West Asians (1.3 and 0.5%, respectively). Among self-reported East Asians, all had evidence of East Asian genetic ancestry; a sizable proportion (21.7%) also had evidence of Pacific Islander genetic ancestry, but this likely represents difficulty in differentiating East Asian and Pacific Islander genetic ancestry. A modest subgroup (3.4%) had evidence of European/West Asian genetic ancestry (majority are self-reported Filipinos), while small proportions had evidence of African or Native American genetic ancestry (0.1 and 0.5%, respectively). Among the Latinos, nearly all had evidence of European/West Asian genetic ancestry; a similar high proportion (94.2%) had evidence of Native American genetic ancestry, and an additional 27.7% had evidence of African ancestry. A substantial number of self-reported Pacific Islanders had evidence of East Asian genetic ancestry (91.3%) in addition to Pacific Islander genetic ancestry (66.3%); these results are again likely due to close genetic similarity between East Asians and Pacific Islanders. There is also evidence of substantial European/West Asian and South Asian genetic ancestry in this group (57.6 and 26.1%, respectively). The former reflects a high rate of European admixture among some self-reported Pacific Islander groups, while the latter likely reflects Fijians of Indian origin. Most self-reported South Asians have evidence of South Asian genetic ancestry; a substantial proportion also has evidence of European or East Asian genetic ancestry, likely due to inability to cleanly separate South Asian genetic ancestry from West Asian or East Asian (Li et al. 2008). Among those reporting Native American race/ethnicity, 14.4% have evidence of Native American genetic ancestry, and all have evidence of European/West Asian genetic ancestry.

Table 2. Proportion of individuals with genetic ancestry from each of six ancestral populations by self-reported race/ethnicity.

		Genetic ancestry
Race/ethnicity	n	EW	AF	EA	NA	PI	SA	% female	Mean age (SE)
One group	91,502							0.59	62.92 (0.04)
EW	76,401	1.000	0.003	0.004	0.009	0.000	0.043	0.59	63.71 (0.05)
AA	2,679	0.910	0.997	0.005	0.013	0.000	0.021	0.57	61.28 (0.25)
EA	6,389	0.034	0.001	1.000	0.005	0.217	0.008	0.58	58.51 (0.18)
NA	674	0.999	0.022	0.022	0.144	0.000	0.037	0.55	64.34 (0.51)
LT	4,807	0.999	0.277	0.008	0.942	0.000	0.024	0.58	57.92 (0.21)
PI	92	0.576	0.000	0.913	0.000	0.663	0.261	0.48	56.89 (1.49)
SA	460	0.307	0.007	0.109	0.004	0.050	0.961	0.39	54.29 (0.67)
Two groups	5,476							0.67	57.37 (0.19)
EW/AA	123	1.000	0.976	0.024	0.033	0.000	0.081	0.67	52.76 (1.50)
EW/EA	572	0.960	0.005	0.942	0.014	0.063	0.080	0.68	49.13 (0.65)
EW/NA	2,548	1.000	0.008	0.007	0.096	0.000	0.024	0.68	61.63 (0.26)
EW/LT	1,564	1.000	0.071	0.010	0.710	0.000	0.068	0.68	54.05 (0.38)
EW/PI	48	1.000	0.000	0.813	0.042	0.625	0.021	0.79	59.64 (2.00)
EW/SA	44	0.955	0.000	0.068	0.045	0.000	0.682	0.66	53.55 (2.26)
AA/EA	29	0.655	0.931	0.828	0.034	0.000	0.069	0.56	50.06 (2.46)
AA/NA	99	1.000	0.99	0.000	0.051	0.000	0.030	0.68	59.67 (1.30)
AA/LT	114	0.991	0.596	0.018	0.754	0.000	0.026	0.34	55.09 1.42)
AA/SA	13	0.167	0.167	0.167	0.083	0.250	0.833	0.17	54.33 (4.23)
EA/LT	95	0.789	0.042	0.926	0.642	0.063	0.000	0.67	56.07 (1.44)
EA/PI	40	0.275	0.025	1.000	0.000	0.475	0.025	0.60	56.93 (2.37)
EA/SA	17	0.059	0.000	0.765	0.000	0.059	0.235	0.47	62.06 (2.88)
NA/LT	129	1.000	0.140	0.031	0.953	0.000	0.047	0.68	58.22 (1.19)
LT/PI	12	1.000	0.417	0.250	0.917	0.000	0.167	0.64	53.93 (3.95)
LT/SA	10	0.600	0.000	0.400	0.600	0.200	0.500	0.63	61.50 (4.56)
Three groups	512							0.70	53.52 (0.75)
EW/AA/NA	115	0.991	0.991	0.000	0.043	0.000	0.017	0.74	59.71 (1.58)
EW/AA/LT	23	0.957	0.696	0.043	0.522	0.000	0.087	0.52	50.09 (4.11)
EW/EA/NA	32	0.969	0.000	0.875	0.250	0.000	0.125	0.69	46.06 (3.11)
EW/EA/LT	48	1.000	0.041	0.857	0.490	0.000	0.061	0.72	45.98 (2.49)
EW/EA/PI	35	0.943	0.000	1.000	0.029	0.486	0.000	0.67	51.92 (3.02)
EW/NA/LT	198	1.000	0.066	0.000	0.803	0.000	0.086	0.70	53.83 (0.99)

Only those with at most three self-reported race/ethnicities and three genetic ancestries are included; race/ethnicity categories with at least 10 members are shown. For individuals self-reporting two or three races/ethnicities, the correspondence between self-report and genetic ancestry is generally quite high. For example, for those reporting European/West Asian and East Asian race/ethnicity, 96 and 94% have evidence of European/West Asian and East Asian genetic ancestry, respectively; for those reporting African/African American and East Asian race/ethnicity, 93.1 and 82.8% have evidence of African and East Asian genetic ancestry, while 65.5% have evidence of European/West Asian genetic ancestry. Among those reporting European/West Asian and Native American race/ethnicity, 9.6% have evidence of Native American genetic ancestry; for those reporting African/African American and Native American race/ethnicity, 5.1% have evidence of Native American genetic ancestry. EW: European/West Asian; AA: African/African American/Afro-Caribbean; EA: East Asian; NA: Native American/Alaska Native; LT: Latino; PI: Pacific Islander; SA: South Asian. Genetic ancestry abbreviations are the same except for AF, which represents sub-Saharan African ancestry.

For those with missing or mis-scanned self-reported race/ethnicity, and whose race/ethnicity was derived from KP administrative databases (Table 3 and Table S7), results align closely with those in Table 2. For individuals self-reporting two or three races/ethnicities, the correspondence between the self-report and genetic ancestry is generally quite high (Table 2).

Table 3. Proportion of individuals with genetic ancestry from each of six ancestral populations by race/ethnicity as determined by KP administrative databases.

		Genetic ancestry
Race/ethnicity	n	EW	AF	EA	NA	PI	SA
White	4575	1	0.007	0.009	0.017	0.001	0.030
African American	102	0.941	0.990	0.000	0.020	0.000	0.020
Asian	311	0.106	0.003	0.952	0.006	0.167	0.074
Latino	255	0.988	0.192	0.043	0.816	0.000	0.035
Other/uncertain	84	0.929	0.131	0.357	0.167	0.071	0.083

Abbreviations are the same as in Table 2.

We also observed a decrease in average age and increasing proportion of females with the number of different race/ethnicity/ancestry groups reported (Table 2). While the different minority groups, and in particular the self-reported East Asians and Latinos, are younger on average, those reporting mixed race/ethnicity are even younger. These patterns likely reflect increasing exogamy over time. As expected, these patterns are also reflected in the genetic PC scores, where, for example, the proportion of mixed East Asian/European genetic ancestry increases with decreasing age. The excess of females among those reporting mixed race/ethnicity appears to reflect a reporting preference, as there was no significant difference in the proportion of individuals with mixed genetic ancestry by sex.

A more in-depth examination of the distribution of continental genetic ancestry for the various self-report race/ethnicity groups is provided in Table S8.

Relatives

We were able to clearly identify first-degree relative (parent–child and full sib) and MZ twin pairs and categorized them based on self-reported race/ethnicity (Figure S9 and Table S9). We also observed thousands of likely second- and third-degree relatives (Figure S9); however, the figure also indicates substantial overlap between these groups based on kinship estimates.

The 34 MZ pairs, who are perfectly concordant for genetic ancestry, are also perfectly concordant for self-reported race/ethnicity. Sib pairs are also (virtually) identical for genetic ancestry. We identified a total of 2018 sib pairs, 1936 (96%) of whom are concordant for self-reported race/ethnicity. Among the 82 discordant pairs, the majority (n = 66) involve pairs where one self-reports Native American or Latino race/ethnicity (solely or in combination with European/West Asian race/ethnicity) while the other reports only European/West Asian race ethnicity (Table S10); in most of these cases, the genetic ancestry is solely European/West Asian, although in some there is also evidence of Native American genetic ancestry. A modest number of pairs are also discordant in their reports of East Asian race/ethnicity, and, again for most of these, the genetic ancestry is solely European/West Asian. Similarly, a few pairs with mixed genetic ancestry including African are discordant in terms of self-reporting of African American race/ethnicity.

We identified 3741 parent–child pairs, of which 3478 (93%) were concordant for self-identified race/ethnicity. The lower rate of concordance compared to the sib pairs is not surprising as parent and child reports may differ if the child’s parents are of different race/ethnicity. In 116 of 263 discordant pairs (Table S11), the child has genetic ancestry that her/his parent does not (Native American in 69 cases, East Asian in 41 cases, and African in 11 cases), and this difference is reflected in the self-report, where the child is self-reporting a race/ethnicity that the parent is not. By contrast, in only 9 cases did the parent have a genetic ancestry that the child did not, and in 8 of these 9 cases the parent has a low level of Native American ancestry (but >5%) whereas the child is below our 5% threshold. Interestingly, in 5 of these cases the parent self-reports as Latino race/ethnicity but the child does not, whereas the opposite is true in 3 of the 8 cases. In an additional 114 cases, the genetic information for parent and child matches but the self-reports for race/ethnicity are different. The largest subgroup (49) of these cases reflects differences in the reporting of Native American or Latino race/ethnicity; and in 47 of these, there is no evidence of Native American genetic ancestry in the parent or child; it is approximately equally split as to whether the parent or child reports the Native American race/ethnicity. Among 53 cases where parent and child are discordant for self-report of Latino race/ethnicity, in ∼2/3 it is the child who self-reports Latino race/ethnicity whereas the parent does not. There are 11 cases of discordance for self-report of East Asian race/ethnicity, and in nearly all of them there is no evidence of continental East Asian genetic ancestry. In slightly more than half of these cases it is the parent who self-reports East Asian race/ethnicity.

Discussion

The RPGEH GERA cohort provides an excellent opportunity to characterize a large, representative northern California population from the perspectives of self-reported race/ethnicity/nationality and genetic ancestry. Overall, the cohort is 80.8% non-Hispanic white and 19.2% minority and includes a broad spectrum of races/ethnicities/nationalities. The results of our PC analyses to characterize genetic structure within each of the major race/ethnicity groups are largely consistent with prior reports.

For the non-Hispanic white individuals, we see a broad spectrum of genetic ancestry ranging from northern Europe to southern Europe and the Middle East. Within that large group, with the exception of Ashkenazi Jews, we see little evidence of distinct clusters. This is consistent with considerable exogamy within this group. By comparison, we do see structure in the East Asian population, correlated with nationality, reflecting continuing endogamy for these nationalities and also recent immigration. On the other hand, we did observe a substantial number of individuals who are admixed between East Asian and European ancestry, reflecting ∼10% of all those reporting East Asian race/ethnicity. The majority of these reflected individuals with one East Asian and one European parent or one East Asian and three European grandparents. In addition, we noted that for self-reported Filipinos, a substantial proportion have modest levels of European genetic ancestry reflecting older admixture.

As expected, most self-reported African Americans show some degree of European genetic ancestry, with an overall average of 26%. Among individuals self-reporting as African American and East Asian, all showed evidence of genetic ancestry from three continents: Africa, Europe/West Asia, and East Asia.

Latinos are the most complex from a genetic perspective, as they can possess genetic ancestry from essentially any of the major continents. Most of the Latinos in our study derive from Mexico and Central/South America, with smaller proportions from Puerto Rico and Cuba. These individuals have varying proportions of Native American, European, and African genetic ancestry. We also found evidence of East Asian genetic ancestry in some individuals, but these were primarily individuals who self-reported both East Asian and Latino nationalities.

Of note, ∼17% of the cohort had evidence of genetic ancestry from more than one continent. However, this does not mean that all or even most of these individuals represent recent continental admixture. As has been true in other analyses (Li et al. 2008), genetic similarity between West Asians and South Asians (and to some degree South Asians and East Asians) did not allow for a clear distinction among these genetic ancestries. As such, while some individuals were estimated to have South Asian genetic ancestry, this more likely reflects the difficulty in demarking West Asian vs. South Asian genetic ancestry. A similar situation holds for Pacific Islanders and East Asians, where we and others have shown strong genetic similarity for some Pacific Islander groups with East Asians. Also, some individuals may have reported more than a single race/ethnicity that may reflect recent country of origin in addition to, or rather than, more distant ancestry, with Indo-Fijians as one example.

If we include only individuals with genetic admixture from nonadjacent continents, the proportion with continental admixture is ∼12%. However, we also note that this fraction depends on our cutoff of 5% for defining genetic admixture as well as some imprecision in the admixture estimation. Of course, a lower threshold would increase the proportion of the cohort that is considered to be genetically admixed, while a higher threshold would do the opposite.

As expected, in a large cohort such as this, we were easily able to identify a substantial number of close relatives: specifically, 34 identical twins, 2018 full sibs, and 3741 parent-child pairs. We also had clear evidence of a large number of likely second-and third-degree relatives, but these kinship groups did not separate clearly from each other. More refined methods may be able to provide more precise kinship estimates.

A major goal was to examine the relationship between self-reported race/ethnicity and genetic ancestry. By and large, there was very high correspondence between the two, allowing for the broad range of genetic ancestry that exists among African Americans and Latinos. We were also able to compare the self-report data of identical twins, parent–child pairs, and sib pairs. All MZ twin pairs were concordant, as were most of the sib pairs. However, we did note that for some sib pairs the self-report data differed. For the majority of these, the discordance related to reporting of Native American or Latino race/ethnicity.

The results obtained here are important for the study of complex genetic disease in this large, population-based cohort through association studies, admixture analysis and admixture mapping and in particular for investigating observed ethnic variation in diseases and traits. As described previously (Risch et al. 2002; Tang et al. 2006), the strong correspondence, also observed here, between the social categories of race/ethnicity and genetic ancestry makes dissection of racial/ethnic differences challenging. The patterns that we observed reflect historical and recent mating practices and their impact on genetic variation. On a global level, geography continues to create strong local endogamy, which is also reflected among the recent U.S. migrant populations. However, the increasing frequency of interracial individuals that we observed in this cohort—a reflection of increasing exogamy—will enhance both the complexity of such analyses and the opportunities to investigate the genetic and environmental contributors to racial/ethnic differences. While the advent of myriad genetic markers can provide accurate estimates of individuals’ genetic ancestry, the social aspects of race/ethnicity may be more challenging to characterize. For example, in our study, considering the various combinations of 7 race/ethnicity categories that an individual could endorse, we observed 50 different combinations, and this does not include individuals who endorsed >3 (although they were few in number). While overall 6% of the cohort endorsed more than a single category, that number is likely to grow as mating patterns continue to evolve.