The use of race, ethnicity and ancestry in human genetic research

Abstract

Post-Human Genome Project progress has enabled a new wave of population genetic research, and intensified controversy over the use of race/ethnicity in this work. At the same time, the development of methods for inferring genetic ancestry offers more empirical means of assigning group labels. Here, we provide a systematic analysis of the use of race/ethnicity and ancestry in current genetic research. We base our analysis on key published recommendations for the use and reporting of race/ethnicity which advise that researchers: explain why the terms/categories were used and how they were measured, carefully define them, and apply them consistently. We studied 170 population genetic research articles from high impact journals, published 2008–2009. A comparative perspective was obtained by aligning study metrics with similar research from articles published 2001–2004. Our analysis indicates a marked improvement in compliance with some of the recommendations/guidelines for the use of race/ethnicity over time, while showing that important shortfalls still remain: no article using ‘race’, ‘ethnicity’ or ‘ancestry’ defined or discussed the meaning of these concepts in context; a third of articles still do not provide a rationale for their use, with those using ‘ancestry’ being the least likely to do so. Further, no article discussed potential socio-ethical implications of the reported research. As such, there remains a clear imperative for highlighting the importance of consistent and comprehensive reporting on human populations to the genetics/genomics community globally, to generate explicit guidelines for the uses of ancestry and genetic ancestry, and importantly, to ensure that guidelines are followed.

Introduction

The completion of the Human Genome Project over a decade ago has led to intensified studies of genetic variation in human populations. Much of this work uses specific population identities to categorize groups, for example Caucasian, Korean, South Asian and Yoruban, and in addition often uses the generic terminology ‘race’ and ‘ethnicity’ to refer to them. The validity of using socially-visible groups in biomedical research has been an ongoing controversy. However, there is now resurgent interest in the subject (Burchard et al. 2003; Cooper et al. 2003; Duster 2005; Risch et al. 2002; Schwartz 2001; Stevens 2003), because technology advances are increasing opportunities to clarify the relationship between social identity, genetic diversity and health, and to move beyond old prejudices about human difference (Ali-Khan and Daar 2010; Editorial 2004a; Rotimi 2004). Accordingly, in the last 2 years several multidisciplinary groups including our own, have convened to examine these issues afresh (Caulfield et al. 2009, Lee et al. 2008).

A fundamental difficulty raised by the use of socially-visible population labels—whether they are referred to as ‘races’, ethnicities, nationalities, or by other language—is that their meanings and parameters are context-dependent (Kressin et al. 2003; Rotimi 2004), and have powerful ramifications beyond the domain of science (Bamshad and Olson 2003; Clayton 2002; Gould 1981; Lewontin 1995; Provine 1973). Lack of clarity and consistency in the description of research populations and inadequate justification for their use has been a persistent source of concern in biomedical research (Bhopal 1997; Clayton 2002; Collins 2004; Comstock et al. 2004; Editorial 2004b; Lee 2004; Sankar and Cho 2002), and can have adverse scientific and social consequences, particularly in the context of genetics research. As such, failure to define a group label or describe how membership was ascertained makes it difficult to know who exactly is being studied, challenging the reproducibility of research findings and limiting portability to other geneticists, disciplines, and the clinic (Brown 2007; Editorial 2004b; Sankar et al. 2007). Further, such ambiguity can encourage racial/ethnic stereotypes and over-simplifications that stymie, rather than promote, understanding of genetic diversity (Bamshad et al. 2004; Race Ethnicity and Genetics Working Group 2005). Likewise, failing to explain why a particular population was studied with respect to the research question can imply that social identity is the basis for any observed phenotypic differences. Such interpretations may divert from further study to identify true underlying mechanisms (Sankar et al. 2004), and have dangerous clinical consequences by encouraging reliance on social identity for prescription or prognosis (Braun et al. 2007; Geiger 2003; Lee 2005).

Ongoing concern has prompted journal editors, professional societies and expert commentators to repeatedly offer guidelines for the use and reporting of race and ethnicity in genetic research. These have largely converged on four key points; (1) define the race and ethnicity, or more broadly the population terms, used in the context of the study (Anonymous 2005; Burchard et al. 2003; Cooper et al. 2003; Editorial 2004b; Iverson et al. 1998; Kaplan and Bennett 2003; Race Ethnicity and Genetics Working Group 2005; Sankar and Cho 2002; Winker 2004); (2) explain how the terms or categories relate to the research hypothesis, or why the particular population was chosen for study by the researchers (Anonymous 2005; Editorial 1996, Editorial 2004a; International Council of Medical Journal Editors 2010; Iverson et al. 1998; Kaplan and Bennett 2003; Lee et al. 2008; Race Ethnicity and Genetics Working Group 2005; Rivara and Finberg 2001; Sankar and Cho 2002; Winker 2004); (3) describe how participants were assigned to the research populations (Anonymous 2005; Editorial 2004b; Lee et al. 2008; Race Ethnicity and Genetics Working Group 2005; Sankar and Cho 2002; Winker 2004); and (4) describe the limitations of the study with respect to the populations to which the research findings can be generalized (Anonymous 2003; Anonymous 2005; Davis et al. 2001; Ioannidis et al. 2004; Osborne and Feit 1992). Various of these have been endorsed by biomedical journals, and by the International Council of Medical Journal Editors http://www.icmje.org/journals.html#S (for review, see Caulfield et al. 2009). However, studies assessing compliance in genetic research published over 2001–2004 (Editorial 2004b; Sankar et al. 2007; Shanawani et al. 2006); indicated that guidelines were not widely followed.

Since those data were collected, the advent of high-resolution genome-wide genotyping is allowing more empirical description of individuals and populations, by the inference of genetic or ‘biogeographical’ ancestry (Bamshad et al. 2004; Li et al. 2008; Novembre et al. 2008; Rosenberg et al. 2002; Royal et al. 2010; Shriver et al. 2004; Via et al. 2009). Used to determine and quantify genetic background, this technology can augment or supersede the use of proxy methods, such as self-identified race/ethnicity, physical appearance, language-spoken, or ancestry based on geographical origin, to stratify research participants and maximize their relative genetic homogeneity. Thus, some have suggested the use of ‘ancestry’ rather than race/ethnicity to describe group differences and genetic variation, because of its more objective basis, and perceived distance from negative connotations associated with ‘race’ (Ali-Khan and Daar 2010; Bamshad et al. 2004; Race Ethnicity and Genetics Working Group 2005; Smart et al. 2006). Our study had two goals; (1) to assess current compliance with recommendations for the use of race and ethnicity—or more broadly social identity—in genetic research; and (2) to examine the use of ‘ancestry’ as a generic terminology to describe study populations, and also in the sense of ‘genetic ancestry’ by the use of empirical genomic methods to categorize research groupings.

Authors who previously examined the use of race and ethnicity in genetic research considered all specific population identifiers used in the context of humans as ‘race and ethnicity’ terms (Sankar et al. 2007; Shanawani et al. 2006). We do not disagree with this, but for the purposes of the second part of our analysis we went beyond previous analyses, sub-dividing our data by the generic terminology used to refer to the specific named study populations in articles, in order to compare articles which used the generic terms ‘race and/or ethnicity’, with those using ‘ancestry’, or ‘other’ terms. In addition, we note that we did not define race or ethnicity or ancestry for the purposes of this work, but rather kept the study open-ended with the goal of observing how these terms are currently put to use by authors. In this vein, we add that the goal of this study was not to assess which of these terms should be used by authors. However, we agree with previous commentators that the study of DNA within the context of socially-identified groups in no way justifies the definition of subgroups of individuals as biologically distinct races (Collins 2010).

Materials and methods

Study design

In this work we undertook a systematic analysis of scientific articles reporting genetic research in the context of human populations. Our analysis was divided into two parts. In the first part, to evaluate how use and reporting of this research has changed over time we assessed selected metrics adapted from previous studies (Sankar et al. 2007; Shanawani et al. 2006). We also asked new questions to examine the use of ‘ancestry’ to describe populations, the use of genotyping data to assign ancestry and thus verify research group membership, and whether discussion of social and ethical implications of the reported research was included in articles. In the second part of the study, to assess differences between articles using different generic terminology to refer to study populations we sub-divided the data by articles using; (1) race and/or ethnicity; (2) ancestry; and (3) otherterminology. We then compared the metrics obtained in part one of the study across these sub-divisions of the data. We also collected qualitative data with respect to how ‘ancestry’ was used in articles.

Sample selection

We conducted a Pubmed search strategy to obtain a sample of journal articles for analysis. We used the keywords (race OR ethnicity OR ancestry) and the genetic terms (polymorphism OR CNV OR SNP), with the limits; humans, English, and publication dates between January 1st 2008 and December 31st 2009 (N = 3536). The use of ‘race’, ‘ethnicity’ and ‘ancestry’ in Pubmed captures articles in which these words occur in the text, articles that use these as words as MeSH headings, and the hierarchy of terms occurring under these headings. For example, ‘race’, is a synonym for the MeSH heading ‘Continental Population Groups’, which includes; ‘American Continental Ancestry Group’; ‘American Native Continental Ancestry Group’; ‘Asian Continental Ancestry Group’; ‘European ‘Continental Ancestry Group’; and ‘Oceanic Continental Ancestry Group’. Likewise, each of these terms captures all the specific groups classified to these geographic regions. For example ‘American Continental Ancestry Group’ includes; ‘Indians Central America’; ‘Indians North America’, ‘Indians South America’; and ‘Inuits’; and likewise, when expanded each of these terms captures a range of specific population identifiers. For example ‘Inuits’ corresponds to; ‘Inuit’; ‘Inupiat(s)’; ‘Eskimo(s)’; ‘Kalaallit(s)’; and ‘Aleut(s)’ (see: http://www.ncbi.nlm.nih.gov/mesh).

We decided to direct our sample toward articles that are most likely to reflect the state of the art in the field of genetic research, and to be of high quality. Our rationale was that such articles may be most likely to have wider scientific and social influence, by serving as models and hypothesis generators for other researchers, and by infiltrating the non-geneticist community by being reported in the popular press. To capture articles most likely to be of this type, we identified a convenience sample of 10 leading population-based geneticists, genetic epidemiologists and genome scientists based in the United States and Canada, and asked them to rate the top 5 most influential journals in which to publish their work. We then limited our article collection to the top 6 highest ranked journals from this survey. These were; the American Journal of Human Genetics; Human Genetics; Nature; Nature Genetics; PLoS Genetics; and Science (N = 197) (search completed February 2010) (see Table 1). We note that three of these have published policy on the use and reporting of race and ethnicity (Brown 2007; Editorial 2004a, b). However, none are listed as explicitly endorsing the ICMJE’s Uniform Requirements for manuscripts submitted to biomedical journals (see http://www.icmje.org/journals.html#S).

Table 1 Sample set characteristics, N (%)

The abstracts and MeSH information for each article were then reviewed to exclude all but original research articles from the sample—news, comments, letters, reviews and meta-analyses were removed. Finally, the entire articles were downloaded and reviewed in detail to verify each described original research studying human genetic variation using human tissue samples or human subjects. This process yielded a final study sample of 170 articles for analysis.

Data analysis

Part one

The study articles were saved as PDFs, and printed out, read and examined by hand to extract data. In addition, the full Medline format information on each article was uploaded to a Refworks database, and the data from our analysis were recorded in customized fields. To enable comparison to previous study on articles published from 2001 to 2004 (Sankar et al. 2007, Shanawani et al. 2006), the coding and analytical framework we used was adapted principally from Sankar et al. (2007), and with reference to the analysis by Shanawani et al. (2006). Sankar’s group developed content codes to analyze how the research populations were described, and the main components and structure of scientific articles. In addition to using these, we developed additional codes to assess the use of ancestry, of empirical genomic methods to measure ancestry, or assign or verify membership in research populations, and the discussion of ethical and social aspects in articles. An initial set of codes was tested by SEA, RT and TK. These codes were subjected to several rounds of consensus coding (Jenkins et al. 2005; Sankar et al. 2007) and discussion amongst all the authors. When interpretation and conceptual issues were resolved, and the codes were deemed to adequately capture relevant article features, a coding guide was generated listing coding rules, definitions and examples. The final study analysis was carried out by SEA.

Coding

The analysis codes evaluated four main areas: (1) basic article features; (2) reasons researchers gave for how and why they used named populations in the study design; (3) the role of the named populations in the research design or the description of the research; (4) use of empirical genomic means to assign or verify membership in the research populations; and (5) discussion of social or ethical implications of human genetic research. We analyzed each article by looking for text corresponding to these codes, or pieces of information, as described below and scored them as a yes/no variable. Additionally, for many of the codes, we collected qualitative data for further analysis, by recording the text content as well. We also noted the country of the institution of the first author, how the research was funded, whether or not informed consent was reported for the research populations involved, and whether or not a conflict of interest statement was provided.

Basic features

Each article was analysed with respect to three basic features providing fundamental information about the study it reported. Each code was scored as a yes/no variable. (1) hypothesis was defined as the presence of a founding idea or assumption stated as the starting point for investigation. Text identified for this code included formally stated hypotheses, and more general research questions, goals or aims. In each case the text had to state or imply that the idea provided the basis for the study; (2) limitations were statements that described the factors that restricted the generalizability of study findings. Statements had to be explicit and related to study design to be coded as limitations. Hypothesis and limitations are standard aspects of scientific research articles. Inclusion of a specific hypothesis is important as this is where readers might expect to find an explanation for how identifying a study population as a specific race, ethnicity or ancestry group relates to the study premise or research question. A limitations statement offers the opportunity to explain how widely the findings can be applied to populations beyond the study sample that might be associated with the race, ethnicity or ancestry terms used in the study. Note that some articles were not included in the limitations analysis because we judged their analysis to not require such a qualification; and (3) sample origin was defined not as the geographical region from which the samples were obtained, but where and how the researchers acquired the tissue samples or genetic data, for example—whether they were obtained from a tissue or databank, collected at a hospital, or were already in researchers’ possession.

Reason for using populations

To examine authors’ explanations for why research was conducted using race and ethnicity or ancestry terms, articles were classified based on three features that have been recommended by expert commentary, journals, and professional societies (Ali-Khan and Daar 2010; American Academy of Pediatrics: Committee on Pediatric Research 2000; American Anthropological Association 2000; Bamshad et al. 2004; Editorial 1996; International Council of Medical Journal Editors 2010; Lee et al. 2008; Race Ethnicity and Genetics Working Group 2005; Rivara and Finberg 2001; Winker 2004), and see, http://www.icmje.org/urm_full.pdf. (1) Why populations was used to label text that gave reasons for pursuing the research question by using a population so identified; (2) Why this population was used to code reasons provided for studying the particular population(s) in question. Reasons could be practical (e.g. because the sample was available) or theoretical (e.g. because the condition of interest was known to occur frequently in a particular group); (3) Basis for assigning population term was defined as the method by which membership in the study population was determined, or the population label was assigned to research participants. For example, self-reported by subjects, taken from existing records, assumed because of the geographical region where subjects were recruited, or assigned based on genomic inference. If an article provided any of these means it was coded yes. Thus, a yes/no variable and how, as qualitative data, was collected.

Use of genotyping data to infer genetic ancestry

To begin to evaluate the nature and the degree to which high resolution genome-wide genotyping—or genetic ancestry testing—is being used to assess the genetic background or ancestry of research participants or samples, we labelled text that described such methodologies. Only studies using these approaches as part of their process of assigning or verifying the membership of participants or samples to research groupings, or to assess for population stratification were coded as ‘yes’. The use of such methods to analyse the genetic structure of populations as the main goal of the reported research were coded ‘no’. Both a yes/no variable and how, as qualitative data, was collected.

Defines race and/or ethnicity, or ancestry terms

To assess the degree to which authors defined and described the terms and identities used to refer to research populations we labelled text according to the following codes (1) Defines generic ‘race’, ‘ethnicity’ or ‘ancestry’ was applied when text explicitly defined race, ethnicity or ancestry as a genetic, social or biological concept, or provided a reference to information that did so. Further, we began to assess the comprehensiveness of research population definitions. We based our criteria on those recommended by commentators who underlined the importance of thorough, and multi-dimensional description of populations (Bamshad et al. 2004, Editorial 2004b). Thus, we applied (2) Defines specific race or ethnicity, ancestry term (or population identifier) to articles only when they provided all of the following information; (1) the identifier or name of the research population; (2) the geographical location where the participants were recruited or the community where they were resident; (3) their ‘racial’, ethnic, or geographical ancestral origin; and (4) specified how this label was assigned (for example, by self-report, by genomic ancestry inference, based on multiple generations of the participants’ family etc.). In addition, if text, or a figure (e.g. a principal components plot) described the group genomically, defining parameters for group exclusion or inclusion, this was also coded as yes.

The role of named populations in genetic research

To examine the various ways that articles used race and ethnicity, or ancestry, text was labelled that referred to the following 5 codes: (1) Label for study population only was applied to text where race, ethnicity, ancestry or other populations terms were used to label the study population only, and not as a research variable; (2) Independent and; (3) Dependent were applied respectively when race, ethnicity, ancestry or other population terms were employed as independent or dependent variables in the research being reported; (4) DNA with label indicated where authors had labelled DNA—for example, alleles, chromosomes, haplotypes, or mutations—with a race, ethnicity or ancestry term, as in ‘Mexican and Caucasian T allele (Plaisier et al. 2009) Codes (1–4) could co-occur, but codes (2) (Independent) and (3) Dependent) were mutually exclusive.

Social and ethical implications related to human population genetic research

We looked for statements discussing social or ethical implications of population-based genetic research. Such content had to discuss implications arising from the genetic research being reported—e.g. text relating to the potential for study results to stigmatize the research population. We coded these as a yes/no variable, and if found, the issues discussed were recorded as qualitative data.

Categorization of articles by general field of interest

We also categorized the articles by their general field of interest; ‘population genetics’ was defined as including population genetic and studies examining inter or intra-population genetic structure, genetic anthropology, and whole genome sequencing articles; ‘medical’ included disease and pharmacogenomics-related articles; ‘methods’, reported new methodologies or analytical approaches in genetic research; and ‘non-medical’ was defined as articles reporting studies of non-medical-related phenotypes, for example height or hair colour. For the purposes of this analysis these categories were exclusive.

Part two

Generic terminology used to refer to research populations

We recorded all the generic terminology used to describe the research populations in each study, and the specific research population names or identifiers. If an article referred to the research populations by ‘race’ or ‘ethnicity/ethnic’ anywhere in the main article body or supplementary materials we coded the article as ‘race and/or ethnicity’. Likewise, if an article referred to populations as ancestries or ancestry groups (but not race or ethnicity), it was coded ‘ancestry’. Articles using only a specific population identifier such as whites, African Americans, Han Chinese etc., or that described populations using any other terminology such as origin, descent, etc. were coded ‘other’. We also recorded examples where the generic terms ‘race and ethnicity’ were used synonymously with ‘ancestry’, or that used them in a conceptually distinct fashion, and collected qualitative data regarding the use and application of ‘ancestry’ in articles.

We note that for the purposes of this analysis and consistent with others (Sankar et al. 2007; Shanawani et al. 2006), we did not distinguish conceptually between race and ethnicity. Despite some commentators offering distinct definitions of these (Editorial 2004b; Harrison 1995; Kalow 2001; Wood 2001), it appears that in practice they are most often used interchangeably (Condit 2007; Oppenheimer 2001; Sankar and Cho 2002). Thus, in this work, we considered them together as one category.

Supplementary and additional data

Many articles provided additional supplementary information or methods online. These were downloaded, examined for relevant information, and coded as part of the analysis for each article. Some referred readers to previously published literature for details about research procedures or the study population. These articles were also downloaded, examined and relevant statements were used as the basis for assigning the codes to the original article. If these articles did not provide the relevant details but in turn referenced another paper, we scored the article as ‘no’ for the code in question.

Statistical analyses

After sub-dividing the data by the generic terminology used to refer to the study populations in articles, as in (1) race and/or ethnicity; (2) ancestry; and (3) other. We then compared frequencies of individual codes across the resulting subsets of the data, assessing the significance of any differences via the chi-square statistic. Statistical tests were performed using SigmaStat statistical software (Version 3.5).

Results

Sample set characteristics

We reviewed and analyzed 170 research articles published in 2008 and 2009 reporting genetic research in the context of human groups. Basic characteristics and categorization by the articles’ general field of interest are shown in Table 1.

Part one—compliance with recommendations for the use and reporting of populations in genetic research

Basic article features

We were able to identify a clearly stated hypothesis or research questions in almost every article in our sample (99.4%, N = 169) (Table 2). Likewise, most papers described the origin of their research samples. Fewer articles described the limitations of their studies with respect to the populations investigated (52.4%, N = 87). Most of these limitations statements were not extensive, but rather comprised of a sentence in the article discussion stating that the study findings should be validated or further investigated in diverse populations, or in other ‘racial’, ethnic or ancestry groups (see for example, Ganesh et al. 2009).

Table 2 Sample set coding frequencies, N (%)

Reason for using populations

About two thirds of articles explained why they chose to study labelled populations (65.9%, N = 112), or why they chose to study the particular populations featured in the research (68.8%, N = 117) (Table 2). Most of these explanations were based on the phenotype or condition under study being of high prevalence in the study population, or the fact that this group was understudied in comparison to others, for example ‘Because neuroblastoma in the United States is demographically a disease of Caucasians of European descent, we limited our initial analyses to this racial group to minimize phenotypic variability’(Diskin et al. 2009). A key scientific consideration in selecting samples for association studies is that they be drawn from the most genetically homogeneous population possible—thus striving to avoid spurious associations resulting from population stratification (Cardon and Palmer 2003; Marchini et al. 2004). However, few articles (4.7%, N = 9), specifically linked this notion to the use of labelled populations in their study, or to the particular population investigated. Of the articles that did not explain why they chose to study labelled populations, all but one were association studies or other analyses to identify a trait’s genetic basis. In many of these articles populations/samples were ostensibly used because of their availability to researchers, although this was not explicitly stated.

Most articles also provided some basis for how the population label was assigned to research participants (88.2%, N = 150). Most indicated that this was by self-reported race, ethnicity, geographical origin or ancestry, and/or was determined based on the geographical region where participants were recruited or resided, and/or was assigned or verified using genomic data (see following section). Assigning population labels on the basis of more than one generation of the research participants’ family has been recommended (Tang et al. 2005). However, only 18 articles (10.6%) in our sample described using such approaches.

Use of genotyping data to infer genetic ancestry

Just over half the articles (51.8%, N = 88) described using genomic data to assess the genetic ancestry of research participants to assign or verify the research groupings, and/or to guard against population stratification (Cardon and Palmer 2003; Marchini et al. 2004) (Table 2). This is important because such approaches can substantiate the genetic similarity of individuals stratified using proxy methods, and provides another element to the description of research populations.

Most of these determinations were described in the methods section of articles where statistical analyses or quality control issues were described, and fell into three broad categories; (1) genome-wide SNP genotypes or ancestry informative markers (AIMs) were used to infer the ancestry proportions of individual participants’ DNA samples. Those whose ancestry percentages fell below a specified cut-off were excluded from further analysis (23.3%, N = 20), see for example, (Trevino et al. 2009); (2) genome-wide SNP data was used to assess the genetic homogeneity of study populations, by principal components cluster analysis, sometimes in comparison to HapMap reference populations. Samples outlying from population clusters of interest were excluded from further analysis (41.9%, N = 36), see for example, (Yamaguchi-Kabata et al. 2008); (3) text briefly states that potential population stratification was examined in the research populations, but no further details are provided. These articles simply state that population genetic structure was not evident, or that it was found and corrected (36.4%, N = 32). Articles featuring this latter wording (3), were not coded ‘yes’ as providing the basis for assigning the population label, because no details were provided as to how samples were included or excluded from the research groups. Likewise, such text was not coded ‘yes’, as constituting a genomic description of the population for the same reason (see section below). Thus, genetic ancestry testing was described in a variety of ways, and at varying levels of detail by authors.

Defining race or ethnicity, and ancestry

No article in our sample set specifically defined the meaning of the generic terms ‘race’, ‘ethnicity’ or ‘ancestry’ in the research reported (Table 2). This was surprising to us given the high percentage of articles which explained the basis for how they assigned the population label used (88.2%). Further, we expected that articles using terms of race, ethnicity or ancestry to categorize their research samples, or that used groups so labelled as independent research variables (51.2%, see section below), would also discuss or define the meaning of these concepts in the context of their study. Although no article provided explicit definitions, several studies whose goal was to analyze genetic substructure in populations, did begin to outline a distinction between population identifiers/names or self-identified ethnicity, and ancestry, which was framed in terms of genetic background (see for example, Li et al. 2008; Reich et al. 2009; Tishkoff et al. 2009).

A critical component of many recommendations has been to use as specific population labels as possible, to carefully define their meanings (Bamshad et al. 2004; Kaplan and Bennett 2003; Sankar and Cho 2002), providing as much information on the population ‘as is compatible with ethical review board requirements’ (Editorial 2004b). For the purpose of this study, we considered a definition to include (1) the name or population identity of the group; (2) the geographical region of recruitment or the community in which the research participant resides; (3) their ethnic identity and or/the geographical origin of their ancestors; and (3) a specific indication of how the latter was determined. To be scored yes, an article needed to provide all of this information. More than half the articles in our dataset defined the specific population identifier used according to these parameters (60%, N = 102). Of these, 54.9% (N = 56) included a genomic description (i.e. groups (1) and (2) described in the previous section). Of articles that did not ‘define’ the population, many noted the geographical location of recruitment or residence, and/or the race/ethnicity, or ancestry of participants or samples, but not how these latter categorizations were determined. For example, they might state that ‘all subjects were of full Japanese ancestry’ (Yasuda et al. 2008), but not explain precisely what this meant in context, or how it was determined.

Ways of using labelled populations in genetic research

About half of the articles (51.7%, N = 88) used the named populations as either dependent or independent variables (Table 2). The remainder used population identifiers only to label their study populations, but not to test a hypothesis related to the named group. A number of papers (13.5%, N = 23) labelled DNA by population. In most of these cases the population identifier was used to label the inferred ancestral identity of DNA sections in admixture mapping or similar studies, see for example (Hancock et al. 2009). A few articles used wording such as ‘ethnic-specific locus’ (Lei et al. 2009) or an ‘Asian mitochondrial DNA haplotype’ (Keyser et al. 2009).

Social and ethical implications

No articles mentioned or discussed social or ethical implications arising from genetic research in general, or from the research being reported (Table 2). On the one hand this was not really surprising given that geneticists and social scientists have not traditionally collaborated, despite calls for interdisciplinary perspectives (Ali-Khan and Daar 2010; Bonham et al. 2005; Condit 2007; Lee et al. 2008; Via et al. 2009). Conversely, given that authors from both disciplines have engaged these issues (Bamshad et al. 2004; Burchard et al. 2003; Caulfield et al. 2009; Cooper et al. 2003; Duster 2005; Lee et al. 2008; Rotimi 2004), we anticipated finding some discussion, however cursory, in articles.

Comparison of current data with previous studies

To begin to get perspective on how researchers’ reporting has changed over the course of the past decade, we compared our findings with a previous study which analyzed genetic research articles published over 2001–2004, and with which we specifically aligned part of our study methodology and analysis (Table 3). We found marked increases in the numbers of articles providing hypothesis statements for the research reported (30% in the earlier study, compared to 99.4% of articles in our dataset, P = <0.001), and likewise describing the origin of their research samples (62.4% compared to 95.9%, P = <0.001). Compared to earlier in the decade, more authors described the limitations of their research findings with respect to the population-based data reported (22.7% compared to 52.4%, P = <0.001). However, still only about half the articles provided limitations.

Table 3 Comparison of current data with earlier study

There were substantial increases in the proportion of articles (1) explaining why samples/participants in the study were grouped by race and ethnicity, or ancestry labels (10.9% compared to 66.5%, P = <0.001); and (2) justifying why these particular population groups were studied (11.2% compared to 68.2%, P = <0.001). In contrast, there was no change over time in articles defining the generic terms ‘race’, ‘ethnicity’, or ‘ancestry’ in the context of their study—no articles in either dataset defined these terms. This was surprising given the intensification of population studies using these terms in the last 10 years, and the continued scrutiny of measurement, communication and identity issues over this time (Caulfield et al. 2009; Clayton 2002; Foster and Sharp 2004; Lee et al. 2008; Rotimi 2004). It also suggests that researchers consider the meanings of these terms self-evident.

Part two: cross-comparison of articles using different terminologies—race and ethnicity, ancestry or other

We recorded the generic terminology used to refer to research populations in each article. Most described the studied populations as races or ethnicities (N = 80, 47.1%) (Table 4). For example, ‘The 67 populations analyzed in this study represent 41 ethnic nationalities living in China and other eastern Asian regions’ (Shi et al. 2009), and ‘All genetic association analyses were stratified by self-identified race (white vs. African American)…’(Rasmussen-Torvik et al. 2009). Only 4 articles (2.4%) used race only as a terminology, the rest used ‘race/ethnicity’, or ethnicity only, to describe populations. ‘Ancestry’ or ‘ancestry groups’ were used in 22.4% of articles (N = 38), for example, ‘Risk allele frequencies of rs12970134 are higher among individuals of Indian Asian ancestry than those of European ancestry’ (Chambers et al. 2008). The remainder of articles referred to populations only by a specific population identifier or name such as ‘European American’, ‘Hadza’ or ‘Japanese’, or by using various descriptors—most often origin, descent and derived, for example ‘this difference is maintained in American children of Japanese descent resident in the US’ (Burgner et al. 2009) (see Table 5 for a list of terms and ways of describing populations compiled from our sample set). We cannot comment on how the relative use of these terminologies has changed over time, as previous studies did not examine this parameter.

Table 4 Presence of coded article features by generic terminology used

Table 5 Terms used, and ways of describing populations compiled from our sample set

To assess potential differences in the way researchers use ‘race and/or ethnicity’, compared to ‘ancestry’ or ‘other’ kinds of terminology to describe research populations or samples, we sub-divided our data by the generic terminology used, and analyzed the frequency of our research codes and categorizations across these sub-groups (Table 4). For the basic article features, there was no significant difference between terminology sub-groups. However, articles using race and/or ethnicity were significantly more likely to provide a justification for why the research studied populations so labeled (75.0%, N = 60) compared to those using ancestry (44.7%, N = 17) or other terminology (69.2%, N = 36) (P = 0.004). Articles using race and/or ethnicity were also significantly more likely to report medical-related research (P = <0.001). Consistent with these findings, during our analysis we noted that medical–related articles often investigated health disparities between groups framed in terms of race or ethnicity, and rationalized the study of their research populations on this basis. On the other hand, articles using ancestry were less likely to provide a justification for the use of this terminology to label populations, or why particular populations were studied (Table 4).

Finally, we hypothesized that articles using ‘ancestry’ to label populations would be more likely to use genotyping data to assess the genetic background of their research groups, in order to assign the population label. Indeed, there was a significant relationship between the use of ‘ancestry’ and of empirical genomic methods (65.8%), compared to ‘race and ethnicity’ (55.0%), or articles using ‘other’ ways of referring to their research populations or samples (36.5%) (P = 0.017) (Table 4). The importance of controlling for population stratification through the assessment of genetic ancestry has been a key consideration in the context of genetic association studies (Cardon and Palmer 2003; Marchini et al. 2004). Consistent with this, medical (53.3%) and non-medical-related articles (62.5%) in our sample set—of which 41.5% and 50.0%, respectively reported genome-wide association studies—were more likely to use genomic methods, while population-related articles were less likely to (23.1%) (P = 0.006) (Table 6). Again this was consistent with our observations that population genetics type articles—which often mapped genetic substructure across populations—mostly relied on language-spoken, geographical location of residence or self-identified ethnicity to assign group membership. Conversely, case–control studies aiming to identify new genetic variants and striving to minimize population stratification, most often analyzed genotyping data or inferred genetic ancestry to stratify samples.

Table 6 Use of empirical genomic methods by article field of interest

Uses of ancestry in our sample set

To further understand how ‘ancestry’ is being by employed in research practice, we catalogued how the term was used by authors in articles. We found the terms ‘ancestry’ or ‘ancestry group’ were used in three main ways. Most commonly, they were used, as described above, to refer to the geographical origin of populations, for example ‘individuals of European ancestry’, or the line of heritage or descent of a group, for example, ‘Ashkenazi Jewish ancestry’ (Bronstein et al. 2008). In particular, ancestry was often used to describe populations/individuals for whom the geographic origin of their predecessors is different from their current place of residence (for example African Americans, or European Americans). ‘Ancestry’, ‘geographic ancestry’ or ‘biogeographic ancestry’ was also used to refer to the genetic background of individuals, or to sections of DNA along a chromosome, as inferred by the analysis of multi-locus genotypes. Sometimes these latter applications were distinguished by being specified as ‘genetic ancestry’ (for example, see (Li et al. 2008)). Most often ancestry or genetic ancestry used in this sense was framed in terms of continental origin—African, European, Native American or Asian, as determined by the use of HapMap I populations or other continental reference SNP collections. However, a few studies analyzed the genetic ancestry of populations on a regional scale (see for example, (Novembre et al. 2008; Reich et al. 2009)).

Confusing or interchangeable uses of race and ethnicity and ancestry

Careful definition and precise use of terms used to refer to populations would facilitate clarity about who is being studied, aid in dissecting genetic from environmental influences on phenotype, and assist in deconstructing conflation between social identity, and genetic background. However, as noted, none of the articles in our sample set defined ‘race and/or ethnicity’ or ‘ancestry’ in the context of their reported research (Table 2). A minority of articles used both race/ethnicity, and ancestry, to refer to the same populations in their reported research (21.2%, N = 36). Of these, about half used the terms distinctly, for example, Choudhry et al. (2008) specified that the ethnicity of participants was Puerto Rican (based on the reported ethnicity of the participants’ biological parents and all four biological grandparents), and then analysed their genetic ancestry in terms of West African, European and Native American background. However, some articles used race and ethnicity, and ancestry, interchangeably or indistinctly (Table 7). Most notably, while most articles which analysed genotypes to infer population genetic identities framed these in terms of ancestry, for example, ‘genetically-inferred individuals of European ancestry (Trevino et al. 2009), a few articles described these in terms of race (Yeager et al. 2009), or ethnicity, (Glessner et al. 2009) (Table 7). Finally, in a number of articles, ancestry or ‘other’ terms were used to refer to research populations in the main article body, while in supplementary materials race or ethnicity was used to describe the same populations. In some cases, this was because the research population’s inferred genetic ancestry only was discussed in the article body, and the method—including the ‘racial’ or ethnic groups from which the ‘ancestry’ group was imputed—was provided in supplementary materials. In other cases, race, ethnicity and ancestry were used in indistinct and interchangeably ways in supplementary text suggesting that less care was taken in the preparation of these materials (Table 7).

Table 7 Examples of indistinct, interchangeable or confusing usage of race and ethnicity and ancestry compiled from our sample set

Discussion

Recent advances in high resolution genetic analyses and access to larger and more diverse population samples are now offering unprecedented opportunities for biomedical progress, and for understanding human identities, histories and relationships. To maximize the benefits of this research, it is crucial that authors precisely define and describe who is under study, the constructs by which they are grouped, and how this is relevant to the research hypothesis. To evaluate the current state of research practice, we examined published articles with two goals: (1) to investigate how recommendations for the use of race and ethnicity—or more broadly social identity—in human genetic research are currently being followed; and (2) to examine the use of ‘ancestry’ as a generic terminology to describe study populations, and also in the sense of ‘genetic ancestry’ by the analysis of genomic data to stratify participants/samples.

We show that there has been marked improvement in compliance with many of the key published recommendations for the use and reporting of population-based genetic research over the last decade—at least in this sample of mainly high impact journals. However, our analysis highlighted considerable shortcomings. Below we discuss some of the main findings, and offer recommendations to improve on the current situation derived from our analysis (see Box 1).

Box 1 Recommendations for the genetics community and biomedical journal editors from our analysis, for the reporting of genetic research in human populations

‘Ancestry’ was used to refer to research populations in more than a fifth of articles in our sample set. More than 50% of articles used genetic ancestry inferences to assign participants/samples to research population groupings—most often the label ‘ancestry’ or ‘genetic ancestry’ was conferred in the context of such a genomic analysis. Genetic ancestry inferences were carried out in a variety of ways, and described with a variety of levels of detail. However, about a third of the articles in which genetic ancestry or population stratification was assessed and corrected, did not describe the method by which this was done at all (Box 1. (1) and (3)).

No article explicitly defined the meaning of the generic terms race, ethnicity or ancestry in context, or in relation to one another, even when both concepts were used within the same article. This was despite the terminology being used to label independent research variables in more than 50% of articles, and the acknowledged ambiguity of the construct of ‘race’ (Anonymous 2002; Long and Kittles 2009). Likewise, the concept of ancestry, despite its ostensibly objective basis, can be understood in multiple ways—for example genetic ancestry, geographical ancestry, biogeographical ancestry etc. (Royal et al. 2010; Via et al. 2009). Similarly, only one article explicitly discussed the relative and heuristic nature of inferred genetic ancestries and population models (Reich et al. 2009). Requiring authors to specifically define and differentiate of concepts of race, ethnicity, and ancestry would promote clarity for the reader about juxtapositions between genetic variation, population history and social identity. Equally productively, it might also engage researchers themselves in deeper thinking about these constructs (Box 1. (2)).

Alternatives to race—for example ethnicity—often seem to come to be used and understood in the same way as race (Condit 2007; Oppenheimer 2001, Sankar and Cho 2002). There is some evidence of a similar definition slippage with respect to ‘ancestry’ in our dataset, most insidiously where inferred populations labels assigned through genetic ancestry assessment were referred to as races or ethnicities (Table 7). Again, requiring the definition of race, ethnicity and ancestry by authors would highlight their differing utility in addressing different biomedical questions, and assist in prising apart conflation between social and genetic identity.

No article in our sample set discussed ethical or social implications of the reported research (Box 1. (4)), despite recent evidence suggesting geneticists are sensitive to these issues (Ali-Khan and Daar 2010; Caulfield et al. 2009; Lee et al. 2008; Smart et al. 2006). We note that the focus of the six journals from which our study sample was drawn is reporting scientific advances within the field of genetics. Thus, the absence of socio-ethical statements is perhaps not surprising, particularly given that geneticists themselves have emphasized a need for greater awareness and expertise on these issues amongst the authors of genetic studies (Ali-Khan and Daar 2010). Geneticists should consider building their capacity in this area, and/or include such experts on research teams. An important alternative, or in addition to requiring such statements by authors within research articles, would be the regular commissioning by genetics research journals of opinion and review articles by social scientists on the socio-ethical implications of recent genetics advances. In addition, we note a more recent article which provides an example of how socio-ethical concerns can be considered in implementing and reporting genetic studies (Patterson et al. 2010).

About a third of articles did not provide a justification for why they studied the particular research population, or how stratifying by race, ethnicity, ancestry etc., was relevant to the hypothesis under investigation. Notably, articles using race/ethnicity were more likely to specify this information. While this is heartening with respect to the uptake of guidelines for race/ethnicity, it suggests there should be explicit discussion and extension of these to address the uses of ‘ancestry’. We note that many of the articles that did not state the reason for the use of a particular population, ostensibly used the samples for practical reasons not directly related to the research hypothesis—because they were available. In such cases, noting that ‘available samples were of ‘X’ origin’ where applicable would contribute to the transparency of the reported research, and may minimize the possibility for misinterpretation vis-à-vis the relationship between genetic and social identity.

Various biomedical journals endorse different combinations of guidelines regarding race/ethnicity, culture, and nationality. However, many do not emphasize them in their online instructions to authors. Journal editors, as gatekeepers of publication standard, seem the intuitive choice to impose such requirements on authors. However, evidence suggests that editors do not feel qualified to develop and apply concrete rules with respect to race and ethnicity (Bhopal et al. 1997; Smart et al. 2006). More importantly, careful consideration of population naming, measurement and definition should occur during study design and research participant recruitment, not ad hoc. A lack of standards on application, definition, classification and measurement of race, ethnicity and ancestry within the genetics community has been noted (Royal et al. 2010; Smart et al. 2006), and personal communication from Dr. Steve Scherer. Such guidelines and standards should be most effective if they are generated through widespread consensus by the genetics community itself. Despite the attention directed to the use and reporting of populations in biomedical study over the last 15 years, our analysis suggests there is still an urgent need for the explicit engagement of these issues by geneticists ((Box 1. (5)), and in particular, to extend the discussion to the uses of ancestry and genetic ancestry ((Box 1. (1), (2), (3), (5)). This might be best achieved through the formation of a dedicated working group including representatives from the spectrum of countries and cultures. Such a group should spearhead discussion of extant guidelines, highlighting their importance for both scientific and socio-ethical reasons, and perhaps their revision and extension in light of the findings of the current study. Broad agreement on, and endorsement of guidelines by the genetics community globally would be a fundamental step forward. Such standards/requirements must then be supported, and consistently enforced by biomedical journal editors (Box 1. (6)).