How Ancestry Influences the Chances of Finding Unrelated Donors: An Investigation in Admixed Brazilians.

A match of HLA loci between patients and donors is critical for successful hematopoietic stem cell transplantation. However, the extreme polymorphism of HLA loci - an outcome of millions of years of natural selection - reduces the chances that two individuals will carry identical combinations of multilocus HLA genotypes. Further, HLA variability is not homogeneously distributed throughout the world: African populations on average have greater variability than non-Africans, reducing the chances that two unrelated African individuals are HLA identical. Here, we explore how self-identification (often equated with "ethnicity" or "race") and genetic ancestry are related to the chances of finding HLA compatible donors in a large sample from Brazil, a highly admixed country. We query REDOME, Brazil's Bone Marrow Registry, and investigate how different criteria for identifying ancestry influence the chances of finding a match. We find that individuals who self-identify as "Black" and "Mixed" on average have lower chances of finding matches than those who self-identify as "White" (up to 57% reduction). We next show that an individual's African genetic ancestry, estimated using molecular markers and quantified as the proportion of an individual's genome that traces its ancestry to Africa, is strongly associated with reduced chances of finding a match (up to 60% reduction). Finally, we document that the strongest reduction in chances of finding a match is associated with having an MHC region of exclusively African ancestry (up to 75% reduction). We apply our findings to a specific condition, for which there is a clinical indication for transplantation: sickle-cell disease. We show that the increased African ancestry in patients with this disease leads to reduced chances of finding a match, when compared to the remainder of the sample, without the condition. Our results underscore the influence of ancestry on chances of finding compatible HLA matches, and indicate that efforts guided to increasing the African component of registries are necessary.

Introduction

Since the first allogeneic transplant in the late 1950s, there has been significant progress in the technical procedures and success rate of hematopoietic stem cell transplantation (HSCT). Nowadays, HSCT has become the standard treatment for several hematological diseases such as hematologic malignancies (leukemias, Hodgkin’s lymphoma), and also a curative treatment for some congenital or acquired disorders of the hematopoietic system (sickle cell anemia, severe aplastic anemia, thalassemia and inborn metabolism errors), as well as a therapeutic option in the treatment of some solid tumors (1, 2).

The success of allogeneic HSCT (allo-HSCT) is in part due to a better understanding of the role of HLA genes in the immune response. Classical HLA genes encode antigen-presenting proteins which are recognized by T-cell receptors. When HLA molecules bind a non-self antigen, cellular and humoral immune responses are triggered. Thus, in allo-HSCT the match between patient and donor HLA is critical, since different HLA alleles can generate distinct antigens detected as non-self during cellular immunological inspection. As a consequence, the patient’s immune system sees the donor cells with incompatible HLA as foreign and mounts an immune response, leading to rejection of the HSCT, and/or to graft versus host disease after grafting. Therefore, the gold standard for allo-HSCT is full compatibility between patient and potential donor at the 2 alleles of HLA-A, -B, -C, -DRB1, and -DQB1 (10/10 match) [see review in Tiercy, 2016 (3)]. Nevertheless, in some cases mismatches may be allowed (9/10 match), especially at alleles which do not generate an anti-HLA antibody (4, 5), or when new therapeutic protocols of haploidentical transplant are used, in which the donor and recipient share a common HLA haplotype (6).

The extremely high number of alleles at HLA loci (more than 27,000 classical HLA alleles were described in the Immuno Polymorphism Database IMGT/HLA until July 2020, https://www.ebi.ac.uk/ipd/imgt/hla/stats.html) makes relatives of the patient the first option in searching for a donor. The patient’s probability of having the same HLA haplotypes as one of their siblings is 25%, and increases with the number of siblings (43.7% with 2, 57.8% with 3, 68.4% with 4 siblings, etc.) (7). However, according to the USA National Marrow Donor Program (NMDP), only 30% of allo-HSCT donors are chosen from close relatives, with unrelated allo-HSCT donors accounting for about 70% of cases (8). Therefore, public donor registries play a key role in meeting the demand for unrelated donors.

According to the World Marrow Donor Association (WMDA), which includes 53 associated countries, the number of registered unrelated donors exceeds 35 million individuals and 700,000 cord blood units (WMDA, 2019; https://statistics.wmda.info/). Despite this large number, a major concern has been to understand how representative existing registries are of the general population, since the high diversity of HLA loci, their geographical heterogeneity (9, 10), and biases in volunteer recruitment, can result in certain groups within the population having reduced access to donors.

This scenario may be critical in the case of admixed populations, which trace their ancestry to different geographic regions, and has been the focus of several recent studies on how ethnicity influences the probability of finding a donor (8, 11–13). In the USA, an NMDP study found that while 75% of patients who self-identify as White-European found a donor with 7/8 HLA matching, only 16% for African-Americans found a match (8). A Memorial Sloan Kettering Cancer Center study of 7/8 or 8/8 HLA matching found that 78% of patients with Northwestern European ancestry found a donor, while this number fell to 44% for those with South European ancestry and 22% for those with African ancestry (13). Theoretical studies by Bergstrom et al. (11, 14) also found that the probability of not finding a match in the NMDP is highest among African-Americans, when compared to other groups. These findings have been interpreted as an outcome of a combination of factors, including a lower representation of African-Americans in databases, as well as the greater genetic diversity of populations originating in Africa, which decreases the likelihood that two unrelated individuals will share a multi-locus HLA genotype (15).

While the studies discussed above make a convincing case regarding the impact of self-identification upon the chances of finding a donor, they do not address the low correspondence between self-identification and genetic ancestry (16–18). In admixed populations, individuals who self-identify to the same group frequently have extremely variable genetic ancestries. This is expected, since self-identification involves a complex interplay of social and genealogical components (19) and varies among geographic regions in Brazil (20, 21).

Here, we investigate the role of genetic ancestry in determining the chances that an individual will find a match in REDOME (Registro Nacional de Doadores Voluntários de Medula Óssea), the Brazilian Marrow Donor Registry. We investigate how three layers of information affect the chances of finding an HLA match: (i) the self-identification; (ii) the genetic ancestry; (iii) the genetic ancestry of the MHC region (). Brazil harbors the largest number of individuals with African ancestry outside Africa (https://www.slavevoyages.org/) and has one of the most admixed populations in the world. In addition, REDOME is the third largest marrow registry in the world, with more than 5 million registered individuals (as of July 2020). We use samples from two cohorts of Brazilians for which we have information on self-identification, and that were previously genotyped with high density SNP arrays (22, 23), providing a total of 8,037 individuals. For each individual we estimate their genetic ancestry, their ancestry within the MHC, and we impute their HLA genotypes using the SNPs flanking the classical HLA loci. We then query the REDOME to identify how many matches are present for each individual, and use these results to compare how different measures of ancestry influence the chances of finding a match ().

Figure 1

Schematic representation of the strategy used in this study. Individuals are classified by three criteria: “IBGE categories” are self-identified categories defined by the Brazilian Institute of Geography and Statistics (IBGE), based on phenotypic characteristics such as skin color, and can be “Black”, “Mixed”, “Indigenous”, “White”, and “Yellow”; “Genetic Ancestry” is inferred with genetic markers spread throughout the genome, and describes the proportion of an individual’s genome that can be assigned to a continent of origin (with red and blue corresponding to African and European proportion ancestries respectively); “MHC ancestry” is the genetic ancestry of the segment of chromosome 6 including the MHC region within an individual (again, with red and blue corresponding to African and European ancestries). For each individual, classified with these criteria, we queried REDOME for the presence of a matching genotype and recorded the number of matches, when they were found.

Material and Methods

Datasets

Individuals from two Brazilian cohorts were queried for matches in REDOME, as described below. Although only a subset of individuals in these cohorts are in fact patients for whom there is an indication of transplantation, for the purposes of our analyses we will treat all samples as patients, given that our goal is to establish relationships between ancestry and chances of finding a potential donor among Brazilians with varying ancestries.

The Brazilian National Ethical Committee for Research (CONEP, resolution 15.895 and resolution 1.297.627), local ethics committees at each participating center, the Institutional Review Boards at the University of California, San Francisco and the REDS-III data coordinating center, RTI International, all reviewed and approved the study.

The Recipient Epidemiology and Donor Evaluation (REDS)-III Brazil Sickle Cell Disease (SCD) Cohort

The REDS-III Brazil SCD cohort was established to study the epidemiology and transfusion outcomes of SCD in Brazil, and includes 2,795 individuals recruited between 2013 and 2015 in 4 reference centers: Fundação Hemominas (Belo Horizonte, Juiz de Fora, and Montes Claros), Fundação Hemope (Recife), Fundação Hemorio (Rio de Janeiro), and Instituto da Criança Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo (São Paulo) (). Samples were previously genotyped with a high density SNP array (Axiom Transfusion Medicine Array, TM Array, Affymetrix, Santa Clara, CA, USA) and are available in the dbGAP (phs001972.v1.p1). After filtering (calling < 97%, Hardy-Weinberg p-value > 10-8, sample missing data <5%) the dataset had 2,703 individuals and 820,837 SNPs (22).

EPIGEN Brazil Initiative

The Brazilian EPIGEN Initiative (EPIGEN) is the largest Brazilian resource (n=6,487) for population genomics. EPIGEN includes three cohorts: Salvador (n=1,309), from northeast Brazil (24); Bambuí (n=1,442), from the southeast (25); and Pelotas (n=3,736), from south Brazil (26) (). The samples were previously genotyped with the HumanOmin 2.5 (Illumina, San Diego, CA, USA) SNP array (23) and are available in the European Nucleotide Archive (PRJEB26388 (ERP108374)), under EPIGEN Committee Controlled Access mode. We selected samples for which both genotype data and self-identification were available, resulting in a total of 5,334 samples (Salvador, n = 918; Bambuí, n = 765; and Pelotas, n = 3,651).

HLA Imputation

For admixed and non-European populations, Attribute Bagging is the most accurate approach for imputing HLA alleles from SNP data (27–29). Here we use the HLA Genotype Imputation with Attribute Bagging (HIBAG) R package (30). HIBAG uses a reference panel with data on both HLA alleles and SNPs in the MHC region to infer HLA alleles for samples with only SNP data.

We built a multi-ancestry imputation model for HLA-A, -B, -C, -DQB1, and -DRB1 with two-field level of resolution using the 1000 Genomes phase III (2,504 individuals) as a reference panel (31, 32). We selected SNPs within the MHC region which are present in both the TM array (10,711 SNPs) and HumanOmin2.5 (9,187 SNPs). For each SNP array, HIBAG models were built using SNPs within 500 kb flanking each HLA locus, and 100 bootstrap samples as classifiers. Out-of-bag estimated accuracies for each model are reported in , and models are available upon request. HLA imputation was performed separately for REDS-III and EPIGEN, and the posterior probability estimated by the model was used as a confidence score, allowing inference of the predicted accuracy of the imputation for each HLA genotype. We used the empirical cumulative distribution function (ECDF) to compare the posterior probability distribution of the imputed HLA genotypes among IBGE categories and genetic ancestries ( and ).

Since Native American populations are not well represented in the reference panel, HLA imputation for individuals of this ancestry is uncertain, so we chose to focus exclusively on the effects of African and European ancestry in subsequent analyses. Understanding of how Native American ancestry impacts the chances of finding a donor in REDOME will require direct typing of HLA alleles.

Finding Matches in REDOME

The Brazilian Bone Marrow Donor Registry (REDOME) was established in 1993, funded by the Brazilian Ministry of Health. As of April 30, 2019, REDOME had 4,869,224 registered volunteers. All individuals are genotyped at HLA-A, -B and -DRB1, and a subset also at HLA-C (n=125,248) and -DQB1 (n=123,298), with HLA resolution ranging between low (e.g., serological assays) to medium/high (e.g., SBT-PCR, Next Generation Sequencing) ().

For each individual, we queried REDOME for potential donors at both low and medium resolutions (one and two HLA allele fields, respectively). At low-resolution an allele such as A*34:02 is compatible with those from donors carrying any variant of A*34 (e.g., A*34:01, A*34:02, etc.). At medium-resolution, only variants of A*34:02 (for example A*34:02:01G) are considered compatible, and any NMDP allele code containing the 34:02 allele.

We searched REDOME for full matching at three, four and five HLA loci, hereinafter referred to as 6/6, 8/8 and 10/10, respectively. Searches for 6/6 matching were for HLA-A, -B and -DRB1; 8/8 searches further include HLA-C; 10/10 searches further include HLA-DQB1. We also carried out analyses that allowed mismatches at a single allele (5/6, 7/8 and 9/10).

To evaluate the impact of the predicted accuracy of imputation on our results, we also performed searches on a subset of individuals with posterior probability of HLA imputation greater than 0.8 at all surveyed loci (2,554 individuals in total; REDS, n = 914; Bambuí, n = 433; Salvador, n = 219; Pelotas, n = 988). In the main text, all results refer to the dataset of 8,037 individuals, without filters for predicted accuracy of HLA imputation, and we refer to for results on the high confidence subset when appropriate.

Inference of Genetic Ancestry

To infer the genetic ancestry for Brazilians, we used the three parental populations: 502 African and 503 European individuals from the 1000 Genomes Project phase-III (33), and 234 Native American individuals (34).

Genetic ancestry for each Brazilian individual was estimated with ADMIXTURE v.1.23 (35), which uses a maximum likelihood framework, based on multilocus SNP genotypes. We performed a supervised analysis with K=3 (corresponding to African, European and Native American ancestries), 2000 bootstrap replicates, windows of 50kb and step size of 10kb, and LD R2 threshold of 0.1.

To infer the genetic ancestry of the MHC region for each individual, we used RFMix v.2 (36). First, genotypes for chromosome 6 were phased with the SHAPEIT v.2.12 software (37). Parental populations were subsampled to have equal sample sizes (234 individuals). We ran RFMix with default parameters, a time since admixture of 8 generations, and 2 EM iterations. Local ancestry was estimated for the whole of chromosome 6, but we subsequently focused on the ancestry of the subset of the MHC region encompassing the classical HLA loci, delimited by HLA-A and -DQB1, and spanning 2,724,220 bp.

IBGE Categories, Genetic Ancestry and the Chance of REDOME Match

The Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e Estatística, IBGE) defines a widely used classification, which identifies individuals as “Black” (Preto), “Mixed” (Pardo), “Indigenous” (Indigena), “White” (Branco) and “Yellow” (Amarelo), creating categories that confound skin color, social self-identification and genealogy (38). Despite critiques (39), this classification remains used in epidemiological studies, the Brazilian census, blood centers, and by REDOME. All individuals in the REDS-III and EPIGEN cohorts, as well as those in the REDOME database, are self-identified according to these categories. Here, we use only “Black”, “Mixed” and “White”, since “Indigenous” and “Yellow” contribute to only 1.8% of our sample (). Our use of these IBGE categories does not assume they are natural groupings with a biological basis. Rather, our goal is to examine how this widely used classification predicts the chances of finding a match in REDOME, and how results based on this classification compare to those obtained when groups defined by genetic ancestry are used.

We used a univariate logistic regression model, assuming “match in REDOME” (yes/no), as the outcome and IBGE categories or genetic ancestry as predictors, computing the odds ratio for each contrast. For genetic ancestry (estimated as percentages of African, European and Native American ancestries per individual), individuals were classified into quartiles of increasing African ancestry (Q1, Q2, Q3, Q4). For local ancestry analyses, we identified three genotypic classes for the MHC region in each individual: African/African, African/European or European/European. Because the ancestries in the MHC have a trimodal distribution (), we could classify individuals into 3 groups according to number of chromosomes with African MHC: 0, 1 and 2 (rounding intermediate values, resulting from chromosomes with mixed ancestries, to the nearest integer).

Results

IBGE Categories Explain Only a Small Amount of Genetic Ancestry

According to IBGE census (2019) (40) the Brazilian population has composition of 45.22% “White”, 45.06% “Mixed”, and 8.86% “Black”, 0.47% “Yellow” and 0.38% “Indigenous”, whereas REDOME’s composition is 54.64% “White”, 23.44% “Mixed”, 7.17% “Black”, 0.46% “Indigenous”, 3.31% “Yellow” and 10.97% “Non Informed”, indicating a deficit of the “Mixed” category in REDOME with respect to the Brazilian population as a whole.

The REDS-III and Salvador (EPIGEN) cohorts have high proportions of “Mixed” and “Black” and high African ancestry, whereas Bambuí and Pelotas are predominantly “White” and have high European ancestry (). The merged dataset has 44.1% “White”, 32.6% “Mixed” and 19.8% “Black” individuals, and the average genetic composition was 62.4% European, 30.9% African and 6.1% Native American.

Table 1

Brazilian Institute of Geography and Statistics (IBGE) categories and genetic ancestry across cohorts.*

Dataset	IBGE categories (%)						Genetic Ancestry (%)
	Black	Indigenous	Mixed	White	Yellow	Non Informed	African	European	Native American
REDS-III	26.7	0.5	58.6	10.7	0	3.5	49.7	43.6	6.7
EPIGEN Salvador	25.4	0.1	61.9	7.5	0	5.1	50.3	43.8	5.9
EPIGENBambuí	6.3	0	34.4	59.3	0	0	14.2	79.4	6.0
EPIGEN Pelotas	16.1	1.8	5.6	74.8	1.7	0	15.5	77.3	7.2
Merged Dataset	19.8	1	32.6	44.1	0.8	1.7	30.9	62.4	6.1

*Percentages refer to the proportion of individuals in each IBGE category, or the average per individual of each genetic ancestry.

We initially examined the relationship between IBGE categories and genetic ancestry (). Individuals categorized as “Black” have, on average, greater African genetic ancestry than those categorized as “White” (ANOVA p-value < 2x10-16; Tukey test p-value < 0.00001). Despite the statistical significance, much of the variation in ancestry among individuals is not captured by the IBGE categories, with about 36% of the variation in African and European genetic ancestry not being explained (r2 = 0.64; ). The poor correlation between self-identified categories and genetic ancestry is even more apparent for the REDS and Salvador cohorts, both of which have a high proportion of individuals in the “Black” and “Mixed” categories (), and in which most of the variation in genetic ancestry is not captured by the IBGE categories (r2 = 0.31).

Figure 2

Relationship between Brazilian Institute of Geography and Statistics (IBGE) categories (“Black”, “Mixed”, and “White”) and genetic ancestry (African, European and Native American) estimated using ADMIXTURE software in the merged dataset (REDS-III + EPIGEN). The white circles represent the averages and the vertical white lines correspond to one standard deviation, and the black shapes describe the distribution of ancestries within each IBGE category.

Our finding of a weak correlation between IBGE categories and genetic ancestry is consistent with previous studies of admixed Brazilians (18, 41), and suggests that the higher the admixture proportion, the lower the correlation between them. This suggests that there may be important differences between the influence of IBGE categories and of genetic ancestry on the chances of finding a match in REDOME, and it is this question we address in subsequent next sections.

The Success of Finding Matches in REDOME

The proportion of individuals in the merged dataset for whom we find at least one match in REDOME at low resolution was 87.7% (6/6), 15.1% (8/8), and 13.0% (10/10). For medium resolution, matches were found for 51.3% (6/6), 6.1% (8/8), and 2.0% (10/10) of the individuals ().

Table 2

Proportion of individuals with at least one match in Brazilian Bone Marrow Donor Registry (REDOME), median number of matches and the maximum number of matches at low and medium resolution.

Loci	Low resolution			Medium resolution
	% at least one match	Median number of matches	Maximum number of matches	% at least one match	Median number of matches	Maximum number of matches
6/6	87.7	15	3,157	51.3	5	2,229
8/8	15.1	3	81	6.1	2	56
10/10	13.0	3	79	2.0	3	49

The total number of potential donors per individual varies depending on the number of loci queried and resolution (). Among those individuals who find at least one match, the median number of potential donors at low resolution is 15 (6/6), 3 (8/8), and 3 (10/10); at medium resolution the values were 5 (6/6), 2 (8/8), and 3 (10/10).

Due to the small number of matches for medium resolution (and low statistical power to investigate the effect of ancestry), we subsequently only report results for low resolution matching. The results for all HLA combinations of low/medium resolution and numbers of loci are available in the ().

Self-Identification as “Black” and High African Ancestry Correlate With Lower Chances of Finding a Match

We first quantified the proportion of individuals with at least one match in REDOME (, top row). In the 6/6 queries, 91.1% of the individuals classified as “White” find at least one donor compared to 84.7% and 82.9% of those categorized as “Mixed” and “Black”, respectively. In the 10/10 queries 16.9% of “White” individuals find a match, while only 7.3% of “Black” individuals do.

Figure 3

Percentage of individuals from the merged dataset with at least one match in Brazilian Bone Marrow Donor Registry (REDOME) for low resolution 6/6, 8/8, and 10/10. Top row: Brazilian Institute of Geography and Statistics (IBGE) categories; Middle row: Quartile of African ancestry; Bottom row: Number of chromosomes with African MHC per individual.

When we compare the chances of finding a match across groups defined by their proportion of African genetic ancestry, a similar pattern emerges, with the chance of finding a match decreasing as African genetic ancestry increases (, middle row). For 6/6 queries, 92.1% of the individuals in the first quartile (with less than 0.61% African genetic ancestry) find at least one potential donor, as opposed to 81.7% of those in the fourth quartile (more than 51.31% African genetic ancestry). Again, in the 10/10 queries the difference is more extreme (18.6% vs. 7.12%).

While genetic ancestry refers to an average of the entire genome, it is possible to assign an ancestry specifically to the MHC region. We found that the greater the African ancestry in the MHC, the lower the chances of finding at least one potential donor. The percentage of individuals with 0 versus 2 chromosomes with African MHC who find a match is 93.0% vs. 76.5% at 6/6, and 19.2% vs. 4.4% at 10/10, respectively (, bottom row).

Using a univariate logistic regression, all comparisons between groups with least African ancestry (i.e., “White” or Q1 of African ancestry) and most African ancestry (i.e., “Black” or Q4 of African Ancestry) are significant (p-value < 2×10-16; see for complete set of OR).

Among individuals with at least one match, the number of potential donors varies substantially (). We therefore assessed how the IBGE categories, genetic ancestry and MHC ancestry influence the number of potential donors found. Individuals categorized as “Black” and those with great African genetic ancestry, on average have a smaller number of potential donors, as compared to those categorized as “White” or having less African ancestry (). For the 6/6 queries, these differences are significant for all layers of ancestry (IBGE categories, African genetic ancestry, and MHC ancestry; p-value < 0.05, Mann-Whitney U test, ).

Therefore, ancestry affects the chances of finding a match in two ways. First, in comparison with individuals categorized as “White” or having more European ancestry, a lower proportion of individuals classified as “Black” or carrying higher African ancestry find a match in REDOME (). Second, among those who do find a donor, those classified as “Black” or having a greater African ancestry, on average have a smaller number of potential donors than “White” or genetically more European individuals ().

Matching of Ancestries Between Donors and Recipients

Overall, our results support that ancestry influences the chances of finding a match in REDOME. Thus, we next asked whether the ancestry of an individual is the same as that of his/her potential donors. Since we do not have genetic ancestry data for REDOME, we analyze IBGE categories.

For individuals with at least one match, we computed the average proportion of potential donors from each IBGE category. We find that individuals in the “Black” IBGE category, as compared to those in “White” and “Mixed”, match proportionally more to “Black” potential donors (5.7%, 10.1%, and 14.5% of “Black” donors for “White”, “Mixed” and “Black” individuals, respectively, at 6/6) (, top row). When considering genetic ancestry, individuals with progressively more African ancestry match proportionately more to “Black” potential donors (4.8%, 6.4%, 10.4% and 15% of “Black” donors for individuals in Q1, Q2, Q3, Q4 at 6/6) (, middle row). This trend is more pronounced when we consider ancestry in the MHC, where individuals who carry two chromosomes with African MHC match a pool of donors which is 20% “Black”, versus 4.5% for individuals with no chromosomes with African MHC (results for 6/6 matching) (, bottom row).

Figure 4

Mean frequency of Brazilian Institute of Geography and Statistics (IBGE) category of the potential donors for individuals in different IBGE categories, African ancestry quartiles, or with different numbers of chromosomes with African MHC. For each individual we calculate the average IBGE category of his/her potential donors, then for each category or genetic ancestry group we computed the group-level average (average of individual averages). Individuals who are “Black” or have more African genetic ancestry find proportionately more donors in the same category/ancestry.

Finding Donors for Sickle Cell Disease Patients

Within our merged dataset, the individuals who are from the REDS-III cohort are diagnosed as having sickle cell disease (SCD). We therefore repeated our previous analyses on these individuals, so as to assess the influence of ancestry for a set of individuals who are eligible for HSTC (42).

Sickle Cell Disease (SCD) is a genetic disorder with an African origin, affecting 3,500 newborns annually, and estimated to affect between 25,000–30,000 people in Brazil (43). The only curative treatment available is HSCT. Although transplantation among family members is recommended, parents or siblings may be carriers of the sickle cell trait, which compromises them as donors. Thus, unrelated allo-HSCT transplantation or the haploidentical transplant, are useful options.

Based on pre-established criteria by the Brazilian Ministry of Health, Flor-Park and col. (42) identified a subset of 417 patients within REDS-III who are eligible for transplantation. This subset consists of 30.3% “Black”, 54.7% “Mixed”, 10.3% “White” individuals, with 4.8% in other IBGE categories. The average genetic ancestry of these individuals is 52.0% African, 41.9% European and 6.1% Native American.

Among these SCD patients, 8.1% find a match in REDOME (10/10 low resolution), compared to 14.4% among individuals without SCD in the EPIGEN cohort (z-score=−3.563, p-value= 0.0004) (). Thus, consistent with the previous analyses, we show that individuals with increased African ancestry – in this case a set of patients with SCD and an indication for transplantation – have a decreased chance of finding a match in REDOME when compared to individuals from a group which has lower African ancestry ().

Discussion

We investigated how ancestry is related to an individual’s chance of finding a match in REDOME, the third largest bone marrow registry in the world. For a sample of 8,037 Brazilians, we used three approaches to assess ancestry: (i) self-identified categories (e.g., “Black”, “White”, or “Mixed”), (ii) genetic ancestry (i.e., the percentage of an individual’s genome which is African or European; ), and (iii) genetic ancestry specific to the MHC region (i.e., how African or European are an individual’s chromosomes within the MHC region; ).

A Hierarchy of Effects: Self-Identification, Genome-Wide Genetic Ancestry, and Genetic Ancestry in the MHC

As in other studies (17, 44, 45), we find that individuals categorized as “Black” or “Mixed” have a lower chance of finding a potential donor than “Whites”. We also show that African genetic ancestry, both genome-wide and within the MHC, is associated with a marked reduction in the chances of finding a match. To directly compare the differences in chances of finding a match between classes such as “Black” and “White’’ to those between extremes of African ancestry, we performed an additional analysis, in which we matched the size of groups being compared (, explanation in legend). We document a hierarchy of effects of African ancestry on the chances of finding a match, with the chances progressively decreasing from IBGE categories, to genomic ancestry, and reaching a maximum decrease when comparing classes based on ancestry in the MHC region.

Table 3

Percent decrease in the chances of finding at least one match for “Black” or most genetically African individuals, relative to “White” or least genetically Africans, respectively.*

Query(low resolution)	Black vs. White	Most vs. Least African (genome)	Most vs. Least African (MHC)
6/6	9%	11%	18%
8/8	51%	54%	65%
10/10	57%	60%	75%

*All contrasts are referenced to that between “Black” (n=1,589) and “White” (n=3,544) sample sizes, involving a partition of ancestry of 3,544 individuals into two groups.

The association between African ancestry and decreased chances of finding a match has previously been explained as a consequence of the higher polymorphism of African populations (15). Humans originated in Africa, and spread over the world experiencing a series of bottlenecks (46, 47), which caused non-Africans to have lower genetic diversity. This reduced variation in non-Africans extends to HLA loci (9, 48, 49), and as a consequence the chance that two unrelated Africans will share a multi-locus HLA genotype is lower than that seen between two Europeans. In this study, we show that African ancestry in the MHC region is the strongest predictor of decreased chances of finding a match in REDOME. Our interpretation is that the IBGE categories and genetic ancestry genome-wide capture information about the ancestry within the MHC, thus accounting for their effects on the chances of finding a match. However, the ancestry within the MHC region is most directly associated with the chances of finding a match, since it captures information about the ancestry of the HLA allele an individual carries, which drives the chances of finding a match.

In addition to the role of African ancestry on the chances of finding a match, admixture in Brazilians may also be an important factor. The EPIGEN cohort (used in the present study), identifies that populations from different regions of the African continent contributed to the Brazilian gene pool (50), so that populations that never met in the African continent may experience gene flow in Brazil. If we transpose this issue to the MHC region, new HLA genotypic combinations may emerge among individuals of African ancestry, adding to the difficulty in finding unrelated voluntary donors. New genotypic combinations created by admixture may also be a more general issue, regardless of ancestry. For example, patients who do not find donors frequently carry alleles which are present in REDOME, but for which the genotype combination was not found (51). Thus, future studies should also model and quantify the role of admixture in shaping the chances of finding a match in REDOME.

Clinical Implications

Our dataset contains a group of patients with sickle cell disease (SCD) and a recommendation for transplantation. At low resolution, 8.1% of SCD patients surveyed find a 10/10 match, compared to 14.4% for those without SCD (in the EPIGEN cohort; p-value=0.0004). This provides a concrete example of how patients with a higher than average African ancestry, when seeking transplantation, have reduced chances of finding a match.

There are still few studies with HSTC from unrelated donors for SCD patients [review in Oevermann and Sodani, 2020 (52)]. This scarcity is due to the difficulty of finding a match with unrelated donors in international public registries (52, 53). As a consequence, many of these unrelated allo-HSTC were performed with some level of HLA mismatch and were associated with a high prevalence of engraftment failure as well as graft-versus-host disease (53, 54). Recently, important advances have been reported, both with improved match between unrelated patients and donors (HLA high resolution full match or haploidentical) and in treatment with post-transplant drugs (55). However, these studies are not sufficient to guarantee the safety of this treatment (52), which is still evaluated as experimental. For this reason, HSCT between siblings is the gold standard for patients with SCD, and in Brazil it is the only modality recommended by the Ministry of Health (ordinance SAS/SCTIE n° 5/2018) with support from the Public Health System (Sistema Único de Saúde - SUS). Therefore, despite the curative potential of unrelated allo-HSCT for many hematological malignant diseases, it is not yet used routinely in clinical treatment of SCD. This is explained, to a large degree, by the high African genetic ancestry in patients with SCD, which decreases the chance of finding a compatible donor. As a consequence, the small number of transplants performed from unrelated donors have often allowed mismatches.

The drawbacks associated with the impossibility of locating HLA-compatible donors for HSCT are potentially severe. In the case of some hematological malignant diseases, such as acute lymphoblastic leukemia in adults, a lack of a HLA-compatible donor may mean withdrawing the therapeutic cornerstone, which is HSCT, and consequently compromising patient survival (56). This scenario is similar for other important non-malignant diseases, i.e. aplastic anemia, in which the HSCT plays a pivotal therapeutic role. However, it seems even more complicated in the case of SCD patients, for whom the low chance of finding a compatible unrelated donor makes it difficult to gather large samples to evaluate appropriate post-transplant parameters and treatment options. Therefore, at the moment, most SCD patients who are eligible for transplantation would not benefit from the positive effects of unrelated allo-HSCT, including decrease of vaso-occlusive episodes, donor-derived erythropoiesis and restoration/stabilization of function of damaged organs (57).

HLA Diversity and Donor Recruitment Challenges

Previous studies have shown that individuals with the same ancestry tend to share HLA alleles more often than those from different populations (12, 58). Our study confirms this trend, with the number of “Black” donors increasing with the recipients’ degree of African ancestry (). On the other hand, many potential donors for “Black” individuals are in fact “White”. This is expected in an admixed population for several reasons: many HLA alleles are shared around the world; REDOME is predominantly “White” and “Mixed”; self-identified categories do not accurately capture the genetic ancestry; potential “White” donors may contain African ancestry in the MHC region.

Faced with this complex scenario, it is natural to ask how large a registry should be, in order to adequately cover the diversity of a population. Expanding the size of the registry indefinitely is not an economically viable option, so strategies have been proposed to define an “optimal registry size”. These efforts model the frequency of the most common HLA haplotypes, the distribution of individuals among regions within a country, and probability of matching (59–62). This task is particularly complex in admixed populations, since genetic diversity varies among groups.

As is the case for the NMDP (11, 14), REDOME has two features indicating that African ancestry is underrepresented, despite the registry’s large size. First, there are proportionally fewer individuals classified as “Mixed” in REDOME as compared to the census for Brazil. Second, individuals with greater African ancestry have decreased chances of finding a match. Increasing REDOME’s African ancestry component, as shown by our findings, is expected to increase the chance of individuals of African ancestry finding a match ().

Once the underrepresented group has been identified, what is the best strategy for donor recruitment? It is economically unfeasible to determine the genetic ancestry of all donors. An alternative is to direct recruitment campaigns to regions/cities where the underrepresented group is more common, or to work with governmental and non-governmental organizations that represent these groups. The impact of a strategy directed to increase the presence of underrepresented groups in REDOME is likely to extend beyond Brazil, since between 2016-2018 REDOME exported 282 hematopoietic cells to Europe, the United States, and Asia (63).

Caveats

In this study we resorted to imputation based on the SNP data to make HLA calls. The accuracy of imputation is reduced when the reference panel inadequately covers the diversity of the sample being imputed. Because the number of samples in available reference panels is still scarce, especially for admixed populations (64–67), alleles may have been incorrectly imputed. To evaluate the impact of imputation errors, we replicated our analyses for a subset of individuals with > 80% probability of correct imputation in all 5 HLA loci. In this dataset of HLA genotypes with a prediction of high accuracy, we find a greater proportion of individuals with at least one match (, and ). This reflects the higher predicted accuracy of imputation for non-rare alleles and non-Africans, for which matching probabilities are also higher. However, the influence of ancestry on matching is highly concordant with the original conclusions, although some statistics are no longer significant (, and ).

Our study was unable to explore the effects of Native American ancestry, which reaches 20% in Northern Brazil (39, 68). While a study of induced pluripotent stem cell (iPSC) banks in the USA showed similar match rates for Caucasians and Native Americans (48% and 46%, likely due to their low genetic diversities) (69), this remains to be explored in Brazil. However, the fact that most Native American alleles are present in REDOME (78% of endemic alleles were observed; https://genevol.ib.usp.br/wp-content/uploads/2020/05/AFND.html), and that their genetic diversity is very low, makes it likely that the finding of Pappas et al. (69) will also apply to Brazil, with match rates for Native Americans being similar to those of Europeans.

The nature of REDOME’s data provides further challenges to our analyses. The data results from samples collected over a span of 30 years, resulting in substantial heterogeneity in the resolution of typing, and the number of loci typed per individual (). The samples collected earlier are at low resolution, whereas the most recent samples have the medium and high-resolution (including Next Generation Sequence data). As a consequence, our results for low and medium resolution, and across different numbers of loci, involve different sets of individuals and sample sizes. Given our focus on the role of ancestry and self-identification, we investigated whether there are differences in the distribution of IBGE categories across the data at different levels of resolution (or typed at a different number of loci). We found that the differences are very small (), and therefore should not affect our findings.

Over most of the time of REDOME’s operation, newly recruited voluntary donors were only typed for HLA-A, -B and -DRB1, with genotyping at HLA-C and -DQB1, as well as high resolution typing conditioned to an initial match at 6/6 (or 5/6). Throughout the paper, we analyze 6/6, 8/8 and 10/10 low resolution typing, despite the fact that successful transplantation requires full or partial matching at high resolution. Our results for medium resolution, 8/8 and 10/10 matching show that the effect of African ancestry becomes even stronger when additional loci are used. While the numbers presented cannot be used directly for clinical practice, the trends we identify are expected to hold in an analysis with high-resolution typing. We therefore believe that the true impact of ancestry on chances of finding a match may in fact be found to be even stronger than we report, when high resolution data for 5 loci become widely available in REDOME.

A final caveat refers to the fact that our study only queries the presence of a matching in REDOME, whereas the chance of finding a donor also involves the ability to contact the potential donors, their willingness to perform the transplant, confirmatory HLA typing at high resolution, and a medical evaluation of the donor’s health. When these factors were taken into account, Gragert et al. (2014) (8) found that in the US donor availability for “Black” patients was 23%, compared to 51% for “White” patients. Since we do not have all this information for the Brazilian sample, we were unable to address these effects. However, the results of Gragert et al. (2014) (8) indicate that donor availability will be lower than the match rates we report, and that the effect of African ancestry on the chances of finding a match may be even greater than we document.

Conclusion

We have shown that among patients seeking HSCT, those with a higher African ancestry face greater difficulty in locating potential donors. Given that more than half of the Brazilian population self-identifies as “Mixed” or “Black” [45.4% and 9.1%; (40)], and that these categories on average have increased African ancestry, this implies a reduced access to HSCT for a large proportion of Brazilians. Difficulty in obtaining a donor for HSTC is one among other forms of inequality faced by “Black” or “Mixed” Brazilians, who also carry a higher burden with respect to infant mortality (70), maternal mortality, risk of stroke (71), and more recently in the proportion of deaths due to COVID-19 (72). In these cases, self-identification is mainly a predictor of mortality as a consequence of the association to socio-economic status (71). In the case of access to donors in REDOME, we show that African ancestry is correlated with decreased chances of finding a match as an outcome of higher population-level diversity among African populations, associated with an underrepresentation of Brazilians of African ancestry in REDOME.

REDOME has a substantially higher frequency of “White” (54.6%) than “Mixed” or “Black” individuals (23.4% and 7.2%, respectively), implying a marked deficit of non-white categories in relation to that of Brazil as a whole. To illustrate a way to address this deficit, consider that individuals with two chromosomes with an African MHC have more matches to “Blacks” than to “Whites” (27.4% vs. 26.6%), even though “Blacks” represent only 7.2% of REDOME. This indicates that an increase in the proportion of individuals with African ancestry within REDOME will decrease the inequality in access to HSCT. This can be achieved, for example, by directing recruitment of new voluntary donors to regions where African ancestry is greater.

Data Availability Statement

Publicly available datasets were analyzed in this study. This data can be found here: The data reported in this paper have been deposited in the European Nucleotide Archive European Nucleotide Archive (PRJEB26388 (ERP108374)), under EPIGEN Committee Controlled Access mode. And in the dbGAP (phs001972.v1.p1).

Ethics Statements

The studies involving human participants were reviewed and approved by Comissão Nacional de Ética em Pesquisa - Brazil (CONEP), the local ethics committees at each participating center, the Institutional Review Boards at the University of California, San Francisco, the REDS-III data coordinating center, RTI International, approved the study. The patients/participants provided their written informed consent to participate in this study.

Author Contributions

KN, LP, and DM designed the study. KN, VA, MS, and AS analyzed the data. KN, VA, and DM wrote the manuscript. DO, CD, FK, ET-S, VR, AC-P, PL, MF-P, CM, SK, BC, BW, ES, and LP discussed, reviewed, and edited manuscript. All authors contributed to the article and approved the submitted version.

Funding

This work was supported by the United States National Institutes of Health - NIH (R01 GM075091; KN, VA, DM, BW); National Institutes of Health, National Heart, Lung, and Blood Institute (HHSN268201100007I; REDS-III); Conselho Nacional de Desenvolvimento Científico e Tecnológico - CNPq and Departamento de Ciência e Tecnologia da Secretaria de Ciência, Tecnologia e Insumos Estratégicos do Ministério da Saúde - Decit/SCTIE/MS (ET-S); Brazil Health Ministry – Transplant National System (proc. 25000.210075/2012-70; LP). São Paulo Funding Agency - FAPESP (2012/09950-9 KN and 2012/18010-0 DM). Minas Gerais State Research Agency (FAPEMIG RED-00314-16 ET-S).

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.