The HLA genes are the most polymorphic in the human genome, and they sit tightly linked on chromosome 6, so alleles at neighboring loci are inherited together in characteristic combinations called haplotypes. Most published haplotype tables are estimates: statistical algorithms such as Expectation-Maximization (EM) reconstruct haplotypes from unphased genotypes. That works well for common haplotypes in samples of millions, but the authors cite evidence that only 39 percent of HLA-A ∼ B ∼ DRB1 haplotypes matched between family-segregated assignments and EM-imputed ones in a comparable study, a striking figure given the diversity of the HLA system in small samples.

GERBase takes the opposite approach. Its database contains 17,260 haplotypes from 8,630 unrelated candidates for hematopoietic cell or kidney transplantation, all typed at the IGEN-AFIP laboratory in São Paulo between 2017 and June 2026. In 98.5 percent of cases the haplotypes were segregated by direct observation within the immediate family, and in the remainder within the extended family. Any locus whose segregation could not be definitively determined was simply excluded rather than guessed. The result is a dataset in which every haplotype was defined by allele segregation and cannot be the product of inference error.

The typing itself is high-resolution and modern. All samples in the fully phased subset were analyzed by next-generation sequencing, using kits such as AllType 11 Loci, AlloSeq Tx 17, Holotype HLA, and Mia Fora, sequenced on Illumina MiSeq and, more recently, on DNBSEQ-G99 and Oxford Nanopore PromethION platforms. Of the 17,260 haplotypes, 9,026, or 52.3 percent, are phased across all nine classical loci, spanning HLA-A, -C, -B, -DRB3/4/5, -DRB1, -DQA1, -DQB1, -DPA1, and -DPB1, and 67.4 percent are phased across the five-locus combination HLA-A, -C, -B, -DRB1, and -DQB1. The diversity is enormous: 4,562 of the fully phased haplotypes were observed only once, giving a haplotype diversity of 0.99972.

A critical validation step was to ask whether transplant candidates adequately represent the broader Brazilian population. The authors compared GERBase allele frequencies with the largest available high-resolution Brazilian reference dataset, the Common, Intermediate, and Well-Documented (CIWD)-Brazil catalog derived from the REDOME voluntary bone marrow donor registry. Across 238 alleles at eight loci, the Pearson correlation was 0.9988 with P < 0.0001, and the regression slope was close to unity. Only 17 alleles differed significantly after Holm-Bonferroni correction, and just five alleles classified as common in CIWD-Brazil were absent from GERBase, all at frequencies of 0.01 to 0.02 percent, consistent with sampling rather than a coverage gap.

The web tool is built around a design decision the authors emphasize: the user defines which loci are displayed, not the tool. Queries can combine specific alleles with a wildcard character that leaves a locus open, returning the distribution of possibilities at that locus. Results are normalized to antigen recognition domain (ARD) groups, a representation that collapses alleles differing only outside the antigen recognition region, yet the tool retains the ability to inspect the individual alleles within each group, including second- and third-field variants. Stratification by self-reported race, following Brazilian census categories, is also supported.

Linkage disequilibrium metrics are a central output. For any allele pair the tool reports Lewontin’s normalized D′, the observed haplotype frequency by direct counting, the expected frequency under linkage equilibrium calculated as the product of individual allele frequencies, and the statistical significance of the association by chi-square or Fisher’s exact test with Holm-Bonferroni adjustment. A color-coded arrow system visualizes the results: black arrows mark pairs in equilibrium, orange arrows mark positive D′, blue arrows mark negative D′, arrow length is proportional to the absolute value of D′, and direction points from the less to the more frequent allele. Hovering over an arrow reveals the full statistics for that pair.

The practical applications are concrete. In one case, a multiplex genotyping assay returned an apparently homozygous HLA-DQA101:02 result together with HLA-DRB107:01 and HLA-DQB102:02. A GERBase query showed that DRB107:01 ∼ DQB102:02 haplotypes in this population almost invariably carry HLA-DQA102:01, making the observed combination improbable and suggesting allele dropout; repeat testing confirmed an HLA-DQA102:01 allele the first assay had failed to amplify. In another scenario, a transplant candidate showed strong reactivity to HLA-C06:02, but the donor had only HLA-B typing available. Entering the donor’s HLA-B information into GERBase revealed a probability below 1 percent that a donor carrying HLA-B35 and -B51 also carried HLA-C*06:02, supporting a negative virtual crossmatch and the acceptance of the organ offer. Conversely, for a patient with strong antibodies against Cw0401, the same query returned a probability above 85 percent of a donor-specific antibody, and the offer could be declined.

The tool may also enhance donor prioritization for allogeneic hematopoietic cell transplantation. In the Brazilian Unrelated Donor Registry, search reports often contain only HLA-A, -B, and -DRB1 at medium resolution, a limitation common in the umbilical cord blood setting as well. GERBase can help infer missing HLA-C and HLA-DQB1 alleles in donor-recipient pairs, informing which unrelated donors or cord blood units should be prioritized for high-resolution typing, and can anticipate potential donor-specific antibodies against HLA alleles absent from incomplete search reports.

The authors acknowledge limitations: the database is small relative to HLA diversity, race is self-reported and subject to reporting bias, ARD grouping affects LD estimates, and regional variation within Brazil was not evaluated. Yet the validation against REDOME is compelling, and the observation-based view of haplotype structure offers a complement to registry-scale statistical estimates. GERBase is accessible at the IGEN website, requires no registration, is free to use, and the underlying dataset is publicly available on Zenodo. Annual updates are planned, making this a living resource for histocompatibility laboratories in Brazil and beyond.

Subject of Research: Immunogenetics

Article Title: GERBase: an online tool for high-resolution HLA allele and haplotype frequencies and linkage disequilibrium analysis in a Brazilian population

Article References: da Silva, J. S., Alves, G. V., Campos, J. H. C., Rampim, G. F., Lima, A. C. M., Gerbase-DeLima, M., & de Marco, R. (2026). GERBase: an online tool for high-resolution HLA allele and haplotype frequencies and linkage disequilibrium analysis in a Brazilian population. Frontiers in Genetics, 17, Article 1973772. https://doi.org/10.3389/fgene.2026.1973772

Image Credits: AI Generated

DOI: 10.3389/fgene.2026.1973772

Keywords: HLA, immunogenetics, transplantation, haplotypes, linkage disequilibrium, Brazil, bioinformatics, GERBase, online, tool, high-resolution, allele