Over 570,000 patients had end-stage renal disease (ESRD) at the end of 2010. Treatment modalities include hemodialysis, peritoneal dialysis, and kidney transplantation. Kidney transplant is the most effective treatment for patients with ESRD. However, despite a waiting list of over 90,000 patients, only 10,622 deceased donor kidneys were transplanted in 2010. The median time a patient with blood type “O” listed in 2003 waited for a deceased donor graft was > 5 years. Longer time on dialysis is one of the strongest risk factors for inferior transplant outcomes. The pursuit of kidney donation from living donors is essential to addressing the long and ever-growing waiting list (Scientific Registry of Transplant Recipients, 2011; http://​www.​srtr.​org/​).

Many transplant candidates have potential living donors, but approximately one-third of these potential donations do not lead to a transplant because of blood group type or crossmatch incompatibility. Kidney paired donation (KPD), also referred to as kidney exchange or live-donor paired exchange, provides an opportunity for donors to indirectly donate to their intended recipient, thereby ensuring living donor kidney transplantation without the barrier of incompatibility. Congress defined and supported KPD by passing the Charlie Norwood Living Organ Donation Act, signed in 2007. Support and interest in KPD has markedly grown in the last decade. However, KPD remains grossly underutilized. Nearly all the initial donor exchange transplants in the USA have been performed through independent, regional KPD registries. The logistics of maintaining a multicenter database, crossmatching potential pairs, and timing the transplants have become increasingly complex, raising concerns regarding programmatic efficiency and efficacy (reviewed in Ref. 9).

Transplants are performed using donor–recipient pairs sharing HLA-A, HLA-B, and HLA-DR antigens from one or two haplotypes. The analysis of HLA matching for kidney graft survival over the period of 1985–1999 is based primarily on serologic typing data. The half-life survival for a graft from an HLA-identical sibling is 23 years, compared to a one-haplotype-related donor, which has a half-life of 12.8 years [6, 10].

Data excerpted from the United Network for Organ Sharing (UNOS) transplant registry [9] indicate that HLA serologic matching appears to have an effect on long-term graft survival of kidneys from cadaveric donors. Although the 1-year graft survival is not very different in cases involving a complete match from those who are completely mismatched, after 5 years the gap in percent survival widens significantly, suggesting that the immunosuppressive drugs are potent in avoiding early graft loss due to acute rejections. However, as the drug dosages are tapered over time, the HLA-matching effect becomes significant. Patients receiving organs with zero mismatches for HLA-A, HLA-B, or HLA-DR antigens achieve a graft survival half-life of 11.3 years compared to a half-life of 6.3 years for grafts that were completely mismatched for these antigens. HLA matching is especially beneficial in second transplants and in patients with preformed antibodies. The effect of HLA-A, HLA-B, and HLA-DR matching remains significant even with the most recent forms of immunosuppression [10].

The selection of a recipient for any given random cadaveric donor is based on an HLA-matching algorithm defined by UNOS (http://​www.​unos.​org). The current allocation policy considers the degree of HLA matching at HLA-A, HLA-B, and HLA-DR. The organs are distributed locally first, and if no match is found, they are offered regionally, and then nationally, until a recipient is found. Nationwide organ sharing is mandatory for pediatric and sensitized candidates with zero HLA-A, HLA-B, and HLA-DR mismatches; local and regional organ sharing is based on one or two HLA-DR mismatches. All other match grades are allocated without regard to mismatching. The main purpose of this algorithm is to try to transplant more minority patients while at the same time not reducing overall graft outcome.

HLA typing for heart and lung transplant patients and their potential donors is an important step for donor selection by virtual crossmatching. Virtual crossmatching is accomplished by avoiding HLA class I and class II antibody specificities identified by screening patients’ sera prior to transplant. HLA typing in these patients usually includes low-resolution HLA-A, HLA-B, HLA-C, HLA-DRB1, HLA-DRB3, HLA-DRB4, HLA-DRB5, and HLA-DQB for patients and their donors. In cases where anti-DQA, anti-DP, or allele-specific antibodies are present, HLA typing of donors and recipients also may include HLA-DQA, HLA-DP, or allele-level typing as needed.

The effect of HLA-A, HLA-B, and HLA-DR matching on the survival of heart and lung transplants is statistically significant. In addition, donor-specific antibodies to HLA antigens are associated with graft failure and poor survival. Antibody-mediated rejection appears in about 10–20 % of heart transplant patients, correlating with factors of poor outcome such as increased incidence for hemodynamic compromise rejection, greater development of cardiac allograft vasculopathy, and higher incidence of mortality. The effect of HLA matching for liver transplants is less clear.

Typically, HLA typing for hematopoietic cell transplantation (HCT) initially would include low-resolution typing for HLA-A, HLA-B, HLA-C, HLA-DRB1, HLA-DRB3, HLA-DRB4, HLA-DRB5, and HLA-DQB1 loci. Allele-level typing is subsequently performed in the final identification of suitable donors for HCT. Three different categories of donors usually are considered in the following order of preference: HLA genotypically identical siblings, HLA-mismatched relatives, and matched or mismatched unrelated donors.

The goal when screening for an HLA-matched sibling donor is to identify which of any siblings have inherited the same HLA haplotype from their parents. This requires typing all siblings and, if possible, both parents for HLA-A, HLA-B, HLA-DRB1, and HLA-DQB1. Given the current average family size in the USA of 2.7 siblings, the approximate probability that a patient will have an HLA match within the family is 30–35 %. For those patients who do not have an HLA-matched sibling within the family, an HLA-mismatched relative who shares one haplotype can be considered. However, the clinical data clearly indicate that with increasing disparity for HLA-A, HLA-B, HLA-DRB1, or HLA-DQB1 loci, there is increased risk of graft-vs-host disease (GVHD) following HCT.

When a matched sibling donor does not exist for a patient requiring allogeneic HCT (70 % of cases), searching for extended family members or donors from unrelated bone marrow registries would be the next option. Registries of volunteer bone marrow donors as well as cord blood exist in most developed countries. The largest of these is the National Marrow Donor Program (NMDP) in the USA (http://​www.​marrow.​org).

The probability of finding a matched donor at HLA-A, HLA-B, HLA-DRB1, and HLA-DQB1 depends on the ethnic origin of the patient and the composition and size of the donor registry being searched. Identifying a patient’s haplotypes can help predict the probability of finding matched donors and assist in developing a search strategy, because some alleles and haplotypes are more common than others, and they are distributed at different frequencies in different racial and ethnic groups. When searching for a donor, for some alleles, an allele-level match is more likely to be found among persons of a particular ethnicity [11–14]. The NMDP matching algorithm HapLogic^SM is based on this principle: identifying the donors or cord blood units (CBUs) with the highest potential to match the patient. This allows transplant physicians searching the Be The Match Registry^® (operated by the NMDP) to identify more quickly and efficiently the best immunogenetically matched donor or CBU for their patients. The transplant center is responsible for selecting an unrelated donor or CBU for a patient. The NMDP recommends that, when possible, patients and adult donors (marrow or peripheral blood stem cells) should be fully matched (8 of 8 loci) at high resolution for HLA-A, HLA-B, HLA-C, and HLA-DRB1. Matching for CBUs is less stringent, and the NMDP recommends fully matched (6 of 6) CBUs at HLA-A, HLA-B (antigen level), and HLA-DRB1 (allele level). This does not imply that availability of a partially matched donor or CBU is a contraindication to transplant. Instead, a less-than-optimal match is another risk factor to be considered in developing the patient’s treatment plan.

Today, patients are more likely to find an unrelated donor or CBU than previously possible. Through the NMDP, three potential hematopoietic cell options for patients in need of an unrelated donor can be available: marrow, peripheral blood stem cells (PBSC), or umbilical cord blood. A patient’s likelihood of finding a potential unrelated donor or CBU has increased with the continuing growth and increasing diversity of the Be The Match Registry. As of 2012, the registry includes nine million potential marrow or PBSC donors and nearly 145,000 CBUs. Through international connections, the NMDP searches more than 18.5 million potential donors as of 2012 and nearly 550,000 CBUs. Nearly all (>95 %) patients are able to find at least one potential 4 of 6 HLA-matched CBU on the Be The Match Registry, the largest in the USA, and the majority will find a potential 5 of 6 match. However, patients belonging to racial groups that are not well represented in the registries have a considerably decreased probability of matched donor identification [12].

HLA compatibility affects not only the ability to achieve sustained engraftment following HCT but also the risk of developing acute and chronic GVHD [6, 13, 15]. Posttransplant risk of graft failure, GVHD, and mortality can be affected by quantitative and qualitative characteristics of donor and recipient HLA mismatching. Moreover, studies have reached different conclusions regarding the relative contributions of HLA class I and class II mismatching because of population-based differences in the specific HLA-mismatch combinations between patients and donors ([13] and references therein). In an analysis of HCT for chronic myelogenous leukemia (CML), the risk of graft failure is affected primarily by donor disparity for HLA class I including HLA-C. The incidence of rejection correlates with the number of donor-incompatible alleles. The incidence of graft rejection is 0.7 % for zero, 8 % for a single, and 19 % for multiple class I allele incompatibilities. Donor disparity for class II does not increase the risk of rejection [6, 13].

The level of matching is more critical for HCT than for solid organ transplantation because of the risk of GVHD. GVHD is a major cause of mortality in HCT and is more frequent than allograft rejection. Analysis of HLA and GVHD focuses on mismatches in the host recognized by donor T cells. Generally, host disparity for class II is thought to convey greater risk for GVHD than class I disparity. In an analysis of HCT for CML, patients with a single class II mismatch at HLA-DR or HLA-DQ have a hazard ratio of 1.8 compared to HLA matches. Single class I mismatches are well tolerated with respect to GVHD; however, combined mismatching at class I and class II confers a hazard ratio of 2.0. Allele mismatches for HLA-A, HLA-B, HLA-C, HLA-DRB1, and HLA-DQB1 also are a significant factor for survival. Patients mismatched for more than one class I allele and those mismatched at both class I and class II alleles have a significantly lower survival than patients and donors fully matched for HLA-A, HLA-B, HLA-C, HLA-DRB1, and HLA-DQB1. A single class I mismatch or single class II mismatch does not appear to affect survival [6, 13, 15]. It is likely that different modes of patients’ immunosuppression after transplant would have an impact on the relative risk associated with various HLA loci. Criteria for “permissible” mismatching in HCT are center dependent. However, very few centers would perform HCT across more than one HLA mismatch at HLA-A, HLA-B, HLA-DR, or HLA-DQ. Interestingly, recent studies have shown that GVHD rates and survival outcomes in pediatric and adult patients receiving cord blood transplants with up to two HLA antigen mismatches (i.e., a 4 of 6 match) are similar to those for 6 of 6 HLA-matched unrelated donor marrow transplants, as long as the recipient received an optimal CD34 cell dose adjusted for the recipient’s body weight [16–19].

The extensive polymorphism observed for HLA loci requires a rigorous approach to data analysis [20]. The complex patterns of LD commonly observed for HLA data can make it difficult to identify specific causal genetic variants. Disease association studies of highly polymorphic HLA data may require the consideration of multiple units of analysis, including amino acid, allele, genotype, and haplotype levels, as well as consideration of gene–gene or gene–environment interactions. The selection of the appropriate statistical tests is critical and will be dependent on the nature of the data set as well as the specific research hypotheses being tested; there is no single analytical approach that can be applied to all data.

Population genetic analyses of gene (allele) frequencies, tests for fit Hardy–Weinberg equilibrium (HWE) expectations, and estimations of haplotype frequencies are the primary steps in analyzing immunogenetic data and provide additional means to assess underlying population substructure, as well as additional units of analysis on which association analyses will be performed. Basic methods for these analyses are described below.

In the age of molecular typing techniques, HLA gene (allele) frequencies (GF) no longer need to be estimated from phenotype data. In most cases, GF for HLA data can be obtained via direct counting, where the number of observations for a given allele is divided by the number of chromosomes (2n, where n = sample size) under study.

The Hardy–Weinberg (HW) model is useful for primary quality control (QC) verification of the integrity of genotype data, as molecular genotyping errors can result in both individual genotype deviations and overall (locus-level) deviations from HWE. In addition, HW testing also is useful for detecting sampling errors (see below) in population samples. Confidence in the accuracy of HW testing therefore is crucial for confidence in subsequent analyses, as many analytical methods (e.g., LD and haplotype estimation) are predicated on an assumption of HWE in the data set.

In a HW test, observed genotype counts are compared to those expected under HWE proportions (HWEP), as calculated from a table of all possible genotypes generated using an appropriate statistical method [21]. The relationship between the allele and genotype frequencies under HWEP is given as
 and 
where p _i is the allele frequency of allele A _i and p _j is the allele frequency of allele A _j . Populations in HWE will not depart significantly from these allele and genotype frequencies.

Tests of overall locus-level HWEP compute a p-value to estimate the significance of observed deviations across all genotypes. Significant deviation of observed genotype counts from expected HWEP can result from factors that include sampling errors (the sampling of admixed, stratified, substructured, or some other form of blended populations), inbreeding or other nonrandom mating, natural selection, and genotyping errors. Tests for deviation from HWEP have low power, and significant deviation from HWEP is not common. Genotyping errors (e.g., failure to detect a specific allele, resulting in an excess of homozygotes) are the first consideration when significant deviations from HWEP are detected (especially when such deviations are detected only at a single locus in a multi-locus analysis), rather than the operation of selection, admixture, or nonrandom mating, unless the data are suspected to be from an unusual population.

Haplotype-level analyses are important to studies of the etiology of human disease, selective forces acting on populations, and optimal bone marrow donor registry (BMDR) size. Multi-locus analyses can be used to identify associations between markers and disease loci that are not evident with single-locus analyses. Haplotypes are used for disease association mapping, quantitative trait locus (QTL) mapping, and imputing underlying genetic markers.

Study design and subject recruitment depend on which of the alternative approaches of identifying haplotypes by segregation analysis in families or estimating haplotypes from population samples of phase-unknown unrelated individuals will be applied. Estimated haplotypes and haplotype frequencies (HFs) play a central role in most genetic studies [21].

It is important to understand that haplotype estimation always results in population-level frequency estimates, rather than haplotype assignments for individuals. Due to the limitations of current haplotype estimation methods, estimated haplotypes observed fewer than three times in a population should be considered unreliable.

The case–control study is the most common study design in HLA disease association studies. Case–control studies can be very sensitive to population stratification within sample cohorts. This issue is particularly important for HLA data, as HLA allele frequency distributions can vary considerably between human populations. If study subjects are not chosen with scrupulous attention to homogeneity of ancestral background, population-based genetic differences between cases and controls may be misinterpreted as disease associations.

The results of the basic population-level analyses outlined above can be used in case–control association analyses. Depending on the study hypothesis, the units of analysis may include gene, carrier, genotype, and/or haplotype frequencies in cases and controls. Assurance that the control population genotypic distributions conform to expected HWEP is critical.

A contingency table is used to test the independence of frequency distributions for categorical variables. Tests for heterogeneity between specific groups (e.g., cases and controls) for HLA data can be performed using contingency table testing and a standard Chi-square (χ ²) measure.

A contingency table always is constructed utilizing raw counts, rather than frequency data. The preferred approach in this type of analysis for highly polymorphic HLA data is to first examine heterogeneity in overall allele frequency distributions in cases vs controls for a particular locus, using a 2 × k row by column (r × c) contingency table. If overall heterogeneity is detected at the locus level, the contributions of individual alleles to the overall deviation can be examined in a series of 2 × 2 tests.

Low-frequency alleles pose a particular challenge in traditional χ ² association tests in disease studies. The χ ² test can lead to false acceptance or rejection of the null hypothesis when the expected genotype counts in a contingency table are small. It is not unusual for 30 or more HLA alleles to be observed at a locus, with a wide range of frequencies, and the preferred approach is to combine low-frequency alleles into a single “binned” category, by combining alleles with an expected value of less than five in cases or controls prior to calculation of the χ ² statistic. The χ ² statistic for a contingency table analysis of case–control data for a genetic association is calculated as
where

O _i  = the observed count of allele i

E _i  = the expected count of allele i

The expected count for each cell in the r × c table is calculated as
where

column total = sum of the counts in the column

The degrees of freedom (df) for the χ ² analysis are the number of alleles with expected values in cases and controls of five or greater, plus the combined category, minus 1 (i.e., k − 1). A p-value is obtained by comparing the test statistic to the χ ² distribution for the appropriate df.

Odds ratios (ORs) are used in disease association studies to describe the strength of association between two variables and reflect the ratio of odds of an outcome (e.g., disease) in one group (e.g., individuals with a particular allele) to the odds of this outcome in another group (individuals without that allele). Relative risk (RR) is a different measure that describes the risk of having the outcome of interest relative to exposure (e.g., positive or negative for a particular allele). RR is more applicable to prospective studies, while OR is preferred for retrospective studies.

OR and RR values are only meaningful when presented with a corresponding confidence interval (CI). The standard probability level is generally 95 %, meaning that there is a 95 % chance that the OR or RR value of interest is in that range. Because the null hypothesis of both tests is that OR or RR = 1, a CI that includes 1 is not considered statistically significant.

Logistic regression (LR) analyses also can be used in case–control studies. LR is a form of generalized linear modeling for data with a binary outcome variable, such as case–control data. LR provides a means to develop association models that include the contribution of numerous covariables. LR is a critical tool in the analysis of complex multivariate data sets.

LR analysis always is performed using software, but extreme care must be taken in the coding of HLA data, as most LR software is not designed to handle high levels of polymorphism common in these data sets. A regression analysis produces a model that includes all of the variables that are useful in predicting the (binary) outcome variable. LR provides an OR for each variable involved in predicting outcome and can be particularly useful in HLA studies that must consider a large number of cofactors.