The relationship between donor-recipient genetic distance and long-term kidney transplant outcome

Cohort

Kidney donor and recipient pairs (n=2,094) were recruited via the United Kingdom and Ireland Renal Transplant Consortium (UKIRTC)¹¹. All transplants were deceased donor grafts and paired on the basis of preferential HLA matching and recipient waiting time. Transplants in the UKIRTC cohort took place between 1981–2007. Only data on the first transplant was considered. All individuals were GWAS genotyped as part of the UKIRTC and the Wellcome Trust Case Control Consortium 3 (see 11 for further details).

The following inclusion/exclusion criteria were applied: participants must have 1) been of European ancestry (determined via principal components analysis (PCA) – see Extended data, Supplementary methods for details)¹², 2) have donated/received a renal transplant and 3) be unrelated to level of 3^rd degree relative (PIHAT<0.1 calculated using PLINK v1.9^13,14).

Clinical predictors of graft survival were tested using STATA v13 in a Cox proportional hazards model and significant variables were included in the analysis as covariates. Acute rejection status was available for 1,099 individuals in our cohort. See Extended data, Supplementary methods¹² for details on imputation. Details on HLA typing were published elsewhere¹¹.

Ethics approval and consent to participate

Ethics approval was granted by the Hammersmith and Queen Charlotte's & Chelsea Research Ethics Committee REC No 08/H0707/1.

Genome-wide shared ancestry

We quantified shared genetic ancestry between donors and recipients at a genome-wide scale using three methods.

Method 1: Identity by descent (IBD). IBD here refers to the amount of shared genetic material between two individuals that is inherited from a recent common ancestor. We quantified the proportion of the genome that is IBD between the donor-recipient pairs using the PIHAT function in PLINK^13,14. The data was pruned for linkage disequilibrium using a 1000-kb window size, 100-kb step size and r² threshold of 0.2.

Method 2: Identity by state (IBS). IBS quantifies the number of shared alleles between two individuals. Alleles that are IBS can be inherited from a recent common ancestor or shared by chance. We quantified the proportion of shared alleles across the genome between each donor and recipient pair using the DST measure in PLINK^13,14. Data was pruned for linkage disequilibrium following the same criteria as outlined above for the IBD method.

Method 3: Chromopainter. We created a mosaic of an individual’s genome from the haplotypes of all the other individuals in the dataset using Chromopainter¹⁰. We calculated the matrix of donated haplotype lengths and associated haplotype-donors, for each individual in our dataset (i.e. the “chromolength” output from Chromopainter). The correlation between the chromolength chunks (donated from each individual in our dataset) of the donor and recipient in a given donor/recipient pair was calculated using the “corr” function in Microsoft Excel (with a perfectly matched mosaic genome having a score of 1). The correlation was then used as the measure of shared genetic ancestry.

These three measures of shared ancestry were tested against time to graft failure using a Cox Proportional Hazards Model using the R survival package (version 2.41–3) and the cox.ph function^15,16. We analysed two different survival analysis models; in the ‘full’ model we included significant clinical predictors as covariates, in the ‘baseline’ model we excluded clinical covariates.

Per genic region analyses

We examined whether shared ancestry between donor and recipient pairs at any specific gene, calculated using the IBD PIHAT and IBS DST methods, was associated with graft survival. Hg19 (human genome assembly 19) coordinates of protein-coding loci were obtained using the UCSC Genome Browser Table Browser via the “Genes and Predictions” group on the USCS gene track¹⁷. The coordinates with a 50-kb buffer region added were used to extract the imputed genotype information for each individual in each donor recipient pair using PLINK, creating a file per genic locus. The files were pruned for LD in PLINK (“–indep-pairwise” command) using 1000-kb window size, 100-kb step-size and r² threshold of 0.2. The PIHAT and DST scores were calculated for each of the thinned genic region files using PLINK’s --genome command. The PIHAT and DST scores for each gene were then extracted for each donor recipient pair.

PIHAT and DST scores for each gene was tested against time to graft failure, adjusting for significant clinical covariates, using R’s survival package¹⁵. We corrected for multiple testing via the Bonferroni method. The total number of loci tested was 40,053 but greater than 50% of these were transcripts coded from overlapping regions and so we only considered independent loci (n= 16,765) when correcting for multiple testing.

As sanity check for the genic approach, we assessed whether genic shared ancestry measures could detect the known effect of the HLA on transplant outcome. As the UKIRTC dataset is composed of donor recipient pairs who were paired by preferential HLA matching and thus skewed towards pairs with fewer mismatches, we focused on the subset of transplant pairs with 0, 5 or 6 serological HLA mismatches. Time to death censored graft failure was tested against the total serological HLA mismatch measure, and against the IBD and IBS measures, in a Cox proportional hazards model using R. As this is a replication of the HLA mismatch measures, we did not apply Bonferroni correction and used a standard significance threshold of p=0.05.

Shared ancestry across transmembrane protein coding genes

As a subgroup analysis, we focused on the scores for transmembrane protein coding genes specifically, calculated in donor-recipient pairs using the IBD PIHAT and IBS DST measures (see methods section 5.2). The list of transmembrane protein coding was obtained using the same methods described previously⁹. PIHAT and DST scores were extracted for these loci from the per-genic region results (see methods section 5.3). The PIHAT score was summed across all of the transmembrane genes to create an overall score. This transmembrane PIHAT score was then normalized, so the mean was 0 and standard deviation was 1. The same process was repeated for DST. The normalized transmembrane DST and PIHAT scores were then tested against time to graft failure in a Cox proportional hazards model using the survival package in R. Models were adjusted for significant clinical predictors of time to graft failure (see Extended data, Supplementary Table 1¹²).

We also analysed the same set of transmembrane genes applying a univariate approach, where the PIHAT/DST calculated for each gene was considered separately and tested against time to graft failure as per the previous paragraph. To correct for multiple testing within this sub-group analysis, each gene was corrected for as opposed to each transcript (as some genes had multiple overlapping transcripts). In total, we corrected for 4,873 tests.

Finally, we carried out an exact replication analysis of the nonsynonymous transmembrane mismatch score which was recently reported to predict time to graft failure⁹. Following the protocol as described by Reindl-Schwaighofer et al.⁹, we imputed the HLA eplet mismatches using HLA matchmaker. Next, using VEP we annotated the imputed genotypes. We used the DR-mismatch program⁹ to calculate the nonsynonymous transmembrane mismatch score. We also calculated the overall genotype mismatch score and the overall nonsynonymous mismatch score (i.e. not limited to transmembrane protein encoding genes). Matching the parameters of the analysis presented in 9, we censored graft survival to 10 years post-transplant and excluded individuals with less than 3 months of follow-up post-transplant (see Extended data, Supplementary Table 5¹² for cohort description). We carried out a Cox proportional hazards model against time to death censored graft failure of the interquartile range of the nonsynonymous transmembrane mismatch score including genome-wide mismatch score, genome-wide nonsynonymous mismatch score, HLA serological mismatch, HLA eplet mismatch, donor age, donor sex, transplant centre and IBD (using PIHAT score as measure). We also tested the nonsynonymous transmembrane mismatch score in a univariate model which was not adjusted for additional covariates.

Genome-wide shared ancestry

We set out to test the effect of the three different genome-wide shared ancestry measures (IBS, IBD and Chromopainter) on time to death censored graft failure using a Cox proportional hazard model.

These shared ancestry measures were successfully calculated on 1,815 donor-recipient pairs (see Extended data, Supplementary Figure 1¹² for distributions of Chromopainter, IBD and IBS measures). Of these pairs, 927 had no detectable IBD relationship (i.e. PIHAT = 0) with 888 pairs having a PIHAT greater than 0. Due to the large number of pairs with PIHAT = 0, we carried out subsequent analyses on two groups; 1) those pairs who shared an IBD of less than or equal to 3^rd degree relative, excluding those with no detected IBD and 2) pairs with an IBD of less than or equal to 3^rd degree relative, including those with no detected IBD (see Extended data, Supplementary Table 2¹²).

These shared ancestry measures were then tested against time to graft failure in a Cox proportional hazards model. None of the genome-wide shared ancestry measures were significant in our full (with clinical covariates) or baseline models (without clinical covariates, see Table 1).

Shared ancestry per genic region results

Given the lack of correlation between graft outcome and donor-recipient genome-wide-level shared ancestry measures, we next tested if the shared ancestry between a donor and recipient at any one particular gene could predict graft outcome.

The HLA is a known predictor of graft outcome. To determine if our shared ancestry measures could reproduce the known effect of the HLA on graft outcome, we tested the correlation between IBD and IBS scores calculated across the HLA region, and the clinically reported combined serological HLA mismatch score (at HLA-A, B and DR). Both IBD and IBS were highly significantly correlated with total serological mismatch scores (P = 3.57x10^-223/r² = 0.48 and 5.74x10^-222/0.47, respectively (see Figure 1 and Figure 2).

Figure 1. Correlation between total HLA mismatches and PIHAT across HLA (p = 3.57x10^-223 and r² of 0.48).

Total mismatches were calculated across HLA-A, B and DR using typing by serology. PIHAT (measure of IBD) was calculated for HLA-A, B and DR individual and combined by summing the PIHAT scores for each locus.

Figure 2. Correlation between total HLA mismatches and DST across HLA (p= 5.74x10^-222 and r² = 0.47).

Total mismatches were calculated across HLA-A, B and DR using typing by serology. DST (measure of IBS) was calculated for HLA-A, B and DR individual and combined by summing the DST scores for each loci.

As donor recipient pairs in the UKIRTC were selected on the basis of preferential HLA matching, the data was skewed towards having a low number of HLA mismatches. In this context, we selected donor-recipient pairs who had either zero-, or 5/6 HLA mismatches for regression analyses testing HLA distance with graft outcome. In this subset of donor-recipient pairs, we found that the total serologically typed HLA mismatch was associated time to graft failure (p=0.009, N=186, N events = 40, HR = 1.22, SE = 0.08).

Having reproduced the known predictive ability of HLA serological scores, we next compared, in the same group of individuals, the IBD and IBS HLA scores, to time to graft failure in a Cox proportional Hazards model. IBD and IBS scores were both significantly associated with time to graft failure with more shared IBD, or a greater amount of IBS associated with a longer time to graft failure (see Table 2). When we conditioned on serological HLA matches, the IBD/IBS signals were no longer significant, suggesting that the IBD/IBS scores are capturing similar variability as the serological typing.

The concordance statistic measures the proportion of pairs of individuals in which the individual with the higher-risk predictor had a failure event before the individual with the lower-risk predictor. We tested this statistic for the serological HLA score and the IBD/IBS scores (serological mismatch concordance = 0.56, standard error (SE) = 0.03; IBS across HLA concordance = 0.56, SE =0.05; IBD across HLA concordance = 0.57, SE = 0.05). This indicates that IBS/IBD per gene measures can capture known associations with genes impacting graft outcome.

We then went on to test shared ancestry in individual genes across the entire genome. We compared the IBS and IBD measures calculated from each coding region in the genome (that contained at least one SNP after quality control measures were applied) to the afore-mentioned outcome measures. No gene reached genome-wide significance, for either the IBD or IBS measures (see Extended data, Supplementary tables 9–12¹²).

The HLA-A (IBD p = 0.017 and IBS p = 0.045) and DR (IBD p = 0.009 and IBS p = 0.01) loci came out in the top 5% of the distribution for the time to graft failure analysis, but did not survive correction for multiple testing (see Supplementary table 3).

We also examined the per gene IBD and IBS scores in a baseline model which excluded clinical covariates. No single gene survived correction for multiple testing (n = 1,804, n events = 375; see Extended data, Supplementary tables 11–12¹²).

Shared ancestry across transmembrane protein encoding genes results

Given previously published analysis^8,9 showing a significant association between donor and recipient allelic mismatches across genes encoding transmembrane proteins and graft outcome, we tested the association between the amount of shared genetic ancestry calculated across these transmembrane proteins and time to death censored graft survival. Neither the normalized transmembrane IBD nor IBS score were a significant predictor of time to graft failure even before correction for multiple testing in our baseline (IBD p = 0.50, IBS p=0.35, n = 1,804, n events = 395) or full models (n= 1030, n events = 160, see Extended data, Supplementary tables 13–17¹²).

Transmembrane genes were also analysed in a univariate model, where the IBD/IBS measure for each transmembrane gene was tested against time to graft failure. None of the transmembrane genes reached significance after correcting for multiple testing (corrected for 4783 tests, see Extended data, Supplementary tables 13 –17¹²).

Finally, we went on to attempt to replicate the findings of Reindl-Schwaighofer et al.⁹, who demonstrated a significant correlation between a transmembrane nonsynonymous donor-recipient mismatch score and time to ten-year and death censored graft failure. We did not find a significant correlation between in allogenomic mismatch score and time to graft failure in the univariate or multivariate (adjusted for additional covariates - see Methods section) models (see Extended data, Supplementary tables 7 and 8¹² for results).

Underlying data

The data used and/or analysed in the presented study was GWAS genotype chip array data and clinical data collected as part of the UKIRTC. Due to the sensitivity of the clinical patient data and genotype data used in this study, we are unable to share the data publicly. Access can be requested by researchers at accredited institutions. Access to data can be requested by emailing Peter J. Conlon (peterconlon@beaumont.ie) or Gianpiero L. Cavalleri (gcavalleri@rcsi.ie).

Extended data

Harvard Dataverse: Supplementary Information: The relationship between donor-recipient genetic distance and long-term kidney transplant outcome. https://doi.org/10.7910/DVN/ZXAXTN¹².

File ‘Supplementary Information - shared ancestry manuscript’ (DOCX) contains the following extended data:

Supplementary Tables 9–16 are available as individual DOCX files as part of the repository project.