An exome-wide study of renal operational tolerance

Background Renal operational tolerance is a rare and beneficial state of prolonged renal allograft function in the absence of immunosuppression. The underlying mechanisms are unknown. We hypothesized that tolerance might be driven by inherited protein coding genetic variants with large effect, at least in some patients. Methods We set up a European survey of over 218,000 renal transplant recipients and collected DNAs from 40 transplant recipients who maintained good allograft function without immunosuppression for at least 1 year. We performed an exome-wide association study comparing the distribution of moderate to high impact variants in 36 tolerant patients, selected for genetic homogeneity using principal component analysis, and 192 controls, using an optimal sequence-kernel association test adjusted for small samples. Results We identified rare variants of HOMER2 (3/36, FDR 0.0387), IQCH (5/36, FDR 0.0362), and LCN2 (3/36, FDR 0.102) in 10 tolerant patients vs. 0 controls. One patient carried a variant in both HOMER2 and LCN2. Furthermore, the three genes showed an identical variant in two patients each. The three genes are expressed at the primary cilium, a key structure in immune responses. Conclusion Rare protein coding variants are associated with operational tolerance in a sizable portion of patients. Our findings have important implications for a better understanding of immune tolerance in transplantation and other fields of medicine. ClinicalTrials.gov, identifier: NCT05124444.

Background: Renal operational tolerance is a rare and beneficial state of prolonged renal allograft function in the absence of immunosuppression. The underlying mechanisms are unknown. We hypothesized that tolerance might be driven by inherited protein coding genetic variants with large e ect, at least in some patients.

Methods:
We set up a European survey of over , renal transplant recipients and collected DNAs from transplant recipients who maintained good allograft function without immunosuppression for at least year. We performed an exome-wide association study comparing the distribution of moderate to high impact variants in tolerant patients, selected for genetic homogeneity using principal component analysis, and controls, using an optimal sequence-kernel association test adjusted for small samples.

Introduction
Renal transplantation is considered the best treatment option for end-stage renal disease. However, graft survival comes at the price of life-long immunosuppression that may cause cardiovascular disease, cancer and infection. Exceptional patients will however maintain long-lasting allograft function, in spite of complete discontinuation of immunosuppression (1)(2)(3). The absence of destructive immunological response toward the allograft, contrasting with global immunocompetence and a normal response against other immunological challenges is referred to as 'operational tolerance'. Renal operational tolerance is associated with prolonged allograft and patient survival (1,2,4,5). The discontinuation of immunosuppression in tolerant patients generally results from patient non-compliance or from medical decisions grounded on a life-threatening infection or cancer. However, under current transplantation protocols, and despite non-adherence being so common (6) spontaneous tolerance is very rarely observed. In a recent European survey, clinicians identified tolerance in 3/10,000 transplanted patients (2). Because non-compliance is usually undisclosed and in the absence of reliable biomarkers, most tolerant kidney recipients are discovered by chance, suggesting that the incidence of tolerance may be underestimated. Inducing renal tolerance in human recipients is possible through strategies of combined kidney and bone marrow transplantation, resulting in transient chimerism (7). Unfortunately, these strategies are challenging and far from routine clinical implementation. Thus, a better understanding of the molecular pathways and mechanisms involved in spontaneous tolerance might help design innovative tolerance induction protocols and medications.
Most studies of renal operational tolerance to date aimed at identifying reliable biomarkers for the detection of patients with reduced immunosuppression needs or for monitoring tolerance induction protocols [for review, see (3)]. Such studies included antigen-specific functional assays, circulating cell immuno-phenotyping, transcriptomic analyses and mechanistic studies. All together they showed that operational tolerance relies on the expansion of diverse alloreactive immune cells with inhibitory phenotypes and suppressive properties that maintain graft protection against inflammation injury. It remains unclear however whether the changes observed were causes or consequences of tolerance and currently, nothing explains why a kidney recipient initially receiving standard immunosuppression will eventually develop a form of immunoquiescence toward an alloreactive graft.
Because various extreme phenotypes are enriched in traitcausing alleles (8)(9)(10)(11)(12)(13)(14), we hypothesized that some tolerant patients may carry inherited, high impact gene variants allowing tolerance to develop. We reasoned that even Mendelian or near-Mendelian inheritance of tolerance would go unnoticed, because the phenotype is only revealed by transplantation, which is never performed experimentally in asymptomatic relatives of transplanted patients. As most high impact variants are found in protein-coding sequences of the genome (15), we sampled one of the world's largest DNA collections of tolerant patients (n = 40) and performed an exome-wide association study.

Eligibility criteria
Tolerant patients (i) were transplanted with an allogenic kidney; (ii) were 18 y or older; (iii) maintained good allograft function (serum creatinine <1.7 mg/dL and proteinuria ≤1 g/day or /g creatinine; or values above these limits but stable with a maximum variability of 20%) in the absence of immunosuppression for at least 1 year. Patients with a history of bone marrow/other solid organ transplantation or who had been included in a protocol of tolerance induction were not eligible. Control patients were (i) healthy persons or persons with a condition unrelated to kidney or immunological disease; (ii) of European ancestry according to the investigator. All participants signed a written informed consent form approved by their local institutional review board.

DNA samples
Following a large European survey involving 256 renal transplantation centers in 28 countries (2), we collected DNA samples and clinical data from 40 tolerant patients originating from 12 countries. Twenty-two patients' DNA samples were available at Nantes University, FR (IRB protocol number: RC14_0431) and 18 other patients were prospectively sampled as part as the TOlerance MOlecular and Genome-wide studies with Renal Allograft recipient Material [TOMOGRAM study, launched by the ERA-EDTA DESCARTES working group (http://wwwa.era-edtaworkinggroups.org/en-US/group/ descartes)]. The samples originating from the Nantes University biocollection were sequenced over 100 base pairs from both ends on an Illumina HiSEq2000 (Aros Applied Biotechnology, DK) with a mean depth of 68 ± 7. The remaining 18 patients DNAs were paired-end sequenced over 125 base pairs on an Illumina HiSeq 1500 platform (Brightcore, BE) with a mean depth of 121 ± 28. The exome sequencing data from the 40 tolerant patients were compared with the dataset of 197 in house controls previously sequenced the same way as the TOMOGRAM cases. Details on DNA sequencing and variant analysis are available in the Supplementary material. Characteristics of tolerant patients are given in Supplementary Table 1.

Principal component analysis
PCA of exome data is a commonly used multivariate analysis that reduces the data's dimensionality while their covariance remains preserved. This analysis reduced the high dimensional dataset to a small number of dimensions termed principal components. Here we performed a principal component decomposition of the whole set of exome data (n = 85,899 variants after applying the filters described above), without LD pruning, to identify outliers due to ancestry using the function snpgds PCA of the R Bioconductor package SNPRelate v1.12.2.2.

Case-control analysis
The rarity of operational tolerance-our cohort of 40 patients is one of the world largest-(3) represents a major obstacle to the exhaustive study of its underlying genetics. However, extreme phenotypes are often driven by highly penetrant genetic variants of large effect making small (9,13) or very small samples (16-18) sufficient to identify the causative gene. We reasoned that our limited number of patients would allow for discovering such variants, if present, and may encourage building larger cohorts in the future. To demonstrate our hypothesis, we built a pilot study comparing 40 tolerant patients to 197 controls from the general population with the objective to demonstrate enrichment of variants with large effect in the tolerant patients. This design called "single extreme vs.
control" has proven to be an effective and reasonable method to increase the study power without resorting to a high number of, less accessible, exome-sequenced patients with the opposite phenotype (19). Healthy controls are specially relevant when we study a rare trait because the contamination in the control population will remain low with little or no impact on the study power (18). Finally, given the high number of genes/variants tested, we reported both false discovery rates (FDRs) and pvalues following Bonferroni correction as measures of statistical significance after correction for multiple hypothesis testing (20).
In practice, we filtered exome data for moderate to highimpact variants affecting exonic, gene promoter or splice-site regions (for details see Supplementary material) and performed a principal component analysis (PCA) to identify outliers by ancestry. We studied exome-wide association using an optimal sequence kernel association test with small sample adjustment (aSKAT-O) which behaves well over a wide range of allele frequencies (21). This test can detect differences in both rare and common variants but upweights the contribution of rare variants that are likely trait-causing. The aSKAT-O can furthermore be used as a burden test or a variance component test by changing the kernel weight (Burden configuration when Rho = 1; SKAT configuration when Rho = 0) (Supplementary Table 2). We introduced principal .
/fmed. . components (PCs) 1 to 3 as covariables into our aSKAT-O model in order to correct for a possible population stratification artifact and an additional PC issued from the comparison of patients according to the sequencing machine in order to correct for a possible batch effect. We performed both bygene analysis, with the aim to capture allelic heterogeneity, and by-variant analysis.

Control of the variants
All genes and variants arising from aSKAT-O analyses had to pass three control steps to ensure they were true and not resulting from a batch effect due to the use of two different sequencers. Firstly, all variants were carefully examined (at the level of the BAM files) for any irregularity in the alignment .
/fmed. . processes. All variants but one (rs529437974-lack of material) were confirmed by Sanger sequencing. Secondly, a new aSKAT-O was performed with the sequencer as an outcome (instead of the phenotype) and the genes or variants predicting the sequencer were excluded (Supplementary Figure 1). Thirdly, we performed a new SKAT analysis comparing our controls (n = 192) to the tolerant patients after exclusion of the 9 non-European (3 Maghrebin, 6 Turkish) (n = 31) in order to rule out a spurious genetic association due to population stratification. Of note, three variants shared by a pair of tolerant patients each were further investigated to rule out a family relationship between the carriers that could bias the genotype-phenotype association. Patients carrying the same variants were compared for the proportion of single nucleotide polymorphisms (SNPs) with a minor allele frequency (MAF) > 5%) (Supplementary Table 2).

Principal component analysis
The filtration steps (see Study Flow Chart in Figure 1) yielded 85,899 moderate to high impact variants in 16,390 genes. PCA of exome data excluded 5 outliers by ancestry out of 197 controls and 4 (T26, T43, T53, and T63) out of 40 tolerant patients (Figure 2).

Per gene and per variant comparisons
We compared the distribution of variants within each gene (n = 16,390) in the 36 tolerant patients vs. 192 controls (Table 1, Figure 3). We identified two genes HOMER2 (entrez ID 9455) .

Discussion
Overall, our results show a statistically significant association of spontaneous renal tolerance with moderate to high impact coding variants in HOMER2 and IQCH/C15orf61, and a likely association with LCN2. HOMER2 is an actin-binding, scaffolding protein expressed at various specialized primary cilia (23), and a critical regulator of intracellular Ca 2+ signaling (24). Homer2 and 3 compete with calcineurin, resulting in a negative regulation of T-cell activation and Homer2,3deficient mice develop an autoimmune-like pathology (25).
IQCH encodes the IQ motif-containing protein H also known as SPATA17 (spermatogenesis associated 17), a cilia and flagellaassociated protein involved in spermatogenesis (26). IQ motif containing proteins bind calmodulin (27), the key transducer of intracellular calcium signals in T cells, and alter its binding to calcium. C15orf61 (chromosome 15 open reading frame 61) is an uncharacterized coding gene located immediately downstream of IQCH in a non-imprinted portion of chromosome 15. LCN2 is expressed at the primary cilium (28,29). It encodes NGAL (neutrophil gelatinase-associated lipocalin), an iron-trafficking protein involved in multiple immunological processes (30)(31)(32), including common variable immunodeficiency (33), but its role in transplantation remains obscure (34,35). Furthermore, HOMER2, IQCH, and NGAL are all cilia-associated proteins. The T cell cilium is viewed as a specialized compartment that assembles at the plasma membrane upon TCR binding and interaction with the antigen-presenting cell or target cell, producing the so-called "immune synapse" (36,37). The immune synapse is a place of intense TCR/CD3 recycling which directly impacts on the duration and intensity of TCR signaling. This process is dependent upon actin polymerization and intracellular calcium release and is critical for the development of T cell full effector potential (38,39). A spurious association between renal tolerance and the primary cilium resulting from a cluster of ciliopathies in our cohort appears unlikely. While 2/10 variant-carrying tolerant patients suffered from a recognized ciliopathy (Table 2), none of the genes above are associated with a renal disease, either in the OMIM database or in a large whole-exome study (40). Furthermore, our results are consistent with the reported association of tolerance with a coding variant in PARVG, encoding an actin-binding protein (41), and also with a GWAS of the mirror phenotype, acute rejection of renal allograft, that showed a signal close to the CCDC67 locus (42), encoding a key protein for ciliogenesis (43). Finally, accumulating reports of immunological disorders linked to dysfunctional actin dynamics at the immune synapse further support the causality of the association (44, 45). It might seem contradictory that patients carrying putative tolerance variants would still be immunocompetent. It is however wellknown that transplantation tolerance in primates cannot be achieved by blocking a single mechanism (46,47). It is therefore likely that renal allograft tolerance results from a combination of the hereditary predisposition that we showed here, and   (48). The unusually high proportion (25%) of donor-recipient pairs that were fully matched for the HLA A, B and DR loci, also argues for a possible contribution of donor-recipient genetic interactions (49). Our study has several limitations. First, because documented cases of allograft tolerance in renal transplantation are extremely rare, we could not replicate our results in an independent cohort. Our study stems from a near-complete survey of more than 218,000 renal transplant recipients in 28 countries, where tolerance was reported in 3 per 10,000 renal recipients (2). Contrasting with other successful genetic projects on common phenotypes (diabetes, chronic kidney disease, and many more), our recruitment was necessarily limited. We designed our study specifically to overcome this issue (SKAT adjusted for small samples, FDR, enlarged control group), resulting in several significant rare variants in biologically relevant genes. While a plethora of GWA studies for many kinds of traits has called for stringent guidelines on p-values and replication cohorts, the number of tolerant patients needed to meet these standards is simply not feasible. Nevertheless, the FDRs of ∼0.03 are highly convincing, indicating only a 3% chance that the results are false discoveries (20). Second, while our study allowed to unmask a few cases of possibly Mendelian toleranceakin to studies of rare Mendelian diseases where only a few patients can uncover the underlying gene (16, 17)-it was underpowered to detect heterogeneous and/or polygenic factors in the remaining patients. We may also have missed variants of large effect as we did not cover every exon of the genome, partly because of the limits of exome capture kits (50). Finally, in the context of a rare trait and a posteriori identification we must acknowledge that data collection could not be exhaustively controlled. Nevertheless, this is the first exome-wide study of tolerance and provides a springboard for candidate gene studies and meta-analyses.
In conclusion, our work strongly suggests that spontaneous renal allograft tolerance stems from moderate to high impact, coding variants of primary cilium genes in at least a sizeable subgroup of patients (10/36, 27%). Further research should confirm our results and investigate the mechanisms involved.

Data availability statement
The bam files of the 38 tolerant patients who consented to further use of their data by all bona fide researchers are deposited in the European Genome-Phenome Archive (EGAS00001007154 and 155).

Ethics statement
The studies involving human participants were reviewed and approved by Erasme-ULB Ethics Committee, Université Libre de Bruxelles, Brussels, Belgium. The patients/participants provided their written informed consent to participate in this study.

Acknowledgments
We are in debt to the patients, clinicians, and study nurses, who took part in this study and especially to the members of the DESCARTES ERA-EDTA working group and the DIVAT consortium for mounting the biocollection. We thank Ms. Prokopova and Mr. Snelders for administrative support, Eric Trepo for interesting discussion, the cytogenetics unit of Erasme Hospital, Brussels, Belgium for preparing cell lines, as well as the biological resource centre for biobanking from Nantes (Nantes Université, CHU Nantes, Centre de resources biologiques (CRB), F-44000 Nantes, France) (BRIF: BB-0033-00040).