LILRB3 genetic variation is associated with kidney transplant failure in African American recipients

Abstract

African American (AA) kidney transplant recipients exhibit a higher rate of graft loss compared to other racial and ethnic populations, highlighting the need to identify causative factors. In the Genomics of Chronic Allograft Rejection (GoCAR) cohort, the pre-transplant blood RNA sequencing revealed a cluster of four consecutive missense SNPs, within the Leukocyte Immunoglobulin-Like Receptor B3 (LILRB3) gene, strongly associated with death censored graft loss (DCGL). This SNP cluster (named LILRB3-4SNPs) encodes missense mutations at amino acids 617–618 proximal to a SHP-1/2 phosphatase-binding ITIM motif. LILRB3-4SNPs is specifically enriched within AA individuals and exhibited a strong association with DCGL and eGFR decline in the AA participants from multiple transplant cohorts. In two large Biobanks (BioMe and All-of-Us), the LILRB3-4SNPs was associated with the early onset of end stage renal disease (ESRD) and acted synergistically with the Apolipoprotein L1 (APOL1) G1/G2 allele to accelerate disease progression. The SNPs were also linked to multiple immune-related diseases in AA individuals. Lastly, by multi-omics analysis of blood and biopsies, the recipients with LILRB3-4SNPs showed enhanced inflammation and monocyte ferroptosis. While larger and prospective studies are needed, our data provide insights on the genetic variation underlying kidney transplant outcomes.

Introduction

African American (AA) kidney transplant recipients face a higher risk of graft failure compared to individuals of European or other ancestries^1–3. This elevated risk has been validated in individuals with genetic markers of African ancestry^4,5 while not attributable to higher HLA mismatches compared to other groups¹. Several genome-wide association studies (GWAS) have identified genetic polymorphisms in HLA and non-HLA regions associated with post-transplant graft failure in a mixed population^5,6, whether non-HLA SNPs contribute to the reduced graft survival in these patients remains inadequately tested³.

While SNP arrays have been commonly used for genotyping in GWAS, these arrays do not adequately probe SNPs in exonic regions or in the under-represented populations and do not provide quantitative association analysis of variant alleles with diseases, all of which limit interpretation of findings based on this methodology. RNA sequencing (RNAseq) overcomes these deficiencies by enabling the simultaneous detection of coding variant and quantification of allele-specific expression. Previous application of RNAseq data in SNP genotyping demonstrated consistency between SNP array and RNA sequencing, particularly when there is sufficient expression coverage of exonic regions^7,8. This capability makes RNAseq a complementary tool for identifying functional SNPs and related gene/pathway signatures associated with human diseases in a quantitative manner⁹.

Herein, we employed RNA sequencing of pretransplant blood from kidney transplant recipients to investigate the functional genetic polymorphisms involved in post-transplant graft loss. Through this analysis, we identified a cluster of four consecutive missense single-nucleotide polymorphisms (SNPs) in the close proximity to the immunoreceptor tyrosine-based inhibitory motif (ITIM) of the Leukocyte Immunoglobulin-Like Receptors (LILR) family B3 gene (LILRB3, an inhibitory immune response regulator), named LILRB3-4SNPs that strongly associated with DCGL, kidney disease and other immune mediated disease progression specifically in patients of African ancestry. Through initial mechanistic studies linking the genotype to function we determined that LILRB3-4SNPs augment inflammation and ferroptosis, thereby identifying a therapeutic target that could potentially improve outcomes in these patients.

Results:

Overview of Study Cohorts and Design.

We analyzed individuals from four kidney transplant cohorts, GoCAR^10,11, SIRPA, CTOT19¹², VericiDx, and two Electronic-Health-Record (EHR)-linked biobank cohorts (BioMe¹³ and All-of-Us¹⁴) with detailed cohort description in Methods and demographic/clinical characteristics tables (Extended Data Table 1 and Table S1–3). The ancestries of the GoCAR and CTOT19 cohorts were genetically determined using SNP array or RNA-seq, while other cohorts provided self-reported race (Methods).

As depicted in Figure 1, we analyzed RNAseq profiles of pre-transplant blood from 170 kidney transplant recipients in the GoCAR cohort to identify SNPs associated with DCGL, from which we detected African American specific SNPs. The prevalence and race disparity of the identified SNPs was further evaluated in four transplant cohorts with RNA or targeted DNA sequencing data and two large biobanks with whole exome sequencing (WES) data. The SNP association with transplant outcomes (DCGL and eGFR decline) were further investigated in 261 AA recipients from four transplant cohorts (GoCAR, n=83 AAs; SIRPA, n=54 AAs; CTOT19, n=47 AAs and VericiDx, n=77 AAs). SNP association with end stage renal disease (ESRD) progression or immune related conditions were examined in two biobank cohorts (BioMe, n= 7,096 AAs and All-of-Us, n=50,969 AAs). Last, we performed multi-omics analysis on the pre-/post-transplant blood and biopsies to investigate the functional roles of the SNP and revealed the potential causal link in in vitro system.

Figure 1. Cohort and study design.

4 independent kidney transplantation cohorts (GoCAR (n=264; 83 AAs), SIRPA (n=54; 54 AAs), CTOT19 (n=128; 47 AAs) and VericiDx (n=77; 77 AAs)), and two EHR-linked biobanks, BioMe (n=30,099; 7,096 AAs) and All-of-Us (n=245,388; 50,969 AAs) were included in this study. Pre-transplant blood RNAseq (n=170) in conjunction with SNP array data (n=588) were initially used to identify the AA-specific expressional SNPs (LILRB3-4SNPs) associated with post-transplant renal failure and eGFR decline, validated in SIRPA, CTOT19, VericiDx and BioMe. The association of LILRB3-4SNPs with the ESRD progression or other immune related diseases was evaluated in BioMe and All-of-Us AA cohort. The functional roles associated with LILRB3-4SNPs were investigated by meta-analysis on whole blood transcriptome by RNAseq (GoCAR, CTOT19 and VericiDx), single-cell RNA sequencing of pre- and post- transplant PBMC, post-transplant biopsies and in vitro functional experiments in THP1 macrophages cell line.

Graft failure associated allele specific expression.

We developed a bioinformatic pipeline to detect and quantify exonic SNPs from RNA sequencing data within the GoCAR cohort with a high sensitivity and specificity of SNP detection comparable to the SNP array (Extended Data Figure 1). From DNA genotyping arrays (n=588)¹⁵ and pretransplant blood RNAseq (n=170)¹¹ data in GoCAR cohort, 410,855 exonic SNPs were extracted for allele-specific expression analysis (n=170)¹¹ (Methods). At each SNP, we quantified the alternative allele expression fraction (AEF) as the fraction of the reads supporting alternative allele among all the reads supporting both alternative and reference alleles of the SNP (Methods). Within these exonic SNPs, 1,438 SNPs exhibited significant associations of AEF with DCGL by Cox univariate regression with a p value ≤ 0.05 (Figure 2A). This list was subsequently refined to 342 after being adjusted by baseline and post-transplant factors including recipient’s genetic ancestry, donor status (living/deceased donor), donor age, HLA mismatch, induction therapy, delayed graft function (DGF), post-transplant infection, acute rejection^15–18 (Table S4). These refined SNPs were situated within 283 distinct genes, displaying a significant enrichment in immune-related functions, including neutrophil degranulation, macrophage activation, and interferon gamma response (Figure 2B and Table S5). These genes were also over-represented in the genes associated with neutrophil counts, Crohn’s disease and rheumatoid arthritis in GWAS studies from GWAS Catalog¹⁹ (Figure S1). Upon tallying the count of significant SNPs within each gene region, multiple members of LILR gene family (LILRA2, LILRB1, LILRB3 and LILRA1 with immune regulatory functions²⁰) occupied prominent positions within the top 20 genes (Figure 2C and Table S4).

Figure 2. The association of expressional SNPs (eSNPs) in the pretransplant blood with graft loss in the GoCAR cohort.

A) Manhattan plot of the eSNP association (- log10(P value)) with DCGL by univariate Cox regression model in the GoCAR RNAseq cohort (n=170); Each dot represents a SNP aligned with its genome coordinate. B) The enriched functions of the genes harboring significant eSNPs (P value ≤ 0.05); Enrichment was evaluated by one-sided hypergeometric test. C) The occurrence of significant eSNPs in the top 20 recurrent genes.

Identifying LILRB3-4SNPs as a risk factor for graft failure.

Among the 342 DCGL-associated SNPs, 25 SNPs (7.3%) belong to LILR gene families, which was surprisingly over-represented compared to other genes (Figure 2C and Table S4). 16 out of these 25 SNPs are nonsynonymous SNPs that could change protein functions (Table S4). Due to the pivotal role of LILR genes as regulators of immune responses and their established connection with autoimmune diseases²⁰, we focused on discerning significant SNPs within the LILR gene families. Of notable interest was SNP (rs549267286 located at chr19:54721007 of hg19), a missense SNP with a high prevalence in individuals of AA ancestry, which exhibited an intriguing correlation with DCGL (Figure 3A). Upon further inspection of the read alignments covering rs549267286, we found this SNP is situated within a cluster of four highly-linked and inheritable consecutive missense SNPs positioned at chr19:54721006–54721009 (hg19) (rs139094141, rs549267286, rs567676351, and rs761515451, a region that we named LILRB3-4SNPs) (Figure 3B and C, Figure S2 and S3, Table S6–8). Two adjacent synonymous SNPs (rs113314747 and rs60566950) within the LILRB3 exon were also found to be in high Linkage Disequilibrium (LD) with those 4 consecutive SNPs (Figure 3B and C and Figure S2). The remaining three SNPs within this cluster were not imputed via SNP array data or identified through RNA-seq genotyping, potentially attributed to the computational challenges for consecutive SNPs calling²¹.

Figure 3. Identification of a cluster of 4 consecutive missense SNPs in LILRB3 gene (named LILRB3-4SNPs) associated with post-transplant renal failure.

A) Two-way cluster view of the AEF data by 25 DCGL-associated eSNPs (Cox regression, P<0.05,) within LILR gene families (horizontal) and patient’s demographics/outcomes (vertical). The grey color represents missing values); B) Sequencing alignment of reads covering 4 consecutive SNPs and their adjacent SNPs within LILRB3 by IGV (chr19:54720990–54721060, hg19); C) Heatmap of the R² and D’ (R squared and D prime measuring the linkage of SNPs) between the LILRB3-4SNPs and their 2 adjacent SNPs; D) Protein structure modeling of the interaction of SHP-2 with LILRB3 harboring LILRB3-4SNPs reference (left) or alternative (right) allele. E) Protein sequence alignment of LILRB3 orthologs across organisms, human (Homo sapiens), chimpanzee (Pan troglodytes), monkey (Macaca mulatta), dog (Canis lupus dingo), rat (Rattus norvegicus) and mouse (Mus musculus). E617 is conserved across species and close to the fourth immunoreceptor tyrosine-based inhibitory motif (ITIM) in LILRB3 protein.

LILRB3-4SNPs results in amino acid substitution at 617–618 Glu-Pro (EP) -> Pro-Ser (PS). This substitution occurs in the linker region between ITIM3 and ITIM4 of LILRB3, raising the possibility that it could impact the binding of LILRB3 to the SHP-1/2 protein²². A model of the complex between SHP2 and the ITIM3-linker-ITIM4 by AlphaFold suggest that the EP->PS mutation may alter the interactions between the ITIMs with SH2 domain of SHP2^23,24 (Figure 3D). Worth noting is the conservation of the amino acid 617E across the LILRB3 orthologs from diverse organisms (Figure 3E).

LILRB3-4SNPs is predominantly detected in AA population.

As LILRB3-4SNPs cannot be accurately genotyped or imputed from SNP arrays, we developed a targeted DNA sequencing array to assess the SNPs in 127 non-European individuals within the GoCAR cohort. Analyzing the subset of 32 individuals with both RNA and targeted DNA sequencing data verified the genotyping of the variant from both technologies. Among 264 AA recipients with RNA or target DNA sequencing, LILRB3-4SNPs was predominantly detected in AA population (19.3%, 16/83) vs European (1.2% 1/82) (Extended Data Table 2). Using RNA or DNA sequencing, we also identified LILRB3-4SNPs within recipients of the CTOT19 cohort (AA: 10.6%, 5/47 vs European 0.0%, 0/27) and AA recipients from the SIRPA cohort (14.8%, 8/54) and the VericiDx cohort (12.9%, 10/77) transplant cohorts (Extended Data Table 2). Taken together, the overall prevalence of LILRB3-4SNPs in four transplant cohorts is 14.9% (39/261) in AA recipients vs 0.9% (1/109) in European recipients. Next, we confirmed the racial disparity of LILRB3-4SNPs in two large electronic health record (EHR)-linked biobanks with WES data (BioMe: 8.6% (608/7,096) in AA vs 0.1% (8/9,555) in European and All-of-Us: 14.9% (7,571/50,964) in AA vs 0.03% (42/125,843) in European) (Extended Data Table 2)

LILRB3-4SNPs associates with DCGL and eGFR decline.

When we stratified patients into two groups based on the AEF of LILRB3-4SNPs in the GoCAR RNAseq cohort: Risk (AEF ≥ mean AEF) and Non-risk (0 ≤ AEF < mean AEF), the high expression of this SNP was predominant in the non-European population (AA + Hispanic), with a notably much higher frequency within the AA population (Figure 4A). Patients demonstrating relatively elevated expression of LILRB3-4SNPs exhibited a significantly heightened risk of graft loss within the entire RNAseq cohort (n=170, p < 0.0001, Figure 4B), within non-European recipients (n=75, p=0.0018, Figure S4A), and within AA recipients (n=40, p=0.00091, Figure 4C). It was worth noting that the overall expression of LILRB3 was not associated with graft loss (Figure S4B), implicating that the LILRB3-4SNPs contributes to graft loss through the modulation of LILRB3 function rather than the regulation of LILRB3 gene expression. Next, we scrutinized whether the DNA genotype of this variant correlates with graft loss and further discovered a significant association between the LILRB3-4SNPs and DCGL within the 127 non-European individuals with LILRB3-4SNPs genotyping performed by targeted DNA sequencing (n=127, p=0.019, Figure S4C), as well as within AA samples (n=52, p=0.025, Figure S4D). As anticipated, the LILRB3-4SNPs genotype from the combined RNA and targeted DNA sequencing cohorts demonstrated a notable association with graft loss in AA (n=83) (p=0.00074, Figure 4D) or AA + Hispanic (p=0.0011, n=169) populations (Figure S4E). Among 83 AA recipients, 16 (19.3%) patients possessed LILRB3-4SNPs, of whom 11 (68.8%) lost their grafts within 5 years post-transplant. No significant difference of other demographic and baseline or post-transplant clinical characteristics was found between the AA patients with and without LILRB3-4SPNs risk allele (Table S10). As APOL1 exonic variants (G1, G2) also bear relevance to the AA-recipients^25–28, we assessed the co-occurrence of APOL1 variant alleles with LILRB3-4SNPs and found that APOL1 variant and LILRB3-4SNPs were not correlated by allele expression in the GoCAR cohort (Figure S5) or genetic linkage in the BioMe and All-of-Us biobanks (Table S11–12). By multi-variable Cox regression analysis, LILRB3-4SNPs is associated with graft loss (hazard ratio (HR)=4.42, p=0.005) adjusted for APOL1 risk alleles, donor status, donor age, DGF, post-transplant infection, acute rejection and HLA mismatches (Extended Data Table 3). The logistic regression also showed a strong association between LILRB3-4SNPs and the graft loss with adjustment of the same variables (p=0.031 and odds ratio=7.54, Extended Data Table 3).

Figure 4. The association of LILRB3-4SNPs with post-transplant renal failure in the GoCAR cohort.

A) AEF of the LILRB3-4SNPs within AA (n=40), Hispanic (n=37) and European (n=83) patients in GoCAR pre-transplant RNAseq cohort. The association of LILRB3-4SNPs with DCGL in the entire RNAseq cohort (n=170, B), AA population by RNA sequencing (n=40, C) and AA population by RNA + DNA sequencing (n=83, D). Note: In (B) and (C), “Risk” group represents the patients with AEF ≥ mean(AEF). In (D), “Risk” group represent the patients carrying LILRB3-4SNPs variant genotype determined by RNAseq or DNAseq. E) Meta-analysis of the association of LILRB3-4SNPs with DCGL in the GoCAR and SIRPA (n=54) cohorts (Cox model adjusted by APOL1 risk allele number (not available for SIRPA), donor status, donor age, DGF, ACR, post-transplant infection (not available for SIRPA), and HLA mismatch) (fixed-effect model). The hazard ratio was showed in dot with horizontal bars showing 95% CI (an arrow was shown when the line passed the axis limit). The P value in the KM survival plot was calculated by log-rank test. The full multi-variate Cox model was detailed in Extended Data Table 3 and Table S15.

As validation, we tested the association between LILRB3–4 SNPs and DCGL in 54 AA recipients from an independent cohort (SIRPA), an observational transplant cohort with a up to 7-year follow-up. This analysis confirmed a higher incidence of graft loss in SNP+ recipients (50%, 4/8) vs SNP− patients (21.7%, 10/46), with an association at HR=3.44, p=0.058 by Cox model and OR=7.35, p=0.044 by logistic regression model, adjusted for donor status, donor age, DGF, acute rejection and HLA mismatch (Table S14). As a note, APOL1 genotype and post-transplant infection status were not available for the SIRPA cohort. As induction therapy was administered to all AA recipients, it was not adjusted for in the DCGL association analysis in the GoCAR and SIRPA cohorts. A meta-analysis of both cohorts confirmed the significant association of LILRB3–4 SNPs with DCGL (p=0.0008 and HR=4.00) (Figure 4E). We did not evaluate the DCGL/LILRB3–4SNPs association in the CTOT19 or VericiDx cohorts due to the limited number of DCGL cases within the ≤2 years follow-up time.

Since early eGFR decline is a biomarker associated with late graft loss^29–31, we next examined the SNP association with low eGFR (<30, ml/min/1.73 m²) or eGFR decline (>30%) within 2 years across four transplant cohorts (GoCAR, SIRPA, CTOT19, VericiDx, total n=261). Increased cases of declined eGFR were observed in SNP+ vs SNP− recipients in all four cohorts, with an odds ratio >1.5 by logistic regression after adjusting for donor status, donor age, DGF and HLA mismatches (Table S15). The combined odds ratio is 4.06 at p=0.006 by meta-analysis of the four transplant cohorts (Table S15). As an independent validation, we also observed that SNP+ kidney transplant recipients (n=9) in the BioMe biobank experienced accelerated eGFR decline over 2-years post-transplant compared to SNP− patients (n=90) (Extended Data Figure 2).

Together, these data indicated that LILRB3-4SNPs represent a distinct, strong genetic risk factor for graft failure in AA kidney transplant recipients.

LILRB3-4SNPs is linked to the progression of kidney diseases.

Individuals of African ancestry have a higher prevalence of CKD (particularly ESRD) than other populations and CKD is commonly associated with immune dysfunctions³². As LILRB gene family members play a central role in regulating immune responses²⁰, we hypothesized that expression of LILRB3-4SNPs impacts the development and/or severity of immune-related conditions beyond allograft failure, including kidney diseases. To explore this, we first tested for associations between LILRB3-4SNPs and kidney diseases (defined by ICD code) in AA participants (Methods). However, we did not find a significant association between LILRB3-SNPs and the prevalence of kidney diseases (Table S16). To explore whether LILRB3-4SNPs might affect kidney disease progression rather than disease prevalence, we evaluated historical eGFR values in each patient and estimated the age with the first significant decline in eGFR value (<15, ml/min/1.73 m²) in ESRD patients in both BioMe (n=489) and All-of-Us (n=250) cohorts. Survival analysis showed that ESRD patients carrying LILRB3-4SNPs experienced a significantly earlier decline compared to those without the risk alleles in the AA population within the BioMe cohort (p=0.013, HR=1.57, Figure 5A left) and the All-of-Us cohort (p=0.047, HR=1.61, Figure 5A middle), as well as in a meta-analysis combining both cohorts (p=0.002, HR=1.58, Figure 5A right).

Figure 5. The association of ESRD onset (the age with eGFR decline <15, ml/min/1.73 m²) with LILRB3-4SNPs.

(A), APOL1 risk alleles (B) and combined genotypes (LILRB3-4SNPs +APOL1 double risk alleles) (C) of the ESRD patients in the BioMe (n=489 ESRD, left), All-of-Us (n=250 ERSD, middle) biobanks and meta-analysis on both cohorts (right) (The hazard ratio was showed in dot with horizontal lines showing 95% CI). P value and hazard ratio (HR) were calculated with cox model (versus Non-risk group: no risk allele) and the meta-analysis was conducted with fixed-effect model (Methods). APOL1 genotype: single risk allele (G0/G1 or G0/G2) and double risk alleles (G1/G1, G2/G2, or G1/G2))

We then examined the association of the APOL1 risk allele with kidney diseases in the BioMe and All-of-Us cohorts. Consistent with previous studies^28,33,34, we first observed the association of APOL1 risk alleles with the prevalence of kidney diseases (Table S16). Secondly, the presence of two APOL1 risk alleles was associated with an earlier progression to ESRD in the BioMe cohort (p=3.85e-04, HR=1.62, Figure 5B left) and the All-of-Us cohort (p=2.96e-03, HR=1.96, Figure 5B middle), as well as in a meta-analysis of both cohorts (p=4.77e-06, HR=1.70, Figure 5B right). After adjusting APOL1 genotype, LILRB3-4SNPs association with early eGFR decline remain significant (BioMe p= 0.013, HR=1.59; All-of-Us p=0.117, HR=1.46; meta-analysis of the two, p=0.004, HR=1.52), suggesting independence of APOL1 risk alleles (Table S17).

Finally, we examined whether patients carrying both LILRB3-4SNPs and APOL1 risk alleles experienced accelerated ESRD progression. When we grouped patients based on the presence of only APOL1 risk alleles, only LILRB3-4SNPs, or combined (with two APOL1 risk alleles and LILRB3-4SNPs), the combined group showed a higher hazard ratio for ESRD progression in the BioMe cohort (p=1.64e-03, HR=2.56, Figure 5C left), the All-of-Us cohort (p=7.19e-05, HR=3.37, Figure 5C middle), and in a meta-analysis of both cohorts (p=5.02e-07, HR=2.93, Figure 5C right), suggesting a synergistic effect of both genotypes on ESRD progression.

Beyond the association with kidney diseases, our analyses showed that LILRB3-4SNPs significantly associated with the prevalence of health conditions related to dysregulated immune functions including asthma^35–37 and depression^38–40 in both BioMe and All-of-Us biobanks (Table S18).

LILRB3-4SNPs enhances inflammation and monocyte ferroptosis.

To gain an insight into potential mechanisms linking LILRB3-4SNPs to kidney graft dysfunction/failure, we undertook multi-omics analyses on the blood and biopsy samples from AA transplant recipients with and without LILRB3-4SNPs.

Meta-differential analysis of three RNAseq datasets from pretransplant blood in AA recipients (GoCAR, CTOT19, and VericiDx) identified 1555 (892 upregulated and 663 downregulated) meta-differentially expressed genes (mDEGs) in the recipients with vs without the LILRB3-4SNPs (Figure 6A–B and Table S19). The upregulated genes displayed a strong connection to functions or pathways that include T/B cell activation, chromatin organization, whereas downregulated genes were linked to functions including erythrocyte/myeloid cell differentiation, oxygen transport, oxidative phosphorylation, and mitochondrial function signatures (Figure 6C). Upregulation of T/B cell signatures and down-regulation of monocyte specific signatures were confirmed by serum profiling (Figure S6 and Table S20). Among these mDEGs, the Ferritin Light Chain (FTL) emerged as the most down-regulated gene and also a key driver that regulates many downstream genes in the gene expression correlation network (Figure S7). Given that FTL functions as a model for iron storage, the reduced expression of FTL can elevate cellular iron levels, consequently fostering ROS accumulation and culminating in ferroptosis-associated cell death⁴¹. Consistent with bulk RNA seq data, single-cell RNA sequencing (scRNAseq) of the pre-transplant PBMC detected increased B/T and decreased monocyte populations in the patients with LILRB3-4SNPs (n=1) vs without (n=1) (Figure 6D). The expression of LILRB3 and ferroptosis-associated genes were predominantly detected in the monocytes (Figure S8). Differential expressional analysis of the immune sub-populations with vs without LILRB3-4SNPs identified up-regulated genes linked to proinflammation response and down-regulated genes involved in negative regulation of ferroptosis in monocytes and the activation of T and B cell signatures (Figure 6E–F and Table S21).

Figure 6. Identification of transcriptomic signatures associated with LILRB3-4SNPs in pre-transplant blood.

A) Volcano plot of meta-DEGs (P<0.05) in the pretransplant blood from AA patients with (GoCAR, n=9; CTOT19, n=5; VericiDx, n=11) versus without (GoCAR, n=24; CTOT19, n=10; VericiDx, n=20) LILRB3-4SNPs in three blood bulk RNAseq cohorts (GoCAR, CTOT19 and VericiDx). Two-sided Z-score test was performed on combined Effect Size (EZ) comparing the profiles between SNP+ vs SNP− patients from three cohorts to identify significant meta-differentially-expressed genes, and corresponding p values were corrected by Benjamini-Hochberg method⁴⁹. B) Dot plot of top 20 up- and down- regulated meta genes across three bulk RNAseq cohorts (ES, effect size, P value < 0.05 from meta-analysis); C) the functional categories enriched with meta-DEGs (red: enriched with up-regulated genes, blue: enriched with down-regulated genes, P<0.05)). D) UMAP of single cell RNA sequencing of pretransplant PBMC from two patients (one with LILRB3-4SNPs risk allele and one without as non-risk) showing the cell proportion of each cell type within each sample. E) Function enrichment of genes significantly up- and down- regulated in monocytes in patients with vs without risk allele. F) Heatmap showing the log2(fold change) of representative DEGs of B cell, T cell and Ferroptosis signatures between SNP+ vs SNP− cells in each cell type. The gene-function enrichment was evaluated with one-sided hypergeometric test showing -log10(P values) in the bar charts in C and E.

We also performed bulk RNAseq on the post-transplant blood within 6–24 months (n=20, 10 LILRB3-4SNPs+ vs 10 LILRB3-4SNPs-) or biopsies at 3 months (n=6, 3 LILRB3-4SNPs+ vs 3 LILRB3-4SNPs-) and scRNAseq (n=6, 3 LILRB3-4SNPs+ vs 3 LILRB3-4SNPs-) on post-transplant PBMC at 24 months from the AA recipients with matched demographic and clinical characteristics (age, sex, induction type, HLA mismatch, donor status). These analyses showed persistent activation of T/B cell signaling pathways and monocyte ferroptosis post-transplant in the patients carrying LILRB3-4SNPs (Extended Data Figure 3 and Table S22).

To test for causal association of the LILRB3-4SNPs expression with enhanced inflammation and monocyte ferroptosis. we performed in vitro functional studies of macrophage cell line (THP-1) overexpressing the LILRB3-4SNPs variant or reference allele. LPS stimulation for 24 hours increased expression of inflammation markers (TNFα, TNFAIP3 and IL1β) and decreased expression of ferroptosis-negatively-associated genes in the cell line overexpressing the variant vs reference allele (Extended Data Figure 4A and B). Cell viability analysis demonstrated a reduced viability upon LPS stimulation in cells with the SNPs which was notably reversed upon addition of a ferroptosis inhibitor (Extended Data Figure 4C and D).

Discussion

Genetic factors may contribute to a high risk of renal failure in AA kidney transplant recipients. Using RNA sequencing along with targeted sequencing and whole exome sequencing in the cohorts from multiple studies, we identified the AA-specific LILRB3-4SNPs, as a cluster of highly-linked consecutive four missense SNPs within LILRB3 gene, as a strong risk factor for kidney transplant failure and it also has a broad association with the progression of kidney and other immune related diseases in AA patients.

We detected the predominant presence of the LILRB3-4SNPs in AA vs European population in transplant cohorts (14.9% in AA vs 0.0% in European) and large biobanks (8.6%~14.9% in AA vs 0.03%~0.1% in European). The previous evidence of recent positive selection of LILR locus were found in Asian population⁴² as well as the Brazilian population with the centuries of admixtures among Native American, European, and African populations⁴³. While the precise target(s) of selection in this region remain to be determined by the sequencing data of large multi-race populations, such evidence does suggest that positive selection has acted on the LILR region in recent human evolutionary history, consistent with the notion that genes with immune-related functions often participate in evolutionary arms races with pathogens⁴⁴, with MHC genes (which interact with LILR genes) showing particularly strong evidence of recent selection^45,46.

LILRB3 as one member of LILR gene family is primarily expressed in myeloid cells. It functions as a negative immune response regulator by binding to SHP-1/2 through ITIM to suppress the NFKB pathway, resulting in reduced cytokine release and inflammation²². The ligand for LILRB3 is not clear although its upregulation in synovial tissue was associated with RA and elevated expression was detected in breast and colon cancers²⁰. Although other LILR gene family members (e.g., LILRB1 and LILRB2) were known to mediate transplant tolerance through binding to HLA-G²⁰, the role of LILRB3 in kidney transplant or diseases is largely unknown. Here we demonstrated the missense polymorphisms in the proximity to the ITIM motif but not the total expression of LILRB3 gene in peripheral blood had the effect on the transplant outcomes. Our multi-omics analysis indicated the SNPs enhanced the adaptive immune responses and ferroptosis in monocyte in pre- and post-transplant peripheral blood and biopsy and initial functional experiment further established a possible causal link. Based on these data, we proposed the following the mechanism of the functional role of LILRB3-4SNPs in the progression of transplant outcomes (Extended Data Figure 5). Activation of LILRB3 causes binding and activation of SHP1/2 phosphatases that, through crosstalk, limit inflammatory signals initiated by TLR stimuli (e.g., LPS) among other stimuli. The expression of the LILRB3-4SNPs reduces the capability of LILRB3’s intracellular ITIM domain to bind to and activate SHP1/2 phosphatases, resulting in amplification of the inflammatory response (e.g., TNFα and cytokine release) and Janus kinase/signal transducer and activator of transcription (JAK/STAT) activation, facilitate induction of ferroptosis, ultimately leading to graft damages. More studies involving LILRB3 to SHP-1/2 binding assays, cellular and functional experiments and kidney transplant model in SNP knocked-in mice are needed.

In two EHR-linked biobanks, LILRB3-4SNPs is associated with the severity but not the prevalence of kidney diseases (particularly ESRD), as opposed to the APOL1 G1/G2 alleles which are primarily associated with the onset of the kidney diseases. This could be partially explained by distinct molecular functions of these two proteins. Within kidney cells APOL1 is primarily expressed in epithelial cells including podocytes, arteriolar endothelium and proximal tubular epithelium. APOL1 plays a role in lipid transport and APOL1 variants have been associated with enhanced opening of cation channels, resulting in multiple putative mechanisms of cell damage, endolysosomal-, or mitochondrial- dysfunction, altered autophagy, endoplasmic reticulum stress. In contrast, LILRB3 is exclusively expressed in immune cells (primarily in myeloid cells) and function as negative immune response regulator. LILRB3-4SNPs may induce systematic inflammation and macrophage/monocyte ferroptosis, leading to tissue damages in kidney damage or ESRD. This also explain why LILRB3 has a broad effect on the immune related diseases (respiratory diseases and mental disorders).

Ferroptosis, an iron-dependent form of non-apoptotic cell death, has emerged as a significant mechanism in the development of chronic kidney diseases⁴⁷. It presents a potential therapeutic target for managing CKD⁴⁸. Despite this, our understanding of ferroptosis in the context of kidney transplants remains limited. The discovery of the association between LILRB3-4SNPs and ferroptosis in monocytes in the bloodstream and post-transplant kidneys is a compelling advancement. Notably, our findings demonstrate that a ferroptosis inhibitor effectively reduced cell death in THP1 cells overexpressing LILRB3-4SNPs, suggesting its potential utility in conjunction with ATG induction for treating recipients carrying these SNPs. However, it is crucial to evaluate the efficacy of the ferroptosis inhibitor in an in vivo mouse transplant model to validate its therapeutic potential.

The major limitation of this study is the relatively small numbers of participants with LILRB3-4SNPs in the four retrospective transplant cohorts. Therefore, the SNP association with transplant outcomes will need to be prospectively validated in a large cohort of AA kidney transplant recipients. Future validation efforts will also explore additional loci within the LILR gene family and extensively investigate downstream biological mechanisms of these loci, along with LILRB3-4SNPs, which will lay a foundation for interventions to improve outcomes within the AA community affected by renal diseases.

In a summary, our data show that LILRB3-4SNPs represents a genetic risk factor for the severity of kidney transplant outcomes and immune-related diseases within the African American population, with potential for a therapeutic intervention using ferroptosis inhibitors.

Methods

Ethics statement.

Four kidney transplant (GoCAR, SIRPA, CTOT19 and VerciDx) and BioMe biobank cohort studies included in this study were approved by the Institutional Review Board of participating sites or registered with ClinicalTrials.gov and all participants enrolled in these cohorts provided the written informed consent: The GoCAR cohort was approved by the Institutional Review Board (IRB) of Icahn School of Medicine at Mount Sinai (STUDY-11–01259-CR003) and also registered with ClinicalTrials.gov (NCT00611702); The SIRPA cohort was approved by the Institutional Review Board at University of Pittsburg (#22070088); The CTOT19 cohort was registered with clinicaltrail.gov (NCT02495077)¹² and approved by the Institutional Review Board at each participating site; The VericiDx cohort was registered with clinicaltrial.org (NCT04727788) and approved by Advarra Institutional Review Board (Pro00049177); The BioMe study was approved by Mount Sinai Institutional Review Board (STUDY-07–0529) and the use of de-identified data from BioMe cohort in this study was also approved by the IRB of Icahn School of Medicine at Mount Sinai (STUDY-22–01233).

Study cohorts and sample selection.

Our study incorporates four kidney transplant cohorts with each focusing on unique aspects of kidney transplantation, namely GoCAR^10,11, SIRPA, CTOT19¹², VericiDx, and two EHR-linked biobank cohorts (BioMe and All-of-Us)^13,14 as depicted in Figure 1. As AA population is the focus of this study, the demographic/clinical characteristics of AA recipients across four transplant cohorts included in this study were summarized in Extended Data Table 1. The sample selection from these cohorts was based on the availability of data or specimen in each cohort and no statistical method was used to predetermine sample size. The sex of the participants was determined through self-report. Sex-based analysis was not conducted in relation to graft failure in this study, as there is no consensus in the literature regarding the impact of recipient sex on graft failure.

1. The GoCAR cohort (n=588, genetically-identified 20.6% AA, 63.6% European) is a multi-center observational kidney transplant study with an aim to uncover genetic and transcriptomic markers associated with diverse transplant outcomes such as acute rejection and renal failure with the period of up to 5-year follow up^10,11. The blood samples were collected pretransplant. Kidney biopsies were performed at 3-, 12-, and 24- month surveillance or clinical indications and biopsy core tissues and blood were simultaneously collected. The histological slides were centrally scored by MHG pathology group. All patients underwent genotyping using SNP arrays¹⁵ and the recipients’ ancestry was determined by SNP arrays and validated by RNAseq data.

   A total number of 264 GoCAR patients were included in this study based on the availability of the specimen or genomic datasets. First, RNA sequencing data (150bp pair-ended reads) of pretransplant blood samples from 170 patients in published study¹¹ (GSE112927) was used to identify expressional SNPs associated with graft loss. These 170 patients had similar demographic and clinical characteristics to the entire cohort (n=588, Table S1) except a higher rate of acute rejection as more samples with rejection were selected in the original study for early rejection¹¹. However, there was no difference for graft loss events (Table S1). Second, targeted DNA sequencing of the region covering the identified LILRB3-4SNPs was performed on 127 non-European (AA + Hispanic) recipients to validate the LILRB3-4SNPs. A total of 83 AA recipients had LILRB3-4SNPs genotypes determined by RNA or targeted DNA sequencing and their demographic/clinical features were similar to the 105 AAs in the entire cohort (Table S2). The LILRB3-4SNPs genotypes of these 83 recipients were correlated with graft loss and early eGFR in the meta-analysis with other cohorts. Last, for multi-omics genomic analysis to investigate the functional roles of LILRB3-4SNPs, Mass spectrometry was performed on the serum in pre-transplant blood from 16 AA recipients (8 with and 8 without LILRB3-4SNPs) to correlate with RNA sequencing data. Single cell RNA sequencing was performed on the pre-transplant PBMC from 2 patients (1 with and 1 without LILRB3-4SNPs) and the post-transplant PBMC at 24 months post-transplant from 6 patients (3 with and 3 without LILRB3-4SNPs) to investigate the transcriptomics dysregulation in immune subsets. RNA sequencing data of posttransplant blood samples collected post 6 months biopsies from 20 recipients (10 with and 10 without. LILRB3-4SNPs) and biopsy samples collected at 3 months post-transplant from 6 recipients (3 with and 3 without LILRB3-4SNPs) were used to correlate the SNPs with post-transplant gene expression in the posttransplant blood or graft respectively. The AA recipients with and without LILRB3-4SNPs selected for post-transplant genomic analysis had matched demographic and clinical parameters (age, sex, HLA mismatch, donor status and induction type).

2. The SIRPA cohort is an observational cohort established at University of Pittsburgh aimed to evaluate the relationships among SIRPA polymorphisms and transplant outcomes. A total of 455 adult renal transplant recipients who underwent transplantation between January 2013 and December 2019 at the University of Pittsburgh Medical Center (UPMC) and had DNA available were screened for the study. Of these 455 patients, there were 54 self-reported African American recipients and were included in the analysis. Patients were followed for a maximum of 7 years. Mean follow-up was 55.5 ± 21 months. Unless medically contraindicated, patients underwent two surveillance biopsies at 3 and 12 months in addition to for-cause biopsies. Biopsies were classified as either early (0 to 4 months), delayed (5 to 12 months), or late (12–84 months). Biopsies were adjudicated by a single pathologist (PR) blinded to the study.

   The LILRB3-4SNPs genotypes of 54 AA patients were determined by targeted DNA sequencing and correlated with transplant outcomes including graft loss and early eGFR combined with other transplant cohorts.

3. The CTOT19 cohort represents a multi-site randomized controlled trial investigating the efficacy of Remicade induction therapy for deceased donor kidney transplant recipients within 2 year followed-up. With a cohort of 225 patients, the study spanned 2 years to assess the effectiveness of this therapy. RNA sequencing was executed on pretransplant blood from 128 recipients, utilizing 101bp paired-end (101 PE) reads (Table S3). The ancestries of these 128 recipients were genetically determined from genotyping by RNA sequencing data:47 African American, 27 European, 41 Hispanic, and 13 individuals from other ethnic backgrounds. RNA sequencing data in this cohort was used to validate the variant expression of LILRB3-4SNPs in AA recipients. The RNAseq data in 47 AA recipients was also combined with two cohorts (GoCAR and VericiDx) to identify the meta-gene signatures associated with the LILRB3-4SNPs. The SNPs was also correlated with early eGFR within 24 months in meta-analysis with other three cohorts. The graft loss association analysis was not performed due to limited graft loss cases within two years follow up time,

4. The VericiDx cohort is an ongoing observational clinical study (241 patients (32.0% self-reported AA) enrolled as of 06/30/2023) with a focus on predicting early acute rejection and renal fibrosis post-transplantation. RNA sequencing profiles (101 PE reads) were generated on pretransplant blood, with a subset of 77 AA recipients. The RNA data in these 77 AA recipients was combined with two cohorts (GoCAR and CTOT19) to identify the meta-gene signatures associated with the LILRB3-4SNPs. The LILRB3-4SNPs was correlated with early eGFR values (within two years) in meta-analysis with other three cohorts. However, the relatively short follow-up duration for many patients in this cohort limited the association analysis for graft failure.

Two EHR-linked biobanks (BioMe and All-of-Us) were included in this study to investigate the association of LILRB3-4SNPs with the progression of kidney diseases or other immune related health conditions.

1. The BioMe cohort, established by the Institute of Personalized Medicine at Mount Sinai with an aim enabling researchers to conduct genetic and epidemiologic studies on a large population linked with medical information, comprises over 34,000 patients from the Mount Sinai Health system, as of September 30, 2022¹³. Among this cohort, whole exome sequencing (WES) data were obtained from 30,099 patients, including self-reported 7,096 African American, 9,555 Caucasian, 10,467 Hispanic, and 2,981 individuals from other ethnic backgrounds. The LILRB3-4SNPs were not detected in this cohort and the adjacent perfectly linked SNP (rs113314747) was used as surrogate. This data was correlated with EHR information to assess the connection between identified SNPs and laboratory test results, as well as disease diagnoses, within the African American population and compared to that for APOL1 G1/G2 alleles.

2. All-of-Us cohort, is one of the largest US population-based biobanks with the health and genetic information from the population of diverse ethnic background. The analysis was conducted through All-of-Us online workbench¹⁴. Among 245,388 participants, 50,969 self-reported AA with WES data were investigated in this study. The LILRB3-4SNPs was genotyped as one insertion followed by one deletion in the database, the same as the genotypes we observed in GoCAR cohort (Figure S3). To be consistent with BioMe, we use the highly linked adjacent SNP (rs113314747) (Table S13) as surrogate for LILRB3-4SNPs to correlate with the progression of kidney diseases and other immune related health conditions (determined by ICD9 and/or ICD10 code).

Genetic Ancestry Determination on GoCAR and CTOT19 cohorts.

In the GoCAR cohort, genetic ancestry was determined as previously described¹⁵. Briefly, using genome-wide genotyped data from SNP arrays, we employed ADMIXTURE/1.3.0⁵⁰ to estimate ancestry proportions, with the 1000 Genomes Project Phase I serving as the reference. Patients with a proportion of African ancestry (pAFR) ≥ 0.6 were classified as African American (AA), those with a proportion of European ancestry (pEUR) ≥ 0.9 were classified as European, and patients with admixed ancestry from AA and EUR were classified as Hispanic. For the CTOT19 cohort, we genotyped the whole transcriptome using an RNA-seq genotyping strategy, and GrafPop/1.0⁵¹ was used to estimate ancestry proportions by extracting race-specific SNPs from CTOT19 patients. AA, EUR, and Hispanic patients were defined using the same pAFR and pEUR thresholds as in GoCAR.

Although we observed mild discrepancies between genetic ancestry and self-reported race in both cohorts, the classification of AA patients was quite consistent with self-reported race in both cohorts (Table S9), suggesting the effectiveness of defining genetic ancestry using both SNP array and RNA-seq genotyping.

Detection of expressional SNPs (eSNPs) from RNA sequencing data.

Genotyping based on RNA-seq data has been discussed in previous studies, demonstrating consistency with SNP array and RNA sequencing data, provided there is sufficient expression of exon regions^7,8. It has also been employed as a complementary method for SNP genotyping in similar allele-specific expression studies⁹. The pipeline for expressional SNPs (eSNPs) quantification based on SNP array and RNA sequencing data from pretransplant blood in GoCAR cohort was depicted in Extended Data Figure 1. Briefly, the quality control, SNP calling and imputation was performed on SNP array data of blood DNA samples from the kidney transplant recipients as described in previous study¹⁵. The SNP calling from RNAseq data of pretransplant blood samples from the recipients was performed with the GATK best practice⁵² under GATK/4.2.0.0 and was compared with SNP array genotyping (Extended Data Figure 1). Sequencing reads were aligned to the human genome hg19 with STAR/2.6.1d⁵³ using two-pass mode to get better alignments around splice junctions. Duplicated reads removing and reads sorting were performed with Picard/1.93. SplitNCigarReads from GATK was used to splits reads with N in the cigar into multiple supplementary alignments and hard clips mismatching overhangs for the specific alignment for RNAseq data. Base quality was recalculated with BaseRecalibrator and variants were called with HaplotypeCaller. Variants from different samples were merged together with GenotypeGVCFs and variants with FS > 30.0 and QD < 2.0 were considered as low quality and filtered. Finally, we combined the genotypes from SNP array and RNAseq, following the order of the confidence level as Array > imputation > RNAseq. The SNP genotypes from SNP arrays and RNAseq were annotated with AnnoVar version “2018Apr16”. The exonic SNPs were subsequently extracted for further analysis. To quantify allele expression of these exonic SNPs, the nucleotides of those exonic SNP locus on human genome (hg19) were masked as “N” to avoid mapping bias towards reference allele. RNA sequencing reads were then aligned to the masked human genome with STAR 2.6.1d and duplicate reads were identified by Picard 1.93. The reads mapped to reference and alternative alleles of each SNP were counted with ASEReadCounter. Consistent with previous analysis⁹, SNPs covered by more than 5 reads were kept for further analysis (Extended Data Figure 1). The alternative allele expression fraction (AEF) of each SNP was calculated as read count for alternative allele / (the total read count for reference and alternative allele). The reads covering target SNP site was extracted with samtools 1.9 and the coverage was visualized with IGV 2.8.2.

The pipeline was also applied to CTOT19 and VericiDx RNA sequencing dataset to detect the allelic specific expression of LILRB3-4SNPs identified in the GoCAR cohort.

Targeted DNA sequencing of the LILRB3-4SNPs locus.

The DNA samples from 127 non-Caucasian (AA+Hispanic) recipients retrieved from GoCAR specimen biobank and 54 AA recipients from the SIRPA cohort were subjected to targeted DNA sequencing by following Illumina 16S sequencing protocol. The DNA region covering the LILRB3-4SNPs was amplified using the primers (5′- TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGTTCTGCTGAGTGTGGGGTCT-3′ and 5′- GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGCCTCCCAGGATGTGACCTA-3′) with addition of the Illumina adaptor overhangs. The PCR amplification was performed in a total 20 μl reaction volume for each DNA sample. After purification of amplicons using AMPure XP beads (Beckman) and confirmation of successful amplification by high sensitivity DNA assay for Bioanalyzer, the amplification product was proceeded to attach dual indices and Illumina sequencing adapters using the Nextera XT Index Kit (Illumina) to PCR product for each DNA sample. Library quantification was performed using the Qubit dsDNA HS Kit (Thermo Fisher Scientific), and the size distribution was assessed using a high-sensitivity DNA assay for Bioanalyzer and qualified libraries were then pooled for sequencing. Sequencing of pooled libraries was performed on the NovaSeq 6000 platform (Illumina) at paired-end 150bp with 20M reads.

To determine genotypes of the targeted region around the LILRB3-4SNPs locus, the reference sequences of 200pb upstream and 200bp downstream rs549267286 was extracted from human genome reference (hg19) to build the targeted reference genome using bwa/0.7.15. The DNA sequences were aligned to the targeted reference with bwa mem with default parameters. SNPs within this region were called by bcftools/1.9 with mpileup by default parameters for each sample and finally merged by bcftools merge.

Structure modelling of the interactions of SHP-2 and LILRB3.

To model the interaction between SHP2 and the C-terminal tail of LILRB3, the sequences of the two SHP domains of human SHP2 (residues 1–218) and the tail segment of LILRB3 containing the last two ITIM motifs (residues 581–631) was used as the input for the AlphaFold 2 structural prediction in the ColabFold webserver^23,24. AlphaFold 2 successfully predicted the expected binding modes between the two ITIM motifs and the two SH2 domains in SHP2, respectively, despite the absence of phosphorylation of the tyrosine residues in the motifs. The phospho-groups were added by using Charmm-GUI and manually adjusted in Coot^54,55. The model was then aligned based on the second SH2 domain with the crystal structure of full-length SHP2 in the inactive state (PDB ID: 4NWF), where the first SH2 domain obstructs the active site of the catalytic domain. Combining the SH2 domains from the AlphaFold model with the catalytic domain from the crystal structure led to a composite model of full-length SHP2 bound to the C-terminal tail of LILRB3, in which the first SH2 domain is pulled away from the catalytic domain, indicating an open active conformation. The EP->PS variant of the LILRB3 in complex with SHP2 was predicted in the same manner. The structure figures were rendered using PyMOL (The PyMOL Molecular Graphics System, Version 2.0 Schrödinger, LLC.).

BioMe and All-of-Us biobank WES and EHR data processing and analysis.

Genotyping of the whole exome sequencing data from BioMe biobank, was described in previous publication^13,56. Briefly, BioMe is a diverse DNA biobank linked with electronic health records from the Mount Sinai Health System (MSHS) based in New York, NY. Patients’ plasma and blood samples was taken for DNA sequencing with their unidentified heath records upon consent. The sample preparation, exome sequencing, quality control and SNP calling were conducted at the Regeneron Genetics Center as described in previous publication⁵⁷ with GATK best practice. The 4 consecutive SNP within LILRB3 was not identified while the adjacent SNP rs113314747, was used as an alternative as it is highly linked with those 4 SNPs. The self-reported race information was extracted from the questionnaire of each patient upon enrollment. There are 30,099 patients with WES data and self-reported race including 7,096 African Americans. The disease diagnosis information of each patient was extracted by the International Classification of Disease (ICD)-10 code from EHR records. The ESRD patients was defined as patients with ICD code (N18.6 and/or 585.6 and/or Z94.0 and/or V42.0). To assess the association of the SNP association with the progression of kidney diseases, longitudinal eGFR values for the patients with kidney diseases were extracted from the blood test records. For each patient, the eGFR decline event was defined as the date of first record less than 15 (ml/min/1.73m) with a later confirming eGFR record (less than 15) within 3 to12 months.

Genotyping of the whole exome sequencing data from All-of-Us was described in previous publication and All-of-Us research hub¹⁴. The All-of-Us program is a longitudinal biobank with WES genotyping, enrolled with more than one million individuals across diverse groups within USA. The analysis was performed on the All-of-Us online workbench⁵⁸. The AA population was extracted with self-defined racial and ethnicity information and the disease information was extracted with ICD10 and/or ICD9 code from EHR records. The LILRB3-4SNPs were identified as one insertion and one deletion (Figure S3) same as the genotype we observed in GoCAR, which is highly linked to the adjacent SNP rs113314747 (Table S13). To be consistent with BioMe, we used rs113314747 as a surrogate for LILRB3-4SNPs in disease correlation analysis. The ESRD patients was defined as patients with ICD code (N18.6 and/or 585.6 and/or Z94.0 and/or V42.0) and the eGFR decline event was defined the same way as BioMe as described above.

Bulk RNA sequencing on pre-transplant and post-transplant blood and data processing.

Bulk RNA sequencing on pre-transplant blood from the recipients in CTOT19 (n=128) and VericiDx (n=77) was performed by VericiDx Inc. Total RNA was isolated from the whole blood stored in PAXgene tubes and then subjected to standard RNA sequencing libraries generation by following manufactory protocol. The pre-transplant sequencing libraries were then sequenced on NextSeq sequencer at 101 bp read length and an average throughput of 12.5M reads per sample.

The sequencing libraries of post-transplant blood collected within 6 and 24 months from 20 AA recipients in the GoCAR cohort were generated by stranded ployA selection kit and were then sequenced on NovaSeqXplus sequencer at 150 bp read length and an average throughput of 30M reads per sample at Cornell Genomics Core Facility.

The raw sequencing data from pre- and post- transplant blood were first aligned to human genome hg38 with STAR 2.5.3a⁵³ and gene expression counts in each sample was quantified with HTseq both with default parameters. Limma voom⁵⁹ was used to test the gene expression between risk and non-risk samples with raw gene expression counts. Genes with p value < 0.05 were identified as differentially expressed (DEG).

Meta differential gene expression and network analysis.

Meta-analysis was performed across the three datasets (GoCAR, CTOT19 and VericiDx) using combined effect size (ES) approach to identify differential meta-gene signatures between samples with and without LILRB3-4SNP. Meta-genes were selected if the following conditions were met: 1) genes were mapped to at least 2 out of 3 data sets. 2) p-value of the meta-ES < 0.05; 2) individual ES was in the same direction, either up or down, in all mapped data sets. These meta-genes (1,555 genes) were further used for meta-network construction. Significant Meta-Coexpression-Network was identified by selecting connections with meta |r| >= 0.6. We constructed the network for samples with LILRB3-4SNPs risk alleles only and performed Markov Cluster Algorithm (inflation parameter=2) to identify sub-clusters from the network. To identify key driver genes within each sub-cluster, we performed z-normalization per sample for case samples in each data set and combined normalized expression data to construct directional Bayesian network. Key Drivers (KD) were then selected as genes that had large downstream numbers and out-degrees. If a KD is not in the downstream of any other KDs in the same module, it becomes a global driver otherwise it is a local driver. Detailed methods of meta-analysis were described in our previous publication⁴⁹.

Serum mass spectrometry of pretransplant serum.

The relative abundances of serum proteins in pretransplant blood of selected 16 (8 SNP+ and 8 SNP−) AA recipients were quantified using mass spectrometry (MS)-based proteomics. Preprocessing of 20 μl of each serum sample was carried out using the Enrich-iST preparation kit, following the manufacturer’s protocol with a slight modification of conducting an overnight digestion at 37°C. Specifically, 20 μl of serum from each sample was mixed with magnetic beads (En-Beads) for 30 minutes on a thermoshaker at 30°C with 1200 rpm. The proteins binding on beads were then reduced, alkylated, followed by on-beads trypsin digestion. The resulting peptides were desalted and analyzed using an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Scientific) as previously described⁶⁰. Briefly, peptides were separated for two hours on C18 columns using a binary gradient of 2% acetonitrile in 0.1% formic acid and 85% acetonitrile in 0.1% formic acid. Peptides were then introduced into the mass spectrometer via nanospray Flex ion source. The spectra were acquired in a positive mode at 2 kV and 275°C, scanning within a range of m/z 375 to 1500 with 120,000 FWHM on an orbitrap. Peptides with charge states 2–7 were selected for further MS/MS analysis. The acquired spectra were then searched against the Uniprot validated human proteome (downloaded on September 11, 2023) using Sequest search engine through Proteome Discoverer (v. 2.4). Mass tolerance was 10 ppm for MS and 0.6 Da for MS2. Methionine oxidation and acetylation of N-termini were set as variable modifications, and cysteine carbamidomethylation was set as a fixed modification. Maximum allowed false discovery rate was 1%.

The expression of each protein between samples with and without risk allele was tested using two-sided Student’s T test. Proteins with P value < 0.05 was identified as differentially expressed (DEP). WebCSEA⁶¹ was used to predict the enriched cell type of up regulated (log2(fold change) > 0) and down regulated (log2(fold change) < 0) DEP separately.

Single cell RNA sequencing on pre- and post-transplant PBMC.

Single cell RNA sequencing was performed on PBMCs isolated from the pretransplant blood of 2 AA recipients (one without and one without LILRB3-4SNPs) and post-transplant (24 months) blood of 6 AA recipients (three without and three without LILRB3-4SNPs) with matched recipient and donor’s demographic and baseline clinical features (age, sex, donor status, HLA mismatches and induction types). The PBMC samples were retrieved from the liquid nitrogen tank and thawed for scRNAseq library generation. Briefly, the viable PBMC were loaded into chromium microfluidic chips and barcoded with a 10x Chromium Controller (10x Genomics). RNA from the barcoded cells was subjected to sequencing library generation with reagents from a Chromium Single Cell 3′ v.2 Reagent Kit (10x Genomics) according to the manufacturer’s protocol. The library was sequenced on Illumina sequencer (NovaSeq 6000) at a 150 bp paired-end sequence length and 400 million read throughput per library.

The pre- and post-transplant PBMC sequencing data were analyzed separately as they were sequenced in different batches. The raw sequencing data was aligned to human genome hg38 and then subjected to gene expression quantification with cellranger (7.1.0) with default parameters. The gene-cell count matrix was submitted to Seurat 4.1.1 28 for QC, normalization and clustering⁶². Genes expressed in less than 3 cells were removed and cells expressing less than 200 genes and more than 7000 genes were removed as low quality and potential doublets. Cells with more than 30% RNA expressed from mitochondrial genes were also removed as low quality. The gene expression of each cell was normalized with its total expression and multiplied with scale factor 10000 and then log-transformed. After batch variations between individuals were corrected with RunHarmony⁶³, unsupervised clustering was performed with FindNeighbours with the first 30 “harmony” dimensions and FindClusters with resolution of 0.8 with the default Louvain algorithm. FindMarkers was then used to identify the markers of each cluster by comparing each gene with other clusters with Wilcoxon Rank Sum test. Each cluster was annotated with classic markers for immune subsets in PBMC as published previously⁶³. The differentially expressed genes (DEG) between risk and non-risk sample in each cell type were identified with FindMarkers by Wilcoxon Rank Sum test. Genes with a p value < 0.001, log2(fold change) <= 0.25 or >= 0.25 and expressed in at least 10 percent of cells in each condition were considered differentially expressed. Function enrichment analysis of DEGs was performed with Enrichr/2.1 on upregulated (log2(fold change) > 0) and downregulated (log2(fold change) < 0) genes separately

Generation of the construct and cell line overexpressing the LILRB3-4SNPs reference and variant allele.

The N-terminal myc-tagged LILRB3 cDNAs with and without LILRB3-4SNPs were synthesized and cloned into a retroviral vector that also carries a blasticidin resistance gene under the control of IRES (internal ribosomal entry site). The LILRB3 cDNA contains a human growth hormone signal peptide, myc-epitope tag, and the coding sequence of mature LILRB3 protein.

Cell viability assay.

THP-1 Cells overexpressing LILRB3–4SNPs variant or reference allele were seeded in 96 well culture plates at a density of 1 × 104 cells/well and treated with 200ug/L LPS and 0.625umol liproxstatin-1. At the end of treatment, cell viability was determined using CellTiter-Glo Luminescence Cell viability Assay kit (Promega) by measuring luminescence (ATP level) following the manufacturer’s instructions. Medium without cells were used as background luminescence. Values were presented as an RLU. The cell variability assay was repeated in four biological replicates.

Quantitative PCR.

THP-1 cells overexpressing LILRB3–4SNPs variant or reference allele was after incubated with 200ug/L LPS for 6 hours and 24 hours and the total RNA was then extracted using TRIzol Reagent (Invitrogen). 1000 ng total RNA was reverse transcribed to cDNA with the EasyScript Plus^™ cDNA Synthesis kit (Lambda Biotech). Quantitative real-time PCR for inflammation and ferroptosis markers was performed with the 7500 Real-Time PCR System (Applied Biosystems). The mRNA level was normalized to GAPDH and expressed as fold change. Quantitative PCR were performed on duplicated biological experiments. The primer sets for inflammation and ferroptosis markers were synthesized by Sigma-Aldrich as shown in TableS24.

Statistical analysis.

For the initial genome wide screening of DCGL-associated SNPs in 170 RNAseq transplant recipients in the GoCAR cohort, allele expression fraction (AEF) of the SNP was correlated with DCGL using univariate or multi-variate Cox model. In the multi-variate Cox model, the baseline and post-transplant clinical factors that were adjusted for included donor status, donor age, DGF, post-transplant infection, post-transplant acute rejection, recipient genetic ancestry, induction therapy and HLA mismatch. For the correlation of identified LILRB3-4SNPs with DCGL within AA recipients, Cox model was used adjusting by APOL1 risk allele number, donor status, donor age, DGF, post-transplant infection, post-transplant acute rejection and HLA mismatch in the GoCAR cohort. All factors were adjusted for in the association analysis for the SIPRA cohort except APOL1 risk allele and post-infection status as they are not available for SIRPA cohort. Induction therapy and genetic ancestry was not adjusted within AA cohorts, as induction therapy was applied to all AA patients (Extended Data Table 1). The correlation of eGFR decline in AA recipients (GoCAR, SIRPA, CTOT19 and VericiDx) with LILRB3-4SNPs was evaluated by Logistic regression model, adjusted by donor status, donor age, DGF and HLA mismatch. In BioMe and All-of-Us cohort, the clinical diagnosis (defined by ICD9 and/or ICD10 code) was tested with LILRB3-4SNPs or APOL1 G1/G2 alleles using Logistic regression model. The association of eGFR decline event with LILRB3-4SNPs and APOL1 risk alleles in BioMe and All-of-Us biobank was tested with Cox regression model. Meta analysis among data sets were conducted with “metafor/2.4” R package using fixed-effects. Due to the small size of GoCAR discovery cohort, the statistical tests for genome-wide identification of genetic variations associated with the transplant outcomes or gene/pathway signatures associated with LILRB3–4SNPs were not adjusted for multiple comparisons but the significance of the candidate signatures was cross-validated in independent external cohorts or other experimental approaches. The cell variability of THP-1 cell lines overexpressing the variant or reference allele upon LPS and liproxstatin-1 stimulation was compared by two-sided Students’ T-test.

Extended Data

Extended Data Figure 1. Evaluation and quantification of allele specific expression in the GoCAR cohort.

A) Overall work flow of eSNP identification and allele expression fraction (AEF) calculation. (B-D) The distribution of AEF of homozygous genotype of reference allele (0/0) (B), heterozygous genotype (0/1) (C) and homozygous genotype of alternative allele (1/1) (D) in the GoCAR cohort. Most alleles showing a balanced expression of reference and alternative alleles (AEF around 50%) while some alleles showed a higher expression of either the reference allele or the alternative allele (AEF > 50% or AEF < 50%), and a few sites exhibited mono-allelic expression at both ends (AEF=0 or AEF=1). This distribution aligns with previous studies on allele-specific expression. (E) The sensitivity and specificity of RNAseq-based genotyping by comparing to the SNP array-based genotyping for heterozygous (upper) and homozygous (lower) calls with various read coverage depths in the GoCAR cohort (n = 153 with both RNA-seq and SNP array data). Each dot represents a sample and box and whiskers plot showing the distribution (thick bar, median; box, 25th to 75th percentile, whiskers reach to the largest/smallest observations within 1.5 box-heights of the box). Overall, the RNAseq-based genotyping strategy achieved over 90% sensitivity and specificity for both heterozygous and homozygous detection with more than 10 reads. With 5–10 reads, we achieved over 75% sensitivity and 100% specificity for heterozygous calls, and 100% sensitivity and over 99% specificity for homozygous calls. These data indicate that our informatic pipeline effectively detected exonic SNPs from RNA-seq data in the GoCAR cohort with high sensitivity and specificity.

Extended Data Figure 2. Comparison of post-transplant longitudinal eGFR values (mean of the records of each month) of AA kidney transplant patients (ICD Z94.0 or V42.0) carrying the LILRB3-4SNPs variant (“Risk”) vs reference (“Non-risk”, blue) allele in the BioMe cohort.

A) longitudinal eGFR values of risk and non-risk AA kidney transplant patients within 48 months after transplant. Bold curves indicate the fitted regression lines for two groups. Comparison (Students’ T test, two-sided) of average eGFR between risk (n=9) and non-risk (n=90) patients within 3 months (B), 3–24 months (C), and 3–48 months (D) after transplant. Each dot represents a sample and the box and whiskers plots showing the distribution (thick bar, median; box, 25th to 75th percentile, whiskers reach to the largest/smallest observations within 1.5 box-heights of the box).

Extended Data Figure 3. Transcriptomic dysregulation of post-transplant blood and kidneys in AA recipients carrying LILRB3-4SNPs variant (“Risk”) vs reference (“Non-risk”) allele.

A) Gene Set Enrichment Analysis (GSEA) enrichment plot of the pathways showing gene upregulation involved in Th17 cell differentiation, T cell receptor signaling and B cell mediated immunity in bulk RNA sequencing of the blood samples collected after 6 months post-transplant in AA recipients with (n=10) vs without (n=10) LILRB3-4SNPs. GSEA analysis was performed on post-transplant blood expression profiles of the recipients with and without the SNP to identify the pathways associated with the SNP (P <0.05). B) UMAP of single cell RNA sequencing of the PBMCs isolated from 6 AA patients (with (n=3) and without (n=3) LILRB3-4SNPs) at 24-month after transplantation. C) Cell proportion of each cell type in two groups. The increased T cell and decreased monocyte populations were detected in the SNPs carrying recipients. D) Function enrichment of significant DEGs between patients with and without LILRB3-4SNPs in each cell type demonstrating gene dysregulation involved in T/B cell activation and ferroptosis. DEGs in the subpopulation was identified by two-sided Wilcoxon Rank Sum test at P value <0.05. The gene-function enrichment was evaluated with one-sided hypergeometric test. E) Heatmap showing the log2(fold change) of selected DEGs of B, T cell activation and ferroptosis signatures between SNP+ vs SNP− cells in each cell type. F) Significantly-dysregulated functions (NES: normalized GSEA enrichment score) in 3-month post-transplant biopsies from 6 AA recipients with (n=3) and without (n=3) the LILRB3-4SNPs. G) GSEA enrichment plot showing down-regulation of ferroptosis-negatively associated genes comparing the patient with (n=3) and without (n=3) LILRB3-4SNPs in 3-month biopsies. The shared transcriptional dysregulation among recipient’s pre- and post- transplant blood, transplanted kidneys implied the persistent inflammation in the blood stream post-transplant causes kidney damage. P values in GSEA and functional enrichment analysis are unadjusted.

Extended Data Figure 4. In vitro functional analysis of THP-1 macrophage cell line overexpressing LILRB3-4SNPs variant (“Risk”) or reference (“Non-risk”) allele.

qPCR on expression changes for immune response (A) and ferroptosis-negatively-associated genes (B) upon LPS stimulation from 0 to 6 hours (left) or from 6 to 24 hours (right). The heatmap colors (blue color for positive values and brown for negative values, respectively) along with the numbers indicate the average of log2(fold changes) from duplicated biological experiments. Cell viability analysis within 48 hours upon LPS stimulation (C) and in conjunction with ferroptosis-inhibitor (lipro-1) treatment (D) in THP-1 cells overexpressing LILRB3-4SNPs variant (n = 4) or reference allele (n = 4) (Student’s t-Test, two sided). The bar plots represent the mean values of the cell viability measurements from four biological replicates and error bars represent one standard deviation. Following a 6 h LPS stimulation, all cell lines exhibited increased expression of crucial inflammatory response markers linked to the LILRB family, including TNFα and IL1β, and the expression of these genes decreased from 6 to 24 hours. The cell line overexpressing the variant allele produced greater quantities of TNFα, TNFAIP3 and IL1β upon 6-hour LPS treatment, with less attenuation between 6 and 24 hours than the reference allele, implicating increased inflammation associated with the SNPs (A). Expression of 5/6 ferroptosis negatively-associated genes in cell line with the SNPs decreased more between 6 and 24 hrs post LPS treatment, consistent with enhanced ferroptosis at 24 hours linked to the SNPs (B). Cell viability analysis demonstrated a reduced viability upon LPS stimulation in cells with the SNPs (C, upper, orange vs green) but not those without the SNPs (C, lower, orange vs green). This phenomenon for the SNPs was reversed by ferroptosis inhibitor, Liproxstatin-1 (Lipro-1, at 0.625 umol) that targets lipid peroxidation (D, orange bar, upper). Lipor-1 had no effect on the cells without SNPs (D, orange bar, lower).

Extended Data Figure 5. The schematic model of the role of LILRB3 in inflammation and ferroptosis.

Activation of LILRB3 causes binding and activation of SHP1/2 phosphatases that, through crosstalk, limit inflammatory signals initiated by TLR stimuli (e.g., LPS) among other stimuli. The expression of the variant LILRB3-4SNPs risk allele reduces the capability of LILRB3’s intracellular ITIM domain to bind to and activate SHP1/2 phosphatases, resulting in amplification of the inflammatory response (e.g., TNFα and cytokine release) and JAK/STAT activation, facilitate induction of ferroptosis, ultimately leading to graft damages. This figure was created in BioRender https://BioRender.com/i68z465.

Extended Data Table 1.

Demographic and clinical informa8on of AA kidney transplant recipients in the GoCAR, SIRPA, CTOT19 and VericiDx cohorts.

	GoCARa (n=83)	SIRPA (n=54)	CTOT19 (n=47)	VericiDx (n=77)
Recipient information
Recipient sexb (male), n (%)	53/83 (67.5)	36/54 (66.7)	25/47 (53.2)	44/77 (57.1)
Recipient age, mean ± SD	52.9 ± 11.7	50.1 ± 13.3	51.8 ± 10.3	52.5 ± 12.9
Transplant outcome
Death censored graft lossc, n (%)	29/83 (34.9)	14/54 (25.9)	2/47 (4.3)	1/77(1.3)
Graft loss time (month), mean ± SD	31.9 ± 26.7	44.1 ± 16.3	1.9 ± 0.0	2.8
Delayed graft function, n (%)	34/83 (41.0)	15/54 (27.8)	21/47 (44.7)	20/77 (26.0)
Acute rejectiond, n (%)	22/83 (26.5)	13/54 (24.1)	5/30 (16.7)	24/77 (31.2)
Post-transplant infection, n (%)	19/83 (22.9)	-	-	-
Underlying disease, n (%)
Diabetes	10/77 (13.0)	20/54 (37.0)	15/47 (31.9)	2/77 (2.6)
Diabetes hypertension	25/77 (32.5)	-	-	29/77 (37.7)
Hypertension	35/77 (45.5)	13/54 (24.1)	15/47 (31.9)	38/77 (49.4)
Glomerular disease	-	6/54 (11.1)	6/47 (12.8)	3/77 (3.9)
Other	7/77 (9.1)	15/54 (27.8)	11/47 (23.4)	5/77 (6.5)
Induction therapy, n (%)
Induction	83/83 (100.0)	54/54 (100.0)	128/128 (100.0)	77/77 (100.0)
Donor information
Donor age, mean ± SD	42.2 ± 16.1	40.7 ± 13.9	41.7 ± 13.6	40.8 ± 14.0
Donor sexb, (male), n (%)	53/83 (67.5)	26/54 (48.2)	27/47 (57.4)	44/77 (57.1)
Donor status, n (%)
Living donor	17/83 (20.5)	21/54 (38.9)	0/0 (0.0)	23/77 (29.9)
Diseased donor	66/83 (79.5)	33/54 (61.1)	47/47 (100.0)	54/77 (70.1)
Donor ancestrye, n (%)
African American	20/64 (31.2)	13/54 (24.1)	6/47 (12.8)	20/74 (27.0)
European	29/64 (45.3)	40/54 (74.1)	35/47 (74.5)	54/74 (73.0)
Hispanic	15/64 (23.4)	-	-	-
Other	-	1/54 (1.9)	6/47 (12.8)	-
HLA mismatch score f
0	3/83 (3.6)	2/54 (3.7)	1/47 (2.1)	0/75 (0.0)
1	6/83 (7.2)	3/54 (5.6)	2/47 (4.3)	7/75 (9.3)
2	27/83 (32.5)	15/54 (27.8)	11/47 (23.4)	27/75 (36)
3	47/83 (56.6)	34/54 (63.0)	33/47 (70.2)	41/75 (54.7)

^a:The GoCAR cohort including the 40 AA patients with RNA-sequencing data and other AA patients included in target sequencing. Genetic ancestry was inferred from genome-wide genotyping data (SNP array of GoCAR and RNAseq of CTOT19) as described in Methods. The race information from SIRPA and VericiDx was based on self-reported.

^b:Both the donor and recipient sex were based on self-reported information.

^c:The follow up time is 5 years for GoCAR¹⁰, 7 years for SIRPA, 2 years for CTOT19¹² and about 1 year VericiDx

^d:The ACR events include borderline rejections.

^e:The donor ancestry information of GoCAR was inferred from genome-wide genotyping data by SNP array and the donor race information from other 3 cohort was based on self-reported.

^f:HLA-mismatch score was derived from 2-digit HLA allele typing (HLA-A, B and DR). Following previous reports for GoCAR¹⁰, the raw mismatch score (scaling from 0 to 6) was categorized as follows: 0 (no mismatches); 1 (1–2 mismatches); 2 (3–4 mismatches); and 3 (5–6 mismatches). The HLA mismatch score from other 3 cohort was calculated the same way as GoCAR.

Extended Data Table 2.

The prevalence of LILRB3-4SNPs variant (Risk) or reference (Non-risk) allele among different ancestry/races in mulGple cohorts.

LILRB3-4SNPs	European	African American	Hispanic	Other
GoCAR (RNAseq + DNAseq, n=264) a
non-Risk	81	67	77	19
Risk	1 (1.2%)	16 (19.3%)	3 (3.8%)	0 (0.0%)
SIRPA (DNAseq, n=54) b
non-Risk	-	46	-	-
Risk	-	8 (14.8%)	-	-
CTOT19 (RNAseq, n=128) c
non-Risk	27	42	41	12
Risk	0 (0.0%)	5 (10.6%)	0 (0.0%)	1 (7.7%)
VericiDX (RNAseq, n=77) d
non-Risk	-	67	-	-
Risk	-	10 (13.0%)	-	-
BioMe (WES, n=30,099) e
non-Risk	9,547	6,488	10,223	2,950
Risk	8 (0.1%)	608 (8.6%)	244 (2.3%)	31 (1.0%)
All-of-Us (WES, n=245,388) f
non-Risk	125,801	43,393	45,401	21,826
Risk	42 (0.0%)	7,571 (14.9%)	840 (1.8%)	509 (2.3%)

^a:genePc ancestry was inferred from SNP array whole genome genotyping (Methods)

^b:self-reported race

^c:genePc ancestry was inferred from RNAseq whole transcriptome genotyping (Methods)

^d:self-reported race

^e:self-reported race

^f:self-reported race, Hispanic was defined as White (race) + Hispanic (ethnicity) and European was defined as White (race) + non-Hispanic (ethnicity).

Extended Data Table 3.

Multivariate Cox model and logistic regression of LILRB3-4SNPs associating with DCGL within AA patients in the GoCAR cohort (n=83) adjusted by APOL1 risk allele number, donor status, donor age, DGF, ACR, post-transplant infection and HLA mismatch.

	%95 confident interval	Hazard (Odds) ratio	P value
Cox model
LILRB3-4SNPs risk alleles (vs non-risk)	(1.55, 12.65)	4.42	0.005
APOL1 single risk allele (vs non-risk)	(0.80, 26.66)	4.62	0.086
APOL1 double alleles (vs non-risk)	(1.14, 37.22)	6.52	0.035
Donor status (deceased vs living)	(0, Inf)	Inf	0.998
Donor age	(1, 1.08)	1.04	0.012
DGF (yes vs no)	(0.18, 1.73)	0.56	0.310
Infection (yes vs no)	(0.17, 2.01)	0.59	0.399
Acute rejection (yes vs no)	(0.60, 6.82)	2.03	0.251
HLA mismatch	(0.22, 1.56)	0.58	0.288
Logistic regression
LILRB3-4SNPs risk alleles (vs non-risk)	(1.38, 62.32)	7.54	0.031
APOL1 single risk allele (vs non-risk)	(0.39, 32.62)	3.10	0.300
APOL1 double risk alleles (vs non-risk)	(1.22, 91.40)	8.50	0.045
Donor status (deceased vs living)	(0, Inf)	Inf	0.992
Donor age	(0.98, 1.07)	1.02	0.336
DGF (yes vs no)	(0.09, 1.77)	0.42	0.252
Infection (yes vs no)	(0.24, 9.25)	1.41	0.706
Acute rejection (yes vs no)	(0.30, 11.86)	1.86	0.494
HLA mismatch	(0.20, 1.91)	0.61	0.372

Data Availability

The bulk RNAseq data of pre-transplant blood from kidney transplant recipients were deposited on Gene Expression Omnibus (GEO) under accession numbers GSE112927 (GoCAR), GSE252272 (CTOT19) and GSE281721 (VericiDx), with race (self-reported), sex and LILRB3-4SNPs risk allele information. The scRNAseq data of pre- and post-transplant PBMCs in the GoCAR cohort was deposited in GEO under the accession number GSE252273. The bulk RNAseq data of post-transplant blood from the GoCAR recipients was deposited in GEO under the accession number GSE261408. The targeted sequencing of GoCAR and SIRPA data was deposited on SRA database with accession number PRJNA1203265. The SNP array data of the GoCAR cohort were published previously¹⁵ and can be requested from the corresponding author within 2–4-week response time frame upon data usage agreement (DUA). The de-identified individual demographic and clinical data can be requested from the corresponding author within 2–4-week response time frame upon data usage agreement (DUA). The genotype and phenotypic data of BioMe biobank was generated by Regeneron and was not publicly available. However, the data will be available for the purpose of validating the results by contacting the corresponding author (weijia.zhang@mssm.edu) within 2–4-week response time frame upon appropriate collaboration and data sharing agreement between the institutes. The genotype and phenotypic data All-of-Us biobank can be accessed through the All-of-Us online workbench (https://www.researchallofus.org).

Code Availability

The customized pipeline for allele-specific analysis and codes for generating figures were deposited under https://github.com/ZephyrSun13/LILRB3_4SNPs.git. The clinical data for validating these codes can be available from the corresponding author within 2–4 weeks respond time frame. The versions for each software can be found in the scripts and also described in the Methods part of the manuscript.