Donor-recipient mismatch at the SIRPA locus adversely affects kidney allograft outcomes

Abstract

Donor-recipient mismatches in histocompatibility antigens recognized by lymphoid cells have been demonstrated to adversely affect allograft outcomes. In contrast, it remains unclear whether mismatches sensed by innate myeloid cells have a similar effect. We investigated the consequences of mismatch in the polymorphic gene encoding signal regulatory protein α (SIRPα) on kidney allograft pathology and survival in mice and humans. We found that SIRPα variants elicit monocyte activation by binding to CD47 and that eliminating SIRPα mismatch or recipient CD47 expression prevented chronic allograft pathology in mice receiving major histocompatibility complex (MHC)–mismatched renal allografts. Human genomic analysis identified two haplotype categories, A and B, encoding SIRPα variants with distinct CD47 binding interfaces. In kidney transplant recipients (N = 455), SIRPα mismatch was associated with increased acute rejection and graft fibrosis in the first posttransplant year, and A recipients of B kidneys had reduced long-term graft survival (hazard ratio, 3.2; 95% confidence interval, 1.5 to 6.9; P = 0.002), a finding that was confirmed in an independent validation cohort (N = 258). Moreover, monocytes in these graft recipients had an activated phenotype. The effects of SIRPα mismatch were independent of ancestry, human leukocyte antigen mismatch, donor-specific antibodies, and delayed graft function. Therefore, these data demonstrate that a donor-recipient mismatch that causes innate immune activation is a determinant of kidney transplantation outcomes.

INTRODUCTION

Current methods for prognosticating transplantation outcomes rely heavily on analysis of histocompatibility antigens that differ between donors and recipients and are sensed by adaptive (T and B) and innate lymphoid (natural killer) cells. Recent data indicate that innate immune cells that belong to the myeloid lineage also sense major histocompatibility (MHC) and non-MHC disparities between the donor and recipient, a process referred to as “innate allorecognition” (1–3). Innate allorecognition in mice drives monocyte differentiation to antigen-presenting dendritic cells (DCs) essential for T cell activation (4–7) and to macrophages that kill allogeneic cells and promote tissue fibrosis (7, 8). Key to initiating this monocyte alloresponse is a mismatch between donors and recipients in the polymorphic gene Sirpa that encodes signal regulatory protein α (SIRPα), an immunoglobulin (Ig) superfamily transmembrane protein that modulates monocyte, DC, and macrophage function (6, 9). At least 10 SIRPα IgV domain variants have been identified in laboratory mice, and amino acid polymorphism among these variants influences their affinity to their binding partner, CD47, which is monomorphic (6, 10). SIRPα and CD47 are coexpressed on monocytic cells but have opposite functions: SIRPα is inhibitory, whereas CD47 is stimulatory (11). As shown in a murine bone marrow plug transplantation model (6), allografts expressing a higher-affinity SIRPα variant than the recipient’s cause host monocyte differentiation to DCs by tipping the balance in favor of activating CD47 signals; in contrast, allografts matched with the recipient at the Sirpa locus do not induce host monocyte differentiation. These effects were independent of MHC disparities (6). Whether SIRPα mismatch influences solid organ allograft outcomes in mice is not known.

The human SIRPα gene (SIRPA) is also polymorphic. An initial study identified 10 distinct alleles on the basis of combinatorial variation in the amino acid sequence of the CD47-binding IgV domain of SIRPα (12). Among the 22 IgV domain residues that interact with CD47, two are polymorphic, dividing the 10 alleles into two groups: v1-like (referred to in this manuscript as A) that contain a leucine at position 66 and a proline at position 99 and v2-like (referred to in this manuscript as B) that contain a serine at position 66 and a proline deletion at position 99. Two studies have provided preliminary evidence that SIRPA A/B donor-recipient mismatch could influence transplantation outcomes in humans. Saliba et al. (13) reported in human leukocyte antigen (HLA)–matched, allogeneic hematopoietic stem cell transplantation that SIRPA mismatch confers a higher risk of chronic graft-versus-host disease. However, the impact of SIRPA mismatch on solid organ transplantation is less clear. A small, HLA-matched, living donor kidney transplantation study showed a trend toward more graft failure and graft inflammation in SIRPA-mismatched cases (14). Therefore, the effect of SIRPA donor-recipient mismatch on short- and long-term graft outcomes after kidney transplantation and the immunological mechanisms that underlie it remain to be established. Such an association, if proven, would provide a tool for predicting allograft outcomes on the basis of a histocompatibility mismatch sensed by the host’s innate myeloid cells. Here, we studied the influence of Sirpa mismatch on kidney allograft rejection in mice. We further interrogated the effects of SIRPA A/B donor-recipient mismatch on histological and clinical allograft outcomes in two kidney transplant cohorts that included both HLA-matched and mismatched donor-recipient pairs as well as living and deceased donor allografts. We also investigated the mechanisms that underlie the effects of SIRPA mismatch on the host’s alloimmune response by analyzing monocytes from patients who received SIRPA-matched or -mismatched renal allografts.

RESULTS

SIRPα mismatch drives chronic pathology in mouse kidney allografts

SIRPα mismatch triggers mouse monocyte and macrophage activation (4), suggesting that it could play a role in driving tissue pathology associated with chronic rejection. To test this possibility, we investigated the effect of SIRPα mismatch on allograft outcomes in a mouse kidney transplantation model in which allografts are not acutely rejected but survive long term (>1 year) and undergo slow histological changes (15). We compared kidneys transplanted between nonobese diabetic (NOD) and C57BL/6 (B6) mice, which are mismatched for MHC and SIRPα, with kidneys transplanted between NOD.B6-Sirpa and B6 mice, which share the same SIRPα but are MHC mismatched (12). NOD SIRPα binds with greater affinity to CD47 than B6 SIRPα and causes monocyte activation in B6 recipients (6, 10). We also transplanted NOD kidneys to B6.CD47^−/− mice to ascertain whether any differences in outcomes observed between NOD and NOD. B6-Sirpa allografts are mediated by donor SIRPα binding to recipient CD47. Ten to 12 months after kidney transplantation, histological changes consistent with chronic rejection (nodular cellular infiltration, fibrosis, and immune cell accumulation that includes T cells, monocyte-derived DCs, and macrophages) were present in Sirpa-mismatched (NOD to B6) allografts but were reduced in Sirpa-matched (NOD. B6-Sirpa to B6) allografts (Fig. 1, A and B). Interferon-γ (IFN-γ) production by CD8 T cells was also significantly (P = 0.03) lower, suggesting decreased activation of intragraft cytotoxic T cells in the absence of Sirpa mismatch (Fig. 1B and fig. S1). Chronic rejection was similarly ameliorated in B6.CD47^−/− recipients of NOD kidneys (Fig. 1, A and B), indicating that the Sirpa mismatch effect was mediated through CD47. Therefore, SIRPα mismatch is a driver of immune-mediated chronic allograft pathology in mice. This finding supports investigating the effect of SIRPα mismatching on kidney transplantation outcomes in humans.

Fig. 1. SIRPα mismatch drives chronic pathology in mouse kidney allografts.

Allogeneic NOD or NOD.B6-Sirpa mouse kidneys were transplanted to bilaterally nephrectomized wild-type (B6) or CD47 knockout (B6.Cd47^−/−) recipients (n = 3 to 6 mice per group). NOD.B6-Sirpa mice are NOD congenic mice that are identical to the parental NOD strain except for an ~2–mega–base pair segment that includes B6-Sirpa instead of NOD Sirpa. (A and B) Shown are graft histopathology with quantification of graft infiltrate and fibrosis (A) and enumeration of graft immune cell accumulation (B) at time of harvest 300 to 360 days posttransplantation. Scale bars in (A) indicate a resolution of 500 μm, and red arrows indicate areas of inflammatory infiltrate. Data points in (A) and (B) represent individual mice, and horizontal bars indicate mean. Groups were compared by one-way ANOVA with Tukey’s correction for multiple comparisons. H&E, hematoxylin and eosin; Mono-DC, monocyte-derived dendritic cells.

SIRPA is polymorphic in humans

Data on SIRPA polymorphism in humans are limited to a single published study of 37 unrelated individuals in whom the SIRPA exon encoding the CD47-binding IgV domain was sequenced (12). To define SIRPA allelic variation more broadly, we analyzed 5008 human genome sequences available through the 1000 Genomes Project to identify all annotated single-nucleotide polymorphisms (SNPs) in SIRPA, derive known haplotypes, and determine their distribution across ethnic groups. We uncovered 77 SNPs that span all exons, 29 of which (the largest group) resided in the IgV-coding domain (fig. S2A). These translated to ~150 haplotypes, most of which, 102, were defined by SNPs in the IgV domain (fig. S2B). Only 13 of these were present at a frequency of >1%, and the top 10 accounted for ~92% of variation in the human population, with haplotype or variant 1 (v1) being the most common (Fig. 2A). v1 was the most prevalent in all ethnic groups, with the highest representation in Africans and Europeans (69 and 62%, respectively), whereas v2 and v3 were most prevalent in East Asians (23 and 17%, respectively) (Fig. 2B). As previously reported (12, 13), variation at two amino acid residues in the SIRPα IgV domain that interact with CD47 divided the 10 most common variants into two classes, v1- and v2-like, that we refer to as A and B in this study. A alleles (v1, v4, v5, v6, and v9) contain a leucine at position 66 and a proline at position 99, and B alleles (v2, v3, v7, v8, and v10) contain a serine at position 66 and a proline deletion at position 99 (fig. S3). The ethnic distribution of A and B alleles is depicted in Fig. 2C. We genotyped 48 healthy individuals and found 42% to be homozygous for A alleles (AA), 15% homozygous for B (BB), and 44% heterozygous (AB) (fig. S4), a distribution consistent with the higher prevalence of A alleles across ethnic groups.

Fig. 2. SIRPA haplotype distribution in the human population.

Five thousand eight human genome sequences available through the 1000 Genomes Project were analyzed to identify SNPs in the SIRPA gene. (A and B) The 10 most common haplotypes were derived, and their distribution across the whole population (A) and across ethnic groups (B) is shown. (C) Variants v1 to v10 were divided into two classes, A (or v1-like) and B (or v2-like), on the basis of polymorphisms in two amino acids that alter the SIRPα-CD47 binding interface. A and B haplotype frequencies across ethnic groups are shown.

SIRPA mismatch was associated with increased acute kidney allograft rejection and fibrosis in humans

To investigate associations between SIRPA mismatch and kidney allograft outcomes in humans, we genotyped all adult kidney transplant donor-recipient pairs transplanted between 2013 and 2020 at the University of Pittsburgh Medical Center (UPMC) from whom DNA samples were available (N = 455). Baseline characteristics of the study population are shown in Table 1. Most of the patients were Caucasian (84%) and were recipients of either living (64%) or deceased (36%) donor allografts. Most of the recipients (87.5%) were maintained on corticosteroid-free immunosuppression. Overall, 48 patients (10.5%) had delayed graft function, and 121 (27%) had either clinical or subclinical acute rejection (AR, Banff ≥ 1A) in the first posttransplant year. AR rates at our center reflect clinical AR diagnosed on for-cause biopsies as well as subclinical AR detected by 3-month and 12-month surveillance biopsies performed on all patients unless a contraindication existed. SIRPA genotype frequencies (AA, BB, and AB) in donors and recipients (Fig. 3A) were consistent with genotype distribution in healthy volunteers (fig. S4). On the basis of SIRPA genotypes, patients were classified into three groups: matched (n = 174, 38%), A→B mismatched if more B alleles were present in the recipient (n = 151, 33%), and B→A mismatched if more B alleles were present in the donor (n = 130, 29%) (Fig. 3B). We chose this classification because it focuses on amino acid variations that could potentially influence the binding of SIRPα to its ligand, CD47 (fig. S3). As shown in Table 1, relevant clinical covariates were comparable among the three groups.

Table 1.

Patient baseline characteristics.

Variable	All patients	Matched	A→B	B→A	P value
Numbers	455	174	151	130
Recipient age (in years)*	53 ± 14	54 ± 14	53 ± 13	52 ± 15	0.6
Recipient gender (male gender %)	62%	61%	66%	58%	0.4
Recipient ethnicity					0.5
Caucasian	84%	86%	83%	83%
African American	14%	13%	14%	16%
Others	2%	1%	3%	1%
Primary renal diagnosis					0.6
Diabetes/hypertension	41%	41%	44%	37%
Glomerular disease	18%	17%	19%	20%
Congenital/inherited	13%	14%	15%	11%
Others	28%	28%	23%	32%
Donor age (in years)*	43 ± 13	42 ± 13	44 ± 13	43 ± 13	0.5
Donor type					0.3
Living	64%	67%	65%	58%
Deceased	36%	33%	35%	42%
KDPI > 50%	15%	16%	13%	15%	0.8
HLA mismatches (10-antigen)*	6.3 ± 2.2	6.1 ± 2.3	6.5 ± 2.1	6.3 ± 2.2	0.2
Class I PRA ≥20%	14.5%	15.5%	10.6%	16.9%	0.3
Class II PRA ≥20%	15.4%	15.5%	10.6%	17.7%	0.2
Preformed DSA	5%	5%	6%	4%	0.5
Cold ischemia time (min)*	333 ± 401	306 ± 391	315 ± 383	390 ± 430	0.1
Warm ischemia time (min)*	37 ± 10	37 ± 10	36 ± 9	38 ± 10	0.2
Thymoglobulin induction	98%	97%	98%	99%	0.3
% Patients on maintenance corticosteroids	12.5%	11.5%	13.2	13.1%	0.9

^*Mean ± SD.

DSA, donor-specific antibody. Blank entries indicate “not applicable.”

Fig. 3. Donor and recipient SIRPA genotypes and mismatch categories.

(A and B) Prevalence of SIRPA AA, AB, and BB genotypes (A) and distribution of SIRPA matched, A→B mismatched, and B→A mismatched patients (B) in the study cohort (455 renal transplant donor-recipient pairs).

We first examined the relationship between SIRPA A/B mismatch and early graft pathology [AR and interstitial fibrosis and tubular atrophy (IFTA) in the first year posttransplantation]. Both B→A (P = 0.004) and A→B (P = 0.04) mismatches were associated with significantly increased incidence of AR compared with the matched group (Fig. 4A). There were no differences in the incidence of borderline rejection (Fig. 4A), the histological grade of AR, the incidence of clinical versus subclinical AR, or microvascular inflammation score among the groups (fig. S5). Given that delayed AR, defined in this study as AR between 5 and 12 months that is either de novo (no prior rejection) or persistent/recurrent (after an earlier rejection), is associated with increased graft loss (16, 17), we analyzed the relationship between SIRPA mismatch and delayed AR (Fig. 4B). B→A mismatched patients had a threefold increase (P = 0.0004) and A→B patients had a twofold increase (P = 0.05) in delayed AR compared with matched patients. The difference between the B→A and A→B groups was also statistically significant (P = 0.02). Similarly, the B→A group experienced higher incidence of late AR (>1 year posttransplantation) compared with the A→B group (P = 0.009) (Fig. 4C). Together, the data indicate that either mismatch is associated with increased AR in the first posttransplantation year but that the B→A SIRPA mismatch puts recipients at higher risk of persistent or recurrent allograft inflammation. Histological type and grade of late AR are shown in table S1.

Fig. 4. SIRPA mismatch is associated with increased acute kidney allograft rejection and fibrosis in humans.

(A to D) SIRPA matched (M), A→B mismatched (A→B), and B→A mismatched (B→A) patients were compared on the basis of rejection diagnosis anytime within the first year after transplantation [no rejection (NR), borderline rejection (BL), or Banff ≥ 1A AR (≥1A-AR)] (A); de novo or persistent delayed AR 5 to 12 months posttransplantation in patients who had at least two biopsies (persistent was defined as AR in someone who had AR in both the early and delayed biopsy, irrespective of treatment) (B); AR-free graft survival (GS) beyond the first year (C); and IFTA categories [minimal (score = 0 to 0.5), mild (1 to 2.5), and moderate or severe (≥3)] (D). Numbers within or above the bars in (A), (B), and (D) represent the percentage of graft recipients in that group. Patient groups were compared by chi-square test in (A), (B), and (D). Survival analysis is reported using the Kaplan-Meier method, and survival curves were compared by log-rank test in (C).

We next investigated the association between SIRPA mismatch and IFTA by 1 year posttransplantation, which is another important predictor of graft loss (18). A significantly greater proportion of B→A and A→B mismatched patients had moderate or severe IFTA (score ≥ 3) compared with matched patients (P < 0.0001 and P = 0.002, respectively) (Fig. 4D). In contrast, a significantly greater proportion of matched patients had minimal or no IFTA (score < 1) compared with either mismatched group (P < 0.0001 and P = 0.001) (Fig. 4D). We also observed a correlation between B→A mismatch and delayed graft function (P = 0.02), although we did not observe a significant difference in de novo donor-specific antibody development (P = 0.1) (fig. S6). Multivariable analysis demonstrated that the association between SIRPA mismatch and AR is independent of these as well as other related, confounding variables such as cold ischemia time, kidney donor profile index (KDPI), and the HLA mismatch [B→A mismatch: adjusted odds ratio (OR), 2.3; 95% confidence interval (CI), 1.3 to 4.1; P = 0.004] (table S2). The same was true for IFTA (B→A mismatch: adjusted OR, 5.0; 95% CI, 2.3 to 11.0; P < 0.0001) (table S2). In summary, SIRPA B→A donor-recipient mismatch was associated with delayed graft function, persistent allograft inflammation, and premature IFTA, all of which portend poor long-term allograft outcomes. In comparison, A→B mismatch correlated only with early AR and increased IFTA in the first year.

SIRPA mismatch was associated with shortened kidney allograft survival in humans

Given the correlation between SIRPA B→A mismatch and multiple risk factors for shortened long-term allograft survival, we compared 7-year death-censored and overall graft survival among the three patient groups. Compared with the matched group, B→A mismatched patients had significantly worse death-censored (75% versus 92%, P = 0.001) and overall graft survival (53% versus 74%, P = 0.003) (Fig. 5, A and B). In contrast, A→B mismatched patients had comparable death-censored (86% versus 92%, P = 0.3) and overall graft survival (69% versus 74%, P = 0.7) to the matched patients. In addition, B→A mismatched death-censored and overall graft outcomes were significantly worse than the A→B mismatched group (P = 0.04 and 0.009, respectively). The increased hazard for death-censored graft loss in the B→A mismatched patients was independent of HLA mismatches, delayed graft function, and donor-specific antibody development [adjusted hazard ratio (HR), 2.7; 95% CI, 1.2 to 5.7; P = 0.01] (table S3). The same was true for overall graft loss (adjusted HR, 1.8; 95% CI, 1.1 to 2.9; P = 0.015) (table S4). Because the effect of B→A mismatch on graft outcomes was dependent on either AR or IFTA (tables S3 and S4), we hypothesized that inflammation and fibrosis mediate the relationship between the mismatch and graft loss. Mediation analysis revealed a significant indirect effect of SIRPA B→A mismatch on graft loss through both inflammation and IFTA (coefficient, 0.33; 95% CI, 0.04 to 0.8; P = 0.04), whereas the direct effect of the mismatch on graft loss was not significant (coefficient, 0.09; 95% CI, −0.45 to 0.63; P = 0.74) (Fig. 5C). Therefore, B→A mismatch leads to shortened allograft survival, an effect that is mediated through inflammation and premature allograft scarring and is independent of HLA mismatch, delayed graft function, and donor-specific antibody development.

Fig. 5. SIRPA mismatch is associated with shortened kidney allograft survival in humans.

(A and B) Death-censored (A) and overall GS (B) in matched (M), A→B mismatched (A→B), and B→A mismatched (B→A) patients. Survival analysis is reported using the Kaplan-Meier method, and survival curves were compared by log-rank test in (A) and (B). (C) Mediation analysis of B→A mismatch and death-censored graft loss (GL) demonstrating that GL in B→A mismatched (mm) patients is fully mediated by a combination of tubulointerstitial inflammation (INFL) and IFTA. Shown are the derived coefficients with 95% CIs of the coefficients in parentheses. SIRPA mm status was used as the risk variable, GL as the outcome variable, and both allograft INFL and IFTA by 1 year posttransplantation as the mediators. HLA mismatch, delayed graft function, and donor-specific antibody were the covariates in this model. Black dashed arrows (middle) depict individual effects, the red dashed arrow (top) depicts the total indirect effect, and the black solid arrow (bottom) depicts the total direct effect.

SIRPA mismatch effect on kidney allograft outcomes was not influenced by ancestry or relatedness

Given that the distribution of SIRPA A and B alleles is not uniform among ethnic groups (Fig 2C), we investigated the contribution of donor and recipient ancestry to differences in graft outcomes between mismatched and matched patients. We first assessed the distribution of donor and recipient self-reported ancestry in the matched versus A→B versus B→A mismatched groups. There were no differences among the three groups (fig. S7, A and B), with the great majority of donors and recipients being Caucasian. Individuals of East Asian ancestry, who have the highest prevalence of the B allele, were very rare in our cohort (five recipients and two donors). Moreover, Caucasian-to-Caucasian transplants were the most common across all groups (fig. S7C). A subgroup analysis of Caucasian-to-Caucasian transplants still demonstrated a strong association between B→A mismatch and increased AR, IFTA, and long-term graft loss (fig. S7, D to G).

Because donor-recipient relatedness can influence the effect of a gene mismatch on graft outcomes, we next compared HLA epitope load between matched, A→B mismatched, and B→A mismatched patients as a surrogate measure of relatedness. We also compared the prevalence of unrelated versus related transplants among the three groups. Neither HLA epitope load nor prevalence of related versus unrelated transplants were different among the groups (fig. S8). Moreover, multivariable analyses showed that SIRPA mismatch was associated with worse graft outcomes (AR, IFTA, and graft loss) independently of donor-recipient relatedness (table S5).

The association between SIRPA mismatch and shortened kidney allograft survival was validated in an independent patient cohort

We evaluated the effect of SIRPA mismatch on long-term kidney allograft survival in an independent cohort of 258 living donor-recipient pairs transplanted at Northwestern University. Fifty-one percent of the pairs in this cohort were SIRPA matched (n = 132), 30% had an A→B mismatch (n = 78), and 19% had a B→A mismatch (n = 48). Baseline characteristics are shown in table S6. No differences were observed among the matched and mismatched groups except for a higher degree of HLA mismatch in the A→B group. A comparison of baseline characteristics between the UPMC and Northwestern cohorts is shown in table S7. Northwestern patients were younger and less likely to be Caucasian, and all received living donor allografts and were better HLA matched. Moreover, they received alemtuzumab (Campath) instead of thymoglobulin induction, and 24% were maintained on corticosteroids versus 12.5% in the UPMC cohort. Death-censored graft survival was significantly lower at 4 years in the B→A mismatched group than either the matched or A→B group (P = 0.036 and P = 0.05, respectively; Fig. 6A). The groups, however, were not significantly different at 5 years (P = 0.1 and 0.2). Similarly, overall graft survival was significantly worse in the B→A mismatched group compared with the matched group (P = 0.04) at 4 years (Fig. 6B). HRs for death-censored graft survival at 4 and 5 years for B→A mismatched compared with matched donor-recipient pairs were 3.0 (95% CI, 1.04 to 8.4) and 2.1 (95% CI, 0.8 to 5.5) in the Northwestern versus 3.5 (1.5 to 8.5) and 2.9 (1.3 to 6.6) in the UPMC cohort, respectively (Fig. 6C). HRs for the combined cohorts after bias-corrected bootstrap validation confirmed the adverse impact of the B→A mismatch on graft survival (Fig. 6C).

Fig. 6. SIRPA mismatch is associated with shortened kidney allograft survival in humans (Northwestern and combined cohorts).

(A and B) Death-censored (A) and overall GS (B) in matched (M), A→B mismatched (A→B), and B→A mismatched (B→A) patients in the Northwestern University cohort. Survival analysis is reported using the Kaplan-Meier method, and survival curves were compared by log-rank test at 4 and 5 years (*P = 0.036, B→A versus M; P = 0.05 B→A versus A→B; **P = 0.1, B→A versus M; P = 0.2 B→A versus A→B) (#P = 0.04, B→A versus M; P = 0.08, B→A versus A→B; ##P = 0.1, B→A versus M; P = 0.17 B→A versus A→B). (C) The forest plot demonstrates HRs and CIs for 5-year and 4-year death-censored GS in B→A mismatched compared with matched patients for the Northwestern University (NW) and UPMC cohorts separately and together. Also shown are the HRs for death-censored graft survival for both cohorts together after bias-corrected bootstrap validation (1000 and 3000 bootstraps at 99% CI, respectively). HRs were derived using unadjusted Cox proportional hazards analysis.

SIRPα mismatch affects monocyte phenotypes in kidney transplant recipients

Because the B→A mismatch correlated with the worst long-term allograft outcomes, we asked whether this mismatch was associated with recipient monocyte activation as observed in mice (6). We performed spectral flow cytometry on peripheral blood mononuclear cells (PBMCs) from SIRPA matched (n = 18, 8 rejectors and 10 non-rejectors) and SIRPA B→A mismatched kidney allograft recipients (n = 19, 10 rejectors and 9 nonrejectors). Rejectors were defined as graft recipients who had at least one episode of AR (≥Banff 1A) during the first year after transplantation, whereas nonrejectors were those who did not reject in the first year. All PBMC samples were collected 3 months after transplantation before treatment for any AR episode. After gating on all monocytes, t-distributed stochastic neighbor embedding (t-SNE) analysis of 22 surface markers was performed (fig. S9, A and B). No difference in proportion of classical, intermediate, or nonclassical monocytes was observed between the two groups, with classical monocytes representing ~80% of all monocytes (fig. S9C). However, a conspicuous shift in overall monocyte phenotype was observed in the mismatched versus matched group at 3 months after transplantation (Fig. 7A). The phenotypic shift was characterized by increased expression of programmed cell death ligand 1, CCR2, CCR7, CD11c, CD74, and CD163 in B→A mismatched patients (Fig. 7B), suggesting an activated phenotype. A similar shift in monocyte phenotype was observed irrespective of whether a B→A mismatched patient was a rejector (had at least one episode of AR in the first posttransplant year) or not (Fig. 7C). Analysis of individual monocyte subsets in the same patients demonstrated the up-regulation of these activation markers predominantly in the classical monocyte subpopulation and, to a lesser extent, in the intermediate and nonclassical subpopulations (fig. S10). Monocyte phenotypes were generally similar in the matched and mismatched groups before transplantation (fig. S11). Therefore, SIRPA B→A mismatch is associated with monocyte activation after transplantation irrespective of rejection status.

Fig. 7. B→A SIRPA mismatch is associated with an activated monocyte phenotype in kidney transplant recipients.

(A) Shown are t-SNE plots based on 22 markers using concatenated files of 30,750 monocytes per patient from matched (M, n = 18) and B→A mismatched (B→A, n = 19) renal allograft recipients. (B) Expression as MFI of individual markers from matched (M) and B→A mismatched (B→A) patients. (C) Expression (as MFI) of individual markers from M and B→A patients in the rejector (top) and nonrejector (bottom) groups. Data are presented as mean ± SEM. Data points represent individual patients. Patient groups were compared by Mann-Whitney U test in (B) and (C). PD-L1, programmed cell death ligand 1.

We also performed multiplex immunofluorescence staining analysis on 3-month renal allograft biopsies with acute cellular rejection from B→A mismatched and matched patients. The tissue was stained for common monocyte and DC cell surface markers, including CD11c, CD14, CD68, CD163, HLA-DR, and SIRPα (fig. S12A). There were no differences in the density of CD11c⁺ DCs or CD68⁺ macrophages. However, we observed a greater density of M2-polarized macrophages, identified as CD68⁺CD163⁺, in matched than B→A mismatched grafts (P = 0.028), whereas we did not observe a significant difference in the density of CD68⁺HLA-DR⁻ macrophages (P = 0.06; fig. S12B). No differences were noted for other cell subsets or markers. This result suggests that macrophages with regulatory or reparative function are increased in matched grafts.

DISCUSSION

Detecting mismatches in genes that encode major or minor histocompatibility antigens has been a mainstay of donor selection and risk stratification in solid organ and bone marrow transplantation because these antigens trigger potent cellular and humoral adaptive immune responses in the host. Here, we expanded the spectrum of clinically important donor-recipient mismatches to include SIRPA, a non-HLA polymorphic gene encoding a membrane protein directly involved in innate allorecognition and activation of monocytic cells. We demonstrated a role for SIRPA mismatch in chronic renal allograft pathology in mice and established its correlation with graft inflammation, graft fibrosis, and graft loss in human kidney transplantation. The key clinical observation that SIRPA B→A mismatch is associated with worse allograft survival was validated in an independent, external cohort. The negative influence of SIRPA mismatch on allograft outcomes was independent of HLA mismatch, was associated with monocyte activation, and was likely mediated through increased risk of allograft inflammation and premature fibrosis. These findings suggest that innate allorecognition by monocytic cells, which has so far been investigated only in murine models, has important implications in humans and indicate that donor and recipient SIRPA genotyping can be used to further risk stratify renal allograft recipients.

A key observation in this manuscript is that SIRPα mismatch in one direction alone, B→A, is associated with worse long-term graft outcomes. This observation is supported by mouse experiments demonstrating that mismatch in one direction (higher-affinity non-self SIRPα variant is on graft cells) triggers greater recipient monocyte differentiation (6). It is unclear whether A and B human SIRPα variants differ in their binding to CD47. Prior biophysical studies have shown that SIRPα B variants have slightly greater affinity to CD47 than A variants (19, 20). Although small, the observed difference could be meaningful because of the weak nature of the SIRPα-CD47 interaction, reminiscent of MHC-peptide binding to the T cell receptor, wherein minor variations in affinity lead to substantial differences in T cell responses (21–24).

We also observed increased risk of AR and IFTA in the first year after transplantation in both mismatched groups, but the risk in the A→B group did not translate to increased delayed AR (5 to 12 months after transplantation), increased late AR (beyond the first year), or worse long-term graft survival. This finding could potentially be explained by a graft-versus-host reaction in the A→B mismatch scenario. When an A kidney is transplanted to a B recipient, passenger donor leukocytes, such as resident DCs and macrophages, in the graft could respond to the B variant on infiltrating donor leukocytes and exacerbate graft inflammation and fibrosis. Given that donor-derived passenger leukocytes dissipate over time, as early as the first year after transplantation (25), the graft-versus-host reaction is not expected to last long enough to sustain chronic graft inflammation and affect long-term outcomes. In contrast, in bone marrow transplantation where donor cells persist, graft-versus-host reactions can be clinically meaningful, as suggested by a study that found an association between SIRPA mismatch and chronic graft-versus-host disease (13).

In addition to its effects on AR, IFTA, and graft loss, B→A mismatch also increased delayed graft function risk. The role of the innate immune system in ischemia-reperfusion injury is well established (26), and the contribution of CD47 activation is supported by murine studies (27, 28). We also observed a trend toward more de novo donor-specific antibody development in the B→A mismatch group, which could be due to a mature immune response driven by monocyte-derived DCs and subsequent T cell–dependent B cell activation when the SIRPA mismatch is present (6, 7). Despite these associations, B→A mismatch worsened allograft outcomes independently of delayed graft function and donor-specific antibody development. Mediation analysis pointed to fibrosis as an important intermediary, driven directly by monocyte differentiation to inflammatory DC or macrophages and indirectly by T cell–dependent AR (7). Monocyte phenotyping revealed signatures of activation and differentiation in B→A recipients irrespective of AR. This is consistent with the mediation analysis and could explain why many patients advance to chronic allograft fibrosis without experiencing AR (29).

Our study has some limitations. Despite the independent association between SIRPA B→A mismatch and shortened graft survival, the retrospective study design limits our ability to confirm a causal relationship. It remains to be seen whether the SIRPA mismatch effect applies to patients on corticosteroid maintenance, which is standard practice at approximately half of all transplant centers in the US. The smaller sample size of patients on maintenance corticosteroids in our study cohort precluded this analysis. Last, the possibility that SIRPα differences between donor and recipient would result in antibody formation directed at the SIRPα polymorphism could not be tested here but will be a focus of future studies.

The findings of our study have important biological and clinical implications. First, innate allorecognition, the ability of myeloid cells to distinguish between self and allogeneic non-self, was initially found and characterized in mice (4, 6, 7, 30). Here, we have provided evidence that this phenomenon is relevant to human organ transplantation and that it possibly enhances adaptive alloimmune responses by activating innate immune cells. Second, we have identified a SIRPA mismatch (B→A) present in approximately one-third of all patients as an independent risk factor for premature fibrosis and adverse allograft outcomes. Thus, SIRPA genotyping can be readily used as a meaningful risk stratification strategy to individualize patient monitoring and immunosuppression. It will also aid clinical trial design by identifying and targeting high- or low-risk populations for immunosuppression augmentation or immunosuppression minimization, respectively.

MATERIALS AND METHODS

Study design

To investigate whether SIRPα mismatch influences kidney transplant pathology and survival, we performed mouse-to-human investigations that included (i) a mouse kidney transplantation model in which allografts undergo chronic rejection, (ii) in silico SNP analysis using the 1000 Genomes Project public database containing 5008 human genome sequences to define human SIRPA allelic variation, and (iii) retrospective observational studies of two independent cohorts of adult renal transplant recipients who underwent transplantation at the UPMC and at Northwestern Memorial Hospital (independent validation cohort). All patients with DNA available from both donors and recipients were included in the analysis. We used Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines by the Enhancing the Quality and Transparency of Health Research (EQUATOR) network for reporting (31). A completed STROBE checklist is provided (data file S1).

Mice

C57BL/6J (B6), NOD/ShiLtJ (NOD), and B6.129S7-Cd47^tm1Fpl/J (B6.CD47^−/−) mice were purchased from the Jackson Laboratory. NOD.NOR-Ila-D2Gul482 (NOD.B6-Sirpa) mice were obtained from J.D. (University of Toronto) (12). All mice were maintained at the University of Pittsburgh animal facility under specific pathogen–free conditions. All animal procedures were performed with approval of the Institutional Animal Care and Use Committee at the University of Pittsburgh.

Mouse kidney transplantation

Mouse kidney transplants were performed as previously described (32). Recipient native kidneys were removed before transplanting the donor kidney. Allograft rejection was monitored by visual observation of recipients for signs of uremia (lethargy, decreased mobility, and ruffled hair) or death.

Mouse histopathology

Kidney allograft tissue was fixed in formalin, paraffin-embedded, sectioned, and stained with hematoxylin and eosin and Masson’s trichrome (MT) stains (Magee-Women’s Research Institute Histology and Microimaging Core, University of Pittsburgh). Slides were scanned on a Zeiss Axioscan.Z1 with a 20× objective. Fibrosis was quantified using the ImageJ software on the basis of MT staining. The total tissue area was determined, and the fibrotic area (stained in blue with MT) was measured by applying a blue color filter. The percentage of fibrosis was calculated by dividing the fibrotic area by the total area. Quantitation was performed by D.Z. in a blinded fashion.

Mouse flow cytometry

Single-cell suspensions of kidney allografts were prepared as previously described (32). Leukocytes were isolated by gradient centrifugation using lympholyte M (Cedarlane Labs). Biotin- or fluorochrome-conjugated antibodies (clone name indicated) against mouse targets are described in table S8. Staining was performed in phosphate-buffered saline (PBS) supplemented with 2% fetal calf serum and 2 mM EDTA for 30 min at 4°C protected from light. CD8⁺ T cells were stained for intracellular IFN-γ as previously described after a 16-hour stimulation with donor-specific allogeneic splenocytes (32). Samples were acquired on an LSRII Fortessa (BD Biosciences) and analyzed by FlowJo (BD Biosciences). Flow gating was performed as previously described (6).

Human genomic analysis

SIRPA SNP genomic analysis was performed in silico using the 1000 Genomes Project public database containing 5008 human genome sequences. This query returned 77 SNPs distributed throughout the SIRPA exons (29 IgV, 18 IgC2, 8 IgC1, 3 transmembrane, and 19 intracellular). These SNPs were uploaded and analyzed using the National Cancer Institute LD-HAP tool (33), which output >150 haplotypes (102 haplotypes from the SIRPA IgV domain). This IgV haplotype list was trimmed to 10 haplotypes that occurred with >1% frequency across five different ethnic backgrounds and accounted for ~92% of variation in the human population. Ethnic backgrounds included were East Asian, South Asian, European, African, and American.

Human genotyping by polymerase chain reaction

SIRPA typing was performed using three sets of primers in three separate reaction tubes as previously published (13). The first two sets (reactions 1 and 2) were used to identify the A and B alleles, respectively, and the third set (reaction 3) was used to confirm the B allele. All reagents were a gift from J.-H. Lee, One Lambda (a Thermo Fisher Scientific brand).

Clinical study

For the UPMC cohort, we performed a retrospective observational cohort study of 455 adult renal transplant recipients who underwent transplantation between January 2013 and December 2019 at the UPMC and had DNA available from both donors and recipients. Patients were followed for a maximum of 7 years. Mean follow-up was 55.5 ± 21 months. The study protocol was approved by the University of Pittsburgh Institutional Review Board (IRB approval no. 22070088).

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 to 84 months). A total of 796 biopsies were performed in 455 patients. Three hundred fifty-six patients (78%) had at least one biopsy in the first posttransplant year. Biopsies were adjudicated by a single pathologist (P.R.) blinded to SIRPA genotype using Banff 2019 criteria. Patients were screened for anti-HLA donor-specific antibodies at 0, 1, 3, 6, 9, 12, 18, and 24 months, and an adjusted median fluorescence intensity (MFI) ≥ 1000 was used as the cutoff.

Induction immunosuppression consisted of thymoglobulin or basiliximab (<3% of patients) and methylprednisolone, followed by a 7-day prednisolone taper. Patients were maintained on mycophenolate mofetil and tacrolimus (target trough concentrations: 8 to 12 ng/ml for the first 3 months and 6 to 10 ng/ml afterward). Patients with calculated panel reactive antibodies (cPRA) > 90% were maintained on oral prednisolone (5 mg/day) (12.5% of patients). Banff 1A or 1B T cell–mediated rejection (TCMR), whether clinical or subclinical, was treated with three daily doses of methylprednisolone. Banff ≥ 2A and steroid-resistant TCMR were treated with thymoglobulin. Acute antibody-mediated rejection was treated with plasmapheresis and intravenous Ig. Prednisolone (5 mg/day) was added to the maintenance regimen in all patients who had rejection.

The number of mismatched eplets, referred to as eplet load in this study, was used as a surrogate for quantifying the genetic disparity between donors and recipients. To calculate HLA eplet load, we first extrapolated the four-digit HLA typing of donors and recipients from the available two-digit data using the HaploSFHI (V2.05) software (34). The number of eplet mismatches for each donor-recipient pair was then calculated using the HLAMatchMaker Eplet Matching software (v2.1).

The validation cohort from Northwestern University consisted of 258 adult kidney transplant recipients who underwent living donor transplantation at Northwestern Memorial Hospital and had DNA available from both donors and recipients for the SIRPA genotyping. Molecular HLA typing for HLA-A, -B, -C, -DRB1, -DQA1, and -DQB1 loci was also available for all donors and recipients. Patients were followed for a mean of 53 ± 15 months. Induction immunosuppression consisted predominantly of alemtuzumab and methylprednisolone (72%). Most patients (89%) received calcineurin inhibitors and mycophenolic acid as maintenance immunosuppression. Seventy-six percent of patients were maintained on corticosteroid-free maintenance immunosuppression. Patients were followed for a maximum of 5 years. Mean follow-up was 53 ± 14.6 months. The study protocol was approved by the Northwestern University Institutional Review Board (IRB no. STU00221880).

Human monocyte phenotyping

PBMCs were separated from peripheral blood by Ficoll density centrifugation and cryopreserved in the Starzl Transplantation Institute Biorepository (IRB approval no. 19030383). For analysis, PBMC aliquots were thawed and assessed by spectral flow cytometry using a Cytek Aurora flow cytometer. Cells were stained using 22 surface markers in PBS supplemented with 2% fetal calf serum and 2 mM EDTA for 30 min at 4°C protected from light (table S9), and individual marker expression was determined after gating on CD14⁺ monocytes using FlowJo and Cytobank software. Nonclassical, intermediate, and classical monocyte subpopulations were identified and quantified (35).

Quantification of intragraft myeloid cells

Multiplex immunofluorescence staining of intragraft myeloid cells was performed using a panel of six antibodies: CD11c (1:400; clone, D3V1E; Cell Signaling Technology), CD14 (1:500; clone, D7A2T; Cell Signaling Technology), CD68 (1:800; clone, D4B9C; Cell Signaling Technology), CD163 (1:400; clone, 10D6; Biocare Medical), HLA-DR (1:100; clone, CR3/43; Abcam), and SIRPα (1:500; clone, D6I3M; Cell Signaling Technology). Slides were scanned using the Pheno Imager HT spectral imaging system (Akoya Biosciences), as described (36). Thirty-nine regions of interest were identified using Phenochart (matched, n = 20; B to A mismatched, n = 19). Spectral libraries were then analyzed using the InForm image analysis software (Akoya Biosciences). Cell density was calculated as the number of target cells divided by the total number of 4′,6-diamidino-2-phenylindole–positive cells in each tissue category.

Statistical analysis

Continuous variables are presented as mean ± SD or SEM; categorical variables are presented as percentages. Continuous variables were compared using independent samples t test or Mann-Whitney U test. Multiple groups were compared using analysis of variance (ANOVA) with Dunnett’s or Tukey’s post hoc correction or Kruskal-Wallis test. Chi-square test was used to compare categorical variables. Univariate and multivariable logistic regression models were used to examine the association between SIRPA mismatch and AR or IFTA. Overall graft survival and death-censored graft survival were analyzed by the Kaplan-Meier method, and survival curves between various patient groups were compared by log-rank tests. A multivariable Cox proportional hazards model was used to assess the independent effect of SIRPA mismatch on graft loss. Mediation analysis [PROCESS macro (v.4) for SPSS, model 6 with 5000 bootstrap iterations] was performed with SIRPA B to A mismatch as the risk variable, death-censored graft loss as the outcome variable, and allograft tubulointerstitial inflammation (Banff t + i scores) and IFTA as mediating variables. HLA mismatches, self-reported ancestry, donor-specific antibody development, and delayed graft function were the confounding factors. Log-transformed intragraft myeloid cell densities were compared between the patient groups using a nested ANOVA by fitting a mixed-effects model with patient group (matched versus B→A mismatch) as the fixed term. Two-sided statistical tests were used to test statistical significance, and a P value of < 0.05 was considered significant.

Supplementary Material

Supplementary Materials

The PDF file includes:

Figs. S1 to S12

Tables S1 to S9

Legend for data file S1

Other Supplementary Material for this manuscript includes the following:

Data file S1

MDAR Reproducibility Checklist

Data and materials availability:

All data associated with this study are present in the paper or the Supplementary Materials. Individual-level data are available at DOI: 10.5061/dryad.pvmcvdnx9.