Circulating Donor-Specific Anti-HLA Antibodies Associate With Immune Activation Independent of Kidney Transplant Histopathological Findings

Despite the critical role of cytokines in allograft rejection, the relation of peripheral blood cytokine profiles to clinical kidney transplant rejection has not been fully elucidated. We assessed 28 cytokines through multiplex assay in 293 blood samples from kidney transplant recipients at time of graft dysfunction. Unsupervised hierarchical clustering identified a subset of patients with increased pro-inflammatory cytokine levels. This patient subset was hallmarked by a high prevalence (75%) of donor-specific anti-human leukocyte antigen antibodies (HLA-DSA) and histological rejection (70%) and had worse graft survival compared to the group with low cytokine levels (HLA-DSA in 1.7% and rejection in 33.7%). Thirty percent of patients with high pro-inflammatory cytokine levels and HLA-DSA did not have histological rejection. Exploring the cellular origin of these cytokines, we found a corresponding expression in endothelial cells, monocytes, and natural killer cells in single-cell RNASeq data from kidney transplant biopsies. Finally, we confirmed secretion of these cytokines in HLA-DSA-mediated cross talk between endothelial cells, NK cells, and monocytes. In conclusion, blood pro-inflammatory cytokines are increased in kidney transplant patients with HLA-DSA, even in the absence of histology of rejection. These observations challenge the concept that histology is the gold standard for identification of ongoing allo-immune activation after transplantation.


INTRODUCTION
Despite marked improvement in short-term graft outcome through reduced incidence of acute T-cell-mediated rejection (TCMR) with the current immunosuppressive armamentarium, long-term kidney allograft survival remains suboptimal (1,2). An important cause for this long-term graft failure is acute rejection, especially antibody-mediated rejection (ABMR) (3,4). ABMR is initiated by donor-specific antibodies (DSA), either to human leukocyte antigens (HLA) or less commonly, but also less easily detectable, to other donor-recipient mismatched antigens (5,6). Part of the explanation for the impaired graft survival associated with ABMR (7-10) is the lack of proven effective treatments to prevent or treat ABMR (4,11,12).
Therefore, we evaluated the complex landscape of 28 cytokines, chemokines, and growth factors in the blood of patients with a broad range of acute rejection types and histological lesions and analyzed the cellular origin of the most relevant cytokines, through publicly available single-cell RNASeq data and through in vitro models of HLA-DSA-mediated activation of NK cells, monocytes, and endothelial cells.

Patient Population
We included all adequate for-cause ("indication") biopsies performed within the first year after transplantation between August 07, 2012, and July 13, 2016, from consenting patients who received a single kidney transplant at the University Hospitals Leuven, Belgium. All patients gave written informed consent for collection and analysis in the Kidney Transplant Biobank, approved by the local ethical committee (S53364 and S61971). Details on the data collection are described in the Supplementary Appendix.

Histopathology
Histological lesions were semiquantitatively scored according to the Banff consensus (32), as reported previously (33). Diagnosis of the phenotypes of ABMR, TCMR, and borderline changes was established using the Banff 2019 criteria (34). More details are described in the Supplementary Appendix.

Detection of Circulating Anti-HLA Antibodies
The follow-up of anti-HLA antibodies was systematically monitored in one histocompatibility laboratory (HILA-Belgian Red Cross Flanders); details on this assessment were previously published (35), and further details are provided in the Supplementary Appendix.

Cytokine Quantification
Peripheral blood serum samples were analyzed for 28 cytokines, chemokines, and growth factors with Bio-Plex immunoassay using Luminex magnetic beads following the manufacturer's instructions (Bio-Rad, M50-0KCAF0Y; Temse, Belgium) using a 27-multiplex panel and an additional single-plex one for CXCL9. Details are provided in Supplementary Table S1 and Supplementary Appendix.

Single-Cell RNA Sequencing Data Analysis
Previously published human single-cell data from two ABMR biopsies and five healthy references corresponding to transplant surveillance biopsies were used. The associated raw counts or matrices were downloaded from the Gene Expression Omnibus (GEO, GSE145927, https://www.ncbi.nlm.nih.gov/geo) (36) and Kidney Precision Medicine Project (https://atlas.kpmp.org/ repository). Details on the gene expression analyses are described in the Supplementary Appendix. The CellChat R package (37) was used to analyze the cell-cell communication between these cells.

NK Cell and Monocyte Sorting and Coculture With Endothelial Cells
Peripheral blood mononuclear cells were isolated from the blood of healthy volunteers from the Etablissement Francais du Sang (Lyon, France) or from the Belgian Red Cross Flanders by Ficoll gradient centrifugation (Eurobio, Courtaboeuf, France). Cell sorting methods are detailed in Supplementary Appendix. Purified NK cells and non-classical monocytes were then cocultured with glomerular endothelial cells. In a first experiment, NK cells were cocultured with the human conditionally immortalized glomerular endothelial cell line (38) (ciGENC; HLA-A2), in combination with either anti-HLA-A2 DSA-containing serum or control human serum. After 4 h of coculture, supernatants were collected. In a second experiment, NK cells and non-classical monocytes were cocultured with primary glomerular endothelial cells (GENC, Cell Systems, USA), after incubation with either anti-HLA-A, -B, -C purified antibody (BD Biosciences, San Jose, CA, USA; cat #560187), or control isotype (BD Biosciences, cat #553447). After 24 h of coculture, supernatants were collected. Details on the two experiments are specified in the Supplementary Appendix.
Levels of 27 cytokines, chemokines, and growth factors in the supernatants were assessed using the same 27-multiplex analysis as used for the clinical samples, following the manufacturer's i n s t r u c t i o n s ( B i o -R a d , M 5 0 -0 K C A F 0 Y ; B i o -R a d , Nazareth, Belgium).

Statistical Analysis
We report descriptive statistics using mean and standard deviation (or median and interquartile range for skewed distributions) for continuous variables or numbers, and percentages for discrete variables. Protein levels were log10transformed for analysis. We restricted the Pearson correlation analyses to the first sample per patient, to avoid dependent measurements. Principal component analysis (PCA) and heatmap analysis were done using Euclidean distance to obtain the distance matrix and complete agglomeration method for clustering in the heatmap analysis. In logistic regression analysis, we corrected for repeat sampling by using mixed models with a random intercept. We controlled for multiple testing using the false discovery rate (FDR). Cox regression analyses and Kaplan-Meier survival curves with log-rank testing were used to assess associations with death-censored graft failure. Patients were censored at the time of death with a functioning graft or at the time of the last follow-up date. Time to failure was counted starting at the day of sample. We assessed the differences in protein levels in the in vitro experiments using paired T-tests on the log10-transformed protein levels. We used SAS (version 9.4; SAS Institute, Cary, NC), R (version 3.6.3, R core team, 2014, packages corrplot, FactoMineR, heatmap3), and GraphPad Prism (version 8; GraphPad Software, San Diego, CA) for statistical analysis and data presentation. Figures were created with BioRender.com.

Blood Cytokine Profiles Associate With HLA-DSA
In the first samples of each patient (N = 192; Figure 1A, Table 1 and Supplementary Table S2), strong positive correlations were demonstrated among cytokines CXCL10/IP10, PDGF-BB, IL-6, GM-CSF, IL-9, CCL5/RANTES, basic-FGF, CCL4/MIP-1b, TNF-a, IFN-g, CCL2/MCP-1, and IL-17a, followed by IL-15, IL-1b, and IL-2, and negative correlations with IL-1Ra, IL-7, and G-CSF ( Figure 1B). PCA based on the 28 blood proteins in first samples revealed two distinct clusters ( Figure 1C). The proinflammatory cytokines IFN-g, TNF-a, IL-6, IL-9, IL-17a, CXCL10/IP-10, CCL4/MIP1b, CCL5/RANTES, basic-FGF, and GM-CSF ( Figures 1D, E) were the strongest contributors to differentiate a cluster of 20/192 (10.4%) samples, which was hallmarked by positivity for HLA-DSA (15/20; 75%), whereas the majority of the other cluster were HLA-DSA negative (169/ 172; 98.3%) ( Figure 1C). Unsupervised hierarchical clustering confirmed this differentiation in the levels of pro-inflammatory cytokines, corresponding to the HLA-DSA status, and the histological lesions scores (cluster I) ( Figure 1F). In 14/15 HLA-DSA+ patients in cluster I, HLA-DSA were present at time of the biopsy (12 persistent pretransplant, 1 de novo, and 1 both pretransplant and de novo HLA-DSA), whereas in 1 patient the HLA-DSA were present pretransplant but had resolved at the first biopsy. In the other five patients (25%) of cluster I, no HLA-DSA was detected at the time of transplantation or at any time post-transplantation. No differences in specificity (class) or strength (MFI) of the HLA-DSA were noted between the HLA-DSA+ samples in cluster I (N = 15 HLA-DSA) vs. cluster II (N = 3 HLA-DSA) (Supplementary Table S3). When comparing patients in cluster I vs. II, differences were noted in HLA-DSA, repeat transplantation, induction therapy, and recipient sex, all reflecting the background risk and management of HLA-DSA, but no differences in C-reactive protein or polyomavirus or cytomegalovirus viremia ( Table 2). When repeating the analyses in all samples (N = 293), including the repeat sample-biopsy pairs, a similar cluster of HLA-DSA+ samples resulted from the PCA and heatmap analyses (Supplementary Figure S1). Follow-up samples from patients whose first biopsy was clustered in cluster I clustered together in the pro-inflammatory cluster.
As the clustering approach indicated a strong association between the global protein levels and HLA-DSA, we next evaluated which proteins individually associated with HLA-DSA using logistic mixed models. All proteins except IL-4, IL-5, IL-10, IL-12p70, IL-13, CCL11/Eotaxin, and VEGF significantly associated with HLA-DSA ( Figure 2A). A high number of samples had levels below the detection limit in both clusters for IL-5, IL-10, IL-12p70, IL-13, and VEGF (Supplementary Table S1 and Supplementary Figure S2).
When considering all biopsies, significant associations were found between cytokines and diagnosis of ABMR (N = 26) vs. no ABMR (N = 267) ( Figure 2C and Supplementary Figure S3). However, after adjustment for HLA-DSA, none of the cytokines remained significantly associated with ABMR ( Figure 2D). Conversely, the associations between HLA-DSA and the individual proteins remained largely unaltered after adjustment for ABMR ( Figure 2B). This illustrates that HLA-DSA status is a stronger determinant of the cytokine profiles than ABMR histology. Compared to ABMR, fewer cytokines associated with TCMR (N = 71) vs. no TCMR (N = 222), and none associated with borderline changes vs. no rejection (Supplementary Figure S3 Figure S3). These findings were corroborated by the associations between protein levels and inflammatory histological lesions. Basic-FGF, IL-1b, IL-9, IL-2, IL-17a, CCL4/MIP1b, CCL5/ RANTES, CXCL9/MIG, IFN-g, and TNF-a positively and IL-7 negatively associated with microvascular inflammation, the hallmark of ABMR (Supplementary Figure S4). Proteins with a positive association with tubulo-interstitial inflammation, the hallmark of TCMR, were CXCL9/MIG and CXCL10/IP10 (Supplementary Figure S4). When considering intimal arteritis, which can occur both in ABMR and in TCMR, upregulated proteins were IL-6, CXCL8/IL-8, CXCL9/MIG, CXCL10/IP10, IFN-g, and TNF-a; downregulated proteins were IL-7, IL-13, and G-CSF (Supplementary Figure S4).  Based on the observed association between cytokines and HLA-DSA, we hypothesized that these cytokines originated from the transplanted kidneys as the allograft is the site where the DSAmediated injury occurs. To determine the cellular origin of the cytokine profiles in the allograft, we used publicly available singlecell data from kidney biopsies. Briefly, scRNASeq was performed on two renal allograft biopsies with DSA-mediated injury, namely, ABMR (one with class I and II anti-HLA-DSA (B8, DQA1) and one with class II HLA-DSA [DR1; DR53 and DQ2 (36)], and five reference biopsies. After quality control and filtering (Supplementary Figure S5), 33,216 cells were detected and clustering revealed 14 clusters corresponding to the main renal cell subtypes but also infiltrating immune cells ( Figures 3A, B).

Cell-Cell Communication
As cytokines and chemokines are the mediators of cell-cell communication, we used the CellChat package to infer the intercellular communication from the scRNASeq data. Twenty signaling pathways showed significant communication between the allograft cells. Among these were the CXCL, CCL, CX3C, PDGF, VEGF, and CSF signaling pathways ( Figure 4A). ECa were important senders but also receivers of CXCL and CCL signaling ( Figure 4B). Lymphocytes were important senders of CCL signaling, whereas APCs were receiving CX3C and CSF signals from ECa in these biopsies. When looking at all signals received by ECa, sent by ECa, APCs, and lymphocytes, the atypical chemokine receptor (ACKR1), also known as DARC, was the identified receptor for all ligand-receptor pairs ( Figure 4C).

Cytokine Secretion in Antibody-Mediated NK Cell and Monocyte Activation
Given that the most clinically relevant cytokines corresponded to expressed genes on activated glomerular endothelial cells, nonclassical monocytes, and NK cells, we next assessed the cytokine production by these cells. Indeed, endothelial injury is the hallmark of DSA-mediated injury in kidney allografts and both NK cells and monocytes are key players in ABMR (40)(41)(42)(43)(44)(45). Both NK cells and non-classical monocytes express the Fcg receptor, through which antibody-dependent cellular cytotoxicity can be induced against a target cell, i.e., the endothelial cell in the setting of ABMR. We tested the specificity of the studied proteins for these cell types in an in vitro model of antibody-mediated NK cell and/or monocyte activation in interaction with glomerular endothelial cells.  control ( Figure 5B and Supplementary Figure S7). After 24 h of coculture of NK cells, non-classical monocytes and GENC, IFNg, IL-5, CXCL8/IL-8, IL-9, IL-15, CXCL10/IP10, CCL2/MCP-1, CCL4/MIP1b, and GM-CSF were significantly higher in the NK cell and monocyte activation condition, compared to control, suggesting a synergistic effect between these two cell types ( Figure 5B and Supplementary Figure S7). When comparing in vitro models of GENC alone, with an HLA class I-specific antibody vs. non-specific isotype antibody, no significant cytokine production was noted (Supplementary Figure S8). Figure S9).

DISCUSSION
In this representative cohort of consecutive samples from kidney transplant recipients at time of allograft dysfunction, we demonstrate that pro-inflammatory cytokines in the peripheral blood associate with presence of HLA-DSA and impaired kidney transplant survival. These pro-inflammatory blood signals associate with HLA-DSA, rather than with rejection phenotypes as identified in kidney transplant biopsies. This illustrates that HLA-DSA can lead to immune activation that is not always reflected in histology of the concomitant kidney transplant biopsies. Single-cell data analysis demonstrated that the most contributing cytokines to the clustering of the clinical samples corresponded to genes expressed on activated glomerular endothelial cells, non-classical monocytes, and NK cells. Subsequent in vitro models confirmed that the cytokines observed in patients' serum can be induced by HLA-DSA-mediated cross talk between glomerular endothelial cells, NK cells, and monocytes. To our knowledge, this is the first study demonstrating the strong association of pro-inflammatory blood cytokine profiles with presence of HLA-DSA. One previous study found a panel of 6 pro-inflammatory cytokines (TNF-a, IFN-g, IL-1b, IL-6, IL-8/ CXCL8, and IL-17) to be associated with angiotensin II type 1 receptor antibodies, a non-HLA antibody, but not with presence of HLA-DSA (13). Most of the cytokines that associated with HLA-DSA in our study are known as potent mediators of the alloimmune responses and associated injury. The function of cytokines and chemokines is characterized by pleiotropism, redundancy, synergism, and antagonism (46), clearly illustrated by the strong correlations we observed between the proteins. The signals that stand out primarily reflect IFN-g-related inflammation including CXCR3 and CCR5 signaling. receptors in rejection has been demonstrated in literature, as well as their potential as therapeutic targets (16,17,27,(47)(48)(49)(50)(51)(52). The presence of both CXCR3 and CCR5 appears to identify subsets of NK cells and T cells in blood with a strong predilection for homing to inflammatory sites (53,54), and CXCL9/MIG and CXCL10/IP10 have been well characterized in kidney allograft rejection (20,55).
Next, the scRNASeq analysis on kidney transplant biopsies suggested that, although the different cytokines are highly collinear, they do not originate from the same cells. CXCL9/ MIG is mainly expressed by activated endothelial cells, whereas CXCL10/IP10 is expressed by both activated endothelial cells and CD16+ monocytes. These results are in line with reports showing that CD16+ non-classical monocytes present an increased capacity for antibodydependent cellular cytotoxicity (56). Other relevant identified cytokines are the pro-inflammatory cytokines IL-6, IL-8, IL-9, and IL-17a.
As the scRNASeq analysis suggested expression of the relevant pro-inflammatory cytokine profiles in kidney endothelial cells, monocytes, and NK cells, we next performed in vitro studies to evaluate whether these cytokines were indeed supernatants (58,59), supernatants of glomerular endothelial cells alone in the presence of a specific anti-HLA antibody vs. isotype antibody were also analyzed. No significant differences in cytokine secretion were noted between the two conditions. Together with the single-cell data, this suggests that endothelial cells express cytokines, but production is only observed when interacting with immune cells. For this interaction with immune cells through chemokines, the ACKR1 receptor, or DARC, was identified as a common receptor on activated endothelial cells for multiple incoming chemokine signals of the CXCL and CCL families. This receptor is known to regulate the dynamics of inflammatory chemokine presentation on vascular endothelial cells and chemokine concentrations in plasma (60). The importance of this receptor in ABMR has been reported previously (28,61,62).
Although older studies did demonstrate that HLA class I antibodies can directly trigger cytokine secretion from endothelial cells (15,63), our data are in accordance with the study of Wei et al., where HLA class I antibodies did not induce cytokine production directly from endothelial cells derived from aortic rings, but cell-cell contact of monocytes with antibodyactivated endothelial cells was required for cytokine production (57). The observation that this is also true for glomerular endothelial cells used in our experiments, which are more relevant for kidney microvascular injury, is novel. The discrepancy with previous studies (15,63) could be explained by differences in vascular origin of the endothelial cells or in the type/concentration of the HLA antibodies.
Our present work challenges the vision that kidney transplant histology is the gold standard for identification of ongoing alloimmune processes. The immune activation detectable in the peripheral blood of patients with HLA-DSA could be an explanation for the bad outcome observed in patients with HLA-DSA, even in the absence of histological lesions in the biopsy (35). This could be related to the observer dependency of kidney transplant biopsy reading and the risk for sampling error, in addition to the potential that classic light microscopy is not sensitive enough for detecting kidney transplant inflammation. As illustrated previously (64), increased intrarenal rejectionassociated gene transcripts can be observed in biopsy samples of DSA-positive/ABMR-negative patients, potentially proceeding to later overt histological injury (65). Given that the pathological evaluation of biopsies is observer-dependent and subject to sampling error, inflammatory lesions could be missed or remain below the thresholds of the Banff classification, yet be reflected by inflammatory changes detectable in the graft/ circulation. Markers for inflammatory changes in the graft below the threshold of the Banff classification have been suggested before, by studying intragraft transcriptional changes ("molecular microscope") (66) or donor-derived cell-free DNA (67) as markers for severity of graft injury to complement the histological assessment using Banff classification. Our study illustrates that also blood inflammatory cytokine profiles could help in objectifying deleterious inflammation in patients at high risk of rejection but with negative histology. This is further supported by our finding that patients with HLA-DSA and enhanced pro-inflammatory signals had worse outcome than HLA-DSA-positive patients without pro-inflammatory signals in the blood. All of this seems to be in contrast to the study of Parajuli et al., who did not observe different outcomes in patients with and without HLA-DSA if the index biopsy was negative for histological inflammation (68). This previous study was however focused on first biopsies in the absence of rejection, the study design is not comparable to our strategy with a relatively short follow-up time, and the interpretation of these data could be hampered by inherent selection bias.
There are some limitations to this study. The single-center nature of our study population, consisting largely of white Europeans with low to intermediate immunologic risk, treated mainly with a tacrolimus-based regimen, limits the generalizability of these findings to other and higher-risk populations and warrants external validation. Second, this study consisted of biopsies performed at time of graft dysfunction with possible confounders and inherently excluding cases with subclinical rejection. All samples were taken in the first year after transplantation, and the majority in the first 3 months after transplantation, reflecting the real-life timing of the majority of indication biopsies but obviating generalization of the results to biopsies later after transplantation and chronic rejection phenotypes. Induction therapy was different between the two clusters, with more induction therapy given to sensitized patients, and could have confounded cytokine levels early after transplantation. However, we would expect the induction therapy to cause a lesser immune activation, whereas here the opposite was observed. T cell sensitization was not measured, and its effect could therefore not be explored in this study. Also, we used a preselected multiplex assay of cytokines, chemokines, and growth factors, but other potentially important proteins were not studied. Downstream analyses were confined to the main clinical association, namely, DSA-mediated injury. This implies that other rejection types that could also contribute to cytokine expression and secretion were not studied in the downstream analyses. Nevertheless, we did not find indications for their contribution in the clinical associations. With regard to the in vitro analysis, we chose to focus on NK cells and non-classical monocytes, while other cell types capable of antibody recognition like CD16+ CD8 T cells (69), NKT cells, gd T cells, neutrophils, and macrophages could also play a role in the chemokine profiles associated with DSA. Moreover, we did not study the effect of complement, which can cause endothelial injury from antibodies independent of immune cells (70). We also did not study the effect of missing self, which could enhance NK cell activation in addition to DSA (43,71). Finally, the in vitro experiments in this study are not adequate to define the exact cellular source of the secreted cytokines.
This study did not intend to explore the biomarker potential of these cytokines. Although biomarker potential is not excluded, different study design and analyses would be needed to assess this. Furthermore, one could question the value of peripheral blood cytokines as clinical biomarkers as we can easily measure the gold standard, which are the circulating HLA-DSA.
In conclusion, this study demonstrates that blood proinflammatory cytokines are increased in kidney transplant patients with HLA-DSA and associate with worse graft survival, even in the absence of histology of rejection. These observations illustrate that HLA-DSA can lead to immune activation that is not always reflected in the histology of the concomitant kidney biopsy, challenging the concept that histology is the gold standard for identification of ongoing allo-immune activation after transplantation.

DATA AVAILABILITY STATEMENT
All of the individual de-identified participant data that underlie the results reported in this article (text, tables, figures, and appendices) can be made available on a collaborative basis following institutional review board approval. Proposals should be directed at the corresponding author.

ETHICS STATEMENT
The studies involving human participants were reviewed and approved by the local ethical committee (S53364 and S61971). The patients/participants provided their written informed consent to participate in this study.