Glomerulocapillary miRNA response to HLA-class I antibody in vitro and in vivo

Changes in miRNA expression of glomerular capillaries during antibody-mediated rejection (ABMR) are poorly understood and could contribute to the deleterious inflammation and fibrosis of ABMR via suppression of target genes. A better understanding could lead to novel diagnostic tools and reveal novel therapeutic targets. We explored deregulated miRNAs in an glomeruloendothelial in vitro model of ABMR due to class I human leukocyte antigen (HLA) with and without complement activation. We studied a set of 16 promising candidate miRNAs in microdissected glomeruli a confirmation set of 20 human transplant biopsies (DSA+) compared to 10 matched controls without evidence for ABMR. Twelve out of these 16 glomerulocapillary miRNAs could successfully be confirmed as dysregulated in vivo with 10 upregulated (let-7c-5p, miR-28-3p, miR-30d-5p, miR-99b-5p, miR-125a-5p, miR-195-5p, miR-374b-3p, miR-484, miR-501-3p, miR-520e) and 2 downregulated (miR29b-3p, miR-885-5p) in DSA+ vs. controls. A random forest analysis based on glomerular miRNAs identified 18/20 DSA+ and 8/10 controls correctly. This glomerulocapillary miRNA signature associated with HLA class I-DSA could improve our understanding of ABMR and be useful for diagnostic or therapeutic purposes.

The main cause for chronic kidney transplant loss besides recurrence of the primary disease is antibody-mediated rejection (ABMR) 1,2 . The diagnosis of acute ABMR is challenging, with current Banff guidelines resting on four cornerstones: 1) donor-specific antibody (DSA), particularly against human leukocyte antigen (HLA), 2) the vascular lesions endarteritis, thrombotic microangiopathy, glomerular and peritubular capillaritis, 3) C4d positivity of endothelial cells and 4) mRNA expression profiling 3 . In clinical practice, fulfilment of the criteria is often not possible due to the varying specificity and sensitivity of the respective test. Moreover, the pathogenesis of endothelial damage or activation is far from understood, despite recent advances in the investigation of endothelial changes upon anti-HLA-class I alloantibody binding with and without complement activation. miRNAs are regulators of mRNA levels and mRNA translation 4 . As small RNA fragments of about 20 nucleotides in length they suppress whole pathways 5 and are involved in pathologic processes like cell death 6 , inflammation 7 and fibrosis 8 .
As a first step towards an improved understanding of miRNA changes in endothelial cells secondary to HLA-class I DSA and to identify diagnostic glomerulocapillary miRNA expression signatures, we investigated an in vitro model of anti-HLA-class I ABMR with and without complement activation which was based on a standard diagnostic test recommended by the American Society For Histocompatibility and Immunogenetics 9 .
In a retrospective pilot study, candidate miRNAs were confirmed for their in vivo relevance in microdissected glomeruli from human transplants with anti-HLA-class I DSA and compared to matched controls. The target pathways of deregulated glomerular miRNAs were explored in silico.

Differential expression of miRNAs in vitro.
The cytoplasm showed a spindle to cobblestone shape.
Sequential incubation with specific anti-A2 without complement as a model for anti-HLA-class I ABMR without complement activation (in vitro C−), incubation with the respective controls irrelevant anti-A1 without complement (in vitro control without complement) and with complement (in vitro control with complement) did not show significant alterations in cell morphology. Only specific anti-A2 with complement as a model for anti-HLA-class I ABMR with complement activation (in vitro C+) caused severe cytoplasmic retraction and shrinking (Fig. 1). All expression data from the in vitro model can be downloaded as a supplementary file (TBA). Volcano plots show the differential miRNA expression in the in vitro models of complement-independent and HLA-class I DSA complement-binding (Figs 2 and 3). Based on inspection of the volcano plots and an in silico analysis of regulated pathways with DIANA miRPath v.2.0 10 we chose a set of 16 miRNAs for confirmation in microdissected glomeruli from patients with only HLA-class I DSAs: miR-let-7c, miR-28-3p, miR-29b, miR-30d, miR-99b, miR-125a-5p, miR-133a, miR-138, miR-146b, miR-195, miR-374b-3p, miR-484, miR-501-3p, miR-520e, miR-625-3p, miR-885-5p (Table 1).

Clinical and histological findings of the confirmation set of human renal transplant biopsies.
Clinical findings of patients are given in Table 2. Except for a higher proportion of living donors in the control cohort (6/10 vs. 3/20; p = 0.030) which should not be a confounder at this time after transplantation, we could not find a significant difference between the cohorts. Further splitting of the DSA+ cohort into subcohorts was avoided due to insufficient sample sizes for pairwise comparisons. Each two biopsies of the DSA+ and the control cohorts were derived from AB0-incompatible transplants, the rest from AB0-compatible transplants. All DSA+ patients had de novo DSA.
The relevant Banff components and the score for glomerular C4d for all control and DSA+ biopsies are shown in Fig. 4. We could not find a significant difference between DSA+ patients and controls for these components.

Confirmation of differentially expressed glomerular miRNAs in human transplant biopsies.
We decided to retain the each two biopsies from AB0-incompatible transplants (shown as grey symbols in Figs 3,4 and 5) in the analysis, because we could not find any obvious effect of the AB0 incompatibility on the glomerular expression of the 16 miRNAs examined.
Random forest analysis of glomerular miRNAs for the diagnosis of anti-HLA-class I DSA. In this pilot study we tried to show the potential of a combined assessment of our candidate glomerular miRNAs as markers of HLA-class I DSA. To this end we performed a random forest analysis for the discrimination of DSA+ (n = 20) vs. controls (n = 10). Of the 20 DSA+ biopsies, 18 were classified correctly using the 16 miRNA expression levels (OOB classification error rate 2/20 = 10%). Of the 10 control samples, 8 were classified correctly (OOB classification error rate 2/10 = 20%). The number of trees used in the analysis was the default value of 500; use of  None  C−  DSA+   TGFB1, TGFB2, TGFB3, HDAC4, COL4A1,  COL4A2, COL1A1, SP1, DNMT3A, MCL1,  DNMT1, VEGFA, MMP15, GRN, FGA, FGB,  FGG, COL3A1, MMP2, ADAM12, HMGA2,  CDC42, TBX21, IFNG, SERPINH1, TET2   other values instead of the default returned similar values. The top three miRNAs which contributed the most to the classification performance were miR-125a-5p, let-7c-5p, and miR-99b-5p.

Discussion
This report is the first description of glomerulocapillary miRNA expression in transplants with DSA. With our in vitro model of ABMR with HLA-class I-DSA with and without complement activation which was based on a standard immunogenetical diagnostic test, we identified miRNAs that could explain the glomerulocapillary inflammation and remodelling of ABMR and that appear promising for diagnostic use. Out of the 16 miRNAs that were subjected to confirmation on microdissected glomeruli from DSA+ biopsies, 12 were found to be differentially regulated compared to controls without DSA in our cohort. Only miR-501-3p showed a discrepancy between the in vitro and the in vivo findings. In vitro, this particular miRNA was lower both in C− and in C+, whereas in vivo it was higher in microdissected glomeruli of DSA+ compared to the controls. The reason for this discrepancy is unclear. One could speculate that this discrepancy might be due to the different time course. In vitro, DSA and complement were effective for less than two hours only. In the human transplants, the DSA would have been effective for a much longer time and a steady state regarding miRNA expression may have been reached.
Our compartment-specific approach adds to the previously discovered miRNAs in whole biopsy tissue with AMBR. With miR-146b-5p only one of the six miRNAs previously described as upregulated in ABMR was among our candidate miRNAs. However, the difference in glomerulocapillary expression between DSA+ and controls did not reach statistical significance for this particular miRNA in our study. There are at least two explanations for this discrepancy: firstly, in contrast to our DSA+ cohort, their ABMR cohort was compiled of C4d-positive cases with histological signs of ABMR, DSA data were not mentioned; secondly, our compartment-specific study examined glomeruli, whereas they analysed whole biopsy cores. With our cohort sizes larger than theirs, a type II error seems less likely but cannot be excluded.  Table 2. Clinical data of the 20 patients with HLA-class I DSA (DSA+) and the 10 controls. The only significant difference between the two cohorts was found in the proportion of living donor transplants, which were more frequent in the controls. All numerical data are given as the median and the interquartile range (IQR). Abbreviations: autosomal dominant polycystic kidney disease (ADPKD), estimated glomerular filtration rate (eGFR), focal and segmental glomerulosclerosis (FSGS), glomerluonephritis (GN), granulomatosis with polyangiitis (GPA), hemolytic-uremic syndrome (HUS), not otherwise specified (NOS). * n = 9 for the controls, because data were not available for one patient transplanted in 1992.
We could find a range of interesting targets for the differentially regulated glomerulocapillary miRNAs, some of which point to novel pathways in ABMR.
The upregulation of let-7c-5p, one of the top three performers in the random forest analysis, in DSA+ biopsies should have a negative effect. It has been described to initiate apoptosis of endothelial cells via targeting of Bcl-xl 11 . Bcl-xl has long been considered an endothelial protective factor in transplantation 12 , particularly in ABMR 13,14 . Interestingly, in vitro experiments similar to ours with a mouse monoclonal antibody against HLA-class I without complement have shown upregulation of BCL2L1 independently of PIK3/AKT1 signalling; this finding was confirmed in human heart transplants 15 . The downregulation of let-7c-5p in our in vitro experiments without complement could contribute to the induction of BCL2L1 in this setting. Complement reversed the effect and upregulated let-7c-5p. Similarly, the upregulation of let-7c-5p in transplant glomeruli should lead to detrimental downregulation of BCL2L1. Moreover, suppression of HMGA2 by let-7c-5p should be detrimental, because HMGA2 suppression has been shown to promote senescence in endothelial progenitor cells 16 . miR-28-3p was one of three miRNAs, which were lower in DSA+ than in controls. miRPath did not yield validated targets or pathways. An in vitro study in cardiomyocytes describes suppression of the PDPK1/AKT1/ MTOR signalling pathway as a function of miR-28-3p. Less suppression of this pathway would result in less activation of allogeneic regulatory T-cells by endothelial cells 16 and less suppression of proinflammatory VCAM1 in endothelial cells via mTORC2 17 . Thus, less miR-28-3p in DSA+ could have proinflammatory effects.
Similarly, the decreased expression of miR-29b-3p should have a detrimental, procoagulant, proinflammatory and profibrotic effect. miR-29b-3p has been shown to target a host of collagen remodelling factors 18,19 and profibrotic cytokines 20 as well as extracellular matrix components relevant in kidney transplant fibrosis, transplant vasculopathy and glomerulopathy, among them MMP2 21,22 , TGFB1 23 , collagen type I, type III and type IV 24 and LAMC2 25 .
miR-99b-5p and miR-125a-5p both belong to the same cluster and are known to be expressed in endothelial cells 26 . Both were found upregulated in DSA+ glomeruli and were among the top three performing miRNAs in the RF analysis. Both miRNAs are known to activate the proinflammatory 27 NFkappa-B pathway 28 . Moreover, miR-125a-5p inhibits LIN28A; LIN28A in turn inhibits hsa-let-7c-5p biogenesis [29][30][31] , which could indirectly contribute to the rise of let-7c-5p we have observed. LIN28A has a role in tissue repair 32 , which would be diminished upon upregulation of miR-125a-5p. Lastly suppression of angiogenesis promoting ERBB2 and ERBB3 33 could result in a blunted angiogenic response.
Another miRNA upregulated in DSA+ was miR-374b-3p; its target NMMLCK contributes to endothelial barrier dysfunction in inflammation 43 . miR-374b-3p also targets RECK 44 , which is important in angiogenesis. All biopsies from an AB0-incompatible transplant are shown in grey, the others in black. The p-value above the bracket between DSA+ and controls relates to a Wilcoxon test between these two cohorts; the p-value on the right relates to a non-parametric pairwise comparison (Steel-Dwass) between controls, C+, C− and C+/C−. For none of the individual components could we find a significant difference in the comparison between DSA+ and controls and in the pairwise comparisons between the DSA+ subgroups and the controls.
Suppression of RECK results in defective angiogenesis and extracellular matrix composition through the lack of inhibition of MMP2 and MMP9 45 .
miR-484 was also upregulated in DSA+, which should have detrimental effects. Targets include VEGFR2 46 , which is expressed in glomerular endothelial cells; activation of VEGFR2 decreased glomerular endothelial barrier function in vitro 47 . VEGFR2 activation also results in protective eNOS activation in glomerular endothelial cells 48 . miR-520e, which was only included in the validation set because of the interesting target complement regulator CD46 49 was upregulated in glomeruli of DSA+. CD46 inactivates C3b and C4b in the initial steps of complement activation as a cofactor of CFI 50,51 . CD46 is constitutively expressed in glomerular capillaries 52 , suppression should lead to aggravation of complement-mediated injury. miR-885-5p was one of the three downregulated miRNAs in glomeruli of DSA+. Targets include CASP3 53 , a key executor of apoptosis 54 . Interestingly, both miR-30d and miR-885-5p were increased in biopsies with glomerular C4d positivity compared to negative biopsies. Whereas miR-30d suppresses TP53 55 , miR-885 activates it 56 . The net effect of the upregulation of both miRNAs on TP53 upon complement activation is unclear, however, terminal complement activation has been reported to increase TP53 expression in vitro and in vivo 57 .
Our pilot study suggests that glomerulocapillary miRNAs could be valuable biomarkers for anti-HLA-class I DSA. Using the random forest method, we could identify 80-90% of biopsies correctly based on the glomerular miRNA expression, which clearly would not have been possible by conventional histopathology including C4d immunostainings. Random forest is a robust, automatic and cutting-edge prediction tool with minimum parameter tuning and does not overfit. It appropriately estimates the validation error rate during the analysis by the out-of-bag error rate, essentially carrying out training and validation by one step 58 . Random forest suited our low sample size, as it does not further reduce it by reserving a portion for validation. Nevertheless, before implementation as a routine diagnostic method prospective validation would still be required. Our method is complementary to conventional histology and can be used after routine histological workup, which would be advantageous compared to the already established methods which consume an entire biopsy core which is then lost for histopathology 59 . Yet, in order to improve ABMR-diagnostics our method will need refinement in prospective studies in Figure 5. Differential expression of glomerular miRNAs in human transplant biopsies with HLA class I-DSA (DSA+) vs. controls. All AB0-incompatible transplants are shown in grey, the others in black. The bar represents the median of each cohort. Glomerular miR-let-7c-5p (a), miR-28-3p (b), miR-30d-5p (d), miR-99b-5p (e), miR-125a-5p (f) and miR-195-5p (j), miR-374b-3p (k), miR-484 (l), miR-501-3p (m), miR-520e (n) and miR-885-5p (p) were higher in DSA+ than to controls. Glomerular miR-29b-3p (c) and miR-885-5p (p) were lower in DSA+ than in controls. We could not find any significant difference for the other four glomerular miRNAs examined. order to identify glomerocapillary miRNAs indicative of impeding structural and functional deterioration with ABMR, not just reflecting the mere presence of HLA-class I DSA.
In summary, we have identified 12 glomerulocapillary miRNAs dysregulated with HLA-class I DSA. Their dysregulation seems to be detrimental, in particular proinflammatory, antiangiogenic and profibrotic. The precise function of these miRNA in endothelial cells should be further investigated, preferably in flow co-culture with other cellular mediators of ABMR. Moreover, our results from biopsy glomeruli are a first step towards the development of novel diagnostic tools for ABMR which could complement traditional histology. An analogous approach could be used to investigate the clinically more important effect of HLA-class II DSA, the combined effects of both HLA-class I and II DSA and of non-HLA-DSA on endothelial miRNA in more elaborate experiments which would require stable transfection of endothelial cells with HLA-class II or stimulation with IFNG. Finally, these glomerulocapillary miRNAs could be novel therapeutic targets in ABMR. All qRT-PCR curves and respective C T values were generated with RQ manager 1.2 (Life Technologies). Thresholds were set as 0.25. Each group was analysed in a separate RQ manager file. Detectable transcripts were defined as those with a C T value below 40 and a regular sigmoid-shaped amplification curve. Amplification curves were subject to a stringent visual quality control. Baseline values were adjusted if necessary. All experiments were done in triplicates. 229 miRNAs detectable in less than 2 samples in all 4 groups were omitted from further analysis. 60 miRNAs, which were not detectable in at least one sample in all four groups were also omitted from further analysis. RNU48 and U6 snRNA served as reference transcripts. Relative glomerular expression was calculated as 2^(mean(C T reference transcripts )-C T target ). The remaining 382 miRNAs (after omission of irrelevant species on the card) were screened for differential expression between in vitro C− or in vitro C+ vs. the respective control to discover complement-independent changes in miRNA expression upon anti-HLA-class I antibody binding and between in vitro C+ with complement vs. the respective control to discover changes in miRNA expression anti-HLA-class I antibody binding with complement activation.

Selection of miRNAs for the confirmation in human transplant biopsies. We selected 15 miRNAs
for validation in microdissected glomeruli of human transplant biopsies based on visual inspection of the volcano plots (Figs 1 and 2) and on in silico data about validated targets and pathways with miRPath 60 . To these 15 miRNAs we added miR-520e, although it was detected only in one sample of in vitro C+, because the predicted targets included complement regulator CD46.

Confirmation in glomeruli of human transplant biopsies.
Based on HLA-class I and II DSA data we selected a total of 30 indication biopsies from our archive; all were from different patients. The tests for anti-HLA-class I and II antibodies were performed with the LuminexTM technology (Luminex, Austin, TX, USA) using the LABScreen TM Mixed Assay, One Lambda/Thermo Fisher, Canoga Park, CA) according to the manufacturer's instructions. The discrimination of anti-HLA specificities in LABScreenTM mixed-positive samples was achieved with LABScreen TM Single Antigen Beads (One Lambda/Thermo Fisher).All biopsies from patients with anti-HLA-class II DSA were excluded from the study. All 20 remaining biopsies with only HLA-class I DSA (DSA+) were divided into complement-binding and non-complement binding class I DSA with the C1q Screen assay (One Lambda). Four had only complement-fixating anti-HLA-class I DSA (C+), 6 had only complement-independent anti-HLA-class I DSAs (C−) and 10 had both complement-fixating and complement-independent anti-HLA-class I DSAs (C+/C−). Complement-fixating anti-HLA-class II DSAs not detectable in the standard Luminex assay were excluded in these patient by a negative lymphocyte cytotoxicity (LCT) tests.
Ten indication biopsies served as controls and were negative for anti-HLA-class I and II DSA based on negative standard Luminex tests and on negative LCT tests.
Clinical data were retrieved from the patients' files. The estimated glomerular filtration rate (eGFR) was calculated according to the modification of diet in renal disease (MDRD) formula 61 for all but one pediatric control patient, for whom we had to use the Schwartz formula 62 . Histological findings including the Banff components according to the most recent update 63 were determined on our routine set of stainings including C4d on paraffin SCIENTIFIC REPORTS | 7: 14554 | DOI:10.1038/s41598-017-14674-5 embedded tissue after epitope retrieval with citrate buffer at pH 6.0; glomerular C4d was reported in analogy to the Banff component C4d.
Glomeruli from the remaining tissue in the paraffin blocks were microdissected, glomerular miRNA was isolated and transcribed as recently described 64 and quantified in analogy to the method described above.
Statistical methods. For the comparison of our in vitro model for complement dependent and complement-independent effects of anti-HLA-class I DSA we drew volcano plots 65 . These plots show the log2 of the fold change in the mean relative expression on the x-axis and the -log10 of the p-value of a t-test comparing the replicates of the experiments on the y-axis.
All values are given as the median and the interquartile range. Continuous parameters were compared with Wilcoxon tests between two cohorts. P-values below 0.05 were considered significant in two-sided tests. However, in the retrospective analysis of biopsies, p-values can only be regarded as descriptive. In this explorative study we did not correct for multiple testing.
In order to assess the value of glomerular miRNA quantification for the discrimination between all patients with HLA-class I DSA and controls, a random forest analysis 66 was performed. Random forest is an ensemble machine learning technique which generates many decision trees (e.g. 500) using a randomly chosen subset of predictive variables (e.g. 1/3 of them), on bootstrap (or sample with replacement) versions of the original data. The individual samples not used in the bootstrap version are nevertheless used in the calculation of the so-called "out-of-bag" (OOB) error estimation. The OOB error estimated is very similar to the leave-one-out cross-validation error rate, but these are obtained without an extra cost (without a collection of extra samples, as well as without extra computational work). Random forest is an attractive method for several reasons: it is a multivariate classifier and the contributions from these variables are non-additive; this method is much less prone to overfitting than others (e.g. single decision tree, regression, boosting, etc.); it evaluates the importance of each variable concerning its contribution to the classification performance. For more information on the pros and cons of random forest and comparison with other similar classifiers/predictors, see the review by Efron and Hastie 58 . The R (www.r-project.org) package randomForest was used with the default setting.
Ethical approval. All transplants were performed according to the Declaration of Istanbul 67 . The study was performed in accordance with the Declaration of Helsinki 68 . Ethical approval was obtained from the Ethics Committees of Hannover Medical School and the University Hospital Essen with informed consent of the patients.