2.1 Study population

Computerized biopsy registries from 2 French hospitals were screened to identify renal allograft biopsies with a histological diagnosis of pure ABMR or pure TCMR according to the Banff 2017 criteria.9 Biopsies were either protocol (at 3 and 12 months) or for-cause biopsies.

For ABMR, biopsies with at least moderate microvascular inflammation ([g + ptc] ≥ 2) and presence of circulating donor-specific antibodies (DSA) at the time of biopsy and/or positivity for C4d were included. Four cases of suspected acute/active ABMR, defined by at least moderate microvascular inflammation ([g + ptc] ≥ 2), but without circulating DSA at the time of biopsy nor positivity for C4d, were also included. Biopsies with an interstitial inflammation involving > 10% of the total cortex (ti > 0) or an interstitial inflammation involving > 10% of nonsclerotic cortical parenchyma (i > 0) were excluded.

For TCMR, biopsies fulfilling the criteria for acute TCMR – interstitial inflammation involving > 25% of nonsclerotic cortical parenchyma (i2 or i3) with moderate (t2) or severe (t3) tubulitis) or chronic active TCMR (interstitial inflammation involving > 25% of the total cortex (ti2 or ti3) and > 25% of the sclerotic cortical parenchyma (i-IFTA score 2 or 3) with moderate (t2) or severe (t3) tubulitis – were included. Cases with glomerulitis (g ≥ 0) were excluded.

Control biopsies (NR) were randomly selected (preimplantation, protocol or indication biopsies graded as category 1 according to the Banff classification).

All participating patients provided written informed consent.

2.2 Conventional immunohistochemistry (IHC)

Paraffin-embedded tissue sections were manually immunostained against NKp46 (clone 8E5B,17 dilution 1:4000, Innate Pharma, Marseille, France) using the ABC peroxidase procedure as previously described.18 Sections from normal human spleen and tonsils were used as positive controls. Labeled NK cells were manually counted at high power field ×40.

2.3 Multiplex immunofluorescence (IF) staining

We set up a multiplex immunofluorescence staining (Figure 1A,B) using Opal reagents (PerkinElmer, Waltham, MA). Opal is a method for multiplex immunofluorescence in paraffin-embedded tissue. It is a tyramide-based method involving deposition of fluorophore-conjugated tyramide at the site of the antigen of interest, effectively amplifying the signal. The permanent nature of the tyramide-antigen bond, accomplished through its covalent binding with tyrosine residues on the antigen of interest allows for heat-mediated removal of antibodies, but preservation of the fluorescence signal associated with the antigen of interest. In this manner, sequential applications of targeted antibodies can be utilized with the flexibility of using multiple antibodies raised in the same species, without concern for the cross-reactivity.

Briefly, 3-4 µm thick paraffin-embedded tissue sections were deparaffinized in xylene, rehydrated through an ethanol gradient and fixed in 10% neutral-buffered formalin for 20 minutes. Antigen retrieval was performed using microwave treatment (MWT) for 15 minutes in antigen retrieval solution pH6 or pH9 (AR6 or AR9) according to the target being detected. At each of the four cycles of staining, primary antibodies anti-NKp46 (clone 8E5B,17 Innate Pharma), anti-CD163 (clone 10D6, Leica Biosystems), anti-CD34 (clone QBEnd-10, Dako), and anti-CD3 (clone SP7, Thermo scientific) were incubated for 30 minutes or 1 hour at room temperature or +37°C, followed by Opal Polymer HRP Ms + Rb Kit (PerkinElmer) for 10 minutes at room temperature. Next, tissue sections were incubated with TSA opal fluorophores (Opal 620, 690, 520 and 540) for 10 minutes (Table 1). MWT was performed at each cycle of staining to remove antibody-TSA complex with AR6 or AR9. Finally, all slides were counterstained with DAPI for 5 minutes. Sections from normal human spleen and tonsils were used as positive controls.

The detailed protocol is given in Figure S1.

2.4 Images collection

Whole slides (cortex + medulla) were scanned using the PerkinElmer Vectra (v3.0; PerkinElmer) at low magnification (×10). Under pathologist supervision, the renal cortex was sampled from individual consecutive fields or regions of interest (ROI) measuring 669 µm × 500 µm (0.3345 mm² each) by using the Phenochart 1.0.4 viewer (PerkinElmer) to scan at high resolution (×20) (Figure 1A). The medulla was not scanned at high resolution. For ABMR and TCMR biopsies, the whole cortex was scanned except for artifact regions (ABMR 72.4 ± 3.9% of the cortex, TCMR 77.2 ± 4.0% of the cortex, P = .37). For NR biopsies, the renal cortex was randomly sampled to analyze at least 50% of the entire cortex (52.4 ± 7.6%). The total surface of renal cortex was used to calculate the cellular density (ie, number of immune cells/mm²).

2.5 Fluorescence analysis and automated cell count

The 45 renal biopsies were analyzed using the software inForm Tissue Finder 2.3.0 (PerkinElmer). Only scanned images at ×20 (ROI fields) were analyzed by inForm.19 This software is based on a supervised machine-learning algorithm. The full procedure is detailed in Figures S2 and S3.

Briefly, the first step consists of creating a “spectral library” as recommended for multiplex analysis: single-stained (CD34, NKP46, CD3, CD163) and unstained slides were analyzed in inForm to integrate the corresponding spectra in a spectral library. The spectral library eliminates fluorophore crosstalk and interference from tissue autofluorescence, allowing precise measurement of each fluorescence signal within a tissue sample.

The second step is the creation of a training set called a “project” containing representative images of the total ROI scanned. This “project” is ultimately used at the end of the study for the “batch analysis,” that is, the final analysis of all the images of the cohort. The representative images are opened using the spectral library then a three-step machine-learning algorithm is performed: tissue segmentation, cell segmentation, and cell phenotyping. The “tissue segmentation” consists to train the software to recognize intravascular and extravascular areas based on the CD34 labeling. The “cell segmentation” consists to train the software to identify every individual cell, based on the recognition of the nuclei using a counterstain-based approach (DAPI). Lastly, the “cell phenotyping” consists in the recognition of different cell subtypes based on CD3, CD163, and NKp46 labeling. Basically, the pathologist identifies at least 5 cells in each phenotype in order to proceed. We chose 376 CD3⁺ cells, 339 NKp46⁺ cells, and 380 CD163⁺ cells. We also chose 412 “other cells” (ie, CD3⁻ NKp46⁻ CD163⁻).

At the end of the training of the project, a batch analysis is performed after opening all the images of the entire cohort, and we obtain an absolute cell count of each individual cell subtype and their localization.

To further distinguish between glomerular and peritubular capillaries, we manually counted and phenotyped the cells within the glomeruli on every image of the study.

2.6 RNA extraction and real-time quantitative reverse transcription PCR

Total RNA was isolated from all biopsies with frozen tissue sample available, using the miRNeasy Mini Kit (Qiagen, Hilden, Germany) and used to make cDNA using a mix containing RNAse inhibitor, dNTP mix, Random Hexamer, MgCl₂ Solution, and MultiScribe Reverse Transcriptase (all from Thermo Fisher, Waltham, MA). PCR reactions were assembled with TaqMan 2x Fast Univ. PCR Master Mix (Thermo Fisher) and predeveloped TaqMan gene expression assays or “home-designed” primers and probes purchased from Thermo Fisher (Table S1). mRNA levels of NKp46 and KLRF1 (NK cells), CD3 and CD25 (T cells), and CD14, CD16, CD68, and CD163 (monocytes/macrophages) were detected by real-time qPCR on a ViiA7 Real-Time system using QuantStudio Real-Time PCR software (Thermo Fisher). Gene expression levels were normalized to the mean of GAPDH and HPRT1 mRNAs, and 18S rRNA. The ΔΔCt method was used to calculate relative expression using human spleen total RNA as reference (AM7970, Thermo Fisher).

2.7 Statistical analysis

Statistical analyses were performed using RStudio, version 1.1.456 for Mac OS X. We compared means between 2 groups using the Mann-Whitney U test, and means among three groups using the Kruskal-Wallis test followed by the Dunn's post-hoc test with Bonferroni adjustment. We compared proportions using the Fisher exact test. Correlation coefficients were determined using Spearman's Rank-Order Correlation. Unless noted otherwise, results are expressed as mean ± SEM.

3.1 Description of the study population

The study population consisted of 45 renal allograft biopsies in 44 patients (demographic data are detailed in Tables 2 and 3). The biopsies were performed at a median time of 5.0 months posttransplantation (interquartile range: 1.0-28.0 months). The mean estimated glomerular filtration rate (eGFR) was 41.3 ± 3.6 mL/min/1.73 m² at the time of rejection diagnosis. Primary pathological diagnoses were 20 ABMR, 20 TCMR, and 5 NR. The mean number of glomeruli and the mean surface of biopsy were 11.3 ± 0.9 and 4.7 ± 0.3 mm², respectively, with no significant difference between the three groups (Kruskal-Wallis, P = .90 and .81). The mean follow-up duration for the cohort was 53.6 ± 7.8 months. During the follow-up, one patient died and five patients lost their graft. All graft losses were attributed to rejection.

3.2 Multiplex IF is a reliable technology to study the alloimmune response in kidney transplantation

To characterize the allograft inflammation during rejection, we developed an in situ mIF technique with automated counting (Figure 1). NKp46, CD3, and CD163 positive cells were quantified in the renal cortex only, using consecutive individual fields scanned at ×20.

To validate this new strategy, we also realized an anti-NKp46 labeling using conventional IHC on a serial tissue section (Figure 2A) and an RT-qPCR from a frozen sample obtained at the same time to compare cellular densities (ie, number of cells/mm²) (Figure 2B) and gene expression levels (Figure 2C). Among the 45 included biopsies, a frozen sample was available for RT-qPCR for 24 biopsies (11 ABMR, 10 TCMR, and 3 NR).

We first correlated the NKp46⁺ cell density manually measured by IHC to the NKp46⁺ cell density automatically measured by mIF (Figure 2B) and found a strong positive correlation between the 2 techniques (r = .91, P < .001). Then we correlated the densities of NKp46⁺, CD3⁺, and CD163⁺ cells measured by mIF to the relative mRNA expression levels of NKp46 and KLRF1 (for NK cells), CD3 and CD25 (for T cells), and CD14, CD16, CD68, and CD163 (for monocytes/macrophages) measured by RT-qPCR (Figure 2C). For all the tested genes, we found a positive correlation between the two techniques (r between .47 and .61, P value between .001 and .021).

3.3 CD163, CD3, and NKp46 positive cells are recruited during kidney allograft rejection

Using mIF, we first analyzed whether leukocyte densities were associated with rejection (Figure 3B). Figure 3A illustrates 2 images obtained during ABMR and TCMR, with the visualization of extravascular and intravascular inflammatory cells based on the CD34 staining. We found that the total immune cell density was low in NR biopsies (45.8 ± 19.6/mm²) and significantly increased during ABMR (435.1 ± 70.1/mm²) and TCMR (1303.2 ± 254.3/mm²), with significantly more immune cells in TCMR than ABMR (P = .003) (Figure 3B). NKp46, CD3, and CD163⁺ cells were significantly more abundant in grafts with either ABMR or TCMR compared to NR biopsies and were higher in TCMR compared to ABMR. In particular, the NKp46⁺ cell density was higher in TCMR (31.7 ± 7.8/mm²) compared to ABMR (9.6 ± 2.2/mm², P = .01). In term of relative proportions, CD3 and CD163⁺ cells were the 2 predominant cell types involved in both ABMR and TCMR (51.8 ± 6.0% and 45.3 ± 5.8% in TCMR respectively; 48.0 ± 4.4% and 49.3 ± 4.4% in ABMR respectively) (Figure 3C). Conversely, NKp46⁺ cells represented a rare population (NR: 0.6 ± 0.4%; ABMR: 2.7 ± 0.7%; TCMR: 2.9 ± 0.6%), with no significant difference between ABMR and TCMR (P = .58).

3.4 The composition of the inflammatory infiltrate is different between the intravascular and extravascular compartments

After a period of machine learning by a pathologist, the image analysis software was able to distinguish the intravascular (ie, glomerular and peritubular capillaries) and extravascular compartments, according to the CD34 labeling as endothelial cell marker (Figure 1C). Using this marker, the localization (intra- or extravascular) of each cell was determined.

CD3⁺ and CD163⁺ cells were the 2 predominant cell types in the extravascular compartment (Figure 4A), for both TCMR (CD3: 708.1 ± 207.2/mm²; CD163: 456.2 ± 89.3/mm²) and ABMR (CD3: 108.2 ± 26.8/mm²; CD163: 205.5 ± 45.8/mm²). On the other hand, CD3⁺ cells were the predominant cell type in the intravascular compartment (Figure 4C) for both TCMR (CD3: 89.4 ± 2.6%; CD163: 7.8 ± 2.5%; NKp46: 2.8 ± 0.8%) and ABMR (CD3: 79.3 ± 3.8%; CD163: 15.4 ± 3.6%; NKp46: 5.3 ± 1.2%).

Interestingly, NK cells and macrophages densities were significantly higher in the intravascular compartment during ABMR compared to TCMR (5.0 ± 1.2/mm² vs 1.5 ± 0.5/mm² for NK cells (P = .006) and 14.5 ± 3.1/mm² vs 5.0 ± 1.5/mm² for macrophages (P = .013)).

As expected, we observed much more leukocytes in the extravascular compartment than in the intravascular compartment during TCMR (1194.5 ± 234.7/mm² vs 108.7 ± 28.0/mm², P < .001) (Figure 4B). But surprisingly, we made the same observation in ABMR (318.3 ± 62.9/mm² vs 116.8 ± 18.4/mm², P = .012), even if the difference was less pronounced, and despite the fact that we included only ABMR cases with less than 10% of interstitial inflammation in the nonatrophic cortex (i-score = 0) or in the total cortex (ti = 0). In ABMR, a careful analysis of all the images of the cohort showed that even if we observed numerous immune cells in the intravascular compartment (ie, glomerulitis and peritubular capillaritis), there were also many leukocytes in the extravascular compartment, in particular around the vessels.

3.5 Differential composition of the microcirculation inflammation in peritubular and glomerular capillaries

Distinction between microcirculation compartments revealed a significantly higher total leucocyte density within the glomerular compartment during ABMR compared to TCMR (18.8 ± 5.6 vs 2.2 ± 0.4, P < .001) (Figure 5A). We did the same observation for each cell type, with higher NKp46⁺, CD3⁺, and CD163⁺ glomerular cell densities during ABMR. Furthermore, we did not observe a significant difference between TCMR and NR biopsies regarding the total leucocyte density in the glomeruli (2.2 ± 0.4 vs 0.3 ± 0.2, P = .06).

On the other hand, we did not find any significant difference concerning the total leukocyte density within the peritubular capillary compartment between ABMR and TCMR (107.6 ± 18.4 vs 108.2 ± 27.9, P = .36) (Figure 5A). However, we found more NKp46⁺ cells and CD163⁺ cells in the peritubular capillaries during ABMR than TCMR (4.7 ± 1.2 vs 1.5 ± 0.5, P = .01 and 11.6 ± 2.5 vs 5.0 ± 1.5, P = .02, respectively).

CD3⁺ cells were the predominant cell type in the peritubular capillaries and glomeruli (Figure 5B), for both TCMR (peritubular capillaries: CD3 87.6 ± 3.3%; CD163 10.5 ± 3.3%; NKp46 2.0 ± 0.6%; glomeruli: CD3 64.6 ± 7.9%; CD163 28.4 ± 6.8%; NKp46 7.1 ± 3.4%) and ABMR (peritubular capillaries: CD3 81.1 ± 3.5%; CD163 14.0 ± 3.3%; NKp46 4.8 ± 0.9%; glomeruli: CD3 59.8 ± 5.9%; CD163 34.2 ± 6.0%; NKp46 6.1 ± 1.9%).

We correlated the total immune cell densities with the respective Banff inflammatory components (ie, correlation between glomerular density and glomerulitis score, peritubular capillary density and peritubular capillaritis score, and extravascular density and inflammation score). The results showed a good correlation between glomerular density and glomerulitis, as well as between the extravascular density and the “i” score. However, the correlation was not as good for peritubular capillaritis, due to the presence of many inflammatory cells within the peritubular capillaries during TCMR, in the inflammation clusters (Figure S4).

3.6 Important heterogeneity in the global composition of the inflammatory burden during renal allograft rejection

Interestingly, we observed an important heterogeneity in the proportion of CD3⁺ cells and CD163⁺ cells whatever the type of rejection (Figure 6A). Variations between the highest and the lowest fraction of CD3⁺ cells ranged from 10.0% to 92.7% for TCMR and from 9.9% to 81.0% for ABMR, whereas for CD163⁺ cells, the variations ranged from 7.0% to 89.0% for TCMR and from 18.8% to 87.7% for ABMR (Figure 6A). These extreme situations are illustrated in Figure 6B.

We did not observe any difference between ABMR DSA- (n = 16) and ABMR DSA⁺ (n = 4) cases (Figure 6A). Furthermore, a patient with TCMR rejection had 2 biopsies 1 month apart (biopsies #10 and 11) and we did not observe a significant difference in the global immune composition between the 2 biopsies (Figure 6A).

We compared clinical, biological, and histological data between rejection biopsies with high and low proportion of CD163, CD3, and NKp46 cells, respectively, according to the median percentage of each cell type (44.4% for CD163, 53.0% for CD3, and 2.1% for NKp46 positive cells) (Table S2). Despite a trend toward significance regarding the time of biopsy in T cell-enriched cases compared to low T cell rejection cases (40.5 ± 11.6 vs 20.6 ± 11.0 months, P = .071), we could not find any significance differences between groups.