Kidney transplantation offers the greatest survival benefit for patients with end-stage kidney failure.1 Long-term outcomes have not significantly improved in recent decades, and rejection remains the major cause of transplant failure after the first year.2–4 Rejection in the kidney transplant is defined by the presence of immune cells within the kidney and is broadly subcategorized into 2 types based on where in the kidney immune cells are located.5 Understanding how immune cells interact with other immune cells and with allograft cells must be of critical importance in rejection.
Bulk transcriptomic studies of biopsies have sought to gain understanding of the molecular mechanisms driving kidney transplant rejection. Some have found natural killer (NK) cell and endothelial cell transcript levels associate with antibody mediated rejection (AMR), suggesting an important relationship between these cell types.6,7 However, these studies lack cellular resolution and are unable to accurately attribute gene expression to cell type. Single-cell RNA-sequencing (scRNA-seq) studies add cellular resolution.8–13 Cell-cell contact is critical for the coordination of the alloimmune response during rejection. Many computational tools assume cell–cell physical interaction based simply on cognate gene (ligand and receptor) expression averaged across 2 cell populations in a scRNA-seq experiment.14–17 These approaches infer physical interaction between cells without information regarding cell–cell distance and are unable to differentiate the transcriptional changes of interacting cells from noninteracting cells.
A percentage of droplets generated in a microfluidics-based scRNA-seq experiment will contain 2 cells due to the chance encapsulation of 2 cells at the time of droplet formation.18 The multiplet frequency can be estimated following Poisson statistics.19,20 These resulting doublets are subsequently removed from the final analysis by computational methods, as they are all assumed to represent artifact. However, physically interacting cells in source tissue, such as a kidney biopsy, may remain in contact after careful tissue digestion in a single-cell experiment.21 These reflect biologically interacting cells and coencapsulation of such doublets, or physically interacting cell complexes (PICs), offers an opportunity to study cell–cell interaction at the transcriptional level.
Distinguishing PICs from artifactual doublets can be accomplished by comparing the estimated frequency of doublets to the observed frequency. PICs will appear at a higher-than-predicted frequency. Alternatively, surface protein-based cell isolation prior to encapsulation can also be used to separate PICs from artifact doublets. For example, human renal microvascular endothelial cells do not express CD45 surface protein.22,23 Therefore, physical isolation of CD45pos immune cells may result in coencapsulation of a CD45neg endothelial cell if they were physically interacting at the time of biopsy preparation. We took advantage of this phenomenon to identify heterotypic PICs containing immune cells in the rejecting kidney transplant and investigate the transcriptional programs associated with cell–cell contact. To do this, we used a computational method known as sequencing physically interacting cells (PIC-seq).21 PIC-seq compares the transcriptomes of a PIC and the 2 contributing nonconjugated singlet cells to determine the relative transcript contribution of each cell in a PIC and the differential expression of transcripts between a cell in a PIC and its corresponding nonconjugated singlet cell.
To assess the feasibility of the PIC-seq method applied to human biopsy samples, we generated single-cell transcriptomes and surface protein data from 11 kidney transplant biopsies enriched for CD45pos cells using Cellular Indexing of Transcriptomes and Epitopes by sequencing (CITE-seq). We investigated the transcriptional programs of interacting lymphocyte-myeloid cells and lymphocyte-endothelial cells that revealed interesting proinflammatory gene expression patterns increased on cell–cell contact.
MATERIALS AND METHODS
Biopsy Samples
Research core biopsy samples were obtained at the time of indication kidney transplant biopsy at Washington University under an institutional review board–approved protocol (No. 201102312).
Single-cell Isolation and Enrichment
Tissue was minced in digestion solution (PBS containing 0.25 mg/mL Liberase and 40 U/mL DNase I), and thermomixed at 37°C and 250 rpm. Cells were isolated as per Supplemental Methods (SDC, https://links.lww.com/TP/C853). Viability was determined using trypan blue. TotalSeq-C (Biolegend) CD31, CD138, CD27, CD79b, IgG-Fc, CD19, CD3, CD16, CD4, CD11c, CD56, CD14, CD8, and CD45 antibodies were used for CITE-seq as per Supplemental Methods (SDC,https://links.lww.com/TP/C853). CD45pos and CD45neg cell fractions were separated as per Supplemental Methods (SDC,https://links.lww.com/TP/C853).
Single-cell Data Analysis
A detailed description of library preparation, sequencing, data analysis, and the PIC-seq pipeline is provided in the supplemental methods https://links.lww.com/TP/C853 and by Giladi et al.21
Immunofluorescence
Fixed-frozen human kidney biopsy sections were washed with PBS for 10 min, permeabilized with 0.25% Triton x-100 in PBS for 10 min, and blocked with 5% bovine serum albumin in PBS for 1 h. The primary antibodies were added and incubated at 4°C overnight. The secondary antibodies were incubated for 1 h at room temperature. The sections were counterstained with DAPI for 5 min and mounted with ProLong Gold.
RESULTS
Doublet Clusters in Single-cell Data From Allograft Biopsies Can Be Biologically Relevant
CITE-seq was performed on 11 for-cause kidney allograft biopsies from 11 different patients. These included 4 biopsies with a diagnosis of AMR, 3 biopsies with T cell–mediated rejection (TCMR) (including with borderline acute rejection), 2 biopsies with mixed rejection, and 2 biopsies with no rejection (Table S1, SDC, https://links.lww.com/TP/C853). Three of the 5 biopsies with a clinical diagnosis of TCMR or no rejection had histologic evidence of glomerulitis (Banff g lesion) and/or peritubular capillaritis (Banff ptc lesion). Half of the nonrejection and pure AMR biopsies had tubulitis (Banff t lesion). All the of mixed and AMR biopsies (n = 6) had DSAs identified at the time of biopsy. However, 2 of 6 of these biopsies had no C4d staining the peritubular capillaries. Cells from each biopsy were enriched for CD45pos cells to improve our ability to identify immune cell PICs. We performed CITE-seq on enriched and nonenriched cells from each biopsy. We used the R package Harmony to correct for batch effect and the Seurat package to cluster cells based on transcriptional similarity (Figure S1, SDC, https://links.lww.com/TP/C853). We identified all major kidney and immune cell types based on lineage specific markers (Figure 1). Our final combined dataset included 31 203 cells across 19 clusters. Identification of the major immune cell types was confirmed by surface protein expression of CD3, CD8, CD56, CD14, CD16, and CD19 (Figure 1B). Each of the 4 clinical diagnoses were represented within each cell type cluster (Figure S2, SDC, https://links.lww.com/TP/C853).
;)
FIGURE 1.: Combined integrated CITE-seq dataset from 11 biopsy samples. A, 31 203 single-cell UMAP plot of combined dataset. D1 and D2 denote doublet clusters. B, Violin plot highlighting lineage specific markers for each cluster identified in A. Immune cell types are also identified by surface protein markers, CD3, CD8, CD56, CD14, CD16, and CD19. D1 and D2 express lineage specific markers from 2 cell types, thus, are heterotypic doublet clusters. UMAP, Uniform Manifold Approximation and Projection.
We identified 2 clusters of heterotypic doublets in our final combined dataset (Figure 1A): D1, expressing myeloid lineage markers and B-cell lineage markers and D2, expressing endothelial cell markers and proximal tubular cell markers. Most scRNA-seq analysis pipelines would consider such doublet clusters artifact and remove them from the final analysis. However, D1 was found at a frequency much greater than expected for artifactual doublets (4.7X; Figure S3A, SDC, https://links.lww.com/TP/C853). Therefore, the D1 cluster was considered a biologically relevant PIC, and we chose to further investigate its relevance in the alloimmune response in the kidney allograft. D2 was considered an artifact as expected, and observed frequency was similar.
Physically Interacting Dendritic Cells and B Cells Are Found in Rejecting Kidney Transplant Biopsies
The D1 cluster (n = 72, 93% from rejecting samples; Figure S4A, SDC, https://links.lww.com/TP/C853) was composed of macrophages or dendritic cells interacting with B cells or plasma cells as these PICs coexpressed CD68 and CD79A or MZB1 gene transcripts (Figure 1B). We performed a PIC-seq analysis of the D1 doublet cluster to determine the transcriptional profile of these PICs using the D1, macrophage, dendritic cell, B-cell, and plasma-cell clusters to computationally deconvolve each sequenced PIC into 2 contributing single cells and determine the gene expression of these cells. The PIC-seq linear regression model used to calculate the ratio of transcripts from each of the 2 cells in a PIC (the mixing factor) performed well (R2 = 0.84; Figure S5A, SDC, https://links.lww.com/TP/C853). PICs were enriched for a dendritic cell subtype (we called DC_B) and CD20pos B cells (Figure 2A). The physical interaction of myeloid cells and B cells in the rejecting allograft was confirmed in an independent allograft biopsy (Figure 2B). A functional analysis of genes increased in these PICs revealed biologic processes that included immune cell activation, regulation of immune responses and cytokine production (Figure 2C). Genes with increased expression in PICs included LILRA4, IRF7, PTGDS, and GZMB and genes downregulated in PICs included LYZ and CD1C (Figure 2D). In total, there were 106 genes whose observed PIC expression (log2(UMI)) was greater or less than expected based on singlet gene expression (fold change 1.5; Table S2, SDC, https://links.lww.com/TP/C853). LILRA4 encodes for a leukocyte immunoglobulin-like receptor, which is classified as an activating receptor, and expression has to date only been described in plasmacytoid dendritic cells (pDC).24,25 The granzyme B gene, GZMB, was also differentially expressed in PIC dendritic cells (Figure 2E). Granzyme B is known to be expressed in activated pDC and that activated pDC express IRF7 (Figure S6A, SDC, https://links.lww.com/TP/C853).26 DC_B cells in PICs express CD68 and not CD1C consistent with a pDC identity (Figure S6B,C, SDC, https://links.lww.com/TP/C853). The finding of increased expression of LILRA4 and its cognate receptor gene, BST2, in these PICs suggests cell–cell interaction via LILRA4 receptor and the BST2 ligand (Figure 2E). These data suggest that DC_B cells are pDCs and that these cells are activated within the kidney allograft on physical contact with B cells.
FIGURE 2.: PIC-seq analysis of D1 cluster. A, Shown are the relative contribution of myeloid cell types and BC or plasma cells to this set of PICs. This shows DC_B dendritic cells and B-cell interactions predominate. B, Immunofluorescence image (×60) of an independent human kidney transplant biopsy with a clinical diagnosis of chronic active AMR (g2, ptc3). CD20pos cells (red) physically interacting with CD68pos cells (green) denoted by a white arrow. * Denotes the location of the inset. C, Pathway analysis of genes with increased expression in PICs. D, Observed gene expression levels in PICs plotted against their expected levels as determined by PIC-seq, pooled over all PICs. Genes with observed:expected ratio >1.5 are highlighted and colored by their specificity in the B cell (blue) or DC (brown) expected contributions (log2(fold change) between the 2 background populations). E, Gene expression of PC or BC interacting with DC_B cells shown in gray bars. Gene expression from singlet myeloid cells (green bars) and singlet lymphoid cells (red bars). Gold sphere connected to the dark blue sphere suggests a physical ligand-receptor interaction between DC_B and B cell. BC, B cells; DC, dendritic cell; PC, plasma cells; PICs, physically interacting cell complexes.
Macrophages and Dendritic Cells Physically Interact With T and NK Cells in Rejecting Kidney Transplant Biopsies and Express Genes That Augment NK Cell Activity
We reclustered macrophages and dendritic cells from the parent dataset and detected 3 macrophage subtypes we labeled as CD14pos, CD16pos, and macrophage1 and dendritic cell clusters based on gene and surface protein expression (Figure 3Aand B). We also detected 2 doublet clusters, we named D3 (n = 517, 93% from rejecting samples; Figure S4B, SDC, https://links.lww.com/TP/C853) and D4 (n = 31) (Figure 3A,B,D). D3 was a large doublet cluster expressing macrophage, dendritic cell, NK, and T-cell markers (Figure 3B). D3 was found at a much greater frequency (9.5X) than expected for artifactual doublets (Figure S3A, SDC, https://links.lww.com/TP/C853). Therefore, the D3 cluster was considered a biologically relevant PIC. D4 was considered an artifact as expected and observed frequency was similar. We identified T-cell-macrophage physical interactions in an independent rejecting biopsy (Figure 3C).
FIGURE 3.: A, UMAP plot of cells subset from macrophages and dendritic cells from the parent object. B, Violin plot of cell type specific markers and surface protein expression of CD14, CD16, CD3, CD8, and CD56. C, Immunofluorescence image (×60) of an independent human kidney transplant biopsy with a clinical diagnosis of chronic active AMR (g2, ptc3). CD3pos cells (green) and CD68pos cells (red) with physical interaction denoted by a white arrow. */** Denotes the location of the insets. D, Table of total cell numbers from each cluster shown in A. PICs, physically interacting cell complexes; UMAP, Uniform Manifold Approximation and Projection.
We performed a PIC-seq analysis of D3 to identity the transcriptional profiles of these PICs. The PIC-seq linear regression model used to calculate the mixing factor performed well on this dataset (R2 = 0.85; Figure S5B, SDC, https://links.lww.com/TP/C853). NK cells were enriched in these PICs (Figure 4A). A functional analysis of genes increased in PICs revealed biologic processes that included NK cell chemotaxis, cytokine-mediated signaling, and cell killing (Figure 4B). There were 37 genes whose observed PIC expression (log2(UMI)) was greater or less than expected based on singlet gene expression (fold change 0.75; Table S3, SDC,https://links.lww.com/TP/C853). Gene expression increased in PICs included LYPD2, CXCL2, CXCL3, CCL2, CCL3, CCL3L1, and CCL4L2 (Figure 4C). LYPD2 is one of 20 genes identified in a bulk transcriptome analysis of DSA-positive versus DSA-negative AMR kidney biopsies.27 Increased LYPD2 expression occurred on dendritic cell interaction with NK cells, CD8 T cells, and CD4 T cells (Figure 4D). This gene is known to be expressed in nonclassical monocytes and to a lesser extend in dendritic cells (Figure S7, SDC, https://links.lww.com/TP/C853). CCL2 expression was increased in NK cell-macrophage1 PICs (Figure 4D), while CCL3 expression was increased in NK cell PICs involving CD14pos macrophages and macrophage1 (Figure 4D). CCL4L2 expression was increased in CD14pos macrophage-CD8 T-cell PICs (Figure 4D). CCL3 is also known as MIP-1a and has been shown to enhance cytotoxicity in resting NK cells and can also be produced by NK cells.28–31CCL3 and CCL4L2 are known to be differentially expressed in human AMR biopsy samples.32 NK cells activated by PDGF in the tumor microenvironment setting have been shown to increase expression of CCL3 and CCL4L2.33 Myeloid-lymphocyte interactions (PICs) can be identified in CITE-seq data and associate with a proinflammatory gene expression pattern in the kidney allograft.
FIGURE 4.: PIC-seq analysis of D3 cluster. A, Shown are the relative contribution of myeloid cells and lymphocytes to this set of PICs. This shows a relative increase in the proportion of NK cells in PICs. B, Pathway analysis of genes with increased expression in PICs. C, Observed gene expression levels in PICs plotted against their expected levels as determined by PIC-seq, pooled over all PICs. Genes with observed:expected ratio >0.75 are highlighted and colored by their specificity in the myeloid cells (red) or lymphocytes (green) expected contributions (log2(fold change) between the 2 background populations). D, Mean observed and expected gene expression levels in PICs grouped according to their myeloid and lymphocyte contributor identities (as in Figure 4A). PICs, physically interacting cell complexes.
Endothelial Cells Physically Interact With NK and T Cells in Kidney Transplant Biopsies and Express Immune Cell Migration and Cytotoxicity Genes
We subclustered the endothelial cell cluster from the parent object to resolve the endothelial cell subtypes found in kidney transplant biopsies. This analysis detected 3 subclusters of endothelial cells, glomerular endothelial cells uniquely expressing EHD3, proximal tubular endothelial cells (PTC) expressing PLVAP, and a cluster of endothelial cells expressing IFNγ-related genes such as CXCL9 and CXCL11 (IFNγ-EC). We also detected 2 doublet clusters D5 and D6 (Figure 5Aand B). The D5 cluster (n = 107, 91% from rejecting samples; Figure S4C, SDC, https://links.lww.com/TP/C853) expressed endothelial cell markers, CD3 (transcript and protein), CD8 protein, and CD56 protein as well as other T cell– and NK cell–associated genes (Figure 5C). D5 was found at a much greater frequency (3.6X) than expected for artifactual doublets (Figure S3A, SDC, https://links.lww.com/TP/C853). Therefore, the D5 cluster was considered a biologically relevant PIC. D6 was considered an artifact as expected and observed frequency was similar. As further evidence for the biologic relevance of D5, we used an immunomagnetic isolation assay (Figure S8A, SDC, https://links.lww.com/TP/C853). For 5 samples, CD45pos cells were captured and sequenced independently of CD45neg cells. Consistent with previous studies, singlet endothelial subclusters did not express CD45 surface protein or PTPRC (Figure S8B, SDC, https://links.lww.com/TP/C853).22,23 The proportion of cells in the nonenriched fraction resembles in vivo cell proportions where the endothelial-to-proximal tubule cell (EC/PT) ratio is 0.3 (Figure S3B, SDC, https://links.lww.com/TP/C853). However, the EC/PT ratio in the enriched sample is reversed suggesting CD45neg endothelial cells are captured in the CD45pos enriched fraction due to cell–cell interaction with CD45pos cells (Figures S3B and S8C–F, SDC, https://links.lww.com/TP/C853). The presence of endothelium–NK cell and endothelium–T-cell interactions in rejecting kidney is confirmed by immunofluorescence of an independent biopsy sample (Figure 5D,E).
FIGURE 5.: A, UMAP plot of cells subset from endothelial cells from the parent object. B, Table of total cell number found in A. C, Violin plot of cell type specific markers and surface protein expression of CD3, CD8, CD56, CD14, and CD16. D, Immunofluorescence images (×60) of an independent human kidney transplant biopsy with a clinical diagnosis of chronic active AMR (g2, ptc3). CD56pos (green) and CD31pos cells (white) with physical interaction denoted by a white arrow. * Denotes the location of the insets. E, Immunofluorescence image (×60) from the same biopsy as in D. CD3pos (green) and CD31pos cells (red) with physical interaction denoted by a white arrow. * Denotes the location of the insets. UMAP, Uniform Manifold Approximation and Projection.
We performed PIC-seq analysis on the D5 cluster. The PIC-seq linear regression model used to calculate the mixing factor performed well on this dataset (R2 = 0.91; Figure S5C, SDC, https://links.lww.com/TP/C853). There were 12 genes whose observed PIC expression (log2(UMI)) was greater than expected based on singlet gene expression (fold change 0.75; Table S4, SDC, https://links.lww.com/TP/C853). There was a slight enrichment of CD4 T cells, NK cells, and NKT cells but not of CD8 T cells in these PICs (Figure 6A). Most interacting cells included PTCs with slight enrichment of IFNγ-ECs. Cell killing was the top biologic pathway identified using genes increased in these PICs (Figure 6B). Increased expression of CXCL9 and CXCL10 genes occurred in the endothelial cells of PICs, genes well known to be increased in association with rejection (Figure 6C). Other increased genes in PICs included MZB1 and TNFRSF6B (Figure 6C). MZB1 is a molecular chaperone and is important in B-cell and plasma-cell differentiation.34 However, MZB1 is also expressed in other cell types including NK cells (Figure S9A, SDC, https://links.lww.com/TP/C853).35 Focusing on NK cells, we found increased expression of GNLY, CCL3, and CCL4 from NK cells physically interacting with endothelium in allograft biopsies, although the fold increase was <0.75 (Figure 6D). GNLY, CCL3, and CCL4 are known to be expressed by NK cells (Figure S9C, SDC, https://links.lww.com/TP/C853). We also found increased FCGR3A expression in these PICs likely from bound NK cells given its known expression in NK cells (Figure S9C, SDC, https://links.lww.com/TP/C853). However, it is possible that endothelial cells are expressing FCGR3A in these PICs. IFNγ-ECs interacting with NK cells have increased expression of CX3CL1 when compared with singlet IFNγ-ECs (Figure S9B, SDC, https://links.lww.com/TP/C853). This highlights the possibility that endothelium–NK cell interaction might be mediated by fractalkine in the setting of rejection. These data support a proinflammatory role for NK cells in the kidney allograft through physical interaction with kidney endothelium.
FIGURE 6.: PIC-seq analysis of D5 cluster. A, Shown are the relative contribution of endothelial cells and lymphocytes to this set of PICs. B, Pathway analysis of genes with increased expression in PICs. C, Observed gene expression levels in PICs plotted against their expected levels as determined by PIC-seq, pooled over all PICs. Genes with observed:expected ratio >0.75 are highlighted and colored by their specificity in the endothelial cells (red) or lymphocytes (green) expected contributions (log2(fold change) between the 2 background populations). D, Mean observed and expected gene expression levels in PICs grouped according to their endothelial and lymphocyte contributor identities (as in A). PICs, physically interacting cell complexes.
Our experimental workflow is outlined in Figure S10 (SDC, https://links.lww.com/TP/C853).
DISCUSSION
In this proof of principle study, we show that doublet clusters in single-cell RNA-seq datasets from human kidney allografts can be biologically relevant physically interacting cells. We can identify these doublet clusters, or PICs, from small amounts of tissue such as a kidney transplant biopsy. Previous single-cell RNA-seq studies used computational tools that infer cell–cell contact based on the average expression of ligand and cognate receptor genes from 2 clusters of single cells. Such inferences are made without any spatial information regarding the physical distance between 2 cell types. Furthermore, these tools are unable to differentiate the transcriptional changes of interacting cells from noninteracting cells within each cluster as ligand or receptor gene expression is averaged. PIC-seq does not rely on these assumptions and can quantify the gene expression occurring on actual cell–cell contact. This approach allows for the study of the transcriptional consequences of such immune cell–cell contacts. This analysis pipeline can also be applied post hoc to existing scRNA-seq datasets.
Our study identified 3 PIC clusters (D1, D3, D5) from a total of 6 doublet clusters. We used the CITE-seq method to increase confidence in identifying immune cell types. This was important as a transcript to protein expression mismatch occurs in some immune cell types (eg, CD4 T cells). Enrichment of samples for CD45pos cells was performed for 2 reasons: first, to increase the probability of identifying immune cell PICs and second, to validate PICs composed of a CD45pos cell and a CD45neg cell. Technical artifact doublets occur due to the stochastic coencapsulation of 2 cells during a microfluidic single-cell experiment. Cell titration experiments have shown a linear relationship between the multiplet rate and number of recovered cells consistent with Poisson loading of cells.19,20 Furthermore, capture efficiency is independent of cell size.20 Therefore, the size of artifact doublet clusters is determined by the number of input cells loaded into the microfluidics circuit and is not influenced by cell size (Supplemental Methods, SDC,https://links.lww.com/TP/C853). Our group has observed doublet clusters in prior datasets that were larger than predicted. Our current dataset confirms the presence of doublet clusters with high observed-to-predicted ratios (Figure S3A, SDC, https://links.lww.com/TP/C853). Enrichment and sequencing of CD45pos cells independently of CD45neg cells showed an EC/PT ratio that was reversed with respect to that expected in the nonenriched fraction (Figure S3B, SDC, https://links.lww.com/TP/C853). Previous studies have demonstrated the lack of CD45 protein expression in kidney endothelial cells even in the setting of increased HLA-DR protein expression.22,23 This was confirmed in our dataset, which gave us confidence in the CD45 enrichment approach to validate our PIC-seq findings. These findings are consistent with doublets that can be biologically relevant in kidney transplant biopsy tissue. To validate our findings, we confirmed the presence of the 3 PIC types described using immunofluorescence imaging.
The PIC-seq analyses of myeloid-lymphocyte PICs (D1+D3) suggest that immune cell activation is occurring within kidney tissue during rejection. Dendritic cells (DC_B) express GZMB and IRF7 on interaction with B cells and dendritic cells (CD1Cpos) expressing LYPD2 on interaction with T cells and NK cells. Immune cell interaction with endothelium is thought to play a central role in kidney transplant rejection. Large-scale bulk transcriptomic studies have identified increased NK cell and endothelial cell transcripts in AMR biopsies, including FCGR3A.6,36 These studies suggest NK cells are activated on antibody binding to the Fc receptor expressed on NK cells. Other studies have looked for alternative mechanisms of NK cell activation on binding endothelium in rejection, such as the “missing self” mediated activation of NK cells.37 Our PIC-seq analysis of lymphocyte and endothelial cell PICs (D5) confirms that proinflammatory mediators such as CXCL9, CXCL10, and FCGR3A are increased on cell–cell contact. Although these genes were previously identified in bulk transcriptomic studies, their role in cell–cell contact in vivo could only be assumed.
Our findings are consistent with prior assumptions regarding the cells involved in rejection and offers an alternative tool to investigate the transcriptional changes occurring as a consequence of cell–cell interaction in the kidney allograft. Our data also highlight CD4 T-cell contact with endothelium, a previously underappreciate cell–cell interaction in the setting of rejection. The endothelium is the interface between donor tissue and recipient immune system and is important in AMR (peritubular capillaritis and glomerulitis).38 Therefore, our approach has the potential to increase our understanding of cell–cell contact at this interface and in other rejection-related microenvironments of the kidney allograft.
The original PIC-seq method used a cell sorting method to physically isolate PICs from singlet cells before sequencing.21 We believe that this step would lead to a significant loss of cells and PICs. Therefore, we relied on observed-to-predicted doublet ratios and CD45pos cell enrichment to determine which doublet clusters were artifact and which were PICs. Our study has limitations. First, the low number of cells sequenced limits our power to detect more PIC associated genes. Therefore, we did not perform a statistical analysis of the rejection subtypes present in each of the PIC clusters. However, most PICs were from rejecting biopsies. Interestingly, most D3 PICs were from AMR biopsies whereas D1 and D5 PICs had similar numbers from TCMR and AMR biopsies. Second, the number and fold change of differentially expressed genes in PICs is also a limitation. These limitations could be mitigated by increasing sample number and cells sequenced per sample. However, cost then becomes an additional limitation. Third, we did not validate each differential gene identified in PICs. However, we did validate the presence of PIC combinations at a cellular level. Finally, our findings likely do not include all PICs present in vivo due to the process of tissue dissociation. Refinement of the dissociation protocol may favor other types of PICs not identified in this work. A multiomics approach to investigate cell–cell interactions in the rejecting kidney would be the ideal approach. For example, a single-cell resolution, unbiased, spatial transcriptomics method would help to validate the PIC-seq findings and add spatial information. Current receptor-ligand computational tools, such as CellChat, are useful for analysis of remote, or secreted, signaling interactions between cells, but they focus on ligand and receptor genes only.14,16,17 Therefore, such tools are complementary to PIC-seq.
In conclusion, we successfully applied an approach to investigate physical cell–cell interactions at the transcriptomic level from a kidney transplant biopsy, a limited tissue source. The PIC-seq method does not rely on assumptions regarding cell contact, a limitation of other ligand-receptor analyses tools in single-cell RNA-seq experiments. The PIC-seq approach complements current receptor-ligand computational methods and has the potential to uncover important cell–cell interactions driving the alloimmune response and lead to targets for future treatments in rejection.
REFERENCES
1. Wolfe RA, Ashby VB, Milford EL, et al. Comparison of mortality in all patients on dialysis, patients on dialysis awaiting transplantation, and recipients of a first cadaveric transplant. N Engl J Med. 1999;341:1725–1730.
2. Matas AJ, Leduc R, Rush D, et al. Histopathologic clusters differentiate subgroups within the nonspecific diagnoses of CAN or CR: preliminary data from the DeKAF study. Am J Transplant. 2010;10:315–323.
3. Famulski KS, Reeve J, de Freitas DG, et al. Kidney transplants with progressing chronic diseases express high levels of acute kidney injury transcripts. Am J Transplant. 2013;13:634–644.
4. Halloran PF, Pereira AB, Chang J, et al. Microarray diagnosis of antibody-mediated rejection in kidney transplant biopsies: an international prospective study (INTERCOM). Am J Transplant. 2013;13:2865–2874.
5. Loupy A, Haas M, Roufosse C, et al. The Banff 2019 Kidney Meeting report (I): updates on and clarification of criteria for T cell- and antibody-mediated rejection. Am J Transplant. 2020;20:2318–2331.
6. Hidalgo LG, Sis B, Sellares J, et al. NK cell transcripts and NK cells in kidney biopsies from patients with donor-specific antibodies: evidence for NK cell involvement in antibody-mediated rejection. Am J Transplant. 2010;10:1812–1822.
7. Sis B, Halloran PF. Endothelial transcripts uncover a previously unknown phenotype: C4d-negative antibody-mediated rejection. Curr Opin Organ Transplant. 2010;15:42–48.
8. Luecken MD, Theis FJ. Current best practices in single-cell RNA-seq analysis: a tutorial. Mol Syst Biol. 2019;15:e8746.
9. Malone AF, Wu H, Fronick C, et al. Harnessing expressed single nucleotide variation and single Cell RNA sequencing to define immune cell chimerism in the rejecting kidney transplant. J Am Soc Nephrol. 2020;31:1977–1986.
10. Wu H, Malone AF, Donnelly EL, et al. Single-cell transcriptomics of a human kidney allograft biopsy specimen defines a diverse inflammatory response. J Am Soc Nephrol. 2018;29:2069–2080.
11. Thareja G, Suryawanshi H, Luo X, et al. standardization and interpretation of RNA-sequencing for transplantation. Transplantation. 2023;107:xxx–xxx. doi: 10.1097/TP.0000000000004763
12. Peereboom ETM, Matern BM, Spierings E, et al. The value of single-cell technologies in solid organ transplantation studies. Transplantation. 2022;106:2325–2337.
13. Malone AF, Humphreys BD. Single-cell transcriptomics and solid organ transplantation. Transplantation. 2019;103:1776–1782.
14. Efremova M, Vento-Tormo M, Teichmann SA, et al. Inferring cell-cell communication from combined expression of multi-subunit ligand-receptor complexes. Nat Protoc. 2020;15:1484–1506.
15. Wang Y, Wang R, Zhang S, et al. iTALK: an R package to characterize and illustrate intercellular communication. bioRxiv. 2019;507871.
16. Jin S, Guerrero-Juarez CF, Zhang L, et al. Inference and analysis of cell-cell communication using CellChat. Nat Commun. 2021;12:1088.
17. Cabello-Aguilar S, Alame M, Kon-Sun-Tack F, et al. Inference of intercellular networks from single-cell transcriptomics. Nucleic Acids Res. 2020;48:e55.
18. Xi NM, Li JJ. Benchmarking computational doublet-detection methods for single-cell RNA sequencing data. Cell Syst. 2021;12:176–194.e6.
19. Bloom JD. Estimating the frequency of multiplets in single-cell RNA sequencing from cell-mixing experiments. PeerJ. 2018;6:e5578.
20. Zheng GX, Terry JM, Belgrader P, et al. Massively parallel digital transcriptional profiling of single cells. Nat Commun. 2017;8:14049.
21. Giladi A, Cohen M, Medaglia C, et al. Dissecting cellular crosstalk by sequencing physically interacting cells. Nat Biotechnol. 2020;38:629–637.
22. Muczynski KA, Ekle DM, Coder DM, et al. Normal human kidney HLA-DR-expressing renal microvascular endothelial cells: characterization, isolation, and regulation of MHC class II expression. J Am Soc Nephrol. 2003;14:1336–1348.
23. Marcu R, Choi YJ, Xue J, et al. Human organ-specific endothelial cell heterogeneity. iScience. 2018;4:20–35.
24. Abdallah F, Coindre S, Gardet M, et al. Leukocyte immunoglobulin-like receptors in regulating the immune response in infectious diseases: a window of opportunity to pathogen persistence and a sound target in therapeutics. Front Immunol. 2021;12:717998.
25. Cao W, Bover L. Signaling and ligand interaction of ILT7: receptor-mediated regulatory mechanisms for plasmacytoid dendritic cells. Immunol Rev. 2010;234:163–176.
26. Rissoan MC, Duhen T, Bridon JM, et al. Subtractive hybridization reveals the expression of immunoglobulin-like transcript 7, Eph-B1, granzyme B, and 3 novel transcripts in human plasmacytoid dendritic cells. Blood. 2002;100:3295–3303.
27. Halloran PF, Madill-Thomsen KS, Pon S, et al. INTERCOMEX Investigators. Molecular diagnosis of ABMR with or without donor-specific antibody in kidney transplant biopsies: differences in timing and intensity but similar mechanisms and outcomes. Am J Transplant. 2022;22:1976–1991.
28. Robertson MJ. Role of chemokines in the biology of natural killer cells. J Leukoc Biol. 2002;71:173–183.
29. Taub DD, Ortaldo JR, Turcovski-Corrales SM, et al. Beta chemokines costimulate lymphocyte cytolysis, proliferation, and lymphokine production. J Leukoc Biol. 1996;59:81–89.
30. Taub DD, Sayers TJ, Carter CR, et al. Alpha and beta chemokines induce NK cell migration and enhance NK-mediated cytolysis. J Immunol. 1995;155:3877–3888.
31. Maghazachi AA, Al-Aoukaty A, Schall TJ. CC chemokines induce the generation of killer cells from CD56+ cells. Eur J Immunol. 1996;26:315–319.
32. Suviolahti E, Ge S, Nast CC, et al. Genes associated with antibody-dependent cell activation are overexpressed in renal biopsies from patients with antibody-mediated rejection. Transpl Immunol. 2015;32:9–17.
33. Barrow AD, Edeling MA, Trifonov V, et al. Natural killer cells control tumor growth by sensing a growth factor. Cell. 2018;172:534–548.e19.
34. Andreani V, Ramamoorthy S, Pandey A, et al. Cochaperone Mzb1 is a key effector of Blimp1 in plasma cell differentiation and β1-integrin function. Proc Natl Acad Sci U S A. 2018;115:E9630–E9639.
35. Schmiedel BJ, Singh D, Madrigal A, et al. Impact of genetic polymorphisms on human immune cell gene expression. Cell. 2018;175:1701–1715.e16.
36. Sis B, Jhangri GS, Bunnag S, et al. Endothelial gene expression in kidney transplants with alloantibody indicates antibody-mediated damage despite lack of C4d staining. Am J Transplant. 2009;9:2312–2323.
37. Koenig A, Chen CC, Marçais A, et al. Missing self triggers NK cell-mediated chronic vascular rejection of solid organ transplants. Nat Commun. 2019;10:5350.
38. Loupy A, Haas M, Solez K, et al. The Banff 2015 Kidney Meeting report: current challenges in rejection classification and prospects for adopting molecular pathology. Am J Transplant. 2017;17:28–41.
