Spatiotemporal immune atlas of the first clinical-grade, gene-edited pig-to-human kidney xenotransplant

Pig-to-human xenotransplantation is rapidly approaching the clinical arena; however, it is unclear which immunomodulatory regimens will effectively control human immune responses to pig xenografts. We transplanted a gene-edited pig kidney into a brain-dead human recipient on pharmacologic immunosuppression and studied the human immune response to the xenograft using spatial transcriptomics and single-cell RNA sequencing. Human immune cells were uncommon in the porcine kidney cortex early after xenotransplantation and consisted of primarily myeloid cells. Both the porcine resident macrophages and human infiltrating macrophages expressed genes consistent with an alternatively activated, anti-inflammatory phenotype. No significant infiltration of human B or T cells into the porcine kidney xenograft was detected. Altogether, these findings provide proof of concept that conventional pharmacologic immunosuppression is sufficient to restrict infiltration of human immune cells into the xenograft early after compatible pig-to-human kidney xenotransplantation.


Abstract: 41
Pig-to-human xenotransplantation is rapidly approaching the clinical arena; however, it is unclear 42 which immunomodulatory regimens will effectively control human immune responses to pig 43 xenografts. We transplanted a gene-edited pig kidney into a brain-dead human recipient on 44 pharmacologic immunosuppression and studied the human immune response to the xenograft 45 using spatial transcriptomics and single-cell RNA sequencing. Human immune cells were 46 uncommon in the porcine kidney cortex early after xenotransplantation and consisted of primarily 47 myeloid cells. Both the porcine resident macrophages and human infiltrating macrophages 48 expressed genes consistent with an alternatively activated, anti-inflammatory phenotype. No 49 significant infiltration of human B or T cells into the porcine kidney xenograft was detected. 50 Altogether, these findings provide proof of concept that conventional pharmacologic 51 immunosuppression is sufficient to restrict infiltration of human immune cells into the xenograft 52 early after compatible pig-to-human kidney xenotransplantation. 53 54 55 Kidney allotransplantation is a life-saving therapy for people with end-stage kidney disease, but 56 there are not enough human kidneys to meet the growing demand 1,2 . Xenotransplantation is a 57 promising solution to increase organ supply; however, it is unclear whether immunomodulatory 58 strategies currently used in allotransplantation will effectively control human immune responses 59 to xenografts. Although studies of porcine kidney xenografts in nonhuman primate (NHP) 60 recipients suggest that many mechanisms of immune injury are conserved between allotransplant 61 and xenotransplant recipients 3 , crossmatch-compatible xenotransplantation has not yet been 62 achieved in NHPs despite genetic knockdown of carbohydrate antigens that promote crossmatch 63 compatibility between pig cells and human sera 4-6 . While short-to intermediate-term graft survival 64 is possible in limited numbers of incompatible xenotransplant NHP recipients, overall recipient 65 survival is poor due to complications of intensive immunosuppression 7 . Crossmatch-compatible 66 transplantation remains thus a primary goal of current xenotransplantation efforts 8 , as this strategy 67 will likely result in safest outcomes for human transplant recipients. However, immunomodulatory 68 strategies which provide effective immune control in compatible pig-to-human 69 xenotransplantation cannot be assessed in incompatible NHP models. Evaluation of immune 70 control in human recipients of compatible porcine xenografts is therefore critically needed in order 71 to ensure recipient safety in forthcoming first-in-human clinical trials of xenotransplantation. 72 73 We developed a preclinical model of pig-to-human kidney xenotransplantation to test the 74 hypothesis that current standard-of-care pharmacologic immunosuppression would control 75 human immune responses in the porcine kidney early after crossmatch-compatible 76 xenotransplantation. As previously reported, porcine kidney xenografts were procured from a 77 domestic pig with 10 genetic modifications 9 and transplanted into a nephrectomized, crossmatch-78 compatible, brain-dead human recipient (Extended Data Fig. 1a) 10 . Immunomodulation of the 79 recipient was established through both genetic modification of the porcine kidney and 80 conventional pharmacologic immunosuppression. In addition to knockdown of the growth 81 hormone receptor and three carbohydrate xenoantigens (GGTA1, B4GALNT2, CMAH), the  82   porcine kidneys expressed human transgenes (CD55, CD46, THBD, PROCR, CD47, HMOX1)  83 intended to reduce inflammation and prevent thrombotic complications within the kidney. 84 Pharmacologic immunosuppression consisted of induction therapy with methylprednisolone, anti-85 thymocyte globulin, and rituximab, while maintenance immunosuppression included tacrolimus, 86 mycophenolate mofetil, and prednisone 10 . The kidney xenografts made urine but did not clear 87 creatinine 10 ; H&E sections of the xenografts revealed no evidence of acute cellular rejection or 88 binding of IgM, IgG, or complement proteins 10 . The experiment was terminated approximately 74 89 hours after transplant due to hemodynamic instability 10 . 90 91 Sequential needle core biopsies of the gene-edited porcine kidneys were taken immediately prior 92 to transplantation, in situ on postoperative days 1 and 3, and immediately prior to explant 93 (Extended Data Fig. 1b). Biopsies were analyzed using spatial transcriptomics and single-nuclear 94 RNA-sequencing (snRNA-seq) approaches. As detection of immune populations can be limited 95 in samples analyzed by snRNA-seq 11 , we also performed single-cell RNA-seq (scRNA-seq) on 96 CD45+ immune cells that were FACS-enriched from the explanted xenografts (Extended Data 97 Fig. 1b). In order to distinguish human from porcine cells, we aligned all sequenced reads to a 98 custom human-porcine hybrid reference genome and annotated clusters based on established 99 marker genes 12 . 100 101 As discrimination of porcine from human immune cells was necessary for this study, we assessed 102 the specificity of our mapping with three different approaches. First, we aligned independent 103 control biopsies from human and pig kidneys to the hybrid reference genome and found that key 104 immune genes mapped to the appropriate species reference (Extended Data Fig. 2). Although 105 these analyses evaluated the performance of our pipeline on samples derived from a single 106 species, we hypothesized that pipeline performance might differ for mixed species samples (i.e. 107 the xenograft). We therefore evaluated the mapping specificity of individual reads sequenced from 108 CD45+ immune cells sorted from the explanted xenograft. Cells were initially clustered according 109 to both cell type and species as individual genes had species-specific annotation after alignment 110 to the hybrid reference (Extended Data Fig. 3). We found that 97.8% of 18,833,529 transcripts 111 recovered from 6,513 immune cells associated with a single species, and 98.8% of cells 112 possessed >90% of transcripts from a single species (Extended Data Fig. 4). These analyses 113 further revealed that ~10% of reads in pig macrophages mapped to 17 human genes (0.27% of 114 human genes), likely as a consequence of high sequence homology (Extended Data Fig. 5). 115 Finally, to assess the impact of homology on the species assignment of individual cells, we 116 employed an alternative mapping strategy using custom modified reference genomes composed 117 of more species-specific genes (see Methods & Extended Data Fig. 6). We found a high 118 correlation in the species assignment of cellular barcodes between the two methods (Extended 119 Data Fig. 7), suggesting that the presence of gene homology did not significantly impact the 120 species assignment for most cell types when using the hybrid reference genome. Collectively, 121 these analyses highlighted the benefit of using the entire transcriptome to make cell assignments 122 in lieu of individual genes that may not be specific to species or cell type. 123

124
We therefore used Cell2location 13 to deconvolute our spatial transcriptomic data from the 125 xenograft biopsies into pig and human cell types based on input of reference transcriptomes. 126 Reference transcriptomes were derived from sorted CD45+ cells and parenchymal cells from the 127 porcine xenograft biopsies (Extended Data Fig. 8). We restricted the input to Cell2location to 128 reference transcriptomes recovered from the xenografts given the unknown impact of known to be relatively specific to these cell types ( Fig. 1 & Extended Fig. 9). Altogether, these 138 experiments reveal limited infiltration of the renal cortex by human immune cells.  Table  144 1) 16 . We found increased individual gene expression ( Fig. 2a,b) and composite gene expression 145 scores of M2 compared to M1 genes in both pig and human macrophages (Fig. 2c). Although 146 differences in gene expression between the species precluded comprehensive formal comparison 147 of donor-versus recipient-derived macrophages, we performed a limited comparison of pro-and 148 anti-inflammatory cytokines and select genes of interest known to be important in macrophage 149 activation and function 17 (Fig. 2d). Collectively, these data suggest that macrophages populating 150 the kidney xenograft express a more alternatively activated, anti-inflammatory transcriptome, 151 independent of species. 152 153 Altogether, this study of the immune census of porcine kidney xenografts after pig-to-human 154 xenotransplantation suggests that current genetic and pharmacologic immunomodulation 155 strategies can prevent early T and B cell infiltration of the xenograft and modulate inflammation 156 provoked by innate immune responses. Despite these encouraging results, our study has several 157 limitations. First, we were not able to determine the impact of immunomodulatory strategies at 158 later time points due to the short duration of the experiment. Second, our data do not distinguish 159 the specific impact of pharmacologic interventions from genetic edits within the pig kidney. Third, 160 we may have overestimated infiltrating human cell number as we could neither distinguish cells 161 in the kidney interstitium from potential blood contaminants nor correct for alignment error for 162 homologous genes. We attempted to overcome these limitations by using the whole transcriptome 163 to make species and cell type assignments to reduce the impact of homologous genes on our cell 164 calls. We also focused our analyses on cells unlikely to represent blood contaminants (i.e. Technologies; Vancouver, BC), depending on the blood volume. Once the PBMCs were isolated, 301 the cells were washed with buffer two times, once at 300xrcf for 8 minutes and once at 120xrcf 302 for 10 minutes with no brake. After removing the wash buffer, RBCs were lysed using 4 mL of 303 room temperature ACK lysis buffer (Quality Biological; Gaithersburg, MD) for 2 minutes on ice 304 and then the ice cold DPBS was added to the 14 mL mark and mixed. Cells were pelleted by 305 centrifugation, 400xrcf for 5 minutes at 4°C. The buffer was removed and 0.5-1 mL fresh, ice cold 306 DPBS without Ca2+ and Mg2+ was added and the cells were suspended. Cells were counted 307 and delivered to the UAB Single Cell core, and scRNA-seq was performed. 308 309

4) Human kidney 310
The native kidneys which were removed from the brain-dead recipient prior to xenotransplantation 311 were offered for allotransplantation but ultimately declined. After exhaustion of the transplant list, 312 the kidneys were transported to the laboratory and processed for spatial transcriptomics analysis 313 as described below. 314 315

5) Control pig kidneys 316
Porcine control kidneys were recovered from 1) a 208-day-old Chester White Cross wild-type sow 317 (Identification number 817D), weighing 154 kgs, and 2) an 8-day-old Chester White Cross 10-GE 318 male piglet (identification number 817D-1) weighing 1.3 kg at the UAB XPC. The kidneys were 319 transported to the laboratory and processed for spatial transcriptomics analysis as described 320 below. 321 322 Spatial Transcriptomicssample preparation 323 Frozen biopsy OCT blocks were equilibrated to -10°C before use. Using a cryostat, a 10 μm 324 section was placed onto a Visium Spatial Gene Expression Slide (10X Genomics) and processed 325 according to the manufacturer's protocol. In summary, slides were fixed in methanol for 30 326 minutes and stained with Hematoxylin and Eosin. Brightfield images were taken using a Keyence 327 BZ-X700 microscope. Slides were placed in specialized slide cassette holders and 328 permeabilization enzyme was added for 12 minutes at 37°C to release the RNA onto the slide. 329 Using the captured RNA, cDNA and subsequently second-strand DNA was created and amplified. • Spatially barcoded libraries prepared from core biopsies of the porcine kidney xenograft 372 ( Fig. 1; Extended Data Fig. 9). 373 • Spatially barcoded libraries prepared from control kidney samples, including human 374 kidney, 10-GE pig kidney, and wild-type pig kidney (Extended Data Fig. 2). 375 376 2) scRNA-seq data processing and analysis of CD45+ cells sorted from the porcine xenograft at 377 explant: 378 Cell Ranger (v.6.1.1) was used to pre-process sequenced reads aligned against the hybrid 379 human-pig genome reference to generate raw count matrices. Downstream analysis on the 380 filtered UMI expression profile for each droplet was completed using R (v.4.2.1) and Seurat 18 381 (v.4.2.0) using default parameters unless otherwise specified. Before conducting additional 382 analyses, background noise from ambient RNA was removed using SoupX 19 (v.1.6.1). The overall 383 contamination fraction (rho) was parameterized using the autoEstCont function to remove > 2% 384 background contamination in our dataset. The SoupX-corrected count matrix was then loaded 385 into R using the Read10X function and was further used to create a Seurat object. Cells with < 386 200 unique features, > 3000 unique features, and > 12% mitochondrial gene expression were 387 filtered. Features expressed in < 5 cells, and cells with doublet scores > 0.3 were also removed. 388 Data were then normalized by a scale factor of 10,000 and log1p-transformed using the 389 LogNormalize function. We then identified the top 3000 variable genes, ranked by coefficient of 390 variation, using FindVariableFeatures. Using the variable genes, we scaled and centered the 391 genes across the cells using the ScaleData function followed by the identification of principal 392 components (RunPCA). Thirty principal components were found and used to construct a k-393 nearest neighbor (KNN) graph using FindNeighbors. Clustering was subsequently performed 394 using FindClusters which employs a shared nearest neighbor (SNN) modularity optimization based 395 clustering algorithm. Cluster information was used as input into the uniform manifold approximation 396 and projection algorithm (RunUMAP) which further aided the visualization of cell manifolds in a 397 low-dimensional space. We ran Seurat's implementation of the Wilcoxon rank-sum test 398 (FindMarkers) to identify differentially expressed genes in each cluster. The expression of cluster-399 specific canonical markers was used to annotate each cell cluster. 400 Module scores were determined for the Seurat object using the AddModuleScore function with 401 the M1/pro-inflammatory or M2/anti-inflammatory gene lists as inputs (Extended Data Table 1  the reference dataset. The merged dataset was then normalized, scaled, and centered as 439 described above. Thirty principal components were found and used to construct a k-nearest 440 neighbor (KNN) graph and clustering was performed using modularity optimization, resulting in 441 17 reference cell clusters which were ultimately collapsed to broader cell types (n=13 clusters). 442 Cell clusters were visualized in low-dimensional space on a UMAP, and marker genes were used 443 to annotate each cell cluster in the reference dataset (Extended Fig. 8). After providing the reference cell type signatures, spatial data, and hyperparameters specified 475 above as inputs, the Cell2location model was trained using max_epochs = 30000 on the full 476 using SCTransform before visualizing gene expression in the respective spatial transcriptomics 500 landscapes. 501 502 503 Species assignment validation analyses using the modified species-specific reference genome 504

2) Construction of the species-specific reference genomes & data analysis 509
Sequenced reads were processed with Cell Ranger (version 6.1.1). Reference genomes 510 utilized included the hg38 and ss11 genomes as provided by Ensembl as well as the pre-511 compiled hg38 reference genome provided by 10x Genomics. Pre-transplantation porcine and 512 human samples were initially mapped to the opposite species' reference genome to identify 513 genes. Any gene with > 3 counts assigned was then subsequently identified and removed from 514 the original reference, thereby resulting in a modified reference genome for each species. All 515 samples were subsequently mapped to both species' modified reference genomes and gene 516 mapping rates on a per cell basis were compared between the two to enable identification of 517 porcine cells from human. As some level of mapping consistently occurred when processing 518 samples against the opposite species' reference genome, cells were identified as porcine if they 519 had a ratio of human to porcine mapped genes < 0.75 and subsequently identified as human if 520 the ratio was > 1.33. Cells that fell between the two ratios were labeled as ambiguous and 521 Spatial transcriptomics was performed on serial needle core biopsies of 10-GE porcine kidneys before and after transplantation into a brain-dead human recipient. Biopsies were obtained from either the right (pre-transplant and Day 3 samples) or left (Day 1 and Day 3T) xenografts. Cell type signatures were identified from reference transcriptomes using Cell2location or expression of individual marker genes (CD3E and CD19). a) Top: Human myeloid cells are detected in biopsies of the porcine kidney xenograft three days after transplantation. Capture spot color corresponds to cell abundance, and color scales to the right of each spatial plot indicate cell abundance. Note that scaling is conserved across time for a given cell type but differs between macrophages (cell abundance range: 0-2) and neutrophils (cell abundance range: 0-1). Visualization of cell abundance in a given capture spot is thus capped at 1 or 2 cells. Bottom: No detection of human T or B cells in xenograft biopsies at any time point. b) Calculated total (left) and normalized (right) cell abundance for various immune cell types in the indicated biopsies. For clarity, quantification of B cells is not shown. Note that cell abundance for human T cells and all B cells was imputed from expression of hg38-CD3E, ss11-CD19, and hg38-CD19 genes as shown in (a). "Pre-tx" = pre-transplant. Day 3T biopsy was taken on post-transplant day 3 at study termination.  scRNA-seq was performed on CD45+ immune cells sorted from the right porcine kidney xenograft at explant (see Extended Fig. 3), and macrophage clusters were selected for analysis. a) Expression of M1 (red) and M2 (blue) genes in human and pig macrophages (see Extended Data Table 1 and ref 16 for full gene list). b) Expanded view of top 50 most highly expressed M1 and M2 genes in each species. Note M2 > M1 genes for both species. c) Composite gene expression score of pig and human macrophages of M1-like pro-inflammatory (red) and M2-like anti-inflammatory (blue) gene signatures (ref 16). UMAPs were generated from re-clustering of macrophage clusters selected from Extended Fig. 3. d) Expression of select anti-and pro-inflammatory cytokine genes in pig and human macrophages. Average gene expression is visualized such that the mean of the scaled expression dataset is set at 0 with a standard deviation of 1. ss11-IL6 was not detected.