对于3D成像和免疫荧光染色实验，在至少两个单独的个体或单独的区域重复每个染色
。对于免疫荧光验证研究 ，使用了市售的抗体。还使用SNCV3或SCCV3分析了15个组织样品中的13个 。对于ISH
，分析了6个组织样品（4个活检和2个肾切除术）
。对于幻灯片序列，分析了67个组织冰球（6个个体） ，还使用SNCV3或森林分析了2个个体。对于visium
，成像了23个肾脏组织切片（22个个体），其中6个也使用SNCV3或SCCV3分析了6个 ，并使用Slide-Seq进行了检查 。空间转录组注释的正交验证揭示了SNCV3/SCCV3和这些技术中的标记基因表达相似
，以及与组织学验证的景点映射相对应的空间定位
。尽管来自同一样本的多构想数据将是最有用的，但在技术上仍然具有挑战性。但是 ，在可能的情况下
，在同一患者的样品子集上进行了几种技术，在某些情况下 ，使用相同的组织块来生成多模式数据（扩展数据图1A和补充表3） 。这种异质抽样方法可确保细胞类型的发现，同时最小化测定依赖性的偏见或伪影时
，使用了不同的肾脏组织来源
。我们认识到，多种技术的样本源的异质性是应有物流和患者活检材料有限的潜在局限性。 　　我们遵守了与这项研究有关的所有道德法规 。所有关于人类样本的实验遵循所有相关指南和法规
。根据知情同意书获得了作为KPMP联盟（https://kpmp.org）的一部分收集的人类样本（补充表1） ，并经华盛顿大学机构审查委员会的毕马威（KPMP）单一IRB批准
，并根据协议批准了批准。作为Hubmap财团的一部分，样本是由肾脏翻译研究中心（KTRC）根据华盛顿大学机构审查委员会批准的协议（IRB 201102312）收集的 。获得了华盛顿大学所有参与者的数据和样本的知情同意 ，包括接受部分或全肾切除术的活人或已故捐助者拒绝肾脏。通过印第安纳大学的知情同意
，获得了来自石材疾病患者的皮质和乳头状活检样本，并获得了印第安纳大学机构审查委员会（IRB 1010002261）的批准
。对于森林空间基因的表达 ，根据知情同意书或印第安纳州的活检生物库同意
，从毕马威（KPMP）获得参考肾切除术和肾脏活检样本，并经豁免同意书获得了印第安纳大学制度审查委员会（IRB 1906572234）的批准。在密歇根大学IRB HUM00150968下的人类肾脏移植转录组（HKTTA）下 ，根据人类肾脏移植转录组（HKTTA）获得知情同意
，获得了活着的供体活检 。在约翰·霍普金斯（Johns Hopkins）机构审查委员会（IRB 00090103）批准的豁免同意书中，从约翰·霍普金斯大学病理档案库中获得了剩下的剩菜冰冻冻结的COVID-AKI肾脏活检
。 　　对于单核OMIC测定 ，根据在线提供的协议（https://doi.org/10.17504/protocols.io.568g9hw）处理组织。为了进行核制剂，收集了约7个40 µm厚度的部分
，并将其存储在Rnalater溶液（RNA分析）中或将其保存在干冰（可访问的染色质分析）上，直到加工或使用新鲜 。为了证实组织组成 ，获得了5 µM的侧面这些厚切片的组织学
，并确定了包括肾小球在内的皮质或髓质组成的相对量
。对于单细胞OMIC测定，使用低温稳定器（Stemcell Technologies）保留了使用的组织（15 CKD ，12个AKI和18个活体活检核心）。 　　Nuclei were isolated from cryosectioned tissues according to a protocol available online (https://doi.org/10.17504/protocols.io.ufketkw) with the exception that 4′,6-diamidino-2-phenylindole (DAPI) was excluded from the nuclear extraction buffer and used only to stain a subset of nuclei used for counting.核直接用于OMIC测定 。 　　根据在线可用的协议（https://doi.org/10.17504/protocols.io.7dthi6n）从冷冻组织中分离出单细胞
。单细胞悬架立即被转移到密歇根大学高级基因组学核心设施进行进一步处理。 　　10x single-nucleus RNA-seq and 10x single-cell RNA-seq were performed according to protocols available online (https://doi.org/10.17504/protocols.io.86khzcw and https://doi.org/10.17504/protocols.io.7dthi6n, respectively), both using the 10x Chromium Single-Cell 3′试剂套件V3 。使用GRCH38（HG38）参考基因组使用10X细胞Ranger（v.3）管道进行样品反复分解
，条形码加工和基因表达定量，除SCCV3实验的子集外 ，使用HG19（补充表1中指示）
。对于单核数据
，内含子被包括在表达估计中。 　　如前所述，SNARE-SEQ217是根据在线提供的协议（https://doi.org/10.17504/protocols.io.be5gjg3​​w）进行的 。使用300周期和200个循环试剂盒分别对Novaseq 6000（Illumina）系统（Novaseq Control Software V.1.6.0和V.1.7.0）分别对可访问的染色质和RNA库进行了测序。 　　SNARE2数据的详细逐步处理已被概述
。18。现在 ，这是作为自动数据处理管道开发的
，可在GitHub（https://github.com/huqiwen0313/snarepip）上获得 。SNAREPIP（V.1.0.1）用于处理所有SNARE2数据集
。该管道为复杂的单细胞分析提供了一个自动框架，包括质量评估 ，DoubleT删除，细胞聚类和识别
，稳健的峰值产生和可及的区域识别，并具有灵活的分析模块以及用于质量评估和下游分析的摘要报告。定向的无环图用于合并整个数据处理步骤 ，以更好地控制误差控制和可重复性 。对于RNA处理
，这涉及使用CASADAPT（v.3.1）51，Dropest（v.0.8.6）52去除可访问的染色质污染读数 ，以提取细胞条形码和星形（版本2.5.2b）53
，以使标记的标记读数与基因组（GRCH38）保持一致
。对于可访问的染色质数据，这涉及Snaptools（V.1.2.3）54和Minimap（V.2-2.20）55 ，以与基因组对齐（GRCH38）。 　　通过10倍细胞游舱过滤器的细胞条形码用于下游分析 。删除线粒体转录本（MT-*）
，使用DoubleTectection软件（v.2.4.0）56鉴定了双线，然后删除
。将所有样品在实验和细胞条形码中组合在一起 ，大于400
，而检测到的基因少于7,500个基因进行下游分析。为了进一步删除低质量数据集，使用Pagoda2（https://github.com/hms-dbmi/pagoda2）应用了基因UMI比率滤波器（Gene.vs.molecule.cell.filter） 。 　　作为质量控制步骤 ，截止 <50% mitochondrial reads per cell was applied. The ambient mRNA contamination was corrected using SoupX (v.1.5.0)57. The mRNA content and the number of genes for doublets are comparatively higher than for single cells. To reduce doublets or multiplets from the analysis, we used a cut-off of >500和 <5,000 genes per cell. 　　Cell barcodes for each sample were retained with the following criteria: having an DropEst cell score of greater than 0.9; having greater than 200 UMI detected; having greater than 200 and less than 7,500 genes detected. Doublets identified by both DoubletDetection (v.3.0) and Scrublet (https://github.com/swolock/scrublet; v.0.2.2) were removed. To further remove low-quality datasets, a gene UMI ratio filter (gene.vs.molecule.cell.filter) was applied using Pagoda2. 　　Cell barcodes for each sample that had already passed quality filtering from RNA data were further retained with the following criteria: having transcriptional start site (TSS) enrichment greater than 0.15; having at least 1,000 read fragments and at least 500 UMI; having fragments overlapping the promoter region ratio of greater than 0.15. Samples were retained only if they exhibited greater than 500 dual omic cells after quality filtering. 　　Clustering analysis was performed using Pagoda2, whereby counts were normalized to the total number per nucleus, batch variations were corrected by scaling expression of each gene to the dataset-wide average. After variance normalization, all 5,526 significantly variant genes were used for principal component analysis (PCA). Clustering was performed at different k values (50, 100, 200, 500) on the basis of the top 50 principal components, with cluster identities determined using the infomap community detection algorithm. The primary cluster resolution (k = 100) was chosen on the basis of the extent of clustering observed. Principal components and cluster annotations were then imported into Seurat (v.4.0.0) and uniform manifold approximation and projection (UMAP) dimensionality reduction was performed using the top 50 principal components identified using Pagoda2. Subsequent analyses were then performed in Seurat. A cluster decision tree was implemented to determine whether a cluster should be merged, split further or labelled as an altered state. For this, differentially expressed genes between clusters were identified for each resolution using the FindAllMarkers function in Seurat (only.pos = TRUE, max.cells.per.ident = 1000, logfc.threshold = 0.25, min.pct = 0.25). Possible altered states were initially defined for clusters with one or more of the following features: low genes detected, a high number of mitochondrial transcripts, a high number of endoplasmic-reticulum-associated transcripts, upregulation of injury markers (CST3, IGFBP7, CLU, FABP1, HAVCR1, TIMP2, LCN2) or enrichment in AKI or CKD samples. Clusters (k = 100) that showed no distinct markers were assessed for altered-state features; if present, then these clusters were tagged as possible altered states, if absent then clusters were merged on the basis of their cluster resolution at k = 200 or 500. If this merging occurred across major classes (epithelial, endothelial, immune, stromal) at higher k values, then these clusters were instead labelled as ambiguous or low quality (including possible multiplets). For k = 100 clusters (non-epithelial only) that did show distinct markers, their k = 50 subclusters were assessed for distinct marker genes; if present, then these clusters were split further. The remaining split and unsplit clusters were then assessed for altered-state features. If present, they were tagged as possible altered states, if absent they were assessed as the final cluster. Annotations of clusters were based on known positive and negative cell type markers11,12,58,59,60 (Supplementary Table 5), the regional distribution of the clusters across the corticomedullary axis and altered state (including cell cycle) features. For separation of EC-DVR from EC-AEA, the combined population was independently clustered using Pagoda2 and clusters associated with medullary sampling were annotated as EC-DVR. For separation of the REN cluster, stromal cells expressing REN were selected on the basis of normalized expression values of greater than 3. Final overall assessment of clustering accuracy was performed using the Single Cell Clustering Assessment Framework (SCCAF v.0.0.10) using the default settings, and compared against that associated with broad cell type classifications (subclass level 1). 　　To overcome the challenge of disparate nomenclature for kidney cell annotations, we leveraged a cross-consortium effort to use the extensive knowledge base from human and rodent single-cell gene expression datasets, as well as the domain expertise from pathologists, biologists, nephrologists and ontologists11,12,22,58,59,60,61 (see also Supplementary Tables 4 and 5 and the HuBMAP ASCT+B Reporter at GitHub (https://hubmapconsortium.github.io/ccf-asct-reporter)). This enabled the adoption of a standardized anatomical and cell type nomenclature for major and minor cell types and their subclasses (Supplementary Table 4), showing distinct and consistent expression profiles of known markers and absence of specific segment markers for some of the cell types (Extended Data Fig. 2a and Supplementary Table 5). The knowledge of the regions dissected and histological composition of snCv3 data further enabled stratification of distinct cortical and outer and inner medullary cell populations (Fig. 2b and Extended Data Fig. 1). The cell type identities and regional locations were confirmed through orthogonal validation using spatial technologies presented here and correlations with existing human or rodent stromal, immune, endothelial and epithelial datasets4,25,58,59,61,62 (Extended Data Fig. 2b–l). 　　Our atlas now includes a higher granularity for the loop of Henle, distal convoluted tubule and collecting duct segments, now resolving three descending thin limb cell types (DTL1, 2, 3); different subpopulations of medullary or cortical thick ascending limb cells (M-TAL/C-TAL); two types of distal convoluted tubule cells (DCT1, 2); intercalated and principal cells of the connecting tubules (CNT-IC and CNT-PC); cortical, outer medullary and inner medullary collecting duct subpopulations (CCD, OMCD, IMCD); and papillary tip epithelial cells abutting the calyx (PapE). Molecular profiles for rare cell types important in homeostasis were annotated, including juxtaglomerular renin-producing granular cells (REN); macula densa cells (MD); and a cell population with enriched Schwann/neuronal (SCI/NEU) genes NRXN1, PLP1 and S100B. Major endothelial cell types were stratified, including endothelial cells of the lymphatics (EC-LYM) and vasa recta (EC-AVR, EC-DVR). Specific stromal and immune cell types were distinguished, including distinct fibroblast populations across the cortico-medullary axis and 12 immune cell types from lymphoid and myeloid lineages. 　　Integration of snCv3 and SNARE RNA data was performed using Seurat (v.4.0.0) using snCv3 as reference. All counts were normalized using sctransform, anchors were identified between datasets based on the snCv3 Pagoda2 principal components. SNARE2 data were then projected onto the snCv3 UMAP structure and snCv3 cell type labels were transferred to SNARE2 using the MapQuery function. Both datasets were then merged and UMAP embeddings were recomputed using the snCv3 projected principal components. Integrated clusters were identified using Pagoda2, with the k-nearest neighbour graph (k = 100) based on the integrated principal components and using the infomap community detection algorithm. The SNARE2 component of the integrated clusters was then annotated to the most overlapping, correlated and/or predicted snCv3 cluster label, with manual inspection of cell type markers used to confirm identities. Integrated clusters that overlapped different classes of cell types were labelled as ambiguous or low-quality clusters. Segregation of EC-AEA, EC-DVR and REN subpopulations was performed as described for snCv3 above. 　　Integration of snCv3 and scCv3 data was performed using Seurat v.4.0.0 with snCv3 as a reference. All counts were normalized using sctransform, anchors were identified between datasets based on the snCv3 Pagoda2 principal components. scCv3 data were then projected onto the snCv3 UMAP structure and snCv3 cell type labels were transferred to scCv3 using the MapQuery function. Both datasets were then merged and UMAP embeddings recomputed using the snCv3 projected principal components. Integrated clusters were identified using Pagoda2, with the k-nearest neighbour graph (k = 100) based on the integrated principal components and using the infomap community detection algorithm. The scCv3 component of the integrated clusters was then annotated to the most overlapping or correlated snCv3 subclass, with manual inspection of cell type markers used to confirm identities. Cell types that could not be accurately resolved (PT-S1/PT-S2) were kept merged. Integrated clusters that overlapped different classes of cell types or that were too ambiguous to annotate were considered to be low quality and were removed from the analysis. Segregation of EC-AEA, EC-DVR and REN subpopulations was performed as described above. 　　As described above, we used the demonstrated Seurat v.4.0.0 integration strategy63 to project query datasets (scCv3, SNARE2 RNA) into the same PCA space as our snCv3 reference. These imputed principal components were used to generate an integrated embedding and integrated clustering through Pagoda2. Query datasets within these integrated clusters were manually annotated on the basis of co-clustering with the reference data, predicted subclass levels and the manual inspection of marker genes. This process was necessary to account for misalignments that occurred for altered states showing more ambiguous marker gene expression profiles, especially for mapping between single-nucleus and single-cell technologies. To assess the accuracy in our alignments, we performed correlation of average expression signatures between the assigned query cell populations and the original reference cell populations (Extended Data Fig. 3e). Although several samples were examined using more than one platform (Supplementary Table 3 and Extended Data Fig. 1a), not all conditions could be covered by all technologies, with AKI/CKD biopsies too limited in size to process with SNARE2 and deeper medullary region capture being less likely for needle biopsies. Despite the differences in patient conditions and regions sampled, we were able to confirm cross-platform sampling with minimal batch contributions for a majority of our subclass (level 3) assignments (77 total). This was demonstrated through integrated bar plots for assay, patient, sex and condition contributions (Extended Data Fig. 3e). The degree to which cells/nuclei between assays were mixed within these subclasses was confirmed using normalized relative entropy weighted by subclass size64, with an average assay entropy across subclasses (covered by more than one technology) of 0.71 and an average patient entropy of 0.71 (out of 1). Mixing within the subclasses was also assessed on the cell embeddings (principal components) using the average silhouette width or ASW (scib.metrics.silhouette_batch function of the scIB package v.1.0.365), with an average score of 0.86 for assays and 0.82 for patients (out of 1). Finally, the average of k-nearest neighbour batch effect test (kBET) score per subclass, computed for all patients using the scib.metrics.kBET function of the scIB package, was 0.49 (out of 1), which is consistent with other integration efforts65. 　　Integration with published data was performed using Seurat v.4.0.0 with snCv3 as a reference. All counts were normalized using sctransform, anchors were identified between datasets on the basis of the snCv3 Pagoda2 principal components. Published data were then projected onto the snCv3 UMAP structure and snCv3 cell type labels were transferred to the published dataset using the MapQuery function. Ref. 12 snDrop-seq data are available at the Gene Expression Omnibus (GEO: GSE121862). Ref. 15 single-nucleus RNA-seq and ref. 14 single-cell RNA-seq count matrices and metadata tables were downloaded from the UCSC Cell Browser (Cell Browser dataset IDs human-kidney-atac and kidney-atlas, respectively). 　　To identify a minimal set of markers that can identify snCv3 clusters and subclasses (subclass.l3), or scCv3 integrated subclasses (subclass.l3), we used the Necessary and Sufficient Forest66 (NSForest v.2; https://github.com/JCVenterInstitute/NSForest/releases/tag/v2.0) software using the default settings. 　　For correlation of RNA expression values between snCv3 and scCv3, or SNARE2, average scaled expression values were generated, and pairwise correlations were performed using variable genes identified from Pagoda2 analysis of snCv3 (top 5,526 genes). For comparison with mouse single-cell RNA-seq data of healthy reference tissue59, raw counts were downloaded from the GEO (GSE129798). For comparison with mouse single-cell RNA-seq from IRI tissue4, raw counts were downloaded from the GEO (GSE139107). For human fibroblast and myofibroblast data25, raw counts were downloaded from Zenodo (https://doi.org/10.5281/zenodo.4059315). For each dataset, raw counts were processed using Seurat: counts for all cell barcodes were scaled by total UMI counts, multiplied by 10,000 and transformed to log space. For comparison with mouse single-cell types of the distal nephron61, the precomputed Seurat object was downloaded from the GEO (GSE150338). For mouse bulk distal segment data61, normalized counts were downloaded from the GEO (GSE150338) and added to the ‘data’ slot in a Seurat object. Bulk-sorted immune cell reference data were obtained using the celldex package67 using the MonacoImmuneData()62 and ImmGenData()67,68 functions and log counts imported into the ‘data’ slot of Seurat. For correlation against these reference datasets, averaged scaled gene expression values for each cluster were calculated (Seurat) using an intersected set of variable genes identified for each dataset (identified using Padoda2 for snCv3 and Seurat for reference datasets). For immune reference correlations, a list of immune-related genes downloaded from ImmPort (https://immport.org) was used instead of the variable genes. Correlations were plotted using the corrplot package (https://github.com/taiyun/corrplot). Immune annotations within our atlas were further confirmed by manual comparison with recently reported data14. 　　For mouse single-nucleus RNA-seq data from IRI tissue4, raw counts were downloaded from the GEO (GSE139107). Integration was performed using Seurat v.4.0.0 with snCv3 as a reference. All counts were normalized using sctransform, anchors were identified between datasets on the basis of the snCv3 Pagoda2 principal components. Mouse data were then projected onto the snCv3 UMAP structure and snCv3 cell type labels were transferred using the MapQuery function. 　　To identify markers associated with altered states (degenerative; adaptive—epithelial or aEpi; adaptive—stromal or aStr; cycling), snCv3 and scCv3 data were independently used to identify differentially expressed genes between reference and corresponding altered states for each subclass level 1 (subclass.l1). To ensure general state-level markers, differentially expressed genes were identified using the FindConservedMarkers function (grouping.var = “condition.l1
”, min.pct = 0.25, max.cells.per.ident = 300) in Seurat. A minimal set of general degenerative conserved genes was identified as enriched (P < 0.05) in the degenerative state of each condition.l1 (reference, AKI and CKD) and in at least 4 out of the 11 (snCv3) or 9 (scCv3) subclass.l1 cell groupings. A minimal set of conserved aEpi genes was identified as enriched (P < 0.05) in the adaptive state of each condition.l1 (reference, AKI and CKD) in both the aPT and aTAL cell populations. This aEpi gene set was then further trimmed to include only those genes that were enriched within the adaptive epithelial population (aPT/aTAL) versus all others using the FindMarkers function and with a minimum P value of 0.05 and average log2-transformed fold change of >0.6
。一组最少的保守ARTAR基因被确定为富集（P< 0.05) in the adaptive state of each condition.l1 (reference, AKI and CKD for snCv2; reference and AKI for scCv3) for stromal cells. To increase representation from MyoF in scCv3 showing a small number of these cells, MyoF-alone enriched genes (average log2-transformed fold change ≥ 0.6; adjusted P < 0.05) were included for the scCv3 gene set. The aStr gene sets were then further trimmed to include only those genes that were enriched within the adaptive stromal population (aFIB and MYOF) compared with all others using the FindMarkers function and with a minimum P value of 0.05 and average log2-transformed fold change of >0.6。一组最小的骑自行车相关基因被确定为富集的基因（调整后的P< 0.05 and average log2-transformed fold change >0.6）在所有相关子类别的循环状态下 。L1细胞组。 　　对于每个单元，从平均归一化计数中计算了改变状态
，细胞外基质和与衰老或SASP相关的基因集的分数，仅使用与平均平均全基因集的最小相关性的基因（参考文献25）（ref
。25）（https://github.com/mahmoudibribrahim/kidnemap）。衰老和SASP基因是从参考文献中获得的 。48（在衰老肾脏中上调的前20个基因）48 ，参考
。69（来自表S3的基因
，Group.Age A），参考。70（图2C的SASP基因）或参考 。71（从表S1（IR表上皮SASP） ，具有正AVE log2比率）71
。 　　为了计算APT和ATAL的基因集富集
，如上所述，在自适应中差异表达的保守基因差异表达。从https://gsea-msigdb.org下载了来自分子特征数据库（MSIGDB）的基因集本体 ，并使用FGSEA72和GAGE73计算了途径富集
，仅保留重要的基因本体学术语（P< 0.05) for both. Redundant pathways were collapsed using the fgsea function collapsePathways and visualized using ggplot. 　　SNARE2 chromatin data were analysed using Signac74 (v.1.1.1). Peak calling was performed using the CallPeaks function and MACS (v.3.0.0a6; https://github.com/macs3-project/MACS) separately for clusters, subclass.l1 and subclass.l3 annotations. Peak regions were then combined and used to generate a peak count matrix using the FeatureMatrix function, then used to create a new assay within the SNARE2 Seurat object using the CreateChromatinAssay function. Gene annotation of the peaks was performed using GetGRangesFromEnsDb(ensdb = EnsDb.Hsapiens.v86). TSS enrichment, nucleosome signal and blacklist fractions were all computed using Signac. Jaspar motifs (JASPAR2020, all vertebrate) were used to generate a motif matrix and motif object that was added to the Seurat object using the AddMotifs function. For motif activity scores, chromVAR75 (v.1.12.0; https://greenleaflab.github.io/chromVAR) was performed using the RunChromVAR function. The chromVAR deviation score matrix was then added to a separate assay slot of the Seurat object. To assess the chromatin data, UMAP embeddings were computed from cis-regulatory topics that were identified through latent Dirichlet allocation using CisTopic76 (v.0.3.0; https://github.com/aertslab/cisTopic) and the runCGSModels function. Only regions accessible in 50 nuclei and nuclei with 200 of these accessible regions were used for cisTopic and downstream analyses. The UMAP coordinates for the remaining nuclei were added to the Seurat object. To ensure high-quality accessible chromatin profiles, only clusters with more than 50 nuclei were retained for downstream analyses (Supplementary Table 7). For joint embedding of SNARE2 accessible chromatin and gene expression, a weighted nearest-neighbour graph was computed using the FindMultiModalNeighbors function (Seurat) based on PCA (RNA) and latent semantic indexing or LSI (accessible chromatin) dimensionality reductions. UMAP dimensionality reduction was performed to visualize the joint embedding. 　　Sites that were differentially accessible for a given cell grouping (subclass) were identified against a selection of background cells with the best matched total peak counts, to best account for technical differences in the total accessibility for each cell. For this, the total peaks in each cell were used for estimation of the distribution of total peaks (depth distribution) for the cells belonging to the test cluster, and 10,000 background cells with a similar depth distribution to the test cluster were randomly selected. Differentially accessible sites (DARs) were then identified as significantly enriched in the positive cells over selected background cells using the CalcDiffAccess function (https://github.com/yanwu2014/chromfunks), where P values were calculated using Fisher’s exact tests on a hypergeometric distribution and adjusted P values (or q values) were calculated using the Benjamini–Hochberg method. For subclass level 2 DARs, VSM/P clusters were merged and the MD was combined with C-TAL before to DAR calling. Subclasses with >100 dars与Q< 0.01 were used for further analysis. Co-accessibility between all peak regions was computed using Cicero77 (v.1.8.1). Sites were then linked to genes on the basis of co-accessibility with promoter regions, occurring within 3,000 bp of a gene’s TSS, using the RegionGeneLinks function (https://github.com/yanwu2014/chromfunks) and the ChIPSeeker package78. DARs associated with markers for each subclass (identified at the subclass.l2 level using snCv3, P < 0.05) and showing q < 0.01 and a log-transformed fold change of >1选择可视化 。为此，将DAR可访问性（峰值计数）进行平均 ，缩放（修剪值至最低为0
，最多为5），并使用Swne Package79的GGHEAT绘图函数可视化
。使用Signac中的超几何测试（FindMotifs函数）计算细胞类型中的基元富集。 　　为了从SNARE2可访问的染色质数据中识别主动转录因子 ，使用发现[所有]标记物通过逻辑回归和使用峰值计数的数量作为潜在变量来评估不同种群之间TFBS的差异活动（或偏差分数） 。仅包括在相应簇，亚类或状态组中检测到的表达的转录因子
。对于PT和TAL簇
，具有差异性活性的TFBS（P< 0.05, average log2-transformed fold change of >0.35）并与在簇之间发现至少2.5％的核（SNARE2）中检测到表达的转录因子相关。从差异性活跃的人的交集中鉴定出常见的APT/ATAL TFBS活性，并在自适应PT和TAL簇中表达 。对于APT和ATAL轨迹模块 ，TFBS显示了模块之间的差异活动（调整后的P< 0.05, average log2-transformed fold change of >在至少2.5％的核/细胞（SNCV3/SCCV3）中检测到的0.35）和表达。对于常见的退化状态TFBS活性
，每个1级子类的参考状态和退化状态之间都确定了差异性活跃的TFBS
。重大退化状态TFBS活动（P< 0.05, average log2-transformed fold change of >0.35）将三个或更多个子类别中的l1缩小到显示在SNCV3/SCCV3的20％以上的退化状态核/细胞中检测到的表达。 　　基于CellChat软件包（V.1.0.0; https：//github.com/sqjin/cellchat）进行配体 - 导体分析 。仅在TAL，亚类2级的TAL ，免疫和基质中（免疫：CDC
，CYCMNP，MAC-M2 ，MAST
，MDC，N ，N
，NCMON，NKT ，NKT，PDC
，PDC，PL ，T和B; Stroma：Myof
，fib，fib ，fib
，dfib，cycmyof and afib）以及用于分泌的ligands的相互作用
。对于TAL轨迹中的单元格 ，我们使用CellChat函数ComputeCommunProb计算了每个模块和其他细胞群之间的细胞间通信概率（有关该方法的详细说明
，请参见参考文献80）。然后，使用自定义的plot_communication函数（代码可用性）根据圆圈图可视化整体扩展的通信概率 。为了进一步了解哪些信号对配体 - 受体（LR）相互作用途径有最大的贡献 ，我们使用plotsigheatMap函数（代码可用性）生成了每种相互作用的途径富集热图
。还使用CellChat软件包中的NetAnalysis_Contribution确定了每种相互作用的重要LR对的贡献。 　　为了将SNARE2细胞类型与肾脏基因组全基因组关联研究（GWAS）特征和疾病联系起来
，我们首先将所有属于同一细胞类型的细胞的二进制峰值可访问性概况概括，以创建一个逐渐划分的峰值峰值可访问性矩阵 。伪库分析可提供更稳定的结果 ，尤其是因为SNARE2可访问性数据可能很少
。为了确保足够的覆盖范围，我们使用了以下修改的子类别2分组：合并了VSM/P簇；MD与C-Tal结合；子类别 <100 DARs with q < 0.01 were excluded. We used g-chromVAR81 (v.0.3.2), an extension of chromVAR for GWAS data, to identify cell types with higher than expected accessibility of genomic regions overlapping GWAS-linked single-nucleotide polymorphisms (SNPs). Running g-chromVAR requires first identifying GWAS-linked SNPs that are more likely to be causal, a process known as fine-mapping. For the chronic kidney failure GWAS traits, we used existing fine-mapped SNPs from the CausalDB database, using the posterior probabilities generated by CAVIARBF82,83. The original GWAS summary statistics files were obtained from an atlas of genetic associations from the UK BioBank84. We manually fine-mapped the CKD, eGFR, blood urea nitrogen and gout traits using the same code that was used to generate the CausalDB database (https://github.com/mulinlab/CAUSALdb-finemapping-pip). The summary statistics for all of these traits are available at the CKDGen Consortium site (https://ckdgen.imbi.uni-freiburg.de/)85,86. We also manually fine-mapped the hypertension trait and the original summary statistics can be found on the EBI GWAS Catalog87. We looked only at causal SNPs with a posterior causal probability of at least 0.05 to ensure that SNPs with low causal probabilities did not cause false-positive signals. Moreover, as g-chromVAR selects a semi-random set of peaks with similar average accessibility and GC content as background peaks, the method has an element of randomness. To ensure stable results, we ran g-chromVAR 20 times and averaged the results. Cluster/trait z-scores were plotted using ggheat (https://github.com/yanwu2014/swne). 　　To link causal SNPs to genes, we used functions outlined in the chromfunks repository (https://github.com/yanwu2014/chromfunks; /R/link_genes.R). This involved the identification of causal peaks for each cell type and trait (minimum peak Z score of 1, minimum peak posterior probability score of 0.025). Sites were then linked to genes on the basis of co-accessibility (Cicero) with promoter regions, occurring within 3,000 bp of a gene’s TSS. Only sites associated with genes detected as expressed in 10% of TAL nuclei/cells (snCv3/scCv3) were included. Motif enrichment within the causal SNP and TAL-associated peaks was performed using the FindMotifs function in Seurat and only motifs for transcription factors expressed in 10% of TAL nuclei/cells (snCv3/scCv3) were included (Supplementary Table 31). For a TAL-associated ESRRB transcription factor subnetwork, peaks were linked to genes using Cicero, then subset to those associated with TAL (C-TAL, M-TAL) marker genes that were identified using the Find[All]Markers function in Seurat for subclass.l3 (P < 0.05). Transcription factors were then linked to gene-associated peaks on the basis of the presence of the motif and correlation of peak and TFBS co-accessibility (chromVAR), using a correlation cut-off of 0.3. Only transcription factors with expression detected within 20% of TAL cells or nuclei (snCv3/scCv3) were included. Eigenvector centralities were then computed using igraph and the transcription-factor-to-gene network was visualized using PlotNetwork in chromfunks. 　　Genes linked with CKDGen consortium GWAS loci for the kidney functional traits eGFR and urinary albumin-creatinine ratio (UACR) were obtained from table S14 of ref. 88. These included the top 500 genes per trait or only those genes also implicated in monogenic glomerular diseases. eQTLs associated with eGFR, systolic blood pressure and general kidney function were obtained from tables S20, S21 and S22 of ref. 89, respectively. Genes associated with the transition from acute to chronic organ injury after ischaemia–reperfusion were obtained from ref. 90 from the following supplementary tables: Acute_Human_Specific (table S3, Human specific column); Acute_Mouse_Overlap (table S3, Shared column); Mid_Acute (table S8, cluster 2 genes); Late_Human_Specific (table S9, Human specific column); Late_Mouse_Overlap (table S9, Shared column); Late_Fibrosis (table S6, positive logFC); Late_Recovery (table S6, negative logFC). Each gene set was assessed for its enrichment within combined snCv3 and scCv3 subclass (level 3) differentially expressed genes (adjusted P < 0.05, log-transformed fold change of >0.25）。使用Fisher的精确测试进行富集
，并使用GGPLOT2对结果进行缩放和可视化 。 　　Neptune91（193名成年患者）和ERCB45（131例）表达数据被用作验证队列，以确定具有不同细胞态基因表达水平的患者之间的重要性
。Neptune（NCT01209000）是一项多中心（21个地点） ，对在第一次临床表明肾脏活检时招募的儿童和成人的前瞻性研究（补充表34）。前瞻性地跟踪研究参与者
，在第一年每4个月进行一次，然后每年待5年 。在每次研究访问中 ，记录了病史
，药物使用和标准的局部实验室测试结果，同时收集血液和尿液样本 ，以测量血清肌酐和尿液蛋白/肌酐比率（UPCR）和EGFR（每1.73 m2每分钟）。末期肾脏疾病（ESKD）定义为透析的启动
，肾脏移植或EGFR <15 ml per min per 1.73 m2 measured at two sequential clinical visits; and the composite end point of kidney functional loss by a combination of ESKD or 40% reduction in eGFR92. Genome-wide transcriptome analysis was performed on the research core obtained at the time of a clinically indicated biopsy using RNA-seq by the University of Michigan Advanced Genomics Core using the Illumina HiSeq2000 system. Read counts were extracted from the fastq files using HTSeq (v.0.11). NEPTUNE mRNA-seq data and clinical data are controlled access data and will be available to researchers on request to NEPTUNE-STUDY@umich.edu. 　　ERCB is the European multicentre study that collects biopsy tissue for gene expression profiling across 28 sites. Transcriptional profiles of biopsies from patients in the ERCB were obtained from the GEO (GSE104954). 　　The gene expression data from the tubulointerstitial compartment of the kidney biopsies from NEPTUNE patients was used to calculate the composite scores for the genes enriched in degenerative, aPT, aTAL and aStr states. The expression of the genes that were uniquely enriched in the cell state (described above) and that were found in both snCv3 and scCv3 were used to calculate the composite cell state score (Supplementary Table 29). As scCv3 did not efficiently identify all stromal cell types, the union of the enriched genes from scCv3 and snCv3 data were used to calculate the aStr cell state score. We also generated a cell state score for the genes that were commonly enriched in aPT and aTAL cells (common). 　　For outcome analyses (40% loss of eGFR or ESKD) in the NEPTUNE cohort, patient profiles were binned according to the degree of cell state score by tertile. Clinical outcomes were available on 193 participants with a total of 30 events. Kaplan–Meier analyses were performed using log-rank tests to determine significance between patients in different tertiles of cell state gene expression. Moreover, for the different cell state scores, multivariable adjusted Cox proportional hazard analyses were performed using five statistical models adjusting for different sets of potential confounding effects given the overall few number of events: (1) age, sex and race; (2) baseline eGFR and UPCR; (3) immunosuppressive treatment and FSGS status; (4) eGFR, UPCR and race self-reported as Black (factors that were associated with outcome in this dataset); and (5) immunosuppressive treatment, eGFR and UPCR (Supplementary Table 30). Note that the adjusted models simply assess whether the association with outcome persists after adjusting for common clinical features (that is, confounding effects), but do not assess for prediction accuracy. 　　In the ERCB cohort, differential expression analyses using multivariable regression modelling were performed between the cell state scores in the disease groups and living donors. Age and sex were used as covariates. The cell state scores for both NEPTUNE and ERCB bulk mRNA transcriptomics data were generated93. In brief, the cell state scores were generated by creating Z scores for each of the cell state gene sets and then using the average Z score as the cell state composite score. These analyses found scores for all adaptive epithelial, but not degenerative, states were significantly higher in the patients with diabetic nephropathy patients compared to that of living donors (Supplementary Table 30). After adjusting for sex and age, both aPT and aTAL were significant when scores from patients with diabetic nephropathy were compared with those of living donors and aPT scores were significant even after correcting for the different disease groups. 　　To find association of patients based on altered-state gene signatures that were used in clinical association analyses (Supplementary Table 30), we performed sample-level clustering. All of the cells from the same patient in snCv3 and scCv3 were aggregated to get pseudo-bulk count matrices on the basis of the associated clinical gene set. The matrices were further normalized by RPKM followed by t-distributed stochastic neighbour embedding (t-SNE) dimensionality reduction. Groups of patients were then identified based on k-means clustering and density-based spatial clustering (DBSCAN) methods in the reduced space. To associate the patient clusters with clinical features, we calculated the distribution of eGFR in each identified group (Code availability). 　　To identify gene sets that best differentiate between AKI and CKD patients in the PT and TAL cell populations, we trained a gene-specific logistic regression model based on the sample-level gene expression. The model was used to assess the predictive power that differentiate patients with AKI and CKD in both snCv3 and scCv3 measured by area under the curve (AUC). The genes with AUC >选择了SNCV3和SCCV3的0.65进行下游分析（补充表32）
。 　　为了鉴定所有细胞类型中AKI和CKD之间差异表达的基因，我们汇总了与每个子类相关的单元（1级） ，以生成每个样品的细胞类型特异性假数
，并使用deseq2方法使用估计PermatimatePercellCelltypede在Cacoa package（V.0.2.0.0.2.0.0.2.0;https://github.com/kharchenkolab/cacoa）。 　　为了找到与疾病进展相关的细胞，我们对PT和TAL细胞进行了轨迹分析 。为了获得准确的伪期定和轨迹估计 ，我们使用Slingshot Package94（v.2.0.0）分别删除了PT和TAL中的退化性细胞群，并分别推断了单核和单个细胞的轨迹
。我们使用Slingshot参数End.Clus指定了正常细胞群作为轨迹推理的终点（PT中的S1 – S3和TAL中的M-TAL）。使用Pagoda2软件包中的Plotemped函数可视化通讯轨迹嵌入 。 　　为了确定基因表达在统计学上是否与推断轨迹显着相关
，我们通过使用公式expi = f（t） + ϵ拟合GAM模型将GAM模型与立方样条回归拟合，将基因的表达建模为估计的假频率的函数 ，其中T是伪的
，F是伪型，而F是立方体spline的功能
。然后将模型与降低的模型expi = f（1） + ϵ进行比较 ，以使用f检验获得p值估计。Benjamin -Hochberg方法用于计算调整后的P值 。为了进一步识别显示AKI和CKD患者之间潜在差异的候选基因
，我们通过使用CKD作为参考来拟合条件平滑的相互作用来扩展基本GAM模型
。 　　为了识别沿估计轨迹的重要基因集的表达模块，我们基于WGCNA软件包95（v.1.70-3）实现的模块检测算法 ，基于使用参数softpower = 10和MinModulesize的平滑基因表达矩阵= 20。总的来说
，我们在TAL细胞中确定了PT和六个模块中的五个不同模块 。对于每个模块中的基因集，我们使用Reactome Online Tool96（https://reactome.org/pathwaybrowser/）进一步执行了途径分析
。通过执行对数比富集测试评估每个模块的临床相关基因集的富集（图6E）。为了预测PT和TAL子类基因的转录因子活性 ，我们使用了Dorothea软件包（V.1.7.2）作为靶标 。从人类和小鼠的证据中策划了多萝西亚的转录因子和转录靶标。根据Viper软件包的Run_Viper函数（v.3.15; https：//bioconductor.org/packages/release/release/bioc/html/html/viper.html）
，根据run_viper函数计算了每种单元类型的转录因子活性评分
。 　　为了识别与每个模块相关的单元，我们开发了一种系统的方法。简而言之 ，对于平滑表达矩阵中的单元格，我们使用PCA进行了降低
，然后进行了卢文集群 。这使沿轨迹识别细胞簇
。对于已识别的细胞簇，我们进行了层次聚类 ，以根据平均基因表达值计算每个模块的相关性
，并通过切割层次结构树将簇与关联模块进一步联系起来。最后，根据其相关簇分配了每个单元的模块标签 。为了将单细胞数据集与单核模块联系起来 ，我们根据PT或TAL单元的关节嵌入进行K-均值聚类
，并根据其最近的K最近邻居（代码可用性）将SCCV3中的单元格分配给模块
。 　　为了进一步研究疾病条件之间无簇的组成变化，我们还进行了细胞密度分析 ，其中我们使用CACOA套件通过2D内核估计来比较AKI和CKD条件之间的归一化细胞密度。计算Z分数以识别显示细胞密度显着差异的区域 。 　　为了验证从人类数据推断的模块的方向
，我们进行了人类和小鼠轨迹的联合比对
。与这两个物种分开推断的个别轨迹（上述弹弓）被对齐，以使用细胞align（https://github.com/shenerorllab/cellalign）进行联合轨迹 ，参数winsz = 0.1 = 0.1
，numpts = 0.1，而numpts = 1000 = 1000。收集组（来自受伤的时间点） 。两种物种之间保守或分歧的基因被指定为重叠/不同的基因集 ，这些基因集经过根据轨迹推断的GAM模型进行了显着性测试（请参见上文）
。 　　使用Velocyto97（V.0.17.17）从Cell Ranger BAM文件中计数剪接和未填充的读数，并使用用Cell Ranger Pipeline预先包装的GRCH38基因注释。从UCSC基因组浏览器下载了重复元素
，并从这些计数中掩盖了 。使用seuratWrappers函数ReadVelocity将相应的织机文件加载到R中，并使用AS.Seurat函数转换为Seurat对象。然后将APT或ATAL轨迹群体子集进行子集 ，并通过基于可能的动力学模型使用SCVELO98（V.0.2.4）计算RNA速度估计
。使用pl.velocity_embedding_stream函数可视化轨迹UMAP上的速度嵌入。计算了基于剪接动力学预测的转录动力学的潜在时间
，并使用PL.HeatMap函数绘制了前300个动力学基因 。为每个TAL模块计算了最高可能性基因，以识别这些状态的潜在驱动因素
。使用PL.Scatter函数生成剪接和未填充或潜在时间散点图。 　　使用Celloracle（v.0.9.1）构建与TAL轨迹模块相关的GRNS ，并在提供的教程中概述了默认参数（https://morris-lab.github.io/celloracle.documentation） 。基本GRN首先是由SNARE2可访问的染色质数据构建的
。使用CICERO（V.1.8.1）鉴定的细胞类型的可接近峰通过其TSS峰链接到基因
，以鉴定可访问的启动子/增强子DNA区域。然后将峰扫描以获取转录因子结合基序（gimme.vertebrate.v5.0）以生成基础GRN 。然后使用SNCV3数据识别特定于TAL状态的GRN
。为了确保使用相关的基因，我们包括在整个ATAL轨迹上变化的基因（补充表17） ，显示了ScvElo分析的动态模块特异性转录（补充表21）
，在TAL细胞（Pagode2）（Pagode2）（Pagode2）或与不同的转录因子活性相关联（补充表20）。通过正规化机器学习回归模型进行GRN推断以修剪无活性（无关紧要或弱）连接，并选择与每个模块或州内的调节连接相关的活动边缘 ，并保留了边缘强度范围内的前2,000个边缘 。然后使用Celloracle绘图函数计算和可视化不同中心度指标的网络评分
。对于硅转录因子扰动分析
，将靶基因表达设置为0，并使用Celloracle的默认参数外推或插值 ，并根据提供的教程将基因表达值推算或插值。如上所述，除了使用包括使用Pagoda2鉴定的可变str基因的基因子集以外
，进行基质GRN构造 。FIB，AFIB ，MYOF的子类3级标记（调整后的P <0.05）;或在至少2.5％的核中检测到表达的转录因子（SNARE2） 并具有在STR亚类之间具有差异性活性的结合位点（P <0.05）。为了确保代表BMP目标SMAD
，还包括SMAD1/5/8
。 　　从新鲜的冷冻肾脏组织中制备组织冰冰球，或者根据逐步协议（https://doi.org/10.17504/protocols.io.bpots.io.b.bvvvv6n69e）。在Novaseq S2 FlowCell（Novaseq 6000）上对库进行测序 ，标准加载浓度为2 nm（读取结构：读取1
，42 bp; index 1，8 bp;读取2 ，60 bp; index 2
，0 bp; index 2，0 bp） 。使用幻灯片seq工具（https://github.com/macoskolab/slideseq-tools/releases/releases/tag/0.1）进行消除 ，基因组对齐和空间匹配
。 　　我们使用giotto100（v.1.0.3）来处理幻灯片seq数据和RCTD101（v.1.2.0）进行细胞类型反卷积。由于仅将参考组织用于幻灯片序列
，因此在反卷积之前，将所有退化状态以及PAPE ，NEU，B和N从SNCV3 Seurat对象中取出 。将seurat对象随机采样至每个级别2（l2）亚型的最多3,000个单元
，并合并了ATL和DTL的1级（L1）子类
。对于每个数据源，即Hubmap或KPMP（补充表2） ，将所有冰球的所有珠子的计数汇总并分层汇编：首先
，除去了少于100 UMI的珠子，并且在少于150个珠子中检测到的基因。Then, the broad l1 subclass annotations in the Seurat object were used to deconvolve all beads (gene_cutoff = 0.0003, gene_cutoff_reg = 0.00035, fc_cutoff = 0.45, fc_cutoff_reg = 0.6, manually adding REN in the RCTD gene list and merging ATL and DTL subtypes as TL).将预测权重标准化为每珠100 。一种细胞类型的相对重量为40％或更高的珠被分类为L1子类
。然后 ，对于每个L1子类
，使用该子类的L2注释以及其余的亚类L1注释（与L1相同的参数）进一步对所有分类珠进行了反vlo。请注意，对于每个L2反卷积 ，使用所有珠子拟合了RCTD中的批量参数
，然后将RCTD对象子集拟合以仅包含L2反卷积的所选珠 。根据分析所需的严格度（图2C，F和扩展数据图4B和其他分析中的30％） ，子类L2处的分类与L1相似
，最大相对权重截止为30％或50％
。对于绘制基因计数，使用命令normalizeGiotto（gobj ，scale -factor = 10000，log_norm = t
，scale_genes = t，scale_cells = f）执行缩放。使用seurat中的点函数绘制标记基因点图（v.4.0.0） 。 　　粗细细胞 - 细胞相互作用可以通过寻找倾向于紧邻的细胞类型来揭示。对于每个冰球 ，我们创建了一个基于Delaunay三角剖分的邻域图
，其中每个珠子都由边缘连接到至少一个相邻的珠子，前提是它们的距离小于50 µm
。对于每对单元类型 ，我们计算它们通过边缘连接的次数。然后
，将单元类型标签随机排列2,500次，以形成连接数量的空分布 。通过此模拟计算出的细胞类型对之间的预期连接数 ，并且将接近度富集定义为观察到的比率比预期的连接频率
。分别使用Giotto的CreateSpatialNetwork和CellProximenrichment命令进行了网络构建和丰富分析。那些最大2级重量小于30％的珠被去除 。我们进一步排除了在组织的视觉边界之外（仅针对名称以“ puck_210113”开头的冰球）之外的虚假珠
，并通过手动指定遵循组织边界的直线
。对于皮质冰球（补充表2），将M-PC ，C-PC和IMCD标签重新标记为PC。m-tal和c-tal为tal；EC-DVR合并为EC-AEA 。除去其他髓样和骑自行车的亚型
。对于髓质冰球
，将M-PC和C-PC重新标记为PC。m-tal和c-tal为tal；所有DTL亚型为DTL；EC-AEA合并为EC-DVR 。除去其他皮质和循环亚型
。 　　为了在扩展数据中生成接近度图。对于皮层和髓质，仅绘制了平均富集值高于0.7和0.8的连接 。 　　制备人肾脏组织并根据森林空间基因表达（10倍基因组学）制造商的方案（CG000240 ，森林组织制备指南）进行成像，如前所述102。肾切除术（n = 6）
，Aki（n = 6）和CKD（n = 11）样品的厚度为10 µm，从Oct-Compound插入的块厚度为10 µm
。这23个样本代表22名参与者 ，因为从CKD的同一参与者那里获得了2个样本（1个皮层和1个髓质）。配备了尼康×10 CFI计划荧光目标的钥匙BZ-X810显微镜用于获取H＆E染色的明亮场镶嵌物
，随后被缝制 。透化12分钟后，将mRNA从染色的组织切片中分离出来
。释放的mRNA被绑定到基准捕获区域中的寡核苷酸。然后将mRNA反转录并进行第二链合成 ，变性
，cDNA扩增和Spriselect cDNA清理（visium cg000239方案），作为文库制备的一部分 。测序是在Illumina Novaseq 6000 System103上进行的
。 　　带有参考基因组GRCH38的太空游侠（V.1.0或更高）用于执行表达分析 ，映射
，计数和聚类。摘要统计数据和质量控制指标包括扩展数据图5和补充表2 。使用SCTRANSFORM104进行了归一化
。最终数据处理是在Seurat（v.3.2.3）中进行的。表达特征图描绘了每个位置中转录表达的强度 。在每个森林样品中，如果剃须刀手动切割边缘 ，则从比较分析中消除了最外层的斑点
。 　　使用Seurat（v.3.2.0）
，使用转移评分系统来评估和绘制每个55 µm斑点产生的签名的比例。转移分数反映了每个位置的签名与给定SNCV3子类（2级）之间的概率 。最高的概率转移分数具有在每个空间转录组斑点图中映射的最高比例。对于细胞类型特征图（图2G和3F以及扩展数据图7i），映射了亚类2个单元格转移评分 ，以传达每个位置下面的签名的比例
。对于细胞状态特征图（图3B），而不是映射亚类2个细胞类型
，在样本中的所有斑点上映射了SNCV3中注释的AEPI细胞状态。我们总结了所有样品所有斑点中与6个单元格状态中每个细胞状态相对应的所有细胞类型产生的特征的比例（图3A） 。使用Fisher的精确测试比较了整个肾切除术，AKI和CKD样品的细胞态比例
。 　　在所有样品中的所有斑点中 ，我们根据图3A所计算的上皮细胞态特征的最高比例将斑点分为健康
，适应性或退化性上皮细胞态。对于基质或免疫细胞类型共定位，我们首先选择了所有23个样品中每种细胞类型的非零传递评分的斑点 。如果其基质或免疫转移评分超过其在所有选定的斑点上的平均转移评分 ，则认为基质或免疫细胞特征的存在被认为与上皮细胞共定位
。计算了具有基质或免疫细胞特征的健康
，适应性和退化性上皮细胞态之间的共定位。 　　为了比较与健康，适应性和退化性细胞态相关的标记基因表达（图3D） ，我们首先将AKI和CKD样品的斑点子分类为5个主要细胞类型中的1个：POD
，PT，PT ，TAL
，TAL，CD或FIB 。对于PT ，TAL和成纤维细胞，如果其签名的最高比例（1级映射）对应于这些细胞类型之一
，则选择一个点
。对于CD子集，如果PC和IC的1级映射比例的总和最大 ，则选择一个点。POD斑点是由1级POD标签产生的至少20％的特征来定义的 。一旦选择了PT
，TAL，成纤维细胞 ，CD和POD斑点的子集
，每个斑点就基于图3A中计算的最高比例的细胞态特征，将每个点进一步分为健康 ，适应性或退化细胞态
。对于POD
，鉴于POD标记的损失与DKD样品内观察到的EC-GC特征的增益有关，EC-GC特征的存在被认为是退化等效的。 　　为了检查与TAL上皮细胞共定位的细胞类型的多样性 ，我们选择了具有超过20％TAL特征的斑点
，其中最高比例的特征来自1级TAL映射 。使用Seurat聚类方法，在Seurat标签转移评分代替基因表达后 ，将选定的斑点重新聚集。由TAL细胞类型和状态引起的具有相似比例的斑点，基质细胞和免疫细胞聚集成13个壁ni
。在23个肾脏样品上映射了壁ni
，并确定每个小裂市场中的标记基因表达。为了描述每种单元类型的相对比例，首先在每个小众市场中计算传递评分平均值 。接下来 ，在壁ches上计算了每种单元类型的Z分数
。 　　为了确定是否适当地映射了74个SNCV3亚类（2级）
，将每个位置的签名比例与由六个由六个无监督的群集组成的组织学验证的集进行了比较（扩展的数据5A）。这六个无监督的簇（肾小球，PT ，Henle的环
，远端曲折的小管，连接小管和收集导管以及间质）的总体比对在H＆E图像中的基础组织病理结构的总比对为97.6％ 。在每个样品中 ，皮质和髓质区域是通过组织学评估定义的
，包括肾小球的存在
。在23个样品中，只有18个样品由仅皮层组成 ，4个样品是皮质和髓质的组合
，而1个样品完全是髓质。 　　KPMP中注册的AKI或CKD患者在OCT中冷冻的肾脏活检核心用于无标签成像，然后在协议中概述的（https://doi.org/10.17504/protococols.io.9.9avh2e）中概述了多重荧光大规模3D成像（https://doi.org/10.17504/prototocols.io.962e6） ，并描述了最近的2.27 。使用低温恒温器将冷冻活检切为50 µm的厚度，然后立即在4％新鲜的多甲醛（PFA）中固定24小时
，然后在0.25％PFA中存储在4°C下
。 　　成像的第一步是使用多光子显微镜的无标签成像组成，以收集安装在非硬化安装介质中的无标记组织的自动荧光和第二个谐波图像。使用安装在直立DM6000显微镜上的Leica SP8共聚焦扫描头进行成像 。为了在亚微米分辨率下对组织进行大规模成像 ，使用高功率
，高数字的孔径物镜收集左侧瓷砖扫描功能
。然后使用Leica LASX软件（v.3.5）将这些组件量缝合到整个样品的单个图​​像量中。设置了扫描仪的变焦和聚焦电机控制，以横向提供0.5×0.5μm的体素尺寸和轴向1μm 。 　　在荧光显微镜之前的组织标记在磷酸盐缓冲盐水（PBS）中洗涤之前 ，并用具有0.1％Triton X-100（MP生物医学）和10％正常驴血清（Jackson Immuno Research）的PBS封闭。首先在室温下施用间接免疫荧光的抗体8-16 h
，然后用0.1％Triton X-100洗涤PBS
。接下来发生了与继发抗体的孵育周期，然后洗涤 ，最后应用直接标记的抗体。除了使用DAPI（补充表35）外
，还使用了针对管状细胞和结构（尿道通道蛋白-1，F-肌动蛋白）和免疫细胞（骨髓氧化酶 ，CD68
，CD3，SIGLEC 8）的抗体抗体 。在最后的洗涤周期后 ，将组织安装在延长玻璃（Thermo Fisher Scientific）中
。 　　使用LEICA×20/0.75 NA多污染物（针对油浸入调节）进行共聚焦显微镜检查，并通过固态激光发射依次提供
，其激光发射的激光线在405 nm，488 nm ，488 nm
，552 nm和635 nm中。Images in 16 channels (emission spectra collected by PMT detectors adjusted for the following ranges: 410–430 nm, 430–450 nm, 450–470 nm, 470–490 nm, 500–509 nm, 510–519 nm, 520–530 nm, 530–540 nm, 570–590 nm, 590–610 nm,为3D MOSAIC的每个面板的每个焦平面收集610–630 nm，631–651 nm ，643–664 nm
，664–685 nm，685–706 nm和706–726 nm） 。然后将所得的16通道图像在光谱上进行横向vall（通过使用Leica LASX线性拆卸软件进行线性混合） ，以区分样品中的八个荧光探针
。先前描述了线性拧紧的验证27。 　　来自皮质或髓质的人肾脏组织样品固定在4％PFA中
，冷冻保存在30％的蔗糖中，并在OCT冷冻层中冷冻 ，并将其切成5μm的切片 。The sections were post-fixed with 4% PFA for 15 min at room temperature, blocked in blocking buffer (1% BSA, 0.2% skimmed milk, 0.3% Triton X-100 in 1× PBS) for 30 min at room temperature and then immunofluorescence microscopy was performed, first by overnight incubation at 4 °C with primary antibodies, followed by labelling with secondary antibodies.主要抗体包括NRXN-1β
，TuJ1，III和III ，Synapsin-1，NPSH-1
，SLC14A2，UMOD ，CD31
，CD31，CD34 ，CD34
，CD11B，PROM1 ，KIM1
，VCAM1，VCAM1 ，AQP1
，AQP1，AQP2 ，AQP2，CD45和S100（补充表36）
。洗涤后
，使用Alexa-488偶联的山羊抗小鼠IgG，Cy3偶联的山羊抗兔IgG或Cy5偶联的驴子抗鸡蛋IgG使用Alexa-488偶联的山羊抗小鼠IgG进行标记 ，在室温下1小时。洗涤后
，将切片用DAPI对核染色进行了反染色 。用尼康80i C1共聚焦显微镜获取图像
。 　　从福尔马林固定，石蜡填充（FFPE）块中用3μM切除人肾组织。使用RNASCOPE探针HS-PROM1（311261 ，高级细胞诊断）
，HS-CST3（528181，高级细胞诊断）和HS-EGF（605771 ，晚期诊断诊断率）（361261
，高级诊断）（361261，高级诊断）（605771 ，高级诊断率）（3222323232323232323232）制造商的协议 。RNASCOPE阳性对照探针HS-UBC（310041
，晚期细胞诊断）用作阳性对照。基于辣根 - 过氧酶的信号扩增系统（322310，RNASCOPE 2.0 HD检测试剂盒 ，晚期细胞诊断）用于与目标探针杂交，然后进行DAB染色
。然后用血久毒素（3535-16
，Ricca Chemical Company）对该切片进行染色。阳性染色是由棕色点确定的 。补液后，将切片浸入周期性酸溶液（0.5％ ，P7875
，Sigma-Aldrich）中5分钟，在三种变化的蒸馏水变化中冲洗 ，与Schiff试剂（3952016
，Sigma-Aldrich）孵育15分钟，然后在Runnion Tap Water中冲洗5分钟
。将核与甲莫莫妥蛋白2（220-102 ，Thermo Fisher Scientific）染色2分钟。然后将切片冲洗在水中
，在酒精中脱水，用二甲苯清除 ，最后用二甲体xyl安装（8312-4
，Thermo Fisher Scientific） 。 　　使用体积组织勘探和分析（VTEA）软件（V.1.0A-R9）进行组织细胞仪和分析
。VTEA是一个3D图像处理工作空间，是作为ImageJ105的插件开发的。VTEA的版本包括细胞的监督和无监督标签以及合并基于空间和功能的门控策略 ，可在GitHub（https://github.com/icbm-iupui/icbm-iupui/volumetric-tissue-tissue-ecploration-analyalis-analyalis-analanalyaliss）上使用 。在此分析管道中，使用强度阈值和内置的VTEA和ImageJ内置的连接组件分割对每个核进行分割
。每个被调查的核成为一个细胞的替代物
，可以注册核周围或核内的位置和标记物染色。这些捕获的信息可用于使用标记强度或空间特征对单元进行分类，使用散点图显示 ，该显示器可以实现各种门控策略和统计分析
，包括导出为所有分段单元格的.CSV文件和相关特征106 。补充表37中总结了根据标记强度分类的细胞
。将门控单元直接映射到图像量中，从而可以立即验证门。此外 ，可以执行图像上的直接门控
，这可以将所选利益区域内的所有单元格回到散点图上的数据显示 。因此，也可以根据图像体积中利益区域（ROI）的直接注释来执行细胞分类。带注释的ROI通过像素的3个专家中的3个专家中的3个分别对每个活检标本进行注释的专家之间确定
。 　　使用组织细胞仪 ，根据以下特征定义了14种细胞类别：（1）PT细胞：皮质中的AQP1+细胞±刷边框染色。（2）C-TAL细胞：皮质中的UMOD+细胞 。（3）基于形态学和F-肌动蛋白染色的ROI
，肾小球细胞（包含POD，肾小球内皮和肾小球细胞）
。（4）皮质大和中容器细胞：基于形态和F-肌动蛋白染色的注释ROI。（5）皮质远端肾单位细胞（远端小管（CD） ，连接小管（CNT）和收集管道（C-CD）或皮质远端肾脏）：aqp1-umod-和基于皮质中独特形态的带注释的ROI 。（6）M-TAL细胞：髓质中的UMOD+细胞
。（7）DTL：髓质中的AQP1+细胞。（8）髓质收集管道：基于髓质中独特的形态的AQP1- umod-和带注释的ROI 。（9）髓质中的血管束：基于延髓和F-肌动蛋白染色的注释的ROI
。（10）中性粒细胞：MPO+细胞。（11）激活的巨噬细胞：MPO -CD68+细胞 。（12）T细胞：CD3+细胞
。（13）改变区域的细胞：基于（无法识别的）管状形态
，扩展的间质，纤维化增加（通过第二次谐波产生成像）和细胞浸润的注释的ROI。（14）未确定：无法根据上述标准进行分类 。 　　使用这种方法 ，将1,540,563个细胞从本分析中使用的所有活检中标记。 　　使用VTEA计算每个活检中的每个细胞，半径为25μM（X和Y中的50素体素
，Z中的25素体素）计算3D邻域
。我们认为，在3D方法中 ，最大的可测量邻域/利基受到成像（z维度）的50μm厚度的限制。因此
，根据Nyquist采样，所使用的半径约为25μm ，这与先前的方法107,108,109一致 。对于每个3D邻域
，使用VTEA来计算特征：每个邻域每个标记的单元格的分数和总和
。活检样本作为.CSV文件，将社区 ，3D的位置及其功能列表出口。 　　在VTEA中生成了通过活检样本的社区CSV文件
，并进口到R（v.4.0.4）中，以删除每个标记的单元格和单型邻域的总和 。这些特征是通过Z标准化缩放的 ，用于Louvain社区检测（R packages fnn（v.1.1.3）和igraph（v.1.2.6））和T-SNE歧管投影（R package rtsne（v.0.15））
。为了了解社区内的相互作用
，对邻域进行了成对的相互作用，并将其绘制在和弦图上（R包：Circlize（V.0.4.12）） ，并计算并绘制了Pearson的相关系数（R packages hmisc（v.4.5.0）（v.4.5.0）和corrplot（v.0.84））。选择了至少一个带有特定标签的细胞的邻里子类，并将其绘制为网络图（R package igraph（v.1.2.6））
，在CD3中具有边缘，并且更改的邻域的更改为所有其他子类别的40％ ，以促进可视化 。所有脚本均作为注释的rstudio笔记本文件（.RMD）提供
。 　　使用seurat生成了SNCV3
，SCCV3，SNARE2和visuium数据的UMAP ，功能
，点和小提琴图。使用Corrplot软件包生成相关图 。使用Signac进行了基因组覆盖图
。与Cirize，GGPLOT2和IGRAPH一起在R中生成了3D细胞仪和邻域分析的图。 　　有关研究设计的更多信息可在与本文有关的自然投资组合报告摘要中获得 。