单细胞分辨率下对细胞因子的免疫反应词典

　　所研究的细胞因子及其替代名称的列表如图1D和补充表1所示。我们选择了86个代表大多数主要细胞因子家族成员的细胞因子，包括以下家庭：IL-1（IL-1α，IL-1β，IL-1β，IL-1R），IL-1RA，IL-18，IL-18，IL-33，IL-33，IL-33，IL-36α和IL-36α和IL-36α和IL-36RRA）；常见的γ链/IL-13/TSLP（IL-2，IL-4，IL-13，IL-7，TSLP，IL-9，IL-9，IL-15和IL-21）；常见的β链（GM-CSF，IL-3和IL-5）；IL-6/IL-12（IL-6，IL-11，IL-27，IL-27，IL-30，IL-31，LIF，OSM，CT-1，NP，IL-12，IL-12，IL-12，IL-23和IL-Y）;IL-10（IL-10，IL-19，IL-20，IL-20，IL-22和IL-24）;IL-17（IL-17A，IL-17B，IL-17C，IL-17D，IL-17E，IL-17E，IL-17F）;干扰素（I型：IFNα1，IFNβ，IFNε和IFNκ； II型：IFNγ；和III型：III：IFNλ2）;TNF（LTα1/β2，LTα2/β1，TNF，OX40L，CD40L，FASL，CD27L，CD27L，CD30L，4-1BBL，TRAIL，TRAIL，RANKL，RANKL，TWEAK，TWEAK，APRIN，BAFF，BAFF，LIGHT，LIGHT，TL1A和GITRL）;补体（C3A和C5A）；生长因子（FLT3L，IL-34，M-CSF，G-CSF，SCF，EGF，VEGF，FGFβ，HGF和IGF-1）。代表性细胞因子（TGFβ1，GDNF，PSPN），催乳素（PRL），瘦素，脂联蛋白（ADIPOQ），抵抗素（ADSF），Noggin，Decorin，Decorin和Dolmbopoietin（TPO））来自其他蛋白质家族。 　　每个重组小鼠细胞因子均从至少两个单独的订单中获得。内毒素水平是 <0.1 ng µg–1 of protein for every cytokine per the information from the vendors (Peprotech and R&D). To preserve cytokine activities, carrier-free cytokines were freshly reconstituted according to the manufacturer’s instructions, stored at 4 °C in sterile conditions and used within 28 h after reconstitution. For each cytokine, 5 μg in 100 μl sterile PBS was injected into each animal. Wild-type female C57BL/6 mice were purchased from the Jackson Laboratory and used in studies as 11–15-week-old young adults after resting for at least 1 week in the facility. Mice were maintained on a 12-h light–dark cycle at room temperature (21 ± 2 °C) and 40 ± 10% humidity. Cytokines were injected under the skin (50% subcutaneous, 50% intradermal) bilaterally in the abdominal flank of each mouse. Bilateral skin-draining inguinal lymph nodes were collected 4 h after injection at 6:00–8:00 and pooled for downstream processing. For each of the 86 cytokines, replicate experiments were performed in three independent C57BL/6 mice to ensure reproducibility. As a control, PBS alone was injected into mice for each experimental batch, totalling 14 PBS-injected mice. All experiments were reviewed and approved by the Broad Institute’s Institutional Animal Care and Use Committee. 　　All samples were processed using an optimized experimental pipeline to ensure quality. In particular, batch effects that arise from experiments performed on different days are known to be a major source of artefact in transcriptomic studies. Therefore, batch-to-batch consistency was strictly experimentally ensured and then computationally verified. Specifically, the mice were ordered from the same batch and housed in the same environment. Animals were randomly allocated to the experimental groups. Lymph nodes were collected at 6:00–8:00 in all experiments to exclude the impact of circadian clocks on transcriptomic profiles. Samples were processed fresh in every experiment and were kept on ice during processing whenever possible. The same researchers performed the same steps of the sample processing and sequencing pipeline following the same, highly optimized procedures. The investigators performing animal experiments and RNA sequencing were blinded from each other during data collection. The number of batches was minimized whenever possible. The three replicated mice for each cytokine were processed in different batches to ensure that batch effects, if any, would not influence biological interpretations. All samples were sequenced on two sequencing runs, with the first sequencing run containing the first set of replicates and the second containing the second and third set of replicates. PBS controls were included in every batch to ensure comparability, and transcriptomic profiles of PBS samples from different batches were computationally compared to verify batch-to-batch consistency (Extended Data Fig. 2a). In brief, Euclidean distances were calculated for each pair of PBS-treated cells of the same cell type based on the entire transcriptome to ensure that the within-batch distances and between-batch distances were comparable. 　　An optimized pipeline for viable cell recovery and more balanced cell-type representation was used to process lymph nodes for scRNA-seq. Lymph nodes were enzymatically digested using a protocol that maximizes the recovery of myeloid and stromal cells while maintaining high viability40. In brief, lymph nodes were placed in RPMI with collagenase IV, dispase and Dnase I at 37 °C, and cells were collected once they were detached. The cells were then immediately placed on ice and washed with PBS supplemented with 2 mM EDTA and 0.5% biotin-free BSA, then filtered through a 70 µm cell strainer. Cells were incubated with Fc blocking antibodies 4 °C, then with a biotinylated anti-CD3 and anti-CD19 antibody cocktail. Antibodies were used at a dilution of 1:100. Streptavidin microbeads were then added and the cells were magnetically sorted using MACS MS columns according to the manufacturer’s protocol (Miltenyi Biotec). After cell sorting, a small fraction of the CD3+ or CD19+ cells was pooled with CD3–CD19– cells for more balanced representation of all cell types and proceeded immediately to scRNA-seq. 　　Cell hashing was used to combine multiple samples into the same single-cell emulsion channel41. The mouse cells obtained from different stimulation conditions were stained with TotalSeq antibodies (BioLegend anti-mouse hashtags 1–8; used at 1:100 dilution), washed 5 times at 4 °C and pooled in PBS with 0.04% BSA according to the manufacturer’s protocol. Next, 55,000 cells were loaded onto a 10x Genomics Chromium instrument (10x Genomics) according to the manufacturer’s instructions. The scRNA-seq libraries were processed using a Chromium Single Cell 3′ Library & Gel Bead v3 kit (10x Genomics) with modifications for generating hashtag libraries41. Quality control for amplified cDNA libraries and final sequencing libraries was performed using a Bioanalyzer High Sensitivity DNA kit (Agilent). scRNA-seq and hashing libraries were normalized to 4 nM concentration and pooled. The pooled libraries were paired-end sequenced on a NovaSeq S4 platform targeting an average sequencing depth of 20,000 reads per cell for gene expression libraries, and on a NovaSeq S4 or SP platform targeting 5,000 reads per cell for hashtag libraries. 　　The raw bcl sequencing data were processed using the CellRanger (v.3.0) Gene Expression pipeline (10x Genomics), including demultiplexing and alignment. Sequencing reads were aligned to the mm10 mouse reference genome, and transcriptomic count matrices were assembled. Hashtag library FASTQ files were processed using the CITE-seq-Count tool (v.1.4.3; github.com/Hoohm/CITE-seq-Count). Gene expression and hashtag were matched using the MULTIseqDemux function of the Seurat R package (v.4.1)42. Cells with multiple hashtags were considered multiplets (for example, doublets or triplets) and were excluded from further analysis. The Seurat R pipeline was used to perform quality control to include only cells with >500个基因，> 1,000个唯一的分子标识符和 <10% mitochondrial gene content. The expression matrix was globally scaled by normalizing the gene expression measurements by the total expression per cell. The resulting values were multiplied by a scale factor of 10,000 and natural log-transformed. 　　For the initial global analysis of all cells, the top 3,000 variable genes were selected for dimensionality reduction analysis. Principal component analysis (PCA) was then used to denoise and to find a lower-dimensional representation of the data. The top 75 principal components (PCs) were used for global clustering and for visualization using a t-SNE map43. Clusters were identified using the Louvain clustering algorithm. This step resulted in a total of 61 non-singleton global (level 1) clusters (Extended Data Fig. 1d). We removed potential multiplets by removing the cells with the top 2% gene counts in each cluster. As different cell types have variations in the numbers of genes detected on average, this step was done at the cluster level rather than for all data. For each level 1 cluster of cells, we then performed another round of clustering (level 2) to further verify the identity of each cluster and to remove potential doublets. This step resulted in a total of 183 global level 2 clusters. The cell-type identity of each level 2 cluster was assigned on the basis of the expression of 115 known marker genes (Supplementary Table 2). Clusters enriched for marker genes of multiple cell types were considered multiplets and removed. The top DEGs between each cell type and others are listed in Supplementary Table 2. 　　A gene expression vector for each biological replicate was created for each cytokine stimulation condition in a given cell type by taking the difference between the average expression vector of cytokine-treated cells and the average expression vector of PBS-treated cells. Genes were included if they were significantly differentially expressed (FDR < 0.05 and absolute log2(FC) > 0.25) compared with PBS controls and were expressed in >用于上调的基因上调的细胞因子治疗细胞中有10％的细胞或> 10％的PBS处理的细胞用于下调的基因。RP，RPL，线粒体基因和未标记的基因被排除在外。然后为这些向量计算成对的皮尔逊相关系数（扩展数据图2B）。 　　为了确定在排水淋巴结中是否可以通过每种细胞类型访问注入的细胞因子，我们检查了用IFNα1或IFNβ处理的细胞中的ISG表达水平（IFNα1/IFNβ）或每种细胞类型的PBS。鉴于在多种细胞类型中对抗病毒程序的强烈诱导，选择了I型干扰素进行此分析。从每种条件中最多采样了100个细胞。通过将每个单元格中ISG的归一化表达式求和得出ISG评分。然后使用这些分数来预测每个细胞是否用IFNα1/IFNβ或PBS处理，并且预测的准确性表示为接收器工作特性曲线（扩展数据图2C）。ISG是从MSIGDB Hallmark基因集获得的。 　　进行DEG的分析以鉴定细胞簇或细胞因子反应基因的标记基因。使用归一化基因表达值的两侧Wilcoxon秩和测试在两组细胞之间进行了DEG的分析。然后调整从测试中获得的P值（Bonferroni或FDR）以解决多个测试。如果基因具有FDR，则将基因视为细胞因子特征< 0.05 and absolute log2(FC) >0.25在> 10％的细胞因子细胞中的细胞因子治疗和PBS处理的细胞之间，用于上调的基因上调的基因上调或> 10％的PBS处理的细胞中下调的基因，至少满足了三只小鼠的FC阈值（以减轻潜在单物体影响力的影响）。RP，RPL，线粒体基因和未标记的基因被排除在特征之外。补充表3中所有主要细胞类型列出了细胞类型的细胞因子特异性细胞因子特异性特异性特异性特征。 　　我们构建了一个全局参考图，该图量化了每种细胞类型中每种细胞因子诱导的总体基因表达变化（图1D）。该地图考虑了两个指标：整个转录组的DEG数量和变化的幅度。DEG的数量是每个细胞因子特征中基因的总数。细胞因子诱导的鉴别表达的总体幅度被计算为细胞因子处理细胞和PBS处理的细胞之间的欧几里得距离。该值归一化为0（低）至100（高）的比例。为了减少异常值在归一化过程中的影响，应用了胜利化，以使高于第95个百分位数的值在归一化之前的第95个百分位数代替。对每种细胞因子处理条件的最多100个细胞进行了每种细胞类型的采样，以计算幅度。每种细胞类型都使用了一个独特的颜色坡道，以强调细胞类型具有不同的特性（例如，平均表达的不同基因数量不同），并进行了独立分析。该分析中包括与五个或更多细胞采样的细胞因子 - 细胞类型组合。 　　GP分析用于识别细胞类型跨细胞类型的每种细胞因子治疗的共同调节基因（图1和补充图1）或在所有治疗条件下的每种细胞类型（扩展数据图5i，J，J，J，J，J，J，J，J，J和8i，J和8i，J和8i，J和补充图2I，J，J，J，J，J，J，3I，J，3I，J，4I，J，4I，J.，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，J，则那么j，5i，J。10i，j和11i，j）。使用R package NMFN（V.2.0）使用非阴性矩阵分解（NMF）算法构建GP。我们去除了与组织分离45和细胞周期以及线粒体基因，RPS和RPL，未标记基因，全球过多的基因以及以少于十个细胞表达的基因相关的基因。 　　在以细胞因子为中心的分析中，使用NMF来鉴定细胞型特异性GP，以响应每种细胞因子。用指定的细胞因子或PBS处理的细胞用于NMF，每种细胞类型最多100个细胞。在此分析中，每种细胞因子分别运行NMF，除了将IFNα1和IFNβ一起处理在一起，并且在图2C和补充表5中一起处理IL-1α和IL-1β。我们鉴定了每一种细胞因子处理40 GPS，其中一些主要与细胞型识别率相对应与细胞型的刺激性相关，并主要对应于细胞型的刺激性。为了定量识别与细胞因子刺激的细胞反应相对应的GPS，使用两侧Wilcoxon Rank-sum检验来鉴定具有细胞因子处理的细胞和PBS处理细胞之间显着差异的重量的GPS。显示了在任何细胞类型中显示出明显上调的GP。每个GP重量最高的30个基因用于使用MSIGDB数据库的标志性基因集上的簇式Profiler（v.4.2.1）46鉴定富集的生物学过程。文本中突出显示生物学意义的基因满足了两个标准：（1）它们是响应细胞因子的GP显着上调的30个最高加权基因之一；（2）在细胞因子特征的DEG分析中，基因在响应细胞因子的响应中也单独显示出明显的上调。 　　在以细胞类型为中心的分析中，使用NMF来分析所有细胞因子治疗条件上每种细胞类型的GPS，以鉴定诱导相似细胞过程的细胞因子。我们使用扩展数据图5-8和补充图中的热图确定了每种细胞类型的十个GP，并对每个单元格的GPS和子观察者之间的关系可视化。2–11。在扩展数据2中显示了每个GP权重最高的顶部基因。5J，6J，7J和8J，补充无花果。2J，3J，4J，5J，6J，7J，8J，9J，10J和11J以及补充表8。 　　IFNγ诱导的基因表达特征用于推断细胞因子诱导的IFNγ的细胞反应水平。每种细胞类型的IFNγ特征评分是通过相对于PBS处理的细胞（FDR调整后的P）构建的。< 0.001 and log2(FC) >1）在相应的单元格类型中。计算每种细胞因子治疗的签名评分和FDR调整后的P值的Log2（FC），并在扩展数据中显示了图4E。 　　为了鉴定细胞因子诱导的细胞极化状态，对每种细胞类型的细胞进行了亚聚集。对于异质细胞类型（例如，巨噬细胞，MIGDC和γδT细胞），分析了最丰富的均质子集，以鉴定在极化状态分析中鉴定细胞因子诱导的态而不是重新衍生细胞子集。我们使用PC根据区分基因（与PBS处理的细胞相比，在任何细胞因子处理的细胞中定义为具有较大的绝对log2（FC）（FC）（FC）（FC）（FC）（FC）（FC）（FC）较大的基因（FC）（FC）（FC）（FC）（FC为0.75至1.5）（在0.75和1.5之间）。我们去除了与组织解离45和细胞周期以及线粒体基因以及RPS和RPL相关的基因。然后，我们使用UMAP48执行了PCA并观察了细胞。计算每个细胞因子或PBS对照的细胞的比例。根据两个标准确定了主要的极化状态：（1）具有显着调整的P的细胞簇< 0.01) more than the expected number of cytokine-stimulated cells using a hypergeometric test; and (2) manual verification of biological relevance of the highly expressed genes or GPs in the subcluster and cytokines inducing the changes. To find discriminating markers and biological functions of each state, we analysed DEGs and co-regulated GPs per state relative to all other cells for the cell type. DEGs were identified using the two-sided Wilcoxon rank-sum test between each polarization state and other cells of the same cell type. The significantly overexpressed genes with the largest log2(FC) are shown. The most strongly polarized states are summarized in Fig. 3. The complete landscape, including less-strongly polarized states, in each cell type can be found in Extended Data Figs. 5–8 and Supplementary Figs. 2–11. We compared the polarization states by calculating the pairwise Pearson correlation coefficients between the gene expression profiles of each polarization state after subtracting the profiles of PBS-treated cells of the same cell type to remove cell-type-specific gene expression. These results are displayed in Extended Data Figs. 5b, 6b, 7b and 8b and Supplementary Figs. 2b, 3b, 4b, 5b, 6b, 7b, 8b, 9b, 10b and 11b. 　　We assigned a unique identifier to each polarization state using the following convention: ‘-’. When applicable, the letters a–d were reserved for type I interferon, type II interferon, IL-1α and IL-1β, and TNF, respectively, which are cytokines that induce polarization states across a large number of cell types. 　　To gain a global view of the 66 polarization states across immune cell types defined in Fig. 3, we used Jaccard similarity index to evaluate similarity between each pair of cell states (Extended Data Fig. 9). The gene expression profile of each polarization state was compared with PBS-treated cells of the same cell type to remove cell-type-specific gene expression. The genes with an absolute log2(FC) >与PBS处理的细胞相比，0.5用于计算Jaccard相似性评分。分别计算上调和下调的基因。使用欧几里得距离上的平均链接方法在层次上进行层次聚类，以识别相似极化状态的组。为了可视化与其他状态相似性较低的独特极化状态，使用强制定向网络说明了相同的结果，较高的圆形大小表示更独特的状态，该状态是根据平均JACCARD相似性值与其他状态的平均值相似性值计算得出的。 　　为了鉴定富含IL-18处理的NK细胞的生物学过程，通过从IL-18处理的NK细胞中的PBS处理的NK细胞中减去PBS处理的NK细胞的平均基因表达值来计算预先排名的基因列表。基因集富集分析是使用基因本体生物学过程基因组的簇源（V.4.2.1）进行的。基因组具有FDR调整后的P< 0.1 are shown. As a comparison, representative cytokines from other NK cell polarization states were analysed using the same method. 　　A map of cell-type-specific production of cytokines was derived from our dataset. Cytokine genes expressed in at least 50 cells were included in the cytokine expression heatmap. The cells were obtained from all conditions (PBS or cytokine treated) to provide a map of cytokine expression under all unstimulated or cytokine stimulation conditions (that is, to account for induced expression). The gene expression level was then normalized relative to the cell type with the maximum expression level (whereby the maximum level is capped at 1 expression unit before normalization) to account for the variation in the number of transcripts produced or detected for each cytokine. A cytokine was considered expressed in a cell type if more than 0.1 normalized expression units were detected. 　　The cytokine–receptor expression map was constructed using the same approach. This included signalling receptors, decoy receptors and receptors that form complexes with cytokines. A list of genes encoding known functional receptors for the 86 studied cytokines are listed in Supplementary Table 1. The cytokine expression map and the cytokine–receptor expression map are shown in Fig. 4a, Extended Data Fig. 10c and Supplementary Table 9. 　　A cell–cell interactome network was constructed to chart available cytokine-mediated cell–cell communication channels. The network was constructed such that the source and sink nodes are cell types and intermediate nodes are cytokines. The paths between source cell-type nodes and sink cell-type nodes through cytokine nodes were established on the basis of the detectability of the cytokine mRNA in the cell population (normalized expression >0.1）以及细胞类型对细胞因子的反应性（相应的细胞因子特征中超过十摄氏度）。For heteromeric cytokines or cytokine complexes composed of two subunits (IL-12, IL-23, IL-27, LTα1/β2 and LTα2/β1), the cytokine is shown as expressed and is annotated with an asterisk if the genes encoding at least one subunit are expressed as there is evidence of extracellular assembly of some components into functional cytokines under健康或病理条件49。为每个源节点分别绘制网络，以易于解释性。 　　为了构建配体 - 受体相互作用组，我们从文献中鉴定出每个细胞因子的功能性同源受体，在补充表1中列出。对于以细胞类型表示的受体，受体的归一化表达值，所需的受体基因的归一化表达值比截止阈值更大（默认表达式单位）。对于杂体受体，需要表达所有成分才能表达受体。对于具有多个功能受体的细胞因子，如果表达任何功能受体，则认为该受体表示表达。然后，我们将细胞因子与表达同源受体的细胞类型联系起来。相互作用组的细胞因子生产部分与配体 - 反应相互作用组中的细胞因子产生部分相同。然后比较配体 - 反应和配体 - 受体网络，以生成这两种方法之间常见或不同的细胞 - 细胞通信路径。 　　我们提供两种类型的IREA分析选项，以评估用户数据中的细胞因子反应，具体取决于输入，这可以是基因集或基因表达矩阵。用户数据中的单元格类型由用户指定。然后，使用以下方法将用户数据与来自免疫词典的同一细胞类型的转录细胞因子响应进行比较： 　　IREA极化分析实现与IREA细胞因子反应分析相同的统计检验。在IREA极化中，将用户数据与极化状态基因表达曲线进行比较。如果至少一个细胞极化状态显着富集（FDR调整后的P <0.05），则显示偏振雷达图。如果没有任何状态显着富集，则雷达图显示每个状态的富集得分为0，这表示输入单元不偏振。在雷达图上，富集评分归一化为0至1之间。 　　我们通过考虑细胞因子的产生和细胞因子反应来构建细胞细胞通信网络的模型。通过检查映射到86种细胞因子中每一个的转录本获得细胞因子的产生。使用IREA评估细胞因子反应，其中包括IREA输出<0.01的细胞因子。对于由两个亚基组成的杂体细胞因子或细胞因子复合物，如果至少一个亚基表示，则显示细胞因子的表达为表达，因为有某些组件的细胞外组装成功能性细胞因子49（与细胞 - 细胞相互作用组相同的方法）。可以看到细胞因子网络，如图5E和扩展数据图12c所示。 　　将SCRNA-SEQ数据下载为10X基因组数据文件36。使用与上述相同的方法处理数据，但进行了较小的修改，从而将40个PC用于下游分析。如出版物中所示，对细胞类型进行了注释36。IREA分析是在抗PD-1处理和每种细胞类型的对照之间使用默认参数进行的。应用了0.05的受体表达阈值，以在细胞因子富集图中产生数据。 　　SCRNA-SEQ数据从人类Covid-19-Blood Study39中下载为Seurat对象。群集注释用于Seurat对象中定义。使用来自COVID-19的通风患者的数据进行IREA分析，并使用默认参数和指定为人类的物种进行默认参数与每种细胞类型的健康个体进行比较。IREA使用国家生物技术信息中心同源数据库（第68页）的最新发布来实现小鼠和人类同源基因转化。应用了0.05的受体表达阈值，以在指南针图中的受体环中产生数据。 　　相应文本中的每个分析描述了使用的统计测试。除非另有说明，否则使用双面统计测试。进行了FDR或Bonferroni调整，以进行多个假设检验的分析。 　　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。

本文来自作者[yjmlxc]投稿，不代表颐居号立场，如若转载，请注明出处：https://yjmlxc.cn/zlan/202506-8917.html