Processed (filtered and annotated) scRNA-seq data

Single cell RNA-seq data generated and reported as part of the manuscript entitled "<b>Dose escalation study of the HLA-A2-WT1 CD3 bispecific antibody RO7283420 in relapsed/refractory acute myeloid leukemia</b>" by Hutchings and Korfi et al.<br><br>Processed (filtered and annotated) data is provided, which can be directly ingested to reproduce the findings of the paper or for ab initio data reuse. processed.h5ad provides raw counts for those cells that passed QC, along with cell type annotation and relevant metadata in the standard H5AD format.<br><br>For instance, to load data in R, try:<br>library(zellkonverter)processed &lt;- readH5AD(file = "./processed.h5ad", X_name = "counts")<br><br>##############################<br>Single-cell RNA sequencing was performed on cryopreserved bone marrow mononuclear cells (BMMCs) and peripheral blood mononuclear cells (PBMCs). After thawing and sorting for viability by flow cytometry, samples were processed and sequenced using the 10x Genomics Single Cell 5’ v2 Gene Expression protocol on the 10x Chromium platform. Raw sequencing data were processed with the Cell Ranger pipeline (v7.1.0) and a custom-built transcriptome reference using the GRCh38 genome and GENCODE v43 annotation, following the exact build steps provided by 10x Genomics. Filtered feature barcode matrices obtained from Cell Ranger were imported and concatenated as a SingleCellExperiment object for downstream analysis in R. Per-cell quality control metrics were computed using the scuttle Bioconductor package. We applied a manual filtering scheme, using a hard cutoff of 10 for mitochondrial gene percentage, and the 5th percentile of total UMI counts and total number of detected genes. Next, we used scDblFinder (v1.12.0) Bioconductor package to exclude potential doublets in the samples. To identify and annotate distinct cell populations, we integrated BMMC and PBMC samples using scanorama (v1.7.4). In brief, highly variable genes were selected using modelGeneVar functionality from the scran (v1.26.2) Bioconductor package, by blocking on individual samples and retaining the top 10% of highly variable genes. We excluded mitochondrial, ribosomal, as well as T- and B-cell receptor genes from the list of highly variable genes to minimize technical and/or donor effects. Feature counts for each cell were divided by the total counts for that cell and multiplied by 10^4, followed by natural-log transformation to yield logCP10k normalized values. Normalized data was z-standardized for the highly variable genes and used as input to scanorama with default parameters. We assembled a Seurat object using the integrated PC space obtained from scanorama to perform shared nearest-neighbor graph construction, clustering, and 2D visualization using Seurat (v5.0.2) R package. We used a set of known marker genes previously reported by Wang B, et al. <i>Nat Commun</i>. 2024 to annotate main cell types. Malignant cells were identified as the three highest WT1-expressing populations with AML and hematopoietic progenitor gene expression profiles and referred collectively to as “AML” for downstream analyses. In the integrated space, T and NK cells clustered together and separated from the rest of cell types. To further split T and NK cells, and their respective subsets, we used an atlas of Bone Marrow hematopoiesis (https://github.com/andygxzeng/BoneMarrowMap) and stratified T/NK cells into 4 categories (CD4T, CD8T, NK, and other), with the “other” category representing proliferating cells and negligible non-T/NK contaminating cells. We further cross-checked the annotations obtained using selected markers as well as those obtained from BoneMarrowMap, with Azimuth BMMC and PBMC reference atlases provided by the HuBMAP consortium (https://azimuth.hubmapconsortium.org) and observed acceptable agreement between them.For downstream analysis of CD8 T cells, we calculated the per-cell enrichment score of naive-like (CCR7, SELL, LEF1, TCF7, IL7R, LTB), cytotoxic (CX3CR1, PRF1, FGFBP2, GZMB, KLRG1, FCGR3A, GZMA, GZMH, GNLY, NKG7, KLRD1), predysfunctional / effector memory (GZMK, CXCR3, ZNF683, CD28, FYN, EOMES, CXCR4, CD44), and exhausted (PDCD1, HAVCR2, LAG3, CTLA4, TIGIT, CXCL13, LAYN, ENTPD1) signatures previously reported by van der Leun et. al. Nature Rev Cancer 2020, using the Adjusted Neighborhood Scoring (ANS) (https://github.com/lciernik/ANS_signature_scoring). Mean enrichment score per sample (median of 700 cells per sample) was consequently used to compare baseline CD8 T-cell states in BMMCs from patients with or without a BM blast reduction.For downstream analysis of AML cells, WT1+ / HLA-A+ / PSMB9+ cells were defined based on a non-zero UMI threshold (UMI&gt;0). We further adopted a recent classification system of AML cells into 7 subsets (leukemia stem and progenitor cell (LSPC)-Quiescent, LSPC-Primed, LSPC-Cycle, granulocyte-monocyte progenitor (GMP)-like, Promonocyte-like, Monocyte-like, and conventional dendritic cell (cDC)-like) as reported by Zeng et. al. Nat Med 2022, to identify different AML subpopulations. In brief, signature genes associated with each of these subsets were obtained from the original publication, and ANS scores were calculated per cell as described above. Finally, each cell was assigned to one of the seven subsets based on its maximum enrichment score.<br><br>For more details, please refer to the publication.

本数据集为Hutchings与Korfi等人发表的题为《复发/难治性急性髓系白血病中HLA-A2-WT1 CD3双特异性抗体RO7283420的剂量递增研究》的手稿所生成并报道的单细胞RNA测序（single cell RNA-seq）数据。

本数据集提供了经过处理（已过滤并注释）的数据，可直接导入以复现论文的研究结果，或用于从头（ab initio）数据复用。processed.h5ad文件以标准H5AD格式存储了通过质量控制（quality control, QC）的细胞的原始计数、细胞类型注释以及相关元数据。

例如，若需在R语言中加载数据，可参考如下代码：
library(zellkonverter)
processed <- readH5AD(file = "./processed.h5ad", X_name = "counts")

##############################

研究对冻存的骨髓单个核细胞（bone marrow mononuclear cells, BMMCs）与外周血单个核细胞（peripheral blood mononuclear cells, PBMCs）进行了单细胞RNA测序。样本经复苏后通过流式细胞术分选活细胞，随后使用10x Genomics单细胞5’ v2基因表达试剂盒在10x Chromium平台上完成处理与测序。原始测序数据通过Cell Ranger流程（v7.1.0）进行处理，并基于GRCh38基因组与GENCODE v43注释构建自定义转录组参考序列，严格遵循10x Genomics提供的官方构建步骤。从Cell Ranger获取的过滤后的特征条形码矩阵（feature barcode matrix）被导入并整合为SingleCellExperiment对象，用于后续R语言分析。使用scuttle Bioconductor包计算每个细胞的质量控制指标。我们采用手动过滤方案：设置线粒体基因占比的硬截断值为10，同时保留总唯一分子标识符（unique molecular identifier, UMI）计数与检测基因数的前5%分位数。随后，使用scDblFinder（v1.12.0）Bioconductor包去除样本中潜在的双细胞。

为识别并注释不同的细胞群，我们使用scanorama（v1.7.4）整合BMMC与PBMC样本。简言之，通过scran（v1.26.2）Bioconductor包的modelGeneVar功能选择高可变基因（highly variable genes），以单个样本作为分组因素进行批次校正，并保留前10%的高可变基因。我们从高可变基因列表中移除线粒体、核糖体以及T、B细胞受体基因，以尽可能降低技术效应与供体差异带来的影响。将每个细胞的特征计数除以该细胞的总计数后乘以10^4，随后进行自然对数转换，得到logCP10k标准化值。将标准化后的数据针对高可变基因进行z标准化（z-score standardization），作为默认参数下scanorama的输入数据。使用从scanorama获取的整合主成分分析（principal component analysis, PCA）空间构建Seurat对象，通过Seurat（v5.0.2）R包完成共享最近邻图构建、细胞聚类与二维可视化。我们使用Wang B等人2024年发表于《Nature Communications（自然-通讯）》的已知标记基因集注释主要细胞类型。恶性细胞被定义为表达WT1水平最高的三个群体，且具有急性髓系白血病（acute myeloid leukemia, AML）与造血祖细胞基因表达特征，后续分析中将其统称为"AML"。

在整合空间中，T细胞与NK细胞聚为一类，并与其他细胞类型分离。为进一步拆分T细胞与NK细胞及其亚群，我们使用骨髓造血图谱（https://github.com/andygxzeng/BoneMarrowMap）将T/NK细胞分为4类：CD4T、CD8T、NK及其他，其中"其他"类别代表增殖细胞与可忽略的非T/NK污染细胞。我们还使用HuBMAP联盟提供的Azimuth BMMC与PBMC参考图谱（https://azimuth.hubmapconsortium.org）交叉验证通过标记基因与骨髓造血图谱获得的细胞注释，结果显示二者一致性良好。

针对CD8 T细胞的下游分析，我们使用van der Leun等人2020年发表于《Nature Reviews Cancer（自然综述：癌症）》的基因特征，通过调整邻域评分（Adjusted Neighborhood Scoring, ANS）方法（https://github.com/lciernik/ANS_signature_scoring）计算每个细胞的富集评分，包括未样型（naive-like，特征基因为CCR7、SELL、LEF1、TCF7、IL7R、LTB）、细胞毒性型（cytotoxic，特征基因为CX3CR1、PRF1、FGFBP2、GZMB、KLRG1、FCGR3A、GZMA、GZMH、GNLY、NKG7、KLRD1）、预功能失调/效应记忆型（predysfunctional / effector memory，特征基因为GZMK、CXCR3、ZNF683、CD28、FYN、EOMES、CXCR4、CD44）以及耗竭型（exhausted，特征基因为PDCD1、HAVCR2、LAG3、CTLA4、TIGIT、CXCL13、LAYN、ENTPD1）。最终以每个样本的平均富集评分（每个样本700个细胞的中位数）比较伴或不伴骨髓原始细胞减少的患者骨髓单个核细胞中基线CD8 T细胞状态。

针对AML细胞的下游分析，我们基于非零UMI阈值（UMI>0）定义WT1+ / HLA-A+ / PSMB9+细胞。我们进一步采用Zeng等人2022年发表于《Nature Medicine（自然医学）》的AML细胞分类系统，将其分为7个亚群：静止期白血病干细胞与祖细胞（leukemia stem and progenitor cell, LSPC）-Quiescent、致敏型白血病干细胞与祖细胞（LSPC-Primed）、周期型白血病干细胞与祖细胞（LSPC-Cycle）、粒-单核细胞祖细胞（granulocyte-monocyte progenitor, GMP）样、前单核细胞样、单核细胞样以及常规树突状细胞（conventional dendritic cell, cDC）样，以识别不同的AML亚群。简言之，从原始文献中获取与每个亚群相关的特征基因，并按照前述方法计算每个细胞的ANS评分。最终根据每个细胞的最大富集评分将其分配至7个亚群中的一个。

更多细节请参阅该发表论文。