A blueprint for tumor-infiltrating B cells across human cancers
人类癌症中肿瘤浸润 B 细胞的蓝图

B cell infiltration varies highly across human cancer types (1). To understand the overall infiltration pattern of B cells, we assessed the abundance of B cell infiltration in The Cancer Genome Atlas (TCGA) Pan-Cancer datasets (n = 8863 tumor samples of 31 cancer types) on the basis of the consensus of six existing deconvolution algorithms (fig. S1A). We selected the cancer types with high and medium intratumor B cell scores for further sampling. CD19⁺ B cells were sorted from 153 samples of 66 patients across 15 human cancer types covering matched tumors, lymph node metastases (LN_Mets), adjacent normal tissues, and peripheral blood, followed by paired single-cell RNA sequencing (scRNA-seq) and single-cell B cell–receptor sequencing (scBCR-seq) (Fig. 1A and fig. S1B). We also integrated diverse published datasets containing TIBs, corrected the batch effects using the Harmony algorithm (25), and finally established a single-cell transcriptional atlas in 477 samples from 269 donors across 20 cancer types. After stringent quality control, we obtained a total of 474,718 single B cell transcriptomes, 69.24% of which were newly generated in this study (fig. S1C and table S1). For scBCR-seq, a total of 166,733 cells from 150,949 clonotypes spanning 61 donors of 15 cancer types were newly produced, carrying at least one pair of productive heavy and light chains, of which 12.76% were clonal (≥two B cells containing the identical BCR pair), corresponding to 21,268 expanded clonotypes. Overall, our single-cell atlas creates a pan-cancer blueprint for B cell investigation (available at http://pancancer.cn/B/).
B 细胞浸润在不同的人类癌症类型中差异很大 (1)。为了了解 B 细胞的整体浸润模式，我们根据六种现有的共识，评估了癌症基因组图谱 (TCGA) 泛癌症数据集（n = 31 种癌症类型的 8863 个肿瘤样本）中 B 细胞浸润的丰度。反卷积算法（图S1A）。我们选择了肿瘤内 B 细胞评分高和中等的癌症类型进行进一步采样。 CD19 ⁺ B 细胞从 15 种人类癌症类型的 66 名患者的 153 个样本中分选，涵盖匹配的肿瘤、淋巴结转移 (LN_Mets)、邻近正常组织和外周血，然后进行配对单细胞 RNA 测序(scRNA-seq) 和单细胞 B 细胞受体测序 (scBCR-seq)（图 1A 和图 S1B）。我们还整合了包含 TIB 的不同已发表数据集，使用 Harmony 算法校正批次效应 (25)，并最终在来自 269 名捐赠者、20 种癌症类型的 477 个样本中建立了单细胞转录图谱。经过严格的质量控制，我们总共获得了474,718个单B细胞转录组，其中69.24%是本研究新生成的（图S1C和表S1）。对于 scBCR-seq，新产生了来自 150,949 个克隆型的总共 166,733 个细胞，涵盖 15 种癌症类型的 61 个供体，携带至少一对生产性重链和轻链，其中 12.76% 是克隆细胞（≥两个 B 细胞含有相同的重链和轻链） BCR 对），对应于 21,268 个扩展克隆型。总体而言，我们的单细胞图谱为 B 细胞研究创建了泛癌蓝图（可在 http://pancancer.cn/B/ 获取）。

Unsupervised clustering analysis identified 15 distinct B cell subsets, all of which were reproducibly observed across cancer types and exhibited distinct tissue- and cancer-type preferences (Fig. 1B; fig. S1, D and E; and table S2). Among them, typical B cell subsets representing different B cell maturation stages were identified, including one naïve B cell (TCL1A, FCER2, and IGHD), three activated B cells (ACBs; CD69 and CD83), one Bm (CRIP2 and ITGB1), three GC B cells [LMO2⁺Light Zone (LZ), CXCR4⁺Dark Zone (DZ), and MKI67⁺cycling], and two ASCs (including TXNDC5⁺PBs and MZB1⁺ PCs) (Fig. 1C and fig. S1F).
无监督聚类分析确定了 15 个不同的 B 细胞亚群，所有这些亚群在各种癌症类型中均可重复观察到，并表现出不同的组织和癌症类型偏好（图 1B；图 S1、D 和 E；以及表 S2）。其中，鉴定了代表不同B细胞成熟阶段的典型B细胞亚群，包括一种幼稚B细胞（TCL1A、FCER2和IGHD）、三种活化B细胞（ACB；CD69和CD83）、一种Bm（CRIP2和ITGB1）、三个 GC B 细胞 [LMO2 ⁺ 亮区 (LZ)、CXCR4 ⁺ 暗区 (DZ) 和 MKI67 ⁺ 循环] 和两个 ASC（包括TXNDC5 ⁺ PB 和 MZB1 ⁺ PC）（图 1C 和图 S1F）。

We next identified four additional subsets that were largely uncharacterized previously (26–30) (fig. S2A). The first was an interferon-stimulated gene-positive naïve B cell subset (B_02) that highly expressed IFIT3, IFI44L, and ISG15, which was associated with injury and was expanded during mucosal healing (31). The second subset was stressed B cells (B_03), predominant in the tumor with high expression of heat shock proteins, as confirmed by 37°C collagenase and 4°C mechanical dissociation and multiplex immunohistochemistry (mIHC) (fig. S2, B and C). In comparison with stressed T cells (32), stressed B cells also resided in the center of TLSs, with high expression of hypoxia-related genes, such as HIF1A, NFE2L2, RELB, NFKB1, and NFKB2 (fig. S2D). The third subset was pre-GC B cells (B_10) that were highly enriched in LN_Mets and tumors as compared with adjacent normal tissues and blood, expressing PSME2, NME1, and ENO1. Pre-GC B cells showed transcriptional similarity to naïve B cells, GC B cells, and PBs, respectively, indicating their transitional stages (Fig. 1C).
接下来，我们确定了之前基本上未表征的四个附加子集 (26-30)（图 S2A）。第一个是干扰素刺激的基因阳性幼稚 B 细胞亚群 (B_02)，其高表达 IFIT3、IFI44L 和 ISG15，与损伤相关，并在粘膜愈合过程中扩增 (31)。第二个亚群是应激 B 细胞 (B_03)，主要存在于热休克蛋白高表达的肿瘤中，经 37°C 胶原酶和 4°C 机械解离和多重免疫组织化学 (mIHC) 证实（图 S2、B 和 C） ）。与应激的T细胞相比（32），应激的B细胞也位于TLS的中心，缺氧相关基因高表达，如HIF1A、NFE2L2、RELB、NFKB1和NFKB2（图S2D）。第三个亚群是GC前B细胞(B_10)，与邻近的正常组织和血液相比，它们在LN_Met和肿瘤中高度富集，表达PSME2、NME1和ENO1。 GC前B细胞分别表现出与初始B细胞、GC B细胞和PB的转录相似性，表明它们的过渡阶段（图1C）。

The fourth subset we identified was a previously unknown AtM B cell subset (B_09) expressing DUSP4, ITGAX (CD11c), FCRL5, ZEB2, and FGR. Previously documented as ABCs, CD27⁻IgD⁻ B_DN, FCRL4⁺ B cells, or exhausted Bm cells reported in aging mice, autoimmune diseases, tonsil, and chronic infection, respectively (16, 19, 23, 24), this subset is now demonstrated to widely exist in the TME (Fig. 1C). We collected and isolated previously published scRNA-seq AtM B cells from autoimmune diseases and hepatitis B virus (HBV) infection datasets (table S1) and performed differentially expressed gene analysis, showing that tumor-infiltrating AtM B cells highly expressed stress-related genes (FOSB and HSPA1A), interferon-induced genes (IFITM1 and IFI30), and activation-related genes (CD83, NR4A1, and CD69), whereas naïve-related genes (TXNIP and IGHD) and metabolic-related genes (MT-ATP6 and NDUFB1) were enriched in autoimmune diseases and HBV infection (fig. S2, E and F, and table S2). Likewise, tumor AtM B cells were enriched in pathways of inflammatory response and hypoxia, whereas angiogenesis and metabolism pathways were dominant in those from autoimmune diseases and HBV infection (fig. S2G). These data highlight that AtM B cells in TME could be functionally distinct, despite their phenotypic concordance with those from autoimmune diseases and virus infections.
我们确定的第四个亚群是以前未知的 AtM B 细胞亚群 (B_09)，表达 DUSP4、ITGAX (CD11c)、FCRL5、ZEB2 和 FGR。先前记录为 ABC、CD27 ⁻ IgD ⁻ B _DN 、FCRL4 ⁺ B 细胞或衰老小鼠中报道的耗尽 Bm 细胞，自身免疫性疾病、扁桃体和慢性感染分别（16、19、23、24），该子集现已被证明广泛存在于 TME 中（图 1C）。我们从自身免疫性疾病和乙型肝炎病毒（HBV）感染数据集中收集并分离了先前发表的 scRNA-seq AtM B 细胞（表 S1），并进行差异表达基因分析，表明肿瘤浸润的 AtM B 细胞高表达应激相关基因（ FOSB 和 HSPA1A）、干扰素诱导基因（IFITM1 和 IFI30）和激活相关基因（CD83、NR4A1 和 CD69），而幼稚相关基因（TXNIP 和 IGHD）和代谢相关基因（MT-ATP6 和 NDUFB1） ）在自身免疫性疾病和 HBV 感染中富集（图 S2、E 和 F，以及表 S2）。同样，肿瘤 AtM B 细胞富含炎症反应和缺氧途径，而血管生成和代谢途径在自身免疫性疾病和 HBV 感染的细胞中占主导地位（图 S2G）。这些数据强调，TME 中的 AtM B 细胞在功能上可能不同，尽管它们的表型与自身免疫性疾病和病毒感染的细胞一致。

Integrative analysis of tissue distribution and transcriptional and repertoire features highlighted the potential tumor-reactive B cells from different tissue sources. TIBs contained more IGHG and IGHA than B cells from nontumor tissues, whereas B cells in blood were enriched with IGHD and IGHM, suggesting that TIBs were antigen-experienced and underwent CSR (Fig. 1D). By the odds ratio (OR) analysis (33), HSP⁺ B cells, EGR1⁺ACBs, and PBs showed a strong preference in tumor; MT1X⁺ B cells, cycling GC B cells, pre-GC B cells, GC_LZ, GC_DZ, and IFIT3⁺ B cells were significantly enriched in LN_Mets; TCL1A⁺ naïve B cells and DUSP4⁺ AtM B cells were highest in blood; and MZB1⁺ PCs were dominant in adjacent normal tissue (Fig. 1E). Again, TIBs had significantly more IGH mutations than those in other tissues, with GC B cells and two ASC subsets exhibiting tumor-specific clonal expansion, indicating their tumor reactivity with strong antigen-driven antibody affinity (Fig. 1F and fig. S2H). In addition, ITGB1⁺ Bm and DUSP4⁺ AtM B cells showed comparable enrichment in tumor, LN_Mets, and blood, slightly higher than adjacent normal tissue, implying intertissue migration potential. Indeed, the STARTRAC migration (pMigr) index (34), which quantifies the possibility of migration between tissues, and the Jaccard index (35), which illustrates the number of BCR clones shared between different tissues, showed that ITGB1⁺ Bm and AtM B cells had moderate mobility between blood and LN_Mets or tumor in about half of the cancers (fig. S2, I to K).
对组织分布以及转录和谱特征的综合分析强调了来自不同组织来源的潜在肿瘤反应性 B 细胞。 TIBs比来自非肿瘤组织的B细胞含有更多的IGHG和IGHA，而血液中的B细胞富含IGHD和IGHM，这表明TIBs经历过抗原并经历了CSR（图1D）。通过比值比（OR）分析（33），HSP ⁺ B 细胞、EGR1 ⁺ ACB 和 PB 在肿瘤中表现出强烈的偏好； MT1X ⁺ B 细胞、循环 GC B 细胞、GC 前 B 细胞、GC _LZ 、GC _DZ 和 IFIT3 ⁺ B细胞中 LN_Met 显着富集； TCL1A ⁺ 幼稚 B 细胞和 DUSP4 ⁺ AtM B 细胞在血液中最高；和MZB1 ⁺ PCs在邻近正常组织中占主导地位（图1E）。同样，TIB 的 IGH 突变明显多于其他组织，GC B 细胞和两个 ASC 亚群表现出肿瘤特异性克隆扩张，表明它们具有强抗原驱动抗体亲和力的肿瘤反应性（图 1F 和图 S2H）。此外，ITGB1 ⁺ Bm 和 DUSP4 ⁺ AtM B 细胞在肿瘤、LN_Mets 和血液中显示出相当的富集，略高于邻近正常组织，这意味着组织间迁移的潜力。事实上，量化组织间迁移可能性的 STARTRAC 迁移 (pMigr) 指数 (34) 和说明不同组织之间共享的 BCR 克隆数量的 Jaccard 指数 (35) 表明，ITGB1 ⁺

ASCs are terminally differentiated B cells that execute effector function by producing antibodies (36). Compared with the existing data classifying ASCs into a homogenous subset (14, 28, 37, 38), our high-resolution map revealed that tumor-infiltrated ASCs exhibited multifaceted diversities in tissue distribution, cancer-type preference, BCR repertoire, and transcriptomic profile. First, according to the Shannon equitability index (39), the proportion of ASCs showed great variation, with tumor-infiltrated ASCs showing higher diversity than those in adjacent normal tissues and blood (P < 0.01), whereas the diversity of non-ASCs was the highest in LN_Mets (fig. S3A). Second, we observed a higher diversity of BCR clonotypes along with an increased number of expanded BCR clonotypes. However, there was a lower clonal frequency observed in tumors, suggesting a higher homogeneity of clone size in tumors than in other tissues, where a small number of expanded clones dominate the repertoire (fig. S3B). Third, cancer types exerted a strong influence on the frequency of ASCs (fig. S1D). Of the two ASC subsets, the median frequency of tumor-infiltrated MZB1⁺ PCs was 12.52% of TIBs, ranging from high (70.18%) in colon adenocarcinoma (COAD) to barely detectable in cholangiocarcinoma (CHOL) (1.85%) and stomach adenocarcinoma (STAD) (3.32%) [analysis of variance (ANOVA), P < 2e⁻¹⁶] (fig. S3, C and D). The median frequency of tumor-infiltrated TXNDC5⁺ PBs was 0.13% but still varied among different cancer types (P = 1.1e⁻⁵). Last, ROGUE analysis (40), which quantifies the transcriptomic purity of identified cell types, further confirmed that of the 15 B cell subsets, 2 ASC subsets exhibited the highest heterogeneity of gene expression (fig. S3E). Subclustering of the ASC compartment (removing TXNDC5⁺ PBs) revealed 10 distinct PC subclusters with specific gene signatures and tissue distribution (Fig. 2, A and B; fig. S3F; and table S2). Among them, PC04.HSPA1A, PC05.NEAT1, and PC08.IFI6 were dominant in the tumor, PC09.RGS13 and PC10.SPINK2 accumulated in adjacent normal tissue, and PC02.RGS13 was prevalent in LN_Mets. In parallel, PC01.NME2 and PC07.CD83 showed comparable enrichment in LN_Mets and blood, PC03.DUSP5 was equally distributed in LN_Mets and tumor, and PC06.SLC3A2 was similarly distributed in the tumor and adjacent normal tissue (Fig. 2C). These findings may help explain the controversial roles of PCs in patient prognosis (5, 41, 42), which could be ascribed to the variability of different PC subsets and cancer types. Collectively, these data indicate that cancer origin has considerable impact on the frequency of ASCs and also imply different B cell origins for ASCs.
ASC 是终末分化的 B 细胞，通过产生抗体来执行效应功能 (36)。与现有将 ASC 分类为同质子集 (14、28、37、38) 的数据相比，我们的高分辨率图谱显示肿瘤浸润的 ASC 在组织分布、癌症类型偏好、BCR 库和转录组谱方面表现出多方面的多样性。首先，根据香农公平性指数（39），ASCs的比例表现出很大的变异性，肿瘤浸润的ASCs的多样性高于邻近正常组织和血液中的ASCs（P < 0.01），而非ASCs的多样性则较低。 LN_Mets 中最高（图 S3A）。其次，我们观察到 BCR 克隆型的多样性更高，并且扩展的 BCR 克隆型数量也有所增加。然而，在肿瘤中观察到的克隆频率较低，这表明肿瘤中克隆大小的同质性高于其他组织，在其他组织中，少数扩展的克隆占主导地位（图S3B）。第三，癌症类型对 ASC 的频率有很大影响（图 S1D）。在两个 ASC 亚群中，肿瘤浸润的 MZB1 ⁺ PC 的中位频率为 TIB 的 12.52%，范围从结肠腺癌 (COAD) 中的高频率 (70.18%) 到胆管癌 (CHOL) 中几乎检测不到。 1.85%) 和胃腺癌 (STAD) (3.32%) [方差分析 (ANOVA), P < 2e ⁻¹⁶ ]（图 S3、C 和 D）。肿瘤浸润的 TXNDC5 ⁺ PB 的中位频率为 0.13%，但不同癌症类型之间仍然存在差异 (P = 1.1e ⁻⁵ )。最后，ROGUE 分析 (40) 对已识别细胞类型的转录组纯度进行量化，进一步证实在 15 个 B 细胞亚群中，2 个 ASC 亚群表现出最高的基因表达异质性（图 S3E）。 ASC 区室的亚聚类（除去 TXNDC5 ⁺ PB）揭示了 10 个具有特定基因特征和组织分布的不同 PC 亚聚类（图 2，A 和 B；图 S3F；和表 S2）。其中，PC04.HSPA1A、PC05.NEAT1和PC08.IFI6在肿瘤中占优势，PC09.RGS13和PC10.SPINK2在邻近正常组织中积累，PC02.RGS13在LN_Met中普遍存在。同时，PC01.NME2和PC07.CD83在LN_Mets和血液中显示出相当的富集，PC03.DUSP5在LN_Mets和肿瘤中均匀分布，并且PC06.SLC3A2在肿瘤和邻近正常组织中相似分布（图2C）。这些发现可能有助于解释 PC 在患者预后中具有争议的作用 (5,41,42)，这可能归因于不同 PC 亚群和癌症类型的变异性。总的来说，这些数据表明癌症起源对 ASC 的频率有相当大的影响，也意味着 ASC 的 B 细胞起源不同。

Although B cell pathways derived from GC (1, 17, 42) or extrafollicular (EF) responses are commonly observed in autoimmune diseases and chronic infections (21, 43, 44), the potential for the tumor ecosystem to influence distinct B cell evolutionary trajectories has not been thoroughly explored. We thus used gene expression to infer the trajectories of ASC differentiation and observed that ASCs were mainly derived from GC B cells (fig. S3G), which is consistent with previous studies (2, 45). However, given that ASC subsets were significantly distinguished from the other subsets by unsupervised clustering on the basis of high expression of PC markers, traditional transcriptome-based trajectory analysis may not accurately reflect the true differentiation state. To address this, we used a BCR clonal-sharing strategy and unsupervised clustering to identify the progenitors of ASCs with STARTRAC [single T cell analysis by RNA-seq and TCR (T cell receptor) tracking] pairwise transition indices (pTrans) (34), Jaccard indices (35), and somatic hypermutations (SHM), because the BCR repertoire is a reliable molecular tag for tracing the lineage of ASCs. We showed that AtM and Bm cells were the two primary B cells clonally shared with ASCs, suggesting that ASCs may originate from both canonical GC and alternative EF pathways in the TME (Fig. 2D and fig. S3, H and I). To validate this, tumor-infiltrated naïve B, Bm, and AtM B cells from liver hepatocellular carcinoma (LIHC) patients were sorted and stimulated by using in vitro B cell differentiation methods (24). Indeed, AtM B cells could differentiate into ASCs more efficiently than naïve B cells (P = 0.035) but more weakly than Bm cells (P = 0.0027) (Fig. 2, E and F). In addition to AtM and Bm cells, we also identified pre-GC, cycling GC, and GC_LZ B cells sharing BCRs with PBs, and IFIT3⁺ B, HSP⁺ B, and EGR1⁺ ACBs sharing BCRs with PCs, which were validated by phylogenetic lineage trees of shared clones between ASCs and these subsets (Fig. 2G and fig. S3, J to L).
尽管源自 GC (1, 17, 42) 或滤泡外 (EF) 反应的 B 细胞途径在自身免疫性疾病和慢性感染中常见 (21, 43, 44)，但肿瘤生态系统有可能影响不同的 B 细胞进化轨迹尚未被彻底探索。因此，我们利用基因表达来推断ASC分化的轨迹，并观察到ASC主要来源于GC B细胞（图S3G），这与之前的研究一致（2, 45）。然而，鉴于ASC子集在PC标记高表达的基础上通过无监督聚类与其他子集显着区分，传统的基于转录组的轨迹分析可能无法准确反映真实的分化状态。为了解决这个问题，我们使用 BCR 克隆共享策略和无监督聚类，通过 STARTRAC [通过 RNA-seq 和 TCR（T 细胞受体）跟踪进行单 T 细胞分析] 成对转变指数 (pTrans) 来识别 ASC 的祖细胞 (34) 、Jaccard 指数 (35) 和体细胞超突变 (SHM)，因为 BCR 库是追踪 ASC 谱系的可靠分子标签。我们发现 AtM 和 Bm 细胞是与 ASC 克隆共享的两种主要 B 细胞，这表明 ASC 可能起源于 TME 中的经典 GC 和替代 EF 途径（图 2D 和图 S3、H 和 I）。为了验证这一点，使用体外 B 细胞分化方法对来自肝细胞癌 (LIHC) 患者的肿瘤浸润的幼稚 B、Bm 和 AtM B 细胞进行分类和刺激 (24)。事实上，AtM B 细胞可以比初始 B 细胞更有效地分化为 ASC（P = 0.035），但比 Bm 细胞更弱（P = 0.0027）（图 2，E 和 F）。 除了 AtM 和 Bm 细胞外，我们还鉴定了前 GC、循环 GC 和与 PB 共享 BCR 的 GC _LZ B 细胞，以及 IFIT3 ⁺ B、HSP ⁺ ACB 与 PC 共享 BCR，这通过 ASC 和这些子集之间共享克隆的系统发育谱系树进行了验证（图 2G 和图 S3，J 到 L）。

We next evaluated the prevalence of the two paths and classified the ASCs sharing BCRs with canonical Bm (66% of GC-derived ASCs), cycling GC (13%), GC_LZ (12%), and GC_DZ (7%) as GC-derived ASCs, whereas those ASCs overlapping with other B cell subsets, including AtM (47% of EF-derived ASCs), ACB (29%), IFIT3⁺ B (8%), HSP⁺ B (6%), pre-GC (5%), and MT1X⁺ B (3%) cells, were classified as EF-derived ASCs at the cellular level (fig. S4A). Subsequently, we used label transfer to acquire more EF- and GC-derived ASCs on the basis of the gene expression profile of already classified ASCs and then validated label-transfer accuracy (fig. S4B). With the defined EF-derived and GC-derived ASCs, we further computed the EF index and integrated with SHM to classify patients into two distinct groups—either EF-dominant or GC-dominant at the patient level—and this classification was further validated and refined by pTrans score (fig. S4, C to E). Both paths could be observed in individual patients and cancers despite obvious patient- and cancer-type preference, indicating both GC and EF responses as universal paths for ASC differentiation (Fig. 2H). COAD and thyroid carcinoma (THCA) used GC_DZ, GC_LZ, and cycling GC B cells, whereas gallbladder cancer (GBC), STAD, lung cancer (LC), and thymoma (THYM) used Bm cells as preferred to the GC pathway. Conversely, the AtM B cell–dominant EF path was enriched in LIHC, pancreatic adenocarcinoma (PAAD), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), and head and neck squamous cell carcinoma (HNSC), whereas pre-GC, ACB, and the IFIT3⁺-dominant EF path were abundant in renal cell carcinoma (RCC), bladder urothelial carcinoma (BLCA), and gastrointestinal stromal tumor (GIST), respectively (Fig. 2H and fig. S4F). AtM B cells had pTrans scores comparable to those of cycling GC and Bm cells in ovarian cancer (OV) and breast cancer (BRCA), respectively, indicating the equal presence of both pathways in these cancers (fig. S4G). In addition to intersubset BCR sharing, the developmental potential of GC and EF pathways could be validated by the expression signatures of non-ASC subsets in ASCs (fig. S4H). For example, PC01.NME2 expressed pre-GC signature genes, including PSME2 and ENO1, and these pre-GC B cells shared BCR with both TXNDC5⁺ PBs and MZB1⁺ PCs in BLCA and RCC. Similarly, PC04.HSPA1A expressed HSPA1A and HSPA1B, and HSP⁺ B cells favored to share BCR with MZB1⁺ PCs in BLCA, suggesting heterogeneous differentiation pathways toward ASCs. Together, these data suggest that two evolutionary paths (GC and EF) are hijacked by distinct cancer ecosystems.
接下来，我们评估了这两种路径的流行率，并将与典型 Bm 共享 BCR 的 ASC（GC 衍生的 ASC 的 66%）、循环 GC（13%）、GC _LZ （12%）和 GC 分类 _DZ (7%) 为 GC 衍生的 ASC，而这些 ASC 与其他 B 细胞亚群重叠，包括 AtM（EF 衍生的 ASC 的 47%）、ACB (29%)、IFIT3 ⁺ B (6%)、GC 前 (5%) 和 MT1X ⁺ B (3%) 细胞，被分类为细胞水平上的 EF 衍生 ASC（图 S4A）。随后，我们根据已分类的 ASC 的基因表达谱，使用标签转移来获取更多 EF 和 GC 衍生的 ASC，然后验证标签转移的准确性（图 S4B）。通过定义 EF 衍生和 GC 衍生的 ASC，我们进一步计算 EF 指数并与 SHM 集成，将患者分为两个不同的组——在患者水平上以 EF 为主或以 GC 为主——并且这种分类得到了进一步验证和验证。通过 pTrans 分数进行细化（图 S4，C 至 E）。尽管有明显的患者和癌症类型偏好，但在个体患者和癌症中都可以观察到这两种路径，表明 GC 和 EF 反应都是 ASC 分化的通用路径（图 2H）。 COAD 和甲状腺癌 (THCA) 使用 GC _DZ 、 GC _LZ 和循环 GC B 细胞，而胆囊癌 (GBC)、STAD、肺癌 (LC) 和胸腺瘤 ( THYM) 使用 Bm 细胞作为 GC 途径的首选。 相反，AtM B 细胞主导的 EF 路径在 LIHC、胰腺腺癌 (PAAD)、宫颈鳞状细胞癌和宫颈内膜腺癌 (CESC) 以及头颈鳞状细胞癌 (HNSC) 中丰富，而前 GC、ACB、 IFIT3 ⁺ 主导的EF路径分别在肾细胞癌（RCC）、膀胱尿路上皮癌（BLCA）和胃肠道间质瘤（GIST）中大量存在（图2H和图S4F）。 AtM B 细胞的 pTrans 评分分别与卵巢癌 (OV) 和乳腺癌 (BRCA) 中循环 GC 和 Bm 细胞的 pTrans 评分相当，表明这两种途径在这些癌症中同等存在（图 S4G）。除了子集间 BCR 共享之外，GC 和 EF 途径的发育潜力还可以通过 ASC 中非 ASC 子集的表达特征来验证（图 S4H）。例如，PC01.NME2 表达前 GC 特征基因，包括 PSME2 和 ENO1，并且这些前 GC B 细胞与 BLCA 中的 TXNDC5 ⁺ PB 和 MZB1 ⁺ PC 共享 BCR和RCC。类似地，PC04.HSPA1A 表达 HSPA1A 和 HSPA1B，HSP ⁺ B 细胞倾向于与 BLCA 中的 MZB1 ⁺ PC 共享 BCR，表明向 ASC 的异质分化途径。总之，这些数据表明两种进化路径（GC 和 EF）被不同的癌症生态系统劫持。

We next explored the BCR repertoire and matched transcriptional features of these two evolutionary paths. The median frequency of EF-derived ASCs in tumors was 1.52% [0.35 to 4.37% (interquartile range)], which was significantly higher than that of GC-derived ASCs (0.45%; 0.13 to 2.29%) (fig. S5A). By contrast, the diversity of gene expression and BCR clonotypes was reduced in EF-derived ASCs (fig. S5, B and C). These cells were characterized by oligoclonal expansions, with the top five clonotypes accounting for 11.51% of the total repertoire compared with 3.20% in GC-derived ASCs. Indeed, GC-derived ASCs exhibited high diversity and relatively homogeneous clone sizes (fig. S5C). Consistent with previous findings in autoimmune disease (43, 46), EF-derived ASCs in tumors displayed significantly lower SHM than those of GC-derived ones (Fig. 2I and fig. S5D). Isotype analysis demonstrated that EF-derived ASCs contained high interferon-induced immunoglobulin G (IgG) isotype and CSR relative to IGHG1 in tumor, along with higher IGHM enrichment and no CSR occurring in adjacent normal tissue and blood (Fig. 2J). By contrast, GC-derived ASCs showed significantly higher expansion and CSR relative to IGHA1 and IGHA2 in tumor, adjacent normal tissue, and blood, which was linked to tumor antigen–specific recognition and antitumor immunity (47) (Fig. 2K and fig. S5E). These differences may stem from the intrinsic distinctions between EF- and GC-derived ASCs, with EF-derived ASCs highly expressing IRF8, CD83, and HLA-DR, exhibiting an earlier immune-response phenotype (48) and enriching for pathways of hypoxia and inflammatory response. Conversely, GC-derived ASCs showed higher expression of immunoglobulin genes and exhibited a more mature PC phenotype—which is consistent with their appearance at later stages of the immune response—and were enriched for several metabolic pathways, including glycolysis, oxidative phosphorylation, and fatty-acid metabolism, highlighting extensive energetic and biosynthetic demands during CSR and GC response (fig. S5, F and G, and table S2). To compare the early and late differentiation stages of EF- and GC-derived ASCs, we used three methods to infer the differentiation state, including cytoTRACE (49), scTour (50), and Monocle3 (51), and showed that EF-derived ASCs were in the early stages compared with terminal differentiation of GC-derived ASCs, indicating an early EF response and a delayed GC response (52) (Fig. 2L and fig. S5H). These results were further validated because EF-derived ASCs were enriched for PC07.CD83, with high expression of HLA-DR and FOXP1, which has been demonstrated to impede germinal center (GC) formation and repress human PC differentiation (53, 54) (fig. S5I). Additionally, the enrichment of the IGHM isotype in PC07.CD83 further aligns with the characteristics of early immunoglobulin M (IgM) ASCs in bone marrow (48) (fig. S5, J and K). GC-derived ASCs were enriched for PC03.DUSP5, PC04.HSPA1A, and PC10.SPINK2. LIHC-infiltrated ASCs exhibited significantly higher EF-derived ASC markers, such as CD20, HLA-DR, CD37, CD74, CD83, and IRF8, than did COAD ones (fig. S5, L and M). These results were also validated with mIHC staining for IRF8 and HLA-DR in ASCs, using tissue microarray (TMA) encompassing both EF-dominant (BLCA, CESC, GIST, LIHC, PAAD, RCC; n = 49) and GC-dominant cancers (COAD, STAD, and LC; n = 23) (fig. S5N). The frequency of IRF8⁺ ASCs (median in EF = 13.67%, median in GC = 6.58%; P = 0.0315) and HLA-DR⁺ ASCs (median in EF = 50%, median in GC = 34.09%; P = 0.0066) among CD79⁺CD20⁻ ASCs was significantly higher in EF-dominant cancers compared with GC-dominant cancers (fig. S5O). Taken together, these results further support the conception that the early response of EF-derived ASCs contrasts with the delayed response of GC-derived ASCs.
接下来，我们探索了 BCR 库并匹配了这两种进化路径的转录特征。肿瘤中 EF 衍生的 ASC 的中位频率为 1.52% [0.35 至 4.37%（四分位距）]，显着高于 GC 衍生的 ASC（0.45%；0.13 至 2.29%）（图 S5A）。相比之下，EF 衍生的 ASC 中基因表达和 BCR 克隆型的多样性降低（图 S5、B 和 C）。这些细胞的特点是寡克隆扩增，前五种克隆型占总库的 11.51%，而 GC 衍生的 ASC 中仅占 3.20%。事实上，GC 衍生的 ASC 表现出高度多样性和相对均质的克隆大小（图 S5C）。与先前在自身免疫性疾病中的发现一致 (43, 46)，肿瘤中 EF 衍生的 ASC 表现出显着低于 GC 衍生 ASC 的 SHM（图 2I 和图 S5D）。同种型分析表明，相对于肿瘤中的IGHG1，EF衍生的ASC含有较高的干扰素诱导的免疫球蛋白G（IgG）同种型和CSR，并且IGHM富集度更高，并且在邻近正常组织和血液中没有出现CSR（图2J）。相比之下，在肿瘤、邻近正常组织和血液中，GC 衍生的 ASC 相对于 IGHA1 和 IGHA2 表现出显着更高的扩增和 CSR，这与肿瘤抗原特异性识别和抗肿瘤免疫相关 (47)（图 2K 和图 2K）。 S5E）。这些差异可能源于 EF 衍生的 ASC 和 GC 衍生的 ASC 之间的内在区别，EF 衍生的 ASC 高表达 IRF8、CD83 和 HLA-DR，表现出早期的免疫反应表型 (48)，并丰富了缺氧和缺氧的途径。炎症反应。 相反，GC 衍生的 ASC 显示出更高的免疫球蛋白基因表达，并表现出更成熟的 PC 表型（这与它们在免疫反应后期的表现一致），并且富集了多种代谢途径，包括糖酵解、氧化磷酸化和脂肪代谢途径。 -酸代谢，突出了 CSR 和 GC 响应期间广泛的能量和生物合成需求（图 S5、F 和 G，以及表 S2）。为了比较 EF 和 GC 衍生的 ASC 的早期和晚期分化阶段，我们使用了三种方法来推断分化状态，包括 cytoTRACE (49)、scTour (50) 和 Monocle3 (51)，并表明 EF 衍生的 ASC与 GC 衍生的 ASC 的终末分化相比，ASC 处于早期阶段，表明早期的 EF 反应和延迟的 GC 反应 (52)（图 2L 和图 S5H）。这些结果得到了进一步验证，因为 EF 衍生的 ASC 富含 PC07.CD83，并高表达 HLA-DR 和 FOXP1，这已被证明可以阻碍生发中心 (GC) 形成并抑制人 PC 分化 (53, 54)（图S5I）。此外，PC07.CD83 中 IGHM 同种型的富集进一步符合骨髓中早期免疫球蛋白 M (IgM) ASC 的特征 (48)（图 S5、J 和 K）。 GC 衍生的 ASC 富含 PC03.DUSP5、PC04.HSPA1A 和 PC10.SPINK2。与 COAD 相比，LIHC 浸润的 ASC 表现出显着更高的 EF 衍生 ASC 标记，例如 CD20、HLA-DR、CD37、CD74、CD83 和 IRF8（图 S5、L 和 M）。这些结果还通过对 ASC 中 IRF8 和 HLA-DR 的 mIHC 染色进行了验证，使用组织微阵列 (TMA) 涵盖 EF 为主的癌症（BLCA、CESC、GIST、LIHC、PAAD、RCC；n = 49）和 GC 为主的癌症（COAD、STAD 和 LC；n = 23）（图 S5N）。 IRF8 ⁺ ASC 的频率（EF 中中位数 = 13.67%，GC 中中位数 = 6.58%；P = 0.0315）和 HLA-DR ⁺ ASC（EF 中中位数 = 50%） ，GC 中的中位数 = 34.09%；P = 0.0066），与 GC 为主的癌症相比，CD79 ⁺ CD20 ⁻ ASC 在 EF 为主的癌症中显着更高（图 S5O）。总而言之，这些结果进一步支持了以下观点：EF 衍生的 ASC 的早期反应与 GC 衍生的 ASC 的延迟反应形成对比。

GC-derived ASCs had more frequent mutations in activation-induced cytidine deaminase (AID) WRC/GYW and WA/TW hotspots than did EF-derived ASCs (55, 56) (fig. S5, P and Q). Specifically, the mutation frequency of GC-derived ASCs in the AID hotspots was 6.05% (4.37 to 8.44%) for IGHA and 4.61% (1.05 to 7.07%) for IGHM. These frequencies were significantly higher than those in EF-derived ASCs, which were 4.65% (2.63 to 6.95%) for IGHA (P = 2.5e⁻¹²) and 0% (0 to 2.93%) for IGHM (P < 2.2e⁻¹⁶). GC-derived ASCs displayed positive selection for replacement mutation in the complementarity-determining region (CDR) of IGHA2 and IGHG2, further supporting their increased CSR (fig. S5R). By contrast, IGHG4 was increased in EF-derived ASCs compared with GC-derived ASCs, which was consistent with a recent report showing IgG4-related autoimmune disease with enriched EF response (57). Likewise, comparing IGHV gene usages between EF- and GC-derived ASCs in tumors revealed that GC-derived ASCs overexpressed IGHV1-8 and IGHV1-24 (fig. S5S), where IGHV1-8 preference was correlated with higher SHM rate (58), indicating their strong antigen-specific affinity. EF-derived ASCs up-regulated a group of IGHV genes (IGHV3-15, IGHV3-23, IGHV3-33, IGHV3-49, IGHV4-34, and IGHV4-59), and increased usage of these IGHV genes has been associated with COVID-19, viral infections, and autoimmune disease (43, 46, 59, 60), implying a polyreactive BCR repertoire. Together, our data indicate that distinct cancer ecosystems hijack two evolutionary paths (EF and GC) and reveal differences in BCR repertoires and transcriptional features between EF- and GC-derived ASCs, highlighting their distinctive roles in antitumor response (Fig. 2M).
与 EF 衍生的 ASC 相比，GC 衍生的 ASC 在激活诱导的胞苷脱氨酶 (AID) WRC/GYW 和 WA/TW 热点中具有更频繁的突变 (55, 56)（图 S5、P 和 Q）。具体而言，AID 热点中 GC 衍生的 ASC 的突变频率对于 IGHA 为 6.05%（4.37 至 8.44%），对于 IGHM 为 4.61%（1.05 至 7.07%）。这些频率显着高于 EF 衍生的 ASC，IGHA 为 4.65%（2.63 至 6.95%）（P = 2.5e ⁻¹² ），IGHM 为 0%（0 至 2.93%）。 P < 2.2e ⁻¹⁶ ）。 GC 衍生的 ASC 在 IGHA2 和 IGHG2 的互补决定区 (CDR) 中显示出对替换突变的正选择，进一步支持了它们增加的 CSR（图 S5R）。相比之下，与 GC 衍生的 ASC 相比，EF 衍生的 ASC 中 IGHG4 增加，这与最近的一份报告一致，该报告显示 IgG4 相关的自身免疫性疾病具有丰富的 EF 反应 (57)。同样，比较肿瘤中 EF 和 GC 衍生的 ASC 之间的 IGHV 基因使用情况表明，GC 衍生的 ASC 过表达 IGHV1-8 和 IGHV1-24（图 S5S），其中 IGHV1-8 偏好与较高的 SHM 率相关 (58) ，表明它们具有很强的抗原特异性亲和力。 EF 衍生的 ASC 上调一组 IGHV 基因（IGHV3-15、IGHV3-23、IGHV3-33、IGHV3-49、IGHV4-34 和 IGHV4-59），并且这些 IGHV 基因的使用增加与COVID-19、病毒感染和自身免疫性疾病 (43, 46, 59, 60)，意味着存在多反应性 BCR 全部。总之，我们的数据表明，不同的癌症生态系统劫持了两条进化路径（EF和GC），并揭示了EF和GC衍生的ASC之间BCR库和转录特征的差异，突出了它们在抗肿瘤反应中的独特作用（图2M）。

Transcription regulation is fundamental for dictating and maintaining B cell identity (61–64), but the tumor-specific regulatory network and the molecular drivers that contribute to TIBs’ identity remain poorly defined. To address this challenge, we applied paired scRNA-seq and scATAC-seq (ATAC, assay for transposase-accessible chromatin sequencing) to characterize the epigenomics of TIBs and obtained chromatin accessibility maps for 53,428 TIBs from nine patients of five cancer types, including GC pathway–dominant cancers (COAD, THCA, LC, and STAD) and EF pathway–dominant cancers (LIHC), with a median of 11,853 distinctly aligned fragments, and ~67.87% of Tn5 insertions within ATAC-seq peaks (Fig. 3A and fig. S6, A to C). By integrating scATAC-seq and scRNA-seq data, we observed eight analogous B cell subpopulations, covering naïve, two ACB, Bm, AtM, GC (DZ and LZ), and PC subsets (Fig. 3A and table S3). As a result, there was strong consistency between subset identities and subset-specific markers annotated in both scRNA-seq and scATAC-seq (Fig. 3B and fig. S6D). Open chromatin accessibility of signature genes for B cell subsets, such as TCL1A, RGS13, ITGAX, and XBP1, further confirmed their cellular identities (Fig. 3C and fig. S6E). To define the transcription factors (TFs) involved in modifying chromatin accessibility of different B cell subsets, we identified TFs whose chromatin accessibility activity was significantly and positively correlated with their gene expression. Such analysis generated a full spectrum of enriched TFs, including known B cell lineage TFs such as PAX5, EBF1, and SPIB (65) (fig. S6, F and G). By combing the footprint and signature of TFs (66), we identified shared and distinctive regulatory patterns across B cell subsets (Fig. 3D). For example, regulatory TF activities harbored shared patterns between similar cell states, such as those active in naïve, ACB, and Bm (i.e., BCL11A, ELF2, SPIB, and STAT6), or in GC_DZ and GC_LZ B cells (i.e., CTCFL, EBF1, HMGA1, and POU2F1) (Fig. 3D). Conversely, a broad spectrum of TFs was specifically enriched in certain B cell subsets (i.e., ETS1 in naïve B; TBX21 and FOSL2 in AtM; POU2F1, MEF2B, and BCL6 in GC B; and IRF2/3/4/5/6/7/8, PRDM1, and ZNF652 in PCs) (Fig. 3, D and E, and fig. S6H). Overall, our data not only unbiasedly shed light on both known and uncharacterized TFs in controlling B cell identity but also provide a comprehensive resource of gene regulatory networks for major B cell subsets across cancers.
转录调控对于决定和维持 B 细胞身份至关重要 (61-64)，但肿瘤特异性调控网络和有助于 TIB 身份的分子驱动因素仍然不明确。为了应对这一挑战，我们应用配对的 scRNA-seq 和 scATAC-seq（ATAC，转座酶可及染色质测序分析）来表征 TIB 的表观基因组学，并获得了来自 5 种癌症类型（包括 GC）的 9 名患者的 53,428 个 TIB 的染色质可及性图谱通路主导型癌症（COAD、THCA、LC 和 STAD）和 EF 通路主导型癌症（LIHC），中位数为 11,853 个明显对齐的片段，ATAC-seq 峰内约 67.87% 的 Tn5 插入（图 3A 和图 S6，A 至 C）。通过整合 scATAC-seq 和 scRNA-seq 数据，我们观察到了八个类似的 B 细胞亚群，涵盖幼稚的、两个 ACB、Bm、AtM、GC（DZ 和 LZ）和 PC 亚群（图 3A 和表 S3）。因此，scRNA-seq 和 scATAC-seq 中注释的子集身份和子集特异性标记之间存在很强的一致性（图 3B 和图 S6D）。 B 细胞亚群特征基因（例如 TCL1A、RGS13、ITGAX 和 XBP1）的开放染色质可及性进一步证实了它们的细胞身份（图 3C 和图 S6E）。为了定义参与改变不同 B 细胞亚群染色质可及性的转录因子 (TF)，我们鉴定了其染色质可及性活性与其基因表达显着正相关的 TF。此类分析生成了全谱富集的 TF，包括已知的 B 细胞谱系 TF，例如 PAX5、EBF1 和 SPIB (65)（图 S6、F 和 G）。通过结合 TF 的足迹和特征 (66)，我们确定了 B 细胞亚群之间共享且独特的调控模式（图 3D）。 例如，调节性 TF 活性在相似的细胞状态之间具有共享模式，例如在幼稚、ACB 和 Bm 中活跃的那些（即 BCL11A、ELF2、SPIB 和 STAT6），或在 GC _DZ 和 GC 中 _LZ B 细胞（即 CTCFL、EBF1、HMGA1 和 POU2F1）（图 3D）。相反，广泛的 TF 在某些 B 细胞亚群中特异性富集（即幼稚 B 中的 ETS1；AtM 中的 TBX21 和 FOSL2；GC B 中的 POU2F1、MEF2B 和 BCL6；以及 IRF2/3/4/5/6/ 7/8、PRDM1 和 PC 中的 ZNF652）（图 3、D 和 E，以及图 S6H）。总体而言，我们的数据不仅公正地揭示了控制 B 细胞身份的已知和未表征的 TF，而且还为跨癌症的主要 B 细胞亚群提供了全面的基因调控网络资源。

Upon closer inspection of Bm, AtM, and PCs, we found that they were predominantly composed of cells from specific EF and GC cancer types, prompting us to investigate potential cancer entity–specific differences in the chromatin landscape of Bm, AtM, and PCs (Fig. 3A). We therefore studied the lineage trajectories of B cells from EF- and GC-dominant cancers (fig. S7, A to C). We identified sequential activities of positive TF regulators, including PAX5, BCL11A, POU2F2, EBF1, REL, TBX21, PRDM1, IRF4, RORA, RUNX2, and XBP1, which promote B cell–lineage specification, maintenance, CSR, and differentiation (62, 64, 65, 67). These activities varied between EF- and GC-dominant cancers along B cell differentiation. Subsequently, we evaluated the TF deviation scores to pinpoint the differential positive TF regulators between EF- and GC-dominant cancers. Bm and AtM in EF-dominant cancers were enriched with AP-1 factors, including FOS and JUN, which are involved in BCR signaling pathway and signify B cell activation and differentiation (68) (fig. S7, D and E). By contrast, Bm and AtM in GC-dominant cancers were enriched with nuclear factor κB (NF-κb) subunits, including REL, RELA, and RELB, which is consistent with their reported functions in GC B cell maintenance and homeostasis (69). Compared with PCs in GC-dominant cancers that showed a high RUNX1-3 deviation score, PCs in EF-dominant cancers showed high activity of TFs that regulate rapid induction of interferon-β (IFN-β) and inflammatory TME, including IRF1/3/4/5/8 and STAT2 (70) (Fig. 3F). To validate this, we confirmed that IRF4 and RUNX2 were enriched in tumor-infiltrated ASCs as compared with other B cells and that IRF4 was up-regulated in LIHC-associated ASCs as compared with COAD-associated ASCs, whereas RUNX2 showed the opposite trend (fig. S7, F and G). These results were substantiated by mIHC staining for IRF4 and RUNX2 in ASCs from a TMA including both EF-dominant (n = 49) and GC-dominant cancers (n = 23) (fig. S7H). The frequency of IRF4⁺ (median in EF = 55.09%, median in GC = 50%; P = 0.0416) in CD79a⁺CD20⁻ ASCs was significantly higher in EF-dominant cancers than in GC-dominant cancers, whereas the frequency of RUNX2⁺ ASCs showed the opposite trend (median in EF = 26.11%, median in GC = 56.89%; P < 0.0001) (fig. S7I). Collectively, these data highlight a dynamic epigenetic regulatory network that functions in fine-tuning B cell differentiation and selection within cancer ecosystems, and they underscore the fundamental role of lineage TFs in managing the balance between EF and GC pathways (Fig. 3G).
经过对 Bm、AtM 和 PC 的仔细检查，我们发现它们主要由来自特定 EF 和 GC 癌症类型的细胞组成，这促使我们研究 Bm、AtM 和 PC 染色质景观中潜在的癌症实体特异性差异。图3A）。因此，我们研究了来自 EF 和 GC 显性癌症的 B 细胞的谱系轨迹（图 S7，A 到 C）。我们确定了正 TF 调节因子的连续活性，包括 PAX5、BCL11A、POU2F2、EBF1、REL、TBX21、PRDM1、IRF4、RORA、RUNX2 和 XBP1，它们促进 B 细胞谱系规范、维持、CSR 和分化（62， 64、65、67）。随着 B 细胞分化，EF 主导型癌症和 GC 主导型癌症之间的这些活性有所不同。随后，我们评估了 TF 偏差评分，以确定 EF 和 GC 为主的癌症之间差异性的阳性 TF 调节因子。 EF 为主的癌症中的 Bm 和 AtM 富含 AP-1 因子，包括 FOS 和 JUN，这些因子参与 BCR 信号通路并意味着 B 细胞激活和分化 (68)（图 S7、D 和 E）。相比之下，GC 显性癌症中的 Bm 和 AtM 富含核因子 κB (NF-κb) 亚基，包括 REL、RELA 和 RELB，这与报道的它们在 GC B 细胞维持和稳态中的功能一致 (69)。与 GC 为主的癌症中表现出较高 RUNX1-3 偏差评分的 PC 相比，EF 为主的癌症中的 PC 显示出调节干扰素-β (IFN-β) 和炎症 TME 快速诱导的 TF 的高活性，包括 IRF1/3 /4/5/8 和 STAT2 (70) (图 3F)。 为了验证这一点，我们证实与其他 B 细胞相比，IRF4 和 RUNX2 在肿瘤浸润的 ASC 中富集，并且与 COAD 相关的 ASC 相比，IRF4 在 LIHC 相关的 ASC 中上调，而 RUNX2 显示相反的趋势。图 S7、F 和 G）。这些结果通过 TMA 的 ASC 中 IRF4 和 RUNX2 的 mIHC 染色得到证实，包括 EF 为主的癌症 (n = 49) 和 GC 为主的癌症 (n = 23)（图 S7H）。 CD79a ⁺ CD20 ⁻ ASC 中 IRF4 ⁺ 的频率（EF 中位数 = 55.09%，GC 中位数 = 50%；P = 0.0416）显着EF 为主的癌症高于 GC 为主的癌症，而 RUNX2 ⁺ ASC 的频率呈现相反的趋势（EF 中中位数 = 26.11%，GC 中中位数 = 56.89%；P < 0.0001）（图S7I）。总的来说，这些数据强调了一个动态表观遗传调控网络，该网络在癌症生态系统内微调 B 细胞分化和选择方面发挥作用，并强调了谱系 TF 在管理 EF 和 GC 途径之间的平衡方面的基本作用（图 3G）。

Recognizing the critical role of AtM B cells as primary progenitors for the EF pathway in the TME, we comprehensively examined their phenotype and function. Compared with the previously defined signature markers in nontumor contexts (21, 22, 24), tumor-infiltrating AtM B cells also showed reduced expression of CD27, CD38, IRF8, and SDC1, together with increased expression of PRDM1, IRF4, and XBP1 in comparison with Bm cells, the well-characterized progenitors for GC-derived ASCs (1). XBP1 is known to dampen B cell proliferation and promote differentiation toward ASCs (71), supporting the differentiation potential of AtM to ASCs (Fig. 4A). AtM B cells exhibited an exhausted and resident memory phenotype, distinguished from the effector phenotype of Bm. In brief, AtM B cells up-regulated immunocheckpoints (PDCD1, CD274, CTLA4, ENTPD1, LAG3, and HAVCR2), exhaustion-associated TFs (TOX, TOX2, ZBED2, BATF, RBPJ, and VDR) (33, 72), and Fc receptor family involved in immunoglobulins and immune complex binding (FCRL2/3/4/5 and CD32A/B) (Fig. 4A). Among them, FCRL4 is linked to tissue-resident memory cells that could inhibit BCR signaling (16). CD32B stands out as the low-affinity inhibitory receptor that inhibits antibody-dependent cellular cytotoxicity (ADCC), antibody-dependent cellular phagocytosis (ADCP), and B cell activation (73). Similarly to exhausted T cells with overactivated TCR signaling, AtM B cells also highly expressed genes of BCR signaling (SYK), immunomodulation, and activation (TLR7/9, CD80, CD86, and CD72). These distinctive phenotypes were further corroborated by flow cytometry (Fig. 4B and fig. S8A). These data collectively demonstrate the distinctive immunophenotypes and molecular features of extrafollicular AtM B cells.
认识到 AtM B 细胞作为 TME 中 EF 途径的主要祖细胞的关键作用，我们全面检查了它们的表型和功能。与之前在非肿瘤环境中定义的特征标记相比 (21,22,24)，肿瘤浸润 AtM B 细胞还表现出 CD27、CD38、IRF8 和 SDC1 表达减少，以及 PRDM1、IRF4 和 XBP1 表达增加。与 Bm 细胞（GC 衍生的 ASC 的特征明确的祖细胞）进行比较 (1)。已知 XBP1 会抑制 B 细胞增殖并促进向 ASC 分化 (71)，支持 AtM 向 ASC 的分化潜力（图 4A）。 AtM B 细胞表现出耗尽且常驻的记忆表型，与 Bm 的效应子表型不同。简而言之，AtM B 细胞上调免疫检查点（PDCD1、CD274、CTLA4、ENTPD1、LAG3 和 HAVCR2）、耗竭相关 TF（TOX、TOX2、ZBED2、BATF、RBPJ 和 VDR）(33, 72)， Fc受体家族参与免疫球蛋白和免疫复合物结合（FCRL2/3/4/5和CD32A/B）（图4A）。其中，FCRL4 与组织驻留记忆细胞相关，可以抑制 BCR 信号传导 (16)。 CD32B 是一种突出的低亲和力抑制性受体，可抑制抗体依赖性细胞毒性 (ADCC)、抗体依赖性细胞吞噬作用 (ADCP) 和 B 细胞活化 (73)。与 TCR 信号过度激活的疲惫 T 细胞类似，AtM B 细胞也高度表达 BCR 信号 (SYK)、免疫调节和激活基因（TLR7/9、CD80、CD86 和 CD72）。这些独特的表型通过流式细胞术得到进一步证实（图4B和图S8A）。这些数据共同证明了滤泡外 AtM B 细胞独特的免疫表型和分子特征。

To further explore the immune repertoire of extrafollicular AtM B cells, isotype analysis indicated that tumor-infiltrating AtM B cells were less switched and contained higher IGHD and IGHM than did Bm cells (P < 0.01), whereas IGHA1, IGHA2, and IGHG2 were prevalent in Bm (P < 0.01) (fig. S8, B and C). Moreover, supernatants of tumor AtM B cells from LIHC patients, collected after 11 days of in vitro activation and differentiation, contained lower IgG, IgA, and IgM than did Bm (fig. S8D). Again, antigen microarrays testing the reactivity of these supernatant antibodies revealed that ASCs derived from AtM B cells produced higher levels of autoantibodies, whereas Bm-derived ASCs produced higher levels of tumor-associated antigen-reactive antibodies (Fig. 4C) for all three types: IgG, IgA, and IgM (fig. S8E). In line with these findings, we observed high SHM in Bm but low or medium levels in AtM B cells (fig. S8F). Collectively, our data show that exhausted AtM B cells exhibit impaired antibody production capacity, bystander activation, and weak tumor reactivity.
为了进一步探索滤泡外 AtM B 细胞的免疫库，同种型分析表明，与 Bm 细胞相比，肿瘤浸润的 AtM B 细胞转换较少，且含有更高的 IGHD 和 IGHM (P < 0.01)，而 IGHA1、IGHA2 和 IGHG2 普遍存在单位为 Bm (P < 0.01)（图 S8、B 和 C）。此外，体外活化和分化11天后收集的来自LIHC患者的肿瘤AtM B细胞上清液中的IgG、IgA和IgM含量低于Bm（图S8D）。同样，测试这些上清抗体反应性的抗原微阵列显示，对于所有三种类型，源自 AtM B 细胞的 ASC 产生更高水平的自身抗体，而源自 Bm 的 ASC 产生更高水平的肿瘤相关抗原反应性抗体（图 4C） ：IgG、IgA 和 IgM（图 S8E）。与这些发现一致，我们观察到 Bm 中的 SHM 水平较高，但 AtM B 细胞中的 SHM 水平较低或中等（图 S8F）。总的来说，我们的数据表明，耗尽的 AtM B 细胞表现出抗体生产能力受损、旁观者激活和肿瘤反应性较弱。

Another crucial issue is whether AtM B cells could enter GC or originate from bona fide GC B cells (19, 74–76). To address this, we jointly analyzed the BCR repertoire, developmental trajectory, epigenomic regulation, and gene expression features. We observed comparable IGHV usage, CDR3 amino acid length, and SHM between AtM and three GC B cells. These data supported the notion that AtM and GC B clones were generally independent rather than interchangeable (fig. S8, G and H). We thus reconstructed the lineage tree for the most expanded clones of AtM and GC B cells, showing that clonal lineages were predominantly derived from either AtM or GC B cells (fig. S8I). Pseudotime trajectory analysis confirmed that AtM and GC B cells were completely separated on two independent branches, with naïve B cells at the beginning and AtM (path 1) plus cycling GC B cells (path 2) in the terminal branches (Fig. 4D). Along the trajectory, high-affinity signature score and CD24, CD38, and SELL gradually up-regulated in path 2, whereas low affinity and exhaustion-associated signature score–related genes (i.e., PDCD1, ENTPD1, and HAVCR2) correspondingly up-regulated in path 1, recapitulating B cell development from the naïve state through the activated state and finally to the AtM state (Fig. 4E and fig. S8, J and K). During this process, the degree of exhaustion progressively escalated. This developmental trajectory was uniformly observed in different cancer types (fig. S9A) and was confirmed by scATAC-seq data (fig. S9B).
另一个关键问题是 AtM B 细胞是否可以进入 GC 或源自真正的 GC B 细胞 (19, 74–76)。为了解决这个问题，我们共同分析了 BCR 库、发育轨迹、表观基因组调控和基因表达特征。我们观察到 AtM 和三个 GC B 细胞之间的 IGHV 使用、CDR3 氨基酸长度和 SHM 具有可比性。这些数据支持这样的观点：AtM 和 GC B 克隆通常是独立的而不是可互换的（图 S8、G 和 H）。因此，我们重建了 AtM 和 GC B 细胞最扩展克隆的谱系树，表明克隆谱系主要源自 AtM 或 GC B 细胞（图 S8I）。伪时间轨迹分析证实AtM和GC B细胞在两个独立的分支上完全分离，初始B细胞在起始处，AtM（路径1）加上循环GC B细胞（路径2）在末端分支（图4D）。沿着轨迹，高亲和力特征得分和 CD24、CD38 和 SELL 在路径 2 中逐渐上调，而低亲和力和耗竭相关特征得分相关基因（即 PDCD1、ENTPD1 和 HAVCR2）相应上调在路径 1 中，概括了 B 细胞从初始状态到激活状态，最后到 AtM 状态的发育（图 4E 和图 S8、J 和 K）。在这个过程中，疲惫的程度逐渐升级。这种发展轨迹在不同的癌症类型中得到了一致的观察（图S9A），并得到了scATAC-seq数据的证实（图S9B）。

We further identified the dynamic expression, chromatin activity, and footprint of AtM. Compared with Bm and GC B cells, AtM B cells exhibited high activity of TFs involved in CSR and exhausted phenotype, and they enhanced immune function of isotype-specific IgG⁺ Bm (33, 67, 77, 78), including TBX21 and BATF (fig. S9, C and D), indicating distinct regulatory programs (Fig. 3D). CSR-related signature score and genes (APEX1, APEX2, XRCC5, XRCC6, POLD2, and AICDA) progressively up-regulated along the trajectory, suggesting that AtM B cells may undergo CSR without experiencing GC response (Fig. 4E and fig. S8K). Similarly, AtM had shared BCR with IFIT3⁺ B in EF-dominant cancers, including HNSC, OV, GIST, BRCA and CESC, suggesting that naïve B cells induced by interferons could directionally differentiate into AtM without entering the GC (fig. S9, E to G). Collectively, these findings suggest that AtM and GC B cells underwent separate developmental processes leading to EF- and GC-derived ASCs.
我们进一步确定了 AtM 的动态表达、染色质活性和足迹。与 Bm 和 GC B 细胞相比，AtM B 细胞表现出参与 CSR 和耗尽表型的 TF 的高活性，并且增强了同种型特异性 IgG ⁺ Bm 的免疫功能 (33, 67, 77, 78) ，包括 TBX21 和 BATF（图 S9、C 和 D），表明不同的监管计划（图 3D）。 CSR相关的特征评分和基因（APEX1、APEX2、XRCC5、XRCC6、POLD2和AICDA）沿着轨迹逐渐上调，表明AtM B细胞可能在不经历GC反应的情况下经历CSR（图4E和图S8K） 。同样，在 EF 主导的癌症（包括 HNSC、OV、GIST、BRCA 和 CESC）中，AtM 与 IFIT3 ⁺ B 共享 BCR，这表明干扰素诱导的幼稚 B 细胞可以在不进入 GC 的情况下定向分化为 AtM （图 S9，E 至 G）。总的来说，这些发现表明 AtM 和 GC B 细胞经历了独立的发育过程，导致 EF 和 GC 衍生的 ASC。

TLSs are often formed by B cells with varying sizes and cellular compositions. However, the spatial architecture and relationship between TLSs and TIBs, specifically AtM B cells, remain unclear. We found that AtM B cells highly expressed TLS signature (7) and were confirmed in pan-cancer TCGA data, indicating their potential localization in TLSs (fig. S10, A and B). We thus performed mIHC staining of TMA containing EF-dominant cancers [LIHC (n = 358; n = 180) and PAAD (n = 48)] and GC-dominant cancers [COAD (n = 98), STAD (n = 52), and LC (n = 90)] to characterize TLSs and AtM B cells (table S1). By using our well-established TLS quantification pipeline (8), we divided tumor tissue into TLS and non-TLS regions and observed a significant correlation between AtM B cells and TLSs, suggesting a potential role for AtM B cells in TLS formation (fig. S10, C and D).
TLS 通常由不同大小和细胞组成的 B 细胞形成。然而，TLS 和 TIB（特别是 AtM B 细胞）之间的空间结构和关系仍不清楚。我们发现 AtM B 细胞高度表达 TLS 特征 (7)，并在泛癌 TCGA 数据中得到证实，表明它们在 TLS 中的潜在定位（图 S10、A 和 B）。因此，我们对含有 EF 为主的癌症 [LIHC (n = 358; n = 180) 和 PAAD (n = 48)] 和 GC 为主的癌症 [COAD (n = 98)、STAD (n = 52) 的 TMA 进行了 mIHC 染色和 LC (n = 90)] 来表征 TLS 和 AtM B 细胞（表 S1）。通过使用我们完善的 TLS 量化流程 (8)，我们将肿瘤组织分为 TLS 和非 TLS 区域，并观察到 ​​AtM B 细胞和 TLS 之间的显着相关性，表明 AtM B 细胞在 TLS 形成中的潜在作用（图 1）。 S10、C 和 D)。

Given the clinical significance of TLS maturation [mature TLSs: stage III and IV GC-like structure containing follicular dendritic cells (fDCs); immature TLSs: stage I and II lymphoid aggregates] and the rarity of GC-like TLS in TME (8, 79–82), we further screened whole-tissue sections with different TLS stages from LIHC (n = 17) and COAD (n = 6). TLSs were classified into four stages on the basis of the cell number of T and B cells together with fDC staining (8) (Fig. 5A). AtM B cells were significantly enriched in immature TLSs, whereas Bm was mainly located in mature TLSs in both COAD and LIHC (Fig. 5, B and C). No difference in ASC density was observed across different TLS stages (fig. S10E). As expected, AtM B cells were predominantly enriched in EF-dominant cancers, such as GIST, CESC, LIHC, and HNSC, whereas GC B cells were highly abundant in GC-dominant cancers, such as COAD, THCA, and LC (fig. S10F), indicating that GC-dominant cancers mainly contained mature TLSs versus immature ones in EF-dominant cancers. Cancer with equal presence of EF and GC paths, such as OV, showed moderate infiltration of AtM and GC_LZ B cells as compared with other cancers. Spatially, we calculated the density of AtM B cells within 200 μm of the inner and outer edges of TLSs or follicles and the distance of AtM B cells to the interface. We observed that AtM B cells were predominantly located in the center of immature TLSs, which is consistent with early EF response and facilitation of TLS formation (Fig. 5, D and E, and fig. S10, G and H). As TLS maturated, AtM B cells migrated to the periphery, while naïve B cells consistently resided in the center of TLSs (fig. S10C). These data validate the GC-independent development of AtM B cells and also suggest that TLS maturation stages drive distinct B cell differentiation in EF- and GC-dominant cancers.
鉴于 TLS 成熟的临床意义[成熟 TLS：含有滤泡树突状细胞 (fDC) 的 III 期和 IV 期 GC 样结构；不成熟的 TLS：I 期和 II 期淋巴聚集体]以及 TME 中 GC 样 TLS 的罕见性 (8, 79–82)，我们进一步从 LIHC (n = 17) 和 COAD (n = 17) 中筛选具有不同 TLS 阶段的全组织切片= 6）。根据T细胞和B细胞的细胞数量以及fDC染色将TLS分为四个阶段（8）（图5A）。在 COAD 和 LIHC 中，AtM B 细胞在未成熟 TLS 中显着富集，而 Bm 主要位于成熟 TLS 中（图 5，B 和 C）。不同 TLS 阶段的 ASC 密度没有观察到差异（图 S10E）。正如预期的那样，AtM B 细胞主要富集在 EF 为主的癌症中，如 GIST、CESC、LIHC 和 HNSC，而 GC B 细胞在 GC 为主的癌症中高度丰富，如 COAD、THCA 和 LC（图 2）。 S10F)，表明 GC 为主的癌症主要含有成熟的 TLS，而 EF 为主的癌症则主要含有未成熟的 TLS。与其他癌症相比，EF 和 GC 路径均等存在的癌症（例如 OV）显示 AtM 和 GC _LZ B 细胞的中度浸润。在空间上，我们计算了 TLS 或滤泡内外边缘 200 μm 范围内 AtM B 细胞的密度以及 AtM B 细胞到界面的距离。我们观察到 AtM B 细胞主要位于未成熟 TLS 的中心，这与早期 EF 反应和 TLS 形成的促进一致（图 5、D 和 E，以及图 S10、G 和 H）。随着 TLS 的成熟，AtM B 细胞迁移到外围，而幼稚 B 细胞始终驻留在 TLS 的中心（图 S10C）。 这些数据验证了 AtM B 细胞的不依赖 GC 的发育，并且还表明 TLS 成熟阶段驱动 EF 和 GC 显性癌症中不同的 B 细胞分化。

This particular location of AtM B cells was echoed by the up-regulation of homing receptors such as CCR7, CXCR3, and EBI2 (encoding GPR183) (Fig. 4A), which could drive AtM B cells into the B-zone–T-zone boundary to await T cell help (52, 83). Indeed, the strong correlation between AtM B cells and T follicular helper (Tfh) or type 1 T helper (Th1) cells, which secrete IFNγ, IL-21, and CD40L in viral infection (76, 84), points to the possibility of cross-talk between these T cell types and AtM B cell development. Flow cytometry and pan-cancer TCGA data deconvoluted by BayesPrism (85), a Bayesian statistical model that infers cell type composition using our paired CD45⁺ scRNA-seq data as reference, showed that AtM B cells positively correlated with PD1^Hi exhausted T cells (Fig. 5F and fig. S11, A to D), especially PD1^HiCD4⁺ T cells exhibiting a PD1^HiCXCL13⁺CXCR5⁻ phenotype (named as peripheral helper T cells, Tph) that can promote TLS formation and B cell terminal differentiation in autoimmune diseases (86, 87). Ligand-receptor analysis provided additional evidence supporting the cross-talk between PD1^HiCD4⁺ Tph cells and AtM B cells, particularly through the IL-21R–IL-21 axis, as compared with Bm cells. This interaction implies a potential role for PD1^HiCD4⁺ Tph cells in promoting AtM B cell differentiation (Fig. 5G and fig. S11E), which was also validated by another independent algorithm, CellChat (88) (fig. S11F). Visualizing CD4⁺PD1^Hi Tph and AtM B cells through published spatial transcriptomics datasets [81 datasets from nine cancer types, including BRCA, CESC, COAD, LIHC, LGG, OV, prostate adenocarcinoma (PRAD), RCC, and skin cutaneous melanoma (SKCM)] revealed a positive correlation between them in TLSs (Fig. 5, H and I, and table S4), which was further supported by mIHC staining on LIHC TMA (n = 180), showing significantly higher CD4⁺PD1^Hi Tph cell density and frequency in close proximity to AtM B cells in TLS than in non-TLS regions of both tumor and adjacent tissues (Fig. 5, J and K). To validate this, LIHC-infiltrated PD1^HiCD4⁺ Tph, Th, and regulatory T (T_reg) cells were sorted and co-cultured with either peripheral total B or sorted naïve B cells from healthy donors (Fig. 5L and fig. S11G). The results showed that Tph cells induced differentiation of AtM B cells more efficiently than Th and T_reg cells, which was significantly attenuated by IL-21R blockade (Fig. 5M and fig. S11H). Together, these data suggest that tumor-infiltrated Tph cells significantly contribute to AtM B cell differentiation.
AtM B 细胞的这一特定位置与 CCR7、CXCR3 和 EBI2（编码 GPR183）等归巢受体的上调相呼应（图 4A），这可以驱动 AtM B 细胞进入 B 区-T 区等待 T 细胞帮助的边界 (52, 83)。事实上，AtM B 细胞与滤泡辅助性 T (Tfh) 或 1 型辅助性 T (Th1) 细胞（在病毒感染中分泌 IFNγ、IL-21 和 CD40L）之间存在很强的相关性 (76, 84)，这表明了以下可能性：这些 T 细胞类型与 AtM B 细胞发育之间的串扰。流式细胞术和泛癌 TCGA 数据由 BayesPrism (85) 解卷积，BayesPrism 是一种贝叶斯统计模型，使用我们的配对 CD45 ⁺ scRNA-seq 数据作为参考来推断细胞类型组成，结果表明 AtM B 细胞与PD1 ^Hi 耗尽 T 细胞（图 5F 和图 S11，A 至 D），尤其是 PD1 ^Hi CD4 ⁺ T 细胞，表现出 PD1 ^Hi CXCL13 ⁺ CXCR5 ⁻ 表型（称为外周辅助 T 细胞，Tph）可促进自身免疫性疾病中 TLS 形成和 B 细胞终末分化 (86, 87)。配体受体分析提供了额外的证据支持 PD1 ^Hi CD4 ⁺ Tph 细胞和 AtM B 细胞之间的串扰，特别是通过 IL-21R–IL-21 轴。与 Bm 细胞。这种相互作用意味着 PD1 ^Hi CD4 ⁺ Tph 细胞在促进 AtM B 细胞分化中的潜在作用（图 5G 和图 S11E），这也得到了另一个独立算法的验证， CellChat (88)（图 S11F）。 通过已发布的空间转录组数据集可视化 CD4 ⁺ PD1 ^Hi Tph 和 AtM B 细胞 [来自九种癌症类型的 81 个数据集，包括 BRCA、CESC、COAD、LIHC、LGG、OV、前列腺腺癌(PRAD)、RCC 和皮肤黑色素瘤 (SKCM)] 显示它们在 TLS 中呈正相关（图 5、H 和 I 以及表 S4），LIHC TMA 上的 mIHC 染色进一步支持了这一点（n = 180） ），与肿瘤和邻近组织的非 TLS 区域相比，TLS 中靠近 AtM B 细胞的 CD4 ⁺ PD1 ^Hi Tph 细胞密度和频率显着更高（图 1）。 5、J 和 K）。为了验证这一点，对 LIHC 浸润的 PD1 ^Hi CD4 ⁺ Tph、Th 和调节性 T (T _reg ) 细胞进行分选并与任一外周血共培养。来自健康捐赠者的总 B 细胞或分选的幼稚 B 细胞（图 5L 和图 S11G）。结果表明，Tph 细胞比 Th 和 T _reg 细胞更有效地诱导 AtM B 细胞分化，并且 IL-21R 阻断显着减弱这种分化（图 5M 和图 S11H）。总之，这些数据表明肿瘤浸润的 Tph 细胞对 AtM B 细胞分化有显着贡献。

We next examined what factors mechanistically drive the primary response toward EF versus GC paths in TME. Pathway analysis of B cells revealed a strong variance of metabolic pathways across human cancer types, suggesting the potential metabolic regulation of B cell identity (fig. S12A). We computed the metabolic pathway scores of Bm and AtM B cells using scMetabolism (89) and unsupervised clustering of metabolism-associated genes (Fig. 6A). Glutamine, glutamate, and sphingolipid metabolism were up-regulated in tumor-infiltrated AtM B cells, whereas arachidonic acid metabolism was significantly up-regulated in Bm cells. The core genes in these pathways, such as glutaminase (GLS), GLA, and PTGES3, showed similar trends (fig. S12B). The results were observed in individual types of cancer without cancer-type preference (fig. S12C). Given the essential role and large abundance of glutamine in TME (90–92), we investigated the association between glutamine and B cell immunophenotypes in our cohort. Through mass spectrometry–based untargeted metabolomics, our data indicated that EF response–dominant cancers (LIHC, CESC, and RCC) have significantly higher glutamine than those in GC-dominant cancers (STAD and LC) (fig. S12D). Moreover, glutamine was significantly higher in tumors than in adjacent normal tissues in LIHC (n = 109) (fig. S12E), whereas COAD and STAD, two GC-dominant cancers, exhibited lower glutamine levels than adjacent normal tissues (93). Thus, we hypothesize that glutamine metabolism may have a potential role in priming EF response in TME.
接下来我们研究了哪些因素在机制上驱动 TME 中对 EF 与 GC 路径的主要响应。 B 细胞的通路分析揭示了不同人类癌症类型的代谢通路存在很大差异，表明 B 细胞身份的潜在代谢调节（图 S12A）。我们使用 scMetabolism (89) 和代谢相关基因的无监督聚类计算了 Bm 和 AtM B 细胞的代谢途径评分（图 6A）。肿瘤浸润的 AtM B 细胞中谷氨酰胺、谷氨酸和鞘脂代谢上调，而 Bm 细胞中花生四烯酸代谢显着上调。这些途径中的核心基因，例如谷氨酰胺酶（GLS）、GLA和PTGES3，表现出相似的趋势（图S12B）。结果是在没有癌症类型偏好的个体癌症类型中观察到的（图S12C）。鉴于谷氨酰胺在 TME 中的重要作用和大量丰度 (90-92)，我们在队列中研究了谷氨酰胺与 B 细胞免疫表型之间的关联。通过基于质谱的非靶向代谢组学，我们的数据表明，EF 反应为主的癌症（LIHC、CESC 和 RCC）的谷氨酰胺含量显着高于 GC 为主的癌症（STAD 和 LC）（图 S12D）。此外，LIHC 中肿瘤中的谷氨酰胺显着高于邻近正常组织 (n = 109)（图 S12E），而 COAD 和 STAD 这两种 GC 为主的癌症的谷氨酰胺水平低于邻近正常组织 (93)。因此，我们假设谷氨酰胺代谢可能在 TME 中启动 EF 反应中具有潜在作用。

To test this hypothesis, we collected B cells from healthy blood and stimulated them with R848 (a TLR7 agonist), IFNγ, and glutamine. Similarly to IFNγ’s role in the autoimmune disease (21, 22), glutamine significantly induced AtM differentiation compared with R848 alone, whereas GLS inhibition (CB839) significantly attenuated it (Fig. 6, B and C). As control, Bm cells were significantly reduced in response to IFNγ and glutamine stimulation, suggesting the opposing regulatory mechanisms for AtM and Bm cells by glutamine. Moreover, AtM gradually increased and peaked around day 4 after in vitro glutamine stimulation in a dose-dependent manner that maximized at around 10-fold of physiological concentration (fig. S12F). We monitored the oxygen consumption rate and adenosine triphosphate (ATP) production of B cells under glutamine stimulation and observed a global reduction of mitochondrial respiration plus glycolysis (Fig. 6D). This indicates that glutamine may remodel the metabolic traits of B cells, leading us to hypothesize that glutamine suppresses the clonal expansion or positive selection of B cells, thus shifting B cell differentiation toward AtM B cells (94).
为了验证这一假设，我们从健康血液中收集了 B 细胞，并用 R848（一种 TLR7 激动剂）、IFNγ 和谷氨酰胺刺激它们。与 IFNγ 在自身免疫性疾病中的作用类似 (21, 22)，与单独使用 R848 相比，谷氨酰胺显着诱导 AtM 分化，而 GLS 抑制 (CB839) 显着减弱 AtM 分化（图 6、B 和 C）。作为对照，Bm 细胞对 IFNγ 和谷氨酰胺刺激的反应显着减少，表明谷氨酰胺对 AtM 和 Bm 细胞的相反调节机制。此外，AtM 逐渐增加，并在体外谷氨酰胺刺激后第 4 天左右达到峰值，呈剂量依赖性，在生理浓度的 10 倍左右达到最大值（图 S12F）。我们监测了谷氨酰胺刺激下 B 细胞的耗氧率和三磷酸腺苷 (ATP) 产量，并观察到线粒体呼吸和糖酵解的整体减少（图 6D）。这表明谷氨酰胺可能重塑 B 细胞的代谢特征，使我们推测谷氨酰胺会抑制 B 细胞的克隆扩增或正选择，从而使 B 细胞分化为 AtM B 细胞 (94)。

We then analyzed the intracellular metabolites of peripheral B cells after R848 and glutamine stimulation and found increased levels of glutamate, α-ketoglutarate (α-KG), and a high ratio of glutamine to glutamate as compared with R848 stimulation alone (fig. S12G). CB839 treatment inhibited this trend, suggesting that GLS plays an essential role in B cell glutaminolysis. The metabolic fate of ¹³C-labeled glutamine after R848 treatment confirmed that glutamine was catalyzed into glutamate (M5 labeled) and entered into the tricarboxylic acid (TCA) cycle through α-KG (M5 labeled), succinate (M4 labeled), fumarate (M4 labeled), pyruvate (M3 labeled), and lactate (M3 labeled), which is consistent with the reported glutaminolysis route (95) (Fig. 6E). Given that glutamine-derived α-KG could regulate T-bet expression in Th1 cells (90, 91), we next asked whether α-KG was also critical for AtM differentiation by supplementing with the cell-permeable α-KG analog dimethyl-αKG (DMαKG). Indeed, AtM differentiation in GLS-inhibited cells was restored when supplemented with DMαKG, comparable to direct stimulation with glutamine. In addition, direct supplementation of DMαKG with R848 induced a similar level of AtM differentiation, suggesting that glutamine-derived α-KG can directly regulate AtM differentiation (Fig. 6F).
然后，我们分析了 R848 和谷氨酰胺刺激后外周 B 细胞的细胞内代谢物，发现与单独 R848 刺激相比，谷氨酸、α-酮戊二酸 (α-KG) 水平增加，并且谷氨酰胺与谷氨酸的比率较高（图 S12G） 。 CB839 治疗抑制了这种趋势，表明 GLS 在 B 细胞谷氨酰胺分解中起着重要作用。 ¹³ C标记的谷氨酰胺在R848处理后的代谢命运证实，谷氨酰胺被催化成谷氨酸（M5标记），并通过α-KG（M5标记）、琥珀酸（M5标记）进入三羧酸（TCA）循环。 M4标记）、富马酸（M4标记）、丙酮酸（M3标记）和乳酸（M3标记），这与报道的谷氨酰胺分解途径一致（95）（图6E）。鉴于谷氨酰胺衍生的 α-KG 可以调节 Th1 细胞中的 T-bet 表达 (90, 91)，我们接下来通过补充细胞渗透性 α-KG 类似物二甲基-αKG 来询问 α-KG 是否对 AtM 分化也至关重要(DMαKG)。事实上，当补充 DMαKG 时，GLS 抑制细胞中的 AtM 分化得到恢复，与谷氨酰胺的直接刺激相当。此外，直接补充R848的DMαKG诱导了类似水平的AtM分化，表明谷氨酰胺衍生的α-KG可以直接调节AtM分化（图6F）。

Considering the impact of α-KG on histone methylation and chromatin accessibility remodeling (91, 96), we next questioned its role in establishing the epigenetic identity of extrafollicular AtM B cells. We found that glutamine stimulation led to increased H3K27 trimethylation, whereas CB839 had no effect on H3K27 itself (Fig. 6C). Further ATAC-seq and RNA-seq revealed that glutamine-treated B cells showed robust activation of AtM identity peaks (Fig. 6G), indicating their acquired extrafollicular identity. Consistently, we also observed enriched TF motifs for AtM cell differentiation, including TBX21 and BATF (fig. S12H). We further used glutamine to stimulate B cells with T-bet overexpression and observed a significant increase in AtM B cells in the overexpression group as compared with the control, providing additional support for the notion that glutamine induces AtM differentiation through the regulation of TBX21 (fig. S12, I and J). Pathway enrichment analysis again demonstrated that EF-associated PI3K-AKT-mTORC1 and xenobiotic-metabolism pathways were activated, which is consistent with our scRNA-seq data (mTORC1, mammalian target of rapamycin complex 1) (Fig. 6H). Several regulators of the mTORC1 pathway were also increased under glutamine stimulation, including PIK3IP1 and AKT1 (fig. S12K). The levels of phospho-mTORC1 and its downstream targets phospho-S6 and phospho-AKT showed a greater increase in tumor-infiltrated AtM B cells than in Bm and naïve B cells, suggesting that mTORC1 signaling might participate in the glutamine regulation (Fig. 6, I and J). Further inhibition of the mTORC1 pathway by rapamycin significantly reduced AtM differentiation under glutamine stimulation (Fig. 6, K and L). Contradictorily, R848-treated B cells were enriched with oxidative phosphorylation and fatty-acid metabolism pathways and showed high expression of CSR family genes versus glutamine-treated B cells, confirming the above hypothesis (fig. S12L). Collectively, these results demonstrate that glutamine metabolism establishes the epigenetic identity of extrafollicular AtM B cells.
考虑到 α-KG 对组蛋白甲基化和染色质可及性重塑的影响 (91, 96)，我们接下来质疑其在建立滤泡外 AtM B 细胞表观遗传身份中的作用。我们发现谷氨酰胺刺激导致H3K27三甲基化增加，而CB839对H3K27本身没有影响（图6C）。进一步的 ATAC-seq 和 RNA-seq 显示，谷氨酰胺处理的 B 细胞表现出 AtM 身份峰的强烈激活（图 6G），表明它们获得了滤泡外身份。一致地，我们还观察到 AtM 细胞分化的丰富 TF 基序，包括 TBX21 和 BATF（图 S12H）。我们进一步使用谷氨酰胺刺激 T-bet 过表达的 B 细胞，并观察到过表达组中 AtM B 细胞与对照组相比显着增加，这为谷氨酰胺通过调节 TBX21 诱导 AtM 分化的观点提供了额外的支持（图S12、I 和 J)。通路富集分析再次证明EF相关的PI3K-AKT-mTORC1和异生物质代谢通路被激活，这与我们的scRNA-seq数据（mTORC1，雷帕霉素复合物1的哺乳动物靶标）一致（图6H）。 mTORC1 通路的几个调节因子在谷氨酰胺刺激下也增加，包括 PIK3IP1 和 AKT1（图 S12K）。磷酸化 mTORC1 及其下游靶标磷酸化 S6 和磷酸化 AKT 的水平在肿瘤浸润的 AtM B 细胞中显示出比 Bm 和幼稚 B 细胞更大的增加，表明 mTORC1 信号传导可能参与谷氨酰胺调节（图 6） ，I 和 J）。雷帕霉素对 mTORC1 通路的进一步抑制显着降低了谷氨酰胺刺激下的 AtM 分化（图 6，K 和 L）。 相反，与谷氨酰胺处理的 B 细胞相比，R848 处理的 B 细胞富含氧化磷酸化和脂肪酸代谢途径，并且显示出 CSR 家族基因的高表达，证实了上述假设（图 S12L）。总的来说，这些结果表明谷氨酰胺代谢建立了滤泡外 AtM B 细胞的表观遗传特性。

In light of the spatial colocalization of extrafollicular AtM B cells and T cells within TLSs, we explored the potential impact of AtM B cells on T cells (Fig. 5A). After stimulation, the sorted LIHC-infiltrated AtM B cells secreted more IL-10 and transforming growth factor–β (TGFβ) and expressed higher PDL-1 than did non-AtM B cells, and this was validated by glutamine-induced B cells as compared with the control (fig. S13, A to C). We next induced peripheral B cells into AtM with glutamine and co-cultured with peripheral CD3⁺ T cells in vitro. AtM B cells could reduce the proliferation of both CD4⁺ and CD8⁺ T cells (Fig. 6M and fig. S13, D and E) and impair the ability of T cells to produce IFNγ, tumor necrosis factor–α (TNFα), and granzyme B (Fig. 6N and fig. S13, F and G). AtM B cells promoted the differentiation toward T_regs and exhausted T cells (fig. S13H), which was akin to previous studies showing that CD27⁻ exhausted B cells and B cells within immature TLSs correlated and colocalized with T_regs (41, 97).
鉴于滤泡外AtM B细胞和TLS内T细胞的空间共定位，我们探索了AtM B细胞对T细胞的潜在影响（图5A）。刺激后，分选的 LIHC 浸润的 AtM B 细胞比非 AtM B 细胞分泌更多的 IL-10 和转化生长因子-β (TGFβ)，并表达更高的 PDL-1，这通过谷氨酰胺诱导的 B 细胞得到验证：与对照相比（图 S13，A 至 C）。接下来，我们用谷氨酰胺诱导外周 B 细胞进入 AtM，并在体外与外周 CD3 ⁺ T 细胞共培养。 AtM B 细胞可以减少 CD4 ⁺ 和 CD8 ⁺ T 细胞的增殖（图 6M 和图 S13、D 和 E），并损害 T 细胞产生的能力IFNγ、肿瘤坏死因子-α (TNFα) 和颗粒酶 B（图 6N 和图 S13、F 和 G）。 AtM B 细胞促进向 T _regs 分化并耗尽 T 细胞（图 S13H），这与之前的研究类似，表明 CD27 ⁻ 耗尽了 B 细胞和未成熟 TLS 内的 B 细胞与 T _regs 相关并共定位 (41, 97)。

Last, we explored the prognostic values of Bm and AtM B cells across cancers. Bm cells were associated with favorable prognosis, whereas AtM B cells correlated with worse survival in four independent cohorts (COAD, STAD, LC, and LIHC) by mIHC, irrespective of EF- or GC-dominant cancer (Fig. 6O and fig. S13I). Moreover, in published cohorts of melanoma (n = 26) (98) and lung cancer (n = 21) (99) with anti-PD1 treatment, the abundance of AtM B cell significantly correlated with treatment resistance, whereas Bm cells correlated with improved responses and longer survival (Fig. 6, P and Q). Collectively, these observations suggest that EF-derived B cells are associated with immature TLSs and exhausted T cells, leading to immunotherapeutic resistance and poor prognosis in certain cancers.
最后，我们探讨了 Bm 和 AtM B 细胞在癌症中的预后价值。 Bm 细胞与良好的预后相关，而 AtM B 细胞与 mIHC 的四个独立队列（COAD、STAD、LC 和 LIHC）中较差的生存相关，无论是 EF 或 GC 为主的癌症（图 6O 和图 S13I） ）。此外，在已发表的接受抗 PD1 治疗的黑色素瘤 (n = 26) (98) 和肺癌 (n = 21) (99) 队列中，AtM B 细胞的丰度与治疗耐药性显着相关，而 Bm 细胞与改善的治疗耐药性相关。反应和更长的生存期（图 6，P 和 Q）。总的来说，这些观察结果表明，EF 衍生的 B 细胞与不成熟的 TLS 和耗尽的 T 细胞有关，导致某些癌症的免疫治疗耐药性和预后不良。

The scRNA-seq data of B cells were obtained from both previously published datasets and newly generated data. For the newly generated scRNA-seq data, paired cancer, adjacent normal, LN_Met, and blood samples were collected from patients who underwent surgical resection at Zhongshan Hospital Fudan University with written consent and approval of the Institutional Review Board–approved protocols. Matched cancer, adjacent normal, and LN_Met tissues were digested with Tumor Dissociation Kit or Collagenase IV using GentleMACS (Miltenyi Biotec), according to the manufacturer’s instructions. CD19⁺ B cells were isolated from mononuclear cells (MNCs) and peripheral blood mononuclear cells (PBMCs) and sorted by FACSAria II (BD Biosciences). In addition, CD45⁺ living cells were also isolated from samples with sufficient cells. The sorted cells were sequenced using 10x Chromium Single-Cell 5′ and V(D)J library construction (10x Genomics, USA), according to the manufacturer’s instructions. We applied CellRanger V7 for read alignment and gene-count matrix generation and BCR sequence assembly. Potential doublets were identified and removed using DoubletFinder (v2.0.3).
B 细胞的 scRNA-seq 数据是从之前发布的数据集和新生成的数据中获得的。对于新生成的 scRNA-seq 数据，配对癌症、邻近正常、LN_Met 和血液样本是从在复旦大学附属中山医院接受手术切除的患者中收集的，并获得了机构审查委员会批准的方案的书面同意和批准。根据制造商的说明，使用 GentleMACS (Miltenyi Biotec) 使用肿瘤解离试剂盒或胶原酶 IV 消化匹配的癌症、邻近正常组织和 LN_Met 组织。 CD19 ⁺ B 细胞从单核细胞 (MNC) 和外周血单核细胞 (PBMC) 中分离，并通过 FACSAria II (BD Biosciences) 分选。此外，还从细胞充足的样品中分离出CD45 ⁺ 活细胞。根据制造商的说明，使用 10x Chromium Single-Cell 5' 和 V(D)J 文库构建（10x Genomics，美国）对分选的细胞进行测序。我们应用 CellRanger V7 进行读取比对和基因计数矩阵生成以及 BCR 序列组装。使用 DoubletFinder (v2.0.3) 识别并删除潜在的双联体。

To integrate different datasets, we applied SCTransform function to identify highly variable genes within each dataset. We then selected the top 2000 genes on the basis of their frequency as highly variable genes for the integrated dataset. All blacklisted genes (immunoglobulin, TCR genes, ribosomal protein coding genes, and tissue dissociation operation–induced genes) were specifically excluded. Harmony (v0.1.0) was applied for dataset integration, and the matrix was used for clustering with the default Seurat pipeline. Marker genes were calculated using FindAllMarker function of Seurat (table S2). The Change-O repertoire clonal assignment toolkit was applied to defined clones, SHM, BCR clonal diversity, and abundances. Paired scBCR-seq data were integrated with scRNA-seq data based on matched cell barcodes. BCR lineage tracing and overlap was calculated by the Jaccard index and the STARTRAC package (v0.1.0).
为了整合不同的数据集，我们应用 SCTransform 函数来识别每个数据集中高度可变的基因。然后，我们根据频率选择前 2000 个基因作为集成数据集的高度可变基因。所有列入黑名单的基因（免疫球蛋白、TCR基因、核糖体蛋白编码基因和组织解离操作诱导基因）均被明确排除。 Harmony (v0.1.0) 用于数据集集成，矩阵用于通过默认 Seurat 管道进行聚类。使用Seurat的FindAllMarker函数计算标记基因（表S2）。 Change-O 库克隆分配工具包应用于定义的克隆、SHM、BCR 克隆多样性和丰度。配对的 scBCR-seq 数据与基于匹配细胞条形码的 scRNA-seq 数据进行整合。 BCR 谱系追踪和重叠是通过 Jaccard 指数和 STARTRAC 软件包 (v0.1.0) 计算的。

EF- and GC-derived ASCs were classified according to the overlap of BCR sequences and gene expression. We classified ASC sharing BCR with canonical Bm and GC B cells as GC-derived ASCs, whereas ASCs containing overlap with other B cell subsets were classified as EF-derived ASCs at the cellular level. For ASCs without clonal sharing, we utilized the label transfer based on the gene expression of already classified ASCs by the TransferData function in the Seurat package. Only the classified cells with a prediction score greater than 0.6 were retained for further analysis. At patient levels, k-means clustering was performed on the basis of the SHM and EF pathway index (the ratio of EF-derived ASCs) to divide patients into two distinct groups. The pTrans score calculated by the STARTRAC package was also used to further validate the classification. We ensured consistent classification of those patients who met the criteria by using both cellular- and patient-level classifications for further analysis. Differential gene expression analysis was performed with the muscat package (v1.16.0) at patient levels.
EF 和 GC 衍生的 ASC 根据 BCR 序列和基因表达的重叠进行分类。我们将与典型 Bm 和 GC B 细胞共享 BCR 的 ASC 归类为 GC 衍生的 ASC，而在细胞水平上将与其他 B 细胞亚群重叠的 ASC 归类为 EF 衍生的 ASC。对于没有克隆共享的 ASC，我们通过 Seurat 包中的 TransferData 函数利用基于已分类 ASC 基因表达的标签转移。仅保留预测分数大于 0.6 的分类细胞进行进一步分析。在患者层面，根据 SHM 和 EF 通路指数（EF 衍生的 ASC 的比率）进行 k 均值聚类，将患者分为两个不同的组。 STARTRAC 软件包计算的 pTrans 评分也用于进一步验证分类。我们通过使用细胞级别和患者级别的分类进行进一步分析，确保对符合标准的患者进行一致的分类。使用 muscat 软件包 (v1.16.0) 在患者水平上进行差异基因表达分析。

To model the PCs’ and non-ASCs’ state transition, we applied multiple methodologies, including scTour (v0.1.3), Monocle3 (v1.0.0), and CytoTRACE (v0.3.3) to separately infer the pseudotime of PC and non-ASC subsets. We used SCENIC to predict and validate the TF regulatory network. CellPhoneDB and CellChat were applied to identify the potential receptor and ligand interaction between B cells and other immune cells. To perform cross-study comparisons, we used our dataset and cell annotation as a reference and trained logistic regression (LR) models using the CellTypist package. The BayesPrism computational method was used to assess the cell-type abundances of bulk transcriptomic data from TCGA.
为了模拟 PC 和非 ASC 的状态转换，我们应用了多种方法，包括 scTour (v0.1.3)、Monocle3 (v1.0.0) 和 CytoTRACE (v0.3.3) 来分别推断 PC 和非 ASC 的伪时间。 ASC 子集。我们使用 SCENIC 来预测和验证 TF 监管网络。 CellPhoneDB 和 CellChat 用于识别 B 细胞和其他免疫细胞之间潜在的受体和配体相互作用。为了进行交叉研究比较，我们使用数据集和单元格注释作为参考，并使用 CellTypist 包训练逻辑回归 (LR) 模型。 BayesPrism 计算方法用于评估来自 TCGA 的大量转录组数据的细胞类型丰度。

Nuclei suspensions of sorted CD19⁺ B cells were prepared according to the manufacturer’s instructions. Mononuclear GEMs with specific 10× barcodes were further generated using a 10x Genomics microfluidic platform. Raw scATAC-seq datasets were processed using the 10x Genomics cellranger-atac count pipeline (v.1.2.0), and downstream analysis was carried out by ArchR (v.1.0.2). Doublet was identified and excluded using the “addDoubletScores” and “filterDoublets” functions of ArchR (v.1.0.2), respectively. Sample batch correction was performed with Harmony (v0.1.0). Cell clustering was performed using the “addClusters” and “addUMAP” functions on the Harmony-corrected data, and “getMarkerFeatures” was applied to identify marker gene scores of each cluster (table S3). scRNA-seq gene expression and metadata was further integrated with scATAC data using the “addGeneIntegrationMatrix” function.
根据制造商的说明制备分选的 CD19 ⁺ B 细胞的细胞核悬浮液。使用 10x Genomics 微流体平台进一步生成具有特定 10x 条形码的单核 GEM。使用 10x Genomics cellranger-atac 计数管道 (v.1.2.0) 处理原始 scATAC-seq 数据集，并由 ArchR (v.1.0.2) 进行下游分析。分别使用 ArchR (v.1.0.2) 的“addDoubletScores”和“filterDoublets”函数来识别和排除 Doublet。使用 Harmony (v0.1.0) 进行样本批次校正。使用“addClusters”和“addUMAP”函数对 Harmony 校正数据进行细胞聚类，并应用“getMarkerFeatures”来识别每个簇的标记基因分数（表 S3）。使用“addGeneIntegrationMatrix”功能将 scRNA-seq 基因表达和元数据进一步与 scATAC 数据整合。

After clustering and cell identification, cell-type-specific peaks were called by MACS2 using the “addReproduciblePeakSet” function. Background peak set was generated with “addBgdPeaks” and used for TF motif enrichment analyses. Peak co-accessibility and Peak2Gene linkages were computed using the “addCoAccessibility” and “addPeak2GeneLinks” functions. The enrichment of chromatin accessibility at different TF motif sequences in single cells was calculated by “addDeviationsMatrix” using the cisbp motif set. Differential peak testing was performed using the “getMarkerFeatures” function, and TF motif enrichment in differential peaks was performed using the “peakAnnoEnrichment” function. TF footprinting analysis was performed using the “getFootprints” function.
聚类和细胞识别后，MACS2 使用“addReproduciblePeakSet”功能调用细胞类型特异性峰。使用“addBgdPeaks”生成背景峰集，并用于 TF 基序富集分析。使用“addCoAccessibility”和“addPeak2GeneLinks”函数计算峰共可访问性和 Peak2Gene 连接。使用 cisbp 基序集通过“addDeviationsMatrix”计算单细胞中不同 TF 基序序列处染色质可及性的富集。使用“getMarkerFeatures”函数进行差异峰测试，并使用“peakAnnoEnrichment”函数进行差异峰中的TF基序富集。使用“getFootprints”函数进行 TF 足迹分析。

MNCs or PBMCs were routinely stained with FVS780 to filter out dead cells. For surface phenotype staining, cells were stained with antibodies in MACS buffer for 15 min at room temperature. For intracellular staining, cells were fixed and permeabilized with a fixation and permeabilization kit for 1 hour at 4°C. Intracellular antibodies were then stained in the permeabilization buffer for 30 min at 4°C. Data were acquired using an LSRFortessa flow cytometer (BD Biosciences) and analyzed using FlowJo software (v10.8.1, BD Biosciences).
MNC 或 PBMC 常规用 FVS780 染色以滤除死细胞。对于表面表型染色，细胞用 MACS 缓冲液中的抗体在室温下染色 15 分钟。对于细胞内染色，使用固定和透化试剂盒在 4°C 下固定和透化细胞 1 小时。然后细胞内抗体在透化缓冲液中在 4°C 下染色 30 分钟。使用 LSRFortessa 流式细胞仪（BD Biosciences）采集数据并使用 FlowJo 软件（v10.8.1，BD Biosciences）进行分析。

Purified tumor-infiltrating naïve, Bm, and AtM B cells were stimulated with 1μg/ml human R848, 10 ng/ml human IL-2, 10 ng/ml human IL-10, and 40 ng/ml human IL-21 in total 200-μl culture medium. On day 11, cells were harvested and stained for plasma cells by fluorescence-activated cell sorting (FACS). Cell-free supernatants were collected and stored at −80°C for further antibody array detection. For AtM B cell differentiation experiments, healthy blood B cells were stimulated with 1μg/ml R848 in the absence or presence of glutamine at different glutamine physiological plasma concentrations and at different times of co-culture. Furthermore, 100ng/ml IFNγ or 10× physiological plasma concentrations of glutamine, or 500 nM CB-839 or 1.5 mM dimethyl-2-oxoglutarate (DMαKG) or 1 nmol rapamycin and 1μg/ml R848 were used to stimulate B cells for 4 days, respectively. AtM B cell phenotypes were stained and acquired by an LSRFortessa flow cytometer (BD Biosciences).
用总计 1μg/ml 人 R848、10 ng/ml 人 IL-2、10 ng/ml 人 IL-10 和 40 ng/ml 人 IL-21 刺激纯化的肿瘤浸润幼稚 Bm 和 AtM B 细胞200 μl 培养基。第11天，收获细胞并通过荧光激活细胞分选术（FACS）对浆细胞进行染色。收集无细胞上清液并储存在-80°C用于进一步的抗体阵列检测。对于 AtM B 细胞分化实验，在不存在或存在谷氨酰胺的情况下，以不同的谷氨酰胺生理血浆浓度和不同的共培养时间，用 1μg/ml R848 刺激健康的血液 B 细胞。此外，使用100ng/ml IFNγ或10×生理血浆浓度的谷氨酰胺，或500 nM CB-839或1.5 mM二甲基-2-酮戊二酸（DMαKG）或1 nmol雷帕霉素和1μg/ml R848刺激B细胞4天。 ， 分别。 AtM B 细胞表型通过 LSRFortessa 流式细胞仪 (BD Biosciences) 进行染色和获取。

For the Tph and B cell co-culture experiment, sorted tumor-infiltrated T_reg, Th, and PD1^HiCD4 Tph cells were stimulated with CD3/CD28 beads or IL-21R and were further plated with isolated total B cells and naïve B cells from healthy blood at a ratio of 1:5 and 1:2 in the final volume of 200 μl for 7 and 5 days, respectively. Phenotypic staining of AtM B cells was performed.
对于 Tph 和 B 细胞共培养实验，用 CD3/CD28 珠或 IL-21R 刺激分选的肿瘤浸润 T _reg 、Th 和 PD1 ^Hi CD4 Tph 细胞，进一步将来自健康血液的分离的总 B 细胞和幼稚 B 细胞以 1:5 和 1:2 的比例分别接种到最终体积 200 μl 中 7 和 5 天。对 AtM B 细胞进行表型染色。

For the AtM B and T cell co-culture experiment, glutamine-induced AtM B cells were collected and co-cultured with CD3/CD28 beads–treated healthy blood T cells at a ratio of 5:1 in the final volume of 200 μl. On day 5, cells were collected, and T cell subsets (CD8, CD4, and T_reg) and functional markers (Ki67, GZMB, IFNγ, TNFα, and PD1) were stained. All of this staining was detected using an LSRFortessa flow cytometer (BD Biosciences).
对于 AtM B 和 T 细胞共培养实验，收集谷氨酰胺诱导的 AtM B 细胞，并与 CD3/CD28 珠处理的健康血 T 细胞以 5:1 的比例在最终体积 200 μl 中共培养。第5天，收集细胞，并对T细胞亚群（CD8、CD4和T _reg ）和功能标记物（Ki67、GZMB、IFNγ、TNFα和PD1）进行染色。所有这些染色均使用 LSRFortessa 流式细胞仪 (BD Biosciences) 进行检测。

Paired adjacent and cancer tissue were immediately frozen in liquid nitrogen. The tissues were then cut into small pieces and homogenized using a homogenizer. Fifty- percent methanol/acetonitrile was added to the homogenized solution for metabolite extraction. The mixture was centrifuged, and the supernatant was dried in a vacuum centrifuge. For liquid chromatography–mass spectrometry (LC-MS) analysis, the samples were redissolved in 50% acetonitrile/water solvent and centrifuged for 15 min, then the supernatant was injected. The intracellular metabolites of glutamine- and CB-839-treated healthy blood B cells were collected with prechilled 80% methanol. The samples were then centrifuged, and supernatant was diluted with LC-MS–grade water. The samples were subsequently centrifuged and injected into the Thermo Fisher Scientific TSQ Quantis system.
配对的邻近组织和癌组织立即冷冻在液氮中。然后将组织切成小块并使用均质器均质化。将百分之五十的甲醇/乙腈添加到均质溶液中以提取代谢物。将混合物离心，并在真空离心机中干燥上清液。对于液相色谱-质谱（LC-MS）分析，将样品重新溶解在50%乙腈/水溶剂中并离心15分钟，然后注入上清液。用预冷的 80% 甲醇收集经谷氨酰胺和 CB-839 处理的健康血 B 细胞的细胞内代谢物。然后将样品离心，并用 LC-MS 级水稀释上清液。随后将样品离心并注入 Thermo Fisher Scientific TSQ Quantis 系统中。

For ¹³C glutamine tracing, healthy blood B cells were treated with R848 and ¹³C-labeled glutamine in glutamine-deficient culture medium for 24 hours. Unlabeled glutamine and R848 treatment were used as negative controls. Cells were reconstituted by 500 μl cold extraction buffer (methanol, acetonitrile, and water). The supernatants were thoroughly lyophilized and resuspended in methanol-water (1:1) before measurement. MS measurement of isotopologue distribution was analyzed using a Thermo Fisher Q-Exactive plus hybrid quadrupole-orbitrap mass spectrometer coupled to a Thermo Fisher Vanquish UPLC system. Data processing and ion annotation based on accurate mass were performed in TraceFinder 5.0 (Thermo Fisher) and Xcalibur 4.0 (Thermo Fisher).
对于 ¹³ C 谷氨酰胺示踪，在缺乏谷氨酰胺的培养基中用 R848 和 ¹³ C 标记的谷氨酰胺处理健康血液 B 细胞 24 小时。未标记的谷氨酰胺和R848处理用作阴性对照。用 500 μl 冷提取缓冲液（甲醇、乙腈和水）重构细胞。测量前将上清液彻底冻干并重悬于甲醇-水（1:1）中。使用与 Thermo Fisher Vanquish UPLC 系统联用的 Thermo Fisher Q-Exactive plus 混合四极杆轨道阱质谱仪对同位素体分布的 MS 测量结果进行分析。基于精确质量的数据处理和离子注释在TraceFinder 5.0 (Thermo Fisher)和Xcalibur 4.0 (Thermo Fisher)中进行。

We performed mIHC to identify ASCs, AtM B cells, PD1^HiCD4 Tph cells, and different TLS structures. Slides were scanned using PerkinElmer’s Vectra 3 platform, and images were quantified using Inform software (v.2.6.0). TLS and non-TLS regions were trained using “tissue segment” functions and quantified on all whole slides. TLS maturity was evaluated by fDC and hematoxylin and eosin (H&E) staining.
我们进行 mIHC 来鉴定 ASC、AtM B 细胞、PD1 ^Hi CD4 Tph 细胞和不同的 TLS 结构。使用 PerkinElmer 的 Vectra 3 平台扫描载玻片，并使用 Inform 软件（v.2.6.0）对图像进行量化。使用“组织片段”功能对 TLS 和非 TLS 区域进行训练，并在所有整个幻灯片上进行量化。通过 fDC 和苏木精和伊红 (H&E) 染色评估 TLS 成熟度。

We used the Spatial Analysis module of HALO software (Indica labs) to analyze the spatial location of AtM B cells between immature and mature TLSs. We then selected the edge of lymphocyte aggregation and the follicular region as the interface and subdivided by 20 μm per band into a total of 20 bands. Infiltration analysis was used to quantify the density and absolute number of AtM B cells in each band inside and outside the interface and the distance of each AtM B cell to the interface. We also used spatstat (v2.3-4) to quantify density and frequency of PD1^HiCD4⁺ Tph cells surrounding 20 μm of AtM B cells in TLS versus non-TLS areas.
我们使用 HALO 软件（Indica labs）的空间分析模块来分析 AtM B 细胞在未成熟和成熟 TLS 之间的空间位置。然后选择淋巴细胞聚集的边缘和滤泡区域作为界面，将每个条带以20μm细分为总共20个条带。使用渗透分析来量化界面内外每个条带中AtM B细胞的密度和绝对数量以及每个AtM B细胞到界面的距离。我们还使用 spatstat (v2.3-4) 来量化 TLS 与非 TLS 区域中 20 μm AtM B 细胞周围 PD1 ^Hi CD4 ⁺ Tph 细胞的密度和频率。