Single-cell landscape of immunocytes in patients with extrahepatic cholangiocarcinoma

Two ECCA patients who did not receive chemotherapy or radiation were enrolled in this study. Patients signed informed consent forms independently and agreed to donate tumor tissues and matched paratumor tissues and peripheral blood for scientific study. Our study was approved by the Tongji Hospital Research Ethics Committee and the Institutional Review Board. The demographic characteristics and clinical information of the patients researched by scRNA-seq are listed in Additional file 3: Table S1. The samples from two patients were collected at different times but by the same methods. After collection during surgery for ECCA, tumor and paratumor tissues were immediately transferred into MACS Tissue Storage Solution (Miltenyi Biotec GmbH), and peripheral blood was obtained before surgery. All fresh samples were stored on ice during transportation.

ScRNA-seq library construction was performed using the Chromium Single Cell 5′ Library & Gel Bead Kit (10× Genomics) for all samples according to the manufacturer’s instructions. Briefly, a single-cell suspension (containing 10000 cells), barcoded gel beads and partitioning oil were loaded onto Chromium Chip A to generate a single-cell gel bead-in-emulsion (GEM). Cells were captured in GEMs during targeted cell recovery. After the reverse transcription step, barcoded cDNA was purified with Dynabeads, and PCR amplification was performed as follows. A 5′ gene expression library was constructed on the basis of amplified cDNA. For gene expression library construction, 50 ng of amplified cDNA was fragmented and end-repaired, 150-bp paired-end reads and raw BCL files were produced after the cDNA was double size-selected with SPRIselect beads and sequenced on a NovaSeq platform (Illumina). Raw data in FASTQ format were assembled from the raw BCL files using Illumina’s bcl2fastq converter.

We used Cell Ranger Software Suite (v.3.1.0) to manipulate the FASTQ files, aligned the sequencing reads to the GRCh38 reference transcriptome build using STAR [15] and generated a filtered UMI expression profile for each single cell. Cell barcodes were then automatically determined based on the distribution of UMI counts. The following inclusion criteria were then applied to each cell: gene number between 500 and 6000, UMI count> 1000 and mitochondrial gene percentage < 10%. Genes expressed in less than three cells were excluded. DoubletDecon was used to detect and remove doublets. After filtration, a total of 13,526 cells were retained for the following analysis. Finally, “FindIntegrationAnchors” and “IntegrateData” in Seurat v.3 were used to integrate the filtered gene-barcode matrix of all samples to remove batch effects across different donors [16]. “ccRemover” was used to identify and remove cell-cycle effect [17].

“LogNormalData” in Seurat v.3 was used to normalize the filtered gene-barcode matrix. Then, the top 2000 highly variable genes were calculated with the “Find Variable Features” function in Seurat. Principal component analysis (PCA) was conducted by using the top 2000 variable genes. Then, the top 50 principal components were used to perform UMAP to visualize the cells. In addition, cell clustering was carried out on the basis of PCA-reduced data for clustering analysis with Seurat v.3. Cell clusters were annotated according to SingleR [18] and marker genes; for example, the phenotypes CD3E+ CD4+ , CD3E+ CD8A+ and CD79A+ were used to identify CD4+ T, CD8+ T and B cells, respectively.

CD4+ T cells, CD8+ T cells, B cells and myeloid immunocytes annotated according to SingleR and marker genes were reintegrated and reclustered by Seurat v.3. Reclustered CD4+ T cells, CD8+ T cells, B cells and myeloid immunocyte subclusters were named by cluster number and cell type; for example, cluster 3 in CD8+ T cells was named C3–CD8. Specifically. With the help of canonical correlation analysis and PCA of the first 50 dimensions, CD4+ T cells, CD8+ T cells, B cells and myeloid immunocytes of all samples were integrated. The parameter k.filter was set to 120. In the clustering step, the parameter resolution was set to 0.8. The “Monocle 3” in Seurat ordered CD8+ T single cells in an unsupervised manner and was used to analyze single-cell trajectories along pseudotime to discover the developmental transitions of CD8+ T subclusters.

GO enrichment analysis of differentially expressed gene sets was carried out in the GOseq R and KOBAS 3.0 packages. GO terms with adjusted P values below 0.05 were considered significantly enriched among differentially expressed genes.

CellPhoneDB (v2.1.7) [19] was used to classify cellular interactions for cells derived from ECCA tumor and matched paratumor tissues. Receptor–ligand crosstalk between two cell populations was identified according to the expression of receptors of one cell type and ligands of the other cell type. The expression frequency of a given receptor ligand gene was required to be > 10% in the corresponding cluster. The P value indicated the likelihood of a given receptor–ligand complex by calculating the proportion of means equal to or higher than the actual mean. The significant mean shows the total mean of the average expression values of each partner in the corresponding cell type interaction pair.

To explore the cellular heterogeneity of immunocytes in ECCA, we analyzed the immunocytes from ECCA patients based on scRNA-seq (Fig. 1A). RNA-seq data derived from a total of 13,526 sorted single cells were divided into 21 clusters and named cluster 1–21 (Fig. 1B). Significantly differentially expressed genes in each cluster are shown in Additional file 4: Table S2. According to the SingleR package [20] and marker genes, subclusters 7, 10, 11, 12, 14 and 15 were identified as CD8+ T lymphocytes with marker genes CD3E and CD8A, subclusters 2 and 18 were CD4+ T lymphocytes with marker gene CD3E and CD4, clusters 1 and 6 were B lymphocytes with CD79A, subcluster 20 was NK cells with KLRD1+ , and clusters 8, 9, 13, 16 and 17 were monocytes/macrophages with CD68 (Fig. 1C, D). Cluster 5 (neutrophils, FCGR3B+), cluster 4 (endothelial cells, PLVAP+), clusters 3 and 19 (epithelial cells, KRT18+) and cluster 21 (fibroblasts, C1S+) were mixed nontargeted cells and excluded from the following study (Fig. 1C and Additional file 1: Fig. S1A).

Our study revealed that the proportion of B lymphocytes gradually increased in peripheral blood (6.93%), paratumor tissues (18.57%) and tumor tissues (33.22%). In contrast, the proportions of CD4+ T lymphocytes gradually decreased from 22.98%, 15.59% 3.31% in peripheral blood, paratumor tissues and tumor tissues, respectively, which was also observed in NK clusters at 13.12%, 3.31% and 2.69%, respectively. Notably, the proportion of CD8+ T lymphocytes in tumor tissues (22.47%) was less than that in paratumor tissues (34.16%) (Fig. 1E, F).

We reclustered the 1657 CD4+ T lymphocytes annotated with the SingleR package and marker genes (Fig. 1C, D), and 5 subclusters emerged and were named C1–CD4 to C5–CD4 (Fig. 2A). Significantly differentially expressed genes in each subcluster are shown in Additional file 5: Table S3. C1–CD4 was characterized by high expression of FOS, FOSB, FOSL2, JUN and JUND known as transcription factor subunits of AP-1 (Fig. 2B). These genes have been described in several satellite cell studies and have been proved to be high expressed genes induced by tissue dissociation [21]. Thus, C1–CD4 may be subcluster resulted from tissue dissociation procedure. C2–CD4 was defined as naïve T lymphocytes specifically expressing the naïve marker genes LEF1, TCF7 and MAL (Fig. 2B) [22]. C3–CD4 highly expressed CD55 and ARID5A, which are known as markers of activated T lymphocytes (Fig. 2B) [23, 24]. C4–CD4 highly expressed genes S100A4, S100A6 and S100A11, and S100A4 was exclusively expressed by memory CD4+ T lymphocytes (Fig. 2B) [25]. The marker genes FOXP3 and IL2RA (CD25) were exclusively prevalent in C5–CD4, which was finally identified as regulatory CD4+ T lymphocytes (CD4+ Tregs) (Fig. 2B). An obvious distinct tissue distribution was found in the above subclusters. The resting CD4+ T lymphocyte subclusters (C2–CD4 and C4–CD4) contained mostly cells from peripheral blood, while activated CD4+ T lymphocyte subclusters (C3–CD4 and C5–CD4) were almost exclusively populated in tumor tissues and paratumor tissues (Fig. 2C, D).

Tregs are characterized as inhibitors of antitumor immunity and promote tumor development and progression. It is of great importance to see a strong enrichment of Tregs (C5–CD4) in ECCA tumor and paratumor tissues in our study (Fig. 2C, D). Furthermore, compared with conventional T cells (Tcon, from C1–CD4 to C4–CD4), Tregs (C5–CD4) exclusively expressed 333 genes in this study, including marker genes (FOXP3 and IL2RA), costimulatory receptors (ICOS, TNFRSF4, TNFRSF9 and TNFRSF18) and coinhibitory receptor/immune checkpoint genes (CTLA4, TIGIT, HAVCR2 and PDCD1) (Fig. 2E and Additional file 5: Table S3). Compared to previous studies, overlapping analysis showed 18 Treg-specific genes, including IL2RA, FOXP3, CCR8, TNFRSF18, BATF, TNFRSF4, CTSC, IL2RB, CTLA4, ACP5, TIGIT, ICOS, IL1R2, TNFRSF9, NCF4, IKZF2, VDR and MAGEH1 (Fig. 2F and Additional file 6: Table S4) [22, 26, 27]. Among them, CTLA4, TIGIT, TNFRSF4 and TNFRSF18 were highly expressed in tumor-infiltrating Tregs. Moreover, ACP5, MAGEH1, TNFRSF9 and CCR8 were especially populated in tumor-infiltrating Tregs and IKZF2 was only highly expressed in paratumor-infiltrating Tregs (Fig. 2G and Additional file 1: Fig. S1B).

To illustrate the CD8+ T subclusters in ECCA, we reclustered 1657 cells identified as CD8+ T lymphocytes and obtained 8 subclusters named C1–CD8 to C8–CD8 (Fig. 3A). Significantly differentially expressed genes in each subcluster are shown in Additional file 7: Table S5. CD8+ Tex subclusters C1–CD8 and C4–CD8 were characterized by high expression of the T cell exhaustion marker genes PDCD1, CTLA4, LAG3 and HAVCR2 and the metallothionein family genes MT1E, MT1F and MT1X (Fig. 3B and Additional file 1: Fig. S1C) [28]. Subclusters C2–CD8, C3–CD8 and C8–CD8, described as effector CD8+ T lymphocytes (CD8+ Teff), presented a strong degree of expression of the cytotoxic effectors PRF1, GNLY and GZM and low expression of exhaustion markers (Fig. 3B, C and Additional file 1: Fig. S1C). The γδ-TCR CD8+ T lymphocyte subclusters C5–CD8 were characterized by high expression of the TCR γ and δ subunit genes TRDV2, TRGV9, TRGV3 and TRGV2 (Fig. 3B). Additionally, the naïve markers SELL, CCR7, LEF1 and TCF7 and the costimulatory receptors ICOS and CD28 were enriched in subcluster C7–CD8 T cells and identified as naïve CD8+ T lymphocytes (Fig. 3B, C and Additional file 1: Fig. S1C). C6–CD8 cells specifically expressed TYMS and PCLAF associated with DNA replication and repair, suggesting a proliferative CD8+ T lymphocyte subcluster (Fig. 3B, C and Additional file 1: Fig. S1C). Moreover, in the analysis of CD8+ T lymphocyte subcluster distribution, we found that CD8+ Tex subclusters C1–CD8 and C4–CD8 showed great preference for tumor tissue (Fig. 3D, E). A previous study proved that CX3CR1+ CD8+ T cells exhibited robust cytotoxicity [29]. Interestingly, CD8+ Teff subclusters C3–CD8 and C8–CD8 highly expressed CX3CR1 were enriched in peripheral blood, while C2–CD8 weakly expressed CX3CR1 concentrated in paratumor tissues (Fig. 3D, E).

CD8+ Tex lymphocytes feature loss of functional effect, upregulation of multiple inhibitory receptor expression and inability to convert to memory CD8+ T lymphocytes to participate in the effective immune response of the body upon receipt of a new stimulus. This subcluster plays an extremely critical role in the TME. Consistent with previous studies, CD8+ Tex lymphocytes in ECCA presented preferential enrichment in tumor tissue with strong expression of coinhibitory receptors. Compared with CD8+ Tn (C7–CD8) and CD8+ Teff (C2–CD8, C3–CD8 and C8–CD8), CD8+ Tex (C1–CD8 and C4–CD8) highly expressed 110 genes (Fig. 3F). Notably, the average expression of the cytotoxic effectors PRF1, GNLY, GZMA and GZMB in CD8+ Tex cells was even higher than that in CD8+ Teff cells (Fig. 3G). Compared with C4–CD8, CD8+ Tex subcluster C1–CD8 highly expressed the costimulatory receptor TNFRSF4 but expressed low levels of the coinhibitory receptors PDCD1, CTLA4, HAVCR2 and LAG3 (Fig. 3C). In addition, pseudotime trajectory analysis confirmed the differentiation trajectory of these subclusters, successively, from Tn (C7–CD8) to αβ-TCR CD8+ Teff (C2–CD8) or γδ-TCR CD8+ Teff (C5–CD8) to CD8+ Tex (C1–CD8) then to Tex (C4–CD8) or proliferative subcluster (C6–CD8) (Fig. 3H). Thus, it is reasonable to identify that the C1–CD8 may be the "intermediate" CD8+ Tex between CD8+ Tn (C7–CD8) and "complete" CD8+ Tex (C4-CD8) (Fig. 3H). A total of 124 genes were highly expressed by "intermediate" CD8+ Tex subcluster C1–CD8 and GO enrichment analysis revealed that these genes were enriched in the pathways of protein folding and binding and response to heat and zinc ions (Additional file 2: Fig. S2A). A total of 266 genes were highly expressed by the "complete" CD8+ Tex subcluster C4–CD8 and GO enrichment analysis revealed that these genes were enriched in the pathways of T cell aggregation, adhesion and activation (Additional file 2: Fig. S2B).

A total of 2038 B lymphocytes were reclustered, and 5 subclusters named C1–B to C5–B lymphocytes were mapped for further study (Fig. 4A). Significantly differentially expressed genes in each subcluster are shown in Additional file 8: Table S6. C1–B and C4–B were classified as naïve B cells (Bn, CD27-IgD+) with high expression of the naïve marker genes IL4R, IGHD (IgD) and IGHM (IgM) (Fig. 4B and Additional file 2: Fig S2C). C2–B showed high expression of the marker gene CD27 and the genes ITGB1 and S100A6, which indicated memory B cells (Bm, CD27+ IgD−) (Fig. 4B and Additional file 2: Fig S2C) [30]. Additionally, C3–B and C5–B subclusters were plasmacytes (PCs) depending on the high expression of the marker genes CD27 and SDC1 (CD138) and the genes IGHG1, IGHG4, MZB1 and XBP1 (Fig. 4B and Additional file 2: Fig. S2C). The classic PCs C5–B (CD27+ CD138+ CD38+) were characterized by high co-expression of the marker genes CD138 and CD38, while C3–B (CD27+ CD138+ CD38−) showed quite poor levels of CD38, indicating a “nonclassic” PC subcluster rarely described in previous studies (Fig. 4B). Compared with peripheral blood, tumor and paratumor tissues contained higher proportions of Bn subcluster C1–B and Bm subcluster C2–B; conversely, the Bn subcluster C4–B and PC subcluster C5–B were observed infrequently. The proportion of “nonclassic” PC subcluster C3–B increased gradually in peripheral blood (0.87%), paratumor tissues (4.27%) and tumor tissues (8.30%) (Fig. 4C, D).

Sixty-three genes were highly expressed in nonclassic PC subcluster C3–B, and IGHG4, IGHG1, MZB1, SDC1, DERL3, XBP1, PRDX4, FKBP11, SSR4 and IGLV3-1 were the top 10 highly expressed genes. GO enrichment analysis revealed that highly expressed genes were enriched in the pathways of immunoglobulin complex and complement activation (Fig. 4E). Analysis of C5–B showed a total of 431 highly expressed genes, and TRIB1, AQP3, B9D1, CYTOR, CNKSR1, FNDC3B, IGF1, TNFRSF17, TXNDC5 and SLAMF7 were the top 10 highly expressed genes. GO enrichment analysis revealed that highly expressed genes were enriched in the immunoglobulin complex and response to protein folding (Fig. 4F).

To characterize myeloid immune cells in ECCA, we reclustered 1657 myeloid immunocytes and identified 8 subclusters named C1-Mye to C8-Mye (Fig. 5A). Significantly differentially expressed genes in each subcluster are shown in Additional file 9: Table S7. C1-Mye highly expressed IL-1β, CXCL2, SOD2, TREM1 and AREG, and C2-Mye highly expressed S100A8, S100A12 and LYZ (Fig. 5B). Both of the subclusters above had gene expression profiles suggesting M1 macrophages [31‐33]. As C4-Mye highly expressed M2-specific genes VCAN and DUSP1 and marker genes CD163 and C6-Mye highly expressed M2-specific genes TGF-β1, A2M, MAF, SLC40A1, FOLR2, GPNMB and MSR1 and the marker genes CD163 and MRC1 (CD206), C4-Mye and C6-Mye were defined as M2 macrophages (Fig. 5B). C3-Mye was classified as CD14+ classical monocytes with high expression of CD14 and FCN1 but low expression of FCGR3A (CD16), and C8-Mye highly expressed FCGR3A (CD16), CX3CR1 and FCN1 but expressed low levels of CD14, indicating CD16+ nonclassic monocytes (Fig. 5B). C5-Mye with high expression of the marker genes LILRA4, IL3RA and CLEC4C resembled plasmacytoid dendritic cells (pDCs), and classical dendritic cells (cDCs) pointed to C7-Mye because of their high expression of the marker genes CD1C and CLEC10A (Fig. 5B) [34, 35]. Furthermore, dendritic cell subclusters C5-Mye and C7-Mye were enriched in tumor and paratumor tissues, while monocyte subclusters C3-Mye and C8-Mye were concentrated in peripheral blood (Fig. 5C, D). The macrophage subclusters C1-Mye (M1) and C6-Mye (M2) were almost exclusively populated with cells from tumor/paratumor tissues, while C2-Mye (M1) and C4-Mye (M2) were almost exclusively populated with cells from peripheral blood (Fig. 5C, D). These different distributions of macrophage subclusters indicated separate transcriptome characteristics of M1 and M2 macrophages between tissues/paratumor tissues and peripheral blood. In addition, the proportion of C1-Mye (M1) in paratumor tissues (70.09%) was higher than that in tumor tissues (42.18%), while the proportion of C6-Mye (M2) in paratumor tissues (2.78%) was lower than that in tumor tissues (25.09%) (Fig. 5D).

As they showed potent protumorigenic and immunosuppressive properties and were enriched in tumor tissues, the M2 macrophage subclusters C4-Mye and C6-Mye were further analyzed. In contrast to C4-Mye enriched in blood, C6-Mye preferred tumor and paratumor tissue and highly expressed 591 genes (Fig. 5E), including the M2 macrophage marker gene MCR1 (CD206), and highly expressed the M2 macrophage-related protumorigenic genes TGF-β, IL10, VEGFA and MMP9 (Fig. 5F). Although studies have focused on pDC differentiation within the myeloid lineage, pDCs in our study showed transcriptome characteristics of both lymphoid and myeloid lineages. The marker genes CD68 (monocytes/macrophages), MZB1 (plasmacytes), PTCRA (immature T lymphocytes) and GZMB (NK cells and cytotoxic T lymphocytes) were the top-ranked genes coexpressed by pDCs (Fig. 5G). pDCs were suggested to serve an endocrine function because highly expressed SCT (secretin) may have endocrine functions (Fig. 5G).

To elucidate cell-to-cell communication in ECCA, CellPhoneDB software was applied to identify potential ligand–receptor interactions in various subclusters originating from ECCA tumor and paratumor tissues (Additional file 10: Table S8). Based on the cell interaction network counts (Additional file 11: Table S9), we found intense cell-to-cell interactions among M2 macrophages (C6-Mye), Tregs (C5–CD4) and CD8+ Tex cells (C1–CD8 and C4–CD8) via chemotactic, immune-suppressive or costimulatory factors (Fig. 6). Among these cells, Tregs highly expressed the TIGIT, CTLA4 and CCR8 genes, which showed receptor–ligand interactions with NECTIN2, CD86 and CCL4 from M2 macrophages (C6-Mye), implying potential cross-talk between M2 macrophages (C6-Mye) and Tregs (C5–CD4). In addition, Treg cells (C5–CD4) were suggested to interact intensively with CD8+ Tex cells (C1–CD8 and C4–CD8) through HLA-E-KLRK1 and its subtypes, which were classified as immune inhibitors because they promote senescent cell accumulation [36] and suppress tumor-specific cytotoxic T lymphocytes [37], and HLA-E-KLRK1 was identified as a novel checkpoint in the tumor microenvironment (TME) [38]. In addition, Treg cells (C5–CD4) were suggested to communicate with CD8+ Tex cells via chemotactic factors (CCL4‐CCR8 and CCL5‐CCR5) [39, 40], adhesion molecules (ICAM3-aLb2 complex) and immune regulatory molecules (LGALS9-CD44) [41], which were abundant in the TME and upregulated the function of Tregs and CD8+ Tex cells. M2 macrophages (C6-Mye) were predicted to secrete the factors CXCL12 and CCL4 to attract CD8+ Tex cells by binding to CXCR3/CXCR4 and CCR5 on CD8+ Tex cells, which are known to recruit immunocytes into cancer tissues. Notably, M2 macrophages were observed to develop self-amplifying feedback via the CXCL12-CXCR4 axis, which was shown to induce tumor-promoting M2 polarization of macrophages [42, 43]. NRP1/2-VEGFA was predicted to interact within M2 macrophages, which could potentially contribute to tumor angiogenesis [44].