Abstract

Background

   Pancreatic angiosarcoma is a rare and highly aggressive tumor
   originating from lymphatic or vascular endothelial cells, with poor
   prognosis and few effective treatments. In this study, we aimed to
   characterize the tumor ecosystem of metastatic pancreatic angiosarcoma,
   along with its potential treatment strategies.

Methods

   Single-cell RNA-sequencing and bioinformatics analysis were performed
   on samples obtained from one patient, including at total of 16,841
   cells from pancreatic angiosarcoma liver metastasis and adjacent normal
   liver tissue.

Results

   Pancreatic angiosarcoma cells exhibited marked upregulation of nuclear
   factor kappa-B (NF-κB), hypoxia-inducible factor 1 (HIF-1), and
   myelocytomatosis oncogene (MYC) proto-oncogene signaling pathways,
   while presenting limited upregulation of actionable therapeutic targets
   except for cyclin-dependent kinase 4 (CDK4) and epidermal growth factor
   receptor. Several immune checkpoint genes, including cytotoxic
   T-lymphocyte-associated protein 4 (CTLA4), lymphocyte-activation gene 3
   (LAG3), programmed cell death protein 1 (PDCD1), and cluster of
   differentiation 86 (CD86), were upregulated in tumor-infiltrating T
   cells, natural killer (NK) cells, and myeloid cells. Furthermore,
   intercellular interaction profiling demonstrated enhanced activity of
   the programmed death-ligand 1 (PD-L1) and CD86 signaling pathways
   within the tumor microenvironment. The gene-set scores of T/NK-cell
   exhaustion, regulatory T cell, and macrophage angiogenesis were
   significantly higher in tumor tissues compared with adjacent normal
   tissues. However, the phagocytosis scores of macrophages within the
   tumor-infiltrating region were significantly lower than those in the
   adjacent normal tissues.

Conclusions

   Our findings outlined an immunosuppressive and angiogenic tumor
   ecosystem in pancreatic angiosarcoma liver metastasis, suggesting that
   pancreatic angiosarcoma may be insensitive to most targeted therapies.
   Conversely, immunotherapies targeting LAG3, PD-L1, and CD86 (e.g.
   isatuximab, Opdualag, and abatacept) and anti-angiogenic agents may be
   therapeutically effective and worthy of subsequent exploration.

   Keywords: pancreatic angiosarcoma, single-cell RNA sequencing,
   immunotherapy, PD-L1, CD86

Introduction

   Angiosarcoma is a rare and highly aggressive tumor originating from
   lymphatic or vascular endothelial cells [[40]1], comprising <1% of all
   sarcomas [[41]2]. With an annual incidence rate of two to three per
   million [[42]3], angiosarcoma occurs at any age (most frequently in
   adults and the elderly), presenting no significant difference in
   incidence between males and females [[43]4]. Angiosarcoma consists of a
   clinically and genetically heterogeneous subgroup of sarcomas that can
   appear in any location of body, mostly in cutaneous tissues (∼60% of
   cases), especially head and neck, but also in soft tissues,
   parenchymatous organs, bone, and retroperitoneal areas [[44]5]. As a
   rare form of angiosarcoma that occurs in the pancreas, pancreatic
   angiosarcoma accounts for only 0.1% of all primary pancreatic tumors,
   with only ∼11 cases reported by investigators to date [[45]6–16].
   Angiosarcoma is prone to metastasis and recrudescence, which leads to a
   very poor prognosis, with an overall survival ranging from 6 to
   16 months [[46]17].

   The diagnosis of pancreatic angiosarcoma remains a big challenge.
   Physicians have trouble in quickly distinguishing angiosarcoma from
   other tumors such as hemangiomas and neuroendocrine tumors due to its
   non-specific morphology and clinical manifestations [[47]17].
   Ultrasound, computed tomography (CT), and magnetic resonance imaging
   have limitations and ultimately histopathology would be required for
   the final diagnosis, which shows spindle, polygonal, epithelioid, and
   undifferentiated round cells, and immunohistochemistry reveals the
   expression of both vascular and endothelial antigens, including
   Factor-VIII-related antigen (Factor-VIIIRA), cluster of differentiation
   34 (CD34), platelet and endothelial-cell adhesion molecule 1
   (PECAM1/CD31), and vascular endothelial growth factor (VEGF) [[48]17].

   The rarity of pancreatic angiosarcoma cases and the delay in diagnosis
   have led to a dilemma in its treatment and prognostic assessment.
   Currently, there are no standardized guidelines for the management of
   patients with pancreatic angiosarcoma, and radical surgery and adjuvant
   radiotherapy are currently the mainstay treatment modalities [[49]6,
   [50]7, [51]17, [52]18]. Cytotoxic chemotherapy is also put into use in
   the face of inoperable tumors. However, these treatments are also
   controversial because of the difficulty in obtaining negative margins
   in surgery and the high recurrence rate after surgery or the severe
   side effects of drugs [[53]17, [54]18]. With the advent of
   gene-sequencing technology in recent years, targeted therapy and
   immunotherapy have also been studied but all gain only mild effects
   [[55]18]. In terms of prognosis, current gene sequencing in
   angiosarcomas has reflected several upregulated cancer-driver genes
   such as ATM serine (ATM), tumor protein p53 (TP53), B-Raf
   proto-oncogene (BRAF), patched 1 (PTCH1), and APC regulator of
   wingless-type mouse mammary tumor virus (MMTV) integration site (WNT)
   signaling pathway (APC) in some patients [[56]19], but characteristic
   markers for pancreatic angiosarcoma remain unidentified. Histological
   grading is also difficult for accurately predicting patient prognosis
   due to the heterogeneity in histologic behavior among the lesions and
   the presence of tiny, undetectable satellite lesions.

   In this study, we analysed the gene-expression landscape of pancreatic
   angiosarcoma liver metastasis based on single-cell RNA-sequencing
   (scRNA-seq) data from a rare case of the disease. By comparing tumor
   tissues with adjacent normal counterparts and analysing the
   tumor-immune microenvironment, we investigated the mechanisms
   underlying the limited efficacy of current targeted therapies in
   pancreatic angiosarcoma, identified drivers of immune evasion and
   metastasis, and explored potential immunotherapeutic targets.

Materials and methods

Human-specimen sampling

   A schematic diagram of the overall tissue-sample-collection and
   processing workflow is shown in [57]Figure 1A. One patient diagnosed
   with pancreatic angiosarcoma liver metastasis was included in this
   study. The patient presented a clear history of constant pain in the
   left upper abdomen for 7 days, with CT imaging demonstrating multiple
   tumor lesions involving both the pancreas and the liver. In order to
   clarify the diagnosis, the patient underwent a fine-needle biopsy
   within the tumor site. A tumor tissue sample within the liver (T) and
   its paired adjacent normal liver tissue sample (N) were obtained during
   the biopsy under the supervision of a pathologist. Pathological
   findings revealed a patchy distribution of malignant cells in the liver
   tissue, some of which exhibited enlarged hyperchromatic nuclei with
   marked cytologic atypia, accompanied by lumen formation of varying
   sizes. Immunohistochemistry presented erythroblast
   transformation-specific transcription factor ERG (ERG) (+), CD31 (+),
   CD34 (+), Kiel 67 antigen (Ki-67) (30%+), chromogranin A (CgA) (−),
   somatostatin receptor 2 (SSTR2) (−), synaptophysin (Syn) (−),
   hepatocyte (−), and glypican-3 (−), as shown in [58]Figure 1B, and the
   final diagnosis was considered to be pancreatic angiosarcoma liver
   metastasis. Prior to needle biopsy of the tumor, the patient did not
   receive any form of chemotherapy, radiation, or drug treatment.
   Adjacent non-malignant liver tissues were procured at a distance of
   ≥5 cm from the tumor. All clinical samples were obtained from The First
   Affiliated Hospital, Sun Yat-sen University (Guangzhou, China) and the
   study has received ethical approval from the Ethics Committee of The
   First Affiliated Hospital, Sun Yat-sen University (Approval No.
   [2023]701). Written informed consent was obtained from the patient for
   both sample collection and data analyses.

Figure 1.

   [59]Figure 1.
   [60]Open in a new tab

   Single-cell transcriptomic analysis in pancreatic angiosarcoma liver
   metastasis and adjacent normal tissues. (A) Workflow of sample
   collection and data analysis in this study. One patient diagnosed with
   pancreatic angiosarcoma with liver metastasis was included in this
   study. Fresh liver metastasis and adjacent normal liver tissues were
   processed and single-cell sequenced to obtain single-cell
   transcriptomic data. (B) Immunohistochemistry (IHC) and
   hematoxylin–eosin staining of angiosarcoma tissue sample. (C) Uniform
   Manifold Approximation and Projection (UMAP) plot of the 16,841 cells
   from two samples of the tumor site, colored by Seurat clusters. (D)
   UMAP plot of the 16,841 cells from two samples of the tumor site,
   colored by manually annotated cell clusters. Each point indicates a
   single cell. (E) Feature plot displaying the expression distributions
   of housekeeping genes, including actin beta (ACTB),
   beta-2-microglobulin (B2M), glyceraldehyde-3-phosphate dehydrogenase
   (GAPDH), and the immune-cell-specific marker protein tyrosine
   phosphatase receptor type C (PTPRC). Each dot represents a single cell,
   with cells exhibiting the highest expression levels visually
   demarcated. (F) Heat map showing scaled, normalized expression of top
   five differentially expressed genes in each cell cluster. (G) Feature
   plot showing the expression distributions of cluster-specific markers
   used to distinguish individual cell clusters. Each dot represents a
   single cell, with peak expression intensity clearly reflected in the
   visual gradient. (H) UMAP showing cell-cluster distribution in tumor
   sample (T) and adjacent normal liver tissue sample (N), and stacked
   histogram shows the proportion of six cell clusters in each sample. HE
   = hematoxylin and eosin staining, NK = natural killer cell, UMAP =
   Uniform Manifold Approximation and Projection, CK = cytokeratin.

Single-cell-suspension preparation

   Tumor tissues and adjacent normal liver tissues were snap-frozen in
   liquid nitrogen within 15 min after sampling for subsequent single-cell
   RNA-sequencing procedures [[61]20]. A total of 2.5 mL of tumor
   digestive enzyme mix (#130–095-929, Tumor Dissociation Kit; Miltenyi
   Biotech, Germany) was added before digestion for 10 min in a 37°C water
   bath. The single-cell suspension of dissociated tissues was then
   filtered through 70- and 30-μm magnetic-activated cell-sorting
   strainers (#130–098-462/130–098-458; Miltenyi Biotech) to remove the
   cell mass. Next, dissociated cells were pelleted (4°C, 400 g, 6 min)
   and resuspended by using a modest red blood cell (RBC) lysis buffer
   (#00430054; eBioscience, USA) for 5–10 min. Two or three times volume
   of phosphate buffer saline (PBS; #C10010500CP; Gibco, USA) was added to
   terminate the RBC lysis. The single-cell suspension was then pelleted
   again and washed once with cold PBS, after which the cells were
   resuspended with RPMI1640 (#01–103-1A; BI, Israel) or cryopreserved in
   90% Fetal Bovine Serum (#FND500; ExCell Bio, China) with 10% dimethyl
   sulfoxide (#D8371; Solarbio, China) for long-term preservation. The
   whole process of tissue dissociation was carried out on ice [[62]21].

Single-cell RNA-seq library preparation

   Single cells were processed with the 10x Chromium Controller (10x
   Genomics, USA) based on the 10x gel bead-in-emulsion code (GEMCode)
   proprietary technology. An appropriate volume of single-cell suspension
   with a cell concentration of 700–1,200 cells/µL was loaded into each
   channel. About 8,000 cells were captured per sample, which were further
   mixed with barcoded beads. After reverse transcription reaction and
   complementary DNA (cDNA) amplification, libraries were constructed
   following the standard 10x Genomics protocol (Single Cell 5′ Reagent
   Kits v5.2 User Guide) and sequenced on Novaseq™ 6000 (Ilumina, USA)
   [[63]21].

Single-cell RNA-seq data preprocessing and cell-type annotation

   The CellRanger software pipeline (Version 7.0.1, 10x Genomics) was used
   to demultiplex cellular barcodes and map reads to the human genome
   (build hg38), and quantify the gene expression in single cells,
   generating a digital gene-expression matrix [unique molecular
   identifier (UMI) counts per gene per cell] for each sample. Expression
   matrices for all samples were then processed by using the R Seurat
   package (version 4.3.0) [[64]22]. As a quality-control (QC) step, we
   filtered out cells with: (i) the number of detected genes <200 or
   >6,000; (ii) the proportion of mitochondrial gene expression of >15%;
   and removed genes detected in fewer than three cells. After that, we
   obtained a total of 16,841 high-quality single-cell transcriptomes from
   all samples. To mitigate the impact of data sources on the analysis, we
   normalized the data by using canonical correlation analysis.
   Subsequently, the “ScaleData” function was employed to perform linear
   regression against the normalized expression levels, total UMI counts,
   and mitochondrial RNA content per cell for each gene. Principal
   component analysis was then performed by using “RunPCA”. The
   gene-expression and clustering outcomes were visualized on a Uniform
   Manifold Approximation and Projection (UMAP) plot by using the top 10
   principal components, obtained through “RunUMAP”. To further cluster
   the cells, we employed the functions “FindNeighbors” and
   “FindClusters”. Cell types were determined by analysing the gene
   expression of established markers. T cells and natural killer (NK)
   cells were identified by using protein tyrosine phosphatase receptor
   type C (PTPRC), CD3D, CD3E, CD8A, and CD4; myeloid cells were
   identified by using CD68, S100 calcium binding protein A8 (S100A8),
   S100A9, complement C1q A chain (C1QA), and B chain (C1QB); B cells were
   identified by using CD19, CD79A, CD79B, and membrane spanning 4-domains
   A1 (MS4A1); endothelial cells were identified by using PECAM1, von
   Willebrand factor (VWF), and selectin E (SELE); fibroblasts were
   identified by using integrin subunit alpha 8 (ITGA8), periostin
   (POSTN), and collagen type I alpha chain 2 (COL1A2); and hepatic cells
   were identified by using albumin (ALB), apolipoprotein A2 (APOA2),
   epithelial cell adhesion molecule (EPCAM), and Keratin 19 (KRT19).

Gene-set signature scores

   In order to probe the status and distribution of immune cells within
   the tumor-immune microenvironment, we assessed the expression levels of
   the several gene sets for naive, cell killing, exhausted, co-stimulate,
   and antigen presentation of immune cells in tumor tissues and adjacent
   normal liver tissues [[65]23]. In addition, tip-like gene-set scores
   were evaluated among endothelial cells [[66]24]. The R package AUCell
   (version 1.24.0) was utilized to assess the average expression of our
   target gene set at the single-cell level [[67]25].

Copy-number variation analysis

   Endothelial-cell copy-number variation (CNV) evaluation was performed
   by using the infercnv R package (version 1.18.1) [[68]21].
   Amplifications and deletions at the single-cell level were predicted
   base on the scRNA-seq data. The gene-expression matrix of tumor cells
   was extracted from the Seurat object to calculate its CNVs and
   fibroblasts, T/NK cells, and myeloid cells were selected as normal
   references. The inferCNV analysis was conducted with parameters