Characterization and digital spatial deconvolution of the immune microenvironment of intraductal oncocytic papillary neoplasms (IOPN) of the pancreas

This study has been approved by the Verona Ethical Committee (project: EPAT-2020, number of approval: 2801/CESC, date of approval: 24-06-2020) and has been conducted in accordance with the Good Practice guidelines, the Declaration of Helsinki, and current laws, ethics, and regulations. All cases with a diagnosis of IOPN were retrieved from the ARC-Net biobank at Verona University Hospital. We selected only cases for which tumor slides and blocks were available.

One representative formalin-fixed paraffin-embedded (FFPE) tissue block was selected for each case for whole-section immunohistochemistry (IHC), which was performed as already described [15, 16]. Briefly, 4 μm, FFPE sections were immunostained with the following antibodies, according to the manufacturer’s instructions: CD3 (clone: LN10; dilution: 1:200; Bio-Optica, Italy), CD4 (clone: 4B12; dilution: 1:100; Novocastra, Germany), CD8 (clone: C8/144B; dilution: 1:200; Dako, USA), CD20 (clone: L26; dilution: 1:100; Novocastra. Germany), CD68 (clone: KP1; dilution: 1:400, Dako, USA), CD163 (clone: 10D6; dilution: 1:200; Novocastra, Germany), PD-1 (clone: NAT; dilution: 1:100; Abcam, UK), PD-L1 (clone: SP263, pre-diluted 0.05 M, Roche, Switzerland), MLH1 (clone: ES05; dilution: 1:30; Dako, USA), PMS2 (clone: MRQ-2; dilution: 1:150; Cell Marque, USA), MSH2 (clone: FE11; dilution: 1:30; Dako, USA), and MSH6 (clone: EP49; dilution: 1:100; Dako, USA). Heat-induced antigen retrieval was performed using a heated plate and 0.01 mol/l of citrate buffer, pH 8.9, for 15 min. Light nuclear counterstaining was performed with hematoxylin (5 min). All samples were processed using a sensitive peroxidase-based “Bond polymer Refine” detection system in an automated Bond instrument (Vision-Biosystem, Leica, Milan, Italy). Sections incubated without the primary antibody served as negative controls.

IHC slides were evaluated separately and in blind by two pathologists (G.P., C.L.). Inconsistencies were resolved by consensus at a multi-headed microscope. For CD3, CD4, CD8, CD20, CD68, CD163, and PD-1, IHC was considered positive when the cell membrane was stained. The evaluation of the expression of these biomarkers was performed using a semi-quantitative (0–5) scoring system, as reported elsewhere [17, 18]: 0 = negative (no positive cells), 1 = rare (1–10 positive cells per high power field — HPF, 40X), 2 = low (11–20 positive cells per HPF), 3 = moderate (21–30 positive cells per HPF), 4 = high (31–50 positive cells per HPF), and 5 = very high (>50 positive cells per HPF). PD-1 evaluation was based on the single most positive HPF, while the scores for CD3, CD8, CD4, CD20, CD68, and CD163 were obtained as a mean value of the five most positive HPFs, as already described [19].

For MLH1, PMS2, MSH2, and MSH6, IHC was considered positive (retained expression) when the cell nuclei was stained, as already described and per current guidelines [20]. The evaluation of the expression of these biomarkers was classified as positive (retained expression) or negative (expression loss). PD-L1 expression was evaluated as reported elsewhere [21,22,23,24], and using two specific scores: (i) tumor proportion score (TPS) which assesses the percentage of positive viable tumor cells, showing partial or complete membrane staining at any intensity, and (ii) combined positive score (CPS) which takes into account the number of tumor and non-neoplastic PD-L1 positive cells, compared with all viable tumor cells.

The patterns of immunohistochemical expression of the lymphocytes markers CD4, CD8, and CD20 (CD4 and CD8 for T cells; CD20 for B-cells) have been also evaluated in terms of stain intensity and spatial disposition using artificial intelligence-based algorithms (Fig. 1). Hematoxylin-eosin (H&E) slides and the matched CD4, CD8, and CD20-stained slides were digitized using the APERIO platform (Leica Biosystems) at 20× of magnification. Each slide was then analyzed using the QuPath open source software platform (version 0.2.3) [25]. A region of interest (ROI) was annotated for each hematoxylin-eosin (H&E) stained slides including the whole tumor and excluding any surrounding normal pancreatic/duodenal tissue, and transferred on the corresponding IHC slide. Cell detection was performed using Stardist [26], a digital pathology toolbox that utilizes star-convex polygons to localize nuclei, for each H&E and IHC slide. A random tree forest classifier for each H&E slide was generated using cell features to classify cells into tumor, immune, and stromal cells. To maximize the accurateness of the algorithm, smoothed features at 25 and 50 μm radius were added and multiple rounds of cell classification review were performed. For each IHC slide, positive cells were detected and transferred to the H&E slide following slides alignment (Fig. 1).

Fig. 1figure 1

Workflow of imaging processing. A random tree forest classifier was generated to identify tumor immune and stromal cells on hematoxylin-eosin (H&E) images (A); on immunohistochemical (IHC) images, positive cells were automatically detected (B). Serial H&E and IHC sections were spatially aligned to virtually project the different immune phenotypes from each IHC section on the H&E image, generating a composite map including cancer cells and all immune subsets (C). On the composite map, spatial relationship among immune cells and cancer cells was computed through distance, density, and “neighborhood” analysis, generating a color-coded positional plot using Cytomap software (D)

First, all IOPNs without an associated adenocarcinoma were analyzed. Then, a specific analysis was performed for IOPNs with an associated adenocarcinoma, where the two components (i.e., intraductal vs. infiltrative) were analyzed separately. Spatial analysis was performed by exploring the organization of immune cells and their relationship among them and with tumor cells using Cytomap, as already described [27]. Briefly, cells were grouped into 200 μm raster scanned neighborhoods, which are cylindrical windows where the position of each neighborhood is evenly distributed across the tissue in a grid pattern.

Then, neighborhoods were clustered into regions based on their cellular composition using “Self-Organizing Map” (SOM) model [28], with the number of regions determined by the Davies-Bouldin criterion [29]. Cellular composition was defined as the number of cells specifically positive for a given IHC marker in each neighborhood, divided by the total number of cells in that neighborhood.

留言 (0)

沒有登入
gif