Multiplex immunofluorescence of whole slide images for enhanced tumor microenvironment immune cell characterization in multiple cancer types
Authors
- BostonGene, Corp., Waltham, USA
Abstract
Single-cell spatial phenotyping using multiplex immunofluorescence (MxIF) imaging has emerged as a promising method to evaluate the heterogeneity of the tumor microenvironment (TME) for immunotherapy response prediction. Here, we used our MxIF imaging pipeline, based on an enhanced cell segmentation tool and spatial analysis, to evaluate the TME immune landscape from whole slide images (WSIs) of various cancers.
Methods:
An MxIF cell segmentation machine-learning algorithm was trained and validated on datasets containing 96,765 and 13,810 cells, respectively, from diverse tissues (Table 1). Accuracy was evaluated using F1-score calculation and comparison with StarDist[1] and CellPose[2]. MxIF was performed on FFPE WSIs (n=13) using the PhenoCycler Fusion 2.0 system. Intratumoral (within tumor) and extratumoral (tumor-adjacent) regions were characterized and samples were assessed as immune-hot (>20% CD8+ T cells) and immune-cold (<10% CD8+ T cells)” based on TME immune cell composition and density. Tumor structural barriers were identified by a high presence of vessels (CD31+/CD31+smooth muscle actin [SMA]+ cells) and dense matrix of fibroblasts (vimentin/fibronectin+ cells), often correlating with immune-cold environments[3,4,5]. RNA-seq was performed and cell populations were calculated using the Kassandra cell deconvolution algorithm[6]; concordance with MxIF data was calculated using Spearman’s rank correlation coefficient.
Results:
The cell segmentation tool demonstrated high accuracy in cell detection (0.91 F1-score) compared to StarDist (0.78 F1-score) and CellPose-v2 (0.86 F1-score). MxIF analysis revealed significant heterogeneity in immune landscapes across tumors (Figure 1). Prostate (n=2) and colorectal (n=2) samples exhibited an immune-cold TME, characterized by a predominance of CD68+CD206+ cells. While these cells were exclusively in extratumoral regions of colorectal samples, indicating a lack of immune cells in the tumor, they were present in both intratumoral and extratumoral regions in prostate samples.
CD45+CD3+CD4+ and CD45+CD20+ cells were present in half of breast carcinoma samples (n=6), indicating both immune-hot and immune-cold regions. The high presence of fibroblasts and vessels in extratumoral regions indicated physical barriers to immune cell entry. Similarly, uveal melanoma samples (n=3) presented a heterogeneous immune environment, with CD45+CD3+CD8+ and CD45+CD20+ cells.
High concordance was found between MxIF and RNA-seq in identifying TME cellular subpopulations (Figure 2), with fairly strong correlations for fibroblasts (R=0.7857, p=0.0014), CD4+ T cells (R=0.7875, p=0.0014), and CD8+ T cells (R=0.7182, p=0.0057).
Conclusions:
Our MxIF WSIs pipeline enabled TME immune cell identification with spatial characterization, and cell phenotype classification that was congruent with RNA-seq. These findings highlight the diverse nature of TME landscapes across cancers and the necessity for tailored immunotherapeutic approaches.
Latest publications
