PD-L1 negative macrophages associate with poor outcomes and high grade irAE development in melanoma patients receiving anti-PD-1 immunotherapy

To the Editor

Immune checkpoint inhibitors (ICI) have revolutionized treatment for melanoma and other skin cancers. Since the anti-CTLA-4 ICI ipilimumab was approved for advanced metastatic melanoma in 2011, anti-PD-1 ICIs pembrolizumab and nivolumab along with the anti-LAG-3 monoclonal antibody relatlimab have all been approved for metastatic melanoma (Knight et al., 2023). Management of advanced nonmelanoma skin cancers has also progressed significantly due to immunotherapy agents. Anti-PD-1 inhibitors pembrolizumab and cemiplimab have been approved for cutaneous squamous cell carcinoma, and cemiplimab has also been approved in basal cell carcinoma (Aboul-Fettouh et al., 2023). Furthermore, anti-PD-1 inhibitor pembrolizumab and anti-PD-L1 inhibitor avelumab have been approved for Merkel cell carcinoma, a rare and aggressive cutaneous neoplasm (Aboul-Fettouh et al., 2023). Despite these therapeutic achievements, treatment with ICIs in cancer patients can be limited by adverse events, particularly high grade immune related adverse events (irAEs) (Goodman et al., 2023). Advanced cancer patients on ICI therapy who initially develop cutaneous irAEs have a tendency to also experience noncutaneous irAEs (Thompson et al., 2021). In particular, mucositis and psoriasiform reactions are associated with significantly increased likelihoods of developing gastrointestinal and endocrine irAEs, respectively. Downstream irAE diagnoses can hinder cancer management, in some cases requiring suspension of initial ICI therapy.

As such, we should ideally be able to identify which patients will do particularly well on ICI therapy, both responding to treatment and avoiding high grade irAEs (≥ grade 3), or particularly poorly, suffering from both treatment failure and getting high grade irAEs. Multiplex immunofluorescence (mIF) has helped profile and quantify key cell populations that contribute toward immunotherapy outcome in melanoma. We have previously demonstrated that a six-marker mIF panel applied to pretreatment melanoma specimens has biomarker value in predicting anti-PD-1 immunotherapy response (Berry et al., 2021). Here, we assess whether we can predict the development of high grade irAEs using this panel. To do so, we used a two-pronged approach. First, we calculated area under the curve (AUC) values from Receiver Operating Characteristic (ROC) curves, which use the true-positive rate (sensitivity) and false-positive rate (1 – specificity) to create a predictive curve for a binary metric. These algorithms are frequently applied in diagnostic decision making (Lu et al., 2019; Mandrekar, 2010). To account for biases, we also used an unsupervised clustering analysis, Factor Analysis for Mixed Data (FAMD), which accommodates both qualitative and quantitative variables. Furthermore, principal component analysis (PCA; included in FAMD analysis) has previously been established in biology as a guiding tool alongside AUC calculations (Nikas & Low, 2011). We combined both analytical approaches in this study to strengthen our predictive tool and reduce bias.

We evaluated density metrics in formalin-fixed paraffin-embedded (FFPE) samples from a cohort of 35 patients with advanced, unresectable melanoma who received anti-PD-1 immunotherapy. We previously stained these samples with a six-marker mIF panel consisting of PD-1, PD-L1, CD8, FoxP3, CD163, and tumor (S100/Sox10), enabling profiling of T cells, myeloid cells, tumor cells, and PD-1/PD-L1 checkpoint expression (Berry et al., 2021). Whole-slide mIF imaging, processing, and analysis was performed using the comprehensive AstroPath platform (Berry et al., 2021). Here, samples were stratified into two groups: responders without high grade irAEs and non-responders with high grade irAEs. Using our previous sampling approach, we evaluated density metrics across the top 30% of all high-power fields (HPFs) with the highest CD8⁺ T cell densities. We found that this top 30% “hotspot” sampling approach optimized AUC values and significance for predicting immunotherapy response (Berry et al., 2021). Across these HPFs, the ten metrics with the highest AUC scores for predicting graded irAEs and response were CD163⁺PD-L1⁻ cells, CD163⁺ cells, PD-L1⁻ tumor cells, PD-L1^low tumor cells, CD8⁺PD-L1^low cells, CD8⁺FoxP3⁺ cells, CD8⁺FoxP3⁺PD-1⁻ cells, CD8⁺FoxP3⁺PD-L1⁻ cells, CD8⁺PD-1⁻ cells, and CD8⁺FoxP3⁺PD-1^mid cells (Fig. 1a). Individually, these density metrics predicted graded irAEs and response at an AUC score range of 0.70 to 0.84 (Fig. 1a). These ten metrics were then used to create a combined binary logistic regression model, which predicted graded irAEs and response at an AUC score of 0.94 (Fig. 1b). We correlated these findings with FAMD clustering analysis for irAE and ICI response (Fig. 2a). Density metrics clustered into two quadrants: the first quadrant predicted immunotherapy response with no high grade irAEs (R, HGirAE−) and the third quadrant predicted no response to immunotherapy with high grade irAE development (NR, HGirAE+) (Fig. 2a). After pooling the top variables across both quadrants, two density metrics overlapped between the two analysis approaches: CD163⁺PD-L1⁻ cells and CD8⁺PD-1⁻ cells (Fig. 2b). Distributions of density metrics for these two cell populations, according to treatment response and high grade irAE status, are shown in Fig. 2c and Fig. 2d, respectively. When CD163⁺PD-L1⁻ and CD8⁺PD-1⁻ cell density metrics were used to generate a binary logistic regression model, an AUC of 0.90 was calculated from the combined ROC curve (Fig. 2e).

Figure 1. AUC analysis and ROC curve combining top 10 density metrics predicting graded irAE response.

(a) Top 10 density metrics with the highest AUC predictive values, sampled across the top 30% of HPFs with the highest CD8⁺ T cell densities. (b) Combined ROC curve for the top 10 density metrics.

Figure 2. Supplementing AUC-ROC analysis with unsupervised FAMD clustering analysis to generate a combined ROC curve from overlapping metrics.

(a) FAMD analysis of qualitative and quantitative variables from our sampling approach, which identified 12 density metrics over two quadrants that had the highest predictive values for responders without high grade irAEs (R, HGirAE−) and non-responders with high grade irAEs (NR, HGirAE+). (b) Two overlapping density metrics, CD163_PDL1_neg and CD8_PD1_neg, were identified from the analysis approaches. Distribution plots of cell densities according to response and high grade irAE status (R, HGirAE− versus NR, HGirAE+) are provided for (c) CD163_PDL1_neg and (d) CD8_PD1_neg cell populations. Error bars are SEM, and significance was assessed using Mann-Whitney U tests (*p < 0.05, **p < 0.01). (e) The two overlapping cell density metrics were used to generate a combined ROC curve, and a corresponding AUC was calculated.

These findings suggest that our mIF panel may offer biomarker potential for predicting poor response to anti-PD-1 immunotherapy and high grade irAE development in melanoma. Specifically, we find that CD163⁺PD-L1⁻ macrophages promote high grade irAE occurrence and no response to anti-PD-1 immunotherapy, whereas CD8⁺PD-1⁻ T cells promote anti-PD-1 response with low likelihood of high grade irAE development. Previously, we have also identified the CD163⁺PD-L1⁻ myeloid phenotype to associate with immunotherapy non-response (Berry et al., 2021). Additionally, using an image patch-based approach to identify interpretable AI-identified spatial TME neighborhoods, we identified a neighborhood associated with high grade irAEs which contained intratumoral CD163⁺PD-L1⁻ cells and lacked FOXP3, PD-L1, PD-1, and CD8 expression (Jimenez-Sanchez et al., 2023). CD163⁺ cells have also been implicated in irAE development; for example, one pilot study showed that advanced melanoma patients treated with nivolumab who developed irAEs had significantly elevated serum levels of soluble CD163 and increased CD163⁺ cell densities in in-transit metastases (Fujimura et al., 2018). Our findings here provide additional support for a role of CD163⁺PD-L1⁻ macrophages in predicting poor clinical outcomes – including both reduced survival and increased immunotherapy related toxicity – to immunotherapy in melanoma. Limitations of this study include the use of a single multiplex immunotherapy panel on a small cohort of patients with the outcomes of interest. Moving forward, we hope to apply these predictive tools in a larger cohort to further validate and enhance our modeling of immunotherapy response alongside high grade irAE development.

Data Availability Statement

The authors affirm that all data necessary for confirming the conclusions of the article are present within the article, figures, and tables. Additional data that support the findings of this study are available from the corresponding author (joelsunshine@jhmi.edu), upon reasonable request.