In this prospective study, we included 131 patients diagnosed with skin cancer (malignant melanoma, cutaneous squamous cell carcinoma, Merkel cell carcinoma), who were treated at the Department of Dermatology, Medical School Hannover from April 2019 to May 2023. All patients were in advanced stages and received either PD-1 monotherapy or a combined checkpoint inhibition including PD-1 and CTLA-4 inhibitors. Longitudinal serum samples were collected per patient at two different time points: pre-treatment (baseline) and 3–4 weeks after the first treatment cycle. The study design is shown in Supplementary Figure S1. All serum samples were centrifuged at 3000g for 15 minutes and supernatants were stored at -80°C until further analyses. Patients were monitored regularly over the course of therapy to allow early detection of irAEs. If irAEs occurred, affected organ, type of irAEs, severity, onset time and treatment regimen were thoroughly documented according to the National Cancer Institute’s Common Terminology Criteria for Adverse Events, version 4.0. (CTCAE).

Serum samples were analyzed using customized highly sensitive electrochemiluminescence immunoassays on the MESO Quickplex SQ120 according to the manufacturer’s instructions (V-Plex 54, U-Plex) (Mesoscale Discovery, Maryland, USA). Data were imported into the Mesoscale Discovery Workbench 4.0 analysis software for quantification (Mesoscale Discovery, Maryland, USA). Standard curves were used to assess the correct concentrations. If concentrations were lower than the detection limit, the parameter was excluded. Additionally, we collected routine laboratory parameters (LDH, white blood cell count, neutrophil counts, eosinophil counts, lymphocyte counts). The study was approved by the ethics committee in Hannover in accordance with the Declaration of Helsinki (vote: Nr 8685_BO_K_2019). Written informed consent of the patients was obtained.

For statistical analysis and visualization of the results GraphPadPrism (version 9.02), SPSS (version 29.0) and R (version 4.5.0.) was used. Illustrations were made in BioRender. Differences in the Log2FC with regard to cytokine concentrations between baseline and 3–4 weeks after therapy initiation were evaluated with the Wilcoxon-signed rank test. Differences between stratified patient groups and clinical characteristics were evaluated using the Chi² Test. In order to evaluate the predictive potential of specific parameters, univariable and multivariable analyses were carried out by the logistic regression model and the Cox proportional hazard model. A p-value less than 0.05 was regarded as statistically significant. To calculate the cytokine risk score, the values (the intercept and the respective regression coefficients) obtained from a multivariable logistic regression analysis were subsequently used based on this formula: cytokine risk score = α + ß₁*X₁ + ß₂*X₂ + ß₃*X₃

α = intercept; ß_1, ß_2, ß₃ = regression coefficients, X_1, X_2, X₃₌ cytokine values.

From April 2019 to May 2023, we enrolled a total of 131 consecutive patients, who received ICI as a standard of care at Hannover Medical School. The median follow-up time from start of ICI was 18 months (inter quartile range: 11–27 months). The median age was 65 years and 55.7% of the patients were male. Melanoma was the most frequent cancer entity (n=122) followed by cSCC, MCC and mucosal or uveal melanoma. Half of the patients were categorized as stage III (49.6%) and the other half as stage IV (50.4%) according to AJCC 2017. Most patients received anti-PD-(L)1 monotherapy (mICI) (76.3%) whereas a minority was treated with a combined ICI (cICI) of anti-PD-1 and anti-CTLA-4 therapy (23.7%). 43.5% of the patients carried a BRAF mutation. The patient characteristics are summarized in Table 1.

63.4% of the patients developed irAEs. Severe irAEs (CTCAE grade 3 and 4) occurred in 24.4% of the patients. The distribution of irAEs grades in regard of treatment regime can be found in Figure 1. The average time from therapy start to first irAEs symptoms was 14,2 weeks (Figure 1). The most frequently observed irAEs affected the skin (19.8%), colon (15.3%) and thyroid (15.3%). Among the key baseline patient characteristics, the use of combined ICI (cICI) with Nivolumab plus Ipilimumab significantly correlated with the occurrence of irAEs (Table 1). The other clinicopathologic variables sex, age, LDH and BRAF mutation showed no significant differences between patients with and without irAEs (Chi²-test, Table 1). We further analyzed baseline laboratory parameters (leukocyte, lymphocyte, eosinophil, and neutrophil counts, as well as the neutrophil-to-lymphocyte ratio [NLR] and derived neutrophil-to-lymphocyte ratio [dNLR]) prior to initiation of immunotherapy. We compared these parameters between patients who did not develop any irAEs and those who developed irAEs of any grade, or grade 3 or higher, or those with the most frequently observed irAEs (skin, colon, thyroid, pituitary gland, and liver).

Applying a Mann-Whitney-Test we did not detect any differences in baseline laboratory parameters between patients with and without irAEs. However, we observed a higher leukocyte count in patients with skin-related irAEs and a lower eosinophil count in patients with thyroid-related irAEs (Supplementary Figure S2).

In our cohort of 131 patients, we analyzed 57 different serum cytokines using electrochemiluminescence both at baseline as well as 3–4 weeks after the first ICI administration. Sixteen parameters were below the quantification limit and were therefore excluded from further analysis. Baseline concentration of the remaining 41 quantifiable parameters (listed in the heatmap in Figure 2) were assessed to determine whether cytokine expression could predict immune related toxicity. Therefore, we compared patients who developed irAEs of any grade to patients who did not experience any irAE at all during the observation period. In addition, we compared cytokine levels between patients who developed severe (CTCAE grade 3-4) irAEs and patients who showed no irAEs. From all tested baseline parameters only IL-27 and BCA/CXCL-13 showed a higher concentration in patients, who did not develop irAEs compared to patients with irAEs (Figure 2B).

To account for the baseline variability in cytokine concentrations and to assess early dynamic changes, we further studied relative cytokine changes and calculated the fold change (FC) in cytokine levels (post-treatment relative to pre-treatment). To visualize the dynamic cytokine pattern across the patient cohort, we performed hierarchical clustering based on longitudinal log2 FC in cytokines levels. Patients were stratified based on the severity of irAEs (0–4). Results are displayed in a heatmap (Figure 2A). We showed that patients with irAEs grade 3–4 display a strong upregulation of pro-inflammatory and interferon-related cytokines such as IFN-γ, IP-10, IL-6, and IL-1β. In contrast, patients without irAEs exhibited lower cytokine activity across the panel. These findings suggest that increasing irAEs severity is associated with a progressive shift towards a pro-inflammatory cytokine milieu. In Figure 2B cytokines are shown which are significantly upregulated on early treatment course in patients who developed irAEs (IFN-γ, IL-1RA, MIP-1alpha, CRP). However, some cytokines were rather downregulated in patients who experienced irAEs compared to patients without irAEs, in particular IL-7 and IL-8. This suggests that both activation and suppression of specific immune pathways may contribute to toxicity profiles.

Furthermore, the five most frequently affected organ systems were analyzed separately within the dataset, yielding distinct cytokine patterns. The most pronounced differences were observed between patients without irAEs and those who experienced adverse events in liver or thyroid, suggesting organ-specific cytokine signatures (Figure 2B).

In order to find early predictors for the occurrence of irAEs, we applied univariable logistic regression analysis with clinical as well as cytokine parameters. As expected, patients receiving cICI therapy exhibited a markedly higher incidence of irAEs (OR = 7,636 [95%CI=2,179-26,764; p<0,001]). In addition to treatment type, we identified further contributing factors such as patient age (OR = 0,971[95%CI=0,946-0,996; p=0,024]) and early changes in cytokine levels, specifically for IL-7 (OR = 0,381 [95%CI=0,193-0,755; p=0,006]), BCA-1/CXCL13 (OR = 2,438 [95%CI=1,173-5,069; p=0,017]), and IL-1RA (OR = 1,901 [95%CI=1,038-3,480; p=0,037]), following the first ICI administration. However, in multivariable analysis, only the type of therapy (p<0,03) and IL-7Log2FC (p<0,017) remained statistically significant (Table 2). The development of severe irAEs was further significantly associated with an increase of IFN-γ, Eotaxin-3, IL-27 and TSLP post treatment (Supplementary Table S1).

To enhance the predictive performance, we calculated a composite cytokine risk score for the occurrence of irAEs including cytokines with a significant p-value in univariable logistic regression model (IL-7Log2FC, BCALog2FC, IL-1RALog2FC). To calculate the cytokine risk score, we used the intercept and the respective regression coefficients of the cytokines derived from the multivariable logistic regression model. Subsequently, a ROC analysis was performed (AUC: 0.710) to determine the optimal cut-off of the cytokine risk score for stratifying the cohort into low- and high-risk groups, based on the Youden Index. The cytokine risk score was further evaluated in a time-dependent Cox regression model and emerged together with the cICI as an independent predictor of outcome in both univariable (HR = 1,801 [95%CI=1,424-2,277; p<0,001]) and multivariable analyses (HR = 1,407 [95%CI=1,072-1,846; p<0,014]) (Table 3). In order to validate the cytokine risk score and check for possible improvement, different methods of preprocessing, filtering or scaling were used. Detailed information can be found in supplements (Supplementary Figure S3 and S4).

Stratification of the cohort according to the cytokine risk score revealed that patients with a high score were significantly associated with an increased hazard of experiencing irAEs (Figure 3). Moreover, further subgroup analysis indicated that patients with a higher cytokine risk score had a significant higher risk of irAEs affecting the skin, colon, thyroid and liver (Supplementary Table S2).