Skip to main content


Fig. 5 | Journal for ImmunoTherapy of Cancer

Fig. 5

From: Distinct predictive biomarker candidates for response to anti-CTLA-4 and anti-PD-1 immunotherapy in melanoma patients

Fig. 5

Parameters that correlate with anti-CTLA-4 and anti-PD-1 clinical response were verified by the automated Citrus algorithm. CyTOF .fcs files were analyzed using the Citrus algorithm to establish a predictive model for anti-CTLA-4 treatment. (a) Citrus clustering of immune subsets (represented by metal_marker channels) based on CyTOF data from baseline PBMCs of anti-CTLA-4 treated patients are shown as graphs. The distribution of each immune subset is presented as an individual graph. The heat spectrum associated with each graph indicates the expression level of each channel in a cluster. (b) A Nearest Shrunken Centroid (PAMR) model described the minimum number of channels and clusters needed to distinguish responders from non-responders to anti-CTLA-4 with the lowest error rate. (c) Comparisons of metal signal intensities (mean) of indicated channels are shown from left to right: CD45RA (153Eu) in base CD4+ and CD8+ clusters, and PD-L2 (172Yb) intensity in the base monocyte cluster in responders vs. non-responders to anti-CTLA-4 therapy. Two clusters from the Citrus predictive model in (b) are not demonstrated in (c) The cluster highlighted for the signal of HLA-DR (157Gd) is not associated with any major immune subsets in this analysis. The cluster highlighted for the signal of CD38 (156Gd) is the base cluster of total PBMC§. §CD38 expression on total PBMC was also assessed, and no significant difference between responders and non-responders was found in the univariate analysis.

Back to article page