Liquid biopsy protein biomarkers to predict responses and elucidate resistance to cancer immunotherapy The response of metastatic melanoma to anti-PD1 is heterogeneous. We performed proteomic profiling of patient plasma samples to build a predictor of immunotherapy response and uncover biological insights underlying primary resistance. An initial cohort comprised 55 metastatic melanoma patients receiving anti-PD1 (Pembrolizumab or Nivolumab) at Massachusetts General Hospital (MGH), and 116 additional patients comprised a validation cohort. Plasma samples were collected at baseline and on-treatment, at 6 weeks and 6 months’ time-points, and profiled for 1000 proteins by a multiplex Proximity Extension Assay (PEA, by Olink Proteomics). A subset of patients had single-cell RNA-seq (Smart-Seq2 protocol) performed on tumor tissue. Group differences and treatment effects were evaluated using linear mixed models with maximum likelihood estimation for model parameters, and Benjamini and Hochberg multiple hypothesis correction. At the baseline, 6 differentially expressed proteins were identified between responders (R) and non-responders (NR) whereas immune suppression marker ST2 and IL-6 were found significantly higher among NR. Kaplan-Meier survival curves stratified by the baseline differentially expressed proteins were highly predictive of overall survival (OS) and progression-free survival (PFS). At 6-weeks on-treatment time point, 80 proteins were found differentially expressed between R and NR including several proteins implicated in primary or acquired resistance (IL8, MIA, TNFR1 among others). Several 6-weeks differentially expressed proteins were highly predictive of survival (ICOSL, IL8, MIA). Furthermore, 160 significantly differentially expressed (DE) proteins were identified across the treatment period majority of which are reflective of immune activation under the pressure of the immunotherapy. Analysis of single-cell RNA-seq data of tumor tissue from a subset of these patients revealed that gene expression of most proteins predictive of response were enriched among tumor myeloid cells, with the remainder of proteins being reflective of exhausted T cell states. These results unveil a putative role of myeloid cells within the tumor microenvironment in anti-PD1 response or primary resistance. Whole plasma proteomic profiling of anti-PD1 treated patients revealed DE proteins between R and NR that may enable a liquid biopsy to predict anti-PD1 response. Importantly, we demonstrate the relationship of serum biomarkers to OS and PFS and are currently attempting to build machine learning classifiers as predictors of response to checkpoint therapy leveraging early and late on-treatment time points.