Background: Checkpoint inhibition with anti-PD1 is able to elicit objective response rates (ORR) of ̃ 20% in patients with advanced hepatocellular carcinoma (aHCC) but molecular biomarkers predicting response remain unknown. Here, we define biomarkers of response and primary resistance to anti-PD1 in aHCC by analyzing molecular features prior to systemic therapy. Methods: Through an international consortium of 13 referral centers, we gathered 111 fresh and archived tumor samples from patients with advanced HCC treated with single agent anti- PD1 therapy. Genomic analysis was performed with whole genome expression arrays, CTNNB1 exon 3 mutation assessment and histological evaluation. Results: Overall, we performed transcriptomic analysis in 83 patients, 28 of which were treated in first-line (ORR: 42.9%) while the remaining 55 patients were treated either in 2nd (41 patients, ORR 29.3%) or 3rd line (14 patients, ORR 7.1%) after prior therapy with tyrosine- kinase inhibitors (TKI) sorafenib or lenvatinib. In patients treated in frontline, response was associated with a significant upregulation in Interferon-γ signaling (IFNγ) and genesets relatedto the antigen presentation machinery (FDR < 0.05). Through differential expression analysis we defined an 11-gene signature that was significantly associated with both major pathways (FDR < 0.01) and was capable of predicting OR as well as progression-free- (PFS) and overall survival (OS) in patients with aHCC. The signature was validated in three independent cohorts of melanoma, lung cancer, and head and neck squamous cell cancer patients were high expression was associated with a significant increase in ORR and longer PFS. In HCC, high expression of the signature was associated with a distinct profile in the immune infiltrate, where an increase in M1 macrophages (p = 0.003), CD4 memory T cell (p < 0.01) and CD4 naïve T cell-infiltration (p = 0.026) was observed. Conversely, low expression of the signature was associated with a marked upregulation of regulatory T cells (p < 0.001). No association was found, however, between either the overall immune infiltrate or CTNNB1 mutations and response. In patients that were treated with TKIs in frontline before receiving subsequent anti-PD1 therapy, the signature was no longer able to predict either OR or PFS/OS, suggesting that TKIs may reshape the microenvironment in a way that renders previously inflamed tumors no longer amenable to anti-PD1. Conclusions: Here, we define an 11-gene signature of response to anti-PD1 in first line aHCC. The signature was validated in patients with other solid cancer types, where it retained its predictive ability. Of note, the signature was not able to identify responders among HCC patients that were treated with TKIs prior to anti-PD1 therapy. The molecular changes induced by different treatment lines would, thus, require fresh biopsies prior to anti-PD1to enable biomarker-driven treatment.