Background: According to retrospective osteosarcoma series, P-glycoprotein (Pgp) overexpression predicts for poor outcome. A risk-adapted treatment strategy, the ISG/OS-2 trial, evaluated the use of mifamurtide, an EMA-approved innate immunity-modulator, in Pgp-positive patients (pts) (NCT01459484). Here, we present a correlative study to develop a predictive classifier based on tumour immune microenvironment gene profiling. Methods: 62 localized osteosarcoma pts were enrolled at diagnosis. RNA was extracted from pre-treatment FFPE and non-decalcified tissues and was analyzed by PanCancer Immune profiling panel (NanoString Technologies, Seattle, WA, US), including 730 immune genes. 33/62 pts (53%) were Pgp-positive and underwent chemotherapy (CT) and adjuvant mifamurtide, Pgp-negative received CT alone. Univariate Cox regression analysis and pts stratification with the MaxStat package were performed. Receiver operating characteristic (ROC) curve analysis to evaluate the model performance and validation in 2 independent sets were applied. Primary objective was the identification of prognostic signatures of osteosarcoma pts at diagnosis and in pts undergoing mifamurtide. Results: No significant differences in terms of overall survival (OS) and event-free survival (EFS) were shown between Pgp-positive and Pgp-negative pts. Therefore, tumour immune gene expression profiles of all 62 pts were analyzed, irrespective of treatment. First, we identified a 21-gene OS-signature able to stratify all pts into high- and low-risk: 5-year OS for high-risk pts 35.7%, and 89% for the low-risk (p < 0.0001, AUC 0.865). The OS-signature was validated in two independent pts cohorts: GSE16091 (n=34) and GSE33383 (n=87) from the Gene Expression Omnibus (GEO) and significantly distinguished in both cohorts high- and low-risk pts (p < 0.0001) - despite in the validations sets different molecular platforms (Affymetrix Human Genome U133A Array and Illumina Human-6 v2.0, respectively) were used. Next, we identified a 31-gene EFS signature, with a 5-year EFS of 32.2% for high-risk and 93% for low-risk pts (p <0.05, AUC = 0.872). Finally, we focused on the subgroup of pts Pgp-positive, treated with CT+mifamurtide: a 54-gene signature able to discriminate high-risk pts, 5-year EFS of 0%, and low-risk pts, 5-year EFS of 100%, was identified (p <0.05, AUC 0.964). A tumor microenvironment deconvolution analysis is ongoing. Conclusions: Tumour immune microenvironment prognostic gene signatures have been identified for risk stratification of pts with osteosarcoma, regardless of treatment with mifamurtide. Importantly, we have developed a mifamurtide-specific signature that predicts EFS. This promising tool might be used to select pts who could benefit from adjuvant mifamurtide. Clinical trial information: NCT03737435.