4556 Background: Both human and rodent studies provide evidence for a role of the microbiome in patients who respond to checkpoint inhibition (CI). So far, no study has unraveled the physiological link between intestinal and tumor microbiome composition in relation to response to CI. The PERFECT trial was a single-arm phase II feasibility study investigating the addition of atezolizumab (PD-L1 inhibitor) to neoadjuvant chemoradiotherapy (nCRT) for resectable esophageal adenocarcinoma (NCT03087864). An exploratory objective of this trial was to evaluate intestinal and tumor microbiome composition including plasma metabolomics as potential biomarkers for immunological and pathological response. Methods: Using 16S rRNA gene sequencing, we analyzed fecal, duodenal and tumor samples at baseline (V0), 3 weeks after start of atezolizumab (V1), and 1 week before surgery (V2). We compared microbiome composition and metabolomics from patients with pathological complete response (pCR; ypT0N0) to patients with a pathological incomplete response. Differences in alpha diversity metrics were tested using mixed linear models. Beta-diversity associations were assessed using permutational MANOVA (adonis) and multilevel PCA (mixOmics). Biomarkers were identified using a machine learning model (XGboost) feature selection. Plasma metabolomics (Metabolon) were determined with liquid chromatography mass spectrometry (LC-MS). Results: Microbiome profiles were significantly altered after start of treatment in all sample types. None of the sample types showed a relation between alpha or beta diversity and pCR. On taxonomical level, we found that the tumor and duodenal baseline samples were weak predictors for response (AUC 0.60 and 0.62, respectively), but better compared to fecal microbiome composition (AUC = 0.49). We identified the top 20 microbes that predicted pCR best in tumor and fecal samples and found significant correlations with metabolites involved in bile acid metabolism. Conclusions: Both tumor and duodenal baseline biopsies were better predictors of pathological response compared to fecal microbiome. Microbes predictive of pCR showed significant correlations with metabolites involved in bile acid metabolism, which is known to indirectly influence immunosurveillance in cancer. Data on immune response in relation to the microbiome and metabolomics are expected Spring 2020. Clinical trial information: NCT03087864 .