3560 Background: Currently, there are no well-established biomarkers available to select mCRC patients (pts) who will benefit most from antiangiogenic therapy. RAS mutant (mt) allele fraction in plasma (plMAF) is an independent prognostic factor in mCRC. Preliminary data from our group suggests the possible predictive role of plMAF in RAS mt pts treated with 1st line chemotherapy (ct) +/- bevacizumab (bev). Methods: Data prospectively/retrospectively collected from RASmt mCRC pts who received 1^st line ct+/-bev treatment in our center, selecting the subset of pts with plMAF sample evaluable with digital PCR (BEAMing) at baseline. Pts were stratified as high (≥ 5.8%) or low ( < 5.8%) plMAF, based on a previously established prognostic cutoff (Elez et al, Mol Onc 2019). We investigated the associations between clinic-pathological variables, overall survival (OS), and progression-free survival (PFS) stratified by plMAF RAS levels using Cox regression models. OS and PFS were calculated by Kaplan-Meier method. Murine PDX were developed from mCRC pts including models from patients with KRASG12/G13 mutations to explore recapitulation of the clinical findings. Results: From October ‘17 to December ‘21, 102 basal plasma samples were analyzed by BEAMing. 47 pts (46%) were classified as high, 39 pts (38%) as low, and 16 (16%) as no-mt detected by BEAMing (non-shedding). OS was significant longer in low plMAF pts than in high plMAF pts (median OS 15.9 vs 37.1 months (mo); HR 0.43; p = 0.001). In high plMAF pts, a trend towards a better PFS was observed in those pts treated with ct+bev compared to ct alone (median 9.4 vs 6.1 mo; HR 0.6; p = 0.18). No differences were observed in low plMAF pts treated with ct +/- bev (median 14.5 vs 14.9 mo; HR 1.2; p = 0.58). Results were not modified when adjusted by the presence of liver metastases. The multivariate PFS model showed no association between RAS plMAF and clinicopathological variables, except for treatment benefit with ct+bev and better outcomes in pts with resectable liver metastases. Interestingly, human ctDNA in one murine PDX model from non-shedding pt was not detectable, whereas human ctDNA was present in another PDX model with a plMAF of 36.2%. Conclusions: This study suggests that plMAF could be a promising predictive biomarker of response to bev in 1^st line RASmt mCRC, but more pts need to be analyzed to confirm this effect. Our results confirm the prognostic role of RASmt plMAF in mCRC. plMAF could partially depend on tumor cell shedding degree and characteristics on the tumor vasculature architecture, this is being investigated in ongoing imaging studies and PDX models. These models will also help to understand the biology behind tumor response to bev and its connection with ctDNA shedding.