Predicting colorectal cancer patient response to neoadjuvant chemotherapy using the MicroOrganoSphere (MOS) platform.

e15518 Background: Neoadjuvant therapy is a critical component of curative treatment strategies for advanced-stage colorectal cancer (CRC). In this setting, identifying the optimal therapy for each patient is paramount. Yet, 30-40% of patients do not respond to first-line chemotherapy, potentially suffering drug-induced morbidity and tumor progression. Methods to rapidly and reliably predict patient response are urgently needed. Droplet-encapsulated miniature 3D tissue models (MicroOrganoSpheres or MOS) enable informed clinical decision making in a relevant time frame [Ding et al., Cell Stem Cell 2022]. Here, we apply MOS precision medicine screening to a clinically annotated retrospective CRC biobank. We hypothesize MOS can predict clinical response in CRC patients. Methods: MOS were generated from 37 resected primary tumor (n=10) and/or metastatic (n=27) samples from 21 patients treated with neoadjuvant chemotherapy. Each input sample yielded 9,000-15,000 MOS droplets, which were then plated, dosed, and longitudinally imaged via high-throughput and precise lab robotics. MOS response to dose-titrated chemotherapy regimens (FOLFOX, FOLFIRI and FOLFOXIRI) was quantified using a novel, AI-based imaging analysis pipeline to generate drug response curves. From this, MOS ChemoPredictor assay scores were determined, to indicate the relative drug sensitivity of the specific MOS model. ChemoPredictor scores were compared to clinical response on both the patient- (RECIST/PFS) and lesion-specific (pathologic response/percent tumor volume decrease) level. Dichotomous patient stratification based on MOS ChemoPredictor score was subsequently used to identify patients at high risk for progression. Results: In vitro MOS droplet sensitivity recapitulated patient response with high accuracy (>80%), both when established from primary tumor lesions and metastases (Table 1). Within this cohort, highest predictive potential was observed for MOS established from liver metastases (sensitivity 100%, specificity 83%). Kaplan Meier plots and log-rank test results based on patient PFS produced clear separation between most and least responsive MOS. Conclusions: MOS ChemoPredictor score represents a functional assay that can potentially drive clinical decision making. Moreover, screening patient tumors with MOS can identify high-risk patients that will benefit from additional and/or alternative treatments. A recently launched prospective study will verify the predictive value of MOS for neoadjuvant CRC patients in a diagnostic setting. Cohort Patients (n) Samples (n) Sensitivity (%) Specificity (%) Accuracy (%) ROC-AUC 4 PT 1 + LM 2 + LNM 3 21 37 91 69 80 0.85 LM 2 15 25 100 83 92 0.92 1 Primary Tumor; 2 Liver Metastasis; 3 Lymph Node Metastasis; 4 Area under the receiver operating characteristic curve .