Predictors of Biologic Efficacy with Lovotibeglogene Autotemcel (Lovo-cel) Gene Therapy in Patients with Sickle Cell Disease

Introduction Lovo-cel, a one-time gene therapy that produces an anti-sickling adult hemoglobin (Hb), HbAT87Q, has been investigated for the treatment of sickle cell disease. Objective To examine the correlation between various lovo-cel drug product (DP) attributes and hematologic parameters to better understand what factors impact the biologic efficacy of lovo-cel. Methods This study used data from 47 patients who received lovo-cel as of February 13, 2023, using the mobilization and manufacturing process applied in HGB-206 (NCT02140554) Group C and HGB-210 (NCT04293185). First, a multivariate analysis was conducted to understand pairwise relationships between postinfusion transduction efficiency metrics and selected hematological and hemolysis indices. Subsequently, a cross-validated random forest (RF) regression was designed to identify attributes predictive of peripheral blood (PB) vector copy number (VCN) using 15 DP characteristics. RF regression uses machine learning to generate a best fit model to the data and identify variables that influence the fit. Missing data were imputed using the k-nearest neighbor algorithm (k=5), and hyperparameters were tuned to minimize the root-mean-square error (RMSE). The final RF model in R yielded an RMSE of 0.78 and an R2 value of 0.28, indicative of a reasonable model fit to the data. Results Markers of hemolysis (eg, reticulocyte count, bilirubin levels, and erythropoietin) were all inversely correlated with measures of postinfusion transduction efficiency (eg, PB VCN, HbAT87Q levels, etc; Figure 1). This indicates that a higher proportion of transduced cells is associated with an increased resolution of sickle cell–associated hemolysis. Additionally, an assay that assesses the propensity of red blood cells to sickle under low oxygen conditions (ie, ex vivo sickling assay) was significantly correlated with total Hb after lovo-cel infusion. These findings prompted us to extend our machine learning framework to understand the relationship between DP attributes that can be measured prior to infusion and PB VCN as the outcome metric. This model demonstrated that DP VCN in the myeloid lineage was the most predictive attribute of PB VCN compared with any other attribute measured prior to infusion (Figure 2). Conclusion Our findings indicate that in clinical studies of lovo-cel, the reduction of hemolysis is significantly correlated with postinfusion transduction efficiency. We conclude that transduction efficiency, as measured by PB VCN, is an important predictor of biologic efficacy and correlates with lovo-cel DP attributes. Consequently, attributes of DP transduction efficiency as well as the resulting PB VCN predict successful treatment outcomes with lovo-cel.