Results
A total of five vectors for hemophilia A therapy and three vectors for hemophilia B therapy were included in this study. The rAAV2-CMV-FIX vector included in Tang et. al.’s report was excluded because plasma FIX levels were less than 1% of normal in hemophilia B patients receiving this vector across all dose levels and it was difficult to distinguish endogenous FIX and FIX produced by GT.5 The preclinical and clinical doses, plasma FVIII activity or FIX levels, normalized FIIIV or FIX in blood circulation, and calculated GEF for the 8 vectors were summarized in supplemental Tables S2 and S3. The data of individual animal or patient were listed in the tables if they were available in literature. Otherwise, only mean values were listed in the tables.
Normalized FIX or FVIII amounts in blood circulation of animals and individual patient were plotted against total vector doses (vg). A power regression model could describe the relationships between normalized transgene protein and dose across three species for the eight vectors (Figures 1 and 2). Four of the eight vectors showed three-species correlation R2 values > 0.8 and the other four vectors (scAAV2/8-LP1-hFIXco, GO-8, SB-525, and SPK-8011) showed three-species correlation R2 < 0.8. The low R2 values of the four vectors were mainly due to large inter-subject and inter-animal variability in transgene product levels and small sample size at each dose level. In most studies, AAV vectors were administered to less than six animals or patients at each dose level. When normalized mean FVIII or FIX amounts were used for regression, R2 values were substantially improved (supplemental Figures S1 and S2). Only scAAV2/8-LP1-hFIXco vector showed a three-species correlation R2 < 0.8 (R2 = 0.7682) (Figure S1C). The low R2 indicated a substantially lower gene expression activity of scAAV2/8-LP1-hFIXco in patients compared to animals, which was likely due to intense T cell responses to this vector in hemophilia B patients.13, 14 Overall, the moderate-to-strong correlation between normalized transgene product and vector dose (supplemental Figures S1 and S2) suggested that it was feasible to normalize dose response in multiple species to a species-invariant scale using an exponent of 0.25.
A power model was also used to generate regression equations based on the data of two preclinical species (mouse and dog or monkey) (Figures 1 and 2). The regression equations derived from preclinical data were then used to predict FIH doses of eight vectors which were anticipated to achieve targeted plasma levels of FIX (250 ng/mL), FIX-Pauda (27.5 ng/mL) or FVIII (12 IU/dL). As a comparison, FIH doses of the eight vectors were also predicted using direct vg/kg conversion and allometric scaling approaches.5, 9 The plots of allometric scaling analyses were shown in supplemental Figures S3 and S4. The FIH doses predicted by the three approaches were compared with clinical doses (Table 1). The FIH doses predicted by the three approaches were ranked as: allometric scaling > dose-response normalization > direct vg/kg conversion for most vectors except for SPK9001. For SPK9001, the predicted FIH doses were ranked as: allometric scaling > direct vg/kg conversion > dose-response normalization. Among the eight vectors, dose-response normalization generated more accurate FIH doses for three vectors (SPK-9001, GO-8, and BMN270). Allometric scaling provided more accurate predictions for three other vectors (rAAV2-hAAT-FIX, scAAV2/8-LP1-hFIXco and SB-525), and direct vg/kg conversion gave more accurate predictions for DTX201 and SPK-8011.
As shown in Figures 1 and 2, the total vector dose (vg) received by patients were usually much higher than that received by animals. To extrapolate dose-response from animals to patients, ideally, animal and human vector dose ranges should overlap. The animal and human dose ranges of scAAV2/8-LP1-hFIXco (Figure 1C), SPK-9001 (Figure 1E), GO-8 (Figure 2A), and SPK-8011 (Figure 2I) overlapped. Compared to direct vg/kg conversion and allometric scaling, dose-response normalization provided more accurate FIH dose predictions for SPK-9001 and GO-8 (Table 1). All the three approaches underestimated FIH dose of scAAV2/8-LP1-hFIXco and dose-response normalization underestimated the FIH dose by 9-fold, which was likely caused by intense T cell responses to the vector capsid in patients. Seven among 8 patients receiving medium-to-high dose of scAAV2/8-LP1-hFIXco reported T-cell responses against capsid, which could eliminate vector-transduced hepatocytes (Table S4).13, 14 All the three approaches overestimated FIH dose of SPK-8011. Dose-response normalization overestimated FIH dose of SPK-8011 by approximately 4-fold and direct vg/kg conversion provided a more accurate prediction. SPK-8011 contains bio-engineered AAV capsid which can transduce human hepatocytes more efficiently than animal hepatocytes. In a humanized mouse liver model, the bio-engineered SPK-8011 vector transduced primary human hepatocytes 100-fold higher than natural AAV8.15 The exceptionally high transduction efficiency of SPK8011 in human hepatocytes might lead to the overprediction of FIH dose.
For vectors rAAV2-hAAT-FIX, SB-525, BMN270, and DTX201, the dose ranges of animals and patients did not overlap. For example, the lowest patient dose and the highest monkey dose of SB-525 were 1.4 × 1014 vg and 1.5 × 1013 vg, respectively (Figure 2C and 2D). The non-overlapping pattern between animal and human doses might compromise animal-to-human extrapolation results. Among the four vectors, dose-response normalization provided an accurate prediction of FIH dose for BMN270 only. Similar to scAAV2/8-LP1-hFIXco, all the three approaches underestimated FIH dose of rAAV2-hAAT-FIX and dose-response normalization underestimated the FIH dose by 4 - 5 fold. Consistently, patients receiving a high dose of rAAV2-hAAT-FIX developed capsid-specific T cell responses (supplemental Table S4). In contrast, T-cell responses against capsid were either not observed or not investigated in the clinical studies of other vectors. The exact reasons for poor FIH dose predictions for SB-525 and DTX201 by dose-response normalization are not clear but may be associated with the non-overlapping pattern between human and animal doses.