(M1130-02-10) Bivariate Statistical Analyses Reveal the Performance of Human FcRn Transgenic Mouse in Predicting Human Pharmacokinetics of Therapeutic Antibodies

Purpose: Cynomolgus monkey (Cyno) has traditionally been a preferred tool for projecting pharmacokinetics (PK) of therapeutic monoclonal antibodies (mAbs) in human (Hu), especially in the linear PK range where nonspecific clearances (CL) mediated by FcRn dominates [1]. However, availability of Cyno has become limited, and there is ongoing effort to conserve the use of Cyno in drug development due to both cost and ethical concerns. Mice carrying transgene for human FcRn (huFcRn) have been developed as a promising tool to replace Cyno in predicting nonspecific CL in humans. This is because the interaction between huFcRn and human IgG, which governs the nonspecific CL of therapeutic mAbs in human, can be recapitulated in these mice [2]. In the present study, using a series of bivariate statistical tests, we compare the performance of huFcRn transgenic mice and Cyno in predicting human PK of therapeutic mAbs.

Methods: PK studies were conducted in Tg32 huFcRn transgenic mice (Tg32, mouse FcRn-/-, huFcRn+/+) using 21 commercially available human IgG mAbs. Serum PK was measured via ELISA following a single-bolus intravenous injection of the mAbs. Non-compartmental analysis (NCA) and two-compartmental model analysis were used to calculate the CL of mAb in Tg32 (Tg32 CL). Clearances of these 21 mAbs in human (Hu CL) and cynomolgus monkey (Cyno CL) were obtained from published literatures. Only CL in linear PK range were used in the analysis. Pearson’s analysis was performed to evaluate the correlation between Tg32 CL vs Hu CL, with Pearson’s correlation coefficient (R) and P-Value (P) calculated to assess the degree of correlation. Receiver operating curve (ROC) analysis was performed to evaluate if Tg32 CL can be used to predict whether a mAb has fast Hu CL ( >5 mL/day/kg). Area-under-receiver operating curve (AUROC) was calculated to evaluate the degree of this predictability, and optimal Tg32 CL cutoff that minimizes the false-positive rate while maximizes the true-positive rate was identified. The Tg32 CL vs Hu CL data were also fitted to the allometric scaling equation to obtain the allometric scaling exponent for Tg32 to Hu CL translation. The accuracy of this translation was assessed by comparing the predict and observed Hu CL. The performances of Tg32 CL in Pearson’s analysis, ROC analysis, and Hu CL translation were compared to that of Cyno CL (Fig 1).

Results: The Tg32 CL of the 21 mAbs were calculated from the serum concentration-time PK profiles, and they range from 4.89 to 12.4 mL/day/kg. Tg32 CL of these 21 mAbs correlated strongly with Hu CL, with R=0.823 and P< 0.0001. Compared to Tg32 CL, Cyno CL was generally well correlated with Hu CL, with R=0.469 and P=0.0319 (Fig. 2a). ROC analysis revealed AUROC=0.93 for Tg32 CL vs Hu CL, indicating Tg32 CL is a highly predictive parameter in identifying mAb with fast Hu CL. ROC analysis also revealed that the optimal Tg32 CL cutoff is 8.5 mL/day/kg, and this CL cutoff provided the highest accuracy in predicting whether a mAb has fast Hu CL. Indeed, majority of the mAb with Tg32 CL that was more than 8.5 mL/day/kg had Hu CL that was above 5 mL/day/kg (Fig. 2b and c). ROC analysis revealed AUROC=0.82 for Cyno CL vs Hu CL, indicating Cyno CL is slightly less predictive, when compared to Tg32 CL, in identifying mAb with fast Hu CL. Using the optimal Cyno CL cutoff of 7 mL/day/kg, 3/21 mAb was mis-identified as having Hu CL < 5 mL/day/kg, while 2/21 mAb was mis-identified as having Hu CL >5mL/day/kg (Fig. 2b and c). Finally, Tg32 CL vs Hu CL data was fitted to allometric scaling equation to identify the optimal scaling exponent of 0.94 for Tg32 to Hu CL translation (Fig. 3a). Using the exponent of 0.94, the predicted Hu CL from Tg32 CL were within 2-fold of observed Hu CL for all 21 mAbs. This prediction accuracy is similar to that of Cyno to Hu CL translation using the reported scaling exponent of 0.85 [3] (Fig. 3b).

Conclusion: Pearson’s and ROC analyses reveal that, when compared to Cyno CL, Tg32 CL is better correlated with Hu CL and is slightly more predictive in identifying mAb with fast Hu CL. Tg32 to Hu CL translation results in accurate prediction of Hu CL. These results suggest that Tg32 is a promising tool to replace cynomolgus monkey in human PK projection.

References: [1] Kamath, Drug Discov Today Technol., 2016; [2] Roopenian et al., Methods in Molecular Biology, 2006; [3] Deng et al., MAbs, 2011

Acknowledgements: All the authors are employees of Genentech Inc.


Fig. 1: Methods used to evaluate the predictability of Tg32 CL for Hu CL. Pearson’s analysis was used to evaluate how well Tg32 CL correlated with Hu CL. ROC analysis was used to evaluate the predictability of Tg32 CL for identifying mAb with fast Hu CL. Allometric scaling equation was used to translate Tg32 CL to Hu CL, and the prediction accuracy was analyzed. The performances of Tg32 CL in these three analyses were compared to that of Cyno CL.


Fig. 2: (a-b) Results of Pearson’s correlation (a) and ROC (b) analyses for Tg32 CL vs Hu CL (top) and Cyno CL vs Hu CL (bottom) for 21 commercially available mAb. Pearson’s R and P are displayed in a. AUROC and optimal CL cutoff are displayed in b. (c) Results of optimal Tg32 or Cyno CL cutoffs in predicting mAb with fast Hu CL (> 5 mL/day/kg), with number of true negative (TN), true positive (TP), false negative (FN), and false positive (FP) noted. TN = slow Tg32/Cyno CL, slow Hu CL. TP = fast Tg32/Cyno CL, fast Hu CL. FN = slow Tg32/Cyno CL, fast Hu CL. FP = fast Tg32/Cyno CL, slow Hu CL.


Fig. 3: (a) Allometric scaling equation was fitted to Tg32 CL vs Hu CL data to identify the optimal allometric scaling exponent of 0.94. (b) Tg32 CL of 21 mAbs was translated to Hu CL using allometric scaling equation and the exponent of 0.94. Cyno CL of 21 mAbs was translated to Hu CL using allometric scaling equation and the exponent of 0.85. The deviations of predict Hu CL from observed Hu CL were calculated for both Tg32 and Cyno translation and compared. Any deviation of more than 2-fold was considered inaccurate.