A comprehensive evaluation of the human Tg32 and Tg276 FcRn transgenic mouse and non-human primate models for predicting antibody pharmacokinetics in human

ABSTRACT

In this study, we present a comprehensive evaluation of two human FcRn transgenic mouse models, Tg276 and Tg32, demonstrating their ability to predict the pharmacokinetic (PK) parameters of IgG-type antibodies in human and monkey (non-human primate, NHP), including molecules with and without half-life extension. To assess the translational relevance of the humanized FcRn mouse models, we integrated a broad dataset comprising in vivo PK parameters in both animals and humans derived from the literature and in-house experiments. Using this dataset, we optimized scaling exponents, performed allometric scaling, and evaluated the predictive performance of the models. The optimized exponents fell within expected ranges, 0.7 to 0.95 for systemic and intercompartmental clearance, and around 1 for central and peripheral volumes of distribution. Our analysis reveals the need to apply distinct scaling exponents for half-life extended versus non-extended antibodies, as well as model-specific exponents. The results demonstrate that human PK predictions from Tg32 and Tg276 mice are comparable to those from the NHP. These models therefore offer a good alternative to the monkey, potentially reducing the need to conduct early in vivo PK studies in NHPs. We also report for the first time the use of an immunocompromised version of the Tg276 model to mitigate anti-drug antibody responses. Our findings show that the Tg276 hemizygous model exhibits a translational performance equal to, and in some cases superior to, that of the Tg32 homozygous strain and the NHP, specifically, in terms of predicted clearance. These insights support the Tg276 model as a valuable tool for early-stage antibody screening and lead optimization in preclinical PK evaluation.

Introduction

Therapeutic monoclonal antibodies (mAbs) offer high target specificity combined with prolonged systemic exposure. The half-life of mAbs is a critical determinant of several clinical parameters, including the first-in-human dose, the amount of antibody required to achieve therapeutic levels, and the dosing frequency.¹ A key contributor to their favorable pharmacokinetic (PK) profile is the neonatal Fc receptor (FcRn), a protein expressed by endothelial cells and circulating monocytes.² Typically, serum proteins are internalized by endothelial cells and directed to lysosomes for degradation and amino acid recycling, resulting in short half-lives of 1–5 days. IgGs are a notable exception: under acidic conditions in the endosome, they undergo conformational changes that promote high-affinity binding between their Fc region and FcRn. This interaction triggers sorting of the IgG – FcRn complex into recycling vesicles that fuse with the cell membrane, where a neutral pH reduces binding affinity, enabling the release of IgGs into the bloodstream. Recognition of FcRn’s central role in regulating antibody PK has led to the engineering of Fc mutations that enhance binding to FcRn at low pH without impairing release at neutral pH.³ Prominent examples are the LS dual mutation M428L/N434S and the YTE triple mutation M252Y/S254T/ T256E.⁴

Given its importance in drug development, considerable effort has been devoted to understanding the factors that influence antibody half-life. Shi⁵ has reviewed how physicochemical properties such as glycosylation, charge, and aggregation affect FcRn interactions and systemic disposition. Dostalek et al.⁶ developed an FcRn transcytosis assay to rank Fc-engineered mAbs by recycling efficiency, correlating in vitro transcytosis with in vivo clearance in rodents and primates. More recently, Jain et al.⁷ integrated FcRn binding, nonspecific interactions, and self-association metrics into a multivariate developability framework, improving early prediction of in vivo clearance in humanized FcRn mice. However, despite advances in in silico and in vitro prediction methods, these approaches remain insufficiently validated, and in vivo studies are still required to accurately determine organismal PK parameters, such as clearance and half-life.

The cynomolgus monkey (a non-human primate, NHP) is a widely used preclinical model for evaluating antibody PK and pharmacodynamics due to its physiological similarities to human. Several studies have assessed the predictive accuracy of NHPs using allometric scaling of systemic clearance (CL) and volumes of distribution (central, Vc; peripheral, Vp). For example, Ling et al.⁸ used 18 mAbs to predict human CL from monkey data, estimating optimal scaling exponents of 0.85 for soluble targets and 0.90 for membrane-bound targets. Wang and Prueksaritanont⁹ reported an exponent of 0.8, while Deng et al.¹⁰ found 0.85 to be optimal. Oitate et al.¹¹ determined exponents of 0.79 and 1.12 for soluble antigens, and 0.96 and 1.00 for membrane-bound antigens, for CL and volume of distribution, respectively. Similar analyses by Dong et al.¹² and Haraya et al.¹³ yielded consistent scaling exponents for CL, intercompartmental clearance (Q), Vc, and Vp, while Haraya and Tachibana¹⁴ explored the scaling factors for half-life extended mAbs. Overall, for non-half-life extended molecules, the scaling exponent for CL from monkey to human typically falls within the range of 0.75–0.85, while it is lower for half -life extended molecules.

Although NHPs provide extremely accurate preclinical PK evaluation of mAbs, their use raises ethical concerns, is expensive, and faces supply chain constraints due to the rapid expansion of the mAb field. For these reasons, the use of NHPs is often limited to later-stage toxicological or dose-range-finding studies, and alternative in vivo models for PK studies are urgently needed, especially during lead selection and optimization. Conventional rodent models are limited by species-specific differences in FcRn – IgG binding: human IgGs bind murine FcRn with abnormally high affinity at acidic pH, leading to overestimated half-lives and underestimated clearance.¹⁵ To address this translational gap, human FcRn transgenic mouse strains such as Tg32 and Tg276 have been developed.¹⁶ These models replace expression of the mouse FcRn alpha chain, encoded by the Fcgrt gene, with that of its human counterpart, resulting in FcRn receptors with human-like characteristics. Tg32 expresses the full human FCGRT gene under its native regulatory sequences, whereas Tg276 uses a ubiquitous CAG promoter, leading to elevated FCGRT expression across tissues.^17,18 Compared to Tg32 mice, Tg276 mice clear human antibodies more rapidly and are more sensitive to Fc-mediated recycling differences among antibody variants (Proetzel and Roopenian, ¹⁹; Valent et al.²⁰). Systematic analyses have assessed the predictive accuracy of Tg32-derived parameters relative to NHP and human. Avery et al.²¹ compared PK data from hemizygous and homozygous Tg32 mice with those from NHPs for 27 mAbs, demonstrating that homozygous Tg32 mice provide superior human PK predictions. Similarly, Betts et al.²² reported that homozygous Tg32 mice and NHPs yielded comparable predictions of human PK parameters, with allometric exponents for CL of 0.90 (Tg32) and 0.81 (NHP), and near-unity exponents for volumes of distribution. In their study, Q scaled with exponents of 0.57 (Tg32) and 0.67 (NHP). Expanding this evaluation, Valente et al.²⁰ directly compared Tg32 and Tg276 mice to NHPs using 16 antibodies – mono-, bi-, and tri-specific – encompassing both wild-type (WT) and Fc-mutated formats and derived CL exponents of 0.91 (0.97 for WT) and volume exponents of 0.93 (0.96 for WT). More recently, Haraya et al.²³ investigated 11 Fc-engineered mAbs (YTE and LS) in Tg32 mice, estimating optimal exponents of 0.73 for CL, 0.60 for Q, 0.95 for Vc, and 0.87 for Vp, thereby suggesting that half-life extended (HLE) antibodies may require distinct allometric factors. While the Tg32 mouse model has been widely adopted due to its physiologically relevant FcRn expression pattern,²² Tg276 mice remain comparatively underutilized, and formal allometric scaling analyses for this model have not yet been reported. Overall, multiple studies suggest that both Tg32 and Tg276 mouse models could serve as useful translational tools. They also indicate that Fc-engineered or multispecific antibodies might require customized allometric exponents for accurate prediction of human PK, although a systematic study to confirm this is still lacking.

In this study, we present the most extensive dataset to date with in vivo PK parameters of antibodies (mainly IgG1, 2, and 4 types) in animals and human with and without half-life extension.

The main objectives of this study were to:

1. Compare the utility of hemizygous Tg276 mice versus homozygous Tg32 mice for predicting monkey and human PK parameters via allometric scaling for HLE and non-HLE molecules.

2. Compare the accuracy of human PK projections from FcRn transgenic mice with that from the cynomolgus monkey.

3. Determine model-specific allometric scaling exponents for HLE and non-HLE antibodies.

4. Evaluate an immunocompromised (SCID) version of the Tg276 model to mitigate anti-drug antibody (ADA) responses and compare PK profiles between SCID and non-SCID Tg276 hemizygous mice.

Results

Allometric scaling exponents

Optimized scaling exponents for CL, Q, and volume of distribution (Vc and Vp) are provided in Table 1. Exponents were optimized using smaller subsets, grouped per projection species (monkey or human), and stratified by non-HLE versus HLE molecules. It is essential to note that the number of HLE molecules in the dataset is significantly lower than that of non-HLE molecules, and the subsets contain relatively few cases (Figure 1). Therefore, the optimized coefficients are strongly dependent on these cases and may change in the future as additional data becomes available. Moreover, due to the still limited data, mono- and multispecific IgG1, 2, and 4 antibodies with different geometries, efficacy targets, and mutation types were pooled into a single experimental data set, and further sub-classification may be required to achieve optimal results.

Table 1.

Optimized allometric scaling exponents for systemic, inter-compartmental clearance and volume of distribution.

Half-life extension	Animal/strain	Systemic clearance, CL (MAE log)		Intercompartmental clearance, Q (MAE log)		Central volume of distribution, Vc (MAE log)		Peripheral volume of distribution, Vp (MAE log)
		Projection to monkey	Projection to human	Projection to monkey	Projection to human	Projection to monkey	Projection to human	Projection to monkey	Projection to human
No	Tg276 (hemi)	0.76 (0.53)	0.83 (0.51)	0.84 (1.31)	0.87 (1.73)	0.98 (0.50)	0.96 (0.47)	0.92 (1.07)	1.03 (0.89)
	Nº molecules	7	12	7	12	7	12	7	12
	Tg32 (homo)	0.95 (0.47)	0.86 (0.86)	0.68 (1.04)	0.70 (1.22)	0.99 (0.51)	0.97 (0.35)	0.83 (0.60)	0.90 (0.66)
	Nº molecules	23	24	23	23	23	24	23	23
	Monkey	–	0.84 (0.58)	–	0.70 (1.02)	–	1.05 (0.36)	–	0.99 (0.62)
	Nº molecules	–	83	–	73	–	83	–	73
Yes	Tg276_hemi	0.84 (0.69)	0.76 (0.41)	0.78 (0.94)	0.85	0.95 (0.49)	0.94 (0.51)	0.93 (0.44)	0.94
	Nº molecules	14	2	13	1	14	2	13	1
	Tg32_homo	0.84 (0.24)	0.72 (0.24)	0.67 (0.58)	0.62 (0.42)	0.92 (0.71)	0.93 (0.49)	0.85 (0.30)	0.91 (0.39)
	Nº molecules	14	11	13	11	14	11	13	11
	Monkey	–	0.63 (0.42)	–	0.74 (0.56)	–	1.00 (0.33)	–	0.99 (0.23)
	Nº molecules	–	13	–	13	–	13	–	13

Figure 1.

Overview of the experimental dataset. A. Antibody classes in the set. B. Mutation types of molecules in the set. C. Distribution of case molecules within the different FcRN humanized transgenic models.

Each exponent was optimized independently. Simultaneous optimization of exponents for central and peripheral volumes of distribution together with the exponents for clearance and intercompartmental clearance was not pursued in this study, as it would require a different estimation framework (e.g., fitting animal concentration – time profiles using a population PK modeling approach with a direct allometric scaling to monkey or human). For visual representations of the PK models and the allometric scaling strategy, see Figures 2 and 3, respectively. The optimized exponents (in particular for CL) deviated slightly from the commonly used value of 0.85. However, most estimates remained within the expected range (0.7–0.95) for projections both from transgenic mice and monkey to human. It is also evident that the analysis yielded different estimates of the exponents Q and CL. Consistent with the mean absolute error (MAE) log, optimization was less accurate for Q than for CL, likely reflecting limitations of the underlying data (e.g., digitized profiles, sparse sampling in the terminal phase, and heterogeneous fitting methods reported in the literature). As additional data becomes available, this optimization should be repeated, ideally splitting a data set into training and test sets.

Figure 2.

A structure of a 1- (A) and 2 (B) -compartment pk model selected to estimate pk parameters in animals and human.

Figure 3.

Allometric scaling strategy applied using 1- and 2-compartment PK parameters.

Optimized scaling exponents for Vc and Vp confirmed the common recommendation in the literature to apply an exponent of 1 and to compare optimized values between non-HLE and HLE molecules. As shown in Table 1, the optimized estimates were close to 1 (particularly for non-HLE molecules), with slightly lower values for HLE molecules.

Observed versus predicted CL of the investigated antibodies in monkey and human, using transgenic mouse models and optimized scaling exponents per HLE and non-HLE molecules, are plotted in Figures 4(A–D) and 5(A–D). Predicted versus observed Q and half-life from the transgenic models to monkey can be seen in Figures S1 and S2, respectively. Figures S3 and S4 show the predicted vs. observed Q and half-life from transgenic mice to humans, while Figures S5 and S6 show the same comparison from monkey to human. Figure 6 shows predicted versus observed clearance of non-HLE and HLE antibodies from monkey to human. Clearance was predicted using the optimized scaling exponents. As shown, the majority of predictions fall within a 2-fold predictive error, with a few clear outliers. The projected half-life depends on multiple parameters, such as predicted CL, Q, Vc, and Vp. In our analysis, predicted Q showed far more outliers than the other parameters, as the concentration points in the distribution and the terminal phase, which are required for robust estimation of Q, were limited in number. Uncertainty in Q propagates into half‑life estimates, as a higher MAE for Q is likely to contribute to the variability in half-life predictions. Inadequate or biased Q estimation may distort the relative contribution of distribution versus elimination phases, leading to over‑ or under‑prediction of terminal half‑life, especially when distribution is slow relative to elimination. As a consequence, half‑life predictions tend to exhibit a higher incidence of outliers and reduced robustness compared to clearance predictions. As larger, more diverse datasets become available, the robustness of Q estimation can be improved by increasing the information content in the early and intermediate time ranges and by pooling data across diverse molecules or studies.

Figure 4.

Observed versus predicted (from transgenic mice) monkey clearance with optimized scaling exponents: A. non-HLE molecules (projection from Tg32 homozygous mouse); B. non-HLE molecules (projection from Tg276 hemizygous mouse), C. HLE-molecules (projection from Tg32 homozygous mouse), D. HLE-molecules (projection from Tg276 hemizygous mouse).

Figure 5.

Observed versus predicted (from transgenic mice) human clearance with optimized scaling exponents: A. non-HLE molecules (projection from Tg32 homozygous mouse), B. non-HLE molecules (projection from Tg276 hemizygous mouse), C. HLE molecules (projection from Tg32 homozygous mouse), D. HLE molecules (projection from Tg276 hemizygous mouse).

Figure 6.

Observed versus predicted (from monkey) human clearance with optimized scaling exponents: A. non-HLE molecules, B. HLE molecules.

Using the compiled dataset, we confirmed that both Tg276 hemizygous and Tg32 homozygous mouse models yielded accurate allometric projections of antibody systemic clearance and terminal half-life in monkey and human for both non-HLE and HLE molecules. The Tg276 hemizygous mouse model performed at least as well as, and, in several instances, better than, the gold-standard Tg32 homozygous mice, supporting its potential as a robust alternative for allometric scaling of human PK parameters. To further illustrate the PK scaling potential of Tg276 hemizygous mice, Figure 7 displays the predicted and observed systemic concentration-time profiles in human and monkey following a single intravenous dose of 1 mg/kg of two proprietary molecules, NVS-2 (non-HLE molecule with available clinical data) and NVS-3 (preclinical HLE molecule). Both molecules were tested in Tg276 and Tg32 models, and mouse PK parameters were determined and projected to either human or monkey. In both cases, projections are accurate (predictions close to the actual PK profiles), with Tg276 hemizygous data indicating a slightly better predictive outcome.

Figure 7.

Predicted and observed intravenous (1 mg/kg) concentration-time profiles of two selected in-house molecules in human and monkey: A. NVS-2 (non-HLE): projection to human, B. NVS-3 (HLE): projection to monkey.

Our investigation confirmed that distinct scaling factors should be applied to HLE and non-HLE molecules. This is clearly illustrated in Figure 8, which compares the predicted versus observed clearance for HLE molecules in human and monkey, using different scaling exponents (optimized for non-HLE or HLE molecules). On average, applying non-HLE scaling exponents results in higher predicted antibody clearance of HLE molecules.

Figure 8.

Observed versus predicted clearance of half-life extended molecules in the dataset using different scaling exponents for: A. Monkey (projection from Tg32 homozygous mouse), B. Monkey (projection from Tg276 hemizygous mouse), C. Human (projection from Tg32 homozygous mouse), D. Human (projection from Tg276 hemizygous mouse), E. Human (projection from monkey), in black: actual optimized scaling exponent for HLE molecules, in gray: optimized scaling exponent optimized for non-HLE molecules; in green: standard scaling exponent of 0.85.

Although the experimental data set for HLE molecules is small, we also estimated the impact of different mutation types, split into three main categories: LS, YTE, and other, on the optimized allometric exponents by performing optimizations for each mutation subgroup. Table S1 in the supplementary material provides the results of this exercise. In short, we calculated up to 20% variability in the exponents for systemic and intercompartmental clearance and volume of distribution.

Comparison of PK data in Tg276 hemizygous non-SCID and SCID mice

Four Novartis proprietary molecules were selected for comparison of their concentration-time profiles in SCID and non-SCID Tg276 hemizygous mice. Figure 9 presents the individual concentration-time profiles for each molecule. Two of these molecules feature YTE-type half-life extension. Although the number of non-SCID individual profiles is greater, the SCID profiles display lower inter-animal variability and no ADA responses. Predictions of monkey and human PK parameters (i.e., clearance and half-life) are comparable between non-SCID and SCID data using identical scaling exponents (see Table 2). The projected clearances for both strains are nearly identical; however, terminal half-life estimates show differences between observations and projections, particularly for NVS-1 (HLE), NVS-2 (non-HLE), and NVS-4 (non-HLE). This discrepancy stems from low confidence in intercompartmental clearance fits driven by limited concentration-time data, strong biphasic profiles, and variability across non-SCID mice. Overall, the predicted PK remains broadly comparable to the one observed in monkey and human, falling within a 2-fold predictive interval.

Figure 9.

Individual concentration-time profiles of four selected internal molecules in Tg276 hemi scid and non-scid mice: A. NVS-1 (HLE), B. NVS-2 (non-HLE), C. NVS-3 (HLE), D. NVS-4 (non-HLE).

Table 2.

Predicted pk parameters from Tg276 hemi SCID and non-SCID mice to monkey and human using identical scaling exponents.

Molecule	Systemic clearance (L/day)			Half-life (days)
	From non-SCID mice	From SCID mice	Actual value	From non-SCID mice	From SCID mice	Actual value
Projection to monkey
NVS-1 (HLE)	0.005	0.005	0.006	28.31	27.10	26.50
NVS-2 (non-HLE)	0.008	0.008	0.011	22.10	20.00	17.72
NVS-3 (HLE)	0.0035	0.004	0.004	29.82	30.19	32.36
NVS-4 (non-HLE)	0.002	0.002	0.0034	9.13	10.02	6.77
Projection to human
NVS-1 (HLE)	0.040	0.036	0.061	67.62	65.84	57.18
NVS-2 (non-HLE)	0.18	0.14	0.19	38.30	34.41	26.17
NVS-3 (HLE)	0.0275	0.032	–	73.71	74.34	–
NVS-4 (non-HLE)	0.174	0.164	0.165	20.22	24.63	21.88

Figure 10 shows the predicted and observed half-life in human and monkey for the selected proprietary molecules: NVS-1 (HLE), NVS-2 (non-HLE), NVS-4 (non-HLE), and NVS-6 (non-HLE), using experimental data from Tg276 hemizygous non-SCID and Tg276 hemizygous SCID mice. Although there are some differences, the projections from non-SCID and SCID mice are, on average, comparable. The differences may be due to half-life being calculated from fitted 2-compartment PK parameters and to the fact that not all four PK parameters were accurately fitted due to limited concentration-time data.

Figure 10.

Predicted and observed half-life of NVS-1 (HLE), NVS-4 (non-HLE) and NVS-6 (non-HLE) internal molecules in human and monkey: A. NVS-1: projection to human, B. NVS-2: projection to human, C. NVS-4: projection to human, D. NVS-6: projection to human, E. NVS-1: projection to monkey, F. NVS-4: projection to monkey.

Discussion

This study demonstrates the translational utility of the Tg32 and Tg276 humanized FcRn transgenic mouse models for predicting PK parameters of therapeutic antibodies, including mono- and multispecific formats and molecules engineered for extended half-life, as depicted in Table 3. By recapitulating human-like IgG – FcRn interactions, these models enable robust simulation of antibody clearance and half-life in the human, thereby improving the reliability of preclinical-to-clinical PK extrapolation.¹⁹ Other mouse models have been developed that express human FCGRT via a knock-in strategy, recapitulating the endogenous mouse expression pattern (e.g.,²⁴). While such models may be useful for evaluating the PK of human and humanized antibodies, the scarcity of published data precluded their inclusion in the present analysis. As additional data become available, comparative evaluation of knock-in models versus Tg32 and Tg276 mice will be of interest to determine whether mouse-specific FcRn expression patterns influence observed PK behavior.

Table 3.

Overview of the models evaluated in the study.

Full Name	Short Name	JAX Strain Number	Genetics	Features	Formats Tested	Applications	Typical study duration*	Notes
B6.Cg-Fcgrttm1Dcr Tg(FCGRT)32Dcr/DcrJ	Tg32	14565	Mouse Fcgrt KO, Human FCGRT BAC Homozygous	Physiological expression of human FCGRT	Monoclonal antibodies (IgG1, IgG2, IgG4), Fc-engineered antibodies (half-life extended and silenced), Bispecific antibodies, Fc-fusion molecules	PK comparison of therapeutic leads; prediction of human PK; evaluation of FcRn antagonists	28–60 days, depending on the format.	Difficult to differentiate the terminal half-life of long-lived molecules; can mount an ADA response.
B6.Cg-Fcgrttm1Dcr Prkdcscid Tg(FCGRT)32Dcr/DcrJ	Tg32 SCID	18441	Mouse Fcgrt KO, Human FCGRT BAC Homozygous, Prkdc scid Mutation Homozygous	Physiological expression of human FCGRT and lack of endogenous antibodies.	Monoclonal antibodies (IgG1, IgG2, IgG4), Fc-engineered antibodies (half-life extended and silenced), Bispecific and multispecific antibodies	PK comparison of therapeutic leads; prediction of human PK; evaluation of FcRn antagonists	28–60 days, depending on the format.	Difficult to differentiate the terminal half-life of long-lived molecules; no ADA response.
B6.Cg-Fcgrttm1Dcr Tg(CAG-FCGRT)276Dcr/DcrJ	Tg276	4919	Mouse Fcgrt KO, Human FCGRT CAG transgene Hemizygous	Over-expression of human FCGRT, shorter serum half-life of human antibodies	WT monoclonal antibodies, Fc-engineered antibodies (half-life extended), Bispecific antibodies	PK comparison for lead selection and optimization, PK analysis of half-life extended molecules	14–28 days	Faster antibody clearance compared to the Tg32 models. Good differentiation of the terminal half-life of lead molecules; prone to mounting an ADA response
B6.Cg-Fcgrttm1Dcr Prkdcscid Tg(CAG-FCGRT)276Dcr/DcrJ	Tg276 SCID	21146	Mouse Fcgrt KO, Human FCGRT CAG transgene Hemizygous, Prkdc scid Mutation Homozygous	Over-expression of human FCGRT; shorter serum half-life of human antibodies; lack of antibody response	Monoclonal antibodies (IgG1, IgG2, IgG4), Fc-engineered antibodies (half-life extended and silenced), Bispecific antibodies, Fc-fusion molecules	PK comparison for lead selection and optimization, PK analysis of life-extended molecules	14.28 days	Faster antibody clearance compared to the Tg32 models.Good differentiation of the terminal half-life of lead molecules; no ADA response

no te: The typical study duration reflects a practical experimental design rather than a limitation of the models.

The main difference between the Tg32 and the Tg276 models lies in the constructs used to express human FCGRT. While the Tg32 model expresses the human gene under its endogenous promoter and closely mimics the native gene’s expression pattern, the FCGRT gene in the Tg276 model is expressed under a synthetic CAG promoter, resulting in high, widespread expression. Surprisingly, the human transgene in Tg276 behaves as a hypomorphic allele, and human antibodies tend to have shorter half-lives in Tg276 mice than in Tg32 mice, despite the transgene’s high expression. While the molecular mechanisms underlying this paradoxical behavior are unclear, Tg276 mice still exhibit a human-relevant recycling mechanism and, by eliminating antibodies faster, enable rapid evaluation of lead candidates.^20,25 This feature can be very useful in applications such as lead selection and optimization that require rapid data turnaround, or in evaluating test articles with a very long half-life. However, due to the perceived artificial nature of the model, the lack of published allometric scaling factors, and their tendency to mount ADA responses, Tg276 mice are rarely used by antibody developers. Our findings demonstrate that, despite the artificial expression of the human transgene, Tg276 hemizygous mice predict human PK parameters with an accuracy comparable to the more established Tg32 homozygous model and NHPs. Additionally, introducing the SCID mutation resolves the ADA issue by eliminating endogenous antibodies without affecting the model’s performance. These data therefore support the use of Tg276 mice in early-stage development, where rapid screening and lead optimization are critical for efficient resource allocation and shorter development timelines. Furthermore, accurate assessment of antibody half-life and distribution in these models can inform dosing strategies and early predictions of therapeutic efficacy, thereby reducing risk and guiding production planning, including decisions on batch size and formulation. The emergence of next-generation IgG antibodies with atypical PK profiles, such as rapid distribution followed by conventional elimination, underscores the value of humanized FcRn models for PK prediction. These profiles may reflect unique FcRn interactions or other biological mechanisms that influence efficacy and safety. Incorporating these models into the lead selection process enables early detection of aberrant PK behavior and provides mechanistic insights to guide the optimization of candidate molecules.

Our analysis confirms previous reports suggesting the need for distinct scaling exponents for non-half-life extended and half-life extended molecules.¹⁴ This distinction is mechanistically plausible because Fc engineering that increases FcRn binding and recycling alters systemic disposition and can shift the cross-species relationship between body size and clearance relative to non-engineered IgGs. We also recognize that half-life extension is not a single mechanism: different mutation types (e.g., YTE-, LS-, and other FcRn-affinity variants) and antibody architectures (mono- vs multispecific formats) may further modulate the appropriate exponents and contribute to the variability observed across datasets. In our study, however, the number of HLE molecules is limited. Compiling more studies in the future will be required to refine the values of the exponents. Additionally, new life-extension mutations have been recently described in multiple publications^26–28 and are making their way to the clinic. For future analyses, we suggest sub-grouping antibodies by geometry, class, or mutation type to help derive allometric coefficients with greater translational value.

Nevertheless, even with our limited datasets, the observed differences in scaling exponents confirm that specific allometric scaling factors are necessary to accurately predict human PK for non-HLE versus HLE molecules. The effect of half-life extension mutation type on the optimized allometric exponents was modest (≤20%) in this small, compiled dataset. However, mutation-specific scaling exponents may be needed, and this may become evident as larger datasets become available. The proposed exponents should therefore be considered preliminary until they can be confirmed in expanded datasets.

The Tg32 exponent values obtained, in particular for non-HLE molecules and projection to human, are in a similar range to those reported by other authors (Tg32, CL exponent: 0.73–0.91, Q exponent ~0.6, monkey: CL exponent: 0.75–0.85 for soluble targets and 0.9–0.96 for membrane-bound targets). For the first time, we calculated the scaling exponents for the Tg276 hemizygous mouse strain and supplemented the exponents analysis with projections from mouse to monkey, which may be useful in the early phases of molecule discovery and development. The exponents derived for the Tg276 hemizygous mouse are, on average, lower than those derived for the Tg32 homozygous mouse, but all range between 0.72–0.95 for systemic clearance, 0.62–0.87 for intercompartmental clearance, and 0.83–1.02 for volume of distribution. For scaling clearance from monkey, we obtained lower exponent values for HLE molecules compared to non-HLE molecules. Exponent values for intercompartmental clearance have lower confidence due to high uncertainty in fitting them to limited concentration-time profiles in animals, especially when using digitized data from the literature. Surprisingly, the calculated exponent values for the central and peripheral volumes of distribution were, in some cases, lower than 1 (HLE molecules, projections from Tg32 homozygous mice). Nevertheless, we still recommend using an exponent of 1, as consistently reported in the literature, for volume of distribution when projecting from any animal model, especially for non-HLE molecules. Because the experimental data set is relatively small, we did not split it into training and test sets; therefore, the optimized exponents may be specific to this data set. However, the resulting values, particularly for Tg32, are well aligned with the literature.

Several key limitations of the transgenic mouse models must be acknowledged. Neither the Tg32 nor the Tg276 mouse model fully recapitulates the complexity of the interaction between human antibody drugs and the immune system, and species-specific differences may affect the extrapolation of specific parameters. For example, immunogenicity and target-mediated drug disposition (TMDD) for non-cross-reactive targets cannot be reliably predicted using these models.^29,30 These constraints underscore the importance of integrating transgenic mouse data with complementary in vitro assays to build comprehensive PK profiles.

Despite these limitations, both Tg32 and Tg276 mouse models offer superior translational value compared to non-humanized rodent models, owing to their expression of human FcRn and performance comparable to that of NHPs, the current reference model for nonclinical PK studies. As such, Tg32 and Tg276 mice represent practical and ethically favorable alternatives to NHPs for PK screening.

Conclusions

Humanized FcRn transgenic mouse models, including Tg32 and Tg276, provide a robust platform for predicting human PK parameters and reducing reliance on NHP in vivo PK studies in preclinical development. Despite the availability of multiple publications, a systematic evaluation of their predictive performance has been lacking. Our analysis, based on both the literature and experimental data, confirms that both Tg32 and Tg276 mice yield reliable PK predictions for antibodies with linear PK in human and NHP, revealing that Tg276 hemizygous mice predict human PK with an accuracy comparable to that of Tg32 homozygous mice and NHPs. The shorter serum half-life observed in Tg276 mice, combined with the feasibility of allometric scaling, makes this model particularly suitable for rapid lead selection and optimization, as well as for evaluating long-lived antibody formats (see Table 3 for model reference). The introduction of the SCID mutation into the Tg276 strain successfully mitigates immunogenicity and reduces inter-animal variability. PK projections from SCID and non-SCID Tg276 mice are consistent, supporting the use of the immunodeficient variant for evaluating antibody drug candidates.

Our results clearly confirm the need for distinct scaling exponents for half‑life extended and non‑half‑life-extended molecules, as suggested by previous work.¹⁴ Although the dataset remains limited, preliminary estimates consistently fall within the expected range of 0.7 to 0.95. We acknowledge that different mutation types may affect these exponents differently. The exponents derived for HLE molecules should be viewed as working estimates rather than definitive values. Future studies with expanded and more uniformly curated datasets, richer early/intermediate sampling to better inform intercompartmental clearance, and external validation (e.g., training/test splits) will be needed to refine these exponents and improve the accuracy and robustness of PK predictions across diverse antibody formats and Fc-engineering strategies.

Materials and methods

Human FcRn transgenic mice

The B6.Cg-Fcgrt^tm1DcrTg(FCGRT)32Dcr/DcrJ (JAX stock #014565) mouse strain, also called Tg32, carries a knockout allele of the murine Fcgrt gene and expresses human FCGRT under its endogenous human promoter. The B6.Cg-Fcgrt^tm1Dcr Tg(CAG-FCGRT)276Dcr/DcrJ (JAX stock #004919) mouse strain, also known as Tg276, lacks the mouse Fcgrt gene and expresses its human counterpart under the control of the ubiquitous CAG (chicken β-actin) promoter. To generate the immunodeficient version of the Tg276 model, we crossed Tg276 mice to FcRn knockout SCID animals (B6.Cg-Fcgrt^tm1Dcr Prkdc^scid/DcrJ, JAX stock #018541), carrying the null Fcgrt allele and the scid mutation in the Prkdc gene. SCID mice cannot perform V(D)J recombination and lack functional antibodies. The resulting B6.Cg-Fcgrt^tm1Dcr Prkdc^scid Tg(CAG-FCGRT)276Dcr/DcrJ (JAX stock #021146) strain, also called Tg276 SCID, is therefore hemizygous for the human FCGRT transgene and has an impaired adaptive immune response.

Mouse standard housing conditions

The experiments were conducted at the Novartis facility in Basel, Switzerland, or at The Jackson Laboratory in Bar Harbor, ME. For the experiments run in Novartis, the animals were kept in Allentown XJ [Tecniplast GM500 or EM500 for DVC studies] individually ventilated cages (IVC) with supply air and exhaust air set to 50 ACH [DVC 75 ACH]. All cages were supplied with Avidity stainless steel automated watering. Cages contained spruce bedding (BK 8–15, J. Rettenmaier & Söhne GmbH + Co. KG, gradual transition to FS 14, J. Rettenmaier & Söhne GmbH + Co. KG until the end of September 2025), both a popsicle stick and a wood block (Pura Aspen Chew Sticks and Pura Aspen Chew Blocks, Labodia) for gnawing, a minimum of 11 g crinkle paper for nesting (pressed paper strips, J. Rettenmaier & Söhne GmbH + Co. KG), a red transparent mouse house (polycarbonate, Tecniplast) for shelter and a transparent handling tunnel (polycarbonate, Zoonlab). Mice were fed ad libitum (mouse and rat maintenance 3892, irradiated, KLIBA NAFAG [if applicable: mouse and rat breeding 3302, irradiated, KLIBA NAFAG) and had unrestricted access to drinking water (reverse osmosis, hyper-chlorinated, 2–3 ppm). The light/dark cycle in the room consisted of 12/12 h with artificial light (on: 6 am, off: 6 pm). The room temperature was set to 22°C (+/− 2K), with a relative humidity of 55% (variable between 45 and 65%). A complete cage change was conducted every other week, with spot changes performed as necessary. During the cage change, nesting material, used but unsoiled, was transferred from the dirty cage to the new, clean cage. Animals were tunnel handled for all procedures. All procedures performed on immunocompromised mice (SCID mutation) were conducted under a biological safety cabinet (Class II). For the experiments performed at The Jackson Laboratory, all animal studies were performed in accordance with the OLAW guidelines and were reviewed and approved by the Institutional Animal Care and Use Committee at The Jackson Laboratory. All mice were maintained under standard conditions with a 12/12 h (on: 6 am, off: 6 pm) light – dark cycle. Animals were housed individually and in positively ventilated polysulfonate cages with HEPA-filtered air at a density of 3–4 mice per cage. Filtered tap water, acidified to a pH of 2.5 to 3.0, and normal rodent chow were provided ad libitum. Animals were weekly checked for welfare and mice presenting more than 15% of maximum body weight loss were considered at humane endpoint and euthanized.

In vivo studies

Male and female Tg276 mice (B6.Cg-Fcgrt^tm1Dcr TG(CAGFCGRT)276Dcr/DcrJ; homozygous/hemizygous) and Tg276 SCID mice (B6.Cg-Fcgrt^tm1Dcr Prkdc^scid Tg(CAGFCGRT)276Dcr/DcrJ; homozygous/hemizygous) were obtained from The Jackson Laboratory (JAX stock #004919 and #021146, respectively).

In experiments conducted at the Novartis facility, 7–8-week-old mice underwent baseline blood sampling. Subsequently, 4–6 days later, each animal received a single intravenous dose of the test compound (5 mL/kg via the lateral tail vein). Blood samples were collected at defined intervals from Day 0 to Day 59 post-dose. Sampling was performed from the vena saphena in conscious mice using CB300Z microvettes (Sarstedt). Samples were allowed to coagulate at room temperature for at least 20 minutes, followed by centrifugation at 2000 × g for 10 minutes at room temperature. The resulting serum was transferred to LoBind Eppendorf tubes and stored at − 80°C until further analysis.

Standard study duration of the transgenic mouse experiments is around 30 days for non-HLE-molecules and up to 60 days for HLE-molecules.

For the studies performed at The Jackson Laboratory, mice were intravenously injected with a single dose of the test compounds (10 mg/kg, in a dose volume of 5 mL/kg). The mice were then bled serially (25 µl samples) at defined intervals from Day 0 to Day 28 post-dose. The blood was collected into K₃EDTA, processed to plasma, diluted 1:10 in 50% glycerol in phosphate-buffered saline, and stored at −20°C until analysis.

Bioanalytical method

At Novareis, a generic ligand-binding assay (LBA) employing Electrochemiluminescence Immunoassay (ECLIA) technology for the quantification of human monoclonal antibodies (mAbs) in standards (STDs), quality controls (QCs), and unknown mouse serum samples was developed. The assay uses a biotinylated goat polyclonal antibody against human IgG as the capture reagent (Southern Biotech, #2049–08), immobilized on streptavidin-coated surfaces (MSD GOLD 96-well Streptavidin plates, Mesoscale Discovery #L15SA-1) to selectively bind human IgG in the samples. A Sulfo-tag-labeled goat polyclonal anti-human IgG antibody serves as the detection reagent (Southern Biotech, #2049–01 in house labeled with sulfo-tag), generating an electrochemiluminescent signal proportional to the concentration of mAbs in the sample. Calibration standards establish a standard curve for quantification, while QCs ensure assay accuracy and precision.

All steps, including sample incubation, washes, and signal generation, are optimized for sensitivity, linearity, and reproducibility. This robust method is designed to reliably quantify human monoclonal antibodies in complex biological matrices for applications in therapeutic monitoring and PK analysis.

At the Jackson Laboratory, plasma samples were assessed via Mabtech (3850–1AD-6) hIgG Fc ELISA to quantify test article concentrations. Total hIgG The Mabtech (3850–1AD-6) hIgG Fc ELISA was validated for sensitivity and dynamic range for all test articles. PK analysis of test articles was determined using PK Solutions Software (Summit Research Services). All statistical analyses of half-life determinations were evaluated using GraphPad Prism software (Version 10.2.3) with ANOVA and Tukey’s multiple-comparison test.

Experimental dataset

The data (134 unique molecules) for this study were compiled from internal (49 molecules) and literature (85 molecules) sources. The set consists of 115 monoclonal antibodies, 9 multi-specific antibodies, and 10 other modalities (engineered proteins). All molecules exhibit linear PK in animals and human (with some concluded with linear PK in animals only or in human at higher, clinically relevant doses), and 34 of 134 molecules have a half-life extension via neonatal Fc receptor (FcRn) modifications (“YTE,” “LS,” and other mutation types). For molecules with known TMDD behavior, only parameters derived from dose ranges reported to be linear were included, and this assumption applies consistently across mouse, monkey, and human projections where possible. Figure 1 shows a graphical overview of the collected data. In vivo PK of each molecule in transgenic mice (Tg32 homozygous, Tg276 hemizygous non-SCID and SCID), monkey and human was described fitting a 1- or 2-compartment PK model (structures in Figure 2) to available individual concentration-time profiles resulting in estimates of systemic clearance (CL), volume of distribution to the central compartment (Vc), volume of distribution to the peripheral compartment (Vp) and intercompartmental clearance (Q). The last two parameters refer only to the 2-compartment model fitting. Human PK parameters are clinical population PK estimates, either published or estimated for in-house data, with the majority of data referring to healthy subjects (typically from clinical Phase 1 studies). However, in some instances, only patient data was available. In most studies, data from intravenous injections were used to derive PK parameters for a healthy population representative (with a standard body weight of around 70 kg, although body weight varies across different clinical trials). Similarly, animal PK parameters refer to a typical mouse (with a body weight of around 25 g) and a cynomolgus monkey (with a body weight of around 3–4 kg). When data were available only for subcutaneous injections, bioavailability was assumed to estimate absolute values of PK parameters. All anti-drug antibody (ADA) positive concentrations for in-house molecules were removed prior to data fitting to prevent them from affecting the estimated PK parameters. No information regarding ADA responses was available for the literature molecules; therefore, the entire published PK profiles were considered, which may have influenced the estimation of scaling exponents.

Table S1 in the supplementary material provides the entire PK dataset, including internal and literature data (with references), along with information on antibody class, type, and half-life extension. Published animal and human concentration-time profiles in the figures were digitized using Digit 1.0.4* software to estimate PK parameters when these parameters were not available, a procedure that may have affected the quality of the data. Monolix* (version 2023 R1) software was used to fit individual (in-house) or mean (digitized from the literature) concentration-time data to estimate PK parameters for all molecules in animals and human.

*https://www.simulations-plus.com

Terminal half-life was calculated using the following equations³¹:

$$1-\mathit{compartment}\mathit{PK}\mathit{model}{T}_{1/2}=\mathit{log}\left(2\right)\ast \left(\mathit{Vss}/\mathit{CL}\right)$$ (1)

2-compartment PK model

$$T_{\frac{1}{2}}=\frac{0.693}{\frac{1}{2}\cdot \left[\left(k_{12}+k_{21}+k_{10}\right)-\sqrt{{\left({\mathit{k}}_{12}+k_{21}+k_{10}\right)}^2-4\cdot k_{21}\cdot k_{10}}\right]}$$ (2)

Where: K12 = Q/Vc, K21 = Q/Vp and K10 = CL/Vc

Individual concentration-time profiles of four in-house molecules were derived from SCID and non-SCID Tg276 hemizygous mice and evaluated for potential immunogenicity, inter-animal variability, and predictive performance using identical scaling exponents as optimized for non-SCID data. The animals were administered single intravenous doses of 10 mg/kg for each compound.

Modeling strategy

Figure 3 illustrates the modeling strategy used, which is based on single-species allometric scaling. Allometric scaling is the projection of PK parameters, such as CL, Q, Vc, and Vp, from one species to another by multiplying each parameter by a body-weight ratio between the two species, corrected by an appropriate exponent. We applied the scaling from mouse to monkey, mouse to human, and from monkey to human. The exponents of CL, Q, Vc and Vp were optimized (although the optimized exponents for volume of distribution served to show that the assumption of 1 is correct and compare results for non- and half-life extended molecules). The analysis was performed separately for half-life extended and non-half-life extended molecules; however, all half-life extension types were pooled into a single dataset prior to optimization, and no differentiation between mutation types was applied. This procedure might have affected the scaling exponents, but the limited dataset would not allow evaluation of each mutation type separately.

The exponents were optimized in this study using the optim function available in R* (package: “stats”), a free software environment for statistical computing and graphics (version 4.4.0). The function is a general-purpose optimization with the selected “Brent” algorithm. This algorithm is efficient for 1D optimization, does not require gradient information, and is robust and reliable for this type of correction.

An objective function was defined to minimize the sum of squared errors (SSE) between the log-transformed predicted and observed clearance values, defined as follows:

$$\mathit{Functio}{n}_{\mathit{obj}}=\mathit{f}\left(\mathit{x},\mathit{k},\mathit{B}{W}_x,\mathit{B}{W}_y,\right)\left\{\mathit{sum}\left({\left(\log (p\mathrm{red}-\log (obs\right)}^2\right)\right\}$$ (2)

Where: x refers to a PK parameter (CL, Q, Vc or Vp) to be scaled, pred and obs to predicted (via x*(BW_x/BW_y)^k) and observed PK parameter, respectively; BW stands for a body weight of species ×and y; and k (α or β) is the exponent to be optimized.

The optimal value (k) was used to calculate predicted values, and the MAE on the log scale was used to assess the model’s validity.

$$\mathit{Mean}\mathit{absolute}log\mathit{error}=\frac{1}{n}\sum _{\mathit{i}=1}^n\left|log\left(\mathit{pre}{d}_i\right)-\log \left(\mathit{ob}{s}_i\right)\right|$$ (3)

where n is the number of molecules

*https://www.r-project.org/

We used the mean absolute error on the logarithmic scale because absolute deviations are less dominated by extreme observations than squared errors, thereby enhancing robustness to outliers. Additionally, the log-scale assessment is appropriate for strictly positive quantities, whose residuals are closer to normality, and for which multiplicative (fold-change) discrepancies are biologically salient. This metric thus yields a robust and biologically interpretable summary of predictive accuracy.

R software was used to create figures using base R packages.

Funding Statement

The author(s) reported that all work was internally funded by The Jackson Laboratory and Novartis AG.

Disclosure statement

The Novartis authors, as indicated by their affiliation, are Novartis employees and own Novartis stocks.

Abbreviations

PK

pharmacokinetics

HLE

Half-life extension

ADA

Anti-Drug Antibody

CL

Clearance

FcRn

Neonatal Fc Receptor

IgG

Immunoglobulin G

IVIG

Intravenous Immunoglobulin

LS

dual-amino acid substitution in the Fc region of IgG antibodies (M428L/N434S]

MAE

Mean Absolute Error

mAbs

Monoclonal Antibodies

NHP

Non-Human Primate

Q

Intercompartmental Clearance

SCID

Severe Combined Immunodeficiency

YTE

triple amino acid substitution in the Fc region of human IgG1 antibodies (M252Y/S254T/T256E)

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2026.2668888