Estimation of Clearance and Bioavailability of Therapeutic Monoclonal Antibodies from Only Subcutaneous Injection Data in Humans Based on Comprehensive Analysis of Clinical Data.

INTRODUCTION: Theoretically, the separate estimation of clearance (CL) and bioavailability (F) requires both intravenous and extravascular injection data. This study investigated whether CL and subcutaneous F of therapeutic monoclonal antibodies (mAbs) in humans can be separately estimated from subcutaneous injection data only.
METHODS: First, the geometric mean of linear pharmacokinetic parameters (CL, intercompartmental CL [Q], volume of distribution in the central compartment [Vc], and volume of distribution in the peripheral compartment [Vp]) after intravenous injection for mAbs in humans that have been reported in public data sources was estimated from 103 mAbs with linear pharmacokinetics and 44 mAbs with nonlinear pharmacokinetics. Next, we estimated the CL and F of 25 mAbs with linear pharmacokinetics from plasma/serum mAb concentration-time profiles after subcutaneous injection in humans by fixing the geometric mean of Q, Vc, and Vp based on the public data. Moreover, the plasma/serum concentration-time profile of 25 mAbs after intravenous injection was simulated using the estimated CL and the geometric mean of Q, Vc, and Vp.
RESULTS: There were no significant differences in parameters among subclasses (immunoglobulin [Ig] G1, 2, and 4) or in linearity (derivation from linear and nonlinear pharmacokinetics). Using only subcutaneous injection data, we successfully estimated the CL of 23/25 mAbs (92%) and F of all 25 mAbs (100%) within 1.5-fold of the observed value. Moreover, overall, the simulated concentration-time profiles were largely consistent with observed data (90.8% within 1.5-fold of the observed values).
CONCLUSIONS: This approach does not require intravenous injection data to separately estimate CL and F after subcutaneous injection in humans and can therefore accelerate the clinical development of mAbs.

Key Points

Introduction

Therapeutic monoclonal antibodies (mAbs) have dramatically changed the treatment of numerous diseases, including cancers, autoimmune diseases, and infections [1, 2]. Since mAbs have large molecular weight, oral formulations are ineffective and they must therefore be injected intravenously, subcutaneously, or intramuscularly [3]. Recently, subcutaneous injection has become more common in clinics because of its convenience [4].

In subcutaneous injection, bioavailability (F) is important for determining the absorption properties of drugs and the required dosage. F is estimated from the ratio of area under the plasma drug concentration–time curve (AUC) after extravascular (oral, subcutaneous, intramuscular, etc.) delivery and intravenous injection. Thus, it is essential to know the AUC after intravenous injection to estimate F, even if an intravenous formulation is not intended as a therapeutic option. Furthermore, clearance (CL) cannot be estimated from subcutaneous injection data alone, but requires comparison of the dose and AUC after intravenous injection. If exposure after extravascular injection is lower than expected in humans, we cannot quantitatively judge which process caused it—absorption or elimination—without intravenous injection data. In fact, several mAbs have been evaluated using intravenous injection in clinical trials even though it would not be used as a therapeutic delivery method [5, 6]. If CL and F could be accurately estimated using only subcutaneous injection data, intravenous evaluations could be skipped in clinical trials.

The linear pharmacokinetic profile of mAbs is reported to be captured well by the two-compartment model in animals and humans [7, 8]. In two-compartment model parameters (CL, intercompartmental CL [Q], volume of distribution in the central compartment [Vc], and volume of distribution in the peripheral compartment [Vp]), CL is determined by the elimination process and Q, Vc, and Vp reflect distribution. Due to its large molecular weight and hydrophilic property, the tissue distribution of mAbs is limited to vascular and interstitial spaces [9]. Therefore, mAbs have been reported to show similar volume of distribution at steady state (Vdss) in cynomolgus monkeys and humans [10]. Moreover, if distribution is similar among mAbs in humans, Q, Vc, and Vp can be fixed according to a typical value. Furthermore, if Q, Vc, and Vp are fixed, CL can be estimated from the slope of the elimination phase (β phase) after subcutaneous injection in linear pharmacokinetics. Since AUC after subcutaneous injection is determined by dose, CL, and F, and is not affected by Q, Vc, and Vp, if CL is estimated from the slope of the elimination phase, then F also can be theoretically estimated from dose and AUC after subcutaneous injection. Thus, CL and F can only be theoretically estimated from the plasma/serum mAb concentration–time profiles after subcutaneous injection.

In this study, to determine the typical values for each pharmacokinetic parameter (CL, Q, Vc, and Vp), a large dataset (103 mAbs with linear pharmacokinetics and 44 mAbs with nonlinear pharmacokinetics) was first constructed from public data and then analysed. We used this large dataset to investigate the effect of IgG subclasses (IgG1, IgG2, and IgG4) and linearity (linear and nonlinear) on linear pharmacokinetic parameters (CL, Q, Vc, and Vp) in humans. We then investigated whether subcutaneous injection data alone were sufficient to separately estimate the CL and F of mAbs in humans by fixing Q, Vc, and Vp as a geometric mean of 147 mAbs.

Materials and Methods

Data Collection

To construct the dataset for analysis, pharmacokinetic data on mAbs in humans was obtained from literature, patents, presentations at scientific conferences, or information provided by the Pharmaceutical and Medical Devices Agency (PMDA), US Food and Drug Administration (FDA) and European Medicines Agency (EMA). When body weight information was unavailable, a body weight of 75 kg was applied. Pharmacokinetic data on mAb concentrations in plasma or serum are generally available. In this study, pharmacokinetic data on mAb concentrations in serum were assumed to be the same as that in plasma. The average values of pharmacokinetic parameters and plasma/serum mAb concentration–time profiles in humans were collected from published data. The plasma/serum mAb concentration–time profiles in humans after a single intravenous/subcutaneous injection were selected for analysis. These profiles were obtained by scanning figures from data sources using UnGraph 5 (Biosoft, Cambridge, UK). The mAbs with linear pharmacokinetics were categorised in Group A and mAbs with nonlinear pharmacokinetics were categorised in Group B. In Group B, the reported two-compartment model parameters (CL, Q, Vc, and Vp) were only used if they were available in published data (as estimated by the Michaelis–Menten [MM] model [11] or the target-mediated drug disposition [TMDD] model [12]). The geometric mean of collected parameters for the subclasses, as well as for linear and nonlinear mAbs, were estimated and compared. All mAbs used in this study had a human IgG sequence as a constant region.

Estimation of Two-Compartment Model Parameters in Group A

Group A consisted of 103 mAbs that showed linear pharmacokinetics in humans. Reported values for two-compartment model parameters were used if available in published data. If unavailable, plasma/serum mAb concentration–time profiles were analysed using the traditional two-compartment model with first-order elimination (electronic supplementary Methods) to estimate CL, Q, Vc, and Vp.

Estimation of Two-Compartment Model Parameters in Group B

Group B consisted of 44 mAbs that showed nonlinear pharmacokinetics in humans. The accurate estimation of linear pharmacokinetic parameters from nonlinear pharmacokinetics is more complex and requires a reliable dataset. Therefore, as mentioned earlier, only two-compartment model parameters that were estimated by the MM or TMDD model were collected from published data.

Estimation of Clearance and Bioavailability After Subcutaneous Injection in Humans

To estimate CL and F, the plasma/serum mAb concentration–time profiles of 25 mAbs in humans after subcutaneous injections were fitted by fixing Q, Vc, and Vp with the geometric mean of 147 mAbs (103 mAbs in Group A and 44 mAbs in Group B). These 25 mAbs were selected because of their linear pharmacokinetics and the availability of plasma/serum mAb concentration–time profiles after both intravenous and subcutaneous injection. Twenty of the 25 mAbs were included in Group A. The estimated CL and F were compared with observed values. The plasma/serum mAb concentration–time profiles after subcutaneous injection were fitted to the traditional two-compartment model with first-order absorption and elimination (electronic supplementary Methods).

Prediction of Plasma/Serum Monoclonal Antibody (mAb) Concentration–Time Profiles After Intravenous Injection in Humans

Using the estimated CL and geometric mean of Q, Vc, and Vp, the plasma/serum mAb concentration–time profiles of 25 mAbs after intravenous injection in humans were then simulated. The dose used in the simulation was the same as that used in the clinical study. Simulated profiles were compared with observed values.

Analysis

All fittings and simulations were performed using SAAMII software (The Epsilon Group, Charlottesville, VA, USA). The Rosenbrock method was used as an integrator and all fittings were performed using the default setting in SAAMII. Relative weight (1/y^2) was used in all fittings. All figures and statistical analyses were prepared using GraphPad Prism 7 (GraphPad Software, San Diego, CA, USA). Since it has been reported that pharmacokinetic parameters show log-normal distribution [13, 14], this was assumed for each parameter in this study.

Results

Estimation of the Geometric Mean of Linear Pharmacokinetic Parameters of mAbs in Humans

A total of 147 mAbs (103 mAbs with linear pharmacokinetics [Group A] and 44 mAbs with nonlinear pharmacokinetics [Group B]) were selected as a dataset to estimate the geometric mean of each pharmacokinetic parameter. Collected linear pharmacokinetic parameters in Groups A and B are summarised in electronic supplementary Tables 1 and 2. The estimated geometric mean of CL, Q, Vc, and Vp for all 147 mAbs, divided into Groups A and B, are summarised in Table 1. In both Groups A and B, the distribution of each pharmacokinetic parameter was similar (Fig. 1), indicating that nonlinear pharmacokinetics does not affect linear two-compartment model parameters estimated by MM or TMDD model analysis. There were 101 IgG1 (74 in Group A and 27 in Group B), 24 IgG2 (14 in Group A and 10 in Group B), and 22 IgG4 (15 in Group A and 7 in Group B) subclasses in the dataset. The estimated geometric mean of CL, Q, Vc, and Vp for mAbs in IgG1, IgG2, and IgG4 are summarised in Table 1. Each pharmacokinetic parameter was similarly distributed in the IgG subclasses (Fig. 2), indicating that the IgG subclass does not affect linear two-compartment model parameters in humans. Thus, all datasets (IgG subclasses and linearity) were combined to estimate the geometric mean of pharmacokinetic parameters of mAbs in humans.

Table 1

Geometric mean of CL, Q, Vc, and Vp of mAbs in humans

Dataset	No. of mAbs	CL	Q	Vc	Vp
		mL/day/kg	mL/day/kg	mL/kg	mL/kg
Total	147	3.32	8.77	44.2	36.7
		(1.11–12.6)	(1.26–86.3)	(21.3–86.7)	(15.6–113)
Group A	103	3.34	8.62	43.5	37.6
		(1.27–12.6)	(1.26–86.3)	(21.4–71.6)	(15.6–113)
Group B	44	3.26	9.11	45.7	34.8
		(1.11–10.1)	(1.76–44.6)	(21.3–86.7)	(16.3–80.0)
IgG1	101	3.55	8.96	44.2	37.0
		(1.17–12.6)	(1.26–86.3)	(21.4–71.6)	(15.6–113)
IgG2	24	2.87	8.45	42.7	36.3
		(1.11–10.1)	(3.70–44.6)	(21.3–58.8)	(16.8–76.3)
IgG4	22	2.85	8.23	45.7	36.1
		(1.41–5.52)	(2.35–24.3)	(26.7–86.7)	(16.3–80.0)

Data in parentheses indicate the range of the observed values

CL clearance, Q intercompartmental clearance, V volume of distribution in the central compartment, V volume of distribution in the peripheral compartment, mAbs monoclonal antibodies, IgG immunoglobulin G

Fig. 1

Effect of linearity on the distribution of a CL, b Q, c Vc, and d Vp. Geometric mean with 95% confidence interval is shown. Closed circles indicate the linear pharmacokinetic parameters derived from mAbs with linear pharmacokinetics, and closed squares indicate the linear pharmacokinetic parameters derived from mAbs with nonlinear pharmacokinetics based on the MM or TMDD models. mAbs monoclonal antibodies, MM Michaelis–Menten, TMDD target-mediated drug disposition, CL clearance, Q intercompartmental clearance, V volume of distribution in the central compartment, V volume of distribution in the peripheral compartment

Fig. 2

Effect of IgG subclasses on the distribution of a CL, b Q, c Vc, and d Vp. Geometric mean with 95% confidence interval is shown. Closed circles, squares and triangles indicate the linear pharmacokinetic parameters of the IgG1, 2, and 4 subclasses. IgG immunoglobulin G, CL clearance, Q intercompartmental clearance, V volume of distribution in the central compartment, V volume of distribution in the peripheral compartment

Estimation of Clearance and Subcutaneous Bioavailability of mAbs in Humans

A total of 25 mAbs were selected because of their linear pharmacokinetics and the availability of plasma/serum mAb concentration–time profiles after both intravenous and subcutaneous injection as a dataset to estimate CL and F. They consisted of 16 IgG1, 4 IgG2, and 5 IgG4 subclasses. Plasma/serum mAb concentration–time profiles after subcutaneous injection were fitted by fixing Q (8.77 mL/day/kg), Vc (44.2 mL/kg), and Vp (36.7 mL/kg) with geometric mean, and CL and F were estimated. Observed and estimated CL and F for the 25 mAbs are summarised in Table 2. The estimated values were plotted with the observed values (Fig. 3a, b). As shown in Fig. 3a, CL of 23/25 mAbs (92 %) was successfully estimated within 1.3-fold of the observed values. CL of all mAbs (100%) was estimated within 2-fold of the observed values. As shown in Fig. 3b, F of 21/25 mAbs (84%) and 25/25 mAbs (100%) was successfully estimated within 1.3- and 1.5-fold of the observed values, respectively. Although the largest IgG subclass for these 25 mAbs is IgG1, there was no apparent difference in estimation accuracy among the three IgG subclasses.

Table 2

Observed and estimated CL and F of 25 mAbs in humans

Antibody	Target	Subclass	Observed		Estimated		Reference
			CL	F	CL	F
			mL/day/kg	%	mL/day/kg	%
Adalimumab	TNFα	IgG1	3.43	52.0	4.22	67.5	[31]
ANB020	IL-33	IgG1	7.20	29.0	6.18	24.8	[32]
Belimumab	BLyS	IgG1	3.02	77.5	2.43	93.5	[33]
Burosumab/KRN23	FGF23	IgG1	3.43	128	3.19	97.1	[34]
Canakinumab	IL-1β	IgG1	2.80	66.5	2.72	88.2	[35]
CNTO5825	IL-13	IgG1	2.31	75.0	2.95	84.6	[36]
Eptinezumab/ALD403	CGRP	IgG1	2.25	71.7	2.72	88.9	[37]
Golimumab	TNFα	IgG1	5.18	51.1	5.62	44.7	[38]
Guselkumab/CNTO1959	IL-23	IgG1	4.61	60.1	4.27	50.7	[5]
MEDI-528	IL-9	IgG1	1.96	79.0	2.14	85.7	[39]
Mepolizumab	IL-5	IgG1	2.92	74.0	3.42	104	[40]
NNC0114-0005	IL-21	IgG1	3.15	60.0	4.03	43.2	[41]
Risankizumab/ABBV-066	IL-23	IgG1	3.80	62.7	2.49	66.1	[42]
Secukinumab	IL-17A	IgG1	2.09	55.0	3.33	68.6	[43]
Sirukumab/CNTO136	IL-6	IgG1	5.11	92.4	4.39	77.0	[44]
Tildrakizumab/MK-3222	IL-23	IgG1	2.20	73.0	2.47	94.0	[45]
AMG403	NGF	IgG2	3.39	73.6	3.30	92.8	[46]
Gevokizumab	IL-1β	IgG2	2.46	69.0	1.90	75.2	[47]
RN317/PF-05335810	PCSK9	IgG2	4.43	61.9	5.37	59.0	[48]
Tezepelumab/AMG157/MEDI9929	TSLP	IgG2	2.52	81.3	2.73	96.4	[49]
ANB019	IL-36R	IgG4	2.96	95.0	2.70	87.7	[50]
BITS7201A	IL-13/17	IgG4	2.43	109	3.00	95.1	[51]
Emicizumab/ACE910	FIXa/FX	IgG4	3.02	85.5	2.58	71.1	[52]
Tralokinumab/CAT-354	IL-13	IgG4	2.93	80.0	3.02	95.7	[53]
Olokizumab	IL-6	IgG4	1.99	84.2	1.88	77.8	[54]

Twenty-five mAbs were selected due to their linear pharmacokinetics and the availability of plasma/serum mAb concentration–time profiles after intravenous and subcutaneous injection

CL clearance, F bioavailability, TNF tumour necrosis factor, IL interleukin, FGF fibroblast growth factor, NGF nerve growth factor, PCSK9 proprotein convertase subtilisin kexin 9, IgG immunoglobulin G

Fig. 3

Estimation of a CL and b F after subcutaneous injection of 25 mAbs in humans. Open circles indicate the observed and estimated CL of 25 mAbs; open squares indicate the observed and estimated F of 25 mAbs; solid line indicates the 100% estimation; and dashed lines and dotted lines indicate the 1.3- and 1.5-fold estimations, respectively. CL clearance, F bioavailability, mAbs monoclonal antibodies

Prediction of Plasma/Serum mAb Concentration–Time Profiles After Intravenous Injection in Humans

Using estimated CL and geometric mean of Q, Vc, and Vp, we simulated the plasma/serum mAb concentration time-profiles after intravenous injection for 25 mAbs. As shown in electronic supplementary Fig. 1, the predicted plasma/serum mAb concentration–time profiles were mostly consistent with the observed values. The relationship between the observed and predicted plasma/serum mAb concentrations is shown in Fig. 4. As a result, 90.8% and 99.7% of time points were accurately predicted within 1.5- and 2-fold of the observed values, respectively. Only one time point (final time point for Risankizumab/ABBV-066) was predicted to be over 2-fold of the observed value.

Fig. 4

Scatter plot of predicted and observed plasma/serum mAb concentrations. Open diamonds indicate the observed and predicted plasma/serum mAb concentrations, and solid line indicates the 100% prediction line. mAb monoclonal antibody

Discussion

In this study, we investigated whether the CL and F of mAbs can be separately estimated using only subcutaneous injection data in humans. Analysis was conducted using a large and comprehensive dataset constructed from a total of 147 mAbs. To be best of our knowledge, this is the largest dataset of pharmacokinetic parameters for mAbs ever reported. It enabled us to investigate the effect of two factors—IgG subclasses and pharmacokinetic linearity—on the linear pharmacokinetic parameters of mAbs in humans.

First, we investigated the effect of IgG subclasses. Previously, Walker et al., examined the effect of IgG subclasses of mAbs on pharmacokinetics in rats and cynomolgus monkeys [15]. They compared IgG1 and IgG2 using four mAbs with different amino acid sequences in the variable region and concluded that the difference in IgG subclass had no significant impact on pharmacokinetics. Furthermore, Tabrizi et al. reported no significant impact of IgG subclass (IgG1 vs. IgG4) on the pharmacokinetics of mAbs in mice, dogs, and cynomolgus monkeys [16]. We and other groups have reported that the pharmacokinetics of mAbs in humans can be accurately predicted using the allometric scaling approach with cynomolgus monkeys [17-19]. Thus, the impact of IgG subclasses should also be minimal in humans; however, evidence in humans has never been demonstrated. To appropriately investigate their effect on pharmacokinetics in humans, we would need to directly compare the pharmacokinetics of IgG subclasses using mAbs with the same amino acid sequence in the variable region. However, since this would be difficult in humans for economic and ethical reasons, it is essential to conduct a comprehensive analysis using a large dataset. In this study, we first demonstrated the impact of IgG subclasses on the pharmacokinetics of mAbs in humans using a large dataset. Our results indicate that IgG subclasses have no significant impact on the linear pharmacokinetic parameters of mAbs in humans.

The second factor we investigated was the effect of linearity on linear pharmacokinetic parameters in humans. This is the first report to examine the effect of linearity on the linear pharmacokinetic parameters of mAbs. Generally, mAbs are classified into two types—those with linear pharmacokinetics and those with nonlinear pharmacokinetics. In most cases, the nonlinear pharmacokinetics of mAbs is caused by the plasma/serum mAb concentration-dependent saturation of target-mediated elimination. To quantitatively and separately estimate both linear and nonlinear pharmacokinetic parameters using nonlinear pharmacokinetic data, plasma/serum mAb concentration–time profiles have been analysed using the MM or TMDD model. In this study, the MM or TMDD model-derived linear pharmacokinetic parameters of mAbs with nonlinear pharmacokinetics were collected from published data and compared with those for mAbs with linear pharmacokinetics. As a result, we found that linearity had no impact on the linear pharmacokinetic parameters. Thus, although we estimated CL and F of mAbs with linear pharmacokinetics using only subcutaneous injection data in humans in this study, this approach could also be used to estimate CL and F of mAbs with nonlinear pharmacokinetics. We will further investigate this possibility in the future using a large dataset of mAbs with nonlinear pharmacokinetics in humans. In this study, since we demonstrated that IgG subclasses and linearity had no significant effect on the linear pharmacokinetic parameters using a large constructed dataset, we combined the data for all 147 mAbs to estimate the geometric mean of each pharmacokinetic parameter.

The estimated geometric means of CL, Q, Vc, and Vp of 147 mAbs in humans were 3.32 mL/day/kg, 8.77 mL/day/kg, 44.2 mL/kg, and 36.7 mL/kg. By fixing Q, Vc, and Vp as a geometric mean, CL and F of mAbs in humans were estimated from plasma/serum mAb concentration–time profiles after subcutaneous injection. Using this approach, we accurately estimated CL and F of 25 mAbs in humans (CL: 92% and 100% within 1.3- and 2-fold of the observed values, F: 84% and 100% within 1.3- and 1.5-fold of the observed values).

The pharmacokinetics of mAbs in humans is influenced by several factors, such as body size, age, plasma serum albumin/IgG levels, or antidrug antibodies [20, 21]. Gill et al. summarised the reported interindividual variability (coefficient of variation [CV]) of CL for several mAbs in clinical trials analysed by population pharmacokinetics [20]. In this report, most of the mAbs showed over 30% CV of CL. Gill et al. also reported that the interindividual variability in subcutaneous F of mAbs in humans was around 40–50% of the CV [20]. Considering the interindividual variability of CL and F of mAbs in humans, the predictability of our approach could be acceptable in the development of mAbs. Furthermore, 90.8% and 99.7% of plasma/serum mAb concentration–time profiles after intravenous injection were successfully predicted within 1.5- and 2-fold of the observed values in this study. Generally, prediction of the plasma/serum concentration–time profile is more difficult compared with that of CL and F because the plasma/serum concentration–time profile is influenced by all pharmacokinetic parameters. Prediction of the later part of the plasma/serum mAb concentration–time profile is especially challenging because it requires the highly accurate prediction of both elimination and distribution. As a result, in this study, although it would normally be very challenging, our approach accurately predicted plasma/serum mAb concentration–time profiles after intravenous injection from subcutaneous injection data.

Recently, technologies such as the recycling antibody [22, 23], sweeping antibody [24, 25], and hyaluronidase [26] are being developed to facilitate the subcutaneous injection of mAbs by reducing the effective dosage or increasing the injectable volume. Our approach should be compatible with these technologies since they only affect CL and/or F of mAbs after subcutaneous injection and show similar tissue distribution to normal mAbs. Clinical data on these technologies will be further analysed to investigate the applicability of our approach. While this study focuses on subcutaneous injection of mAbs, other injection routes, such as intramuscular [27, 28] and intraperitoneal [29, 30], are also used in the clinic. Since injection routes do not affect the systemic tissue distribution of mAbs, our approach can be expanded to include intramuscular and intraperitoneal injection in the future. Our approach can also be used with multiple technologies and in a variety of clinical situations.

Conclusion

This study demonstrated an approach for estimating CL and F after subcutaneous injection in humans using only subcutaneous injection data. This means that an intravenous injection will no longer be required to separately estimate the CL and F of mAbs in clinical trials. Our approach has the potential to change the way clinical trials are designed and to accelerate the development of mAbs.

Below is the link to the electronic supplementary material.

Supplementary file1 (DOCX 301 kb)

A large and comprehensive dataset of linear pharmacokinetic parameters for monoclonal antibodies (mAbs) in humans was constructed using public data on 147 mAbs.

Clearance (CL) and bioavailability (F) of mAbs were accurately estimated using only subcutaneous injection data by fixing intercompartmental clearance (Q), volume of distribution in the central compartment (Vc), and volume of distribution in the peripheral compartment [Vp] according to the geometric mean of 147 mAbs.

Plasma/serum concentration–time profiles of mAbs after intravenous injection were accurately predicted using estimated CL and the geometric mean of Q, Vc, and Vp.

Our efficient approach does not require intravenous data to separately estimate CL and F, and can therefore accelerate the clinical development of mAbs.