Introduction
Trastuzumab emtansine (T-DM1) is an antibody–drug conjugate (ADC) comprising the humanized monoclonal antibody (mAb) trastuzumab conjugated to the highly potent cytotoxic agent DM1. T-DM1 delivers DM1 specifically to human epidermal growth factor receptor 2 (HER2)-overexpressing tumor cells [
1,
2]. Like trastuzumab, T-DM1 induces antibody-dependent cellular cytotoxicity, inhibits cell signaling through the phosphatidylinositol-3-kinase/AKT pathway, and inhibits HER2 shedding [
2,
3]. Single-agent T-DM1 demonstrated superior efficacy in treating HER2-positive metastatic breast cancer (MBC) in the first-line and relapsed/refractory settings [
4,
5] and was recently approved by the US Food and Drug Administration for the treatment of HER2-positive MBC [
6].
The pharmacokinetics (PKs) of various analytes (T-DM1 conjugate, total trastuzumab, and DM1) after T-DM1 dosing were characterized in phase I–III studies by non-compartmental analysis (NCA) [
5,
7]. Phase I data suggested that T-DM1 conjugate, the major analyte correlated with treatment efficacy and safety, exhibits faster clearance (CL) at doses ≤1.2 mg/kg and linear PKs for doses ≥2.4 mg/kg for the every 3-week (q3w) regimen [
8]. In phase II and phase III studies with 3.6 mg/kg q3w, linear PK properties were confirmed, and no significant accumulation was observed, consistent with its terminal half-life of approximately 4 days [
9‐
12]. For the weekly (qw) regimen, modest accumulation of T-DM1 conjugate was observed after the first three doses [
13].
By targeting delivery, T-DM1 improves the therapeutic window of DM1. However, based on the maximum tolerated dose (MTD) identified for both q3w and qw regimens, T-DM1 has a relatively narrow therapeutic window compared with typical mAbs [
7,
8,
13]. Therefore, it is important to assess the effects of demographic and pathophysiologic covariates on the PKs of T-DM1 to determine whether dose adjustments are needed. To estimate typical PK parameter values and interindividual variability (IIV) for T-DM1 conjugate, an interim population PK (PopPK) model was developed using PK data from 273 patients in one phase I (TDM3569g) and two phase II studies (TDM4258g and TDM4374g) [
14]. To support the labeling statement on the PK properties of T-DM1 and dosing recommendation, an updated PopPK model including data from an additional 400 patients from one phase II TDM4450g trial and one phase III registrational trial (EMILIA) is reported here. Relative to the previous report [
14] and to better characterize the effects of covariates on T-DM1 PK in patients with breast cancer, a more comprehensive spectrum of covariates with a wider range of values was evaluated. The updated PopPK model was also externally validated by another phase II study (TDM4688g). The model reported here will be of great value for the current clinical application of T-DM1 in patients with breast cancer.
Materials and methods
Patients, PK serum sampling, and bioanalysis
After the administration of single-agent T-DM1, PK data were collected from patients participating in the TDM3569g, TDM4258g, TDM4374g, TDM4450g, TDM4370g (EMILIA), and TDM4688g trials [
4,
5,
8‐
10,
13,
15] (Supplemental Table 1). All study designs were approved by independent ethics committees and conducted in accordance with the Declaration of Helsinki; all patients provided written informed consent [
4,
5,
8‐
10,
13,
15].
T-DM1 serum samples were analyzed by the Bioanalytical Sciences Department at Genentech, Inc., (South San Francisco, CA) or by PPD (Richmond, VA) using a validated indirect sandwich enzyme-linked immunosorbent assay (Genentech, Inc., data on file). The conjugate assay quantified all conjugated trastuzumab containing ≥1 covalently bound DM1 molecule, while excluding unconjugated trastuzumab. The minimum quantifiable concentration (MQC) of the assay ranged from 0.04 to 0.06 µg/mL [
16]. Observations below the MQC were omitted from the analysis.
Establishment and validation of the PopPK final model
PopPK analysis was performed using nonlinear mixed-effects modeling with first-order conditional estimation with interaction method. Model estimation and evaluation were implemented with NONMEM 7 (version 7.1.2; ICON Development Solutions, Ellicott City, MD) with Intel Fortran Compiler (version 10.1.021; Intel, Santa Clara, CA), PerlSpeaksNONMEM (version 3.2.12; Uppsala University, Uppsala, Sweden), and S-PLUS 6.2 (TIBCO Software Inc., Palo Alto, CA).
A base model was first established, with all covariates likely to impact T-DM1 conjugate exposure (Supplemental Table 2) explored for a possible correlation with key T-DM1 post hoc PK parameters. The clinically relevant covariates tested included those related to demographics, renal function, disease severity, and treatment history. The final model was identified by testing covariates using stepwise forward addition followed by backward deletion. A change in minimum objective function (MOF) at the P < 0.01 level of significance (log-likelihood ratio test) was used for the forward addition step, and P < 0.001 was used for the backward deletion step to retain the covariates in the final model.
Internal validation was performed, including goodness-of-fit diagnostics, visual predictive check (VPC) plots [
17,
18], numerical predictive check (NPC) [
18], bootstrap [
19], and shrinkage [
20] assessments. Compared with internal validation, external validation is more rigorous because the model’s predictability is evaluated against a new dataset [
21‐
23]. PK data from the phase II study TDM4688g were used for external validation (
N = 51) [
15]. Predicted T-DM1 serum concentrations for validation patients were obtained using post hoc Bayesian forecasting by fixing the parameters in the structural and variance models to the final estimates. Population-predicted serum T-DM1 concentrations (PRED) were compared with observed T-DM1 concentrations (DV). Prediction errors (
P
e) were calculated as (Eq.
1):
$$ P_{\text{e}} = \frac{{\left( {{\text{PRED}} - {\text{DV}}} \right)}}{\text{DV}} \times 100\;\% $$
(1)
Bias (mean prediction error [MPE]) was then calculated (Eq.
2):
$$ {\text{MPE}} = \frac{{\sum {P_{\text{e}} } }}{n} $$
(2)
where
n denotes the number of observations.
The predicted clearance (CL
POP,i
) and central volume (
V
cPOP,i
) for each patient (per individual covariate values) were obtained and compared with the maximum a posteriori probability (MAP) Bayesian estimates of clearance (CL
Bayesian,i
) and central volume (
V
cBayesian,i
) for each validation patient based on the available concentration measurements and final PopPK parameter estimates. Prediction errors were calculated for each individual patient (
P
ei) and expressed as a percentage of the MAP Bayesian estimate (Eq.
3):
$$ \begin{gathered} {\text{CL}}\_P_{\text{ei}} = \frac{{{\text{CL}}_{{{\text{POP}},i}} - {\text{CL}}_{{{\text{Bayesian}},i}} }}{{{\text{CL}}_{{{\text{Bayesian}},i}} }} \times 100\;\% \hfill \\ V_{\text{c}} \_P_{\text{ei}} = \frac{{V_{{{\text{c}}\;{\text{POP}},i}} - V_{{{\text{c}}{\text{Bayesian}},i}} }}{{V_{{{\text{c}}{\text{Bayesian}},i}} }} \times 100\;\% \hfill \\ \end{gathered} $$
(3)
The bias of prediction was assessed by MPE (Eq.
4):
$$ {\text{MPE}} = \frac{{\sum {P_{\text{ei}} } }}{N} $$
(4)
where
N denotes the number of patients.
Based on the final model, the effect of extreme values of each statistically significant covariate (5th and 95th percentiles) on T-DM1 PK parameters (CL and V
c) was evaluated. To assess whether T-DM1 PKs differed in various populations, individual Bayesian post hoc CL and V
c were normalized by statistically significant covariates in the final model and compared among clinically relevant populations (e.g., those defined by age, race, region, renal function, disease severity, and treatment history).
Sensitivity analyses
Sensitivity analyses were performed to examine the influence of statistically significant covariates on the expected steady-state exposure of T-DM1 conjugate, including steady-state area under the concentration versus time curve (AUC), maximum concentration (C
max), and trough concentration (C
trough). The simulated exposure of patients with extreme covariate values (5th and 95th percentiles) was compared with a typical patient with median covariate values for each of the statistically significant covariates in the final model.
Model applications: exposure comparison among various populations
To compare the expected T-DM1 conjugate steady-state exposure in populations defined by age, race, region, or renal function, individual T-DM1 exposures (steady-state AUC, C
max, C
trough) were simulated using Bayesian post hoc PK parameters for each patient for multiple doses of 3.6 mg/kg q3w. This simulation accounts for potential correlations among covariates.
Discussion
Results from this PopPK analysis informed the prescribing information for T-DM1 in patients with HER2-positive MBC, specifically its PK properties and dose justification based on the impact of weight, age, race, region, renal function, treatment history, and disease and health status on PKs. The model can reliably predict T-DM1 exposure and IIV irrespective of prior trastuzumab treatment.
Linear PKs were observed in patients receiving clinically relevant T-DM1 doses (2.4–4.8 mg/kg q3w). T-DM1 CL appeared to be faster in the five patients who received T-DM1 doses of ≤1.2 mg/kg q3w, likely due to target-mediated disposition at low doses. However, the PopPK model parameters associated with nonlinear elimination were not estimated with good precision, possibly because of the limited amount of data available for these lower doses of T-DM1. Sensitivity analyses indicated that the inclusion/exclusion of PK data from these five patients did not impact PK parameter estimation.
The T-DM1 CL (0.676 L/day) is approximately three times faster than that of unconjugated trastuzumab (approximately 0.2 L/day) [
26]. The elimination half-life of T-DM1 (approximately 4 days) is shorter than that for a typical immunoglobulin G1 antibody (2–3 weeks) [
27]. These data support multiple mechanisms of T-DM1 CL. T-DM1 undergoes both target-mediated (HER2) and non-specific (partly Fc-mediated) proteolytic degradation, similar to that of mAbs [
28]. Moreover, DM1 deconjugation from T-DM1 may contribute to T-DM1 CL, which may partially explain the faster CL and shorter half-life of T-DM1 [
29]. The central volume of distribution of T-DM1 for a 70-kg patient is 3.127 L, which is similar to that of unconjugated trastuzumab (~2.95 L) ([
26]), other mAbs (~2.4–5.5 L) ([
27]), and ADCs such as brentuximab vedotin (4.16 L) [
30]. The peripheral volume of distribution of T-DM1 is relatively small (0.66 L) and is lower than that of brentuximab vedotin (8.06 L) [
30]. Overall, the PK properties of T-DM1 are more similar to a mAb than to a small molecule drug, with relatively slow CL and a small volume of distribution that is mostly confined to plasma.
The linker used in ADCs may affect CL by impacting the rate of deconjugation of the cytotoxic drug component. Several linkers have been used in various compounds in clinical development. For example, T-DM1 contains a non-cleavable thioether linker (4-(
N-malemidomethyl) cyclohexane-1-carboxylate [MCC]); brentuximab vedotin contains a peptide-based linker (maleimidocaproyl-valine-citruline-p-aminobenzyl-oxycarbonyl [MC-vc-PAB]); inotuzumab ozogamicin and gemtuzumab ozogamicin contain acid-labile linkers (4-(4-acetylphenoxy)butanoic acid [AcBut]); and AVE9633 and SVR3419 contain disulfide linkers (
N-succinimidyl-3-(2′-pyridyldithio) butyrate [SPDB]). T-DM1, with its MCC linker, has a CL rate (~0.7 L/day) similar to ADCs containing SPDB linkers (AVE9633 200–260 mg/m
2, ~0.9–1.1 L/day [
31,
32]; SAR3419, ~0.6 L/day [
33]). However, higher CL values have been observed for ADCs containing MC-vc-PAB linkers (brentuximab vedotin 1.8 mg/kg, ~1.5–1.8 L/day [
30,
32,
34]) or AcBut linkers (inotuzumab ozogamicin 1.8 mg/m
2, ~2.2–3.8 L/day [
32,
35,
36]; gemtuzumab ozogamicin 0.23 mg/kg, ~3.2 L/day [
32,
37]). The lower CL of T-DM1 at clinically relevant doses (i.e., when target-mediated CL is largely saturated) might be attributable to its non-cleavable linker, which results in less rapid deconjugation of the cytotoxic component.
The relatively small IIV of T-DM1 CL and
V
c (~19 and 12 %, respectively, Table
1) is consistent with the expected IIV of mAbs (~30–40 %) [
27]. The IIV estimated for Q and V
p was relatively high in the final PopPK model, likely because of the limited data in the terminal phase of the phase II and phase III trials. A relatively large η-shrinkage of >30 % for
V
c,
Q, and
V
p was observed in the final model, suggesting that PK data may not sufficiently characterize the IIV for these parameters. More intensive concentration–time data may improve the accuracy of the IIV estimation, especially for
Q and
V
p, which had relatively high η-shrinkage in both the base (data not shown) and final models.
The body weight-based regimen of 3.6 mg/kg was established as the MTD for T-DM1 in phase I testing [
8]. Per a theoretical simulation comparing flat versus body weight-based dosing [
38], flat dosing would not be expected to reduce IIV, given that body weight impacts T-DM1 CL and
V
c with exponential function values of 0.5 and 0.6, respectively (Table
1, parameters
θ
6 and
θ
5). Sensitivity analyses suggested that the magnitude of the impact of body weight on T-DM1 PK parameters (CL,
V
c) and exposure (AUC,
C
max,
C
trough) is relatively small (<25 %) (Table
2; Fig.
3).
T-DM1 conjugate was cleared more slowly in patients with lower body weight (Table
2); however, conjugate exposure was lower in lighter patients due to body weight-based dosing (Fig.
3). Of the 671 patients included in this analysis, 68 (10.1 %) weighed ≥90 kg and thus received a greater amount of T-DM1 versus the overall population. These patients had higher mean exposure (21.0 % higher for AUC, 18.4 % higher for
C
max, 19.1 % higher for
C
trough, with largely overlapping intervals of 5th to 95th percentiles), despite faster CL and larger
V
c. Furthermore, based on the exposure–response analysis of T-DM1 3.6 mg/kg q3w [
7,
12,
39,
40], the variability in T-DM1 AUC and
C
max is not expected to have a clinically meaningful impact on overall safety; thus, the current body weight-based regimen remains appropriate, with no further dose adjustment recommended for heavier patients.
The covariates of age, race, and geographic region were not significant, suggesting that no further dose adjustment based on these covariates is necessary. Although Asian patients and patients from Asia have lower mean exposures, these differences are likely due to lower body weight. Based on the exposure–response analysis of T-DM1 3.6 mg/kg q3w [
7,
12,
39,
40], this exposure variability is not expected to have a clinically meaningful impact on safety.
Compared with the earlier PopPK model that included only phase I and phase II data [
14], two additional covariates (ECD and TBL) were identified as significantly impacting T-DM1 CL in the revised updated PopPK model. This may be a consequence of the increased number of patients in the second- and third-line treatment settings, patients who introduced a larger dynamic range for these two covariates. However, the effects of ECD and TBL were not considered clinically meaningful because of the small-to-moderate magnitude of effect on AUC,
C
max, and
C
trough. Mechanistically, ECD and TMBD are highly correlated with total HER2 antigen concentration, and their correlation with T-DM1 CL suggests a potential mechanism of HER2 target-mediated CL of T-DM1. These findings are similar to those from a PopPK analysis of trastuzumab, where ECD was also found to be a statistically significant covariate for trastuzumab CL [
26]. While high baseline TBL resulting from prior trastuzumab treatment may increase total trastuzumab exposure and compete with T-DM1 for the HER2-mediated CL pathway, the effects of this covariate on exposure do not appear to translate into clinically relevant differences in safety [
12,
39,
40].
Although TMBD (and potentially overall disease severity) for most patients decreased over time and after repeated T-DM1 infusions (because of shrinkage of target lesions following treatment), T-DM1 PKs do not appear to change over time. Model diagnostic plots did not suggest the existence of apparent bias of residual versus time (Supplemental Fig. 3), further indicating that inclusion of time-varying covariates for this patient population may not be necessary. Based on NCA, similar PK parameters were observed in cycle 1 and after several doses of T-DM1 (data not shown), suggesting that the PKs of T-DM1 are not largely affected by tumor size or changes in disease severity after treatment; this may be due to the relatively small magnitude of effect of TMBD on T-DM1 CL and exposure.
Given the high molecular weight of T-DM1, renal function is unlikely to impact CL. Based on preclinical studies, T-DM1 is mainly eliminated through bile after conversion to DM1-containing catabolites, with minimal (<5 %) renal elimination [
41]. T-DM1 PK parameters (CL,
V
c) after normalization of other covariates were similar in patients with varying degrees of renal function (normal, mild impairment, or moderate impairment [per calculated CrCL]). Thus, renal impairment is not expected to impact T-DM1 PKs.
Some covariates related to hepatic function were tested in this analysis, with ALBU and AST identified as being statistically significant. A dedicated phase I study (BO25499) of T-DM1 in patients with MBC and normal or mild/moderate hepatic impairment is ongoing; PKs and safety results will be reported separately. PKs in patients with early-stage breast cancer characterized by lower baseline tumor burden and non-metastatic disease status are also being evaluated.