APHINITY Global PK sub-study design
The design of the APHINITY study has been reported previously [13]. APHINITY included an optional Global PK sub-study with a separate protocol from the main trial.
PK sampling
PK samples were collected only in the PK sub-study. A sparse PK sampling approach was adopted. The PK sampling times for pertuzumab and trastuzumab were pre- and post-dose on cycles 1, 10, and 15 (and Cycle 2 pre-dose). PK sampling times for chemotherapy included 3, 5, and 24 h post-dose for paclitaxel, 3 and 5 h post-dose for the metabolite 6-alpha-hydroxy paclitaxel, and 1, 2, 4, and 5 h post-dose for carboplatin.
Bioanalytical methods
Validated assays were used to measure pertuzumab, trastuzumab, paclitaxel and 6-alpha-hydroxy paclitaxel, and carboplatin from blood samples.
The serum concentrations of pertuzumab were determined by an enzyme-linked immunosorbent assay described previously [14]. The assay used a monoclonal anti-idiotype antibody against pertuzumab to capture pertuzumab from serum samples. Bound pertuzumab was detected with a biotinylated monoclonal antibody (10C4; Antibody Engineering, Genentech, Inc., South San Francisco, CA, USA) against a Genentech, Inc. immunoglobulin G framework and horseradish peroxidase-Avidin D conjugate. A peroxidase substrate (tetramethyl benzidine) was used for color development to quantify serum pertuzumab against a standard curve. The lower limit of quantification in human serum was 150 ng/mL with a standard curve reporting range of 150–4000 ng/mL (limit of detection was 62.5 ng/mL). The inter-assay accuracy (percentage difference) ranged from − 8.75 to 3.84% while the inter-assay precision (percentage coefficient of variation) ranged from 3.89 to 15.3%. The presence of trastuzumab did not interfere with the accurate quantification of pertuzumab in this assay.
Trastuzumab serum concentrations were determined by a validated high-performance liquid chromatography with tandem mass spectrometry (LC–MS/MS) detection described previously [15]. An affinity capture approach using streptavidin magnetic beads coupled with biotinylated recombinant human HER2 extracellular domain was used to enrich trastuzumab from human serum. The bound trastuzumab protein was subjected to ‘on-bead’ proteolysis with trypsin, following standard protein denaturation, reduction, and alkylation processing steps. Prior to digestion completion, working internal standard solution was added. The characteristic peptide fragments produced by this procedure were then quantified as surrogates of the total antibody concentration originating from trastuzumab by LC–MS/MS (i.e., multiple reaction monitoring [MRM]). The lower limit of quantification in human serum was 100 ng/mL with a standard curve reporting range of 100 ng/mL to 2500 ng/mL. The inter-assay accuracy (percentage difference) ranged from − 8.08 to − 1.47%, while the inter-assay precision (percentage coefficient of variation) ranged from 3.07 to 8.44%. The presence of pertuzumab did not interfere with the accurate quantification of trastuzumab in this assay.
Plasma concentrations of paclitaxel and its metabolite 6-alpha-hydroxy paclitaxel were determined by a validated liquid chromatography tandem mass spectrometry method. An aliquot of 50 μL of human plasma (K2EDTA) sample containing paclitaxel and 6-alpha-hydroxy paclitaxel was extracted using supported-liquid extraction. The API 5000 Triple Quad™ (Applied Biosystems, Foster City, CA, USA) was operated in MRM mode under optimized conditions for the detection of paclitaxel and 6-alpha-hydroxy paclitaxel positive ions formed by electrospray ionization. Paclitaxel-d5 was used as an internal standard. Paclitaxel concentrations were calculated with the use of a standard curve with a 1/x2 linear regression over a concentration range of 2.00–2500 ng/mL. Concentrations of 6-alpha-hydroxy paclitaxel were calculated using a separate standard curve with a 1/x2 linear regression over the same concentration range of 2.00–2500 ng/mL. The inter-assay relative standard deviation ranged from 1.5 to 9.6% for paclitaxel and from 2.2 to 8.9% for 6-alpha-hydroxy paclitaxel. The inter-assay accuracy ranged from 86.0 to 96.6% of nominal for paclitaxel and from 96.5 to 103.2% of nominal for 6-alpha-hydroxy paclitaxel. Stability of paclitaxel and 6-alpha-hydroxy paclitaxel was established in human plasma for 449 days at − 20 °C and 1280 days at – 70 °C.
Carboplatin plasma concentrations were determined by a validated inductively coupled plasma tandem mass spectrometry method. Human plasma (K2EDTA) samples (50 μL) containing carboplatin were analyzed on a Perkin-Elmer ELAN DRC II mass spectrometer optimized for the detection of platinum from carboplatin. Terbium was used as an internal standard. Platinum concentrations were calculated with the use of a standard curve with a 1/x2 linear regression over a concentration range of 2.00–1000 ng/mL. The inter-assay relative standard deviation ranged from 0.68 to 4.06%, while the inter-assay accuracy ranged from 97.6 to 100.8% of nominal. Stability of platinum was established in human plasma for 195 days at − 20 °C and − 70 °C.
Data handling
Patients were defined as evaluable for pharmacokinetic (PK) analysis if they had at least one documented pertuzumab administration and a corresponding post-dose pertuzumab PK sample collection. Records were excluded if the time of drug administration or sample collection was missing. No imputation of PK values was performed. Observations with missing PK or time values, or those below the minimum quantifiable concentration, were omitted from the analysis.
Outliers were identified by visual inspection of each individual’s concentration versus time profile. Typically, a data point was deemed an outlier if a trough concentration was greater than the peak concentration, or if the absolute residual variability was five times larger than the expected residual standard deviation.
PK analysis
Pertuzumab concentrations from the APHINITY Global PK sub-study were compared with predictions based on a previously developed pertuzumab population PK model [10]. This model was built on data collected from patients with solid tumors, including MBC, during five phase I/Ib studies, six phase II studies, and one pivotal phase III study [10]. Most of the data ( > 95%) used for population PK development were based on pertuzumab without concomitant trastuzumab treatment; seven of the 12 studies included investigated pertuzumab as a monotherapy. In the previously developed model, pertuzumab PK were described by a two-compartment linear model with a CL, central volume of distribution, and terminal elimination half-life of 0.235 L/day, 3.11 L, and 18 days, respectively. The covariates identified as significantly influencing pertuzumab CL were baseline serum albumin and LBW, with 15.5% and 4.1% of the between-subject variability in CL explained by serum albumin and LBW, respectively.
Comparisons of pertuzumab concentrations from the APHINITY Global PK sub-study to the previously developed model predictions were performed using NONMEM version 7.3 software (ICON Development Solutions, Ellicott City, MD, USA). Post-processing of NONMEM analysis results was carried out in R version 3.2.2 (R Development Core Team, R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria; ISBN 3-900051-07-0; URL https://www.R-project.org/). Individual PK parameters were estimated using first-order conditional estimation with interaction.
To evaluate the agreement of the observed PK data in the APHINITY Global PK sub-study with the historical PK data based on the population PK model, a visual predictive check was performed. Pertuzumab serum concentrations for 10 000 subjects were simulated using LBW resampled from the observed LBW in the pertuzumab arm as well as nominal dose times and amounts for each patient. Albumin levels were not measured in APHINITY. Therefore, the median observed baseline albumin level of 4.3 g/dL (range 3.3–5.7 g/dL, N = 258) in HannaH (NCT00950300), a study of subcutaneous or intravenous trastuzumab for EBC [16], was added as the value for those in the APHINITY Global PK sub-study. In the NeoSphere study, the median observed baseline albumin level was 4.4 g/dL (range 3.1–5.3 g/dL, N = 180), indicating that the selected median baseline albumin value of 4.3 g/dL is appropriate for an EBC patient population and can be used to describe pertuzumab PK in the APHINITY study. Median predicted pertuzumab concentrations and a 95% prediction interval were compared with the observed data.
For the purpose of the exploratory exposure–response analysis, individually predicted pertuzumab serum concentrations based on each patient’s observed serum concentrations and covariates were obtained. The predictions were derived by fixing the parameters in the structural and variance model to the parameter estimates in the historical validated population PK model and generating the individual empirical Bayes estimates with NONMEM by setting MAXEVAL = 0. Individual exposure estimates (AUC, Cmin, and Cmax) were subsequently obtained for use in the exposure–response analysis (detailed below). Diagnostic plots of observed data versus population prediction and individual prediction were examined for adequate fit. Plots of conditional weighted residual versus population prediction and versus time (after first and last doses) were inspected for evidence of systematic lack of fit, and to confirm the absence of bias in the error distributions.
DDIs
The DDI analysis was carried out using R version 3.2.2.
The potential effect of pertuzumab on the steady-state PK of trastuzumab was assessed by comparing the arithmetic means of serum trastuzumab concentrations at pre-dose (Cmin,ss) and post-infusion (Cmax,ss) in cycles 10 and 15 in the pertuzumab and placebo arms. In addition, the 90% CIs in the ratio of the geometric means (calculated by standard methods) were constructed. Similarly, the potential effect of pertuzumab on the PK of paclitaxel (and 6-alpha-hydroxy paclitaxel) and carboplatin was assessed by comparing the arithmetic means of Cmax and area under the concentration–time curve over all concentration measurements (AUClast) in Cycle 1 in the pertuzumab and placebo arms.
For paclitaxel and carboplatin, collection of multiple blood samples on Cycle 1 Day 1 allowed characterization of the post-infusion concentration–time curves using noncompartmental methods. Cmax was defined as the maximum observed concentration and AUClast was calculated using the linear trapezoidal rule and nominal observation times. The 90% CIs in the ratio of the geometric means were also constructed. All observations for 6-alpha-hydroxy paclitaxel at 24 h post-dose were reported as below the quantification limit. Thus, AUClast was not calculated.
The potential effect of trastuzumab on the PK of pertuzumab was assessed by comparing pertuzumab Cmax and Cmin observed in the APHINITY Global PK sub-study with the predictions based on the population PK model. An adequate prediction of the observed PK by the historical model would suggest that there was no impact of trastuzumab on the PK of pertuzumab.
Exploratory exposure–response analysis
To generate individual pertuzumab exposure for patients in the pertuzumab arm, a simulation dataset was constructed based on estimated individual parameters for the population PK model, corresponding LBW and median albumin incorporated as described previously. For all patients with at least one available valid post-treatment concentration measurement of pertuzumab, predictions of Cmax,ss, Cmin,ss, and AUC at steady state (AUCss) were derived. The simulation dataset consisted of a loading dose and three maintenance doses. The loading dose and infusion duration were taken from the NONMEM dataset for each patient in the pertuzumab arm; thereafter, three doses of 420 mg pertuzumab were administered using a 30-min infusion with 3 weeks’ dosing interval. Pertuzumab approximate steady state is achieved following the first maintenance dose, and therefore three maintenance doses were selected to ensure steady state across patients. Individual predictions of concentration at the end of infusion (Cmax,ss) and before next dose (Cmin,ss) were generated following the third maintenance dose for all patients. AUCss was calculated by dividing the maintenance dose by individual clearance values.
The efficacy endpoint in the exposure–efficacy analysis was the primary study endpoint, IDFS [13]. IDFS is the time from randomization to recurrence of ipsilateral invasive breast tumor, recurrence of ipsilateral locoregional invasive disease, a distant disease recurrence, contralateral invasive breast cancer, or death from any cause. Patients who had not had an event at the time of data analysis were censored at the date they were last known to be event-free. The primary exposure metrics used in the exposure–efficacy analysis were individual predicted Cmin,ss and AUCss. Box plots were created to compare exposure (Cmin,ss and AUCss) in the subset of patients with PK data who had an IDFS event (N = 4) versus the patients with PK data who did not (N = 31).
Adverse events (AEs) were considered for the analysis if there was a difference of ≥ 5% in the incidence of Grade ≥ 3 AEs between the pertuzumab and placebo arms in the primary analysis [13]. Furthermore, to be considered for the analysis, pertuzumab PK data were required for ≥ 5 patients experiencing the AE; “any grade ≥ 3 AE” (incidence 64.2% in the pertuzumab arm and 57.3% in the placebo arm in the overall population [13]) and Grade ≥ 3 diarrhea (incidence 9.8% in the pertuzumab arm and 3.7% in the placebo arm in the overall population [13]) met these criteria. These safety endpoints were assessed as binary variables. Results were deemed exploratory and did not reflect formal statistical hypothesis testing.