Population pharmacokinetics and exposure–response relationship of amatuximab, an anti-mesothelin monoclonal antibody, in patients with malignant pleural mesothelioma and its application in dose selection

Abstract

Purpose

To characterize amatuximab pharmacokinetics (PK) and the relationship of amatuximab exposure with response in patients with unresectable malignant pleural mesothelioma (MPM) receiving amatuximab with pemetrexed and cisplatin.

Methods

A nonlinear mixed effects PK model was built using data from all of the amatuximab studies conducted to date. Patients received amatuximab alone or in combination with chemotherapy. The influence of demographic, laboratory and disease characteristics on PK parameters was assessed. Exposure–response analyses explored relationships between amatuximab exposure and overall survival (OS), progression-free survival (PFS) and safety. Alternative amatuximab dosing regimens were explored with simulations using population PK and parametric survival models.

Results

Amatuximab PK was best described by a two-compartment model with parallel linear and nonlinear elimination pathways. Body weight and an antidrug antibodies reaction with the titer >64 affected volume of distribution and clearance, respectively. Exposure–response analyses demonstrated that the amatuximab exposure (C_min) showed a significant effect on OS (log-rank test, P = 0.0202). For patients with amatuximab C_min above the median (38.2 μg/mL), the median OS was 583 days (90 % CI 418 –NE). For patients with C_min ≤ 38.2 μg/mL, the median OS was 375 days (90 % CI 325–486). The amatuximab exposure showed similar significant effect on PFS. Exposure–response analysis for adverse events did not reveal any relationship.

Conclusions

In patients with MPM, higher amatuximab exposure in combination with chemotherapy was shown to be associated with longer OS, supporting evaluation of more frequent dosing in future trials to achieve higher exposure and subsequently longer OS.

Malignant pleural mesothelioma (MPM) is an aggressive disease with poor prognosis. Although patients with a limited tumor burden may benefit from surgical resection, most patients have advanced disease at diagnosis and are not candidates for surgery [1]. For patients who are not eligible for curative surgery, the median survival with supportive care alone is approximately 6 months, whereas with the current standard treatment, a combination of cisplatin and pemetrexed, the median survival is 12 months [2, 3].

Mesothelin is a glycosylphosphatidyl inositol (GPI)-anchored membrane glycoprotein, which is present in a restricted set of normal adult tissues such as the mesothelium [4]. In contrast, mesothelin is highly expressed in many epithelial cancers. More than half of all the ovarian cancers and lung adenocarcinomas and nearly all epithelial mesotheliomas and pancreatic ductal adenocarcinomas express mesothelin [5–9]. Although the normal biological function of mesothelin is unknown, growing evidence suggests that it may play a role in tumorigenesis and metastasis in mesothelioma [10]. Its limited expression in normal human tissue and high expression in tumor makes mesothelin an excellent target antigen for antibody-based immunotherapy [11].

Amatuximab (MORAb-009) is a chimeric high-affinity monoclonal IgG1/k antibody targeting mesothelin [12]. In vitro, amatuximab elicits antibody-dependent cellular cytotoxicity (ADCC) against mesothelin-expressing tumor cell lines and inhibits heterotypic cell adhesion of mesothelin-positive tumor cells to CA125-expressing tumor cells. In tumor xenograft studies, combination treatment with amatuximab plus chemotherapy led to a greater reduction in the growth of mesothelin-expressing tumors than either amatuximab or chemotherapy alone. In a Phase I study of patients with mesothelin-expressing cancers, weekly infusions of amatuximab were well tolerated and the maximum tolerated dose was identified as 200 mg/m² [13]. Based on its safety in the Phase I study and preclinical studies showing synergy with chemotherapy, amatuximab was combined with pemetrexed and cisplatin in a single-arm Phase II study in patients with unresectable MPM. This multi-center Phase II study (NCT00738582) demonstrated that amatuximab in combination with pemetrexed and cisplatin was well tolerated and resulted in an overall response rate of 39 % (all PRs), a disease control rate of 90 % and median progression-free survival (PFS) of 6.1 months by independent radiologic review in the primary efficacy population [14]. Although the study did not meet the targeted primary endpoint of 3-month improvement in PFS over historical controls, the median overall survival (OS) of 14.8 months [14] compares favorably with 13.3 months in the fully supplemented subset of patients in the Phase III study of cisplatin and pemetrexed [2]. The extended final plateau of the Kaplan–Meier curve of OS of patients in that study suggests that the combination may be particularly effective in a subgroup of patients. With an aim to understand this observation and to support the future strategy for amatuximab development in MPM, population pharmacokinetics (PK) and pharmacokinetic/pharmacodynamic (PK/PD) analyses were performed.

Materials and methods

Patients and study design

The population PK analysis used data collected from four Phase I and II studies with advanced cancers, pancreatic cancer or MPM, and PK/PD analyses used PK and response data from the recently published Phase II study in MPM [14]. All studies were conducted in accordance with Good Clinical Practice and under ethical principles established by the Declaration of Helsinki, and patients gave written informed consent.

One Phase I study (NCT00325494; 13) was conducted in the USA in 24 patients with advanced mesothelin-expressing tumors who had failed standard chemotherapy. Amatuximab was administered via IV infusion weekly for 4 of a 6-week cycle at ascending doses from 12.5 to 400 mg/m² to sequential cohorts of patients. A second Phase I study (NCT01018784; 15) was conducted in 17 Japanese patients with solid tumors who had failed standard chemotherapy. Amatuximab was administered via IV infusion weekly in 4-week cycles at ascending doses from 50 to 200 mg/m² to sequential cohorts of patients. A Phase II study (NCT00570713) was conducted in 155 patients with advanced pancreatic cancer. This was a randomized, double-blind, placebo-controlled study assessing the efficacy of amatuximab in combination with gemcitabine. Amatuximab was administered weekly via IV infusion at 5 mg/kg for 7 weeks in an 8-week cycle (Cycle 1); in subsequent 4-week cycles, amatuximab was given weekly for 3 weeks. Out of 155, 71 patients from this study contributed to PK database. The Phase II study in 89 patients with unresectable MPM was an international, open-label study assessing the effect of amatuximab in combination with cisplatin/pemetrexed [14]. Patients received up to six cycles of combination treatment with amatuximab, pemetrexed and cisplatin. Amatuximab was administered on Days 1 and 8 of each 21-day cycle. Pemetrexed and cisplatin were administered on Day 1 of each 21-day cycle. Patients who completed combination treatment or discontinued chemotherapy because of intolerable toxicity continued to receive amatuximab as single-agent maintenance therapy on Days 1 and 8 of all subsequent cycles until disease progression. Amatuximab was administered at a dose of 5 mg/kg by infusion for up to 2 h. Amatuximab PK data were available from 87 patients.

PK sampling and bioanalytical method

In the US Phase I study [13], blood samples were taken for the first four doses at pre-dose, mid-infusion, end of infusion and 30, 60 min then 2, 4 and 24 h (only first and fourth dose) after the end of each infusion of amatuximab. In the Japanese Phase I study [15], blood samples were taken (for the first and fourth dose) pre-dose, 30 min after the start of infusion, end of infusion and 30, 60 min then 2, 4, 24 and 72 h (only first dose) after the end of each infusion. For the second and third doses, blood samples were only collected pre-dose. Blood samples were also taken pre-dose on Day 1 of any subsequent cycle. For the Phase II pancreatic cancer study, blood samples were taken pre-dose and at the end of the infusion on weeks 1 and 3 of all cycles. For the Phase II mesothelioma study [14], blood samples were taken pre-dose and at the end of infusion on day 1 week 1 of each cycle.

Serum concentrations of amatuximab, except US Phase I study, were measured using a recombinant mesothelin antigen-based electrochemiluminescent immunoassay (ECLIA) to capture and quantify the serum concentration of free/partially complexed amatuximab. The lower limit of quantitation (LLOQ) for the method is 98 ng/mL in undiluted serum. Serum concentrations of free amatuximab in study samples from US Phase 1 study were measured with a colorimetric ELISA utilizing a recombinant mesothelin capture and a goat anti-mouse IgG horseradish peroxidase reporter. This assay format had a LLOQ of 50 ng/mL in neat serum.

The titer of amatuximab ADA in human serum samples was available with at least fourfold differences representing measureable changes in ADA titer.

Population PK base model development

Population PK estimation and model evaluation were undertaken using nonlinear mixed effects modeling (NON-MEM) software version 7.2.0 (ICON Development Solutions; Ellicott City, MD) interfaced with PDxPop 5.0 (ICON Development Solutions; Ellicott City, MD). PK parameters were estimated using sequential estimation methods of iterative two stage followed by the first-order conditional with interaction. Amatuximab concentrations below the limit of quantification (BLQ) were included in the dataset, and M3 (estimating likelihood that BLQ concentration is less than LLOQ), YLO/M2 (estimating likelihood that all measured concentrations >BLQ are greater than LLOQ) and replacement with ½ BLQ methods were tested [16].

The population PK model was developed in a stepwise manner with evaluation at each step. Two initial PK models were assessed: a two-compartment model either with first-order elimination, a structural model consistently reported for many monoclonal antibodies [17], or with parallel Michaelis–Menten and first-order elimination. The population PK model was parameterized as linear clearance (C_L), volume of distribution of the central compartment (V₁), intercompartmental clearance (Q) and volume of distribution of the peripheral compartment (V₂), and for nonlinear clearance, maximal elimination rate (V_max) and the Michaelis–Menten constant (K_m). All between-subject variability on the PK parameters was described by a lognormal parameter distribution. To account for the difference between model-predicted and actual amatuximab concentrations, various residual error models were tested.

The following criteria were considered when assessing the acceptability of a model: the successful convergence of the minimization procedure; the number of significant digits being >3; the covariance step terminating without any warning message; no correlation between parameters >0.95; and standard errors of the estimates <50 %. The following goodness-of-fit plots were used to evaluate the ability of the PK to describe the available data: population and individual predicted amatuximab concentrations versus observed concentrations; conditional weighted residuals (CWRES) versus time and versus population predicted concentrations.

Covariate model development

Baseline demographic covariates (age, gender, race/ethnicity [Japanese vs. non-Japanese] and weight), disease-related covariates (Eastern Cooperative Oncology Group [ECOG] performance status), chemotherapy regimen (pemetrexed plus cisplatin; gemcitabine), baseline serum albumin, baseline mesothelin levels and the development of antidrug antibodies (ADA) were included in the covariate analysis. For Phase II studies, Karnofsky performance status (KPS) scores were converted to ECOG status. KPS of 100/90 was considered equivalent to ECOG 0 and KPS 80/70 equivalent to ECOG 1. KPS of 70 was combined with ECOG 1 as numbers of patients were few in this category (5/84). For covariate analysis, individual Bayes post hoc PK parameter estimates were generated using the base model. The difference of individual point estimates from the corresponding population value (η) was plotted against the covariates to identify any potential relationships; η-shrinkage was calculated and reported for inter-individual variability (IIV) parameters estimated by the model [18]. Parameters with >30 % shrinkage were excluded from the covariate analysis.

Covariate effects significant at a level of P value ≤0.01 in the univariate analysis (which corresponded to the difference in objective function (OFV) of 6.63; Chi-squared distribution for 1 degree of freedom) were included in the full model. The significance of covariate effects in the full model was then retested by backward deletion. The significance level for retaining a covariate in the final model was P ≤ 0.001, which corresponded to the difference in OFV of 10.83 for 1 degree of freedom. Using data from study in MPM patients only, effect of baseline mesothelin levels (median = 61.3 ng/mL, range 1.44–36 ng/mL) was also tested on amatuximab PK.

Model evaluation

The PK model was qualified by a prediction-corrected visual predictive check [19]. The simulated amatuximab concentrations were obtained for 199 patients using the actual dosing history, covariate information and the final PK model (n = 20). The median and 5th and 95th percentiles (90 % PI) of these simulated concentrations were calculated and plotted with observed amatuximab pre-infusion concentration separately for Phase I and Phase II studies.

Exposure–response analyses in MPM

Exploratory analyses of the relationship between amatuximab exposure (C_min: observed average pre-infusion for last three assessments; C_max: observed average post-infusion concentration for last three assessments) and OS, the time between the patient’s first dose of amatuximab and date of death due to any cause, and PFS, the time between the patient’s first dose of amatuximab and either the date of radiologic progression of their disease or the date of death due to any cause, were undertaken. Patients were categorized according to whether they were above or below the median amatuximab exposure. Data for patients who were known to have not died or progressed at the time of the analysis were censored at the date they were last known to be alive or the clinical cutoff date, whichever occurred first.

Time-to-event analysis for PFS and OS was performed using Kaplan–Meier and Cox regression analyses using survfit() and Coxph() functions, respectively, in S-PLUS version 8.1 (TIBCO Software, Inc., Palo Alto, CA). Predictors were considered statistically significant at P value of 0.05 (log-likelihood ratio test) in the univariate analysis. Difference between two Kaplan–Meier plots was tested using a two-sided log-rank test. To take into account effect of other confounding risk factor, a multivariate parametric survival model was developed to estimate the probability distribution of T, a random variable representing the time from study start to survival, as a function of various covariates (proportional hazard model). To estimate the hazard function, different assumptions on distribution structure were tested (Weibull, extreme, normal, log-normal, logistic, log-logistic). Model parameter estimation was performed using censorReg in S-PLUS 8.1. The following factors were tested as predictors of OS and PFS: demography (age, race/ethnicity and gender); disease characteristics [baseline KPS, baseline tumor size, histology (epithelial vs. mixed), stage of disease (III/IV vs. I/II), baseline levels of CA125, mesothelin and MPF]; and amatuximab treatment [C_min, presence or absence of an ADA of any titer, change in biomarkers post-treatment (CA125 and MPF ratio to baseline at last assessment)]. Patients with missing values were excluded from multivariate analysis. The significant (P < 0.05) independent predictors from the Cox regression analysis were added to the model iteratively in the order of significance. Predictors were kept in the model according to the following criteria of forward inclusion/backward exclusion: log-likelihood ratio test, inclusion P value of 0.05 and exclusion P value of 0.01.

Simulations and dose evaluation

In light of the positive exposure–response relationship for both PFS and OS data and benefit of achieving amatuximab C_min above 38.2 μg/mL for any future trials for better response, the following were considered: a more frequent administration of amatuximab (weekly without interruption) at 5 mg/kg, in order to minimize the development of ADA which, when they occurred, had led to a higher rate of discontinuations due to hypersensitivity reactions, and a higher dose of 7.5 mg/kg also administered weekly. Stochastic simulations were performed using the amatuximab population PK model. Steady-state C_min of amatuximab given a dose of 5 and 7.5 mg/kg administered weekly with and without ADA > 64 was simulated. The administered dose was given weekly as a 1-h constant rate infusion (no interruption) in a 21-day cycle for six cycles. Patients were assumed to have a body weight of 70 kg. For each of the two dosing regimens, the percentage of patients predicted to achieve an amatuximab steady-state C_min above 38.2 μg/mL was calculated.

The parametric model for OS was used to predict the median OS at the median-predicted steady-state C_min given a dose of 5 or 7.5 mg/kg. All patients were assumed to be either negative for ADA or had an ADA > 64 titer and have an epithelial histology for MPM tumor.

Results

Base PK model

The population PK model was developed using 3236 amatuximab serum concentrations collected from 199 patients of which 49 amatuximab serum concentrations were BLQ, mainly when ADA were positive. Most of the patients (64.3 %) were male with age ranging 33–90 years (median, 65 years) and body weight ranging between 35 and 134 kg (median, 74 kg; see Table 1).

Table 1.

Summary of continuous and categorical covariates for patients included in the population pharmacokinetics and exposure-response analyses for amatuximab

Covariate (unit)	Mean	SD	Median	Min	Max
Pharmacokinetic analysis database (N = 199)
Age (years)	64.5	9.55	65.0	33.0	90.0
Baseline weight (kg)	75.1	16.4	74.0	35.0	134
Baseline albumin (g/dL)	3.80	0.53	3.80	2.38	5.46
Gender n (%)	Male = 128 (64.3); female = 71 (35.7)
Race n (%)	Caucasian = 162 (81.4); Japanese = 17 (8.54); Others = 16 (8.04); missing = 4 (2.01)
Baseline ECOG performance status n (%)	0 = 119 (59.8); 1 = 78 (39.2); 2 = 2 (1)
Exposure-response analyses MPM database (N = 89)
Age	65.5	7.59	67.0	46.0	80.0
Baseline tumor size (sum of longest diameter of target lesions in mm)	91.3	66.5	76.0	11.0	334
Baseline mesothelin (N = 84; ng/mL)a	121	138	61.3	1.44	736
CA-125 baseline (N = 79; U/mL)	72.1	171	11	2	1266
CA-125 ratio to baseline at last assessment (N = 76)	16.4	20.2	9.57	0.25	95.3
MPF baseline (N = 83; ng/mL)	18.8	26.5	7.32	1.14	169
MPF ratio to baseline at last assessment (N = 58)	1.36	1.46	1.14	0.03	9.46
Race	Caucasian = 79 (89), others = 10 (11)
Gender n (%)	Male = 69 (78) female = 20 (22)
KPS	100 = 22 (24.7), 90 = 40 (44.9), 80 = 21 (23.6), 70 = 6 (6.7)
Histology	Epithelial 79 (88.8), mixed 10 (11.2)
Stage of disease	IV = 43 (48), III = 35 (39), II = 5 (6), IB = 4 (5), IA = 2 (2)

^aTested in both population PK and exposure–response analyses

The population PK model development started with a two-compartment PK model with linear elimination from the central compartment. Adding a parallel nonlinear elimination pathway resulted in a 142.6 point drop on OFV, improved the overall fit to the available data and reduced the IIV. For handing BLQ data, M3, YLO and replacement with ½ BLQ were tested. Compared to the ½ BLQ replacement method, the M3 and YLO methods resulted in a higher drop in the OFV; however, the PK model became unstable and sensitive to change in OFV. As the percentage of BLQ was low (1.51 %), replacement with ½ BLQ was considered appropriate [20]. The IIV could be estimated on C_L, V₁ and V_max. The distribution of eta on V_max was tailed to the right. To fix this problem, Manly, logit and Box–Cox eta transformations were tested [21, 22]. Manly transformation resulted in a normal distribution of eta on V_max (Supplementary Figure 1). Covariance between V₁ and C_L was estimated. Residual variability was best described by a combined (proportional and additive) error model. However, the additive part could not be estimated and was fixed to a quarter of BLQ.

Plots of eta (parameter) versus covariates from the base PK model demonstrated potential relationships for an effect of gender on V₁ and V_max, ECOG (0 vs. 1 and 2) on C_L and V_max, race (Japanese vs. non-Japanese) on V₁ and V_max and weight on V₁ (Supplementary Figure 2). To test the effect of an ADA reaction on the amatuximab PK, only V₁ and linear C_L were considered. The ADA reaction was categorized into 5 categories: presence or absence of an ADA; presence or absence of ADA with titer >1; presence or absence of ADA with titer >4; presence or absence of ADA with titer >64; and presence or absence of ADA with titer >160. As shown in Supplementary Figure 2, the effect of ADA on C_L was more apparent when the titer was greater than 64.

Final PK model

The final population PK model for amatuximab was a two-compartment model with parallel linear and nonlinear elimination pathways, parameterized in terms of C_L, V₁, V₂, Q, V_max and K_m. The model included IIV on C_L, V₁ and V_max; however, IIV was not estimable for V₂, Q and K_m (Table 2).

Table 2.

Paramteter estimates of population pharmacokinetic model for amatuximab

Parameter [units]	NONMEM estimates
	Point estimate	%RSE	95 % CI
CL [L/h] = ΘCL* ${\Theta }_{\text{ADA}}^{\text{ADA}}$
ΘCL [L/h]	0.0299	4.18	0.0275–0.324
ΘADA (effect of ADA on CL)	1.49	2.17	1.43–1.55
Vc [L] = ΘVc*(WGT/70)ΘWGT
ΘVc [L]	3.89	2.51	3.70–4.08
ΘWGT (body weight effect on Vc)	0.597	18.9	0.376–0.818
Q [L/h] = ΘQ
ΘQ [L/h]	0.0147	6.59	0.0128–0.0166
Vp [L] = ΘV2
ΘVp[L]	2.62	4.24	2.40–2.84
Vmax [mg/h] ΘVmax
ΘVmax [mg/h]	0.173	18.1	0.112–0.234
Km [ng/mL] = ΘKm
ΘKm	790	11.8	607–973
Shape factor for manly transformation of ${\omega }_{V_{\max }}^2$	−0.181	75.7	−0.450 to 0.0875
Inter-individual variability (CV %)
${\omega }_{C_{\text{L}}}^2$	24.1	16.0
${\omega }_{V_c}^2$	24.1	10.8
Correlation between CL and Vc	41.1 %	25.5
${\omega }_{V_{\max }}^2$	124	38.8
Residual variability (CV% or SD)
σ2 prop (CV %)	33.9	0.587
σ2 add (SD ± ng/mL)	24.5 FIXED

%RSE: percent relative standard error of the estimate = SE/parameter estimate * 100; C_L = clearance, V_c = volume of central compartment, V_p = volume of peripheral compartment, Q = inter-compartment clearance from V_c to V_p, V_max maximum elimination rate, K_m, amatuximab concentration at which 50 % of V_max is reached WGT weight, ADA = 1 (presence of ADA with titer >64), ${\omega }_{C_{\text{L}}}^2$, ${\omega }_{V_1}^2$, ${\omega }_{V_{\max }}^2$ = covariance of random effect of C_L, V_c and V_max, respectively, σ² prop, σ² add = proportional and additional component of the residual error model, respectively

Amatuximab C_L was increased 1.5 times when the ADA titer was greater than 64. Amatuximab V₁ was identified to increase with increasing body weight (power = 0.597). All the parameters of the structural model were estimated with good precision (%RSE ≥ 2.17 to ≤18.9 %). IIV in the model parameters was moderate to high ranging between 24.1 % for V₁ and 124 % for V_max. IIV was well estimated with good precision for all parameters (%RSE ≤ 38.8 %). The residual variability in amatuximab concentrations was moderate (proportional: CV = 33.9 %, additive: SD = 0.245 μg/mL). No trend was observed in goodness-of-fit plots.

In order to evaluate the predictive performance of the final PK model for amatuximab, simulations were performed. Based on sampling schemes, plots were separated by Phase I and Phase II studies (Fig. 1). The median (red line) and 5th and 95th percentiles (90 % PI, gray area) of these prediction-corrected simulated concentrations were calculated and plotted with observed amatuximab concentration data. The majority of the observed amatuximab concentrations are within the 90 % prediction intervals. Hence, the amatuximab concentration time course has been reasonably well defined by the final PK model with good predictive performance (Fig. 2).

Fig. 1.

Prediction-corrected visual predictive check of amatuximab population pharmacokinetic model

Fig. 2.

Kaplan–Meier plots for OS and PFS in MPM patients receiving amatuximab in combination with pemetrexed and cisplatin stratified by amatuximab exposure

Effect of covariates

In order to explain IIV in amatuximab PK, the effect of various covariates was tested on C_L, V₁ and V_max. Weight was identified to have effect on V₁. Weight explained 4.3 % IIV on V₁. Population estimate of V₁ for a 70-kg patient is 3.89 L and will vary between 2.49 and 5.55 L for patients weighing 35 and 134 kg, respectively. After taking into account weight effect, age (33–90 years), race (Japanese vs. non-Japanese) and gender did not influence amatuximab PK. Baseline albumin 2.38–5.46 g/dL did not influence amatuximab PK.

Another covariate that affected amatuximab PK was an ADA reaction with the titer being greater than 64. With an ADA reaction of titer >64, the amatuximab linear clearance increased by a factor of 1.5, which results in a lower amatuximab exposure (Supplementary Figure 3).

Exposure–response relationship: overall survival

Summary of the amatuximab exposure parameters (observed C_min and C_max) tested in exposure–response relationships is presented in Supplementary Table 1. For OS, response and amatuximab exposure data were available from 86 patients with median C_min and C_max of 38.2 and 139 μg/mL, respectively. Kaplan–Meier time-to-event analysis for OS was performed, with patients stratified according to median of amatuximab C_min and C_max. Compared to C_max, C_min showed a more significant effect on OS (log-rank test P = 0.0491 vs. P = 0.0202; Fig. 3a, b). For patients with C_min above the median (38.2 μg/mL), the median OS was 583 days (90 % CI 418 –NE). For patients with C_min below the median, the median OS was 375 days (90 % CI 325–486).

Fig. 3.

Percentage of MPM patients predicted, using amatuximab population pharmacokinetic model, to achieve steady-state amatuximab C_min above 38.2 μg/mL when given weekly doses of 5 and 7.5 mg/kg in case of no or positive ADA reaction (titer >64)

Results for the OS Cox regression univariate analyses are provided in Supplementary Table 2. Other baseline factors that found to be significant predictors of OS were baseline KPS and histology. Baseline tumor size and change in MPF post-treatment at last assessment showed trend of correlation but did not reach statistical significance level. A better baseline performance status is associated with longer OS. Patients with the epithelial histology had longer OS than those with mixed histology. To take into account the effect of confounding baseline characteristics on OS, multivariate cox regression analysis was performed. The statistically significant covariates (KPS, C_min and histology) from the univariate analysis were added simultaneously, and results are presented in Table 3. Amatuximab exposure (C_min) was a significant predictor of OS after taking into account KPS status and histology. In this model, OS increases with better KPS, a higher C_min, as well as with epithelial histology.

Table 3.

Results of multivariate Cox regression analysis for OS in MPM patients receiving amatuximab in combination with pemetrexed and cisplatin

Predictor	Wald P value	Hazard ratio (95 % CI)
MPM patients treated with amatuximab in combination with pemetrexed and cisplatin (N = 84)
KPS	0.00028	0.948 (0.921–0.976)
Cmin	0.00036	0.987 (0.979–0.994)
Histology (epithelial vs. mixed)	0.01400	2.680 (1.223–5.875)
MPM patients treated with amatuximab in combination with pemetrexed and cisplatin, who completed at least four cycles of treatment (N = 53)
KPS	0.0087	0.951 (0.916–0.987)
Cmin	0.02	0.989 (0.979–0.998)
Histology (epithelial vs. mixed)	0.032	2.754 (1.093–6.943)

To be able to predict primary end point (OS) of any future trial, a parametric survival model was also developed following univariate Cox regression analysis. For the parametric model, Weibull distribution was selected based on minimum −2 * log-likelihood values for different OS time distribution. Similar to multivariate Cox regression analysis for OS, amatuximab exposure (C_min) was a significant predictor of OS after taking into account KPS status and histology. Parameter estimates, with %RSE, 95 % confidence interval and Wald P value for the significant predictors in the final OS model for the amatuximab are given in Supplementary Table 2.

Exposure–response relationship: progression-free survival

Kaplan–Meier time-to-event analysis for PFS was performed, with patients stratified according to median of amatuximab C_min and C_max. PFS and amatuximab exposure data were available from 76 patients, with median C_min and C_max value of 32.9 and 140 μg/mL, respectively (Supplementary Table 1). Compared to C_max, C_min showed a more significant effect on PFS (log-rank test P = 0.00223 vs. P = < 0.0001) (Fig. 3c, d). Median PFS for patients achieving C_min concentration above 32.9 μg/mL was 238 days (90 % CI 193–322). For patients achieving C_min concentration of less than or equal to 32.9 μg/mL, median PFS was 115 days (90 % CI 94–165). Results for the PFS Cox regression univariate analyses are provided in supplementary Table 2. None of the baseline characteristics affected PFS. However, change in the CA125 post-treatment value at the patient’s last assessment was also a significant predictor for PFS.

Sensitivity analysis/effect of early dropout on exposure–response analysis for OS

Amatuximab has a half-life of approximately 9–10 days [13]. PK data from study 003 indicated that four cycles of drug treatment were required to achieve PK steady state. As shown in Supplementary Figure 4, terminations which occurred prior to completing four cycles of treatment will have lower overall serum exposures compared to terminations occurring after four cycles of treatment simply as a result of these patients not reaching their steady-state PK levels. Kaplan–Meier plots stratified by median C_min in this subset of data (OS, n = 59; PFS, n = 53) showed the same trend of exposure–response relationship (Supplementary Figure 5) although the level of significance dropped. Multivariate Cox regression analysis using this subset of data showed no statistically significant correlation between PFS and exposure (data not shown). However, for OS, a statistically significant effect of KPS, histology and amatuximab C_min was confirmed (Table 3).

Exposure–response relationship: adverse events

Frequency of treatment-emergent adverse events (CTC Grade 3 or higher) was compared in the groups with maximum amatuximab pre-infusion concentrations ≥108 and <108 μg/mL. The value of 108 μg/mL represents the upper quartile of the distribution of the maximum amatuximab pre-infusion concentrations for each subject. No difference in the frequency was observed between the two groups (data not shown).

Simulations and dose evaluations

Simulations of PK profiles showed that given amatuximab dose of 5 mg/kg weekly in case of no ADA reaction will achieve a median C_min of 83.1 μg/mL and ADA reaction with tier >64 will drop median steady-state C_min to 26.9 μg/mL. A weekly dose of 7.5 mg/kg is predicted to achieve a median C_min of 128 μg/mL, and ADA reaction with tier >64 will drop the median steady-state C_min to 45 μg/mL. Amatuximab weekly dose of 5 mg/kg is predicted to achieve median C_min concentration of 83.1 μg/mL. The median and range of steady-state C_min concentration observed in Study 003 were 38.2 μg/mL (0.1–136 μg/mL). The predicted steady-state C_min of 83.1 μg/mL is higher than the observed data. This is due to weekly dosing without interruption in these simulations, compared to 2 weekly dosing with 1-week interruption in study 003. Also, any interruption due to hypersensitivity reactions was not considered for these simulations.

Figure 3 presents percentage of patients that predicted to achieve amatuximab steady-state C_min above 38.2 μg/mL, given an amatuximab dose of either 5 or 7.5 mg/kg administered weekly, with and without ADA >64. Approximately 80 % of patients would be above a steady-state C_min of 38.2 μg/mL by 28 days when giving 5 mg/kg weekly. A dose of 7.5 mg/kg weekly will increase this percentage to approximately 90 % of patients being above 38.2 μg/mL. Parametric OS model that predicted OS is presented in Table 4.

Discussion

Amatuximab is a chimeric monoclonal antibody against mesothelin, a cell surface glycoprotein highly expressed in many cancers including malignant mesothelioma [4–9]. Study 003 was an open-label Phase II study initiated in patients with MPM. Amatuximab was added onto the standard therapy of pemetrexed and cisplatin. Compared to the historical data of the Phase III trial of pemetrexed plus cisplatin in MPM patients [2], no improvement in PFS was observed; however, there appeared to be an improvement in OS [14]. With an aim to understand this observation and to support future strategy for amatuximab development in MPM, population PK and PK/PD analyses were performed. This is the first report on the population PK profile of amatuximab in patients with cancer as well as the exposure–response relationship for amatuximab on OS and PFS in patients with unresectable MPM.

Amatuximab pharmacokinetics was best described by a two-compartment model, incorporating parallel linear and nonlinear (Michaelis–Menten) pathways for elimination from the central compartment. The typical parameter estimates for C_L, V₁ and V₂ were 0.0299 L/h, 3.89 L and 2.62, respectively. These estimates are consistent with previously reported PK for mAbs, reflecting a low clearance and limited tissue penetration of these compounds [15]. The pcVPC confirmed that the final PK model provides a good description of all the available data from Studies 001, 002, 003 and 102.

The nonlinear elimination pathway for amatuximab can be attributed to its interaction with its target, mesothelin, which is overexpressed in virtually all patients with MPM. When the mAb binds to the receptor, the antibody–antigen complex is internalized and degraded within the cell [19]. The covariates, baseline weight and ADA reaction with the titer >64 were significant covariates for V₁ or for C_L, respectively. Giving pemetrexed plus cisplatin concomitantly did not appear to influence C_L or V₁. The age, gender, race/ethnicity or baseline albumin level did not influence the pharmacokinetics of amatuximab.

As noted, there appeared to be an improvement in median OS in patients with MPM in Study 003. In the Cox regression and Kaplan–Meier survival analyses of the OS data, a better baseline KPS, an epithelial histology, and the amatuximab C_min were associated with a longer OS. For PFS, amatuximab C_min and change in CA125 post-treatment at last assessment were significant predictors. Based on the multivariate parametric survival model, amatuximab exposure was a significant predictor of OS even in the presence of an effect of baseline KPS score and histology.

Sensitivity analysis of exposure response relationship demonstrated that early dropout patients affect the relationship; however, the trend of higher exposure associated with better OS outcome was still supported and was confirmed when confounding factors affecting OS were taken into account. It should be noted that lower number of patients in subset of data could also be a contributing factor of loss of statistical power. Overall, the exposure response analyses supported the hypothesis that a longer OS is likely to be achieved with higher amatuximab exposures.

Results of the PK simulations showed that 5 mg/kg of amatuximab, administered weekly, will achieve median steady-state C_min of 83.1 μg/mL, and approximately, 80 % of patients would be above the C_min of 38.2 μg/mL by 28 days of 5 mg/kg weekly dosing. Increasing the dose to 7.5 mg/kg would increase the exposure, but the proposed strategy to administer amatuximab 5 mg/kg weekly in an attempt to reduce the occurrence of hypersensitivity reactions seems to give more benefit. Also, as a higher dose has not been tested previously, it might raise a potential safety concern. Thus, weekly administration of amatuximab, 5 mg/kg, in combination with the standard therapy of pemetrexed and cisplatin may lead to higher steady-state C_min which may lead to a statistically significant and clinically meaningful benefit in OS versus pemetrexed plus cisplatin alone in patients with unresectable MPM. These analyses supported the selection of the dose for an already initiated randomized and controlled safety and efficacy study of amatuximab in combination with the standard therapy of pemetrexed and cisplatin in patients with unresectable MPM (ClinicalTrials.gov, NCT02357147).

In summary, amatuximab PK was best described by a two-compartment model with parallel linear and nonlinear elimination with ADA increasing the elimination. In patients with MPM, higher amatuximab C_min was shown to be associated with longer OS than has been seen with chemotherapy alone, supporting evaluation of more frequent dosing in future trial.