3.1 Summary of Patient Data

Results from 441 patients (79.6% female; 74.1% white), with a median age of 56 years (range 27–87 years) and a median body weight of 67 kg (range 32–129 kg) were included in the evaluation. Median CRCL was 97 mL/min (range 35–304 mL/min); renal function was normal (CRCL ≥ 90 mL/min) in 58.0% of patients, mildly impaired (CRCL 60–89 mL/min) in 32.9% of patients, and moderately impaired (CRCL 30–59 mL/min) in 8.8% of patients (no patients had severe impairment [CRCL 15–29 mL/min]). Hepatic function was normal (bilirubin ≤ upper limit of normal [ULN] and aspartate aminotransferase [AST] ≤ ULN) in 67.1% of patients, mildly impaired (bilirubin ≤ ULN and AST > ULN or bilirubin > 1.0–1.5 × ULN and AST of any value) in 31.3% of patients and moderately impaired (bilirubin > 1.5–3.0 × ULN and AST of any value) in 1.4% of patients (no patients had severe impairment [bilirubin > 3.0 × ULN and AST of any value]). Almost one-third (31.5%) of patients had concomitant use of an acid-reducing agent. Out of the 441 patients, 4522 capivasertib pharmacokinetic samples were reported. The analysis included 3963 observations with a median of six observations (range 2–21) per patient, while 559 observations were excluded for various reasons, such as being below the limit of quantification (141), being pre-treatment observations (411), missing observations (four) or having unexpectedly high values (three). Summary statistics of patient demographics and clinical covariates at baseline are provided in Table 2 (key covariates) and Table 2 of the ESM (other covariates). Figure 1 shows concentration–time profiles and the individual predictions for each study.

Table 2 Summary statistics for key covariates at baselineFig. 1

Individual capivasertib plasma concentration–time profiles after administration in the final model

3.2 PK of Capivasertib

The PK of capivasertib was adequately described by a three-compartment model with two parallel first-order and zero-order absorption mechanisms with lag time and with auto-inhibition of CL/F managed by a sigmoid Emax process (Fig. 2). The model was parameterised in terms of CL/F, inter-compartmental clearances of peripheral compartments (Q3/F and Q/F), V2/F, apparent volumes of distribution of peripheral compartments (V3/F and V4/F), Ka and absorption lag time (Lag). Capivasertib linear elimination was initially implemented from the central V2/F compartment. However, it was later identified that CL/F was time- and dose-dependent. An Emax exponential function was employed, defined by maximal inhibition of CL/F (Imax), the time at which inhibition of CL0/F is reduced by half and a Hill parameter fixed at five to account for this. An exponential model was used to describe the BSV except for the fraction of dose absorbed as first order (F1), which was described using an additive model. The residual variability was modelled using log-transformed data with a combined proportional and additive residual error. The BSV and the proportional residual error were expressed as a coefficient of variation.

Fig. 2

Capivasertib pharmacokinetic model. CL/F apparent clearance, CL0/F initial apparent clearance, D2 duration of zero-order absorption, F1 fraction absorbed as first order, Imax maximal inhibition of CL/F, Ka absorption rate constant, Lag1 lag time for absorption, Q3/F and Q4/F inter-compartmental clearances of peripheral compartments, T50 time at which inhibition of CL0/F is reduced by half, V2/F apparent volume of distribution of central compartment, V3/F and V4/F apparent volumes of distribution of peripheral compartments

As capivasertib was distributed into three compartments, three t½ were calculated corresponding to t½α, t½β and t½γ of 0.37, 4.2 and 31 h, respectively, for the first dose and of 0.42, 4.4 and 31 h, respectively, after multiple doses at 400 mg. For the administration schedule of 400 mg b.i.d. (4/3), the accumulation ratio was 1.58 and the effective t½ was 8.34 h [22]. Therefore, although it was predicted that the steady state would be reached on day 3 (after 41 h), owing to the time-dependent clearance during the first week of treatment and the intermittent (4/3) schedule, the steady state was predicted to be reached on every third and fourth dosing day each week from the second week at 400 mg b.i.d. The apparent volume of distribution at steady state was 255.6 L at 400 mg, suggesting that capivasertib distributes well in the body.

The absorption mechanisms component of the model estimated that 80% of the dose was absorbed using the first-order mechanism and the remaining 20% by a zero-order process where time to maximum concentration was reached at 1.4 h with slight variations depending on the dose level, body weight and food status. There was no Lag when capivasertib was administered orally as a tablet after an overnight fast. However, there was a short lag time when capivasertib was administered orally as a tablet or capsule under semi-fasted (no food intake from 2 h prior to 1 h after) or fed conditions, which was estimated to be 0.46 h for the capsule and 0.21 hours for the tablet. No other differences were detected between tablets and capsules. Diagnostic plots showed adequate fit to the data, with no apparent trends of residuals over time (data not shown).

3.3 Covariate Models

The model included significant covariates on the structural parameters of CL/F (body weight), absorption time lag ([ALAG], food effect and formulation), F1 (body weight) and Imax (dose and paclitaxel concomitant use), where the i suffix corresponds to the individual value for the ith individual. The equations that defined the structural parameters were as follows:

(a)

\(F_ = \frac\left( }} \right)^ }} + \eta LogitF_}}}} \right)^ }} + \eta LogitF_ }},\)

(b)

\(F_ = 1 - F_ ,\)

(c)

\(}_ = \left( }}} \times \left( }} \right) + Lag_}}} \times }} \right)\left( }} \right)e^}1}} ,\)

(d)

\(}_ /}_ = }_ /}\left( } - 67} \right) \times }0\_}} \right)e^}_ /}}} ,\)

(e)

\(}_ /}_ = }_ /}_ \left( }_ }} \times }^ }}^ + }^ }}} \right),\)

(f)

\(I_ = I_ \left( } - 480} \right) \times I}\_}} \right)\left( } \times I}\_}} \right)e^ }} ,\)

where CL0/F denotes the initial clearance after the first dose and CL/F refers to the time-varying CL managed by the sigmoid Emax function. Goodness-of-fit plots (Fig. 3) indicated a good fit of the model for most data where the observed capivasertib concentrations satisfactorily matched the predicted population concentrations or individual prediction concentrations. The observed-versus-predicted plots (Fig. 3a and b) showed random normal scatter around the identity line, indicating the absence of systematic bias and the adequacy of the model to describe the data. The conditional weighted residuals plots (Fig. 3c and d) showed random normal scatter around zero with no specific pattern to suggest model misspecification and a fairly constant distribution versus time with only eight observations of absolute value higher than six. The normalised prediction distribution error plots (Fig. 3e and f) showed a random normal scatter around zero. The correlation between parameter estimates in the correlation matrix lies between − 0.95 and 0.95 for all parameters. The estimated random-effect values were adequately centred around zero as the reported p values in the NONMEM output for CL/F, V2/F, F1, Imax and ALAG were larger than 0.05.

Fig. 3

Goodness-of-fit graphs for the capivasertib population pharmacokinetic model. (a) Observed versus population predicted capivasertib concentrations; (b) observed versus individual predicted capivasertib concentrations; (c) conditional weighted residuals (CWRES) error versus population predicted capivasertib concentrations; (d) CWRES error versus time; (e) normalised prediction distribution error (NPDE) versus population-predicted concentrations; and (f) NPDE versus time. The solid black line or red dashed line is the line of identity, and the blue line is the locally estimated scatterplot smoothing line. Outliers (|CWRES| > 6) IDs are identified in red in panels c and d

In the model, the between-subject random effects, CL/F, V3/F, F1 and Imax, demonstrated a centred distribution, whereas ALAG, which depended on food status, presented a more dispersed distribution. Shrinkages were moderate, by only 15% for CL/F, which was deemed adequate to generate individual predictions. No remarkable relationships were observed among between-subject random effects and structural parameters (data not shown). All other parameters were estimated with good precision, as relative standard errors of the parameter estimates were < 25% of the estimated value. The model parameters are described in Table 3.

Table 3 Parameter estimates and non-parametric bootstrap with 95% CI3.4 Model Assessment

The median values of parameters and 95% CIs obtained from the converged bootstrap runs for capivasertib are presented in Table 3. The median values of parameters were in close agreement with the population estimates in the model, suggesting that the NONMEM parameter estimates were unbiased. The model was evaluated by VPC and prediction-corrected VPC pooling all the studies, which showed adequate predictions (Figs. 1 and 2 of the ESM). Most observed concentrations were within the 95% prediction interval, indicating that the predicted variability did not exceed the observed variability.

3.5 Model Simulation

No differences in CL/F were detected among the dose levels after the first dose; however, a moderate dose-dependent and time-dependent CL/F was observed after multiple doses, as indicated by an initial CL/F of 62.2 L/h (BSV 39.3%), which decreased by 18%, 22% and 54% after approximately 120 h at 400, 480 and 800 mg (Fig. 4). The time at which inhibition of CL/F is reduced by half was 67 h. The median area under the curve for 12 h at steady state (AUC12h,ss) and the maximum concentration at steady state (Cmax,ss) for capivasertib were 7730 μg h/L and 1340 μg/L, respectively, at 400 mg b.i.d. (4/3) after multiple doses. Simulations for the three schedules explored at different dose levels show how the time-dependent inhibition and the dose levels modify the shape of the concentration–time profiles with steady state achieved after approximately 120 h (Fig. 5).

Fig. 4

Apparent clearance (CL/F) time course after twice-daily dosing of capivasertib. h hours

Fig. 5

Deterministic concentration–time profiles at different schedules and twice-daily dose levels (2/5), 2 days on/5 days off; (4/3) 4 days on/3 days off

Forest plots were used to assess the effect of covariates included in the model for AUC12h,ss, Cmax,ss, apparent clearance at steady state (CLss/F) and time to maximum concentration at steady state after multiple doses. The reference population had a body weight of 67 kg, with no paclitaxel usage, administered semi-fasted at 400 mg b.i.d (4/3) with tablets (Fig. 6 and Fig. 3 of the ESM). Capivasertib AUC12h,ss and Cmax,ss increased approximately proportionally between the dose levels of 80 mg and 480 mg after multiple doses but more than proportionally at doses beyond 480 mg (Fig. 6a and b). The CLss/F median ratio for a patient with a body weight of 47 kg (5th percentile) was 0.88 (95% CI 0.80–0.94), and with a body weight of 99 kg (95th percentile) was 1.20 (95% CI 1.09–1.31), against the 67-kg reference (Fig. 3a of the ESM). Patients with concomitant paclitaxel had 20% higher CLss/F (median ratio 1.20, 95% CI 1.12–1.29) [Fig. 3a of the ESM], and AUC12h,ss and Cmax,ss reduced by more than 10% (median ratio 0.84, 95% CI 0.78–0.89 and median ratio of 0.89, 95% CI 0.85–0.93, respectively; Fig. 6a and b) against the reference of no paclitaxel, after multiple doses.

Fig. 6

Forest plot for (a) area under the plasma concentration–time curve (AUC) and (b) maximum plasma concentration. (a) Median AUC for 12 hours at steady state (AUC12h,ss) (μg.h/L) and AUC12h,ss fraction relative to the reference (body weight of 67 kg, dosed at 400 mg twice daily with tablets, semi-fasted food status and no use of paclitaxel). (b) Median maximum concentration at steady state (Cmax,ss) [μg/L] and Cmax,ss fraction relative to the reference (body weight of 67 kg, dosed at capivasertib 400 mg twice daily with tablets, semi-fasted food status and no use of paclitaxel). CI confidence interval

3.6 In Vitro Cellular Kinase Inhibition

After establishing the pharmacokinetic model for capivasertib, the relationship between exposure and kinases potentially targeted by capivasertib was analysed. At the cellular level, capivasertib has the potential to inhibit a number of kinases in the same broad AGC kinase family of which AKT (Fig. 4 of the ESM) is a member including p70S6K (Fig. 5 of the ESM), PKA (Fig. 6 of the ESM) and ROCK (Fig. 7 of the ESM) [2]. Across multiple cell lines, capivasertib has greatest activity versus AKT1, 2 and 3 reducing phosphorylation of its direct and indirect substrates PRAS40 with IC50 values of 0.31–2.33 µM, GSK3β with IC50 values of 0.08–0.76 µM and S6 with IC50 values of 0.07–1.41 µM (Figs. 8 and Fig. 9 and Table 4 of the ESM). This suggests that within the clinical exposure achieved at steady state, capivasertib (400 mg b.i.d.; [4/3]) achieves robust cover over the IC50s required to inhibit AKT1, 2 and 3 signalling during the treatment periods.

Capivasertib has some activity versus p70S6K, PKA and ROCK in kinase assays [2] but potency in cells is reduced relative to the concentration at which it inhibits AKT. In TSC null cells, in which p70S6K is constitutively active, capivasertib reduced p70S6K-mediated phosphorylation of S6 with IC50 values of 0.67–2.52 µM (Table 5 of the ESM). Capivasertib was even less effective versus PKA reducing phosphorylation of its substrate VASP with IC50 values of 1.1–10.37 µM (Table 6 of the ESM), and it was not effective at inhibiting ROCK-mediated phosphorylation of its substrate cofilin (Table 7 of the ESM). Mapping these IC50 values onto the model-predicted capivasertib plasma concentrations indicates that capivasertib at 400 mg b.i.d. (4/3) does not achieve exposures to be effective versus these additional kinases.