Pharmacokinetic and Pharmacodynamic Effects of a γ‐Secretase Modulator, PF‐06648671, on CSF Amyloid‐β Peptides in Randomized Phase I Studies

Abstract

γ‐Secretase modulators (GSMs) represent a promising therapy for Alzheimer's disease by reducing pathogenic amyloid‐β (Aβ) peptide production. Three phase I studies (NCT02316756, NCT02407353, and NCT02440100) investigated the safety/tolerability, pharmacokinetics (PKs), and pharmacodynamics (PDs) of the oral GSM, PF‐06648671. A PK/PD indirect‐response model was developed (using biomarker data) to simultaneously characterize differential effects of PF‐06648671 on multiple Aβ species in cerebrospinal fluid (CSF). Healthy subjects (n = 120) received single doses or multiple‐ascending doses of PF‐06648671/placebo for 14 days. No serious adverse events occurred; severe adverse eventswere deemed not drug related. PF‐06648671 decreased Aβ42 and Aβ40 concentrations in CSF, with greater effects on Aβ42, and increased Aβ37 and Aβ38 levels, particularly Aβ37. No significant change in total Aβ was observed. The PK/PD model well described the tendency of observed CSF Aβ data and the steady‐state effects of PF‐06648671, supporting its use for predicting central Aβ effects and optimal dose selection for GSMs in future trials.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

☑ In Alzheimer's disease, γ‐secretase modulators (GSMs) have been identified as potential therapies; through modulating γ‐secretase activity, the balance of amyloid‐β (Aβ) species production can be altered to reduce Aβ‐associated pathogenicity. However, quantitative models predictive of changes in various Aβ species in human cerebrospinal fluid (CSF) with respect to GMSs are not available to date.

• WHAT QUESTION DID THIS STUDY ADDRESS?

☑ Using pharmacokinetics (PKs) and pharmacodynamics (PDs) data from three studies in healthy subjects, we aimed to quantitatively characterize differential effects of the GSM PF‐06648671 on CSF Aβ species simultaneously using a population PK/PD model.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

☑ PF‐06648671 decreased CSF levels of the more pathogenic Aβ42 and Aβ40 species and increased levels of less pathogenic Aβ37 and Aβ38. The PK/PD model adequately described observed data on multiple CSF Aβ species and predicted the effects of PF‐06648671 on CSF Aβ levels and ratios.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

☑ The PK/PD model developed here may be used to inform dose selection in future PF‐06648671 clinical trials.

Dementia is a progressive, neurodegenerative disorder affecting ~ 50 million people worldwide,1 of which Alzheimer's disease (AD) is most common.1 Treatments including cholinesterase inhibitors and N‐methyl‐d‐aspartate receptor antagonists help with AD symptoms.2 However, a medical need exists for agents capable of modifying and slowing AD progression.

AD pathology in the brain is characterized by senile, amyloid plaques comprising a core of amyloid‐β (Aβ) peptide fibrils,3, 4 and neurofibrillary tangles of primarily hyperphosphorylated tau protein.4, 5 Amyloid precursor protein (APP) is a transmembrane protein that is cleaved during processing by γ‐secretase to generate Aβ peptides of various lengths, including Aβ37, 38, 40, and 42.6, 7 Aβ42 has a higher aggregation tendency than the other peptides 8 and is a major component of amyloid plaques.9 Aggregated Aβ42 oligomers interact with synaptic receptors,10 causing nerve‐cell disruption and death.11 The shorter Aβ37 and Aβ38 species are believed to be less pathogenic than Aβ42.12 Increased Aβ42 production and an increased ratio of Aβ42:Aβ40 has been associated with an aggressive familial form of AD,13 and Aβ42:Aβ40 or Aβ42:Aβ38 ratios in cerebrospinal fluid (CSF) are useful biomarkers for AD pathology and predictors of AD.14, 15

Reducing Aβ42 and Aβ40 by targeting γ‐secretase is a promising therapeutic strategy, leading to the development of γ‐secretase inhibitors (GSIs) and, more recently, γ‐secretase modulators (GSMs). GSIs inhibit APP cleavage and Aβ40 and Aβ42 production16, 17 but are associated with detrimental effects on cognitive functioning and toxicities, including gastrointestinal bleeding, skin cancer, and immunosuppression, likely due to off‐target effects on other processes, such as Notch proteolysis.18, 19, 20, 21, 22, 23 Efforts have been made to design “Notch‐sparing” GSIs that are APP‐selective,24 but their safety has not been demonstrated. Moreover, GSIs can increase production of C‐terminal fragments of APP, potentially causing toxic effects on synaptic function and cognition.20, 25

GSMs modulate, rather than inhibit, γ‐secretase and shift Aβ production from Aβ42 and Aβ40 to Aβ37 and Aβ38, without altering overall Aβ levels.12, 26, 27 GSMs interact with γ‐secretase to increase “processivity” (successive cleavage) of Aβ, leading to these less pathogenic forms of Aβ.28, 29, 30 It has been demonstrated in vitro that the aggregation propensity of Aβ42 is greatly reduced by co‐incubation with Aβ37 and Aβ38, raising the possibility that shorter Aβ peptides may be beneficial in preventing or slowing amyloid plaque formation.30 Importantly, by modulating rather than inhibiting γ‐secretase, GSMs can avoid the mechanism‐based toxicity associated with GSIs.31

Several GSMs are in development;26, 30, 32, 33, 34, 35 to date, published clinical studies of one GSM have provided proof of the GSM mechanism in humans and have confirmed translation of pharmacologic activity from animal models to humans.32, 33 However, quantitative models predictive of Aβ changes in human CSF are yet to be published.

PF‐06648671 is a novel, brain‐penetrable, small‐molecule GSM and a potent modulator of γ‐secretase in vivo and in vitro. In cell‐based assays, PF‐06648671 reduced Aβ42 and Aβ40, with concomitant increases in Aβ37 and Aβ38, without inhibiting the cleavage of Notch or other substrates.36 Research in animals demonstrated reduced Aβ42 within the brain and CSF following acute oral administration of PF‐06648671.36 This change in the CSF Aβ profile represents mechanism‐relevant pharmacodynamic (PD) end points suitable for inclusion in translational research studies in humans.

This analysis utilized data from three phase I studies of PF‐06648671: two single‐dose studies, B7991001 (http://www.clinicaltrials.gov NCT02316756) and B7991003 (NCT02407353), and a multiple‐dose study, B7991002 (NCT02440100). The aim was to develop a pharmacokinetic (PK)/PD model to characterize the differential effects of PF‐06648671 on the various CSF Aβ species in healthy humans.

Results

Study populations

In B7991001, 18 subjects were randomized and received the study drug; all were white men, with a mean age of 35.3 years (range 22–55 years) and mean weight of 79.1 kg (range 52.8–104.0 kg). Subjects received PF‐06648671 doses of 2, 4, 12, and 40 mg in cohort 1, or 120, 240, 360 mg in cohort 2 (n = 6, except 4 mg, where n = 5); eight received placebo (Table 1). One subject discontinued from each of the 2 mg and 4 mg groups and was replaced. All subjects were analyzed for PK (except those receiving placebo); all were included in PD and safety analyses.

Table 1.

Overview of the PF‐06648671 phase I studies used in this analysis

B7991001: First‐in‐human (single‐ascending dose)
Design	Crossover, placebo controlled, placebo substitution
Subjects	Healthy
Data collection	PK, safety, tolerability, plasma biomarkers
PF‐06648671 doses (n = 5–8 per dose)
Cohort 1	2, 4, 12, 40 mg, placebo
Cohort 2	120,a 240, 360 mg, placebo
B7991003: CSF biomarker
Design	Single dose, parallel, placebo controlled, lumbar catheterization for serial CSF sampling
Subjects	Healthy
Data collection	PK, safety, tolerability, serial CSF sampling for time‐course CSF biomarker effects (≤36 hours)
PF‐06648671 doses	150 mg (n = 3), 300 mg (n = 11), placebo (n = 8)
B7991002: Multiple‐ascending doses
Design	Parallel group, placebo controlled, single LP for CSF sample collection
Subjects	Healthy (part 1) and healthy elderly (part 2)
Data collection	Safety, tolerability, PK, PD (trough CSF biomarkers) at:
	Day ‐3 (baseline) and predose on day 14 in one cohort of 40 mg q.d.
	24 Hours postdose on day 14 (in cohorts of 100, 200, and 360 mg q.d.) (steady state)
PF‐06648671 doses
Part 1	4, 12, 40, 40 (no LP), 100, 200, 360 mg (n = 8 for each), or placebo q.d. (n = 2) for 14 days
Part 2	200 mg (n = 8) or placebo (n = 2) q.d. for 14 days

CSF, cerebrospinal fluid; LP, lumbar puncture; PD, pharmacodynamic; PK, pharmacokinetic; q.d., once daily.

^aDosed in fed and fasted states.

In B7991003, 22 subjects were randomized and received single‐dose PF‐06648671 150 mg (n = 3), 300 mg (n = 11), and placebo (n = 8; Table 1). All were men, except one woman randomized to placebo. Eleven subjects were white, eight were black, one was Asian, and two were classified as “other.” Mean age was 36.1 years (range 21–55 years), mean weight was 76.0 kg (range 56.0–99.6 kg). All subjects completed the study, except one assigned placebo who was no longer willing to participate and was replaced. All subjects were analyzed for PK (except those receiving placebo), PD, and safety.

B7991002 comprised two parts: one in healthy subjects aged 18–55 years, and one in healthy elderly subjects aged 65–85 years. In part 1, 70 subjects were randomized; 8 were assigned to each PF‐06648671 dose, and 14 to placebo. The 40 mg q.d. dose was repeated in a second cohort without lumbar puncture (LP), to investigate the underlying cause for severe adverse events (AEs) experienced in three subjects in the first 40 mg q.d. cohort (details below). Six subjects discontinued, four due to AEs (details below). All subjects were men, 63 were white, 4 were black, and 3 were Asian; mean age was 32.2 years (range 18–55 years), and mean weight was 76.1 kg (range 52.4–107.2 kg). In part 2, 10 elderly subjects were randomized to PF‐06648671 200 mg q.d. (n = 8) or placebo (n = 2); 4 were women (all received PF‐06648671). Mean age was 68.0 years (range 65–75 years) and mean weight was 72.3 kg (range 54.9–91.3 kg); nine were white and one was Asian. All subjects completed part 2. All subjects in B7991002 were analyzed for PK (except those receiving placebo), PD, and safety.

Safety

In B7991001, 14 treatment‐emergent AEs were reported in 12 subjects in cohort 1 (9 treatment‐related), and 13 AEs in 11 subjects in cohort 2 (8 treatment‐related; Table S1 ). No deaths, serious AEs (SAEs), severe AEs, dose reductions, or temporary discontinuations due to AEs were reported; one subject discontinued due to a nontreatment‐related AE (pneumonia).

In B7991003, 14 subjects reported 32 AEs, of which 1 was considered treatment related (in the placebo group; Table S1 ). No deaths, SAEs, discontinuations, dose reductions, or temporary discontinuations due to AEs occurred. Two subjects receiving placebo had severe AEs of post‐LP syndrome and procedural pain; neither was deemed treatment related.

In B7991002, 177 AEs were reported in 65 subjects, of which 103 were considered treatment related (Table S1 ). No deaths, SAEs, or dose reductions due to AEs occurred. Severe AEs were reported in three subjects receiving PF‐06648671 40 mg q.d.: nausea and headache in two subjects (one discontinued the study, one temporarily discontinued the study drug), and postural dizziness and headache in one subject (the subject discontinued the study); all were considered related to LP rather than the study drug. Two additional subjects discontinued the study due to AEs: one due to mild hunger in the 40 mg q.d. group and one due to pregnancy of the subject's partner, which was reported during treatment in the 100 mg q.d. group; both were unrelated to the study drug.

The most common laboratory abnormality across all studies was low‐density lipoprotein cholesterol elevated >1.2‐times upper limit of normal; no abnormalities were considered clinically significant by investigators. Vital signs or electrocardiogram data abnormalities were infrequent and not considered clinically significant by investigators.

PK analysis

PF‐06648671 was rapidly absorbed following single and multiple doses; T_max, median time to reach maximal plasma concentration (C_max), was ≤1.5 hours across studies in the fasting state (Table S2 ).

In the single‐dose studies, plasma C_max and area under the concentration‐time curve from time 0 to infinity (AUC_0−inf) increased with dose (Table S2 ). Mean terminal half‐life (t_1/2) values were 12.8–23.1 hours. With food, median T_max was delayed by 3 hours, C_max decreased, and AUC_0−inf increased slightly (Table S2 ). PF‐06648671 plasma concentration over time in B7991003 is shown in Figure 1 a. In B7991003, the ratio of PF‐06648671 CSF exposure (measured as area under the concentration‐time curve from time of administration up to the time of the last quantifiable concentration (AUC_0−last)) to that in plasma (~ 0.003) was substantially lower than the free fraction in plasma (~ 0.013) and the ratio observed in B7991002 (~ 0.013; data not shown). This lower ratio was primarily attributed to nonspecific binding of PF‐06648671 to the long tubing used in serial CSF sample collection, with likely substantial drug loss, whereas no tubing was used in the B7991002 single‐sampling procedure. Therefore, PF‐06648671 CSF concentration was not used in PK/PD modeling.

Figure 1.

PK and PD plots from study B7991003 (serial CSF sampling biomarker study). (a) Plasma concentration of PF‐06648671 over time. (b–f) CSF concentration over time for (b) Aβ42, (c) Aβ40, (d) Aβ38, (e) Aβ37, and (f) total Aβ. Mean profiles are shown as bold lines, profiles from individual subjects are shown as thinner lines. Aβ, amyloid‐β; CSF, cerebrospinal fluid; PD, pharmacodynamic; PK, pharmacokinetic.

In B7991002, steady‐state plasma C_max and AUC from time 0 to time τ (AUC_0−τ, τ = 24 hours for q.d. dosing) increased with dose (Figure 2 a), with mean t_1/2 values of 12.0–20.8 hours across doses (Table S2 ). Observed accumulation ratios for AUC_0−τ and C_max were 1.2–1.8 on day 14. Following multiple ascending doses, plasma PK parameters in elderly subjects were generally similar to those in younger subjects (Table S2 ). PF‐06648671 concentrations in CSF of nonelderly subjects at day 14 ranged from mean (coefficient of variation) 3.49 ng/mL (81%) with 40 mg q.d., to 27.85 ng/mL (31%) with 360 mg q.d. The mean ratio of CSF:total plasma concentration ranged from 0.0124−0.0143, approximately equal to the unbound fraction in plasma (0.013).

Figure 2.

PK and PD plots from study B7991002 at days 0 and 14 (multiple‐ascending dose). (a) Plasma concentration of PF‐06648671 by dose. (b–f) CSF concentration by dose for (b) Aβ42, (c) Aβ40, (d) Aβ38, (e) Aβ37, and (f) total Aβ. Box plot shows median (bold horizontal line) as well as quartiles (25% and 75%), with whiskers to the last data point within 1.5 times the IQR. Black dots represent the outlier values (values outside 1.5 times the IQR above the 75% quartile and below the 25% quartile). Aβ, amyloid‐β; CSF, cerebrospinal fluid; IQR, interquartile range; PD, pharmacodynamic; PK, pharmacokinetic.

Supplementary text provides further details.

PD analysis

In B7991003, data from serial CSF sampling were highly variable (Figure 1 b–f), and a baseline drift of Aβ levels with placebo was observed. Nevertheless, dose‐dependent decreases in Aβ42, and to a lesser extent in Aβ40, were observed up to 36 hours postdose; dose‐dependent increases in Aβ37, and to a lesser extent Aβ38, were observed (Figure 1 b–e and Table S2 ). Total CSF Aβ levels in some subjects fluctuated substantially over the collection period (Figure 1 f). However, no difference was observed between PF‐06648671 and placebo groups. Total Aβ levels did not change significantly with either dose (Figure 1 f and Table S2 ). A delay between PK T_max and the time to maximum PD effect was noted (Figure 1 a vs. 1 b–e). Following single doses, no rebound effect of plasma Aβ levels over baseline was observed.

In B7991002 (in healthy young subjects), single CSF samples were obtained and analyzed at baseline and steady state in cohorts receiving placebo, 40, 100, 200, and 360 mg q.d. Robust dose‐dependent changes in Aβ were observed with a negligible placebo effect (Figure 2 b–e, 0 mg dose). Across all PF‐06648671 doses, placebo‐adjusted Aβ42 and Aβ40 levels decreased, with a commensurate dose‐dependent increase in CSF Aβ37 and Aβ38 levels (Table S2 ). No consistent changes in total Aβ levels were observed across doses (Figure 2 f and Table S2 ).

Supplementary Text provides further details.

PK/PD model

A PK/PD indirect‐response model was used to simultaneously characterize the differential effects of PF‐06648671 on multiple CSF Aβ species, including Aβ42, Aβ40, Aβ38, and Aβ37 (Figure 3). The placebo response was empirically characterized to account for the baseline drift observed over time; parameter estimates for placebo response are listed in Table S3 .

Figure 3.

PK/PD model schematic. γ, sigmoidicity constant; Aβ, amyloid‐β; Conc, plasma concentration of drug; CSF, cerebrospinal fluid; EC ₅₀, plasma concentration at half E_max; E_max, maximum response achievable; IC ₅₀, plasma concentration at half I_max; I_max, maximum inhibition achievable; k _in, zero‐order rate of Aβ production; k _out, first‐order turnover constant of Aβ; PD, pharmacodynamic; PK, pharmacokinetic.

Inhibitory effects of PF‐06648671 on CSF Aβ42 were greater than on Aβ40, demonstrated by higher maximum inhibition achievable (I_max) and smaller plasma concentration at half I_max (IC₅₀) values with Aβ42 vs. Aβ40 (Table 2). Also, PF‐06648671‐induced increases in Aβ37 levels were greater than those for Aβ38, demonstrated by higher maximum response achievable (E_max) and lower plasma concentration at half E_max (EC₅₀) values for Aβ37 (Table 2).

Table 2.

PD parameter estimates in the PK/PD indirect‐response model

Species	Parameters	Estimates	RSE, %
Aβ42	Baseline (pg/mL)	1,001	5.19
	k out (/hour)	0.121	14.8
	Imax	0.744	5.92
	IC50 (ng/mL)	421	28.1
Aβ40	Baseline (pg/mL)	7,584	4.52
	k out (/hour)	0.0835	15.3
	Imax	0.725	15.2
	IC50 (ng/mL)	1,228	38
Aβ38	Baseline (pg/mL)	2,404	4.22
	k out (/hour)	0.144	24.3
	Emax	0.869	87.2
	EC50 (ng/mL)	1,765	142
	Gamma (γ)	1.4	97.3
Aβ37	Baseline (pg/mL)	702	4.3
	k out (/hour)	0.0964	9.93
	Emax	5.63	21
	EC50 (ng/mL)	1,285	37
	Gamma (γ)	1.42	25.8

γ, sigmoidicity constant; Aβ, amyloid‐β; EC₅₀, plasma concentration at half E_max; E_max, maximum response achievable; IC₅₀, plasma concentration at half I_max; I_max, maximum inhibition achievable; k _out, first‐order turnover constant of Aβ; PD, pharmacodynamic; PK, pharmacokinetic; RSE, relative standard error.

Goodness‐of‐fit plots of observed Aβ concentrations vs. population‐predicted and individual‐predicted Aβ concentrations showed that the PK/PD model adequately described the observed Aβ species concentrations from single‐dose and multiple‐dose studies (Figure S2 ); no apparent bias or systematic trends were evident (Figure S3 ). Visual predictive check (VPC) revealed that the observed and simulated medians were in good agreement overall for multiple‐dose data (Figure 4), although variability tended to be higher in the simulation than in the observed data. VPC of single‐dose data also showed a congruence between the simulated and observed data for the central tendency; however, overestimation of variability was greater (data not shown).

Figure 4.

Visual predictive check for multiple‐dose data on day 14 of B7991002. (a) Aβ42. (b) Aβ40. (c) Aβ38. (d) Aβ37. Circles represent observed data; gray solid and dashed lines represent observed median and 90% percentiles; black lines represent the corresponding simulation percentiles. Baseline was not presented for better visualization of dose response. Aβ, amyloid‐β; CSF, cerebrospinal fluid.

Using the PK/PD model, the differential effects of PF‐06448671 at steady state on Aβ species in CSF were investigated. Reduction in Aβ42 was expected to plateau at ~70% at high doses studied. Although the estimated maximum reductions were similar for CSF Aβ42 and Aβ40 (74% and 73%), the doses required to achieve the same level of reduction were considerably higher for Aβ40 than Aβ42 due to higher estimated IC₅₀ for Aβ40 vs. Aβ42. For instance, it was predicted that an average 50% reduction in CSF Aβ42 would be achieved at PF‐06648671 75 mg q.d. (Figure 5 a), whereas PF‐06648671 250 mg q.d. would be required to reduce CSF Aβ40 to the same extent (Figure 5 b). In addition, an increase in CSF Aβ37 of up to nearly 500% was predicted within the simulated dose ranges of PF‐06648671 (Figure 5 c). Predicted steady‐state effects of PF‐06648671 on Aβ38 are not shown, due to large uncertainty in its parameter estimates, especially in EC₅₀ and γ (sigmoidicity; Table 2).

Figure 5.

Predicted average steady‐state effects of PF‐06648671 on CSF samples. (a) Aβ42 percent change from baseline. (b) Aβ40 percent change from baseline. (c) Aβ37 percent change from baseline. (d) Ratio of Aβ42:Aβ40. (e) Ratio of Aβ42:Aβ38.* (f) Ratio of Aβ42:Aβ37. Dotted line represents the median, with the shaded area representing the 95% CI. *95% CI for ratio of Aβ42:Aβ38 prediction was truncated due to missing values. Aβ, amyloid‐β; CI, confidence interval; CSF, cerebrospinal fluid.

The ratio of Aβ42:Aβ40 decreased maximally by 30% with PF‐06648671 between 150 and 200 mg q.d. (Figure 5 d). Similarly, reductions in Aβ42:Aβ37 and Aβ42:Aβ38 ratios were predicted to reach maximum between 150 and 200 mg q.d. (Figure 5 e, f). The median ratios expected at 200 mg q.d. were ~0.09, 0.10, and 0.11 vs. 0.13, 0.42, and 1.42 at 0 mg doses, for Aβ42:Aβ40, Aβ42:Aβ38, and Aβ42:Aβ37, respectively (data not shown).

A statement regarding data sharing is in the Supplementary Text .

Discussion

In three phase I studies of PF‐06648671, no significant drug‐related safety concerns were noted with single‐dose and multiple‐ascending daily doses for up to 14 days in healthy subjects, including elderly subjects. Severe AEs (headache, nausea, and dizziness) were considered to be secondary to the LP procedure rather than drug‐related. The most common AEs of headache, back pain, dizziness, and musculoskeletal discomfort in B7991002 were deemed in most cases to result from post‐LP syndrome. Symptoms resolved with supportive treatment in all subjects. No clinically significant laboratory and electrocardiogram abnormalities were observed.

PF‐06648671 was rapidly absorbed and plasma T_max was prolonged under fed conditions, consistent with other GSMs.32, 35 The t_1/2 of ~20 hours favored once‐daily dosing. The mean CSF:plasma concentration ratio at day 14 of multiple dosing was roughly equivalent to the unbound fraction in plasma, indicating that PF‐06648671 has excellent CSF penetration.

PF‐06648671 resulted in robust inhibition of CSF Aβ42 and Aβ40 production, and a commensurate increase in Aβ37 and Aβ38, with no net change in total CSF Aβ. These observations are consistent with the GSM mode of action. The greater reduction in CSF Aβ42 than Aβ40 following single and multiple doses of PF‐06648671 over the clinically relevant dose range led to the decrease in Aβ42:Aβ40 ratio, indicating that PF‐06648671 had a stronger inhibitory effect on Aβ42 than Aβ40. Similar to other GSMs, PF‐06648671 did not lead to the plasma Aβ‐rebound effect observed with GSIs.37

The placebo response in CSF Aβ observed in B7991003 was likely an artifact of a small CSF volume and frequent sampling, as observed with serial Aβ sampling from CSF elsewhere.33, 38 Alternatively, the drift could relate to the number of samples obtained over the study, with potential impact on intracranial pressure and/or CSF redistribution to lumbar subarachnoid space. Reducing the sampling frequency may minimize such effects.39 The placebo response was negligible in B7991002 using single LP collection.

The PK/PD indirect‐response model was developed to quantitatively and simultaneously characterize differential effects of PF‐06648671 on four forms of CSF Aβ, and total Aβ, and potentially provide mechanistic understanding of these effects. PF‐06648671 inhibited CSF Aβ42 and Aβ40 while increasing Aβ37 and Aβ38, with no effect on total Aβ. To our knowledge, this is the first PK/PD model using clinical data to describe differential effects of a GSM on multiple Aβ species. Although simultaneous characterization of all Aβ species is more challenging than using separate models for each species, it is advantageous as the model can account for correlations in baseline levels or drug‐induced responses between Aβ species. For example, a patient with high baseline CSF Aβ42 levels most likely has high levels of other Aβ species. Furthermore, a patient whose response is greater in one species is likely to have greater responses in other species. Therefore, our model allowed accurate prediction of drug effects not only on the individual Aβ species but also Aβ ratios of interest, and accounted for the observed delay between PK and PD responses in a semimechanistic manner. However, variability tended to be overpredicted because the number of subjects may have been insufficient for robust characterization of inter‐individual variability.

The PK/PD model predicted a 50% reduction in CSF Aβ42 levels with PF‐06648671 75 mg, but quadrupling the dose to 300 mg would only give a predicted reduction of 65%, in keeping with the observed data. The plateau observed in Figure 5 a suggests that dose levels approached the limit of the GSM mechanism in inhibiting Aβ42 production and that higher doses would not lead to greater inhibition. If Aβ42 inhibition or the ratio of Aβ42 to Aβ40 or other shorter species is important to elicit a clinical benefit with GSM (not yet proven), this model would inform optimal dosing regimens in future efficacy trials.

The reductions in Aβ42 and Aβ40, and increases in Aβ37 and Aβ38, correspond with other GSMs. Decreases in CSF Aβ42 and Aβ40 of up to 77% and 74%, and increases in CSF Aβ37 and Aβ38 of up to 545% and 69%, respectively, were reported with multiple‐ascending doses of BMS‐932481 over 28 days in healthy subjects.32 Single dosing of BMS‐932481 in healthy subjects led to a decrease of up to 50% in Aβ42 and up to 12‐fold and 2‐fold increases in Aβ37 and Aβ38, respectively.33 Another study using single‐ascending doses of E2212 reported a decrease in Aβ42 AUC_{0–24 hours} values of up to 44.1%.35

Our analysis has limitations and challenges. Data were obtained from healthy subjects, whose PD responses may differ from those of patients with AD. The number of subjects in each cohort was small and data variability was large. Study designs differed between single‐dose and multiple‐dose studies: rich (serial) PD samples from B7991003 provided information on CSF Aβ dynamics but did not provide a clear exposure–response relationship due to placebo baseline drift and large variability; steady‐state information on the PD response with more robust dose‐response relationships was obtained from B7991002 but without the Aβ time course. Another challenge was characterizing exposure‐response for Aβ38 due to its step like response; PD parameter estimates for Aβ38 were estimated with large uncertainty. Nevertheless, we believe a single PK/PD model developed from these pooled data provides valuable information and utility in designing future studies and selecting doses.

In conclusion, the GSM PF‐06648671 had a good safety profile at doses up to 360 mg q.d. for up to 14 days in healthy young and elderly subjects, was well absorbed, and could be administered once daily irrespective of food. PF‐06648671 resulted in robust, dose‐dependent reductions in CSF Aβ42 and Aβ40 and elevations in Aβ37 and Aβ38, with no effect on total Aβ. The PK/PD model adequately described the relationship between PF‐06648671 exposure and responses across multiple Aβ species and provided quantitative mechanistic understanding of the differential effects. This model could support dose selection and optimization in future PF‐06648671 clinical trials testing the therapeutic benefit of modulating levels of Aβ species in AD.

Methods

Compound

PF‐06648671 (2‐[(1S)‐1‐{(2S,5R)‐5‐[4‐chloro‐5‐fluoro‐2‐(trifluoromethyl)phenyl]tetrahydrofuran‐2‐yl}ethyl]‐7‐(4‐methyl‐1H‐imidazol‐1‐yl)‐3,4‐dihydro‐2H‐pyrido[1,2‐a]pyrazine‐1,6‐dione; Figure S1 ) was prepared using methods described in a Pfizer patent (WO2014045156, 2014; https://patentscope.wipo.int/search/en/detail.jsf?docId=WO2014045156%26recNum=262%26docAn=IB2013058347%26queryString=EN_ALL:nmr%20AND%20PA:pfizer%26maxRec=8201).

Study design

Studies were phase I, randomized, investigator blind and subject blind, sponsor open, and placebo controlled (Table 1). Primary and secondary objectives included safety/tolerability, PK characterization, and PF‐06448671 effects on Aβ species (PD).

B7991001 was conducted between December 2014 and March 2015 in healthy subjects aged 18–55 years, with a single‐ascending oral dose with placebo substitution, in a crossover design. Subjects received a single dose of PF‐06648671 or placebo after fasting (all doses) or after a high‐fat meal (120‐mg fed arm).

B7991003 was conducted between October 2015 and March 2016 in healthy subjects aged 18–55 years. A single oral dose of PF‐06648671 or placebo was administered after fasting.

B7991002 was conducted between May 2015 and October 2016 in healthy subjects aged 18–55 years (part 1) and healthy elderly subjects aged 65–85 years (part 2). PF‐06648671 was administered orally once daily for 14 days, after fasting (days 1, 7, and 14) or >1 hour before/after a standard breakfast (all other times).

A sample size of 8 subjects per cohort (B7991001), 30 subjects (B7991003), and 10 subjects per cohort (B7991002) aimed to minimize exposure of a new chemical entity yet provide adequate safety, tolerability, and PK data at each dose (B7991001, B7991002), or based on variability estimates from sponsor‐conducted studies (B7991003).

Before dosing, randomization numbers were allocated corresponding to a treatment schedule determined by a sponsor‐generated randomization code.

Studies were performed at Pfizer Clinical Research Units in Belgium (B7991001 and B7991002) and the United States (B7991003), in compliance with the ethical principles of the Declaration of Helsinki and International Conference on Harmonisation Good Clinical Practice guidelines. The protocol was approved by the Independent Ethics Committee at each unit. All subjects provided informed consent.

Safety

AEs, and their severity and potential causality, were recorded by investigators. SAEs were defined as any that resulted in death or were life‐threatening, required inpatient hospitalization or prolongation of existing hospitalization, or resulted in persistent or significant disability/incapacity or in congenital anomaly/birth defects. Laboratory evaluations, physical examinations, electrocardiogram monitoring, and vital sign assessments were performed.

PK/PD characterization

For plasma PK analysis, blood samples were collected predose and at regular intervals postdose on day 1 and once daily on days 2–4 (B7991001); predose and at regular intervals on day 1 and at 12–24‐hour intervals to day 4 (B7991003); and predose and at regular intervals postdose on days 1, 7, and 14, and once daily on days 2–6, 8–13, and 15–17 (B7991002).

In B7991003, serial CSF samples for PK and PD were collected 2 hours predose and at 0, 2, 4, 8, 12, 16, 20, 24, 28, and 36 hours postdose via PharMed® (Saint‐Gobain Biopharm, Aurora, OH) tubing connected to a catheter in each subject's lumbar. In B7991002, a single LP was performed 72 hours pre‐first dose and 24 hours post‐day 14 dose, or day 14 predose for those receiving 40 mg q.d.

Supplementary Text summarizes sample and statistical analyses.

PK/PD modeling

Analyses were performed using NONMEM version 7.3 (ICON Development Solutions, Gaithersburg, MD), with a first‐order conditional estimation with interaction method. PK was characterized by a two‐compartment model with linear elimination and first‐order absorption (Table S4 ). In estimating PD effects, PK data were included but population PK parameters were fixed (Table S4 ), using the population PK parameters and data approach.40

In B7991003, an empirical placebo model was developed for each Aβ species using placebo data, to account for baseline drift with placebo (Eq. (1)):

$${\text{Placebo}}_{i,j}=({\mathrm{\theta }}_{{\mathrm{P}}_{\max }}+{\eta }_i)·{\text{time}}_{i,j}^{{\mathrm{\theta }}_{K_p}}$$ (1)

(i, ith individual; j, jth time; θ, fixed‐effects parameter; P_max, coefficient describing magnitude of placebo response; η, random‐effects parameter; K _p, exponent of time to describe placebo response rate.) Placebo parameters (P_max and K _p), and interindividual variance in P_max, were fixed when treatment effects were estimated.

A PK/PD indirect‐response model was used to characterize effects of PF‐06648671 (Figure 3). CSF Aβ responses were described according to whether the production rate of Aβ species decreased (Aβ40 and Aβ42; Eq. (2)) or increased (Aβ37 and Aβ38; Eq. (3)) with respect to PF‐06648671 plasma concentration:

$$\frac{\mathrm{d}A\mathrm{\beta }42,40}{\text{dt}}=k_{\mathrm{in}}·\left(1-\frac{{\mathrm{I}}_{\max }·\text{Conc}}{{\text{IC}}_{50}+\text{Conc}}\right)-k_{\mathrm{out}}·A\mathrm{\beta }40,42$$ (2)

$$\frac{\mathrm{d}A\mathrm{\beta }38,37}{\text{dt}}=k_{\mathrm{in}}·\left(1+\frac{{\mathrm{E}}_{\max }·{\text{Conc}}^{\mathrm{\gamma }}}{{\text{EC}}_{50}^{\mathrm{\gamma }}+{\text{Conc}}^{\mathrm{\gamma }}}\right)-k_{\mathrm{out}}·A\mathrm{\beta }38,37$$ (3)

(Conc, individual predicted plasma PF‐06648671 concentration; E_max, maximum response achievable; EC₅₀, plasma concentration at half E_max; I_max, maximum inhibition achievable; IC₅₀, plasma concentration at half I_max; k _in, zero‐order rate of Aβ production; k _out, first‐order turnover rate constant of Aβ (thus, baseline = k _in/k _out); γ, sigmoidicity constant.)

Overall PD response was the sum of baseline, placebo (single‐dose PD data), and treatment effects. Unique PD parameters were estimated for each Aβ species.

Model evaluations

Concordance plots between observed and model‐predicted values were created to assess the model's goodness‐ of‐ fit. The model was evaluated by VPC, where 1,000 simulations were performed to check agreement between observed data and final model predictions (simulated data).

Model predictions

From the final PK/PD model parameter estimates and variance‐covariance matrix, 1,000 sets of population parameters were sampled assuming a multivariate normal distribution. Aβ responses from the sampled parameters were simulated; 2.5th and 97.5th percentiles were taken to construct 95% confidence intervals.

Funding

This study was funded by Pfizer Inc.

Conflicts of Interest

J.E.A., C.C., F.D.C., T.F., C.K., J.M., L.MdC., and R.Q. are employees of, and hold stock options for, Pfizer Inc. P.H., E.H.‐K., C.L., and R.L. were employees of, and held stock options for, Pfizer Inc at the time of the research.

Author Contributions

J.E.A., T.F., E.H.‐K., P.H., C.K., C.L., R.L., J.M., L.MdC., and R.Q. wrote the manuscript. T.F. E.H.‐K., P.H., C.L., R.L., and R.Q. designed the research. C.C., F.D.C., P.H., C.K., C.L., L.MdC., and R.Q. performed the research. J.E.A., T.F., E.H.‐K., P.H., C.L., R.L., J.M., and R.Q. analyzed the data.

Supporting information

Figure S1. Chemical structure of PF‐06648671: (2‐[(1S)‐1‐{(2S,5R)‐5‐[4‐chloro‐5‐fluoro‐2‐(trifluoromethyl)phenyl]tetrahydrofuran‐2‐yl}ethyl]‐7‐(4‐methyl‐1H‐imidazol‐1‐yl)‐3,4‐dihydro‐2H‐pyrido[1,2‐a]pyrazine‐1,6‐dione.

Figure S2. Goodness‐of‐fit plot for observed Aβ species concentration vs. (a–d) population‐predicted and (e–h) individual‐predicted Aβ species concentration following PF‐06648671 single (150, 300 mg; B7991003) or multiple (40, 100, 200, 360 mg q.d.; B7991002) dosing, using the PK/PD indirect‐response model. (a, e) Aβ42. (b, f) Aβ40. (c, g) Aβ38. (d, h) Aβ37. Solid line represents a line of identity, and dashed line represents loess smooth. Aβ, amyloid‐β; PK/PD, pharmacokinetic‐pharmacodynamic.

Figure S3. Conditional weighted residual plots. (a) Aβ42. (b) Aβ40. (c) Aβ38. (d) Aβ37. The dashed line represents loess smooth. Aβ, amyloid‐β; LOESS, locally weighted scatterplot smoothing.

Table S1. Overview of all causality (treatment‐related) treatment‐emergent adverse events.

Table S2. Summary of plasma PF‐06648671 PK parameter values, and CSF Aβ levels following PF‐06648671, across studies.

Table S3. Placebo model parameter estimates.

Table S4. Population PK parameter estimates.

Supplementary Text.