Ocular Pharmacokinetics of Faricimab Following Intravitreal Administration in Patients With Retinal Disease

Abstract

Purpose: To characterize faricimab ocular and systemic pharmacokinetics (PK) in patients with neovascular age-related macular degeneration (nAMD) or diabetic macular edema (DME) and to assess the effect of faricimab ocular exposure on clinical endpoints.

Methods: A population PK (popPK) model was developed using pooled data from phase 1 to 3 studies in patients with nAMD/DME. The dataset included 1095 faricimab aqueous humor (AH) concentrations from 284 patients and 8372 faricimab plasma concentrations from 2246 patients.

Results: Following intravitreal administration, faricimab PK was accurately described by a linear three-compartment model with sequential vitreous humor (VH), AH, and plasma compartments. Faricimab VH elimination to AH is the slowest process, with an estimated half-life (t_1/2) of 7.5 days. Due to flip-flop kinetics, plasma, AH, and VH concentrations declined in parallel. Disease had no effect on faricimab PK. Plasma exposure was ∼6000-fold lower than VH exposure. Age, anti-drug antibodies, body weight, and sex statistically significantly influenced PK parameters but had no clinically meaningful effect on ocular and systemic exposure. VH t_1/2 alone could not explain faricimab dosing frequency. Exposure–response analyses showed similar gains in best-corrected visual acuity across faricimab exposure ranges and dosing regimens.

Conclusions: Faricimab ocular and systemic pharmacokinetics were quantified and accurately described by the popPK model, developed using a large dataset from patients with nAMD/DME. Exposure–response analyses suggest that faricimab phase 3 dosing algorithms are appropriate to select the most suitable dosing regimen.

Translational Relevance: The popPK analysis suggested that faricimab dosing frequency was influenced by several factors and not by VH t_1/2 alone.

Introduction

Neovascular age-related macular degeneration (nAMD) is a leading cause of blindness in adults 60 years of age and older, whereas diabetic retinopathy is a leading cause of avoidable vision impairment and blindness in patients with diabetes.¹ Upregulation of vascular endothelial growth factor (VEGF)-A is a key driver of vascular leakage leading to vision loss in both diseases.² Vision and anatomical improvements have previously been demonstrated with intravitreal (IVT) anti-VEGF agents, establishing anti-VEGF therapy as the standard of care for patients with nAMD and diabetic macular edema (DME).^3–6 However, early vision gains are often not maintained long term in real-world clinical practice. Regular treatment and frequent monitoring visits are often associated with a high burden on patients, clinicians, and healthcare systems, which can also lead to a risk of over- or undertreatment if fixed treatment intervals become too short or too long.⁷ Alternative dosing approaches for existing anti-VEGF agents, aimed at increasing treatment and monitoring intervals, have been used and evaluated, with variable visual results.⁸ Furthermore, selective VEGF neutralization alone does not address all potential targets or mechanisms underlying disease pathology.⁹ Therefore, targeting additional pathways beyond VEGF may offer new opportunities for novel treatments that increase durability, reduce treatment burden, and potentially improve patient outcomes in clinical practice compared with currently available therapies. 

Faricimab is the first humanized, bispecific, immunoglobulin G monoclonal antibody designed for intraocular use. Faricimab targets two distinct pathways by binding to angiopoietin-2 (Ang-2) and VEGF-A, whereas the fragment crystallizable (Fc) region of faricimab was engineered to reduce Fc-mediated effector functions and increase systemic clearance.¹⁰ The Ang-2 and tyrosine kinase with immunoglobulin-like and epidermal growth factor homology domains (Tie) signaling pathway is a key regulator of vascular stability in the retinal vasculature. Under physiological conditions, Ang-1 mediates endothelial cell survival and cell junction integrity via Tie2 receptors. In retinal vascular diseases, Ang-2 upregulation competitively inhibits Ang-1 binding to Tie2, thereby neutralizing the vasoprotective effects of the Ang-1 and Tie2 signaling pathway.¹¹ Therefore, dual pathway inhibition may synergistically promote vascular stability and improve outcomes beyond current anti-VEGF therapies.¹² 

Faricimab was studied in four phase 3 studies, which assessed the safety and efficacy of faricimab administered up to every 16 weeks (Q16W) in patients with nAMD (TENAYA and LUCERNE)¹³ and DME (YOSEMITE and RHINE).¹⁴ One-year results from TENAYA and LUCERNE demonstrated that faricimab up to Q16W resulted in noninferior vision gains compared with aflibercept every 8 weeks (Q8W)¹³ and offered more rapid improvement in anatomical outcomes during the head-to-head dosing period.¹⁵ Similarly, 1-year results from YOSEMITE and RHINE showed that faricimab Q8W or per a treat-and-extend (T&E)-based personalized treatment interval up to Q16W resulted in noninferior vision gains and greater anatomic improvements compared with aflibercept Q8W.¹⁴ In all of these trials, faricimab demonstrated extended durability, with approximately 50% of patients on Q16W dosing at 1 year.^13,14 Faricimab was also well tolerated, with an acceptable safety profile.^13,14 Two-year findings from TENAYA/LUCERNE and YOSEMITE/RHINE demonstrated that vision gains with faricimab remained similar to those achieved with aflibercept^16,17 and that the improvements in anatomic outcomes in YOSEMITE/RHINE for faricimab versus aflibercept were maintained.¹⁷ Furthermore, at year 2, approximately 80% of patients were on every 12 weeks (Q12W) dosing, and more than 60% of patients were on Q16W dosing.^16,17 These findings demonstrated the potential of faricimab to meaningfully extend the time between treatments with sustained efficacy. 

Here, we present the first comprehensive analyses of the ocular and systemic pharmacokinetics (PK) of faricimab in patients with nAMD or DME based on pooled data from the four registrational phase 3 trials and five additional phase 1 and 2 studies. We also evaluated the relationship between dosing frequency, change in best-corrected visual acuity (BCVA), and faricimab concentrations in the vitreous to further understand patient-specific factors influencing the durability of faricimab. 

Materials and Methods

Data Acquisition

All studies included in the population pharmacokinetics (popPK) analyses were conducted in accordance with the International Conference on Harmonisation E6 Guidelines for Good Clinical Practice and the principles of the Declaration of Helsinki, as well as the laws and regulations of the country in which the research was conducted. All patients enrolled provided informed consent in writing. Trial designs, patient characteristics, and efficacy and safety results have previously been reported.^{9,13,14,16–22} Study JP39844 is the only exception and has not been previously described. An overview of the JP39844 study is therefore provided in Supplementary Figure S1. Hence, only the data collected in the phase 3 studies (∼80% of the overall dataset) are briefly described hereafter. 

TENAYA (NCT03823287) and LUCERNE (NCT03823300) evaluated the efficacy, safety, durability, and PK of faricimab 6.0 mg, administered by IVT injection up to Q16W, based on protocol-defined disease activity assessments in treatment-naïve patients with nAMD. The study design and rationale for TENAYA and LUCERNE have previously been described,¹⁸ and the efficacy, durability, and safety data have been reported.^13,16 A total of 1329 patients were randomly assigned to faricimab up to Q16W (TENAYA, n = 334; LUCERNE, n = 331) or aflibercept Q8W (TENAYA, n = 337; LUCERNE, n = 327). The primary endpoint was the mean change in BCVA from baseline averaged over weeks 40, 44, and 48. Data up to the primary endpoint were included in the popPK analyses. 

YOSEMITE (NCT03622580) and RHINE (NCT03622593) evaluated the efficacy, safety, PK, and optimal treatment frequency of faricimab 6.0 mg, administered by IVT injection Q8W or by a T&E-based personalized treatment interval, with adjustable dosing up to Q16W depending on disease activity assessments in patients with DME. The study design and rationale for YOSEMITE and RHINE have previously been described,¹⁹ and the efficacy, durability, and safety data have been reported.^14,17 A total of 1891 patients with center-involving DME, secondary to diabetes (type 1 or 2), were randomly assigned to faricimab Q8W dosing (YOSEMITE, n = 315; RHINE, n = 317), faricimab T&E (YOSEMITE, n = 313; RHINE, n = 319), or aflibercept Q8W (YOSEMITE, n = 312; RHINE, n = 315). The primary endpoint was the mean change in BCVA averaged over weeks 48, 52, and 56. Data up to the primary endpoint were included in the present analyses. 

Plasma and Aqueous Humor Sampling and Analyses

Plasma samples were collected from all patients, and optional aqueous humor (AH) samples were collected from patients consenting to AH sampling at multiple time points during the studies for the determination of faricimab concentrations. Plasma samples for the detection of anti-faricimab antibodies (antidrug antibodies [ADAs]) were obtained from all patients. The details of sampling times for each study are provided in the Supplementary Material. The number of quantifiable samples by time window included in the dataset is summarized in Table 1. 

Table 1.

 

Number of Quantifiable AH and Plasma Samples Per Time Window

 

All available faricimab samples and ADAs were analyzed using fully validated assays. The concentration of free faricimab in AH and in plasma was measured using a colorimetric quantitative enzyme-linked immunosorbent assay (ELISA). The quantification range was 7.81 to 500 ng/mL in AH and 0.800 to 50.0 ng/mL in plasma. The AH and plasma assays used the same assay format, allowing for comparison of PK results in both matrices. ADAs were determined in tripotassium ethylenediaminetetraacetic acid plasma using a bridging ELISA method. 

PopPK Model Development Approach

Using a nonlinear mixed-effects modeling approach, all available AH and plasma data from the faricimab treatment arms of the nine studies were pooled and used to characterize the ocular and systemic PK of faricimab. All patients with at least one quantifiable faricimab concentration value associated with a documented dosing history were included. Observations that were below the limit of quantification (BLQ) were used in the analysis, and a likelihood-based method for handling BLQ observations was applied.²³ 

The model, as defined by Equations 1 to 3 below, was the starting model, which was developed before phase 3 results were available. This is a three-compartment catenary model (Fig. 1). Following the administration of faricimab into the vitreous humor (VH), linear elimination from VH to AH was observed. The elimination of faricimab from plasma was also observed to be linear. The volume of the VH compartment (V_VH) was fixed to the literature value (0.0045 L),²⁴ whereas all other parameters were estimated. The elimination rate constant from the VH compartment is represented by K_VH, whereas K_AH and K = CL/V_C are the elimination rate constants for the AH and plasma compartments, respectively (Equations 1–3)  

\begin{eqnarray} \frac{{dA( 1 )}}{{dt}} = - {{\rm{K}}_{{\rm{VH}}}}*A( 1 ) \end{eqnarray}

(1)

 

\begin{eqnarray} \frac{{dA( 2 )}}{{dt}} = {{\rm{K}}_{{\rm{VH}}}}*A( 1 ) - {{\rm{K}}_{{\rm{AH}}}}*A( 2 ) \end{eqnarray}

(2)

 

\begin{eqnarray} \frac{{dA( 3 )}}{{dt}} = {{\rm{K}}_{{\rm{AH}}}}*A( 2 ) - \frac{{{\rm{CL}}}}{{{{\rm{V}}_{\rm{C}}}}}*A( 3 ) \end{eqnarray}

(3)

 

In the above equations, A(1), A(2), and A(3) are the amount of faricimab in the VH, AH, and plasma, respectively. The faricimab concentrations in each compartment are given by the ratio of the amount (of faricimab) by the volume in the respective compartments. 

Figure 1.

 

Schematic of the model for ocular and systemic pharmacokinetics of faricimab. AH, aqueous humor; CL, clearance; IVT, intravitreal; k, elimination rate; k_AH, elimination rate constant for AH; k_VH, elimination rate constant for VH; V_A, volume of the AH compartment; V_C, volume of the plasma compartment; V_H, volume of the VH compartment; VH, vitreous humor.

Structural model refinement was driven by the data, and the adequacy of the model was assessed using various goodness-of-fit indicators (e.g., observed vs. predicted concentrations), the minimum objective function value (a measure of the likelihood of the model), and plausibility and precision of the parameter estimates. Interindividual variability in structural model parameters was estimated using a log-normal distribution for the random effects. The initial residual variability was a combined additive and proportional error model, with the random effect, separate for plasma and AH. 

Covariates, such as age, body weight, sex, race, ADAs, disease indication, renal impairment, and hepatic impairment, were selected based on scientific interest, mechanistic plausibility, and exploratory graphics and their influences on model parameters tested. These covariate–parameter relationships were all included in the full model using a multiplicative expression for covariates (using normalized power models for continuous covariates). Inferences regarding covariate effects and their clinical relevance were based on the resulting parameter estimates and measures of estimation precision. Effects within 10% of the null value were also excluded to arrive at a parsimonious model if such removal did not result in a significant increase of the objective function value. The adequacy of the final model and model parameter estimates was investigated with a visual predictive check.^25,26 

Relationship Between PK Parameters and Dosing Regimen

The relationships between faricimab VH PK and dosing regimen were explored using the phase 3 efficacy data. In trials with response-adjusted dosing, an exposure–response relationship using metrics such as area under the concentration curve (AUC) or trough concentrations at steady state may be biased and therefore not appropriate, as those metrics depend on the response-adjusted dose. Instead, the VH half-life (t_1/2) (t_1/2,KVH = log(2)/K_VH), which is independent of dose and dosing regimen and thereby independent of response, was used as an exposure metric. Using the flexible dosing arms in the nAMD and DME trials, binary logistic regression analyses were performed to assess the probability to extend the treatment regimen beyond every 4 weeks (Q4W) or Q8W as a function of VH t_1/2. The dosing regimen was defined by the dosing scheduled at week 52 in patients with DME and at week 48 in patients with nAMD. The analyses were performed by indication. The following generalized linear regression model was used for the logistic regression analysis:  

\begin{eqnarray*} logit( p ) = {a_0} + {b_0}\cdot PREDICTOR \end{eqnarray*}

 

Here, p is the probability to extend the treatment regimen beyond Q8W, logit is the logit function, and PREDICTOR is VH t_1/2. To define confidence intervals (CIs) for the logistic regression function, 1000 bootstrap samples were drawn with replacement from the analysis population, and the logistic regression models were fitted to each of these samples. The effect of relevant covariates, including ADAs and patient baseline characteristics such as pigment epithelium detachment (PED) in nAMD and central subfield thickness (CST) and previous treatment in DME, on the logistic parameters was assessed as appropriate. 

Exposure–efficacy analyses were conducted to assess the relationship between BCVA change from baseline and VH t_1/2. In the phase 3 YOSEMITE and RHINE DME trials, the probability of dropping out of the studies was independent of exposure. Therefore, the 34 patients who did not complete the study were excluded from subsequent analyses. No patients dropped out of the phase 3 TENAYA and LUCERNE nAMD trials. 

Software

The popPK analysis was conducted using nonlinear mixed-effects modeling (NONMEM 7.5.0; ICON Development Solutions, Ellicott City, MD).²⁷ Graphical and all other statistical analyses, including evaluation of NONMEM outputs, were performed using R 4.0.2 for Windows (R Project for Statistical Computing, Vienna, Austria). 

Results

PopPK Analysis

A summary of the data included in faricimab popPK analyses is presented in Supplementary Table S1. The plots of concentration versus time after most recent dose are presented in Figure 2. Visual investigation of the plots indicated an exponential and parallel decay of both AH and plasma concentrations, with no apparent differences between patients with DME or nAMD. Characteristics of patients included in the popPK analyses are summarized in Table 2. 

Figure 2.

 

Individual faricimab concentrations versus time after most recent dose in patients with nAMD and DME (pooled) and patients with nAMD or DME. The asterisks (*) indicate below the limit of quantification; green circles, faricimab 0.5 mg; red circles, faricimab 1.5 mg; blue circles, faricimab 3.0 mg; black circles, faricimab 6.0 mg. DME, diabetic macular edema; nAMD, neovascular age-related macular degeneration.

Table 2.

 

Characteristics of Patients Included in the Population Pharmacokinetic Analyses (N = 2246)

 

PK Model

The starting model was the three-compartment catenary model (Fig. 1) developed using the phase 2 data and described by Equations 1 to 3, as previously mentioned. Interindividual random effects were included for all parameters except VH and AH volume. A full model quantifying the effects of relevant covariates on the PK parameters (i.e., weight, sex, age, disease type, and ADA) was subsequently developed. The final model was the parsimonious model that retained only effects that were statistically significant and clinically relevant. The parameters included in the final model are presented in Table 3. The kinetics of faricimab were characterized by slow release from the VH to AH, corresponding to a VH t_1/2 of 7.46 days. Due to flip-flop kinetics, the faricimab plasma concentration–time profile declined in parallel with the AH and VH concentration–time profiles. No faricimab accumulation was observed in the VH, AH, or plasma compartments, and steady state was reached by the end of the 12-week Q4W initiation dose period. The popPK estimate of the apparent plasma clearance (CL/F) was 2.33 L/day. 

Table 3.

 

Parameter Estimates for the Faricimab Population Pharmacokinetic Model

 

The covariates found to influence faricimab ocular PK parameters were age and presence of ADAs, whereas faricimab systemic PK parameters were influenced by sex and body weight (Fig. 3). A typical 44-year-old patient had a VH t_1/2 that was approximately 31% shorter than a typical 89-year-old patient. Results of the popPK analysis showed that the faricimab vitreous elimination rate constant was 30.4% faster in patients with treatment-emergent plasma ADAs, resulting in shorter vitreous t_1/2, although no changes in plasma clearance were observed. Faricimab plasma clearance was 13.7% slower in females. Plasma volume increased proportionally with body weight (with the power coefficient of 1.00; 95% CI, 0.795–1.21), exactly equal to the expected allometric increase. In nearly perfect agreement with allometric scaling, faricimab plasma clearance increased with weight with a power coefficient of 0.773 (95% CI, 0.699–0.847). As patients with DME were on average heavier (mean weight 86.8 kg vs. 75.2 kg for patients with nAMD), this translates to approximately 10% lower systemic exposure (steady-state AUC) in a typical patient with DME compared with a typical patient with nAMD. No other covariates (including race, patient disease characteristics at baseline, prior medication or treatment, fellow eye treatment, concomitant administration of drug lowering intraocular pressure, or hepatic or renal impairment) affected the faricimab PK parameters. Overall, the fixed-effect parameters and interindividual variability were estimated with good precision. For most parameters, the relative standard errors (RSEs) were below 13%. Only the correlations between the random effects of k_VH and k_AH, and k_AH and CL, had higher RSEs (29.0% and 21.1%, respectively). 

Figure 3.

 

Covariate effects on pharmacokinetic parameters. Covariate effects with 95% CIs are shown for subpopulations relative to a reference patient. The hatched areas represent typical values ± 20%. ADA, anti-drug antibody; AH, aqueous humor; CI, confidence interval; CL, clearance; DME, diabetic macular edema; k_AH, elimination rate constant for AH; k_VH, elimination rate constant for VH; nAMD, neovascular age-related macular degeneration; VC, volume of the plasma compartment; VH, vitreous humor.

Based on the visual predictive check plots, one can conclude that the final popPK model adequately captured both the central tendency and the interindividual variability of the faricimab PK in AH and in plasma (Supplementary Figs. S2–S4). Table 4 shows faricimab steady-state exposure estimates following 6-mg Q8W dosing. 

Table 4.

 

Faricimab Steady-State Exposure Estimates Following 6-mg Q8W Dosing in Phase 3 Studies

 

Analysis of Exposure–Durability Relationships

Phase 3 in nAMD

The probability of Q8W dosing at week 48 was 36.4% for the lowest tertile of predicted VH t_1/2 (3.22–6.78 days) and 8.1% for the highest tertile of predicted VH t_1/2 (8.96–17.1 days). This probability declined with increasing VH t_1/2 (Fig. 4, Supplementary Table S2). However, as shown in Supplementary Figure S5 (left panel), there was an important overlap of the distribution of VH t_1/2 across the different dosing regimens, suggesting that VH t_1/2 alone cannot accurately predict the faricimab dosing regimen. Indeed, baseline PED and ADA status also had significant effects on the probability of Q8W dosing (Supplementary Table S2). The probability of Q8W dosing was 4% lower for ADA-positive patients compared with ADA-negative patients, assuming a consistent VH t_1/2 and PED size. At typical VH t_1/2 with a PED size of 258 µm (mean observed PED in phase 3 trials of patients with nAMD), ADA-negative patients with no PED had a 13% higher probability of extending their treatment regimen beyond Q8W than ADA-negative patients. 

Figure 4.

 

Logistic regression: probability of Q8W dosing in patients with nAMD. The red solid line and green shaded area represent the logistic regression model prediction and 95% CI of predictions. Points show exposure of individual patients with events (p = 1) and without events (p = 0), vertically jittered for better visualization. Black squares and vertical green lines show the observed fractions of individuals with events in each exposure tertile and 95% CIs for these fractions. Dashed vertical lines show the bounds of exposure tertiles. The P value was determined by the glm() function. Q8W, every 8 weeks.

Phase 3 in DME

The probability of dropping out of the study was independent of exposure; therefore, the 34 patients who did not complete the study were excluded from subsequent analyses. The probability of Q4W or Q8W dosing at week 52 was 33.8% for the lowest tertile of predicted VH t_1/2 (2.97–6.06 days) and 20.8% for the highest tertile of predicted VH t_1/2 (7.91–15.3 days). This probability decreased with increasing VH t_1/2 (Fig. 5, Supplementary Table S3). However, as shown in Supplementary Figure S5 (right panel), there was an important overlap of the distribution of VH t_1/2 across the different dosing regimens, suggesting that the VH t_1/2 alone cannot predict accurately faricimab dosing regimen. Indeed, CST, previous treatment, and cataract surgery also had significant effects on the probability for Q4W or Q8W dosing (Supplementary Table S3). A longer VH t_1/2 and a lower CST at baseline decreased the probability of Q4W dosing (Fig. 5, Supplementary Table S3). The probability of extending the treatment regimen to Q12W or Q16W was higher with longer VH t_1/2 and low CST and in treatment-naïve patients. The probability of Q16W dosing was higher in patients with lower CST, longer VH t_1/2, and no history of cataract surgery. 

Figure 5.

 

Logistic regression: probability of Q4W or Q8W dosing in patients with DME. The red solid line and green shaded area represent the logistic regression model prediction and 95% CI of predictions. Points show exposure of individual patients with events (p = 1) and without events (p = 0), vertically jittered for better visualization. Black squares and vertical green lines show the observed fraction of individuals with events in each exposure tertile and 95% CIs for these fractions. Dashed vertical lines show the bounds of the exposure tertiles. The P value was determined by the glm() function.

Exposure–Response BCVA

Linear regression analyses demonstrated flat relationships between BCVA responses to faricimab treatment at weeks 40, 44, and 48 by VH t_1/2 for patients with nAMD and weeks 48, 52, and 56 by VH exposure for patients with DME (Figs. 6, 7). Similarly, flat relationships were established between CST responses to faricimab treatment and the VH t_1/2 at the endpoints at weeks 40, 44, and 48 in patients with nAMD and at weeks 48, 52, and 56 in patients with DME (Figs. 8, 9). 

Figure 6.

 

BCVA changes at weeks 40, 44, and 48 by VH t_1/2 in patients with nAMD in phase 3 studies. Individual values are plotted versus VH t_1/2. The red line shows the Lowess trend line. The blue line shows the linear regression line. BCVA, best-corrective visual acuity; t1/2, half-life.

Figure 7.

 

BCVA changes at weeks 48, 52, and 56 by VH t_1/2 in patients with DME in phase 3 studies. Individual values are plotted versus VH t_1/2. The red line shows the Lowess trend line. The blue line shows the linear regression line.

Figure 8.

 

CST changes at weeks 40, 44, and 48 by VH t_1/2 in patients with nAMD in phase 3 studies. Individual values are plotted versus VH t_1/2. The red line shows the Lowess trend line. The blue line shows the linear regression line. CST, central subfield thickness.

Figure 9.

 

CST changes at weeks 48, 52, and 56 by VH t_1/2 in patients with DME in phase 3 studies. Individual values are plotted versus VH t_1/2. The red line shows the Lowess trend line. The blue line shows the linear regression line.

Discussion

The popPK of faricimab, the first IVT bispecific antibody to independently bind and neutralize two key drivers of vascular instability, Ang-2 and VEGF-A, was characterized in AH and plasma following multiple IVT administrations to patients with nAMD and DME. Data were sourced from two phase 1, three phase 2, and four phase 3 studies. The PK dataset used for the current analyses is unprecedented with regard to the number of patients included (N = 2246), the large number of samples per patient (both plasma and AH samples), and by the consistency of bioanalytical methods used for the analysis. This high-quality dataset provided a representative sample of the target population, allowed for accurate characterization of the faricimab systemic and ocular PK, and facilitated exposure–response analysis where the PK were correlated with efficacy data obtained in the phase 3 trials. 

A data-driven approach with the principle of parsimony was used to develop the faricimab popPK model with the most appropriate minimal structure—namely, the dosing and physiological compartments (AH and plasma) from which the data were obtained. Therefore, it was hypothesized that, from the VH, the drug is transferred to the AH and then to plasma. Yet, it is most likely that, once injected into the VH, the drug will transfer in part to AH and in part to the retina, the site of action. This suggests that an additional retinal compartment could be added to the existing model, resulting in a more mechanistic model than the current model. However, as there are no data observed in the retinal compartment, it is not possible to differentiate the transfer rate from the VH to retina with that from the VH to the AH. This leads to a parameter identifiability issue. Hence, a model including the retinal compartment was not considered. A similar approach was used for the previously described PK model of IVT injection of ranibizumab,²⁸ which used a two-compartment PK model comprised of the VH and plasma (as AH samples were not available). 

The analyses indicated that faricimab concentration–time courses in AH and plasma are well described by a three-compartment catenary linear model. The ability of the model to capture both the central tendency and the interindividual variability of faricimab PK in AH and plasma was established. A comprehensive description of faricimab in the ocular compartment allows for reliable estimation of VH exposure, which is considered the optimal exposure estimate for assessment of exposure–response relationships because it represents exposure in the vicinity of the target organ. 

The popPK model was qualified with various diagnostic plots, including simulation-based diagnostic plots. The population fixed- and random-effect parameters were estimated with good precision, as suggested by their respective relative standard errors (most were <15% and two were between 20% and 30%). Therefore, the population parameter estimates were considered reliable. 

Following IVT administration, faricimab is cleared from the VH through the AH and then the plasma. Faricimab was engineered to abolish binding to the neonatal Fc receptor (FcRn) receptor, leading to a faster systemic elimination (estimated systemic t_1/2 of 0.44 days) compared with normal immunoglobulin G, for which the t_1/2 is approximately 25 days.²⁹ After IVT administration, flip-flop kinetics,³⁰ which were also reported for ranibizumab,²⁸ are observed, where the elimination of the drug from the body is determined by the slowest elimination rate from the VH to the AH. Consequently, plasma, AH, and VH concentrations declined in parallel, and the t_1/2 determined in the plasma reflected the VH t_1/2 (7.5 days for faricimab). Estimates of the ocular t_1/2 of several intravitreal biologics including aflibercept, ranibizumab, and brolucizumab have previously been reported.³¹ The systemic exposure to faricimab following IVT administration of 6 mg faricimab was low, and the maximum free faricimab plasma concentrations were approximately 600- and 6000-fold lower than in the AH and VH, respectively (Table 4). The popPK model described the single- and multiple-dose AH and plasma data. All transfer and elimination processes were first-order linear processes. After multiple administrations, there was no accumulation in the plasma, which is an important finding in the context of risk for systemic exposure-related adverse events. 

Consistent with allometric scaling, plasma volume and clearance increased with body weight as a power function with power coefficients of 1.00 (plasma volume) and 0.773 (clearance). The incidence rate of intraocular inflammation (IOI) was low and ≤3% in each of the disease indications.^13,14,16,17 The effects of faricimab exposure on the incidence of IOI were assessed with logistic regression models (data not shown). The models confirmed that the IOI incidence did not increase with faricimab VH exposure. Faricimab was also well tolerated across the broad plasma exposure range with a low incidence of systemic, non-ocular adverse events; therefore, differences in plasma exposure by body weight were not considered to be clinically meaningful, and no body weight–based dosing was needed. 

Age is a covariate affecting faricimab VH disposition, with longer VH t_1/2 in older patients. This increase in t_1/2 with age is consistent with age-related changes in the VH (fibrin and hyaluronic acid composition changes), leading to the vitreous contracting forward to the vitreous base and resulting in a condensation of the vitreous gel. Therefore, the vitreous becomes smaller but denser, and drugs may diffuse more slowly.^32,33 However, the association between age and VH disposition may be the result of various factors, such as age-related changes in metabolism, AH flow, or the extracellular matrix. As patients with DME/diabetic retinopathy were on average younger (mean age 62 years vs. 76 years for patients with nAMD), this translates to approximately 10% shorter VH elimination t_1/2 in a typical patient with DME (6.90 days) compared with a typical patient with nAMD (7.78 days). The resulting age-based difference in ocular exposure is considered not clinically meaningful in view of the flat exposure–BCVA correlation. 

ADA status was also identified as a statistically significant covariate. ADA-positive patients had a 30.4% higher ocular elimination rate compared with ADA-negative patients. There was a low incidence of ADA-positive patients (approximately 10%) in the phase 3 trials, and the effect on the vitreous exposure in ADA-positive patients was relatively small. The exposure–response analysis also demonstrated a similar response across the range of vitreous exposure in phase 3, confirming that the changes in vitreous exposure in ADA-positive patients are unlikely to be associated with a change in efficacy. Overall, the covariate findings were not considered clinically relevant, and no dose adjustment was deemed necessary. These results are also consistent with the findings previously reported for the intravitreal biologics ranibizumab and ranibizumab biosimilar SBll.³⁴ 

In trials with response-adjusted dosing, an exposure–response relationship using metrics such as AUC or trough concentrations at steady state may be biased and therefore not appropriate given that these metrics depend on the response-adjusted dosing. Instead, the patient-specific parameter VH t_1/2, where t_1/2,KVH = log(2)/K_VH, which is independent of dose and dosing regimen (and thereby independent of response), was used as a metric of exposure. Although drug retention in the VH is a factor contributing to durability, Supplementary Figure S5 shows that in both DME and nAMD the distribution of the VH t_1/2 overlaps greatly between the different dosing regimens, suggesting that the VH t_1/2 alone cannot accurately predict faricimab dosing regimens. Additional patient-specific and disease-related factors contribute to patient dosing regimens. Our analyses identified PED (for nAMD), baseline CST, cataract surgery, and previous treatment (for DME) as contributing factors to dosing frequency. In patients with nAMD, the probability of less frequent dosing decreased with increasing PED size at baseline. This can be expected, as PED size correlates with disease severity in nAMD. In patients with DME, the probability of extending treatment intervals decreased with increasing baseline CST and for previously treated patients. The latter may reflect previously treated patients having a longer disease duration and thereby more severe disease. Patients with DME who had previous cataract surgery also had a lower probability of extending their treatment regimen to Q16W. This suggests that the probability of less frequent dosing is a multifactorial function of several patient-specific and disease-specific factors, in addition to previous treatment. However, based on covariate analyses of the available dataset, when cataract surgery was tested as a covariate on PK parameters, such as drug t_1/2, no significant effects were detected (data not shown). The exposure–response analyses indicate that, despite the high variability in exposure in patients with nAMD and DME, there is no exposure-dependent difference in the change in BCVA at the endpoint analysis. This demonstrates that the T&E-based personalized treatment algorithm (designed to optimize treatment with the least number of injections possible) used in faricimab phase 3 trials is adequate to select the regimen for each patient. 

Herein, we have discussed the popPK of faricimab, a first-in-class bispecific antibody, which selectively targets and neutralizes the VEGF-A and Ang-2 pathways. Faricimab ocular and systemic exposure were accurately described by the popPK model, whereas exposure–response analyses showed that dosing frequency is a function of VH t_1/2 and patient characteristics at baseline. The analyses suggest that faricimab phase 3 dosing algorithms are appropriate to select the most suitable regimen for patients with nAMD or DME. 

Acknowledgments

The authors thank Karl Csaky, MD, and Robert Avery, DO, for their insightful comments and careful review of this manuscript. 

Supported by F. Hoffmann-La Roche Ltd., Basel, Switzerland. The sponsor participated in the study design; collection, analysis, and interpretation of data; writing of the report; and decision to submit the paper for publication. Funding was provided by F. Hoffmann-La Roche Ltd. for the study and third-party writing assistance, which was provided by Neil Norcross, PhD, and Luke Carey, PhD, CMPP, of Envision Pharma Group. 

Disclosure: C. Diack, F. Hoffmann-La Roche (E, I); L. Gibiansky, QuantPharm (E, C); F. Jaminion, F. Hoffmann-La Roche (E, I); E. Gibiansky, QuantPharm (E, C); J. Gaudreault, JJG Pharma Consulting (I, C); K. Bogman, F. Hoffmann-La Roche (E, I); V. Cosson, F. Hoffmann-La Roche (E, I)