Opportunities for using in silico‐based extended dosing regimens for monoclonal antibody immune checkpoint inhibitors

Abstract

Therapeutic drug monitoring (TDM) involves frequent measurements of drug concentrations to ensure levels remain within a therapeutic window, and it is especially useful for drugs with narrow therapeutic indices or extensive interindividual pharmacokinetic variability. This technique has never been applied to immuno‐oncology drugs, but, given recent examinations of clinical data (both exposure and response) on a number of these drugs, further investigations into TDM may be justified to reduce costs as well as potentially reducing the severity and/or duration of immune‐related adverse events. Specifically, all but one of the approved PD‐1 and PD‐L1 inhibitors (pembrolizumab, nivolumab, cemiplimab‐rwlc, atezolizumab, avelumab, durvalumab) have been shown to exhibit a plateaued exposure‐response (E‐R) curve at doses evaluated extensively to date, as well as time‐dependent changes in drug exposure. Furthermore, responders have a greater decrease in drug clearance over time and would, therefore, have supratherapeutic serum concentrations. With frequent trough measurements, it is possible to use pharmacokinetic modelling and simulation to estimate drug clearance via Bayesian methods. Based on patient‐specific estimates for clearance, optimal alternative dosing strategies can be simulated to lower drug and cost burden yet maintain therapeutic levels, especially as the clearance of the drug decreases over time. This review will comprehensively discuss each of the FDA approved PD‐1, PD‐L1/2 and CTLA‐4 inhibitors regarding their indications and current recommended dosing, with evidence supporting the investigation of these types of TDM strategies.

1. INTRODUCTION

The activation of the immune system to treat cancer has been a tremendous achievement in modern oncology. Over the past two decades, multiple “immuno‐oncology” (I‐O) therapies have received approval by the US FDA and European Medicines Agency (EMA). They encompass therapies ranging from chimeric antigen receptor (CAR) T cells, oncolytic viruses and cancer vaccines to chimeric, humanized or fully human engineered monoclonal antibody (mAb) drugs designed to bind a biological target (receptor). This review discusses the potential use of a variation of therapeutic drug monitoring (TDM) to optimize dose and schedule for one class of therapies, mAb drugs targeting immune checkpoints (immune checkpoint inhibitors, ICIs), with the goals of reducing both toxicities and cost.

The ICIs target T‐cell pathways as they aim to inhibit the ability of tumour cells to “override” the activation of T cells towards killing the tumour cell. This mechanism of T‐cell activation involves the binding of either the T‐cell receptor (TCR) to the major histocompatibility complex (MHC) on an antigen‐presenting cell (APC; eg, the tumour cell) or the T‐cell surface receptor (aka CD28) to CD80/CD86 (aka B7) on an APC. In addition, inhibitory pathways exist to deactivate the T‐cell response. Certain tumour cells upregulate receptors associated with these inhibitory pathways to avoid T cells from targeting and killing them. Therefore, designing drugs to target and block these inhibitory pathways allows the immune system (T cells) to target and destroy tumour cells.

One such inhibitory pathway is the binding of the cytotoxic T‐lymphocyte antigen‐4 (CTLA‐4) receptor found on T cells to B7 on APCs, thus preventing the binding of CD28 to B7. To date, ipilimumab, marketed as Yervoy by Bristol‐Myers Squibb, is the only drug approved for use that targets this pathway. The other inhibitory pathway involves binding of PD‐1 on T cells to its ligand PD‐L1 or PD‐L2 on APCs. The PD‐1 pathway is believed to act in concert with CTLA stimulation by regulating activated T cells in the later stages of an immune response. ¹ , ² Currently, there are six ICIs approved by the EMA, the US FDA or both that target and block the PD‐1/PD‐L1 pathway: three that bind PD‐1 (cemiplimab‐rwlc [Libtayo; Regeneron, Eastview, NY, USA], nivolumab [Opdivo; Bristol‐Myers Squibb, New York, NY, USA] and pembrolizumab [Keytruda; Merck, Kenilworth, NJ, USA]), and three that bind PD‐L1 (atezolizumab [Tecentriq; Genentech, San Francisco, CA, USA], avelumab [Bavencio; Pfizer, New York, NY, USA], and durvalumab [Imfinzi; AstraZeneca, Cambridge, England, UK]).

The clinical efficacy and safety of the ICIs were reviewed thoroughly by Heinzerling et al ³ and, even more recently, Centanni et al published an extensive review of the pharmacokinetics (PK) and pharmacodynamics of the ICIs, thus this review article will not recapitulate these topics. ⁴ Instead, the focus of this review will be to analyse the potential opportunity for TDM to optimize the dosing of ICIs that ultimately reduces both toxicities and cost. To support these suggestions into studying the potential use of TDM for these agents, arguments will be made that demonstrate that supratherapeutic steady‐state trough levels are achieved with standard dosing, and to provide evidence that lower doses and trough concentrations achieve comparable efficacy.

2. POTENTIAL OPPORTUNITIES FOR COST SAVINGS

TDM is a technique used to ensure that drug concentrations are in a therapeutic range and/or not a toxic range. While this is theoretically applicable to all drugs, TDM requires coordinated sample collection and rapid drug concentration measurement (therefore a reliable and rapid bioanalytical assay), mathematical analysis (but not always) and subsequent clinical dose adjustment from a coordinated team typically consisting of physicians, pharmacists, pharmacologists and chemists. As a result, TDM has traditionally been limited to drugs with narrow therapeutic indices from established E‐R relationships and extensive interindividual (but low intraindividual) PK variability. ⁵

In general, mAb‐based oncology drugs have not been considered optimal for traditional TDM, yet there is precedent for using this approach on biologics. ⁶ , ⁷ ICIs demonstrate extensive interindividual PK variability, ⁴ and most have commercially available ELISA assays ⁸ and established mechanisms of action (eg, binding to PD‐1 on T‐cells and blocking the interaction with PD‐L1 on tumours). ³ , ⁴ TDM, however, has yet to be applied to ICIs, most likely due to an apparent lack of relationship of exposure to either efficacy or safety over doses studied extensively. Leven et al provides a detailed review of the complex and unique E‐R relationships for each ICI drug in regards to melanoma, but these could conceivably be translated to any disease for which these drugs are effective. ⁹ These intricate relationships dictating the response will be examined more closely for the currently approved drugs, overlaid with details about the PK, and, ultimately, how a variation of TDM could improve the efficiency of dosing in later cycles and potentially reduce the financial toxicity.

As more clinical data are collected (exposure, safety and efficacy), it is becoming evident that TDM and subsequent pharmacometric analysis could provide optimal dose schedules for ICIs. In fact, several recent publications have suggested pharmacometric‐guided dosing, along with TDM, could not only better guide dosing, but also offer the added benefit of improving patient convenience, reducing costs and hypothetically reducing the duration and severity of immune‐related adverse events, ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ as well as improving patient convenience with less frequent clinical visits.

2.1. Anti‐CTLA

2.1.1. Ipilimumab

Ipilimumab was initially approved in the United States by the FDA in 2011. Currently, single‐agent ipilimumab is prescribed as either a 3 mg/kg (unresectable or metastatic melanoma) or a 10 mg/kg (adjuvant melanoma) infusion every 3 weeks (q3wk) for four doses. In the latter case, additional doses of 10 mg/kg every 12 weeks are recommended for up to 3 years pending recurrent disease. ¹⁵ Ipilimumab is also approved in combination therapy in metastatic renal cell carcinoma (RCC), microsatellite instability‐high or mismatch repair deficient metastatic colorectal cancer with nivolumab. Hamad et al demonstrated comparable efficacy of ipilimumab over a dose range of 3 to 10 mg/kg, with similar effects on peripheral T cells, and with lower toxicity at 3 mg/kg. ¹⁶ , ¹⁷ Furthermore, near‐saturation of target binding was determined in vitro at 20 μg/mL, which is achievable with 3 mg/kg q3wk. ¹⁸ Based on pooled data from patients given 0.3, 3 or 10 mg/kg, there was a clear E‐R relationship between steady‐state trough levels (C _MIN,SS) and overall survival (OS) over this wide dose range, with patients in the highest quartile of C _MIN,SS (82.1 μg/mL) having an almost 4‐fold prolonged OS (24.3 months) compared to patients in the lowest quartile (8.52 μg/mL; 6.51 months). ¹⁹ Due to interindividual variability in ipilimumab PK, the dose‐response relationship was not as apparent and, surprisingly, no time‐dependent changes in clearance were observed. ⁴ Based on the distinct exposure‐response relationship observed for ipilimumab that appears to be dosing sufficiently, the lack of time‐dependent changes in exposure and the added clinical benefit from combining with nivolumab, there is no apparent justification for evaluating alternative/extended dosing strategies.

2.2. Anti‐PD‐1

Preventing the interaction of PD‐1 to PD‐L1, by binding either PD‐1 or PD‐L1, results in the same mechanistic outcome – overriding the tumour's ability to evade T‐cell targeting. Although there are certainly differences among the drugs and their target binding (eg, nivolumab and pembrolizumab bind different regions of PD‐1 ²⁰ ), the ultimate mechanistic objective is to block the interaction between PD‐1 and PD‐L1. The molecular structures of the three anti‐PD‐1 inhibitors (cemiplimab‐rwlc, pembrolizumab and nivolumab) are all based on the immunoglobulin G4 family (IgG4) and are identical in their mechanism of action with very similar PK and PD. ⁹ , ²¹ , ²² Cemiplimab‐rwlc, pembrolizumab and nivolumab each demonstrate time‐dependent changes in PK, specifically the clearance pathways (target binding, protein catabolism, etc), to strikingly similar degrees. ⁴ , ²³ As such, for repeated dosing, exposure levels would gradually increase over time as clearance decreases, with the magnitude of change being greater in responders than nonresponders. ²⁴ , ²⁵ , ²⁶ , ²⁷

This inherent property of these IgG4‐based drugs provides an opportunity for cost savings through pharmacometric modelling and simulation (M&S) based on TDM‐measured trough concentrations. Specifically, it is conceivable to optimize doses based on each individual's unique change in clearance related to their clinical response to that drug. For example, if a particular patient has a measured trough lower than the median for that dose, then this concentration, along with the patient's unique covariates (body weight, age, albumin, etc), can be inputted into an appropriate population PK model. Based on the patient's prior dosing history, simulations can provide post hoc estimates for clearance via Bayesian methods. Due to the known relationship between baseline clearance and response for nivolumab ²⁸ and pembrolizumab, ²⁶ nonresponders can be identified earlier. Conversely, if a patient has a higher than median trough concentration, Bayesian methods can simulate an alternative dosing regimen to maintain levels above an established target concentration. In short, these drugs provide a tremendous opportunity to use PK‐guided in silico predictions to personally optimize alternative dose regimens in responders in subsequent cycles of therapy.

Ogungbenro et al recently conducted a comprehensive comparison of various methods of dosing pembrolizumab and nivolumab using pharmacokinetic modelling and simulation, including weight‐normalized (mg/kg) dosing, flat dosing and PK‐derived strategies. ²⁹ The study concluded that alternative dosing strategies using PK modelling and simulation offers the greatest waste reduction of drug vials, and therefore cost reduction (compared to dose banding or fixed dosing) relative to weight‐based dosing, while maintaining exposure and response. The use of PK‐guided in silico predictions could be used, in a manner similar to TDM, to personally optimize alternative dosing strategies for anti‐PD‐1 agents.

2.2.1. Nivolumab

Nivolumab was initially approved for advanced melanoma in December 2014 at a dose of 3 mg/kg q2wk based on the results of the CheckMate‐037 study. ³⁰ Currently, nivolumab is approved in 11 indications in various stages of melanoma, nonsmall cell lung cancer (NSCLC), RCC, hepatocellular carcinoma, head and neck squamous cell carcinoma (HNSCC), classical Hodgkin lymphoma, urothelial carcinoma and colorectal cancer. ³¹ Recent in silico modelling and simulation demonstrated comparable exposure with a flat 240 mg q2wk dose. ³² This in silico modelling and simulation strategy was extended to compare the exposure of 3 mg/kg q2wk to 480 mg q4wk. ³³ Bristol‐Myers Squibb currently offers three vial sizes of nivolumab: 40 mg, 100 mg and 240 mg, with the first two best suited for weight‐based dosing and the latter for flat/fixed dosing. The availability of three different size vials reduces the impact of vial wasting.

The PK of nivolumab has been well characterized, with distinctive covariates (patient variables) that influence the drug's clearance. There is roughly 30% variability in nivolumab clearance between subjects, some of which can be explained by baseline disease status, body weight, renal function, race and sex. ²⁷ , ²⁸ , ³¹ Patients who have a worse disease burden and a poor ECOG score, generally, clear nivolumab faster than patients with a smaller baseline tumour size and better ECOG score. ²⁷ , ²⁸ Also tied to a patient's overall health is their state of cachexia, which can manifest (among other symptoms) as lower albumin due to faster protein catabolism that is not due to simple malnutrition. ³⁴ This, in turn, also affects nivolumab (a protein) catabolism and hence results in a faster drug clearance. These same variables are associated not only with fast drug clearance, but also with lack of benefit, regardless of dose. ²⁸

While there are several factors that affect drug clearance, the most impactful variables in determining the magnitude of time‐dependent decrease in drug clearance is a patient's clinical response and baseline tumour size. ²⁸ Patients clinically scored as complete responders (CR) have a larger magnitude decrease in nivolumab clearance (~45%) as their tumour burden decreases and their overall health improves, compared to patients with a partial response (~35%), stable disease (~25%) or progressive disease (20%). ²⁸ A higher albumin level is associated with a greater percentage decrease in clearance due to the overall health status of the patient, but, interestingly, a larger baseline tumour size (which has faster baseline clearance) is also associated with a greater attenuation in clearance over time.

Given the known associations between patient variables and the mathematical impact on clearance, including changes over time, ²⁷ , ²⁸ measurements of nivolumab serum trough concentrations after one dose (2 or 4 weeks post dose) can be used to assess baseline drug clearance (CL₀), a suggested predictive marker for OS. ²⁸ , ³⁵ , ³⁶ By assessing CL₀ after a single dose, it is possible to more quickly identify responders from nonresponders, which would prevent wasted spending on an ineffective drug. There is reasonable evidence that patients with fast baseline clearance tend to poorly respond to nivolumab and ample evidence of nonresponders being given higher doses and still not responding. ²⁸ Additionally, for those responders who would remain on nivolumab therapy, trough measurements could be used for PK‐guided in silico predictions to optimize an alternative dose schedule as clearance decreases with time and disease improvement. It is hypothesized that a patient's unique alternative dose schedule could be optimized through PK modelling and simulation, thus not only maintaining a therapeutic concentration but also reducing patient and payer costs. By measuring trough concentrations following each dose/cycle, a patient's steady‐state clearance can be estimated using Bayesian post hoc methods to then ultimately adjust when the patient would receive their next dose. These simulations are based on maintaining serum trough concentrations deemed to be clinically effective, and there is convincing evidence that doses lower than the FDA recommended doses are effective.

In early clinical studies, doses as low as 0.1 mg/kg q2wk were demonstrated to be effective in patients with melanoma ³⁷ and RCC. ³⁸ , ³⁹ The median steady‐state trough concentration at 0.1 mg/kg q2wk is 2.5 μg/mL, ⁴⁰ which has been demonstrated to saturate the PD‐1 receptor comparably to 10 mg/kg. ³⁷ , ³⁸ Receptor occupancy, as measured in the (peripheral) blood, has some utility as a biomarker, but there is an incomplete understanding of the relationship between peripheral to tumoral receptor occupancy. ³⁸ In other disease indications, such as NSCLC, ≤1 mg/kg q2wk was suggested to be less effective than 3 mg/kg q2wk; however, there were relatively low numbers of a heterogenous population of patients studied at each dose level in most of these early dose‐finding nonrandomized trials. Therefore, there is reason to further investigate whether lower doses would also be just as effective in larger cohorts as 3 mg/kg q2wk. The 2015 randomized study demonstrating comparable clinical activity in RCC at 0.3 vs 2 vs 10 mg/kg q3wk ³⁹ was an important confirmation that low‐dose nivolumab is clinically effective. In short, there is published clinical evidence suggesting that low‐dose nivolumab has comparable efficacy to the current approved dose, supporting the development of treatment regimens utilizing lower doses or less frequent dosing.

The efficacy results (overall response rate; ORR) in melanoma and RCC could be translatable to other tumour types that failed to demonstrate efficacy due to statistically inadequate dose cohorts. Furthermore, subsequent M&S used to approve 480 mg q4wk flat dose by the FDA have demonstrated, from a totality of clinical evidence, consistent results across all tumour types ⁴¹ (FDA approval package for Opdivo from March 2018, obtained via Freedom of Information Act). This suggests that the previously observed exposure‐response relationship of nivolumab in NSCLC may be misleading and that the clinically effective low doses observed in RCC and melanoma are possible for NSCLC and others. In fact, once adjusted for small sample size and study effects (when comparing across different studies and study populations), there was no significant difference in the OS hazard ratio between 1 and 3 mg/kg in both squamous (Hazard Ratio (HR) 2.08 [95% CI: 0.866, 5]) and nonsquamous (HR 1.02 [95% CI: 0.819, 4.52]). ⁴¹

While not TDM by the traditional definition, it is suggested that PK (trough)‐guided in silico predictions can be used to more quickly determine responders and to optimize a less‐frequent dosing regimen in later cycles for responders who have a significant (40‐50%) time‐dependent decrease in clearance at steady state for subsequent cycles of therapy. In the case of nivolumab, because responders experience increases in exposure in later cycles due to time‐dependent decrease in clearance of ~50%, coupled with clinical evidence that lower doses demonstrate comparable efficacy, it is suggested that in silico‐predicted alternative/extended dosing regimens be explored. Such suggested studies would need to be properly controlled and to that end Ratain et al suggested the execution of randomized clinical trials to compare efficacy and steady‐state serum trough concentrations between standard dosing and in silico‐predicted alternative dosing schedules in a Bayesian noninferiority design. ¹¹ , ¹³ The target trough to maintain in the alternative dosing regimens is not explicitly known, but there is clinical evidence to suggest lower doses are just as effective as higher doses and the aforementioned Bayesian noninferiority trial can be powered to confirm a target trough to achieve. The endpoints of these initial studies would be to determine a valid trough concentration that maintains efficacy in a relatively small patient set (n < 72). If evidence of comparable efficacy is seen, then an additional phase II study could be performed to validate the efficacy of the trough‐guided, in silico‐predicted alternative dosing regimen for that drug.

2.2.2. Pembrolizumab

Pembrolizumab was initially approved in the United States in September 2014 and is currently indicated in 11 different disease settings and 17 unique settings, if various combinations with chemotherapy are included. ⁴² All adult indications now recommend a fixed dose of 200 mg q3wk after a 2017 study used PK modelling and simulation to demonstrate the comparable exposure with the previous FDA recommended dose of 2 mg/kg q3wk. ⁴³ A pharmacoeconomic analysis determined that personalized (weight‐based) dosing of 2 mg/kg would save US payers $825 million (in 2016 USD) annually, compared to fixed 200 mg, due to the lower amount administered (e.g. 160 mg for an 80‐kg adult). ⁴⁴ This estimation related only to patients with PD‐L1 positive lung cancer. However, a major assumption to this cost savings was 100% compliance in vial sharing, which is not always logistically feasible. Once a pembrolizumab vial is opened and diluted/formulated, it has a 6‐hour shelf life at room temperature and only 24‐hour shelf life when refrigerated. ⁴² Thus, any vial sharing would need to be performed between patients visiting the same clinic on the same day. Such coordination, while certainly possible, is unlikely to be actively employed at most clinics in the United States and is only allowed in specialized pharmacies that follow US Pharmacopeial Convention. ⁴⁵ Therefore, the $825 million in cost savings associated with weight‐based dosing is unlikely to be realized.

The PK of pembrolizumab are very similar to nivolumab regarding volume of distribution, half‐life and the time‐dependent nature of clearance. ²⁶ As with nivolumab, pembrolizumab clearance was, at baseline, shown to be correlated to response, time‐dependent and related to albumin (among other covariates). ²⁶ A lingering uncertainty, although largely studied, that should be acknowledged is the true target concentration, which needs to be maintained (steady‐state trough levels) to continuously saturate PD‐1 and induce a durable, robust clinical response. Patnaik et al demonstrated in the KEYNOTE‐001 study that 1 mg/kg dosing was sufficient to saturate receptor (PD‐1) occupancy, which was equal to that of higher doses (2 and 10 mg/kg) that were maintained for 21 days, while also demonstrating clinical activity. ⁴⁶ This study also suggested that a dose of 2 mg/kg was superior to 1 mg/kg, ⁴³ although this suggestion was based on a comparison of a small number of patients treated at different doses in a nonrandomized phase I study. Freshwater et al further demonstrated that 200 mg q3wk fixed dosing provides comparable exposure and efficacy as 2 mg/kg q3wk. ⁴³

It is well established that there is a flat E‐R for patients treated at 2 mg/kg q3wk or 10 mg/kg q3wk, indicating these doses are at the plateau of the E‐R curve. ⁴⁷ This strongly suggests lower exposures are likely still effective. TDM is therefore an ideal way to personally titrate pembrolizumab exposure over time based on alternative dosing schedules using the serum trough measurements of the preceding dose. While there are limited data available for patients treated with low‐dose pembrolizumab, the evidence that low‐dose nivolumab is effective ³⁷ , ³⁹ strongly suggests that similar doses of pembrolizumab are effective, given that the two drugs have been shown to be clinically interchangeable. ⁴⁸ While nivolumab and pembrolizumab have different PD‐1 binding sites and mechanism, ²⁰ they are considered interchangeable as drug therapies based on clinical efficacy and toxicity. ⁴⁸ Although not directly comparable due to differences in clinical trial criteria, the latest results of the KEYNOTE‐029 study, which combines pembrolizumab (2 mg/kg q3wk) with ipilimumab (1 mg/kg q3wk), further supports this similarity between the drugs as it reports overall response rates of 61% are superior to pembrolizumab alone ⁴⁹ and are virtually identical to response rates (61%) to nivolumab with ipilimumab (Nivo1 + Ipi3). ⁵⁰

2.2.3. Cemiplimab‐rwlc

Cemiplimab‐rwlc is the first therapeutic to be approved (September 2018) for the small subset of patients with cutaneous squamous cell carcinoma (CSCC), both locally advanced and metastatic, who are not eligible for curative surgery or radiation. ⁵¹ The drug is administered as a 350 mg fixed dose by infusion q3wk and is provided in a single vial size (350 mg, reconstitution in 7 mL; 50 mg/mL). The approval of cemiplimab‐rwlc was based on single‐arm trials that demonstrated nearly 50% objective response rates (ORR) to 3 mg/kg cemiplimab‐rwlc q2wk. ⁵²

There is very little information available on the population PK of cemiplimab‐rwlc, specifically published models. The drug has similar PK to nivolumab and pembrolizumab, including time‐dependent changes in clearance. ²³ The covariates included in the population analyses were body weight, albumin, body mass index, race, immunoglobuin G and alanine aminotransferase, ⁵³ but none were demonstrated to have a significant impact on cemiplimab‐rwlc exposure. ²³ As with nivolumab and pembrolizumab, there was a flat (ie, plateaued) E‐R relationship with efficacy over the range of doses studied (1–10 mg/kg q2wk) ⁵³ and, ultimately, a dose of 3 mg/kg q2wk or 350 mg q3wk was chosen. Both demonstrated comparable exposure via population pharmacokinetic modelling and simulation. ⁵³ Interestingly, the flat exposure‐ORR relationship was only studied over the steady‐state serum trough (C _MIN,SS) range of 10‐30 μg/mL, yet the C _MIN,SS from 3 mg/kg q2wk (65.7 μg/mL) and 350 mg q3wk (58.7 μg/mL) were both much higher than concentrations demonstrated to be effective. ⁵³

Based on the similar time‐dependent PK and flat E‐R relationship of cemiplimab‐rwlc, as with nivolumab and pembrolizumab, TDM is worth considering for similar reasons. Very few patients in the trials used for the FDA approval were dosed at 1 mg/kg q2wk, which yields a C _MIN,SS of 19.2 μg/mL and is well within the range observed to demonstrate a flat E‐R relationship with ORR, ⁵³ therefore it stands to reason that lower doses may be equally effective.

2.3. Anti‐PD‐L1/2

2.3.1. Atezolizumab

Atezolizumab, developed and marketed by Genentech, is an IgG1‐based anti‐PD‐L1 mAb, receiving accelerated approval by the FDA in 2016 initially for locally advanced or metastatic urothelial carcinoma at a dose of 1200 mg q3wk. ⁵⁴ Subsequent approval for NSCLC, triple‐negative breast cancer (TNBC) and small‐cell lung cancer (SCLC) was granted based on statistically significant improvements in duration of response, ORR or hazard ratios for OS or progression‐free survival (PFS). ⁵⁴ Initially, Genentech only offered one vial size (1200 mg), but later added a second size (840 mg) based on results of population PK modelling and simulation that demonstrated comparable exposure of 840 mg q2wk and 1680 mg q4wk with 1200 mg q3wk. ⁵⁵ The atezolizumab dosing label was recently modified (May 2019) based on these in silico findings to reflect these two alternative dosing options in order to provide patients with flexibility in their scheduled visits to the clinic.

The clinical PK of atezolizumab demonstrate a terminal half‐life of 27 days, a steady‐state volume of distribution of 6.9 L and a clearance of 0.20 L/day (29% Coefficient of Variation (CV) between patients) that decreases ~17% by steady state, which was determined to be clinically nonsignificant. ⁵⁴ , ⁵⁶ The atezolizumab E‐R relationship with ORR was flat over both the first‐dose C _MIN range of 18‐127 μg/mL ⁵⁷ and the AUC_SS range of 2000‐10 000 hr*μg/mL ⁵⁶ from the 1200 mg q3wk dose, which is not surprising given that Genentech determined that the target concentration to maintain efficacy (based on animal tissue distribution and receptor occupancy data) is 4‐6 μg/mL. ⁵⁷ , ⁵⁸ Deng et al used radiolabelled chimeric mouse analogue of atezolizumab in tumour‐bearing mice to depict tumour distribution, as well as receptor occupancy studies in both mice (given the murine chimeric analogue to atezolizumab) and monkeys (given atezolizumab) that demonstrated 96% PD‐L1 saturation at a concentration of 0.5 μg/mL in murine blood lymphocytes. ⁵⁸ While target binding in tumour is not always reflective of target binding in blood cells, it has been shown that PD‐L1 target saturation in nontumour bearing tissue can lead to favourable tumour uptake with sufficient exposure to drug. ⁵⁹ As such, a dose of 3 mg/kg maintained optimal tumour uptake for 120 hours in mice, with corresponding plasma concentrations between 2 and 8 μg/mL. Given that the plasma concentration previously deemed effective in a xenograft mouse model was translatable to humans for cetuximab, ⁶⁰ it is also likely true for atezolizumab that the active plasma concentration in mice is comparable to that in humans. Indeed, Genentech chose 6 μg/mL as the human target plasma concentration based not only on the aforementioned nonclinical tissue distribution data, but also on early clinical PK data, suggesting that 4 mg/kg q3wk would achieve this target in 90% of patients. ⁵⁸ To further support this, clinical activity was observed in doses ranging from 1 to 20 mg/kg q3wk. ⁶¹

A major opportunity for optimizing alternative dosing regimens has so far been overlooked, ¹² as these approved doses produce C _MIN,SS levels that are orders of magnitude higher ⁵⁷ than the manufacturer's target. Even a single dose of 3 mg/kg had a C _MIN of 12.2 μg/mL. ⁵⁷ There is enormous potential for TDM to drastically decrease the dose frequency of atezolizumab. As stated with the PD‐1 inhibitors, which also demonstrated time‐dependent PK and a relatively flat E‐R relationship, it is very likely that prolonging the dosing interval will still provide exposure above an effective concentration. Despite the lower magnitude decrease in clearance over time relative to the PD‐1 inhibitors (IgG1‐ vs IgG4‐based mAbs), there was comparable interindividual variability in the clearance (~30%). The benefits gained from decreasing a patient's dosing interval is even greater with atezolizumab than with the PD‐1 inhibitors due to the massive difference in median C _MIN,SS from approved doses (>100 μg/mL) and the known target concentration (6 μg/mL). This almost 20‐fold difference in C _MIN,SS can be viewed as an opportunity to use PK‐guided intermittent dosing of 1200 mg to provide about 5% (20‐fold lower) of the standard dose burden.

2.3.2. Avelumab

The second anti‐PD‐L1 agent approved by the FDA was avelumab, which was initially granted accelerated approval to treat Merkel cell carcinoma (MCC) in March 2017 based on the results of just 88 patients on a single‐arm trial with an ORR of 32%. ⁶² , ⁶³ Subsequent indications of urothelial carcinoma (9 May 2017) and RCC (14 May 2017) followed, each with the recommended dose of 10 mg/kg q2wk. Novakovic et al demonstrated comparable exposure between 10 mg/kg q2wk and flat 800 mg q2wk, ⁶⁴ which was the basis of the most recent Bavencio label change.

The clinical PK of avelumab are not typical of IgG1‐based mAbs, except with respect to volume of distribution (4.7 L), with a much shorter half‐life (4.5–6.1 days) and faster clearance (0.6 L/day) ⁶² , ⁶⁵ compared to atezolizumab. Faster clearance was associated with the male sex, MCC tumour type, decreased albumin, increased body weight and larger baseline tumour size. ⁶³ There is a significant time‐dependent decrease in clearance impacted by sex, race, C‐reactive protein, baseline weight and albumin, the magnitude of which is greater in responders (vs nonresponders) and in patients with MCC and HNSCC. ⁶⁵ Saturation of the PD‐L1 target (<90%) was achieved with doses ranging from 3 to 20 mg/kg. ⁶³ However, unlike the PD‐1 inhibitors and atezolizumab, avelumab demonstrates an E‐R relationship with best overall response (BOR) taken from 1629 patients. ⁶³ As C _MIN,SS increased from 12.5 to 25 μg/mL, the probability of achieving a response increased from 0% to 9%, or 25 to 37.5 μg/mL (achieved with doses of 10 and 20 mg/kg, respectively ⁶⁶ ). The probability then increased from 9% to 60% and plateaued thereafter. Interestingly, responders had slower clearance (both at baseline and steady state) compared to nonresponders among MCC patients, but the same was not true for urothelial carcinoma patients. Alternative dosing regimens may not be warranted for avelumab given there is sufficient evidence that the E‐R relationship was not plateaued and that the currently approved dosing regimen is likely optimized.

2.3.3. Durvalumab

Durvalumab was developed by AstraZeneca and received initial FDA approval in May 2017 for locally advanced or metastatic urothelial carcinoma and later in February 2018 for unresectable stage III NSCLC. ⁶⁷ The drug was approved in urothelial carcinoma based on safety and efficacy data from 182 patients in Study 1108 (NCT01693562), all given 10 mg/kg q2wk, where the ORR was 17% for all patients (n = 31/182), 26% for patients with high PD‐L1 expression (n = 25/95) and only 4% for those with low/negative PD‐L1 expression (n = 3/73). Of the small percentage of patients whose PD‐L1 expression levels were not evaluable (n = 14), 21% responded to treatment. ⁶⁷ The approval in NSCLC was based on improvements in both OS and PFS from 713 patients randomized (2:1, durvalumab:placebo), where 10 mg/kg q2wk durvalumab significantly improved the hazard ratios (0.68, P = 0.0025 and 0.52, P < 0.0001, for OS and PFS, respectively). ⁶⁷ In currently unpublished clinical data, a very recent press release reported preliminary results from the POSEIDON study (durvalumab + anti‐CTLA drug tremelimumab + chemotherapy) that demonstrated prolonged PFS in previously untreated stage IV NSCLC compared to chemotherapy alone. ⁶⁸

The recommended dose for both approved indications of 10 mg/kg q2wk provides a median C _MIN,SS of 91.9 μg/mL ⁶⁹ resulting from a steady‐state volume of distribution of 5.6 L, half‐life of 18 days and clearance of 8.2 mL/h (0.197 L/day), which is 23% slower than baseline clearance. ⁶⁷ , ⁷⁰ As with the PD‐1 inhibitors, albumin was correlated with the change in clearance, but there was no significant difference in exposure (first‐dose C _MAX, C _MIN or C _MIN,SS) between responders and nonresponders. ⁶⁹ Patients having >20 g/L of immunoglobulin G demonstrated 30% lower first‐dose AUC than patients having <20 g/L, ⁷⁰ comparable to atezolizumab.

It was recently shown through crystal structure‐based molecular modelling that atezolizumab and durvalumab have different binding to PD‐L1. ⁷¹ While there is potential for varying clinical efficacy and response based on binding kinetics and location on the target, this has yet to be definitively proven, unlike for nivolumab and pembrolizumab. ²⁰ , ⁴⁸ It is nonetheless possible that an effective durvalumab concentration is indeed different than for atezolizumab. As seen with avelumab, which demonstrated much faster clearance, a more frequent dosing schedule (q2wk) and amount (10 mg/kg) was needed to induce a clinical effect. The recommended durvalumab dose of 10 mg/kg q2wk (same as avelumab) provides a C _MIN,SS of 91.9 μg/mL, above the target ascribed to durvalumab (50 μg/mL). ⁶⁹ It is possible that lower doses of durvalumab would be effective, but there is not as wide a margin between achievable trough concentrations from approved doses and known target concentrations, as is the case for atezolizumab. The enormous potential for dose reduction and cost savings in atezolizumab could potentially make it the most cost‐effective, and thus most prescribed, anti‐PD‐L1 agent.

3. CONCLUSIONS

Several ICI drugs have evidence of efficacy at lower doses and also demonstrate decreased clearance at steady state that lead to higher exposures. For these reasons, additional investigations into using in silico modelling and simulation in Bayesian noninferiority studies are warranted for these agents (Table 1). By inputting trough concentrations into proven population PK models, simulations can be performed to estimate an individual's clearance to then guide their future dose/schedule. With the pattern of decreasing clearance over time, particularly with responders, and the wide margin between actual and minimally effective concentrations, a tremendous potential cost‐savings could ultimately be incurred.

TABLE 1.

Summary of drugs with the potential for in silico dose re‐optimization

Agent	Atezolizumab	Avelumab	Cemiplimab	Durvalumab	Ipilimumab	Pembrolizumab	Nivolumab
Dose range where efficacy studied (mg/kg)	1–20 54	1–20 62	1–10 23	0.1–20 69	0.3–10 15	1–10 42	0.1–10 31
Exp‐resp	Flat 54	Not flat 62	Flat; 10–30 μg/mL, C MIN,SS 23	Inconclusive 69	Not flat 15	Flat 42	Flat 31
Time‐dependent clearance?	Yes, but not clinically significant 54	Yes 62	Yes 23	Yes 69	No 15	Yes 42	Yes 31
Target conc known	Yes, 6 μg/mL 54	No	No	No	No	No	No
Dose re‐optimization warranted?	Yes	No	Yes	No	No	Yes	Yes

3.1. Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY.

COMPETING INTERESTS

CJP, DAG, RN and JCG declared no conflicts of interest and have nothing to disclose. MJR reports grants from AbbVie, personal fees from Aptevo, personal fees from Cyclacel, other from BeiGene, grants from Dicerna, personal fees from multiple generic pharmaceutical companies, grants from Genentech, personal fees from Pneuma Respiratory, outside the submitted work, and is Co‐founder and Director, Value in Cancer Care Consortium (Vi3C, www.vi3c.org). WDF declared no conflicts of interest and has nothing to disclose.

The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government.

Peer CJ, Goldstein DA, Goodell JC, Nguyen R, Figg WD, Ratain MJ. Opportunities for using in silico‐based extended dosing regimens for monoclonal antibody immune checkpoint inhibitors. Br J Clin Pharmacol. 2020;86:1769–1777. 10.1111/bcp.14369