Challenge 1: ensuring adequate drug supply to enable rapid first-in-human study and seamless transition to phase 2/3

Nirmatrelvir became a lead candidate following preclinical demonstration of potent antiviral activity, clean toxicity profile, and acceptable oral bioavailability. Uncertainties associated with its oral absorption presented challenges to developing strategies to enable evaluation of a wide dose range required to establish safety and pharmacokinetics in phase 1 (NCT04756531). To address these, in silico modeling and profiling results obtained with only hundreds of milligrams of compound enabled us to confidently choose crystalline nirmatrelvir for first-in-human (FIH) study. During the design phase, the goal was maximizing nirmatrelvir concentrations to increase confidence in clinical efficacy and preventing resistance. Accordingly, pharmacokinetic profiles and safety of nirmatrelvir alone and in combination with the pharmacokinetic enhancer (ritonavir) were assessed in phase 1. Ultimately the selected phase 2/3 dose ((NCT04960202) included ritonavir to allow adequate free exposure of nirmatrelvir to achieve clinical antiviral effect and prevent resistance. The latter became increasingly important as SARS-CoV-2 variants emerged.

To enable rapid FIH study progression, minor discovery synthesis process modifications were made to ensure multikilogram nirmatrelvir quantities were available to support early drug product development, regulatory requirements, and clinical studies. Because of the program’s lightspeed development³, the nirmatrelvir commercial synthesis (described in Challenge 2) was developed concurrently, and included crystallization process optimization to ensure material properties suitable for drug product were consistently obtained.

To support FIH study objectives (assess safety/tolerability and pharmacokinetics)⁶, suspension and tablet formulations were developed to provide flexibility for dose escalation and rapid transition to phase 2/3 studies. With limited drug substance quantities available to support early-stage development, predictive experimental and computational tools were leveraged to execute drug product stability, bioperformance, and manufacturability risk assessments. Early in development, parallel accelerated chemical and physical stability evaluations of multiple test formulations were performed as a formulation screening tool. Preliminary pharmacokinetic models were developed using preclinical data and in vitro characterizations of drug substances and drug product dosage forms to provide initial bioperformance risk assessments. Drug product manufacturability was evaluated by assessing key drug substance and formulation properties by a series of material-sparing characterization techniques, and assessments were compared with an extensive internal database.

Concurrent with the FIH study and using preliminary clinical data, population pharmacokinetic model construction expedited dose selection and informed phase 2/3 study designs. To help finalize dose selection for these trials, a relative bioavailability (rBA) study was conducted to bridge suspension bioperformance to a prototype tablet formulation at the end of the FIH study. Using rBA results and the population pharmacokinetic model, a twice-daily nirmatrelvir/ritonavir dose regimen (300 mg/100 mg) was selected for phase 2/3 study⁸.

Challenge 2: supply continuity of clinical and commercial drug substance

Through infrastructure established within the lightspeed paradigm, a fit-for-purpose synthetic route enabled from medicinal chemists’ synthesis experience allowed for early development and clinical continuity. In parallel, early engagement to identify a commercial synthesis protocol supported a rapid nirmatrelvir scale-up. The fit-for-purpose route was completed on a timeline that did not allow exploration of all possible routes. Some key issues that needed overcoming included improving synthetic convergency by avoiding use of protecting groups (potentially extensive procedures to add and remove chemical structures to temporarily mask the reactivity of parts of the molecule), identifying solid intermediates that would not pose undue burdens to handling and isolation, computer-modeling of mixing effects to support rapid and seamless transfer to large-scale manufacturing, and identifying reagents and solvents available in appropriate quantities. An added challenge was the effort expended within the Chemical Research & Development group on development and manufacture of several small molecule components of the mRNA lipid nanoparticle program. Accordingly, other programs were paused to allow requisite focus on both COVID-19 programs.

Nirmatrelvir (1 in Fig. 2) is a polypeptide lending itself to modular disconnection. Key to success of the lightspeed development was having these building blocks from existing commercial sources (2), previous development candidates (3), or from published literature (4)^12,13. Even with this head start, sourcing the necessary >100 metric ton (MT) quantities required robust chemistry development and used reagents and solvents more readily available at scale. These requirements were also important for synthetic steps required to convert the building blocks to nirmatrelvir (Fig. 2), which were designed to avoid use of reagents, such as Burgess reagent in Step 4 dehydration, 4-dimethylaminopyridine (DMAP) in Step 2 amidation, and cyclopentyl methyl ether and 2-methyltetrahydrofuran. Shown in Fig. 2 is a more convergent synthesis, use of well-behaved solid intermediates, and, in the case of intermediate 7, a telescoped process (a method that combines two or more reactions into a single isolation) for Steps 3 and 4 with isolation of the penultimate intermediate 8, the methyl tert-butyl ether (MTBE) solvate of nirmatrelvir (1). Selected solvents and reagents needed to be sustainable and available in volumes required for clinical and commercial supply needs. An initial version of this route used the lithium salt of intermediate 5 and alternate reagents for Step 2 amidation, including the use of DMAP. To achieve enhanced process intensification (see Challenge 4), the route was updated to include sodium salt (5) and amidation conditions shown in Fig. 2.

EDCl 1-ethyl-3-(3-dimethylaminopropyl)carbodiimide hydrochloride salt, HOPO 2-hydroxypyridine-N-oxide, IPAC isopropyl acetate, iPrOAc acetic acid isopropyl ester, MEK methyl ethyl ketone, NaCl sodium chloride, NaOH sodium hydroxide, NMM N-methylmorpholine, MsCl methanesulfonyl chloride, MTBE methyl tertiary-butyl ether, TEA triethylamine, THF tetrahydrofuran, TFAA trifluoroacetic anhydride.

Predictive science tools were applied throughout development of drug substance synthesis; the following examples are representative of those used. First, some early batches of Step 1, a 2-phase liquid−liquid system, suffered incomplete reaction conversions in certain reactor configurations. Computational fluid dynamic modeling explained these failures allowing for equipment and stir rate selection that rapidly resolved the issue across a range of reactor configurations, manufacturing sites, and batch scales. Second, the Step 5 crystallization process was optimized by statistical modeling to achieve parametric control of particle size distribution appropriate for drug product quality without the need of milling. The drug substance particle size target was consistently met with over 2000 consecutive batches at a variety of scales and equipment configurations across multiple global manufacturing plants. Finally, crystal structure prediction tools were applied to the nirmatrelvir drug substance to de-risk crystalline polymorph selection, allowing for enhanced understanding and confidence in selecting the thermodynamically most stable polymorph for development and commercialization.

The drug substance manufacture scale from first nirmatrelvir synthesis in July 2020 to EUA in December 2021 is shown in Fig. 1. The accelerated timeline and parallel execution of multiple activities account for the remarkable scale increase from first laboratory synthesis (7 mg) to EUA and global supply manufacture (~100,000 kg), an increase of >10 billion-fold.

An important component of the development process was determining commercial drug substance specifications, and the appropriate starting material specifications to meet them. Although standard protocol for any development program, the pace amplified the need for this approach for nirmatrelvir. Regulatory toxicology studies were therefore performed with drug substance batches of representative purity, anticipating that process-related impurities in these development batches may be present in the final drug substance specification.

The process shown in Fig. 2 was developed while keeping sustainability in mind by continuously improving yield and reducing solvent and reagent quantities. Process intensification efforts were necessary to make this synthesis viable as a long-term commercial process to support the 100 MT quantities of drug substance needed for a global pandemic. Table 1 shows utilization factor improvements made over a 15-month timespan. This reflects yield improvements in each drug substance synthesis step arising from a combination of reagent and solvent optimization based on process understanding and performance on scale, process optimization via study designs, computational modeling of mixing effects and kinetics to optimize reaction conversions. This improvement represents a substantial reduction in starting material input requirements.

Shortly after EUA, drug substance production substantially increased, from 6.7 MT manufactured within ~1 month after EUA to 44.6 MT within ~3 months. This scale and pace required multiple global sites and key strategic partners. The latter was used for manufacture of the 3 drug substance building blocks (3, 2, 4), which went to 8 distinct manufacturing sites for drug substance.

Challenge 3: accelerated commercial dosage-form design using predictive science to supply pivotal clinical studies with proposed commercial formulation

After completion of the FIH study, phase 2/3 pivotal trials were initiated within 2 months. The drug product team set a breakthrough goal to have the final commercial dosage form and manufacturing process ready to supply these clinical studies, which if successful, would eliminate a need for a pivotal bioequivalence study.

Drug product stability, bioperformance, and manufacturability were key elements in designing the potential commercial drug product dosage form. To meet the accelerated development timeline, predictive tools, including material-sparing experimental approaches and computational predictions, were leveraged for risk assessment and decision-making.

Adequate long-term drug product stability is paramount for commercial dosage form design. Early identification and selection of the most stable crystalline nirmatrelvir polymorph used crystal structure prediction tools and provided the basis for commercial dosage form design. Accelerated chemical and physical stability assessment approaches enabled rapid selection of commercial drug product formulation composition and packaging configuration prior to regulatory stability batch manufacture.

The initial prototype formulation from the rBA study was optimized for enhanced manufacturing robustness alongside adequate bioperformance. Continued pharmacokinetic model optimization provided key insights into the potential impact of drug substance and drug product characteristics on bioperformance, enabling rapid identification of key attributes and targeted development of a suitable control strategy for drug substance and drug product.

Given the lightspeed development, the scale-up of the drug product manufacturing batch process from development (<1 kg) to commercial scale (750 kg) with a limited drug substance supply was challenging. Timelines did not permit running processes to collect data and identify an appropriate operational space for drug product manufacture. Instead, state-of-the-art predictive computational and machine-learning models were used, which helped reduce risks in process development and subsequent scale-up to multiple commercial manufacturing sites. Nirmatrelvir film-coated tablets were manufactured using a dry granulation batch process comprising several key unit operations, which were each developed using computational tools shown in Fig. 3A−D. For blending, computational models investigated mixing behavior and blend homogeneity across development to commercial blender scales, providing vital understanding of mixing dynamics. Hybrid computational models combining mechanistic and machine-learning methods were developed to transfer the roller compaction process from development laboratories to commercial manufacturing sites. A suite of models was developed to optimize and scale-up the tablet compression process to predict key tablet attributes and virtually test and eliminate tablet fracture and breakage risk. Thermodynamic principles were used to determine suitable process conditions for commercial-scale coating. Simulations estimated mechanical stress on the tablets during coating to ensure physical integrity.

A Bin blending; B roller compaction; C tablet compression; D tablet coating. DEM discrete element method, FEA finite element analysis, GPU graphics processing unit, ML machine learning.

The process modeling of each of these unit operations was verified through rigorous quality testing of drug product batches manufactured at multiple sites and scales and was key to delivering Paxlovid global supplies.

Challenge 4: establishing a Paxlovid commercial supply chain

Rapidly scaling the manufacturing process to deliver millions of high-quality Paxlovid doses relied on the experience and capability within the network of manufacturing sites and key strategic partners. As the commercial manufacturing process and anticipated demand became clear, unprecedented agility and coordination among manufacturing site operations and supply chain professionals were needed to replan existing production across the network and to support Paxlovid supplies. Critical to these efforts were colleagues at those facilities and testing sites and across all supportive manufacturing functions, coupled with close collaboration with the research and development team.

Early in development, with lead times exceeding 7 months, scale-up within the existing manufacturing facility network was recognized as only part of the solution to supplying Paxlovid. Another critical consideration was production scale-out to multiple existing external suppliers and forging new supply relationships, which included all supply chain elements. With 25 supply nodes, including internal and partner sites across 10 countries, Paxlovid supply relies on a complex, global supply chain (Fig. 4). A component within this supply chain is ritonavir, which needs to be co-administered with each nirmatrelvir dose⁷. Although ritonavir is widely available from multiple suppliers, any appearance, size, shape, and physical and chemical specification variations limited the number of viable sources. Capacity and availability of sufficient quantities were challenging. Uncertainty in the initial forecasted volumes of finished, packed Paxlovid and varied regulatory status of each ritonavir source added to the complexity. With potential volumes of >100 million finished Paxlovid packs required by the end of 2022, a substantial increase in ritonavir supply was needed to avoid disrupting its existing global availability. As with all key materials sourced externally, incremental ritonavir production by suppliers was funded at risk to ensure sufficient stock availability for final packing operations.

EU European Union, ROW rest of world, UK United Kingdom, US United States.

Real-time packaged product stability was on the critical path to support the maximum duration of shelf life at launch. Supportive bulk stability and modeling tools were important to supplement the limited real-time data available. Packaging configurations that could be quickly scaled but also maximized ease of administration and patient compliance-focused technical solutions to a few options.

Paxlovid was launched in a patient pack configuration containing 5 blister cards per carton. With a 5-day dosing duration⁷, each treatment course contained 30 tablets within the patient pack. Because each blister card was configured with morning and evening doses, each with 2 nirmatrelvir 150-mg tablets and 1 ritonavir 100-mg tablet, each dose needed to be convenient and unambiguous. The limited time to have commercial-ready packaging also dictated the use of foil/foil blisters, which provide the most protective barrier material configuration. This meant that the primary pack would be opaque; it was therefore critical to ensure each cavity contained the correct tablet before lidding. The packaging equipment included a high-resolution camera inspection system, which utilizes an AI-driven algorithm to distinguish between the ritonavir and nirmatrelvir tablets and therefore, confirm that all tablets are in the correct position during high-speed packaging operations.

Before the US EUA for Paxlovid on December 22, 2021⁷, the first shipments were delivered for immediate availability upon this authorization. This would be repeated across the globe as other countries granted EUAs. In mid-November 2021, within weeks after announcing positive interim results in the pivotal EPIC-HR study, Pfizer and the Medicines Patent Pool (MPP), a United Nations-backed public health organization working to increase access to life-saving medicines for low- and middle-income countries (LMICs), announced a licensing agreement. This agreement enabled the MPP to grant sublicenses to other companies for the manufacture and sale of their generic nirmatrelvir equivalent with technical support from Pfizer. Combined with ritonavir, this generic equivalent can be supplied to 95 LMICs comprising ~53% of the world’s population.