Research Article

On-demand continuous-flow production of pharmaceuticals in a compact, reconfigurable system

See allHide authors and affiliations

Science  01 Apr 2016:
Vol. 352, Issue 6281, pp. 61-67
DOI: 10.1126/science.aaf1337

Drug manufacturing in a fridge-sized box

Commodity chemicals tend to be manufactured in a continuous fashion. However, the preparation of pharmaceuticals still proceeds batch by batch, partly on account of the complexity of their molecular structures. Adamo et al. now present an apparatus roughly the size of a household refrigerator that can synthesize and purify pharmaceuticals under continuous-flow conditions (see the Perspective by Martin). The integrated set of modules can produce hundreds to thousands of accumulated doses in a day, delivered in aqueous solution.

Science, this issue p. 61; see also p. 44


Pharmaceutical manufacturing typically uses batch processing at multiple locations. Disadvantages of this approach include long production times and the potential for supply chain disruptions. As a preliminary demonstration of an alternative approach, we report here the continuous-flow synthesis and formulation of active pharmaceutical ingredients in a compact, reconfigurable manufacturing platform. Continuous end-to-end synthesis in the refrigerator-sized [1.0 meter (width) × 0.7 meter (length) × 1.8 meter (height)] system produces sufficient quantities per day to supply hundreds to thousands of oral or topical liquid doses of diphenhydramine hydrochloride, lidocaine hydrochloride, diazepam, and fluoxetine hydrochloride that meet U.S. Pharmacopeia standards. Underlying this flexible plug-and-play approach are substantial enabling advances in continuous-flow synthesis, complex multistep sequence telescoping, reaction engineering equipment, and real-time formulation.

Whereas manufacturing of automobiles, electronics, petrochemicals, polymers, and food use an assembly-line and/or continuous, steady-state strategy, pharmaceutical synthesis remains one of the last industrial processes to apply a noncontinuous or “batch” approach. Moreover, pharmaceutical companies generally assemble the active pharmaceutical ingredient (API) using molecular fragments obtained from different sources, with the final synthesis steps done at the company location. The API is then often mixed with excipients and formulated in the final drug product form at a separate plant. As a result, production of a finished dosage form can require up to a total of 12 months, with large inventories of intermediates at several stages. This enormous space-time demand is one of a myriad of reasons that has led to increased interest in continuous manufacturing of APIs and drug products, as well as in the development of integrated processes that would manufacture the drug product from raw materials in a single end-to-end process (15).

Another major challenge facing the pharmaceutical industry is drug shortages; the U.S. Food and Drug Administration (FDA) has reported well over 200 cases per year during 2011–2014 (6). The root causes of these shortages often trace back to factors reflective of the limitations of batchwise manufacturing, such as variations in quality control and supply chain interruption. Moreover, the small number of suppliers for any particular medicine further exacerbates the challenges faced by batchwise manufacturing to respond to sudden changes in demand or need, such as in epidemic or pandemic instances of influenza outbreak.

To address the above issues, we have developed a continuous manufacturing platform that combines both synthesis and final drug product formulation into a single, highly compact unit (Fig. 1). The utilization of continuous flow (79) within the system enables efficient heat and mass transfer, as well as process intensification (10) and automation. Over the past several years, the merits of flow chemistry in streamlining synthesis (11) have been successfully demonstrated in the preparation of many individual high-profile APIs (12, 13), including artemisinin (14), imatinib (15), efavirenz (16), nabumetone (17), rufinamide (18), pregabalin (19), and (E/Z)-tamoxifen (20).

Fig. 1 Reconfigurable system for continuous production and formulation of APIs.

(A) Labeled photograph of the stack of upstream synthesis modules. (B) Labeled photograph of the downstream purification and formulation modules. (C) Close-up examples of upstream units; PFA tube flow reactors in an aluminum shell for heating (left) and membrane surface tension–based separation units (right). (D) Images of some of the main components in the downstream unit including the (a) buffer tank, (b) precipitation tank, (c) filtration unit, (d) crystallization unit, (e) filtration unit, (f) formulation tank, (g) solution holding tank, and (h) formulated API. Details are in the supplementary text.

Work with colleagues at the Massachusetts Institute of Technology (MIT) on end-to-end, continuous manufacturing of a single API, aliskiren hemifumarate, in a shipping container–sized unit (21) enabled us to identify critical steps in on-demand manufacturing of pharmaceuticals. Specifically, we chose to address challenges in reconfiguration for multiple synthesis of multiple compounds, tight integration of process streams for reduced footprint, innovations in chemical reaction and purification equipment, and compact systems for crystallization and formulation. As a result, the current system is ~1/40 the size and reconfigurable, in order to enable the on-demand synthesis and formulation of not just one, but many drug products. With the necessary regulatory approvals, this proof-of-principle system could enable a gradual phase-in of pharmaceutical production in response to demand. Reproduction of the system would be simpler and less costly to operate than a full batch plant and so could produce pharmaceuticals only needed for small patient populations or to meet humanitarian needs. It could be particularly advantageous for drugs with a short shelf life. Furthermore, the ability to manufacture the active ingredient on demand could reduce formulation complexity relative to tablets needing yearlong stability.

The flexible, plug-and-play refrigerator-sized platform (Fig. 1) [1.0 m (width) × 0.7 m (length) × 1.8 m (height), ∼100 kg] is capable of complex multistep synthesis, multiple in-line purifications, postsynthesis work-up and handling, semibatch crystallization, real-time process monitoring, and ultimately formulation of high-purity drug products. To demonstrate its capabilities, we produced, from raw materials, sufficient quantities to supply hundreds to thousands of consumable oral or topical liquid doses per day of four different pharmaceuticals: diphenhydramine hydrochloride (1), lidocaine hydrochloride (2), diazepam (3), and fluoxetine hydrochloride (4) (Fig. 2) (22). The latter API, fluoxetine hydrochloride (4), was synthesized as a racemic mixture, as approved by the FDA. These generic molecules from different drug classes have differing chemical structures and synthesis routes, thus challenging the capabilities and exploring the technical limits of the continuous-flow system. Moreover, they are ‎drugs commonly found in a chief medic’s toolkit. Diphenhydramine hydrochloride (1), for example, well known by the trade name Benadryl, is an ethanolamine-based antihistamine used to treat the common cold, lessen symptoms of allergies, and act as a mild sleep aid. Lidocaine hydrochloride (2), alternatively, is a common local anesthetic and class-1b antiarrhythmic drug. Diazepam (3), also known as Valium, is a central nervous system depressant. Finally, fluoxetine hydrochloride (4) is a widely used antidepressant recognized by its trade names Prozac and Sarafem.

Fig. 2 Reconfigurable modules and flowcharts for API synthesis.

(A) Diphenhydramine hydrochloride, (B) lidocaine hydrochloride, (C) diazepam, and (D) fluoxetine hydrochloride. The top row represents the different modules. Colored modules are active, gray boxes designate inactive modules. Reagent and solvent numbers refer to the compounds listed in table S2.

As shown in Figs. 3 to 5, the synthesis of each API utilizes simple starting materials and reagents readily available from commercial suppliers and highlighted advantages that flow chemistry offers relative to batch synthesis. Synthetic schemes were first developed in flow on a microliter scale before translating to the platform. The reactions leverage quick exposure at elevated temperatures (130° to 180°C) and pressures (~1.7 MPa) in controlled environments to enable faster reactions with low impurity profiles and reduce total synthesis times from hours to minutes. Reagents were in high concentrations, close to saturation, and in some cases even neat, which ensured high productivity while reducing waste and solvent amounts. This is in contrast to batch conditions that use lower concentrations, as solvents often also serve as a heat transfer medium. Moreover, in the flow system, reaction and purification occurred at the same time at different locations within the same uninterrupted reactor network. In batch, each of the operations would be physically and temporally disconnected and would have much larger time, space, and workforce requirements, hence drastically increasing the global footprint and decreasing the global output of a given process.

Fig. 3 Synthesis of diphenhydramine hydrochloride using the reconfigurable system.

Flowchart detailing the upstream and downstream synthesis.

Fig. 4 Synthesis of APIs via two-step upstream configurations.

(A) Lidocaine hydrochloride and (B) diazepam.

Fig. 5 Demonstration of a multistep API synthesis.

Flowchart detailing the upstream and downstream synthesis of fluoxetine hydrochloride. DMSO, dimethyl sulfoxide; rt, room temperature; DIBAL, diisobutylaluminum hydride.

Assembly of the platform

The system consists of reconfigurable upstream and downstream units (Fig. 1) that, despite having many complex operations, can be managed easily by an individual user. This is unlike typical batch manufacturing, which requires many operators to oversee multiple large-scale reactors and tanks with volumes on the order of thousands of liters and the transport and formulation of the final API in a separate processing plant (23). As shown in Fig. 1A, the upstream unit houses reaction-based equipment for producing APIs (e.g., feeds, pumps, reactors, separators, and pressure regulators) and has a maximum power requirement of 1.5 kW, which is mainly consumed by heating the reactors and operating the pumps. The backside (in Fig. 1A), alternatively, represents the downstream unit (Fig. 1B) dedicated to purification and formulation of the drug product (e.g., tanks to precipitate the crude API from reaction mixtures, crystallizers, and filters) (Fig. 1D). Temperature, pressure, flow, and level sensors are included at strategic positions and coupled with data acquisition units to facilitate operational monitoring and support real-time production control. Because few commercial chemically compatible components were available and suitable for the gram-per-hour size scale combined with elevated temperatures and pressures, we developed most of the unit operations used in the upstream and downstream systems, as detailed in the supplementary text. These include pressure sensors, clamshell reactors with an outer aluminum body, and inner PFA (perfluoro alkoxy polymer) tubing for chemical compatibility with good heat transfer (Fig. 1C and fig. S5), surface tension liquid-liquid–driven extraction units (24) (Fig. 1C), multiline back pressure regulators (fig. S3), automated precipitation, filtration (Fig. 1D and fig. S6), and crystallization tanks, and automated formulation (Fig. 1D and figs. S7 and S8). The ventilation of this system was designed to have a face velocity between 0.4 and 0.5 m/s, which is typical for chemical fume hoods in the United States.

The units were arranged in modules of reactors and separators to enable reconfiguration to produce the four different drug products within the same system (Fig. 2; see table S2 for the numbering scheme). The synthesis schemes demonstrate the ability to reconfigure the system for increasing levels of chemical complexity, starting with diphenhydramine (Fig. 2A) with one reactor, one separator, and four pumps and finishing with fluoxetine (Fig. 2D) with four reactors, four separators, and 11 pumps. An inline attenuated total reflection (ATR) Fourier transform infrared (FTIR) system (FlowIR) (figs. S9, S13, S14, S18, and S19) (25) provided real-time monitoring of the formed APIs. LabVIEW (National Instruments) programs were also implemented, along with the high- and fast-performance modular X Series data acquisition (DAQ) device and sensors for monitoring multiple process parameters—namely, pressure, reactor temperature, and flow rates. The same LabVIEW platform was also used to automate different units, including heating reactors, pumps, gravity-based separators, and multichannel valves. The downstream module alternatively (Fig. 2, right-hand modules) consisted of precipitation, filtration, redissolution, crystallization, filtration, and formulation units. All drug products were purified and formulated to meet U.S. Pharmacopeia (USP) standards. Consistent with the on-demand format, we focused on concentrated aqueous or alcohol-based formulations ready for dilution to target concentrations when needed and stable for at least 31 days (table S1). Solid formulations, such as tablets, would have required substantial additional space to house unit operations of drying, powder transport, solids blending, and tableting—all processes that would be difficult to implement on the gram-per-hour scale. Nevertheless, we are currently pursuing the miniaturization of these processes so that solid formulations may be prepared on the same platform.

Synthesis and formulation of diphenhydramine hydrochloride

As a first demonstration of the capabilities of this compact unit, diphenhydramine hydrochloride (1) was manufactured in its final liquid dosage form. As shown in Fig. 3, the process commenced with the reaction between an excess amount of neat 2-dimethylaminoethanol (5) and neat chlorodiphenylmethane (6) at a temperature of 180°C and a pressure of 1.7 MPa generated with the use of a back pressure regulator (BPR). The reaction was complete within 15 min, in contrast to typical batch processing requiring 5 or more hours at 125°C in benzene for a similar substrate (26). Because the product API has a melting point of 168°C, it could be handled in flow at 180°C in the absence of additional solvent, thereby minimizing the waste generated. The molten salt was then treated with a stream of preheated (140°C) aqueous NaOH (7). An inline purification and extraction process employing a packed-bed column to increase mass transfer, a gravity-operated liquid-liquid separator with automatic level control (fig. S1), and an activated charcoal filter to remove the colored impurities produced the diphenhydramine API as a solution in hexanes in 82% yield.

In the downstream section, the API was precipitated with HCl (10), and the resulting salt was filtered, washed, and dried in a specially constructed device with a Hastelloy filtration membrane (fig. S6) (27). After redissolving in isopropyl alcohol at 60°C, the diphenhydramine hydrochloride (1) was recrystallized in a crystallizer, while being cooled to 5°C. Upon filtering and drying, the crystals were dissolved in water. Real-time monitoring using an ultrasonic probe yielded the final dosage concentration (5 ml at 2.5 mg/ml). High-performance liquid chromatography analysis determined that the purity of the product conformed to USP standards (fig. S12) (28). Overall, the system capacity based on the optimal yield observed in each step was 4500 doses per day.

The facile transition from 1 to the production of lidocaine hydrochloride (2) (Fig. 4; see also Fig. 2B) was next accomplished through simple adjustments of the fluid manifolds to direct the fluids to specific reactors and separators. Whereas 1 was produced via a single upstream reaction, both 2 and 3 were generated through similar two-step upstream configurations, with modifications mainly in the purification and extraction regimens.

Synthesis and formulation of lidocaine hydrochloride

The synthesis of lidocaine hydrochloride (2) began with the acylation of 2,6-xylidine (11) in N-methyl-2-pyrrolidone (NMP) with neat chloroacetyl chloride (12), premixed inline with a stream of NMP (13) to avoid decomposition on standing (Fig. 4A; see also Fig. 2B). Subsequent addition of a stream of KOH and Et2NH (14) in a mixture of polar protic solvents facilitated the installation of the tertiary amine to generate the crude API, without any intermediate purification. A BPR set at 1.7 MPa after reactor II enabled liquid flow at elevated temperatures (120°C and 130°C), allowing liquid operation well above the boiling point of diethylamine (55°C) and some of the solvents used (methanol and water). As a result, the reaction was complete within 5 min versus batch procedures of 60 min in refluxing toluene (29) or 4 to 5 hours in refluxing benzene (30). Overall, complete conversion (99%) of the starting materials to the crude API was realized in only 36 min. To deliver the crude lidocaine solution with sufficient purity for a streamlined downstream process, hexane (15) and a NaCl/NH4Cl saturated solution (16) were then injected through a cross-junction into the outlet product stream. Upon passing through a packed-bed column containing 0.1-mm glass beads and an inline gravity liquid-liquid separator, lidocaine was obtained in 90% yield. The downstream processing next proceeded with the formation of the HCl salt in a manner similar to that of diphenhydramine. After recrystallization, 2 (88% yield) had a purity of 97.7%, thereby meeting USP standards (fig. S17) (31). The API was treated with a premixed aqueous solution of 4% sodium methylcarboxycellulose to yield a final concentrate. Overall, this system can produce 810 doses (dosage strength = 20 mg/ml) of lidocaine hydrochloride per day.

Synthesis and formulation of diazepam

Following the production of lidocaine hydrochloride (2), we next transitioned to diazepam (3), through switching-in charcoal purification and gravity-based extraction units. As shown in Fig. 4B (see also Fig. 2C), the crude API was synthesized in a two-step upstream sequence initiated with the acylation of 5-chloro-2-(methylamino)benzophenone in NMP (18) with neat bromoacetyl chloride (19) premixed inline with a stream of NMP (20). Bromine displacement, followed by an intramolecular cyclization reaction upon addition of a stream of NH3 in MeOH/H2O (21), then furnished the target molecule. Similar to lidocaine, the application of elevated pressure (1.7 MPa) and temperatures (90°C and 130°C) in this sequence enabled liquid flow and complete conversion of the starting materials in only 13 min compared to 24 hours of batch operation at room temperature (32). After a continuous extraction, the organic stream was then passed through the activated charcoal cartridge to remove the dark colored dimer and trimer side-products. After precipitation and recrystallization in the downstream section, the dried diazepam crystals (3) (94% yield) had a purity level that met USP standards (fig. S22) (33). Resuspending in ethanol in the formulation tank then provided a concentrate. At a dosage concentration of 1 mg/ml (one dose is 5 ml at 1 mg/ml), this system can produce ~3000 doses per day.

Synthesis and formulation of fluoxetine hydrochloride

The last of the APIs produced, fluoxetine hydrochloride (4), was specifically chosen to demonstrate the versatility and capacity of this system to carry out a complex, fully integrated, telescoped, multistep, biphasic synthesis (Fig. 5; see also Fig. 2D). A series of individual reactions carried out in flow, with purification and isolation of each intermediate in batch, has been previously demonstrated (34). By integrating four reactors and four inline separation units, however, we realized the continuous end-to-end synthesis of this API as a racemic mixture. As shown in Fig. 5, the entire upstream reactor network was maintained at 1.7 MPa through the use of multichannel BPR located near the end of the upstream unit. The synthesis commenced with a DIBAL (27) reduction of a close-to-saturated solution of 3-chloropropiophenone in toluene (26) at room temperature in the first reactor. A stream of 4 M aqueous solution of HCl (28) was then introduced, and the resulting mixture was subjected to ultrasound in the second reactor to enable fast dissolution of the aluminum salts and ensure long-term and stable operation of the system (35). A two-stage inline extraction and separation sequence with in-house–constructed membrane liquid-liquid/gas separators removed the aqueous waste and gas (24). An additional stream of aqueous HCl (29) injected into the system before the second separation ensured a complete quench of the reaction.

The intermediate alcohol next reacted with aqueous methylamine (30) at 135°C in the third reactor in a biphasic flow. After a residence time of 10 min, tetrahydrofuran (THF) (32) and aqueous NaCl (20 mol %) (31) efficiently extracted the resulting amino alcohol into a suitable organic solvent (THF) for nucleophilic aromatic substitution in the fourth reactor. Upon separation of the aqueous and organic phase, the latter passed through a cartridge containing 0.4-nm molecular sieves to remove residual water. After a short residence time of 2.6 min in the fourth reactor, the fluoxetine solution merged with a stream of water to prevent the precipitation of the KF salt. Extraction and separation produced a solution of fluoxetine in tert-butyl methyl ether (TBME) (36) in 43% yield and at a production rate corresponding to 1100 doses per day (one dose is 5 ml at 4 mg/ml) prior to downstream processing. Similar to the other three APIs, the downstream processing involved a precipitation and recrystallization sequence to provide fluoxetine hydrochloride crystals that met USP standards (fig. S25) (36). Redissolution in water yielded the final concentrate in 100 to 200 doses.

Overall, the total cycle times for the production and formulation of the APIs varied from 12.2 hours in the case of lidocaine hydrochloride to 47.7 hours for fluoxetine hydrochloride (table S3). Whereas the upsteam syntheses required three residence times (total of 0.7 to 1.3 hours) of the sequential reactions to achieve steady state, the downstream processes took much longer and were mainly dominated by the precipitation step. Because the system featured valves, convenient feed swaps (from reagents to solvents) and fast cleaning procedures between each API production were achieved. Appropriate solvent combinations were added to the reactor lines to flush the up- and downstream units. At the shortest, switching the production of lidocaine hydrochloride to diazepam required a total of 15 min for a complete flush of the internal lines in the upstream section. A switchover from the simplest to the most complex synthesis (diphenhydramine hydrochloride to fluoxetine hydrochloride) would take 2 hours. No cross-contamination was detected from run to run, and the results were reproducible within a standard deviation of 0.6% (diphenhydramine hydrochloride) to 4.7% (fluoxetine hydrochloride) yield for each API production within a single run. The downstream purification and formulation units required no reconfiguration—only the aforementioned flushing. Thus, all transitions between production runs could be completed in less than 4 hours. To meet current good manufacturing practices, one could consider replacing the perfluorinated tubing and membranes in the reactors, BPRs, and separators. The units were designed to facilitate such a replacement.


For over a decade, the FDA has been working to stimulate modernization of small-molecule manufacturing, which is largely based on batch manufacturing processes (37, 38). The vision of the FDA’s Pharmaceutical Quality for the 21st Century Initiative is to create a more robust and flexible pharmaceutical sector capable of manufacturing high-quality APIs. Continuous manufacturing is one such strategy for meeting this vision (1, 39). Continuous manufacturing systems benefit from integrated processing and control, which can translate to increased safety (no manual handling) and shorter processing times. The use of highly adaptable smaller equipment, which implements real-time monitoring, may also lower production costs and improve product quality (1, 37, 38). The present implementation of four well-known pharmaceutical drugs demonstrates the concept of continuous, small-scale, on-demand production of pharmaceuticals. Already-demonstrated advances in flow chemistry (1120) could be realized on similar platforms, and with additional research, ultimately enable the continuous synthesis of modern small-molecule pharmaceuticals, including enantiopure APIs. The current system focused on liquid oral and topical dosage formulations commensurate with the on-demand approach. A complete alternative platform to current batch manufacturing would inevitably have to produce pharmaceuticals in the common dosage forms of tablets and capsules as well as sterile injectable solutions, which would require advances in downstream processing. Specifically, classical unit operations of crystallization, drying, powder transport, solids blending, and tableting would have to be miniaturized and integrated. New approaches such as three-dimensional printing of tablets could facilitate these developments. Realization and demonstration of good manufacturing practices and ultimately FDA approval will be critical to future applications of this technology, including production units for hospitals, health care organizations, pharmaceutical development, and humanitarian aid.

Supplementary Materials

Materials and Methods

Figs. S1 to S25

Tables S1 to S3

References (4043)

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
Acknowledgments: Authors are listed in alphabetical order. We thank A. Clayton for elements of the system design, S. H. Harrison for the LabVIEW programming downstream, and A. T. Carlson for assistance with the design of the downstream units. We also thank L. M. Heckman and J. M. Noss for assistance with development of chemistry. This work was supported by the Defense Advanced Research Project Agency (DARPA) and Space and Naval Warfare Systems Center Pacific (SSC Pacific) under Contract no. N66001-l l-C-4147. We thank E. Choi D.T. McQuade, J. Lewin, and G. Ling for their advice and support. The data reported in this paper are available in the article or in the supplementary materials. A.A. is founder of Zaiput Flow Technologies. T.F.J. is a cofounder of Snapdragon Chemistry, Inc., and a scientific adviser for Zaiput Flow Technologies, Continuus Pharmaceuticals, Paraza Pharmaceuticals, and Asymchem. K.F.J. is a scientific adviser for Snapdragon Chemistry, Inc. A.S.M. is a scientific adviser to GenSyn Technologies, Blues Spark Technologies, Continuus Pharmaceuticals, and Goddard Laboratories. MIT has filed a patent on behalf of A.A., M.B., T.F.J., K.F.J., J.C., A.S.M., J.-C.M.M., E.M.R., D.R.S., T.S., N.W., S.Y.W., and P.Z.

Stay Connected to Science

Navigate This Article