Organic Process Research & Development Review pubs.acs.org/OPRD Design of Experiments (DoE) and Process Optimization. A Review of Recent Publications Steven A. Weissman*'^ and Neal G. Anderson*'* fJ-Star Research, 3001 Hadley Road, Unit 1, South Plainfield, New Jersey 07080, United States * Anderson's Process Solutions, 7400 Griffin Lane, Jacksonville, Oregon 97530, United States O Supporting Information ABSTRACT: Statistical design of experiments (DoE) is a powerful tool for optimizing processes, and it has been used in many stages of API development. This review summarizes selected publications from Organic Process Research & Development using DoE to show how processes can be optimized efficiently and how DoE findings may be applied to scale-up. ■ INTRODUCTION Statistical design of experiments (DoE) is a powerful approach to optimizing chemical processes.1 In 2002 the editor of this journal expressed the opinion that DoE deserved to be taught in academia and should be used more often in industry.2 Over the past 11 years the application of DoE to pharmaceutical process development3 has greatly increased,4 as measured by the volume of DoE-related publications in OPRD (Figure l). For the period 2003—2007, the average was 6.0 publications/year, but that figure jumped to 14.7 publications/year for the period 2008— 2013, a 145% increase. 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Figure 1. Number of DoE-related publications in OPRD annually from 2003-2013. Some of the reasons for this upward trend in DoE implementation are as follows: 1. Changing R&D Environment: The pressure of compressed API development timelines is always increasing.5 This, combined with recent reductions in the pharmaceutical work force, means process scientists are being asked to do more with fewer resources. Also, several new FDA designations are aimed at providing expedited reviews of drugs serving unmet medical needs.6 Accordingly, efficiency becomes paramount and DoE is a tool to resolve optimization issues in a timely manner. 2. Technological Advances: The once daunting prospect of setting up the multiple reactions required by a DoE has diminished with advances in parallel reactors and analytical assays/instrumentation. Setting up 19 reactions is no longer considered to be the barrier it was even 15 years ago. High-throughput screening (HTS) tools and flow chemistry setups have been widely adopted. 3. Changing Drug Substance Regulatory Environment: The unmistakable trend away from the traditional quality by testing and toward quality by design (QbD) has led to increased awareness of the concept of design space.7 DoE, as part of process validation, plays a central role defining the acceptable ranges for the critical process parameters. The FDA now expects DoE to be part of the NDA submissions. 4. Increased Implementation of Green Chemistry Principles: Several of the well-known 12 Green Chemistry Principles advocate for using lesser amounts of solvents and reagents and for increased process efficiency as a means toward generating less chemical waste. Accordingly, manufacturers are increasingly implementing programs9 and applying metrics10 to assess the green chemistry performance of their processes. DoE's ability to enable scientists to hone in on the optimal reaction conditions has helped it gain wider acceptance. Clearly the perceived barriers11 to implementing DoE in the process chemistry setting are being overcome, leading to wider adoption of this powerful tool. The traditional optimization approach, varying one factor/ variable at a time (OFAT, also called OVAT),12 suffers from many drawbacks. First, it rarely uncovers the optimal conditions, in large part due to the fact that the outcomes are highly dependent on the starting point. Second, the OFAT approach is unable to separate the "noise" (the inherent run-to-run variation of a system) of a reaction from actual improvement unless a significant number of reactions are repeated using the same conditions. The systematic approach inherent in DoE eliminates researcher bias and often will lead to reaction conditions that one had not considered previously. But the most significant Special Issue: Application of Design of Experiments to Process Development Received: May 28, 2014 Published: June 27, 2014 ACS Publications e 2014 American Chemical Society 1605 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Half Normal plot A = volume B = catalyst (C = base; no significant effect) D = temperature (E = time; no significant effect) Effect! Figure 2. Half-normal plot showing the influence of volume, catalyst charge, and temperature on a Heck reaction yield. Volume and temperature interact (AD) to influence the yield. The farther that data points are from the diagonal line, the greater the influence of the factor. Reprinted with permission from Aggarwal, V. K.; Staubitz, A. C; Owen, M. Org. Process Res. Dev. 2006, 10, 64. Copyright 2006 American Chemical Society. advantage to using DoE is the ability to quickly detect how interactions between factors can affect product yield and quality. Also, by simultaneously varying parameters, the DoE approach can be more efficient than that achieved by the traditional approach of varying one factor/variable at a time. For instance, if the OFAT approach was used to investigate the influence of three factors on a reaction (temperature, concentration, and reagent stoichiometry), eight experiments would be required; but more information could be generated through four experiments in a half-factorial DoE. Running a DoE may seem daunting initially, since the number of experiments to be run is defined at the beginning, unlike the traditional approach. But even if the effort to perform the multiple reactions and assays engendered by a DoE is time-consuming, the quality and thoroughness of the information obtained outweigh the effort. Generally there are three times in the life of a process when DoE may be used: early in process development when the best conditions are not obvious; when exploring robustness to assist in defining a design space in support of a QbD filing with regulatory agencies; and when optimizing conditions for established procedures. In the latter case there can be considerable financial impact by minimizing the charges of reagents and solvents and by optimizing conditions for small improvements in yield or product quality.13 HTS compliments DoE and may be used to select reagents, as in screening various acids for an acid-mediated deprotection. DoE can measure the effects of ranges of variables, such as the effects of charging various equivalents of a specific acid over a range of temperatures.14 The results of DoE experiments are analyzed to find which factors influence the results in a statistically relevant manner and which of those are interdependent, if any. These analyses can be shown through half-normal plots (Figure 2), analysis of variance (ANOVA), and other means. Data from nonlinear studies of reactions can be shown in contour plots and response surface methodology (RSM, Figure 3). DoE software packages used for process optimization include Design Expert (Stat-Ease Inc.), MODDE (Umetrics), DoE Fusion PRO (S-Matrix Corp.), STAVEX (Aicos), Minitab (Minitab Inc.), and JMP (SAS). Most, if not all, of the sponsoring companies provide training, consulting services, and free online tutorials for their products. The software platforms can be used to set up DoE studies that range from the very simple to the very complex and from fractional factorial designs to RSM designs that include settings at multiple levels. The more advanced designs are typically used to preclude confounding interactions of variables, or when practical constraints render certain combinations of factors unfeasible to execute. Unlocking the full potential of a DoE is predicated on systematically proceeding through a prescribed work flow: 1. Define the Objective: What process issues are to be resolved? Frequently, the goal is to optimize a process, but sometimes the process conditions are "locked" and the goal is merely to understand the robustness around the existing conditions. In this case, robustness is defined as probing the impact (if any) of small changes in the continuous factors16 on the outcome. 2. Define the Factors/Variables and Their Ranges: The determination as to which factors/variables to include in a DoE depends on resources. Clearly, more variables will engender more reactions, time, and materials. Thus, it is imperative to prioritize them, using existing process knowledge, into groups such as known to be impactful, suspected, possible, and unlikely. Then assign high and low settings for the factors selected for inclusion in the study, as is the case when using the common two-level factorial design. These settings need to be relevant, achievable, and practical. The end product of this exercise is creation of the design space. 3. Define the Responses: These are the measurable outcomes of the process. Frequently, reaction yield (either isolated or assay) and conversion are the primary outputs. The responses are related to the objectives, and several can be included in a single DoE study without creating more reactions. Accuracy and reproducibility of experimental technique and assays are of paramount importance in order to achieve meaningful results. Typically, the center point experiments17 are performed in triplicate as a probe of reproducibility but also serve to detect nonlinearity in a response when used in a screening design (vide infra). 1606 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review DTBB18C6 yield 48 00 $400 «000 «600 72 00 (a) (b) Figure 3. Contour (a) and three-dimensional response surface (b) plots of the effect of time (X3) and equivalents of base (X4) on reaction yield. Reprinted with permission from Fan, J.; Yi, C; Lan, X.; Yang, B. Org. Process Res. Dev. 2013, 17, 368. Copyright 2013 American Chemical Society. 4. Select the Experimental Design: The choice depends on the objective, number of factors, and resources available, mostly related to the number of reactions required. The most common are screening designs in which qualitative information about the relevant factors is obtained. These designs also enable the factors to be ranked in order of extent of impact on the response. Screening designs are frequently employed to weed out the irrelevant factors to enable focusing on the most relevant factors in a second DoE design, also known as optimization designs.18 These latter designs require more experiments than screening designs but generate a more comprehensive model of the response surface. 5. Generate Reaction Worksheet: Inputting all the above information into the software will generate a list of reactions to perform. Ideally, the reactions are all performed at the same time, but often this is impractical; thus, it is best have the software randomize the run order as a means of mitigating any systemic bias or error. 6. Perform Reactions/Collect Data: Again, it is critical to perform the reactions under identical experimental and analytical conditions. Sometimes it is not possible to do so; for example, perhaps two lots of starting material must be used out of necessity, as one has been depleted. DoE is able to account for this by dividing the reactions into "blocks" with each lot of starting material comprising a block. The statistical output is then able to determine whether the source of starting material imparted any bias into the design. 7. Data Input/Software Analysis: The results are input into the software, and each response is analyzed individually. A mathematical model is selected by the user based upon key metrics, such as the adjusted and predictive R-squared terms. The model can then be used to predict the outcome of a theoretical set of reaction conditions within the design space. Often there is more than one response requirement needed to meet a goal or specification. For example, a conversion must be >98.0 A% while the level of an impurity must be < 1.5 A%. The software can assist in identifying reaction conditions that meet both requirements. These overlapping outputs are commonly known as "sweet spot plots". 8. Confirming Reaction: It is important to note the ideal conditions suggested by the DoE analysis are only a model prediction and must be verified experimentally to validate the results. This review summarizes selected process research examples using DoE published in Organic Process Research & Development between 2003 and 2013.19 The review aims to demonstrate the power of DoE to optimize processes with relatively few experiments, and it highlights the optimal process conditions that may not have been obvious at the beginning of research efforts. The discussions below are loosely grouped by reaction type or operation, such as hydrolysis, crystallization, and analytical development.20 Some perspective is provided, as sometimes process scientists in these examples have used other processing considerations to select optimal scale-up conditions that were not evident based on the results of DoE studies. This review will not discuss the details of the statistical analysis. For each summary, we sought to provide the following information if it was presented: • What was the problem that was solved/studied? • What type of DoE design was employed (e.g., screening, RSM, robustness)? • What continuous factors were investigated? What were the values/ranges? • How many reactions were performed? In what time frame? • What was the overall improvement realized through the DoE? • What scale was the confirming reaction performed on? • What were the optimal conditions used to verify the DoE analysis? • To what extent did the DoE model predict the actual outcome? ■ REVIEW OF SELECTED DOE MANUSCRIPTS PUBLISHED IN OPRD FROM 2003-2013 Hydrolysis. Bristol-Myers Squibb (BMS) workers optimized the deprotection of acetonide 1 to diol 2 in the preparation of BMS-644950, an HMGR inhibitor (Scheme l).22 The parameters examined were the charge of MeCN (7.5—12.5 volumes), HC1 concentration (0.015-0.030 N), HC1 stoichi-ometry (0.03—0.07 equiv), and reaction temperature (35—49 °C). The data analysis revealed that temperature had the greatest impact on the reaction rate, with HC1 concentration also having an impact. A competing epimerization at the C-5 hydroxyl was minimized using a catalytic amount of HC1 at moderate 1607 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 1. Hydrolysis CCVBu ArN CCWBu CO,H cone. HCl (0.9 ml/g) H20 (0.6 ml/g) AcOH (3.6 ml/g) O PhCH3_ " ArN 30 °C 1 (35 kg) Ar = 4-F-phenyl Art) 3 (4.95 kg as pTsOH salt) Ar = 3,5-dichlorophenyl 1) r?-OctSH (2.2 eq.) H2S04 (2.2 eq.) CH3CN RT/0.5 h 2) MTBE 3) heptane OAr 1)H2SO4(7.0 eq.) 4 vol% H20 in CH3CN 70 °C/2 h ■ 2 H2S04 (9, 95%) 1)KOH (powder, 5.4 eq) f-AmOH (10 vol) 90 °C/11 h 1) NaOH(1.59 eq.) THF, H20 -1 to 1 °C/2.75 h 2) HCl, CH2CI2 XN' ''C02Et 3) A, -CH2CI2 H 4) + acetone OAr meglumine salt (92%) 15 (48.7 kg) 16 temperature. The diol ester was telescoped and converted to the ammonium carboxylate as the API. Hydrolysis of the tot-butyl ester 3 led to the formation of 1% of the tert-butylamide impurity 5, along with impurities 6 and 7 (Scheme l), under the conditions initially used for scale-up (concentrated HC1 and AcOH, ca. 1:2, in toluene at 55—65 °C).23 The tert-butylamide was probably formed by a Ritter reaction of nitriles 3 or 4 with the cation generated through deprotection of the tot-butyl ester. Results from the initial DoE screen showed that the interaction of higher temperatures and higher concentrations of HC1 had the biggest impact in increasing the rate of deprotection. The BMS group undertook additional DoE experiments to examine the effects of HC1 and H20 on the formation of 5, 6, and 7 (central composite design: HC1: 1.23, 1.57 mL/g of 5; H20: 0.325, 1.175 mL/g of 5). By increasing the concentration of water and decreasing the concentration of HC1, impurities 5, 6, and 7 were controlled to less than 0.25% in the API 4. DoE screening was used to investigate the removal of the N-THP and N-t-Bu protecting groups in the development of the reverse transcriptase inhibitor candidate 10 (Scheme l).14 Initial conditions provided the tosylate salt analog of 10 in about 65% yield through double deprotection of 8 using 5 equiv of dry TsOH and 25 equiv of TFA in acetonitrile. The use of either acid alone led only to loss of the THP group, and acetonitrile was necessary for double deprotection with minimal levels of impurities. The primary disadvantages of the initial conditions were the need to dry the reaction mixture azeotropically, the need for a large excess of TFA, and the relatively low yield. Through two high-throughput screenings for acids using microtiter plates, cone H2S04 was identified as a simpler alternative for double deprotection; however, under those conditions, the yield of 10 was still about 65% with a large amount of black polymeric material observed, as with the initial conditions. Treating 8 with cone H2S04 at RT smoothly removed only the THP group and produced black polymeric material; when purified 9 was treated with cone H2S04 in acetonitrile at 70 °C, the yield of 10 was essentially quantitative. Thus, the low yields from double deprotection were due to the removal of the THP group, not the tert-butyl group, as had been suspected. The Merck researchers screened reagents to trap the byproducts from the THP group, i.e., dihydropyran and hydrated dihydropyran. When n-octanefhiol was added to solutions of 8 in acetonitrile containing cone H2S04, 9 was rapidly generated and crystallized in high yield; however, heating a mixture of 8, n-octanethiol, and cone H2S04 in acetonitrile produced lower yields of 10, probably because the dihydropyran—thiol adducts were unstable. A DoE study to optimize removing the t-Bu group 1608 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 2. Acylations by Oxygen Nucleophiles H°Y^OH ^OH HO" I O 17 Okg) BzCI (3.0 eq.) H2S04 (0.75 wt%) PhCH3 (0.7 kg/kg 1) 90 °C; 106 °C; 125-130 °C PhC02H (mp 122 °C) + O BzO^Jl BzO' 18 O 2) PhCH3 3) H20 4) 89-92 °C/ 30 min 5) cool, crystallize O OH H,0 BzO. BzO-'V OH OH BzCI (3.05 eq) T T Py< (5-5 eq) HO*OBz OBz HO*^j^ ''OR2 OR3 21, R2 = H, R3 = Bz R2 = Bz, R3 = H 0 19 (92.7%) OBz OBz 2) Bz20 (1.5 eq) 47 °C/26 h ..OBz XI' HO^Y ''OBz BzO' 3) PhCH3, H20 0Bz 47 °C/1 h 22 (45% isolated. 4) aq. NaHC03 94% Purity) 5) A, - PhCH3, + heptane ?HOBn V-^^~OH BnO-\»—-^^-OH OBn D.L-24 (1 g, 4 mg/mL in solvent) Novozyme 435 H2C=CHOAc - hexane (0.3 : 1.7 v/v) 45 °C/24 h ?HOBn V^-^OH BnO-X-'-^^OAc OBn L-(-)-25 (50%, 99% ee) OH Bnu I HO~7---f-^/-OBn BnO D-(+)-24 (99% ee) from 9 with a full-factorial design was executed varying water content (0, 10 vol %), temperature (55, 75 °C), reaction time (2, 8 h), and charge of H2S04 (3, 15 equiv); three center point runs were added for a total of 19 runs. The highest yields were observed at the center point runs. A second DoE (optimization) was carried out, holding the temperature constant at 65 °C with a reaction time of 3 h; the design was a face-centered central composite (CCF) for water content (2, 4.5, and 7 vol %) and H2S04 charges (5, 7, and 9 equiv), totaling 11 runs with three center point experiments. Under all conditions, amine 9 reacted completely, and the charge of H2S04 had little effect on conversion or yield. Smaller charges of water raised the yields somewhat. Although the highest yield was found using 2 vol % H20, experiments showed that the yields dropped off sharply at charges of less than 2 vol % H20, and for robust operations on scale, a charge of 4 vol % H20 was selected. Additional investigations showed no hydrolysis of the nitrile group in 10. AcOH and tot-butyl acetamide were found in the reactions, indicating that acetonitrile was reacting with water and the t-BuNH2. Overall the yield from the two-step processing was improved from about 65% to ~85%. A collaboration between AMRI Global and Icagen optimized the final step in the synthesis of potassium ion channel blocker ICA-17043 (12) using DoE (Scheme l).24 The initial hydration of penultimate 11 was accomplished under acidic conditions but was plagued by formation of 13 and 14, as well as a troublesome workup. Under basic conditions (powdered KOH in tot-amyl alcohol), their preliminary results showed a 95% conversion with 4% impurities. A 22 full factorial design (run in replicate with two center points; 10 reactions) was established to look at maximizing yield and minimizing impurity formation by varying solvent volume (4 and 7 vol) and KOH charge (1.5 and 3.5 equiv). The reactions, performed in a 12-position carousel reactor, were assayed after 6 and 24 h at 100 °C. Each factor and the two factor interaction impacted the reaction conversion, which was maximized with high KOH charge and low solvent charge. Impurity formation was reduced under the higher dilution conditions, while prolonged heating decomposed the product, especially with more KOH and less solvent. The conversion ranged from 76—100% over the design space. Additional post-DoE optimization led to the eventual optimal conditions that balanced the need for high conversion (the nitrile was difficult to remove) and low impurity formation. The hydrolysis of chiral ester 15 was investigated using DoE, in order to minimize racemization of 16 (Scheme 1).2S Of the four factors studied (temperature, solvent amount, NaOH equiv, and addition rate), a low reaction temperature (0 °C) was the most important; only 0.3% epimerization was detected. Acylation. For a large-scale preparation of 0,0'-dibenzoyl-L-(+)-tartaric acid (19; Scheme 2), DoE studies were used to optimize the benzoylation of 17 and the subsequent hydrolysis of intermediate anhydride 18, increasing the yield from 81% to 95% in the telescoped process.26 Twenty runs were carried out using a central composite design with 6 center points looking at solvent amount (0.65—1.15 g of toluene/g of 17), acid equivalents (15— 35 mg of H2S04/g of 17) and temperature (105-125 °C). The results were correlated with the isolated yield of and the amount of color in 19 (measured by absorption at 600 nm). The yield increased as the temperature was raised and the charges of toluene and H2S04 were decreased, while the color increased as the charges of toluene and H2S04 were increased. To prepare 18, the reaction temperature was raised in stages, first to 90 °C to accelerate the reaction, then to 106 °C to encourage the release of gaseous HC1, and finally to 125—130 °C to melt the benzoic acid byproduct. Next the reaction mass was cooled and diluted with toluene and water, and the mixture was heated at reflux at 89-92 °C to hydrolyze anhydride 18. [Safety note: massive evolution of vapors would occur if water is added to the mixture above the boiling point of the toluene-water azeotrope.] The mixture was cooled to crystallize 19. This process was carried out in a 1.5 m3 reactor. A Sanofi-Aventis group27 studied the conditions to optimize the selectivity and conversion for the tetrabenzoylation of D-galactose (20) to produce 22 as part of their synthesis of SGLT inhibitor candidate SAR7226 (Scheme 2). An initial solvent screen identified NMP and DMA as the most desirable solvents 1609 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 3. Acylations with a Carbamoyl Chloride O CI3CO^OCCI3 Boc (0.38 eq.) Et3N (1.5 eq.) EtOAc 5) 26 6) Et3N 7) aq. workup (>85%) 1)C02(g) Et3N(1.3 eq.) EtOAc 25 °C 27 2) TMSCI (1.5eq.) CF, TMSO. CH3 O CH CF, 29 3) pyr (2.0 eq.) - 4) SOCI2 (1.5 eq.) 30 (1.2 eq) Et3N (2.5 eq) EtOAc (0.55 M) reflux/16 h COCH N NaBH(OAc)3 (1.64eq) F HC02H (5.5 eq) CH3CN 15°C/12 h to minimize formation of both the over-reacted pentabenzoyl species 23 (10—13 A%) and the tribenzoylated species 21 (5—8 A%). NMP was chosen for further development for industrial hygiene reasons. Benzoyl chloride was found to be more reactive and less specific than benzoic anhydride. The goals became to benzoylate with BzCl at the more reactive hydroxyl groups (CI and C6) and better solubilize the intermediates, and then complete the benzoylation to 22 through the more selective Bz20. Thus, the two benzoylating reagents were employed in tandem in the DoE study, BzCl and Bz20 (1.5 equiv), while pyridine (5.5 equiv) was selected as the base. The authors chose an optimization design (second-order central composite) looking at the BzCl stoichiometry (2.9—3.2 equiv), solvent volume (7—13 volumes), and reaction time (14— 24 h) at 20 °C. The target was keeping the level of 23 to <15 A% while maximizing conversion, and then adding the final benzoyl group with Bz20 at a moderate temperature. The data indicated that the highest levels of the desired product (the thermodynami-cally equilibrated /J-anomer 22) were obtained with 10 volumes of NMP and 3.05 equiv of BzCl and Bz20 (1.5 equiv). The optimized process provided a mixture of22 (65—70 A%), 23 (8— 11 A%), and 21 (5-9 A%). The DoE study served to identify robust conditions that reduced the formation of the over-reacted species from 13 A% to 8-11 A%. A group from universities in Brazil optimized the kinetic resolution of 1,3,6-tri-O-benzyl-myo-inositol (D,L-24) by acetylation using Novozyme 435, a lipase immobilized on acrylate resin (Scheme 2).2S In the first DoE screen, a fractional factorial design (24_1with three center point runs; 11 reactions) was employed with the goal of maximizing the conversion; factors varied were substrate concentration (2—14 mg/mL), vinyl acetate concentration (187—560 mg/mL), enzyme concentration (54.6—136 units), and temperature (30—50 °C). The concentrations of the enzyme and substrate had the greatest impact on the conversion, while temperature and vinyl acetate charge (a large molar excess) had virtually no impact. The conversion in the design space ranged from 6—43%. In a subsequent central composite optimization design, the acyl donor charge was fixed; the conversion was greatly influenced by the temperature and substrate concentration. An interdependence between substrate and enzyme concentrations was observed. Reaction conditions were selected for high conversions with relatively low enzyme loading. The model predicted highest conversions at 57 °C, but that temperature was not used due to concerns about increased enzyme degradation. A portion of enzyme was reused seven times under optimal conditions before the extent of conversion fell off significantly. Overall, these optimizations reduced the time cycle from 112 to 24 h and increased the productivity 15-fold. GSK researchers used DoE to optimize the formation of urea 28 (Scheme 3).29 Triphosgene was the only reagent found to react with hindered amine 26 to form an active carbonyl derivative and then generate 28 from amine 27 in an acceptable yield. DoE studies were carried out, varying the triphosgene charge (0.35, 0.45 equiv), solution volume of 26 (2, 10 vol), benzylamine 27 charge (1.2, 1.5 equiv), and solution volume of 27 (5, 15 vol); using a central composite design, 30 experiments were run. To minimize the generation of impurities, the optimal conditions required 0.38 equiv of triphosgene and 1.4 equiv of benzylamine 27; solvent volumes were not critical. To avoid the dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 1610 Organic Process Research & Development Review Scheme 4. Acylations by Nitrogen Nucleophiles FmOCOCOCI (1.5 eq) DIPEA Ar (2.0 eq) ch2ci2 0°C 34 Ar = 4-pyridyl OFm 35, X = O (90 - 95%; (EtO)2PMe (3 eq) PhCH3, 30 °C/50 min 36, X = P(OEt)2Me h2n Ph W ( C02H 38 (100 g) 1)SOCI2 (2.0 eq) PhMe 70 °C/20 h HO,, HN CO,H Ph 2) A, - SOCI2 3) L-threonine (1.05 eq) 39 „„ , (82% after Na2C03 (2.2 eq) recrysta||ization) H20, 20 °C/20 h HATU (1.06 eq) (/Pr)2EtN (3 eq) DMF, 0 °C/2 h O C02Et CI (1.2 eq.) pyr(0.1 mol%) PhMe (6 mL/g) 25 °C/1.5h C02Et 1) /-BuB(OH)2(3.5eq) aq HCl, MeOH, heptane 20 °C / 20 h 2) A, - MeOH 3) + EtOAc, aq NaHC03 4) (HOCH2CH2)2NH (1.0 eq), 20°C/12h HO,,. ' H HN— -N Ph 41 (84%) 1)TBD (0.28 eq) 2-MeTHF, -OH PhMe use of phosgene or phosgene derivatives, a second-generation approach was developed. Reaction of amine 27 with C02, Et3N, and TMSC1 probably generated the soluble TMS-carbamate 29, which was then treated with pyridine and SOCl2 to generate the carbamoyl chloride 30. The latter was then condensed with amine 26 to generate urea 28. A DoE examined the effects of varying seven factors in this reaction mixture of gas, liquid, and solid salts: temperature before and after TMSC1 addition, volume of EtOAc, and the charges of Et3N, TMSC1, pyridine, and SOCl2. Twenty experiments were run in a D-optimal design, keeping the flow of C02 constant throughout the processing. Higher charges of SOCl2 increased the yields but produced darker reaction mixtures. Concentrated reactions and lower temperatures gave lower yields. The C02—SOCl2 process was used to manufacture the API through deprotection of 28 and salt formation. GSK workers applied QbD principles in developing the manufacturing process for casopitant mesylate, an NK] 1611 antagonist. They sought to determine the proven acceptable ranges (PAR) for the parameters known to impact the quality of the API using DoE for two of the eight steps. In one stage (Scheme 3), (R)-piperidone 31 was acylated with carbamoyl chloride 30 to yield urea 32. Four process variables were studied: reaction concentration in EtOAc (4.5—5.5 vols), TEA equivalents (2.3—2.7), carbamoyl chloride equivalents (1.0— 1.2), and reaction temperature (75—85 °C). The yield response varied over these reaction ranges from 82 to 99%. Analysis indicated that only the amount of reagent 30 had an impact on yield, with 1.2 equiv being preferred. The subsequent reductive amination of urea 32 with 1-acetylpiperazine to generate 33 was also optimized with DoE to determine the impact on the downstream API. A two-level fractional factorial design (26-3) covering six variables with two center points was undertaken (10 reactions); the responses measured were yield and diastereo-meric ratio. The six variables were NaBH(OAc)3 equivalents dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 5. Oxidations RNH OH 1) DMSO (4 vol) AC2O (5.53 eq.) 60 °C / 2 h 2) H20 (2 vol) 60 °C /1 h 3) EtOAc, H20 4) aq. washes 5) A, - EtOAc 6) + MTBE RNH SCH, 48 (81%; 1) NaOCI (1.12 eq) TEMPO (10 mol% KBr (5 mol%) EtOAc - H20 -5 to 0 °C 2) aq. Na2S204 3) 0.5 M NaOH 4) - EtOAc, + /PrOH 5) + H20 50 (29.6 kg) N N-Me =N 51, R = CHO (93% 02 (40 mL/min) C03O4 (2.5 wt%) NaOH (330 wt%) 10%H2O in ethylene glycol 110°C/12 h 52, R = C02H 53, R = CI, Br CHO 57 (98% conversion) HO 59 (22 kg) 62 (225 mol) 1)Na2W04(5 mol%) H202 (1.2 eq) pH 6.8 buffer C02H DMA 2 90 °C / 2 h 2) ascorbic acid quench RuCI3 (cat) NaOCI (3.9 eq) NaOH added to maintain pH 8.4 water [0.2 M] 5°C C02H 60, R = H (98% AUC) 61, R = OOH H02C 63 (73 - 75%) (1.55—1.72), formic acid equivalents (5—6), reaction temperature (5—25 °C), age period for the NaBH(OAc)3 and formic acid (5—55 min), 1-acetylpiperazine equivalents (2.0—2.5), and addition time (5—55 min). The center point values for the NaBH(OAc)3, formic acid, and piperazine equivalents at 15 °C gave the highest yield (ca. 72%) and were selected for the manufacturing process. Minimal variation of the diastereose-lectivity was observed within this design space (maximum: 75:25). The optimized procedures are described for a 1 kg scale in the paper. GSK workers optimized an acylation and cyclization in the synthesis of GV143253A, a broad-spectrum antibiotic.31 Initially the acylation of the /^-lactam nitrogen of intermediate 34 with FmOCOCOCl proceeded in only 45-50% yield (Scheme 4). A central composite design (10 experiments with 2 center points) was set up to examine the effect of the reagent (1.3—2.2 equiv) and DIPEA (1.5—3.0 equiv). Parallel experimentation was facilitated by the use of an Anachem SK233 apparatus. The second reaction examined was the formation of phosphorane 36 from 35, and another central composite design (20 experiments with 6 center points) was used. The three parameters were phosphonite equiv (2—10), solvent volumes (DCM/toluene; 10—50), and temperature (undisclosed range). Higher amounts of the phosphonite led to improved yields at 40 °C while the reaction concentration had no impact. The phosphorane was cyclized in cyclohexane at 90 °C to produce the desired crystalline intermediate 37 in 60% overall yield from 34. This improved yield compares to the 45% isolated yield obtained before the second DoE optimization. The Process Chemistry team from Teva employed DoE to optimize several continuous variables in the synthesis of 41 (Scheme 4), a derivative of delanzomid (CEP-18770), a peptide boronic acid under investigation as a proteasome inhibitor for the treatment of multiple myeloma.32 Several DoEs using Design Expert were performed to optimize the operating range of each variable of the acylations to yield 39 and 40 and trans-esterification in the three step process. For the first step, the thionyl chloride charge was reduced from 5 to 2 equiv under more concentrated conditions, and by charging more than 1.0 equiv of L-threonine and 1.3 equiv of sodium carbonate, the dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 1612 Organic Process Research & Development Review formation of a troublesome impurity was avoided. To minimize epimerization at the threoninamide methine of 39 in the second acylation, HATU was employed and at least 2.8 equiv of DIPEA was necessary. The other factors were found to be robust. For the transesterification, the amount of isobutylboronic acid (3.5 equiv) and concentrated conditions were imperative to driving the transesterification equilibrium in a timely manner. The collective process improvements resulted in an improved overall yield (25 to 60%) and product quality, and avoided chromatography. Lilly researchers used a full factorial DoE to optimize the acid chloride formation of 42 (Scheme 4).33 Pyridine was selected as the catalyst over DMF, as the latter produced an impurity that was difficult to purge. The three factors examined were temperature (20, 40 °C) and the charges of oxalyl chloride (1.05, 1.3 equiv) and toluene (4, 9 volumes). A higher charge of oxalyl chloride and a higher temperature promoted the formation of the acid chloride 43, while, at lower temperature and a higher volume of toluene, the conversion to 43 was slower. Optimal conditions were selected near the center of the design space. Compound 43 was subsequently used to acylate a hydrazine derivative. Using 1,5,7-triazabicyclo[4.4.0]dec-5-ene (TBD) as an acylation catalyst, a Sanofi team optimized the formation of amide 46 from ester 44 and benzylamine 45 (Scheme 4). DBU or tetramethylguanidine gave <8% conversion in reaction screening.34 A three factor central composite design studying the effects of TBD catalyst (0.10—0.33 equiv), reaction temperature (60— 80 °C), and reaction time (3—5 h) was undertaken to maximize conversion while minimizing the catalyst charge. The analysis revealed the TBD charge was the most influential factor, followed by temperature. The best combination of robustness and the above criteria was satisfied with 0.28 equiv of TBD at 70 °C for 4 h, using a stepwise heating protocol. These simple conditions skirted the need to prepare 46 in separate steps of hydrolysis and acylation. Oxidation. GSK researchers investigated in detail three procedures to oxidize the o-amino alcohol 47 to the ketone API 48 (Scheme 5).3S Key considerations were conversions of >99.5% to ease purification, avoiding epimerizing the center o-to the ketone, minimizing environmental impact, and minimizing any impurities, especially metals. TEMPO oxidations were examined, varying the charges of TEMPO (0.05, 0.5 equiv), KBr (0.05, 0.5 equiv), and NaOCl (l.l, 2.0 equiv), along with the pH of the NaOCl solution (8.0, 10.0) and the agitation rate (300,900 rpm) of the biphasic mixture (aqueous CH2C12). Under half-factorial experimental design, 32 experiments were carried out, while monitoring the disappearance of 47 and the formation of 48 and an impurity. The most critical factors for the optimal conditions included charging 0.35 equiv of TEMPO and acidifying the pH of commercial NaOCl solutions to about 9.0. Unfortunately, these conditions did not prove to be robust, and the NaOCl decomposed to unidentified byproducts, possibly O^ Cl2, or Br2; hence, TEMPO-catalyzed oxidations were abandoned. Oxidation using a polymeric piperidinyloxyl radical (PIPO) catalyst derived from oxidation of Chimassorb 944 was also examined. Again 32 experiments were performed, varying the charges of PIPO (1.67, 6.67 wt %), KBr (0.03, 0.2 equiv), NaOCl (1.1, 1.8 equiv), pH (8.4, 9.8), and agitation rate (300, 900 rpm). High conversion with high selectivity was observed using 3 wt % PIPO, 0.2 equiv of KBr, and 1.1 equiv of NaOCl at pH 9.0. Unfortunately, the stable emulsion generated by the polymeric reagent posed a significant obstacle to further development, and the approach with PIPO was also abandoned. Under Moffatt oxidation conditions (Ac20—DMSO, shown in Scheme 5), the Pummerer side product 49 was the primary impurity, and its formation was promoted by increasing amounts of H20. A DoE was performed on four variables, equivalents of Ac20 (2.0, 4.0), H20 (0.0, 1.0), and DMSO (3.4, 4.5); and temperature (50, 70 °C); 32 runs were carried out, monitoring for conversion to 48 and formation of 49. The charges of Ac20 and H20 were the most critical factors, with an optimum temperature of 60 °C. The authors mentioned that, before charging Ac20, a vacuum distillation with toluene can be used to azeotrope out water brought in with the DMSO or with 47, thus decreasing the formation of byproduct 49. BMS workers found the TEMPO-promoted oxidation of alcohol 50 to aldehyde 51 was complicated by the formation of impurities 52 and 53. An undisclosed DoE was performed leading to the conditions that identified the ideal reagent stoichiometries (TEMPO, 0.1 equiv; KBr, 0.05 equiv; NaOCl, 1.12 equiv). Unbuffered NaOCl was used to circumvent the decomposition of NaOCl seen with more acidic solutions often employed for TEMPO oxidations.22 DoE was used to examine the air-mediated oxidation of p-cresol (54) to p-hydroxybenzaldehyde (57) (Scheme 5).36 Ethylene glycol was charged to trap the benzoquinone methide 55 as the glycol ether 56, and this solvent was chosen, as it was safer than MeOH (based on flammability and flash point considerations), which was initially used. Water was beneficial to the reactions, perhaps because it decreased the viscosity of the solvent. Four factors were examined under DoE screening: the charges of substrate (15, 25 g), base to substrate ratio (2:1, 4:1), water charge (10%, 20%), and temperature (80, 90 °C). A reduced central composite design was carried out, with a total of 21 experiments, monitoring for the formation of 56 and 57. An increased substrate loading decreased the selectivity for 57. Analysis of the DoE results indicated that the process optimum would fall outside the initial ranges chosen, so further optimization was undertaken. Under optimal conditions, 57 was generated in 98% yield, and no p-hydroxybenzoic acid (58) was detected. The oxidation of alcohol 59 to ketone 60 was optimized by Shionogi workers37 en route to the synthesis of endothelin A receptor antagonist S-0139 (Scheme 5). The Jones oxidation initially developed was deemed unsuitable for large-scale production, so it was replaced with a more benign sodium tungstate—H202 reagent. The reaction with Na2W04—H202 initially proceeded in 81% yield in dimethylacetamide (DMA) but was plagued by the formation of the hydroperoxide 61, which led to lactone, diketone, and epoxide side products. A central composite design (16 reactions) was chosen to study the effect of hydrogen peroxide (35% aqueous; 1.0—1.4 equiv) and sodium tungstate (3—7 mol %) stoichiometry as well as the pH of the phosphate buffer employed (1.5—9.1) in the reaction. The reactions were performed in 0.4 M aqueous DMA at 90 °C on a 2.4 mmol scale. The responses were yield of the desired ketone 60 and formation of the difficult-to-reject peroxide 61. The results showed that the impurity formation was minimized at high pH and with fewer equivalents of hydrogen peroxide. The yield of the ketone was maximized at 1.2 equiv of oxidant over a wide pH range using 5 mol % of the tungstate. Due to the instability of the reagent system at higher pH, a buffer of pH 6.8 was selected for the manufacturing conditions; the reaction mixture was quenched and ozonolyzed to the desired product (not shown). 1613 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 6. Halogenation and Nitration Boc C02Me 1)(MeOCH2CH2)2NSF3 (Deoxo-Fluor®, 3 eq., 50% in PhCH3) Si02 (5 wt%) 70°C/5 -10 h O 64 (3.5 kg) 2) 10%aq. NH4OH, pH 8.5 3) A, - PhCH3 4) + r?-hexane Boc C02Me Boc vC02Me N- I" I" F 65 66 (65 : 66 : 64 - 40% : 9% : 37%) 1) aq. NaOH 23 °C / 24 h 2) aq. NaOH R MeOH Boc C02H 23 °C / 4 days N 1) H2S04 (1 mol%) C02Et NCS(1.05eq) MeOAc 13-19°C/2h 2) NaHC03, Na2S03 3) + /'PrOH, - MeOAc 4) KOH 5) aq. AcOH HO,C 3) NaOCI p -p 4) 10% HCl to pH 2.5 65a (39 - 42%) C02H 70 (50 g) Br2 (0.75 eq.) 70% HN03 (20 vol.) 50 °C / 4 h 71 (19%) 1) NaOH (1.0M. 2.5eq.) HgO(1.09 eq.) H20, AcOH 95 °C / 2 days 2) cone. HCI (24.6 eq.). 80 °C Br ^-^ Br 72 73 (93% as 3:1 mix of 72: 73) Ac -Ac Ac-N N-Ac n-n^-(n- H H 74 N20, Ac Ac-N N-Ac ON -(n~ H H Ac NO 2) HN03 / 60 °C 3) H2S04 / 60 °C 4) H20 0,N- -NO, j2^~n1>N 02N-Ni?—^N-N02 i H H 75 (96%) N02 The oxidation of keto-olefin 62 to keto-acid 63 was optimized by a Lonza group en route to an undisclosed azasteroid (Scheme 5). A CCF design was employed, with 16 reactions including 2 center points.39 The authors noted that unless the poorly soluble starting material was micronized, the reaction was very sensitive to agitation and amount of RuCl3 catalyst. The continuous variables were solvent volume (300—600 mL normalized to a 100 mmol charge of 62), NaOCl equiv (3.6—4.4), and reaction pH (7.7—8.9), with an unspecified charge of RuCl3 catalyst. The exothermic nature of the reaction dictated the temperature be fixed at 5 °C. The responses modeled were yield, formation of three unidentified overoxidized side-products, and conversion (unreacted 62). The level of the side-products needed to be kept <0.1%. The data suggested that reaction concentration was not critical if the other two factors (NaOCl equiv and pH) were within their optimal range. A higher charge of NaOCl led to higher conversion with higher impurity formation. Since the starting material was effectively removed during the crystallization, the response of conversion of starting material was removed from the analysis, which predicted the optimal values to be 510 mL of solvent, 3.9 equiv of NaOCl, and pH 8.4. The expected yield of 78% was verified on a 1.5 mol scale and gave a 74% yield of 63. Subsequent production scale (12 batches, 225 mol each) gave yields of 73—75%, with the levels of the side-products and 62 at <0.03%. Halogenation and Nitration. A Pfizer group required a robust process for the synthesis of gem-difluoro compound 65 en route to an HIV protease inhibitor candidate (Scheme 6).40 The fluorination of ketone 64 produced the vinyl fluoride 66 in addition to desired 65, and various treatments were attempted to convert 66 to 65. The following reagent stoichiometry ranges were studied: fluorination with Deoxo-Fluor (1.6—3.0 equiv), DCE (0-0.68 mL/mmol of 64), water (0.1-0.3 equiv), Aliquat 336 (0.1 equiv), NaHC03 (0.2-0.6 equiv), and either spray-dried or amorphous KF (1—2 equiv). All the reactions were run in toluene (1.3 M) at 70 °C. Other reagents and additives studied included AICI3, 18-crown-6, and TBAF, all of which led to partial or complete degradation of 64. A fractional factorial design with 2 center points was utilized (18 reactions) with an HEL Duet 317 automated parallel reactor setup. The only factor that impacted the yield was the Deoxo-Fluor charge, with the higher amount (3 equiv) leading to better yields and helping to decompose 66, thus easing workup. Subsequent investigations showed that NaHC03 was required to buffer the reaction acidity; under optimized conditions, ~37% of starting material remained. Subsequently, the researchers found that a small charge of silica gel promoted fluorination in nonglass reactors, probably through the 1614 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 7. Hydrogenations Pd(OH)2 / C N^N'Bn (20%w/wrelto1) HO,, .^C02H ' Ph MeO,C CH, 89 (60 g, 0.5 mol) 1)90(21 mg) KOmu (5 mmol) H2 (4.5 bar) /PrOH 20 °C/2.5 h II OH 2) HCl (5 mmol), ch2ci2 3) distillation 87, R = H (85%) 88. R = OH (87:88 > 50:1) (xylyl)2 CH3 91 (96%, 99.0% ee) C02Et [(S>PhanePhos Rh COD]BF4 (0.1 wt%), H2 (4 bar) EtOH,~0.46M 40 "C/12 h 92 (120 kg) C02H generation of a fluorosilicate, and the reaction and workup were optimized as shown in Scheme 6. Unreacted 64 was selectively hydrolyzed and purged by extraction; a second treatment with NaOH hydrolyzed 65 and 66, leaving Deoxo-Fluor-related impurities in the organic phase. Treatment with NaOCl to decolorize solutions was found to decompose the carboxylic acid generated from 66. Through these fortuitous findings, extractive workups were used, and chromatographic purification was avoided. Ten batches were carried out on scale through fluorinations in a Halar reactor. In the chlorination of quinolone 67 catalyzed by sulfuric acid, the formation of the dimeric impurity 69 posed a concern (Scheme 6). In order to minimize its formation, DoE screening was used to examine the impact of reaction temperature (13, 21 °C), the equivalents of NCS (1.05, 1.2) and H2S04 (0.8, 1.2 mol %), water content (0, 0.5 vol %), solvent (MeOAc or EtOAc), solvent charge (2, 3 vol), and NCS addition rate (0.05, 0.3 vol/ min). Only 19 experiments were needed by using a 27-3 1615 fractional factorial design plus three center points. The formation of 69 was suppressed by charging higher levels of NCS, adding the reagent more quickly and by using dry solvents and lower temperatures. The hydrolysis of the azetidine intermediate was implicated by the impact of H20 on the formation of 69. Furthermore, by using MeOAc, a smaller excess of NCS was needed. DoE screening was then used to examine the robustness of the selected conditions to control the amount of 69: reaction temperature (13,21 °C), the equivalents of NCS (1.04, 1.07) and H2S04 (0.8, 1.2 mol %), and NCS addition period (30, 75 min). Ten experiments were carried out (24_1 and two center points). The first three factors significantly influenced the formation of 69, but the conditions were judged robust enough to afford batches of 68 with suitably low levels of 69. Upon scale-up in the kilo lab, the API (68) was generated in 91% yield as the N-Me-D-glucamine salt, with the level of the impurity 69 below the limit qualified in a tox batch.42 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Researchers at AstraZeneca used DoE studies to examine conditions for the bromination of 70 and subsequent hydrolysis and decarboxylation of intermediate 71 (Scheme 6).43 To optimize the yield of 71, five factors were varied: Br2 charge (0.75, 0.95 equiv), Br2 addition time (10, 60 min), 70% HN03 charge (15, 20 vol), reaction temperature (50, 65 °C), and reaction time (0.5, 4 h). A total of 11 experiments were run, 2s-3 plus three center points. Increasing the charge of Br2 gave a higher yield but negatively affected the quality. For safety reasons, the solvent volume was not decreased below 20 volumes. Although the DoE investigations did not improve the yield of 71 from the initial conditions, the process proved sufficiently robust to provide material needed for further studies. DoE studies were then used to investigate the conversion of 71 to acid 72. Six factors were investigated: the charges of NaOH (2.5, 4.0 equiv), H20 (6.75, 13.25 vol), HgO (1.08, 1.15 equiv), AcOH (1.0, 2.0 vol), HC1 (10, 20 vol), and a second charge of H20 (0, 12.5 vol). No conditions were developed that improved the ratio of 72:73 above 3:1, and 72 was purified by recrystallization from AcOH. The preparation of the energetic molecule HNIW (CL-20, 7S)44 was optimized using orthogonal array designs, a type of fractional factorial design (Scheme 6).4S Debenzylation/nitro-sation of 74 was carried out by treatment with N204, and after the sequential additions of HN03 and H2S04, the hexa(nitroamine) 75 was generated. Nine parameters were examined (equivalents of N204, HN03, and H2S04; temperature; time for nitrosation; acid addition temperatures; and addition times for the acids), varying four levels of each parameter. A full factorial design would have totaled 262,144 experiments (49), but using the powerful orthogonal array design, a total of only 32 experiments were carried out. The analysis showed that only the time (10 h) and temperature of nitrosation, and the sulfuric acid addition temperature impacted the yield. A 96% yield of 75 was obtained under the optimal conditions. Reduction. The debenzylation and in situ N-acetylation of hexaazaisowurtzitane derivative 76 to yield 74 was optimized using a reduced central composite design (CCD) (Scheme 7).46 The factors studied were the effects of Pd catalyst loading (8— 17%), reaction temperature (39—51 °C), hydrogen pressure (3.2—5.3 bar), and Ac20 stoichiometry (9.1 — 13.6 equiv) on reaction yield. The 21 experiments were performed on a 5 g scale in DMF (10 vol) and yields ranged from 8—70% over the design space. Of the four factors studied, only the hydrogen pressure did not have an influence on the yield. Higher catalyst loading led to better yields, as did temperatures around 48 °C and an acetic anhydride charge of 10.9 equiv. Bromobenzene was charged as a precursor to HBr, which catalyzed the debenzylation. The model predicted a 75% maximum yield, which was validated by a confirmatory experiment that gave 74 in 73% yield. A GSK team in Italy used DoE for both the reduction of imine 77 and the subsequent resolution of 78 to optimize the route to ketopiperazine 79, an intermediate in the synthesis of GW597599, an NK-1 antagonist (Scheme 7)47 Initially the hydrogenation of 77 with Pd in MeOH required prolonged reaction times, high catalyst loading, and high pressure. A catalyst change to Pt/C, followed by a DoE of the continuous variables (hydrogen pressure: 0.3—4.3 bar; temperature: 20—40 °C; catalyst charge: 2—10 wt %) enabled an improved procedure (Scheme 7) with complete reduction within 1 h. The reduction was telescoped with a crystallization-induced dynamic resolution (CIDR)48 and provided a 70% yield of the desired enantiomer as the mandelate salt on a 5 L lab scale; upon scale-up to the pilot plant, the yield dropped to 50%, in part because the batch adhered to the surface of the crystallizer. Thus, a full factorial design studying the effects of agitation rate (200—1000 rpm), water charge (0.10—0.45 equiv), and aldehyde charge (0.03— 0.10 equiv) was undertaken. The agitation most affected yield, implicating the importance of thorough mixing. Computer modeling suggested that poor agitation behind the baffles was responsible for the encrustation found on scale with the CIDR, and for that thick suspension a conical vessel without baffles was selected to avoid dead-mixing zones. The process was successfully scaled up in a pilot plant. GSK workers needed to understand the factors that led to the formation of the des-fluoro impurity in casopitant mesylate 49 This impurity was detected at 0.1% in intermediate 85, and controlling and minimizing its formation at no more than 0.15% were critical to meeting the API specification. The source was traced to the reduction of 80 (H2 over Rh/C) that generated intermediate 81 and not the subsequent hydrogenolysis to 83 (H2 over Pd/C) (Scheme 7). A full factorial design looked at the impact of 5% Rh/C loading (0.76—3.82%), hydrogen partial pressure (0.2-3.0 bar), EtOAc volumes (3-10) with two center points in a 10-reaction matrix Reaction temperature (set at 25 °C) was not included, as prior experience showed it would not have an impact on the impurity formation. Data analysis indicated that the high settings for the catalyst loading and hydrogen pressure led to higher levels of 82, while the solvent amount had no impact. These conditions, in concert with removing the Rh/C catalyst before the second hydrogenolysis, ensured control of impurity 84 and fulfilled requirements for control through QbD. A Merck Process group in the U.K. used DoE to improve the reductive ring closure of intermediate 86 to prepare indole 87 while minimizing formation of 88, the corresponding N-hydroxyl species (Scheme 7).S0 The formation of 88 was associated with stalled reactions, and the purity of 86 was also important for successful reactions. Initially, conversions were only about 50%, and so the researchers resorted to a fractional factorial design for rapid optimization. Using an Endeavor parallel hydrogenation apparatus, they studied the effects of temperature (40—75 °C), hydrogen pressure (10-70 psi), AcOH (0-50 vol %), and 10% Pd/C loading (10-20 wt %). High levels of AcOH and higher temperatures generally led to higher yields while pressures >50 psi correlated with lower yields. A model derived from the main effects proved useful in identifying reaction conditions for scale-up that allowed for decreased solvent charge and lower temperature. Researchers from Chirotech investigated the asymmetric hydrogenation of acetophenone (89) using either of two enantiomeric precatalysts containing ruthenium and a diphos-phine derived from a cyclophane to form 90, for example.51 The purpose of these experiments was to demonstrate that the hydrogenations could be carried out in high yield and with high chiral purity on a multigram scale to yield 91 (Scheme 7). A DoE screen was carried out, varying pressure (3, 13 bar), concentration of 89 (0.2, 0.4 g/mL), and temperature (10, 30 °C), while monitoring for stereoselectivity and rate; 11 experiments were carried out using a 23 design plus three center point runs. Under those conditions, there was slight variation in selectivity. Higher concentrations and higher pressures, in general, increased the selectivity and reaction rate, while higher temperatures increased the reaction rate but decreased selectivity. 1616 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Researchers from Dr. Reddy's Laboratories and Dow Chemical Laboratories jointly investigated the asymmetric hydrogenation of an amino acid precursor en route to 94 for the type II diabetes inhibitor denagliptin.52 After screening substrates, solvents, and precatalysts, conditions were selected for reduction of tetrasubstituted olefin 92 to 93 (Scheme 7). In early optimization studies, the researchers utilized a DoE for the hydrogenation, focusing on minimizing the charge of the catalyst [(S)-PhanePhos Rh COD]BF4. Using a substrate:catalyst ratio of 500:1, the variables were substrate concentration (20—100 g/L), H2 pressure (4—12 bar), and reaction temperature (20—50 °C). For these experiments, the chiral induction was fairly constant at 94—96% ee, so conversion was selected as the primary response. The main factor influencing the extent of conversion to 93 was the concentration of 92, followed by the H2 pressure. The model predicted the best conversions at high concentration of 92, high pressure, and high temperature. Subsequent investigations showed that inhibitory impurities were removed by recrystall-izing 92, allowing the hydrogenation to proceed reliably with lower precatalyst charges. Reductive animation of lactol 95 was used in the manufacture of the fesoterodine intermediate 96, in a route free of protecting groups (Scheme 8).53 The primary side products were the over- 97 98 99 reduced amine 98 and triol 99, which probably arose through the quinone methide intermediate 97. Through DoE studies, the Pfizer (Sandwich) group found that reduction using H2 and Pd/ C gave the highest selectivity, as compared to hydrogenation with other catalysts or reductive animation54 using NaCNBH3 or NaBH4. DoE was used to optimize the charges of NaBH4 and BF3-efher for the diborane-mediated reduction of secondary amide 79 (Scheme 9).SS In initial studies the mandelate salt was broken by treatment with aq NaOH/CHjCl^ and the solvent for the solution of the free base was displaced by THF; subsequent studies showed that 0.2 equiv of nBu4NBr could be added to dissolve the mandelate salt directly in THF. A CCD was carried out at 55—57 °C to assess the conversion to 100 after 5 h by varying the amount of NaBH4 (1.0,4.0 equiv) and BF3-Et20 (0.5, 3.0 equiv); 10 experiments total with two center points. The optimal conditions were ~3.5 equiv of NaBH4 and ~3 equiv of BF3-Et20. To control the diborane emissions, the reduction was carried out in a sealed vessel, and a second CCD was carried out. The results showed that at 55 °C, 90% reduction could be achieved using 3 equiv of NaBH4 and 2.7—3.0 equiv of BF3-Et20, relatively small decreases in the charges of those reagents. For safety considerations, commercially available BF3-THF was used in the optimized process. Researchers from Argonaut Technologies used DoE to optimize the preparation of oxazolidinone 103, beginning with the reduction of L-phenylalanine (101) to L-phenylalaninol (102) through borane generated from NaBH4 and iodine (Scheme 9). The product was treated with triphosgene to afford 103. A DoE was employed for the generation of 102, varying the charges of NaBH4 (1.1—2.0 equiv) and I2 (0.55—1.0 equiv) and reaction time (1—4 h). Eight experiments were run, along with two replicates. Runs with 0.89 equiv of I2 gave essentially the same yield as runs with 1.0 equiv, so the substoichiometric charge was employed. The response surface curve for the data showed that times longer than 1 h did not affect the yield and that there was a plateau for the yield between 1.6 and 2.0 equiv of NaBH4. A charge of 1.8 equiv of NaBH4 was chosen to ensure complete reduction within 1 h on scale, and the process was scaled up smoothly. A Pfizer/DSM collaboration applied DoE to develop a scalable synthesis of pyrrolidine 106, an intermediate57 in the synthesis of ingliforib, a candidate for treatment of diabetes (Scheme 9). Reduction of imide 104 to amine 106 in toluene with BH3 generated from NaBH4/BF3 proceeded in good yield, but generated 3-5% of ether 107. Red-Al (Na-(MeOCH2CH20)2AlH2) was not expected to generate 107 and was chosen as a safer and less expensive alternative to BH3-THF. The initial reaction of Red-Al (3 equiv) at 110 °C did not generate 107 but led to the formation of 27% of aminal 105, which was found to convert to 106 with time and more Red-Al. With extended reaction times the pyrrole byproduct 108 grew. A fractional factorial design (23_1) with 2 center point replicates generated only six reactions in studying the effects of time, temperature, and Red-Al charge. The responses measured were the purity of desired 106 and A% of intermediate 105. No details on the range for each variable were provided, but the desired product purity was strongly correlated with the higher temperature. An inverse addition was used for successful scale-up. Using DoE, the process was rapidly optimized, even before the structure of intermediate 105 had been established.58 Pfizer workers59 used an experimental design approach to gain insight into the reduction of ketone 109 to form the oxindole 110, an intermediate in the synthesis of ziprasidone hydrochloride (Scheme 9). The reactive chloroketone was known to be a skin sensitizer, so forEHS reasons the Friedel—Crafts acylation was telescoped into the reduction. Reaction of 109 with Et3SiH generated 110 and the intermediate alcohol 111, but controlling the amount of over-reduced alkane 112 was difficult. Aluminum chloride-mediated deoxygenation using tetramethyldisiloxane (TMDS) cleanly afforded 110. TMDS was preferred over polymethylhydrosiloxane, in part because of easier workup. Three factors were studied in the DoE: TMDS charge (2-4 equiv), reaction temperature (0—25 °C), and concentration of 109 in dichloromethane (6—10 vol). A full factorial design (23) was employed, plus one center point, for nine reactions. DoE analysis indicated that only the TMDS charge was relevant, but the optimum value was also dependent on the A1C13 charge from the previous step. This was reported to be the first commercially viable process. A team of Codexis/Pfizer collaborators investigated the process optimization of the biocatalytic kinetic resolution of racemic 2-methylvaleraldehyde (113) to prepare (R)-2-methyl- 1617 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 9. Reductions HN H,C NH 1)(r?Bu)4NBr(0.2eq.) THF, 40 °C to 25 °C 0 2)NaBH4(3.6eq.) HO. Y C02H 79 25 °C 3) BF3-THF (7.5 eq.) 35 °C/18 h 4) MeOH, A 5) A, - MeOH + /PrOH 6) HCI in /PrOH HN, H3C NH F ■ 2 HCl 100 (81.5%) Ph NH2 C02H 101 X oJ^o Bn 104,44.7 kg in PhCH3 CI 1) NaBH4(1.8eq.), THF 2) l2inTHF(0.9eq.) -5 to 5 °C/1 h 3) MeOH (5 eq.) 4) 20% NaOH (5 eq.) 5) A to 80 °C (- solvents) 1) 2 added to Red-AI* (4.0 eq.) in PhCH3 @ 20 - 35 °C 2) 110°C/4 h NH, OH Ph 102 N H (2.5 kg) 3) 50% NaOH CICH2COCI (1.2eq) AICI3 O (3.3 eqj ch2ci2 CH, X X^OH Bn 105 CI CI 109 1)KRED(0.09wt%) NADP (0.05 wt%) pH 7.5 buffer /-PrOH (45 vol%) 20 °C/20 h CHO - 113 2) EDTA (400 kg, 3) A, - /-PrOH 2.2 M) 4) NaHS03 114 (45% conversion, 98.5 % ee) 1) + CH2CI2 O 2) CO(OCCI3)2 (0.33 eq.) HN-X 3) 25°C/1.5h ~ - 4) - aq. phase 5) Na2S04 6) + /PrOAc, + heptane 7) A to 80 °C (- CH2CI2) 8) cool to 0 °C Ph 103 (74%) O O HO O' Bn 106 (93%. 99% GC purity) Bn 107 1) (HMe2Si)20 (2.0 eq) CH2CI2 (4 vol) CI 0 - 5 "C/4-6 h 2) H20, added over 3 h V Bn 108 CI N H 110 (87%) CI 111, R, =CI, R2 = OH 112, R, = H, R, = H pentanol (114; Scheme 9).6° Reduction was carried out with a ketoreductase (KRED), with i-PrOH as the bulk reductant. Three variables were studied in a screening design using two 12-place parallel synthesis carousels: loading of KRED (0.75—2.5 g/ L), loading of the cofactor NADP (0.04-0.13 g/L), and reaction temperature (15 °C—25 °C). The substrate concentration (3.5 mL/g), pH (7.4), stirring rate (100 rpm), and % IPA (58% v/v) were kept constant. Two responses, conversion and selectivity (% ee), were measured at 4, 8, and 23 h time points. All three time points showed that the KRED concentration was the most influential factor affecting the conversion, with higher KRED loading leading to higher conversion. Lower temperatures gave higher selectivities. The cofactor concentration had no impact on either the conversion or selectivity, allowing the researchers to decrease the charge of this relatively expensive cocatalyst. On scale, the reduction was quenched by EDTA, and the unreacted aldehydes were removed by codistillation and treatment with NaHS03. C—N Bond Formation. Roche researchers used DoE to optimize the formation of a protected nucleoside of L-ribose, 117 (Scheme 10).61 The mixture of triazole 115 and anomers 116 (in DCM) was treated with Ac20 to remove water, and the volatiles were removed under reduced pressure. A catalytic amount of triflic acid was added, and after about 4 h the mass was cooled and diluted with EtOH to crystallize 117. A central-composite DoE was used to assess the influence of temperature (105—135 °C) and charge of TfOH (0.7—3.0 wt %). Optimal conditions included 0.7-1.5 wt % of TfOH and 110-120 °C; at greater charges and higher temperatures, the reaction mass turned dark, and at lower charges and lower temperatures, the reaction was too slow. A GSK development team62 developed a robust SN2 displacement step in the synthesis of analogues of dopamine D3 receptor antagonist SB-277011. The reaction between chloropropylthiotriazole 118 and amine-HCl salt 119 to provide 120 was plagued by the formation of impurity 121 arising from intramolecular cyclization of 118 (Scheme 10). Impurity 121 was formed in 20—25 A% by HPLC, and the derived impurity 122, isomeric with 120, was difficult to purge by crystallization. The group simultaneously employed DoE and kinetic modeling to gain further process understanding, while maximizing selectivity and yield and minimizing unreacted starting materials. A full 1618 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 10. C—N Bond Formation C02CH3 N, J) H AcO^^/"" OAc 1) Ac20 (0.56 eq.) 90- 120 °C 2) F3CSO3H AcO OAc (0.84 wt%) 110-120 °C 4 h 115 (1.05 eq.) 116 (mixture of anomers) 3) EtOH 119 M N H ■ HCl Ox- N-N TEA (4.0 eq) _CI Kl(1.6eq) Me 118 (1.5 eq.) DMSO Ox-(2.3 vol) 65 °C N-N i % n Me Ox- Mg(OH)2 (2 eq.) Me3S(0)l (2 eq.), .0 H20 (1 eq.) 120 (94%) OCH, NHCbz NHR NMP 100°C/6h 123 ■■ H2C NHCbz NHR 124 C02CH3 N IJ N^ OAc AcÖ OAc 117(74%) © N-N V Me Ox 121 N-N N " Me Ox= H2C H3C H3CNX.OH NHCbz NHR 125 (92%) Ar —N. 122 factorial design with two center points (10 reactions) was performed using the Reactarray-SK233 parallel reaction tool. The four factors included the stoichiometries of triazole 118 (l.1 — 1.6 equiv), triethylamine (2.1—2.6 equiv), and potassium iodide (Kl) (0.2—1.6 equiv) and the reaction temperature (50— 70 °C). The reaction volume in DMSO was fixed at 4 volumes, the volume of the initial process. The results from the DoE study showed an adequate fit to the data and demonstrated that the high setting for all four factors, which corresponded to the previously existing reaction conditions, led to the highest yield within the design space. In other words, the DoE suggested no improved conditions. Some key observations were derived from the study, namely that the reaction conversion benefits from higher levels of KI, the selectivity benefits from high KI levels and lower temperature, and that higher temperature promotes formation of 121. A subsequent fractional factorial design, including the DMSO charge (3.6—4.4 volumes), was performed to assess the robustness around the optimal settings from the first design and determine whether more concentrated conditions would encourage the intermolecular alkylation. Unfortunately, the variability of the data implied a lack of robustness around the investigated design space, perhaps because of the stirrability of the thick reaction mixtures. This design, though, did suggest that the selectivity improves at higher concentration (less DMSO). Kinetics studies of the reaction showed that the halogen exchange reaction was the rate-determining step. The pi>CA's of amine 119 and TEA are similar, and traditional OFAT studies outside the original design spaces showed that a greater excess of TEA would benefit the formation of the free base of 119. Another round of factor screening and robustness testing of the optimal factor settings led to the final conditions outlined in Scheme 10 and confirmed a 94% yield with more robust operations, a 6% better yield than the original conditions. Merck researchers used high-throughput screening and DoE to maximize the N-methylation of raltegravir intermediate 123 and reduce solvent usage (Scheme 10). 3 DoE screening was used to examine the effects of reagent stoichiometries, reaction concentration, temperature, and reaction time. O-Methylation was a significant side reaction, but the researchers found that with extended reaction time and higher temperature the O-methyl impurity 124 was selectively demethylated back to 123, thereby increasing the ratio of 125 to 124. Alkylation conditions using Mel and Mg(OMe)2, higher concentration, higher temperature, and more equivalents of reagents favored formation of 125. To avoid handling the volatile Mel under scale-up conditions, trimethylsulfoxonium iodide was employed, which gradually decomposed to Mel under the reaction conditions. For complete in situ demethylation of 124, at least 1 equiv of H20 was necessary. The 92% yield from conditions shown in Scheme 10 was a substantial improvement over the initial yield of 70%. Condensations. An Astra-Zeneca group developed a one-pot, unsymmetrical Hantzch reaction of 126 to provide 129 (60% yield initially)64 en route to their chiral API candidate 130 (Scheme ll). Four variables were studied: equivalents of dione 127 (0.9-1.3), NH4OAc (2 X 2.5 to 2 X 3.5), and 128 (0.9-1.3), and solvent charge (20—35 vols). They employed a fractional factorial design (24_1) with two center points for a total of 10 reactions. The responses were yield of 129 and the amount of tricyclic impurity 132, originally observed at 15—30%. The reaction concentration and the dione charge had large effects, and they were observed to have an interactive effect as well. The dione charge had a larger impact at higher concentrations than at lower concentrations. But in either instance, a higher dione charge led to a higher yield of 129, but it also led to formation of higher quantities of impurity 132. In order to strike a compromise between reaction throughput, maximizing 129 and minimizing 132, the group settled on the conditions shown in Scheme 11. These changes minimized formation of diester impurity 131 (2—4 A% originally), eliminated the need for chromatography and simplified the isolation. The process was demonstrated to manufacture over one ton of 129 in 58% yield with <0.1% of impurities 131 and 132. 1619 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 11. Condensations o 1) NH4OAc 3) CN (2.5 eq.) J EtOH(18vol) HCTXF. OR CHO 126 (1.0 eq.) 2) & (128,1.1 eq.) ^ NH4OAC (2.5 eq), (127, 0.9 eq,) EtOH(11vol) EtOH(6vol) A/35h R = allyl C02R CCf-3 H OH 129 (58% racemic TBSO 0= OR 133 (r = allyl) N K 1) CH3P(OEt)z *~CH3 (2.3 eq., 2 h addition) PhCH3 (30 vol) 90 °C/2 h 2) aq. washes 3) + fPrOH, A, - PhCH3 h3C NMe2 AcOH 135 <3^> N K H3C N h 136 y h RO 134 (67%, 60%isolated) CH3 H3C^N 137, R = H 138, R = I (67' Nal04, l2 CH3 H3C X,? 139 MeO Ar,CHO (1.0 eq.) Q Aliquat®128 (10mol%) \s -► HN KHC°3 NAr K2C03 144 aq 2-MeTHF 70 "C/2.5 h "Ar! HN "Ar 'Ar 145 146 (145 : 146 < 1 : 98) N 'Ar, HN, CI3CC02H AcOH 80°C/1.5h O HN Ar, 147(92%) The cyclization of 133 to form thiopenem 134 was optimized using a 24 full factorial design with four center points (20 reactions) (Scheme ll). The four factors investigated were reaction temperature (60, 90 °C), concentration (15, 45 volumes), phosphinite stoichiometry (2, 3 equiv), and phosphinite addition time (l, 5 h). The responses quantified were in situ yield at 5 and 21 h. The Pfizer group found that the optimal temperature was about 90 °C, with incomplete reactions at lower temperatures and thermal decomposition at higher temperatures. More dilute reaction conditions were preferred, as expected for an intramolecular reaction. Phosphinite stoichiometry and addition times had relatively little impact on the outcome of this carbene-mediated cyclization. Under optimized conditions, the sulopenem intermediate was isolated in 60% yield, with impurities being effectively purged by crystallization from iPrOH. While the yield was only marginally improved over the scouting reaction, the DoE provided the information necessary to reduce the addition time and the time cycle of the manufacturing process.65 To optimize the formation of imidazole 137 from amidine 135 (Scheme 11), a group from Roche used DoE to study the effects of propargylamine equivalents, temperature, and time.66 Eleven experiments were performed, including three center point runs. When more than 1.25 equiv of propargylamine was charged, the yield of 137 dropped off sharply. Using flow NMR, the group found that amine exchange to form 136 occurred without heating, and imidazole 139 was identified as arising from excess propargylamine. Subsequent experiments showed that cone HC1 was superior to AcOH, formic acid or TFA, and the preferred conditions using cone HC1 were essentially the optimal conditions identified in the DoE. For isolation, 137 was converted to the crystalline iodoimidazole 138. Naphthyridine 143 was found in ~25% yield as a byproduct in the preparation of 142 from 140 and 141 in ethoxyethanol at 130 °C (Scheme ll). When additional 142 was needed for studies, Wyeth researchers used DoE to optimize its formation.67 Four factors were considered in a full factorial design (24 runs plus three center point runs): the charges of 141 (l, 4 equiv) and dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 1620 Organic Process Research & Development Review Scheme 12. Robinson Annulation OMe cone HCI 0.16 eq) H20 OMe 80 °c 30 min (1.25 eq) OH OH BnO 1 (148, 0.21 mol) NaOAc (3.5 eq.) CO,H cone HCl (0.17 eq) O C02H (1.4 eq) H20 5-45 °C (over 2 h) ( OBn 149 (90%) OBn 150 Scheme 13. Palladium-Mediated Couplings O Pd(OAc)2 (0.5 mol%) (nBu)4NBr(1 eq) NaOAc(1.5eq.) 151 (1.5eq) DMF(0.10M) 120 °C/24h PIT Ph 152 (99% by GC-MS, 89% isolated) Pd(OAc)2 (0.1 eq.) P(o-tol)3 (0.25 eq.) HC02H (1.1 eq.) piperidine (7.0 eq.) DMF, 90 °C / 3 h C02CH3 159 (71%) COoCH, catalyst pyr-HCl (0, 5 equiv), temperature (50, 130 °C), and time (8, 24 h). (Some previous experimentation had shown that reaction concentration had little impact on the formation of 142.) Analysis of the results clearly showed that higher temperatures and higher charges of 141 and pyr-HCl gave higher yields of 142, and indicated that the optimal conditions might be outside of those initially chosen. Accordingly, a second DoE was carried out for 8 h at 130 °C, the reflux temperature of the solvent, varying the charges of 141 (1 — 12 equiv) and pyr-HCl (0—9 equiv). With charges of 9 equiv each of 141 and pyr-HCl, the yield was about 70%, but purification would have been difficult given the large excesses of those components. The researchers found that higher temperatures, accessible in microwave reactors, generated 142 in about 80% yield with considerably lower stoichiometrics. GSK workers68 sought to optimize the PTC-mediated aldol condensation between an aromatic aldehyde and thiazalone 144 1621 to generate aldol adducts 145 and 146, which upon acidification yielded olefin 147 (Scheme ll). They employed a D-optimal design to probe the impact of the qualitative factors of solvent (2-BuOH, toluene, 2-MeTHF, MIBK, i-PrOAc), Lewis acid (none, Sc(OTf)3, Y(OTf)3, Yb(OTf)3), base (50% NaOH, 2 N NaOH (with and without brine), 50% K2C03,25% KHC03), and phase-transfer agent (Aliquat 128, (n-Bu)4NHS04, BnEt3NCl). The values for the following continuous variables were fixed: reactant stoichiometry (l:l), reaction concentration (0.25 M), PTC charge (10 mol %), Lewis acid charge (10 mol %), and reaction temperature (60 °C). The collecting and collation of the HPLC data were accomplished with the iChemExplorer software. The regression analysis revealed that Aliquat 128 andYb(OTf)3 were beneficial for high conversion, while i-PrOAc, 2-MeTHF, and 2-BuOH were the most desirable solvents. They found that the aldol reaction was driven by the crystallization of the enol 146. Subsequent experiments revealed that pH 11.5 was optimal, dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 14. Coupling with Grignard Reagents TBDMSO TBDMSO 1) N2; H20 (0.8 eq.) H2C=CHC02Et(1.79 eq.) NiCI2 ■ 6 H20 (18 mol%) R Zn (4.2 eq.), THF/pyr 60-70 °C/30 min; cool, filter 2) 163, in THF/pyr; 24 "CI 2.5 h 3) acetone; H20 R-CH2CH2C02Et (164, 96%) 166(1.2 eq. Mg (1.4 eq) l2 (2 mol%) Br - THF 60 °C/2 h NC CI(CH2)3NMe2-(2.27 eq) CIMg, MgBr A V ch2ci2 F _ -5 "CI 167 40 min p ó 165 (25 kg) ß Bn0^^\| 169 1) (20.7 kg, in THF, CIMg \ added over 1h) (1.5eq.) -10 to °C CuCI (5 mol%, 2) MeOH in THF) 3) 2 M HCl; 1 h 4) MTBE, aq. Na2S203 5) H20 Mg (3.0 eq) l2 (3.2 mol%) toluene/THF 90 °C/3 h ■v ,NMe2 0 °C /1 h OH 170 (94%) OH BnO^ OH 171 (<2%) NMe, which led to the use of 1:1 saturated aq KHC03/K2C03 with 2-MeTHF. Yb(OTf)3 was eventually eliminated, as it had no effect at pH 11.5. Acids with pK, 0—1.5 were optimal for the dehydration of the 145/146 wet cake, and trichloroacetic acid in AcOH was used on scale. Under optimized conditions, olefin 147 was produced in less than 6 h, a marked improvement over the original conditions requiring at least 3 days. Screening was carried out in microreactors with minimal starting materials (3 g) and completed within 2 weeks. Researchers from GSK used a DoE study to optimize a Robinson annulation to generate tropinone 149 (Scheme 12).69 Amine 148, prepared from ethanolamine, was charged as the limiting component. In a total of 10 experiments, seven factors were examined in the DoE: the stoichiometries of the dimethoxy-THF, diacid, hydrated bis-aldehyde, HC1, and NaOAc; the addition temperature; and the rate of heating. The goals were complete consumption of 148 with minimal formation of the pyrrole impurity 150. Lower charges of the dimethoxy-THF (1.05 equiv) led to incomplete consumption of 148, while higher charges (1.55 equiv) increased the amount of 150 formed. Reducing the charge of NaOAc from 4.0 equiv to 3.5 equiv was also beneficial in minimizing formation of 150. Metal-Catalyzed Cross-Couplings and Metal-Mediated Reactions. Aggarwal and co-workers at the University of Bristol and GSK70 optimized the Heck coupling between trans-cinnamaldehyde (151) and iodobenzene (Scheme 13), which had been reported to proceed in only 56% yield (59% conversion). They employed a 2S_1 fractional factorial design with 4 center points (20 reactions; 1 mmol scale) to explore the impact of solvent charge (1.6—5.0 mL), temperature (50—90 °C), Pd catalyst loading (0.5-5.0 mol %), NaOAc charge (1.5-4.0 equiv), and reaction time (4—9 h). The analysis revealed that a high temperature was the most relevant setting to attain a high yield of 3,3-diphenylacrylaldehyde (152), followed by higher catalyst loading and higher solvent charge; there was a significant interaction between temperature and concentration, which would have been overlooked by OFAT experiments. The NaOAc charge and reaction time had no impact on the yield. Yields in the design space ranged from 5—75%, and curvature was detected from yields at the center point conditions, so another DoE was carried out using an RSM approach. Data from the first run showed that good yields could be achieved using low catalyst charges. The second DoE explored some conditions outside the first DoE: higher reaction temperatures (90—120 °C) and more dilute conditions (5—10 mL). With the exploration of that new design space, the authors included reaction time (8—24 h) and low catalyst loadings. The RSM design, a face-centered central composite (CCD) with 30 experiments, led to improved yields, mainly due to the higher temperature. At the lower Pd loading, high yields could only be obtained at the more dilute conditions (99% yield). The authors rationalized that the higher temperature leads to faster rates and the low Pd loading under higher dilution extends the catalyst lifetime, allowing for full conversion. In 3 weeks the authors improved the reaction yield from 56% to 99%. Merck researchers used DoE to optimize a Suzuki coupling72 following a HTS to identify the optimal solvent, base, and ligand. The design studied temperature (35—65 °C) and the ratio of aryl boronic acid 154 to substrate 153 (0.8 to 1.4), while monitoring conversion, regioselectivity, and yield of 155 (Scheme 13). A central composite design was used, and 13 experiments were run. As expected, higher temperatures gave higher conversions but poor regioselectivity, forming undesired 156 and 157. 1622 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 15. Mannich Reaction HCI ) H2C=N(fBu) <5ecl) HCl 2-butanol 40 °C/3 h 2) 5 M HCl 172 (13.1 kg) NH(t-Bu) 173 (70 A%, 59% isolated) NH(t-Bu) 174, R-i = CH2NHfBu, R2 = H 175, R, = H, R2 = CH2NHfBu (on scale 27A% combined) 176 Nt-Bu Scheme 16. O-Alkylation OH ArHN N Ö 177 (72.5 kg) N CH; 178(1.75 eq.) 1) NaOH (s, 3.6 eq.) PhCH3 106-9 °C/9h 2) 75 °C, aq. washes 3) A, - PhCH3 4) + heptanes ArHN H°-^OCH3 1) (8eq.) Cul (0.1 eq.) NaOf-Am (7 eg.) diglyme-DMF 90 °C/2 h 2) filtration 3) HCI 1) NaH (1.2 eq.) Et20, 30 °C/30 min 2) A, - Et20 3) + PhCH3, 24 mL/g ^C02Et (1.0 eq.), 70 °C/3h 5) + AcOH, RT Q^C02H OEt 188 ^O 189 CI 182 (75% from 2b) 0CH3 183, Y=Br 184, Y = H (< 2.5% AUC) 185, Y = OCH3 (~1%AUC) ? OEt 0 187(52%) -O O 190 Eventually, (o-tolyl)3P was chosen over other ligands, as it was cheaper in bulk, gave faster conversions, and required less 154. In the Pd-catalyzed reductive cyclization of 158 to 159 (Scheme 13), significant amounts of hydrogenolyzed impurities 160, 161, and 162 were formed.73 For optimization, five factors were examined by DoE: equivalents of phosphine ligand (0.1, 0.3), HC02H (2.0, 4.0), and piperidine (5.0, 7.0); charge of DMF (10% and 30% v/w); and temperature (50, 90 °C). (The charge of Pd was held constant at 0.1 equiv.) On a 100 mg scale, 17 experiments were run using a half factorial (2S_1) design plus one center point. Higher temperatures and less HC02H were found to be critical for good selectivity, as shown by the half-normal plots. Based on this information, the HC02H charge was reduced to 1.1 equiv. A follow-up RSM design indicated that the maximum yield was 80%, and the optimal conditions were found to be experimentally robust. Scale-up was expected to be straightforward for these homogeneous reaction conditions. A team from Hoffman-LaRoche evaluated the reductive coupling between iodide 163 and ethyl acrylate, which affords 164, a calcitriol precursor (Scheme 14).74 The robustness of the process was a concern for scale-up, as no recourse was available if the catalyst was not regenerated successfully. Water content was expected to be critical for turnover of the Ni(0) catalyst, and high charges of ethyl acrylate were expected to afford lower yields due to polymerization. Zinc and NiCl2-6H20 were handled in a glovebag to preclude any deactivation of the catalyst by 02. A full factorial design with three variables and one center point (9 reactions) was chosen. The variables were additional water (3—6 equiv relative to NiCl2-6H20), ethyl acrylate (1.5—2.2 equiv), and the NiCl2-6H20 charge (0.125—0.25 equiv). The only factor in the DoE study that significantly affected the yield was the ethyl acrylate charge, which was then set at 1.85 equiv. Yields were reproducible and approaching quantitative, with ethyl acrylate charges of 1.85 equiv (±5%). Through DoE, the optimized process was found to be fairly robust over the entire design space. 1623 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Using DoE, a group from Dr. Reddy's Laboratories and the Institute of Science and Technology optimized the process to diol 168, a key intermediate in the synthesis of citalopram.75'76 The one-pot route involves formation of two Grignard reagents (including 167) followed by their sequential reactions, starting with benzolactone 165 (Scheme 14). The full-factorial screening design of the first step studied the amounts of 166 (1.2—1.5 equiv vs 165), Mg (1.2—1.5 equiv vs 165), and iodine initiator (l —10 wt %). The resulting 12 reactions (including 4 center points) were assayed for yield, purity and residual 165. Similarly, the second Grignard reagent was optimized with respect to equivalents of l-chloro-3-(dimethylamino )propane (1.8-3.0), Mg (2.0-4.0) and iodine (1-10 wt %). Both of the designs were augmented with additional eight reactions to generate a RSM design. Based on the assay yield and purity requirements (>75% and 98%, respectively), design spaces showing the acceptable operating limits were generated indicating a wide range of acceptable conditions. Three validation experiments were then performed to corroborate the model predictions and showed excellent agreement between the prediction and the actual outcome. A group from Merck (U.K.) sought to optimize the copper-catalyzed reaction of epoxide 169 with commercially available vinyl magnesium chloride (Scheme 14).77 On initial scale-up they observed that up to 20 A% of chlorohydrin 171 was formed, due to prolonged Grignard addition time. By applying an inverse addition mode, the levels of 171 were suppressed (<1%). A DoE was undertaken to gain further insight into the robustness of this reaction. A 23_1 fractional factorial design with two center points (10 reactions) was performed in THF with CuCl (20 mol %) to examine the effects of temperature (—20 to +20 °C), Grignard stoichiometry (1.5—2.5 equiv), and epoxide addition time (15— 105 min). The response monitored was formation of 170, which ranged from 0.2 to 40% over the design space. The reaction temperature was the dominant factor affecting the outcome, while the other two factors had minimal impact. The reactions run at —20 °C gave lower levels of the chlorohydrin. Additional univariate optimization revealed that the reaction temperature could be raised to —5 °C under the conditions shown in Scheme 14. By quenching with MeOH and then HC1, emulsions and solids were avoided, and the Na2S203 wash reduced the level of Cu in the product. Aryl Alkylation. Carbogen researchers78 optimized the final step in the synthesis of antimalarial agent GSK369796 (173), via a Mannich reaction on phenol 172, initially performed in MeOH (Scheme 15). Driving the reaction to completion was important, as unreacted 172 precipitated from alcoholic reaction solvents and complicated workups. Byproducts were the bis-Mannich adducts 174 and 175 and the benzoxazine 176, which formed with extended reactions. DoE studies in i-PrOH or 2-butanol, chosen for improved product stability, showed that more dilute conditions (35 vol), low reaction temperature (40 °C), and 5 equiv of the Schiff base raised the yield of 173. After Mannich reaction in 2-butanol and acidification to isolate 173, impurity 176 was found to be soluble, so the optimized process used 2-butanol as solvent, avoiding chromatographic purification. These improvements raised the isolated yield from ~30% to 46% with a purity increase from ~80 A% to 99.9 A%. O-Alkylation. An SNAr displacement of 177 was the last step in the manufacture of API 179 (Scheme 16).79 The goals of the optimizations were to increase the isolated yield and provide more robust processing conditions, while changing the solvent from f-AmOH to toluene. Two levels of four variables were examined: equivs of base and alcohol 178, H20 charge, and temperature. A fractional factorial design DoE was run for a total of 10 experiments (24_1 + 2 center points). The AstraZeneca researchers found that more than two equiv of base were not necessary and that charging more equiv of H20 than base decreased the rate of reaction, by hydrating the base. Increasing the reaction temperature or the charge of alcohol 178 increased the rate of reaction, but with the drawbacks of generating more impurities or increasing the overall cost, respectively. Process optimization raised the yield of 179 from 63% to almost 87%. Researchers at GSK employed DoE screening to optimize the preparation of API 182 from dibromo intermediate 180 (Scheme 16).80 Under metal-free conditions, the initial substitution to 183 was rapid, but conversion to 182 required extended reaction times; hence, metal-catalyzed conditions were studied to shorten the overall reaction time. A mixture of DMF-diglyme was chosen as the solvent, as some DMF was necessary for the copper-catalyzed reactions and diglyme was inert under the reaction conditions. The factors varied were the charges of Cul (0.1, 0.7 equiv), NaOtAm (6, 7 equiv), and methoxyethanol (8, 10 equiv); temperature (80-90 °C); and time (2—16 h), while screening for short reaction times, the yield of product 182, and minimal formation of the hydrodebrominated side product 184. (impurity 184 was problematic because it was not completely purged from 182 in routine crystallizations; impurity 185 arose from reaction of MeOH, formed by decomposition of methoxyethanol under the basic conditions.) On a 50 L scale, using the optimized conditions, the yield of 182 was 75%, with <0.1 ppm of Cu in isolated 182. Due to the concerns about minimizing the concentration of copper in the workup streams and controlling the amount of 184 formed, research on 180 was discontinued in favor of reactions with the bis-fluoro intermediate 181, which did not need metal catalysis. The alkylation of phthalimidoylethanol 186 with ethyl chloroacetate was examined by DoE (Scheme 16).81 The alkoxide of 186 was generated in Et20 using an excess of NaH, and the ether solvent was displaced with toluene. In the first investigations, the factors examined were temperature (24, 70 °C), concentration (24, 48 mL/g), and reaction time (3, 27 h); a total of 11 experiments were run, a full factorial design (23) plus 3 center point runs. The highest yield of 187 was obtained by extended reaction at a high temperature under dilute conditions, with the latter probably serving to decrease the formation of the transesterification side product 190. (Compounds 188 and 189 were mentioned as potential side products.) Subsequent investigations did not increase the yield beyond the 40% achieved from the first DoE. As the reaction mixtures darkened with extended reactions, a second DoE screen was performed, holding experiments at 70 °C over 3 h. The factors examined were charges of NaH (l, 1.2 equiv) and ethyl chloroacetate (l, 1.2 equiv), sonication (none, 1 h), and the presence of catalytic KI (none, 5 mol %) or catalytic n-Bu4NBr (none, 5 mol %). A fractional factorial screen was set up comprising 11 experiments: 2s-2 + 3 center point runs. Analysis showed that the small excess of NaH raised the yield, but the addition of the phase-transfer catalyst lowered the yield and increased byproduct formation; the other factors had no effect. Through these investigations, the yield of 187 was raised from 25% to 52%. Flow Chemistry. The value of microreactors for quickly optimizing conditions has been noted,82 and approaches to making large quantities of material under continuous flow conditions have been reviewed.83 With continuous flow reactors, it is necessary to optimize the average residence time, T, which is 1624 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 17. Flow Chemistry Cv^CH3 H2N OH J V (5eq.) H3C^N^CH H3C MeOH, 20 °C t = 100 sec N ^3 ^ HN HI04 (0.31 eq.) OCH3 H2S04 (1 eq.) 191 (99%) 192 aq. CH3CN NH, I PITXH3 193 PrT CH, TBSO" ^ "OTBS TBSO1' ^ "OTBS (194, 5 mg/mL) (195, 96 A%) 10 N NaOH Na O d-5,eq.) N N' II O PhS ~NH2 -'-- CH 70 °C t = 2 min 1.5 eq.; NH, 80 °C (> 99% conversion) t = 4.0 min + ^ 50% aq H2NOH (4-7 eq) CN 196 starting materials dissolved in NMP ^—SPh 197 (90%) (in /PrOH) 100-125 °C t = 5-9 min (20 bar back pressure) N-OH NH, 198 (100% conversion) Me' H2 (1 atm) Raney-Ni - MeOH "Y° (0.05 M) (CH2)3C02Me 5 mL/min 199 70 °C (250 g) O cone. HCI NH N-^OH Me (CH2)3C02Me_ 200 immobilized R. miehi (X-Cube®) N—^ M~" OH HO' "-(CH2)16CH3 203 60 °C 0.4 mL/min 204 (1 eq, 72 mM t = 40 sec 205 (95% conversion) in heptane) Me' (CH2)3C02Me Me (CH2)3C02Me 201 (85%) 202 0^,(CH2)16CH3 O Br 1) n-BuLi (1.2 eq)/hexane THF/-65°C/3.8 min HCl ~ "HP* T 2)B(OMe)3(1.5eq) OMe -65 °C/4.6 min B(OH)2 analogous to the reaction time in batch operations; for this reason, continuous flow reactions have been grouped together. In optimizing the preparation of pyrrole 191 from a Paal— Knorr reaction (Scheme 17) using microreactors, a Dutch group used DoE, focusing on complete conversion with minimal reaction time.84 They reported four levels each of stoichiometry and length of reaction, with three levels for reaction temperature. A total of 58 experiments were run, with an average run time of only 25 s. Essentially, a quantitative yield was achieved, without an acid catalyst. Microreactors were used to screen conditions to remove a poro-methoxyphenyl (PMP) group in amine 192, using HIO4/ H2S04 (Scheme 17).ss Using a D-optimal screening design, 51 reactions were carried out in less than 6 h, consuming approximately 10 mg of 192. Data showed that high temperatures gave faster conversion to 193 (>99% conversion at 90 °C/ 1.3 min), but in a larger microreactor, the boiling solvents (H20, acetonitrile) produced an unsteady flow. Accordingly, a reaction temperature of 80 °C was successfully adopted. Genzyme researchers sought to optimize the continuous photoisomerization of 194 to 195 (Scheme 17), an intermediate in the synthesis of doxercalciferol, a treatment for chronic kidney disease.86 For efficient conversion they used a continuous photochemical reactor, with solutions of 194 being pumped through tubing wrapped around a photolysis cell. They conducted a 24 full factorial design looking at the effects of temperature (10—30 °C), flow rate (5.5—21.8 mL/min), substrate concentration in heptane (5.0—50.0 mg/mL), and the photosensitizer 9-acetylanthracene (9-AA, 4—16 wt %). This approach with 3 center points led to 19 reactions, measuring A% of 195 as the output. A full factorial design was employed because only a few minutes were required to run 100 mg reactions and isolate the crude product. Conversions varied from 62—96% over the design space. The goals were to test the robustness around dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 1625 Organic Process Research & Development Review Scheme 18. Crystallization and Resolution 1) PhNHCH, OH 'DCC C02H PhCH3 80 - 85 °C aq. NaOH Me Ph ,CH-. NaO N N Me L-tartaric acid (0.86 eq.) MeOH : EtOH (30 : 70) 35 L / kg 60 °C to 5°C 1) 5MHCIto pH 1.0 + 0.2 (added over 15-30 min) 30 + 2 °C 2) age at 30 + 2 °C 2 h (no agitation) 3) centrifuge, dry Ph ,CH3 HO N N Me 207 (76 - 77%) recrystallize 4 times HO^XOjH HO" C02H 210 7.1 kg, 50% ee 210 3.9 kg (55%) 97.0 % ee ONa ®CI NH2 X - H,N Me Prf C02H 211 (66-71%) 213 (1.5 kg, 214 89% pure) (1.32 eq.) 1) H20 2) A, - /PrOH NH? NH? 3) NaOH to pH 9, filter A 4) NaOH 80 °C /1 h 215 5)nBuOH (75%, 93% pure as residue) ŕPrOH - PhCH3 (65:35 wt:wt) ZnCI2 (0.26 equiv) 82 °C/16h CHO HN NH, N Me the center point values and to minimize the use of 9-AA, which was anticipated to be troublesome to remove from 195. Data analysis showed that temperature and 9-AA charge had little bearing on the outcome of the reaction; thus, the charge for the latter was set to 4 wt 96. The flow rate and concentration interacted, with high conversions at high concentration/low flow rates and low concentrations/high flow rates. Subsequent work showed that the 9-AA charge could be lowered to 2.5 wt % and was removed by continuous filtration through a bed of carbon and Celite. Overall, the 96 A% outcome was the same as the original conditions, but the 9-AA charge was reduced from 10 to 2.5 wt % and the process was demonstrated to be robust over a wide temperature range. Wyeth researchers employed DoE to optimize the preparation of aminotriazole 197 (Scheme 17).87 Due to safety concerns, continuous operations were selected, and the /?-azidoethyl phenyl sulfide 196 was chosen as a safer alternative to ethyl azide. In the DoE study, the factors varied were temperature (65—95 °C), and charges of cyanoacetamide (1.5, 2 equiv) and NaOH (1.5, 2 equiv). In this design, the 16 experiments were carried out monitoring the formation of 197 and the amount of unreacted cyanoacetamide. Yields of 197 fell off at higher temperatures due to the side reaction of NaOH with cyanoacetamide. Hungarian co-workers applied continuous flow technology to the reaction of benzonitrile with 50% hydroxylamine to afford the corresponding amidoxime (198; Scheme 17), a precursor to oxadiazoles.88 Hydroxylamine is known to decompose exo-thermically with the generation of gases, and the temperature for thermal decomposition is lowered significantly in the presence of metal ions. A microreactor was chosen for process optimization due to the inherently greater safety from heating only small quantities of reactants at any moment, and because the glass surfaces of the microreactor were essentially free of iron and other metals. Optimized conversions were probed looking at the effects of reaction temperature (75-125 °C), H2NHOH stoichiometry (1—7 equiv), and residence time (1—9 min) using a central composite design (19 reactions). Data analysis with the MODDE software revealed high conversions (>97%) could be attained over a wide range of factor settings with a 5 min residence time, namely at 125 °C with 4 equiv of H2NHOH or at 100 °C with 7 equiv of H2NHOH. The DoE runs were performed with a 10 \i~L reactor, and these optimized conditions worked smoothly in a 450 \i~L reactor. Teva process chemists applied fractional factorial design to identify the critical process parameters in the continuous flow synthesis of benzimidazole 201 (via 200), an intermediate in the synthesis of bendamustine hydrochloride (Scheme 17).89 The original process, a batch mode hydrogenation of bis-nitro arene 199, generated high amounts of the N-oxide impurity 202 through incomplete reduction. Batch hydrogenation also posed safety concerns, and specialized reactors would be required on scale to maintain a hydrogen overpressure and to dissipate the heat from the very exothermic reduction of 199. Continuous flow reactors offer advantages through ready heat transfer and efficient mixing; with the H-Cube continuous flow hydrogenator, hydrogen is generated on demand by electrolysis of H20.90 Initial continuous flow hydrogenation was developed using the dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 1626 Organic Process Research & Development Review H-Cube with Raney-Ni as the catalyst and methanol as solvent. Using the maximum rate of generating H2 by the H-Cube, optimization of the flow conditions focused on the following parameters: temperature (20—100 °C), system pressure (1—50 bar), flow rate (0.3—3.0 mL/min), and concentration of 199 in MeOH (0.01—0.06 M). Higher concentrations were not attainable due to limited solubility of 4 in MeOH. The nine reaction study (l center point) was analyzed with Design-Expert software and revealed that the concentration and flow rate were the most impactful factors. The other two factors, temperature and pressure, had little or no impact on yield. With low concentration (<0.025 M) and low flow rates (<0.5 mL/min), the conversion was high (>95%). The optimal conditions (0.02 M, 0.4 mL/min flow at 50-70 °C and 1 -5 bar) resulted in a 97% conversion, consistent with the DoE model prediction. Using data from the DoE for the H-Cube and exploratory experiments with the H-Cube Midi, five hydrogenations were scaled up with the H-Cube Midi using the conditions shown in Scheme 17. de Souza and co-workers applied DoE to optimize a biocatalytic conversion by continuous flow through an immobilized enzyme in an X-Cube reactor.91 The esterification of (B,S)-l,2-isopropylidine glycerol (203) with stearic acid (204) was catalyzed by immobilized lipase from Rhizomucor miehi, providing 205, a precursor to a monoacyl glycerol (Scheme 17). Initial optimization provided 87% conversion at 60 °C at 0.6 mL/min. A 23_1 fractional factorial design (3 center points; 7 reactions) examined the effects of temperature (40—60 °C), substrate concentration (35-100mM), andflowrate (0.2-3.0 mL/min), and analysis showed that only temperature had no bearing on the outcome. Thus, temperature was fixed at 60 °C in a subsequent RSM design (3 center points; 11 reactions). As the first design suggested that generally lower flow rates and higher concentrations were beneficial, the ranges in the RSM design were 0.4—0.8 mL/min and 72—95 mM, respectively. The RSM showed that yields fell off sharply from the optimal conditions. A University of Bergen team developed a multijet oscillating disk minireactor system for performing flow chemistries at cryogenic temperatures. They employed DoE to optimize the continuous synthesis of phenylboronic 206 acid via the lithiation of 4-bromoanisole and quenching with methyl borate (Scheme 17).92 Preliminary experiments had indicated that total residence time and temperature influenced the reaction yield, which was maximized at 71%. A two-factor full factorial design with two center points led to six experiments over the ranges of —70 to —60 °C and 7.3 to 10.95 min for combined average residence times. As expected, both variables were relevant, as was the two-factor interaction term, suggesting that, at the warmer temperatures, shorter reactions were favored, while, at colder temperatures, longer reaction times led to higher yields. In an effort to increase the yield, they extrapolated outside the design space, into a region that suggested that at —75 °C and 10.9 min total residence time that >90% yields could be attained. A reaction at those conditions provided a 69% yield, probably because other variables were not considered. In principle, an output of over 2 kg/day could be achieved under these conditions. Crystallization and Resolution. A team from DuPont Chemoswed required batches of Roquinimex (207, Scheme 18) to have uniform physical properties to afford a suitable dissolution profile (>90% after 1 h).93 Crude amide 207 was prepared by a coupling with dicyclohexylcarbodiimide (DCC) and purified by filtering off residual dicyclohexylurea (DCU) from a solution of the sodium salt. To crystallize 207, 5 M HC1 was added to pH 1. A distinct relationship between the particle size and good dissolution was detected with larger crystals. A 23 full factorial design with 3 center points (ll reactions) of the crystallization variables was performed by studying the influence of temperature (20—30 °C), acid dosing period (15—30 min), and concentration (6—11 volumes water). The single response was the dissolution rate (% after 1 h) and ranged from 63—93% over the design space. The data showed that larger crystals were produced at higher crystallization temperatures and with longer dosing times, as expected. All three factors were statistically relevant, with the higher aging temperature being the most impactful, followed by longer dosing time and lower dilution. A relevant interaction between concentration and dosing time indicated that, at shorter dose times, more dilute conditions were preferred but at the longer dose period, the more concentrated solution gave an improved dissolution profile. The DoE-optimized crystallization was successfully performed on three 40 kg batches and provided 207 with dissolution rates of 94— 95% (vs 77-93% prior to DoE). By controlling the crystallization, the levels of residual N-methylaniline were controlled and the yields improved from 68—72% to 76—77%. Form 1 of the NK] antagonist casopitant mesylate (208) was developed by GSK initially, but late in development it was discovered that the API was actually a mixture of Forms 1 and 3. Competitive slurry experiments showed crystallization of Form 1 was favored below 20 °C and that of Form 3 was favored above 30 °C (i.e., enantiotropic), but under no conditions did one form crystallize exclusively. Careful analysis of the preclinical toxicology batch revealed it contained 27% w/w Form 3; thus, this became the upper limit for future batches.94 As part of a QbD approach to defining the design space for the crystallization of the API, they applied DoE to develop a robust process, defined as a mixture of polymorphs with the Form 3 content capped at 27%. The first design screened eight continuous variables (35 reactions): acetone (3.5—5.5 vol) and EtOAc (3.0—5.0 vol), stirring power (4-320 W/m3), seed amount (0.25-0.75 wt %), seeding temperature (30—50 °C), age time before isooctane addition (0.5—2.5 h), isooctane added (2—4 vol), and isooctane addition time (0.5—1.5 h). Three different batch purities were included in the study (low, standard, and modified) by spiking the crystallizations with known impurities. A second DoE (10 reactions) studied a narrower range of those factors found to be relevant from the first design in order to define the proven acceptable ranges (PARs). The combined data sets indicated that seeding temperature was the most impactful factor, followed by the seeding amount. Higher seeding temperatures and seeding with smaller amounts of Form 1 led to higher levels of undesired Form 3. The remaining parameters had little or no impact on the formation of Form 3. Even under the most unfavorable conditions within the design space, the undesired form was present at no more than 14 w/w%, well below the threshold value. The robustness of the crystallization led to the conclusion that the Form 3 level in the API was not a critical quality attribute (CQA). Abbott researchers used DoE to optimize conditions for the resolution of209 (Scheme 18).9S Three factors were considered: the tartaric acid equivalents, solvent volume, and isolation temperature. Four successive recrystallizations raised the ee of salt 210 from 50% to 97% for eventual conversion to the API. As part of a QbD approach to control genotoxic impurities in a fluoroarylamine (structure not shown), a team from GSK employed DoE to optimize the rejection of mesylate esters resulting from isolating the product as the MSA salt.96 Under the optimal crystallization conditions that had been chosen, the 1627 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Scheme 19. Miscellaneous Reactions 1)MeOH (7 vol) HC(OMe)3 (4.5 eq.) H2S04 (1.78 eq.) 35 °C/7 h 2) Et3N (2.7 eq.) 55 °C 216, R = OH (40 kg) 3) H20, MIBK 217, R = H seed N-NH 1) KOf-Bu, DME 2) CICH2SCH3 H3C NC CH3 (~42 kg) 221 3) aq. NH3 4) + /-PrOAc, aq. washes 5) A, - /-PrOAc 1) H2S04 (1.0 eq.), < 25 °C, 20 min addition time 2) 20 °C/3 h; filter 3) reslurry in CH3CN -SCH3 h,CS- H3C^ NC N-NH y .0 H2S04 •CH3 H3C N-N y o "CH, CH3 221H2S04 (< 1%) nc^f ■ h2so4 CH3 222 H2SO„ (41% isolated) H3C. NC N-N y 0 H2S04 ~CH3 CH3 223 H2S04 (<1%) nra KOtBu (1.05 eq.) -»- |Q DMSO (1.1 eq.) 20-30°C °,s ,CH2 >X 224 =-y HO' 227(48%, q|_) >98% frans-) EW „ CH3 °*s4 1) KOH (1.5 eq.) H20 20 °C/2 h 2) HCl, /-PrOAc 3) A, - /-PrOAc CH3 225, 54.2 kg in DMSO, 80 °C 224 added to 225 over 1 h, 75 - 80 °C / 228 (-4%) EtO' CH3 226 SO,CI 1) Na2S03 (2.0 eq.) Na2HP04 (1-0 eq.) H20 (1 L) 60 °C/16h 2) 2,6-CI2C7H5Br (1-0 eq.) acetone (2 L) 60 °C/2 h 3) H20 Ph. O ,PN-OEt OEt 232 ■ PhCHO KOH Ph^^ Ph + H20+ p _0Et *u OEt moisture content was low (<0.4 wt %) and an excess of methanesulfonic acid relative to the free base was charged; those conditions favor the formation of mesylate esters. At the time of these investigations, the carcinogenicity potency of ethyl mesylate was not available. The impacts of the following parameters were studied: seeding temperature (34—44 °C), volume of EtOAc (3.0-5.0), volume of acetone (4.2-4.8), volume of isooctane (2.5—3.5), and isooctane addition time (0.5—1.5 h). A fractional factorial (2s-2) design was used in a 10 reaction study, including two center points. The measured responses were the residual levels of the methyl, ethyl, and isopropyl mesylate esters in the API, likely arising from the corresponding alcohols being present from upstream processing. In all cases, < 1 ppm of the mesylate esters was found in both the product suspensions and the isolated salts. The robustness of the operations throughout isolation and drying was also demon- strated, permitting the argument that testing batches of the mesylate salt for these mesylate esters would be unnecessary. A key step in the synthesis of GSK's casopitant mesylate, the CIDR on racemic amine 83 to produce (R)-amine salt 211, was optimized with DoE (Scheme 18). ' Epimerization of the methine carbon probably proceeds through an acid-catalyzed retro-Michael/Michael equilibrium with amine 212. Water was added to minimize the formation of three impurities arising from self-condensation of two molecules of 212, which dramatically lowered the isolated yield. The investigators studied the impact of the reaction time, temperature, water equiv (0.5—1.5), solvent volumes (IPA; 11 — 14), and L-mandelic acid stoichiometry (0.9— 1.3) on the yield of 211 and the formation of the unwanted diastereomeric amine salt. The formation of the unwanted diastereomer was kept <1.5% with lower amounts of mandelic acid, water, and the reaction volume. The yield ranged from 47— dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 1628 Organic Process Research & Development Review 63% over the design space and was favored by longer reaction time, lower temperature, and less mandelic acid. To preclude racemization during slow nitrations on scale, residual i-PrOH in the wet cake was displaced with cyclohexane. A subsequent DoE to confirm the robustness over the proposed proven acceptable range (PAR) was run and confirmed with scoping experiments at forcing, mild, and midpoint settings for the relevant parameters. DSM chemists used DoE to optimize the ZnCl2-catalyzed condensation of enolate 213 with amidinium chlorides to produce 4-aminopyrimidines (Scheme 18), used in the industrial synthesis of vitamin B].100 Having settled on an IPA/toluene solvent mixture, the team observed several runs to suffer from reaction mixtures that stuck to the vessel walls and were difficult to stir, and predicting which mixtures would be stirrable was not straightforward. Accordingly, they screened the following three factors to maximize the yield of pyrimidine 215 while ensuring stirrability: (l) % iPrOH in toluene (65-90%), (2) acetamidine (214) charge (0.95-1.40 equiv), and (3) ZnCl2 charge (0.15-0.30 equiv). The yields ranged from 68 to 80% over the design space. Ultimately, a DoE model, using STAVEX 5.1, was obtained (although not disclosed) that resulted in a higher R value of 0.96. The optimized process was demonstrated on an cubic meter scale. Pfizer researchers used DoE to simplify an amidation— animation sequence in the synthesis of CB] antagonist, CP-945,598-01. The large number of process variables for the extractive workup of the amide intermediate led the researchers to develop a direct isolation instead. They also used DoE studies to select the proper solvent composition (30% water/70% THF) and seed temperature (45 °C) for the crystallization of the penultimate aminated species.101 Miscellaneous Reactions. GSK researchers employed DoE to optimize the acid-catalyzed rearrangement of pleuromutilin, 216 (Scheme 19).102 Initially, conditions had to be sufficiently robust to tolerate variable amounts of acetate impurity 217 found in batches of the natural product 216. A half-factorial study was carried out to maximize the consumption of 216 (target <3% unreacted), by varying temperature (20, 40 °C), charges of MeOH (2, 6 vol), H2S04 (0.1, 0.5 vol), and trimethylortho-formate (TMOF, 0.6, 2 vol). Analysis of the data showed that the behavior was considerably nonlinear and that the optimum might lie outside of the ranges initially chosen. A second DoE was conducted using a central composite design, focusing on maximizing conversion to 218 and minimizing the formation of the major impurity, alkene 220 (target <4%). Four factors were examined: temperature (10—40 °C) and the charges of MeOH (0-40 mL/g), H2S04 (0.25-1.0 mL/g), and TMOF (0.4—1.6 mL/g); 30 experiments were carried out, including six center point runs. An interaction between acid equiv and temperature was most influential such that, at 10 °C, the amount of alkene was low (<0.5%) and constant over the range of acid equiv; while, at 30 °C, the alkene level was universally higher (4 to 6.4%) but dependent on the acid amount. The optimal conditions were initially performed at 25 °C (40 kg pilot scale) and gave yields up to 79% with >97% purity. In order to reduce time cycles from 24 to 6 h, acceptable results were achieved at 35 °C with no loss in purity. Prompted by an unexpected 10% yield loss during the synthesis of 222-H2S04 on scale (Scheme 19), a Pfizer group used DoE to minimize decomposition to 221 via loss of the methylthiomethyl group.103 In the initial process work up, the postquench mixture of regioisomers 222 and 223 in i-PrOAc was treated with H2SQ4 at 50 °C, aged for 12 h at 20-50 °C, filtered, and reslurried in acetonitrile. The variables examined included the H2S04 charge (0.8, 1.2 equiv), salt formation temperature (30, 70 °C), H20 charge (0.05, 0.4%), and H2S04 addition time (5, 25 min). Ten experiments were run: 24_1 plus two runs at the center point setting. Temperature was found to be the most critical factor, with increased degradation to 221-H2S04 seen with increasing temperature or with extended hold time before filtration. Based on the DoE analysis, the salt was isolated after stirring at 20 °C for 3 h, which served to reduce the level of 221* H2S04 by up to 50%. In the plant under the optimized conditions, two batches (42 kg input) gave an improved yield (41% vs 37%) and acceptable quality of 222-H2S04 for the current scale-up needs. Lilly researchers used DoE to optimize the cyclopropanation of olefin 225 with dimethylsulfoxonium methylide (Corey's ylide, 224; Scheme 19).104 Five factors were examined: temperature (60, 80 °C), cation (Na+ or K+), anion (t-BuO", HO~), addition type (base or preformed 224), and addition time (40, 80 min). A total of 16 experiments were run (2S_1, no center points). Higher yields of 226 were favored by using t-BuOK to preform ylide 224, with the 40 min addition at 80 °C. Eventually, the researchers found that the highest yield was obtained at 95 °C with a 15 min addition time, but these conditions were avoided due to safety concerns for scale-up. To avoid isolating the volatile 226, hydrolysis to acid 227 was telescoped with the cyclopropanation. The major impurity, 228, was removed by the extractive workup. Researchers from Pharmacia and SUGEN employed DoE studies in the early stage optimization of a reductive sulfonylation (Scheme 19).105 Reductions of the sulfonyl chloride 229 to the sulfinate intermediate (not shown) were carried out with zinc dust or sodium sulfite. Initial scale-up of Zn(0)-mediated reductions gave lower yields of 230 than anticipated, probably due to the agglomeration of the zinc dust, and as much as 5 A% of the sulfide side product 231 was formed. By resorting to reduction with Na2S03, the researchers anticipated that the reaction mixture would be homogeneous and that overreduction would be unlikely. The alkylation of the sulfinate intermediate was found to be robust, so attention was turned to DoE optimization of the reduction. Three factors were considered initially: Na2S03 charge (2, 4 equiv), temperature (60, 100 °C), and time (6, 24 h) (full factorial plus one duplicated run to assess reproducibility for a total of 9 runs). Markedly lower yields were found when a greater charge of Na2S03 was used, and the higher-temperature reaction was extended. The conditions that gave a 93% yield of 230 on screening with a 1 mmol charge of 229 produced 230 in only 70% when scaled up to 10 mmol; poor mass transfer of the biphasic (liquid—liquid) reaction mass was suspected. A second DoE screen was set up to assess whether poor mass transfer was causing low yields, and eight factors were considered: concentration (0.25, 0.5 M), charge of sulfite (2.0, 2.5 equiv), temperature (60, 80 °C), time for reduction (6, 24 h), cation (Na+, K+), presence or absence of a phase-transfer catalyst, reaction time (l, 2 h), and solvent (DMF, acetone). For the second screen, 17 experiments were run (28-4 plus a center point): a low-resolution design due to limited resources. The authors concluded that none of the main effects that could influence mass transfer (concentration, reaction time, use of PTC, e.g.) were significant, implicating an isolated occurrence of poor mixing due to magnetic stirring in the 10 mmol run with the low yield (vide supra). Scale-up of the biphasic reaction mass was carried out using vigorous mechanical agitation. Overall, using 1629 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Na2S03/ the yield of 230 was improved from 40-45% to 88%, and impurity 231 was not observed. DoE was used to analyze a Horner—Emmons reaction between benzaldehyde and diethyl benzylphosphonate, 232 (Scheme 19).106 The variables screened were base addition period (10, 20 min), initial temperature (5, 40 °C), and final temperature (40, 80 °C). High yields were obtained when both temperatures were low and the addition time was long, or when the temperatures were high and the addition time was short. Additional experiments showed that the reaction was second-order in benzaldehyde and that the competing reaction was hydrolysis of232. For best control of the olefination on scale and to minimize hydrolysis of 232, the authors recommended carrying out the reaction at a lower temperature. Analytical Investigations. Researchers in France used DoE to calculate the specific heat capacity (CP) of solvents through analyses in a RC1 reaction calorimeter.107 The accuracy of these values is important to calculate the adiabatic temperature increase of reaction mixtures. Using a central composite design, 20 experiments were carried out on three solvents, which included six center point runs to assess the reproducibility. The factors examined were the experimentally determined heat capacity, the volume of solvent charged to the calorimeter, and the agitation rate. Analysis showed that the greatest error between the experimentally determined CP and the true value of CP occurred with low volumes in the reactor and high agitation rate. A model was created that analysts can apply to reaction calorimeters in their lab. Researchers in the Netherlands analyzed intermediates in a steroidal contraceptive that had been manufactured for 20 years.108 The API was made in more than 10 steps from a plant-derived starting material, and intermediates had been assayed by TLC. In order to comply with QbD directives and to preclude failing batches of intermediates by relatively subjective TLC analyses, the researchers developed an ultrahigh-performance liquid chromatography (UHPLC) method for the last five steps. A DoE study was set up to assess the robustness of the analytical method, by examining performance under conditions that bracketed the optimized conditions. The factors examined were column temperature (30, 50 °C), mobile phase flow rate (0.7, 0.9 mL/min), and composition of the mobile phase (92%, 98% acetonitrile in Eluent B). A full factorial study with three center point runs was set up, for a total of 11 runs; responses were peak resolution and assay times. The assay method was found to be robust and suitable for the last five steps. Not only were TLC assays replaced by a less subjective UHPLC assay, but assay times for the intermediates could be reduced to as little as 25% of the TLC assay time. ■ CONCLUSION DoE is gaining wider acceptance as another valuable tool for process optimization in the pharmaceutical industry. When experimentation and analyses are reproducible, the time required for generating insightful data maybe cut in half or decreased even further. A review of publications in Organic Process Research & Development that used DoE shows that it has been implemented at various stages of process development. As Aggarwal and Owen noted, "The models generated by both the screening and the optimization studies are approximations of reality. They are used to predict the favored settings and then these settings can be implemented in order to validate the model."70 The process chemist and engineer must apply their experience to ensure that both the parameters chosen for the DoE and the results are practicable, reproducible, and relevant to the overall project objectives. We can expect that DoE will be used increasingly for efficient process development. ■ ASSOCIATED CONTENT O Supporting Information Bibliographic information for the publications on DoE in OPRD that appeared from 2003—2013, used for Figure 1. This material is available free of charge via the Internet at http://pubs.acs.org. ■ AUTHOR INFORMATION Corresponding Authors *E-mail: sweissman(2>jstar-research.com. Phone: 908-791-9811 (x520). *E-mail: nganderson(2>wildblue.net. Phone: 541-899-5555. Author Contributions The manuscript was written through contributions of both authors. Both authors have given approval to the final version of the manuscript. Funding No funds were used to support the research of this manuscript. Notes The authors declare no competing financial interest. Biographies Steven A. Weissman, a native of Marblehead, MA, received his H.A. degree in Chemistry from the University of Vermont in 1983 and his Ph.D. from Tufts University in the field of organophosphorus chemistry. He then accepted a postdoctoral appointment with Professor Hebert C. Brown (Purdue University) before joining the Merck Manufacturing division in 1990. In 1997, he moved to Merck Process Research, where he rose to the rank of Senior Investigator and was responsible for developing practical syntheses of pre-clinical candidates, as well as contributed to the commercial routes for Crixivan, Isentress, and Emend. In 2010, he joined Concert Pharmaceuticals as their head of Process Chemistry, and in 2013, he joined J-Star Research as a Director, where he advises clients on process characterization strategies through DoE investigations to facilitate regulatory filings and chemical development. Dr. Weissman has authored over 30 peer-reviewed publications and been cited as an inventor on 11 U.S. patents. 1630 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review Neal G. Anderson received his B.Sc. degree in Honors Biology from the University of Illinois in 1972 and his Ph.D. in medicinal chemistry from the University of Michigan in 1977. After an industrial post-doc, he joined E. R Squibb & Sons in 1979. Working in chemical process R&D, he developed and implemented processes for scale-up of many drug candidates and intermediates, along with processes to manufacture aztreonam, captopril, fosinopril sodium, and nadolol. After 17 years, he left Bristol-Myers Squibb as a Principal Scientist. Since 1997 he has consulted internationally, lectured, and presented short courses on chemical process R&D of "small molecules". He is the author or coauthor of 16 peer-reviewed publications and one book chapter, and he is an inventor on six U.S. patents. He is the author of Practical Process Research & Development—A Guide for Organic Chemists (Academic Press; 2nd edition, 2012). ■ ABBREVIATIONS ANOVA, analysis of variance; CCD, central composite design; CCF, central composite face-centered design; CIDR, crystallization-induced dynamic resolution; DoE, design of experiments; Fm, 9-fluorenylmefhyl; HTS, high-throughput screening; KRED, ketoreductase; QbD, quality by design; OFAT, one factor at a time; OVAT, one variable at a time; RSM, response surface methodology ■ REFERENCES (1) In this review, DoE includes "design of experiments", "design of experiment", "response surface methodology", and "factorial experimental design". (2) Laird, T. Org. Process Res. Dev. 2002, 6, 337. (3) Mills, J. E. Design of Experiments in Pharmaceutical Process Research and Development. In Chemical Process Research; Abdel-Magid, A F., Ragan, J. A, Eds.; ACS Symposium Series 870; American Chemical Society: Washington, DC, 2003; Chapter 6, pp 87-109. (4) Mullin, R Design of Experiments Makes A Comeback. Chem. Eng. News 2013, April 1, pp 25-28. (5) In 2002, it was stated that delaying the introduction of a product by six months could reduce the lifetime profit by 50%: Ritter, S. K. Chem. Eng. News 2002, 80 (47), 19. (6) For example, the FDA now provides the breakthrough NDA filing designation to sponsors for fast track or accelerated approvals and priority review. For more information, see http://www.fda.gov/Drugs/ DevelopmentApprovalProcess. (7) Anderson, N. C; Burdick, D. C; Reeve, M. M. Org. Process Res. Dev. 2011, IS, 162. (8) (a) ICH Harmonized Tripartite Guideline Ql 1, Development and Manufactrue of Drug Substances (Chemical Entities and Biotechno-logical/Biological Entities), May 2012: http://www.ich.org/fileadmin/ Public_Web_Site/ICH_Products/Guidelines/Quality/Qll/Qll_ Step_4.pdf (accessed 4/24/2014). (b) For instance, Vertex researchers used DoE to advance a QbD filing for a hydrogenation: Looker, A R.; Ryan, M. P.; Neubert-Langille, B. J.; Naji, R Org. Process Res. Dev. 2010, 14, 1032. (9) Leahy, D. K.; Tucker, J. L.; Mergelsberg, I.; Dunn, P. J.; Kopach, M. E.; Purohit, V. C. Org. Process Res. Dev. 2013, 17, 1099. (10) Andraos, J. Org. Process Res. Dev. 2005, 9, 149. (11) Lendrem, D.; Owen, M.; Godbert, S. Org. Process Res. Dev. 2001, 5, 324. (12) (a) Pilipauskas, D. R. Using Factorial Experiments in the Development of Process Chemistry. In Process Chemistry in the Pharmaceutical Industry; Gadamasetti, K. G., Ed.; Dekker: New York, 1999. (b) Statistics for Experimenters; Box, G. E. P.; Hunter, W. G.; Hunter, J. S.; Wiley: New York, 1978. (c) Carlson, R; Carlson, J. E. Design and Optimization in Organic Synthesis; Elsevier: Amsterdam, 2005. (13) For instance, Pfizer researchers determined through DoE the optimal catalyst and solvents to minimize hydrogenolysis byproducts from the reductive amination used to manufacture sertraline: Taber, G. P.; Pfisterer, D. M.; Colberg, J. C. Org. Process Res. Dev. 2004, 8, 385. (14) Kuethe, J. T.; Tellers, D. M.; Weissman, S. A.; Yasuda, N. Org. Process Res. Dev. 2009, J 3, 471. (15) Bacchi, S.; Yahyah, M.; Carangio, A.; Ribecai, A.; Oliveros, M. C. C. Org. Process Res. Dev. 2010, 14, 332. This reference provides an example of a more complex design using D-Optimal design. (16) Continuous factors are those variables to which a numerical value can be assigned (temperature, stoichiometry, e.g.) in contrast to discreet factors (solvent, ligand, reagent, base, e.g.). (17) The center point is the mean value of the high and low settings for a given variable/factor. (18) Optimization designs are frequently synonymous with response surface models (RSM) and central composite designs. (19) Only a handful of non-OPRD papers on the topic were published in this time frame. Although they could have been included, we chose to limit our discussion to those from this journal. (20) An overview of the application of DoE to optimization of an HPLC method can be found here: Reid, G. L.; Morgado, J.; Barnett, K.; Harrington, B.; Wang, J.; Harwood, J.; Fortin, D. Analytical Quality by Design (AQbD) in Pharmaceutical Development. Am. Pharm. Rev. published online: Aug. 27, 2013. http://www. americanpharmaceuticalreview.com/Featured-Articles/144191-Analytical-Quality-by-Design-AQbD-in-Pharmaceutical-Development/ (accessed June 25, 2014). (21) The optimized conditions are detailed in the schemes, rather than in the body of the text. (22) Hobson, L. A; Akiti, O.; Deshmukh, S. S.; Harper, S.; Katipally, K.; Lai, C. J.; Livingston, R. C; Lo, E.; Miller, M. M.; Ramakrishnan, S.; Shen, L.; Spink, J.; Tummala, S.; Wei, C; Yamamoto, K.; Young, J.; Parsons, R. L., Jr. Org. Process Res. Dev. 2010, 14, 441. (23) DelMonte, A J.; Fan, Y; Girard, K. P.; Jones, G. S.; Waltermire, R E.; Rosso, V; Wang, X. Org. Process Res. Dev. 2011, IS, 64. (24) Mobele, B. I.; Venkatraman, S.; McNaughton-Smith, G.; Gibb, C; Ulysse, L. G.; Lindmark, C. A; Shaw, S.; Marron, B.; Spear, K.; Suto, M. J. Org. Process Res. Dev. 2012,16, 1385. (25) Banks, A.; Breen, G. F.; Caine, D.; Carey, J. S.; Drake, C; Forth, M. A; Gladwin, A; Guelfi, S.; Hayes, J. F.; Maragni, P.; Morgan, D. O; Oxley, P.; Perboni, A.; Popkin, M. E.; Rawlinson, F.; Roux, G. Org. Process Res. Dev. 2009, 13, 1130. (26) Hajmowicz, H; Wisialski, J.; Synoradzki, L. Org. Process Res. Dev. 2011, IS, 427. (27) Weiberth, F. J.; Gill, H. S.; Jiang, Y; Lee, G. E.; Lienard, P.; Pemberton, C; Powers, M. R; Subotkowski, W.; Tomasik, W.; Vanasse, B. J.; Yu, Y. Org. Process Res. Dev. 2010, 14, 623. (28) Manoel, E. A.; Pais, K. C; Cunha, A. G; Simas, A B. C; Coelho, M. A Z.; Freire, D. M. G. Org. Process Res. Dev. 2012,16, 1378. (29) Guercio, G.; Bacchi, S.; Perboni, A.; Leroi, C; Tinazzi, F.; Bientinesi, I.; Hourdin, M.; Goodyear, M.; Curti, S.; Provera, S.; Cimarosti, Z. Org. Process Res. Dev. 2009,13, 1100. (30) Cimarosti, Z.; Bravo, F.; Castoldi, D.; Tinazzi, F.; Provera, S.; Perboni, A.; Papini, D.; Westerduin, P. Org. Process Res. Dev. 2010, 14, 805. 1631 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review (31) Guercio, G.; Perboni, A.; Tinazzi, F.; Rovatti, L.; Provera, S. Org. Process Res. Dev. 2010,14, 840. (32) Roemmele, R. C; Christie, M. A. Org. Process Res. Dev. 2013,17, 422. (33) Argentine, M. D.; Braden, T. M.; Czarnik, J.; Conder, E. W.; Dunlap, S. E.; Fennell, J. W.; LaPack, M. A.; Rothhaar, R. R.; Scherer, R. B. ; Schmid, C. R; Vicenzi, J. T.; Wei, J. G.; Werner, J. A.; Roginski, R T. Org. Process Res. Dev. 2009,13, 131. (34) Weiberth, F. J.; Yu, Y.; Subotkowski, W.; Pemberton, C. Org. Process Res. Dev. 2012,16, 1967. (35) Goodman, S. N.; Dai, Wang, J.; Clark, W. M., Jr. Org. Process Res. Dev. 2011, IS, 123. (36) Barton, B.; Logie, C. G.; Schoonees, B. M.; Zeelie, B. Org. Process Res. Dev. 2005, 9, 70. (37) Hida, T.; Fukui, Y.; Kawata, K.; Kabaki, M.; Masui, T.; Fumoto, M.; Nogusa, H. Org. Process Res. Dev. 2010,14, 289. (38) Hida, T.; Nogusa, H. Tetrahedron 2009, 65, 270. (39) Quittmann, W.; Roberge, D. M.; Bessard, Y. Org. Process Res. Dev. 2004, 8, 1036. (40) Rossi, F.; Corcella, F.; Caldarelli, F. S.; Heidempergher, F.; Marchionni, C; Auguadro, M.; Cattaneo, M.; Ceriani, L.; Visentin, G; Ventrella, G.; Pinciroli, V.; Ramella, G.; Candiani, I.; Bedeschi, A.; Tomasi, A.; Kline, B. J.; Martinez, C. A; Yazbeck, D.; Kučera, D.J. Org. Process Res. Dev. 2008,12, 322. (41) In glass-lined vessels, the reaction of Deoxo-Fluor with the glass, as evidenced by the observance of pitting, is surmised to produce a fluorosilicate, which is known to be beneficial to this type of reactions (see Radhamani, K. N.; Padma, D. K /. Fluorine Chem. 1993, 95 ). (42) Hanselmann, R.; Johnson, G; Reeve, M. M.; Huang, S.-T. Org. Process Res. Dev. 2009,13, 54. (43) Moseley, J. D.; Moss, W. O.; Welham, M. J.; Ancell, C. L.; Banister, J.; Bowden, S. A; Norton, G.; Young, M. J. Org. Process Res. Dev. 2003, 7, 58. (44) Bescond, P.; Graindorge, H; Mace, H. EP91337A1, 1999. Bescond, P.; Graindorge, H; Mace, H. US 5,973,149, 1999. (45) Bayat, Y.; Hajimirsadeghi, S. S.; Pourmortazavi, S. M. Org. Process Res. Dev. 2011, IS, 810. The hexabenzyl precursor to TADB was prepared from glyoxal and benzylamine. (46) Bayat, Y.; Ebrahimi, H; Fotouhi-Far, F. Org. Process Res. Dev. 2012,16, 1733. (47) Guercio, G.; Bacchi, S.; Goodyear, M.; Carangio, A.; Tinazzi, F.; Curti, S. Org. Process Res. Dev. 2008,12, 1188. (48) (a) Anderson, N. G. Org. Process Res. Dev. 2005, 9, 800. (b) Brands, K M. J.; Davies, A J. Chem. Rev. 2006, 106, 2711. (49) Bravo, F.; Cimarosti, Z.; Tinazzi, F.; Castoldi, D.; Stonestreet, P.; Galgano, A; Westerduin, P. Org. Process Res. Dev. 2010,14, 832. (50) Alorati, A. D.; Gibb, A. D.; Mullens, P. R; Stewart, G. W. Org. Process Res. Dev. 2012,16, 1947. (51) Chaplin, D.; Harrison, P.; Henschke, J. P.; Lennon, I. C; Meek, G.; Moran, P.; Pilkington, C. J.; Ramsden, J. A; Watkins, S.; Zanotti-Gerosa, A. Org. Process Res. Dev. 2003, 7, 89. (52) Appell, R. B.; Boulton, L. T.; Daugs, E. D.; Hansen, M.; Hanson, C. H; Heinrich, J.; Kronig, C; Lloyd, R. C; Louks, D.; Nitz, M.; Praquin, C; Ramsden, J. A.; Samuel, H; Smit, M.; Willets, M. Org. Process Res. Dev. 2013,17, 69. (53) Dirat, O.; Bibb, A J.; Burns, C. M.; Checksfield, G. D.; Dillon, B. R; Field, S. E.; Fussell, S. J.; Green, S. P.; Mason, C; Mathew, J.; Mathew, S.; Moses, I. B.; Nikiforov, P. I.; Pettman, A. J.; Susanne, F. Org. Process Res. Dev. 2011, IS, 1010. (54) Another reductive amination was discussed with Scheme 3. (55) Guercio, G.; Manzo, A M.; Goodyear, M.; Bacchi, S.; Curti, S.; Provera, S. Org. Process Res. Dev. 2009, 13, 489. In the optimized process, («Bu)4NBr was added to solubilize the mandelate salt in THF, avoiding the salt split and some racemization that had been produced by the presence of residual NaOH that was extracted into the organic phase. (56) Vo, L.; Ciula, J.; Gooding, O. W. Org. Process Res. Dev. 2003, 7, 514. (57) Alimardanov, A R; Barrila, M. T.; Busch, F. R; Carey, J. J.; Couturier, M. A; Cui, C. Org. Process Res. Dev. 2004, 8, 834. (58) Other intermediates have been proposed and monitored by React-IR in the reduction of a succinimide to the corresponding pyrrolidine: see ref 15. (59) Nadkarni, D. V.; Hallissey, J. F. Org. Process Res. Dev. 2008, 12, 1142. (60) Gooding, O. W.; Voladri, R.; Bautista, A.; Hopkins, T.; Huisman, G.; Jenne, S.; Ma, S.; Mundorff, E. C; Savile, M. M.; Truesdell, S. J.; Wong, J. W. Org. Process Res. Dev. 2010, 14, 119. (61) Zhang, P.; Dong, Z. E.; Cleary, T. P. Org. Process Res. Dev. 2005, 9, 583. (62) Massari, L.; Panelli, L.; Hughes, M.; Stazi, F.; Maton, W.; Westerduin, P.; Scaravelli, F.; Bacchi, S. Org. Process Res. Dev. 2010,14, 1364. (63) Humphrey, G. R; Pye, P. J.; Zhong, Y.-L.; Angelaud, R; Askin, D.; Belyk, K. M.; Maligres, P. E.; Mancheno, D. E.; Miller, R. A.; Reamer, R A.; Weissman, S. A. Org. Process Res. Dev. 2011, IS, 73. (64) Hopes, P. A.; Parker, A. J.; Patel, I. Org. Process Res. Dev. 2006,10, 808. (65) Brenek, S. J.; Caron, S.; Chisowe, E.; Delude, M. P.; Drexler, M. T.; Ewing, M. D.; Handfield, R.E.; Ide, N. D.; Nadkarni, D.V.; Nelson,J. D.; Olivier, M.; Perfect, H. H; Phillips, J. E.; Teixeira, J. J.; Weekly, R M.; Zelina, J. P. Org. Process Res. Dev. 2012, 16, 1348. (66) Barrios Sosa, A C; Williamson, R T.; Conway, R.; Shankar, A.; Sumpter, R; Cleary, T. Org. Process Res. Dev. 2011, IS, 449. (67) Gopalsamy, A.; Shi, M.; Nilakantan, R Org. Process Res. Dev. 2007, 11, 450. (68) Corona, J. A; Davis, R D.; Kedia, S. B.; Mitchell, M. B. Org. Process Res. Dev. 2010, 14, 712. (69) Bream, R N; Hayes, D.; Hulcoop, D. G.; Whiteman, A. J. Org. Process Res. Dev. 2013, 17, 641. (70) Aggarwal, V. K.; Staubitz, A. C; Owen, M. Org. Process Res. Dev. 2006, 10, 64. (71) These conditions of high temperature, low catalyst loadings, and relatively dilute conditions are analogous to the optimal conditions for other Pd-catalyzed couplings without ligands: (a) de Vries, A H. M.; Mulders, J. M. C. A; Mommers, J. H. M.; Henderickx, H.J. W.; de Vries, J. G. Org. Lett. 2003,5, 3285. (b) Arvela, R. K.; Leadbeater, N. E.; Sangi, M. S.; Williams, V. A.; Granados, P.; Singer, R. D. /. Org. Chem. 2005, 70, 161. (72) Cai, C; Chung, J. Y. L.; McWilliams, J. C; Sun, Y; Shultz, C. S.; Palucki, M. Org. Process Res. Dev. 2007, 11, 328. (73) Nishimura, K.; Kinugawa, M. Org. Process Res. Dev. 2012,16,225. (74) Van Arnum, S. D.; Moffet, H; Carpenter, B. K Org. Process Res. Dev. 2004, 8, 769. (75) Vedantham, R; Vetukuri, V. N. K. V. P. R; Boini, A; Khagga, M.; Bandichhor, R Org. Process Res. Dev. 2013,17, 798. (76) For a review of escitalopram syntheses, see: Harrington, P. J. Pharmaceutical Process Chemistry for Synthesis; Rethinking the Routes to Scale-Up; Wiley-VCH; 2011; pp 30-91. (77) Alam, M.; Wise, C; Baxter, C. A.; Cleator, E.; Walkinshaw, A. Org. Process Res. Dev. 2012, 16, 435. (78) Lawrence, R M.; Dennis, K C; O'Neill, P. M.; Hahn, D. U.; Roeder, M.; Struppe, C. Org. Process Res. Dev. 2008, 12, 294. (79) Raw, S. A; Taylor, B. A; Tomasi, S. Org. Process Res. Dev. 2011, IS, 688. (80) Bullock, K M.; Burton, D.; Corona, J.; Diederich, A.; Glover, B.; Harvey, K; Mitchell, M. B.; Trone, M. D.; Yule, R; Zhang, Y; ToczkoJ. F. Org. Process Res. Dev. 2009, 13, 303. (81) Synoradzki, L.; Jaficzewski, D.; Wlostowski, M. Org. Process Res. Dev. 2005, 9, 18. (82) Kirschning, A. Beilstein J. Org. Chem. 2013, 9, 1696. (83) Anderson, N. G. Org. Process Res. Dev. 2012, 16, 852. (84) Nieuwland, P. J.; Segers, R; Koch, K.; van Hest, J. C. M.; Rutjes, F. P. J. T. Org. Process Res. Dev. 2011, IS, 783. (85) Koch, K; van Weerdenburg, B. J. A; Verkade, J. M. M.; Nieuwland, P. J.; Rutjes, F. P.J. T.; van Hest, J. C. M. Org. Process Res. Dev. 2009,13, 1003. (86) Anderson, B. G.; Bauta, W. E.; Cantrell, W. R, Jr. Org. Process Res. Dev. 2012,16, 967. 1632 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633 Organic Process Research & Development Review (87) Tinder, R; Farr, R.; Heid, R.; Zhao, R; Rarig, R. S., Jr.; Storz, T. Org. Process Res. Dev. 2009,13, 1401. (88) Vörös, A.; Baän, Z.; Mizsey, P.; Finta, Z. Org. Process Res. Dev. 2012,16, 1717. (89) Chen, J.; Przyuski, K; Roemmele, R; Bakale, R. P. Org. Process Res. Dev. 2013, DOI: 10.1021/op400179f. (90) http://thalesnano.com/products/h-cube (accessed 11 April 2014). (91) Junior, I. I.; Flores, M. C; Sutili, F. K.; Leite, S. G. F.; de M. Miranda, L. S.; Leal, I. C. R; de Souza, R O. M. A. Org. Process Res. Dev. 2012.16, 1098. (92) Sleveland, D.; Bjorsvik, H.-R Org. Process Res. Dev. 2012,16,1121. (93) Sjövall, S.; Hansen, L.; Granquist, B. Org. Process Res. Dev. 2004,8, 802. (94) Cimarosti, Z.; Castagnoli, C; Rossetti, M.; Scarati, M.; Day, C; Johnson, B.; Westerduin, P. Org. Process Res. Dev. 2010,14, 1337. (95) Pu, Y.-M.; Grieme, T.; Gupta, A; Plata, D.; Bhatia, A. V.; Cowart, M.; Ku, Y.-Y. Org. Process Res. Dev. 2005, 9, 45. (96) Cimarosti, Z.; Bravo, F.; Stonestreet, P.; Tinazzi, F.; Vecchi, O.; Camurri, G. Org. Process Res. Dev. 2010, 14, 993. (97) (a) Teasdale, A.; Eyley, S. C; Delaney, E.; Jacq, K.; Taylor-Worth, K; Lipczynski, A.; Reif, V.; Elder, D. P.; Facchine, K L.; Golec, S.; Oestrich, R. S.; Sandra, P.; David, F. Org. Process Res. Dev. 2009,13,429. (b) Teasdale, A; Delaney, E. J.; Eyley, S. C; Jacq, K; Taylor-Worth, K.; Lipczynski, A.; Hoffmann, W.; Reif, V.; Elder, D. P.; Facchine, K. L.; Golec, S.; Oestrich, R S.; Sandra, P.; David, F. Org. Process Res. Dev. 2010.14, 999. (98) Bravo, F.; Cimarosti, Z.; Tinazzi, F.; Smith, G. E.; Castoldi, D.; Provera, S.; Westerduin, P. Org. Process Res. Dev. 2010,14, 1162. (99) Another CIDR, using 3,5-dichlorosalicylaldehyde, is discussed with Scheme 7. (100) Letinois, U.; Schütz, J.; Härter, R; Stoll, R; Huffschmidt, F.; Bonrath, W.; Karge, R. Org. Process Res. Dev. 2013,17, 427. (101) Ragan, J. A; Bourassa, D. E.; Blunt, J.; Breen, D.; Busch, F. R; Cordi, E. M.; Damon, D. B.; Do, N.; Engtrakul, A.; Lynch, D.; McDermott, R E.; Mongillo, J. A; OSullivan, M. M.; Rose, P. R; Vanderplas, B. C. Org. Process Res. Dev. 2009, 13, 186. (102) Andemichael, Y; ChenJ.; ClawsonJ. S.; Dai, W.; Diederich, A.; Downing, S. V.; Freyer, A. J.; Liu, P.; Oh, L. M.; Patience, D. B.; Sharpe, S.; Sisko, J.; Tsui, J.; Vogt, F. G.; Wang, J.; Wernersbach, L.; Webb, E. C; Wertman, J.; Zhou, L. Org. Process Res. Dev. 2009, 13, 729. (103) de Koning, P. D.; McManus, D. J.; Bandurek, G. R Org. Process Res. Dev. 2011, 15, 1081. (104) Delhaye, L.; Stevens, C; Merschaert, A.; Delbeke, P.; Briöne, W.; Tilstam, U.; Borghese, A; Geldhof, G.; Diker, K.; Dubois, A; Barberis, M.; Casarubios, L. Org. Process Res. Dev. 2007,11, 1104. (105) Chen, J. J.; Nugent, T. C; Lu, C. V.; Kondapally, S.; Giannousis, P.; Wang, Y; Wilmot, J. T. Org. Process Res. Dev. 2003, 7, 313. (106) Grabarnick, M.; Zamir, S. Org. Process Res. Dev. 2003, 7, 237. (107) Alozie, D.-M.; Courtes, P.; Ha, B.; Prat, D. Org. Process Res. Dev. 2011.15, 1412. (108) Musters, J.; van den Bos, L.; Kellenbach, E. Org. Process Res. Dev. 2013.17, 87. 1633 dx.doi.org/10.1021/op500169m I Org. Process Res. Dev. 2015,19,1605-1633