Design of Experiments: A Rational Approach Toward
Non-Covalent Asymmetric Organocatalysis
P. Renzi*1
M. Bella*
Dipartimento di Chimica, Sapienza Università di Roma, Piazzale Aldo Moro 5, 00185 Roma, Italy
marco.bella@uniroma1.it
polyssena.renzi@chemie.uni-regensburg.de
Received: 06.09.2016
Accepted after revision: 27.10.2016 Published online: 08.12.2016
DOI: 10.1055/s-0036-1588654; Art ID: st-2016-a0591-a
Abstract This account describes, from a personal point of view, the possible strategies to tackle and optimize non-covalent organocata-lyzed reactions. When chemical intermediates are covalently bound, predictive mechanistic scenarios can be depicted. In contrast, there are several organocatalyzed transformations (e.g., those employing cincho-na alkaloids) for which optimization is essentially based on a trial-and-error approach. The experience of the authors is that these reactions can be tackled with a rational approach employing Design of Experi-ments (DoE). This tool is widely exploited in industrial process chemis-try, but is little known within academia. The purpose of this account is to show the effectiveness and utility of DoE in asymmetric non-cova-lently organocatalyzed reactions, discussing selected examples. 1 Introduction: Covalently and Non-Covalently Asymmetric
Or-ganocatalyzed Reactions
2 The Challenge: Optimizing (in a Rational Way) Non-Covalently Organocatalyzed Reactions without Deep Knowledge of the Mechanism. Case Study 1
3 The Solution: DoE Might be the Best Possible Tool to Approach these Issues
4 The Application of DoE to Kinetic Resolution. Case Study 2 5 Conclusions and Outlook
Key words organocatalysis, Design of Experiments, non-covalent or-ganocatalysis, asymmetric synthesis, synthetic methodologies
1 Introduction: Covalently and
Non-Cova-lently Asymmetric Organocatalyzed
Reac-tions
As students in chemistry we have no problem in trust-ing the analytical data showtrust-ing the outcome of a given re-action. Here are the NMR and mass spectra, the HPLC traces, so anyone can verify the identity and the stereochemical purity of a compound. What about the proposed transition
states and intermediates? Sometimes these structures might be useful, but in other cases they can be just a vague picture of the real world. Let’s take as an example the mechanisms in amine-mediated organocatalysis.2 There are two major modes of activation. One is the formation of a covalently bound intermediate, such as the iminium ion and the enamine, the other is based on non-covalent cataly-sis. Iminium ions and enamines are useful thanks to their predictive conformations. Even if the exact nature of the Jørgensen–Hayashi TMS-diphenylprolinol-derived enamine can be debated, the stereochemical outcome of the reaction can be confidently predicted assuming that the Re face of the intermediate is shielded by the bulky TMS diphenyl group, therefore the attack of a nucleophile occurs on the opposite Si face (Figure 1).3
Figure 1 Covalent asymmetric organocatalysis: the proposed interme-diates are valuable in predicting the stereochemical outcome
Even if the elusive enamine intermediate has been iden-tified at least once, in the case of a deprotonated proline-derived intermediate,4 can we state that these are the ‘real’ structures? They are as ‘real’ or as ‘unreal’ as the drawing of a molecular orbital, but in most of the cases they are ‘useful’ because they describe the actual chemical world, and they allow us to foresee at least the stereochemical outcome of reactions. N R O O H E Attack from the
(front) Re face
N
R
OTMS E Attack from the
(back) Si face Ar Ar H R H O E H R H O E
What about the case in which there are no covalent bonds involved, such as in the example of non-covalent or-ganocatalysis? In this instance, the reaction intermediates are vaguely described because they are not connected by covalent bonds, even if it is possible to put forward some reasonable hypotheses, such as the formation of an ion pair between the catalyst and the substrate (Figure 2). According to this scenario, reactions run in solvents which favor a ‘tight’ ion pair (e.g., toluene) show generally a higher enan-tioselectivity with respect to polar solvents (e.g., metha-nol). Also, the higher the dilution, the better is the observed reaction enantioselectivity since the hypothesized ion pairs should have less interactions with the others. Bifunctional organocatalysts usually perform quite well in terms of reac-tion rate accelerareac-tion and stereocontrol, but the under-standing of a plausible transition state, which involves the substrates and the high molecular weight catalyst, is very complex.
Ten years earlier, when Marco Bella was a postdoc in K. A. Jørgensen’s group and started his involvement in the organocatalysis world, these issues were unresolved. Non-covalent organocatalyzed reactions give useful products which could be fully characterized with respect to the yield and stereochemical purity. In contrast, the evidence regard-ing how these products were formed (reaction mechanism) was minimal, a similar analogy being our knowledge in as-tronomy of what happens once the event horizon in a black
hole has been passed, being mostly the realm of science
fic-tion movies.5 Group meetings were frustrating because
there was no sound answer to the questions raised regard-ing the mechanism and rationalization of the stereochemi-cal outcome.
Figure 2 Example of a proposed intermediate for a non-covalently cat-alyzed asymmetric reaction
Reactants are kept together by non-covalent bonds.
Multiple conformational equilibra render depiction of the transition state difficult.
Change of solvent might invert the sense of stereoinduction. Dilution might affect the reaction enantioselection.
Weak interactions such as hydrogen bonding might play a major role. N N H H H S N H R1 O R2 N O O N H Electrophile activation by hydrogen bonding Activation of a nucleophile by deprotonation R3 H Biographical Sketches
Polyssena Renzi was born in Rome, Italy, in 1987. She stud-ied chemistry at ‘Sapienza’ Uni-versity of Rome, where she received her degree in 2011. In 2014, she obtained her Ph.D. at the same institute under the
su-pervision of Professor M. Bella working on the development of new strategies for asymmetric reactions. She then joined the group of Professor R. M. Gschwind at Regensburg Uni-versity (GE) as a postdoctoral
researcher, where she is cur-rently working on the mecha-nistic investigation of organocatalyzed reactions by means of NMR.
Marco Bella was born in 1972 and received his Ph.D. from Sa-pienza University of Rome in 2000. From then until 2005 he was a postdoctoral researcher in the groups of K. C. Nicolaou (La Jolla, USA) and K. A. Jørgensen (Aarhus, Denmark). In 2005, he returned to Sapienza where he is currently Associate Professor
and teaches organic stereo-chemistry. His research inter-ests concern asymmetric reactions, especially those which can be scaled up in an in-dustrial environment, and novel strategies in asymmetric cataly-sis. He is married to Julia, who is a key team player in helping to manage extra-academic
proj-ects such as Luca (7), a special son, and Daniel (5). In the spare time he has left, he is active in the dissemination of science through articles directed to the general public and his own blog on the electronic version of ‘Il Fatto Quotidiano’, one of the major Italian newspapers.
2 The Challenge: Optimizing (in a Rational
Way) Non-Covalently Organocatalyzed
Reac-tions without Deep Knowledge of the
Mecha-nism. Case Study 1
To illustrate the above point, let us take one of our own recent papers as an example.6 The aza-Michael addition of imides to enones is a challenging transformation leading, in a simple way, to the introduction of an amine functionality (Scheme 1), but the weak nucleophilicity of the imides and the reversibility of the conjugated addition (leading to Mi-chael adducts which are not configurationally stable) has, over the years, hampered its development.7 This transfor-mation cannot be realized with simple cyclic enones such as 1a, which are not activated enough. On the other hand, it was successful when employing alkylidene β-ketoesters such as 1b, which have been recognized as a more versatile class of substrates because of their increased reactivity due to the strong polarization of the enone double bond.8 Con-cerning imides 2, their use as nucleophiles in asymmetric synthesis was limited, and only a few reports were known in the literature at the time.9
In this context, we observed no reaction between suc-cinimide and 2-cyclohexen-1-one (1a), but replacing 1a with the activated electrophile, 2-ethoxycarbonyl 2-cyclo-hexen-1-one (1b), in the presence of quinine 5 as a catalyst led to the formation of the desired Michael adduct 3, albeit in low conversion and low enantioselectivity (10% yield, <10% ee after one day of reaction) (Scheme 1 and Figure 3). However, the use of chiral thioureas as catalysts afforded the desired product in a reasonable yield and enantioselec-tivity (see Figure 3 for the structures of catalysts 6–8). Con-sidering the results obtained applying the thiourea-based
compound 6 developed by the Sóos group in 200510 (more
than 95% conversion and 70% enantiomeric excess, Figure 3), we decided to synthesize two different types of multi-functional catalyst derived from cinchona alkaloids linked with amino acids and presenting a different substitution pattern on the thiourea group (compounds 7 and 8, Figure 3,). A screening of these compounds allowed us to select quinine- and cinchonidine-derived catalysts 8a and 8b, which afforded, respectively, high levels of enantioselectivi-ty (83% and 85% ee for compound 4), and excellent conver-sions (>95% conversion for compound 3). However, the yield of the isolated product 4 after reduction was not optimal. Thus, we tested the effect of different additives on the en-antioselectivity of the reaction, finding that the use of inor-ganic Lewis acids, such as palladium acetate [Pd(OAc)2], nickel chloride [NiCl2] or the chiral organic acid camphor-sulfonic acid [(+)-CSA], were beneficial in terms of the en-antioselectivity,11,12 but at the expense of the yield. The best results were obtained by employing camphorsulfonic acid
(0.1 equiv) in the presence of 20 mol% of the catalyst, but the increase in the enantioselectivity was also associated with a reduction of the conversion (from 95% to 74% with catalyst 8b). This behavior can be ascribed to a progressive quenching of the catalyst by the acid. These were mostly empirical observations. In this specific reaction, there were also some key issues regarding reproducibility of yield and the enantiomeric excess of the product in the experimental setting, mainly due to the chemical and stereochemical in-stability of product 3. Longer reaction times would result in higher yield, but also in a higher racemization rate. In fact, the control of parameters such as reaction time, catalyst loading and concentration was crucial for the outcome of this reaction. To tackle these specific issues was well be-yond the possibility of any computational calculation. A tri-al-and-error strategy would appear to be the only feasible one in order to optimize this specific reaction. When the preliminary results of this work were presented at a confer-ence in Siena (EWSDy),13 the questions raised by Professor Stephen Hanessian about how to optimize the reaction were difficult to answer, and the only viable solution would seem to be the use of ‘brute force’, e.g., several students testing all the possible combinations of additive, catalyst and conditions. Is this scientifically sound? Is this some-thing which will be useful to train early-stage researchers (which is one of the key missions of an academic institu-tion) and most of all, is this enjoyable? The main motivation why people put effort into doing science is that they have fun. If science becomes a repetitive exercise, it will no lon-ger remain exciting.
To approach this problem in a sound way and to collect information which can be predictive for the outcome of the reaction, one must know the following parameters: (i) The preferred conformation of the organocatalyst in solution. (ii) The mode in which the electrophile substrate, the unsat-urated β-ketoester 1b, is bound to the organocatalyst. (iii) The mode in which the nucleophile, the imide 2, is bound to the organocatalyst 6–8. (iv) The relative position of these three molecules. (v) The manner in which the solvent is in-fluencing this array and specifically the effect of concentra-tion. (vi) The role of the additive.
Furthermore, one should be able to give an estimate of the energy differences between the two diastereomorphic transition states. Because the uncertainty of the calculation might easily be higher than the energy difference, still to-day it remains clear that no computational approach might give any predictive or reliable information. In fact, the ener-gy difference between the two diastereomorphic states giv-ing rise to a product with 90% ee is less than 2 kcal/mol. A reliable calculation taking into account all the parameters discussed above and which allows discrimination between 80% ee and 90% ee is still unrealistic today.
Scheme 1 Organocatalytic asymmetric addition of imides 2: general scheme
3 The
Solution:
DoE
Might
be
the
Best
Possi-ble Tool to Approach these Issues
The last lecture of the day at the EWDSy in Siena was a presentation by a company representative of a software for DoE (Design of Experiments).14 The lecture started with the discussion of the mathematics involved in DoE, and since chemists are mostly ‘allergic’ to mathematics, most attend-ees left after the first hour. In the second hour, a practical demonstration of the DoE software on computers was pro-vided. This proved enlightening. The application of DoE, specifically in organic synthesis, has been highlighted in a recent review.15
In many fields, it is quite common that there is the wish to optimize the outcome (response) of a procedure without deep knowledge of the factors influencing it (variables). If a reaction experimental setting cannot be fully explained, but has indeed an effect on its outcome, why should that reaction be left aside? This is exactly the problem which is faced in non-covalent organocatalysis: there are surely some parameters which are important for the reaction out-come, but since responses are not linear and influenced by each other, proceeding with trial-and-error might not be the most effective approach. The response can be visualized as a hypersurface in n + 1 dimensions with respect to the number n of variables. Design of Experiments is then a ra-tional strategy to explore the experimental space which has been defined by them.
The following example might help to clarify how to ap-proach such a problem. Let us take a generic reaction, the response (yield) of which depends only on two variables (temperature and pH). This might be the formation of an imine from a carbonyl compound or any other reaction. First of all, one should define the limits of the experimental domain (e.g., 40–80 °C for the temperature, and a pH of 2– 10). The quality of the response will be based on the appro-priate choice of experimental domain, therefore values which have only mathematical but not chemical signifi-cance (e.g., concentration zero or infinite), or that are of un-likely chemical interest (extreme values of pH or tempera-ture), should be avoided. Since the response depends on two parameters, the ‘classic’ approach to optimize this re-action will be to fix one of the two variables (e.g., tempera-ture) and then screen the other (pH). Then, once the maxi-mum of the response (yield) has been found as a function of pH, change the temperature (Figure 4a). However, this strategy (change One Variable At a Time or OVAT) is not the most convenient approach to explore the experimental space, since in most of the experiments the variables are not independent. A more effective strategy is to change si-multaneously both parameters, so that the reactivity space is explored in a more rational way. Specifically, the same number of experiments (9) can be placed at the center of the domain, at the four corners and at the middle points of the perimeter (Figure 4b). The exploration of a bidimen-sional space can be easily designed and visualized. Coming to a multidimensional space, the number of relevant points
NH X O O R 1a R = H 1b R = CO2Et + OH R N O X Catalyst toluene, –20 °C 2 3 OH R N O X 4 NaBH4 (4 equiv), MeOH
Not chemically and configurationally stable
Isolated as single diasteroisomer
–78 °C to –20 °C, 1 h
Figure 3 Catalysts tested in the asymmetric aza-Michael addition and a summary of the results obtained in terms of yield and enantioselec-tivity N HO MeO N H N MeO N H H N H N S F3C F3C 5 6 N MeO N H HN 7 O H N H N S F3C F3C H R1 N R2 N H HN N H S t-BuO2C H R1
R1 = H, Me, i-Pr, CH2i-Pr, CH2Ph, t-Bu
20–95% conversion <5–77% ee 1–5 days reaction time
8
R1 = H, Me, i-Pr, CH2i-Pr, CH2Ph, t-Bu, CH 2Ot-Bu
R2 = H, OMe
34 to >95% conversion 18–82% ee 1–3 days reaction time
t-BuO
R1 =
R2 = H, 8a
R2 = OMe, 8b
increases. If we imagine a tridimensional experimental space, the number of significant points as in the above ex-ample will be 27 (Figure 5a). Therefore, it is convenient to select just some of them. Two kinds of setups are possible: initially, it is advisable to use a screening model in which a limited number of points gives quick information regarding the responses. Examples of screening models are shown in Figure 5b (full factorial), in which the corners and the cen-tral point are preserved, and in Figure 5c (fractional factori-al), a model in which two corners out of four on each face are cut off. Each model includes the central point of the de-sign in which the experiment can be repeated usually three times. This allows an estimate of the experimental error.
It is difficult to imagine an n-dimensional object, there-fore, at this stage, the experimental setting can be better designed by the aid of a computer program. The detailed discussion about how computer programs operate is be-yond the scope of this article. It is helpful, but not neces-sary, to have a deep understanding of the mathematics be-hind DoE.
Figure 4 Methods for the exploration of a bidimensional experimental space. Each circle represents an experiment with the combination of the parameters. OVAT approach (a), and a more rational design (b) for a generic reaction
DoE as a tool, as with all tools, is suitable for specific is-sues. It is especially effective to optimize responses for
con-tinuous variables [temperature, reaction time, pH value,
concentration of the reagents, amount (%) of the catalyst] and to discover non-linear effects. It is less effective to
inves-tigate discrete variables (solvent, catalyst or additive). The
number of experiments in a design grows exponentially with the number of variables [e.g., to include all corners and middle points on the perimeter it is necessary to run nine experiments for a two-dimensional design (Figure 4b) and twenty-seven for a three-dimensional design (Figure 5a)]. Therefore, DoE is not an ideal tool for a wide screening of catalysts, but it can be effectively used for discriminating among a small number of similar-performing catalysts un-der different experimental conditions.
DoE can be seen as a tool to save time and resources, running only the most significant experiments, but its main advantage is that it allows a rational exploration of reactivi-ty space, in contrast with a trial-and-error approach. There-fore, the application of DoE is not limited to large industrial production, where process optimization leads to significant savings in terms of solvents, reagents, usage time of the re-actors, but it should also find wider application in academic research. A field for which it is really useful is non-covalent catalysis.
The application of DoE to organocatalytic asymmetric synthesis in academia is nearly virtually a new field in or-ganic chemistry; in fact, we were unaware of previous sys-tematic applications to new asymmetric reactions, such as the one that will be described in the present work.
Figure 5 Methods for the exploration of a tridimensional experimental space. Each circle represents an experiment with the combination of the parameters. All significant points (27 experiments, a). Some experi-mental design used to simplify the exploration of the reactivity space: full factorial (9 experiments, all corners, b), and fractional factorial de-sign (5 experiments, alternate corners, c); these setups are employed for screening purposes. More sophisticated designs can instead be em-ployed to model the response surface: central composite CCF (15 ex-periments, center of the single faces plus corners, d), and Box–Benhken design (9 points, middle points, e). Usually the central point of the de-sign (which is always present) is repeated three times in order to esti-mate the experimental error
In our case, since the OVAT approach was not successful to solve the problem of progressive racemization with time, we decided to test this new strategy. The impossibility of finding a straightforward solution could be, in fact, due to the presence of interactions between variables, which are not taken into account by the OVAT approach. In order to be able to explore the experimental space by means of a ratio-nal approach taking interactions into account, it is neces-sary to use a multivariate method which allows all perti-nent factors to be considered simultaneously. Therefore, to achieve the real optimization of our reaction, we decided to apply to our system, rather than a trial-and-error-approach,
a rational strategy such as the Design of Experiments (DoE), with the aim, also, to develop a standard protocol for scien-tists dealing with similar issues.
The DoE screening was realized with the aim to opti-mize both the yield and the enantiomeric excess of the aza-Michael addition of interest, studying the non-linear (possi-bly synergistic) effect of using a second Lewis acid as the additive to be employed simultaneously with (+)-camphor-sulfonic acid and the catalysts derived from quinine 8a or cinchonidine 8b.
The parameters we decided to analyze were: (i) The choice of the cinchona-alkaloid-derived catalyst (a discrete variable with two levels). (ii) The presence or absence of palladium(II) acetate as an additive in addition to (+)-CSA (a discrete variable). (iii) The catalyst loading (a multilevel
variable studied at three levels). (iv) The reaction concen-tration (a multilevel variable studied at three levels). (v) The reaction time (a multilevel variable with four levels).
The type of solvent (toluene) and the reaction tempera-ture (T = –20 °C) were fixed. Having five different variables under study, the experimental space representing this reac-tion is a pentadimensional reactivity space, exhaustive ex-ploration of which by applying a full factorial design would have required 144 experiments (= 2 × 2 × 3 × 3 × 4). Per-forming 144 experiments would have required a consider-able investment of working time and chemicals, which can be avoided by selecting significant experiments and by ap-plying a rational approach. We chose a D-optimal design se-lecting 19 runs, which was built by considering five factors [catalyst, addition of Pd(OAc)2, catalyst amount, solution
Table 1 Design of Experiments in the Screening of the Asymmetric Addition of 2a to 1b to Afford Alcohol 4a
Entrya Catalyst Pd(OAc)
2 (1 = yes; 0 = no) Catalyst (mol%) Toluene (mL) Time (d) Yieldb ee (%)c
1 8a 1 5 3 9 32 87 2 8a 0 20 3 9 46 95 3 8b 1 5 2 9 <5 – 4 8b 0 20 2 9 33 75 5 8a 1 20 2 9 20 85 6 8b 1 20 3 9 22 82 7 8b 0 5 3 9 11 84 8 8a 0 5 2 9 <5 – 9 8b 0 10 3 5 14 82 10 8a 0 20 2.5 5 22 88 11 8a 1 5 2 3 25 86 12 8b 0 5 2 2 17 86 13 8b 0 20 3 2 25 90 14 8b 1 5 3 2 15 84 15 8a 0 5 3 2 <5 – 16 8b 1 20 2 2 28 85 17 8a 1 20 3 2 14 85 18 8a 0 10 2.5 2 12 87 19 8a 0 20 2 2 27 90
a Reactions run in the presence of 1b (0.446 mmol), 2a (1.5 equiv), catalyst 8a or 8b (x mol%, where the value of x is indicated in column 4) and (+)-CSA (x/2 mol%); (CSA = camphorsulfonic acid).
b Yield of isolated product.
c Ee values determined by HPLC on a chiral stationary phase.
NH O O O CO2Et 1b + 1. 8a or 8b, (+)–CSA toluene, –20 °C 2a OH CO2Et N O O 4a N R N H HN N H S t-BuO2C H R = H, 8a R = OMe, 8b t-BuO
2. NaBH4 (4 equiv), MeOH,
–78 °C to –20 °C, 1 h
concentration and time], including repeated experiments16 and assuming a priori that specific binary interactions could be considered as not significant. Analyzing the results obtained, only three sets of reaction conditions gave a yield lower than 5% for the isolated cyclohexanol 4a, in all cases, the enantioselectivity was good with moderate yields (12– 46%, 75–95% ee). Satisfying results were obtained in the presence of 20 mol% of cinchonidine-based catalyst 8a and 10 mol% of (+)-CSA, in the absence of the second additive, at the highest dilution (3 mL of toluene) after nine days, al-lowing product 4a to be isolated in 46% yield with 95% ee. These conditions have not been found by serendipity, but through a systematic exploration of the reactivity space. The application of design of experiment allowed us to im-prove the yield from 25% to 46% and the ee from 91 to 95% (see Table 1).
With these optimized reaction conditions in hand, we tried to further extend the scope of the aza-Michael reac-tion of imides. We decided to test different imides to see if the conditions developed for the synthesis of 4a could be a good starting point. Considering the different properties of the imides tested, a separate optimization would have been required in principle for each nucleophile. Good results were obtained for each imide tested either for the yield or for the enantiomeric excess. The results obtained are sum-marized in Figure 6. Besides, the amide and ester function-ality of compound 4b can be easily cleaved to afford biolog-ically interesting cyclic hydroxy amino acid 9 bearing three adjacent stereogenic centers (Figure 6).17
In summary, a simplified application of DoE (we just run a screening setup) in our specific case allowed us to
find the best possible experimental conditions for our par-ticular reaction (46% yield, 95% ee for compound 4a). It should be noted that these results, even if not optimal as absolute values, were not found simply by a trial-and-error approach, but rather by applying a rational approach.
Scheme 2 General reaction scheme for the kinetic resolution of 4-sub-stituted oxazinones rac-10
4 The Application of DoE to Kinetic
Resolu-tion. Case Study 2
18Among the known synthetic transformations, kinetic resolution is probably one of the most affected by experi-mental conditions. Kinetics, in fact, can be influenced by a large number of variables such as temperature, reaction time, concentration of reagents, type of solvent, type of cat-alyst and its loading. The OVAT approach, therefore, can be considered a poor methodology for the optimization of this kind of transformation since variables are likely to interact with each other. So finding the optimum conditions would require that a large number of experiments should be per-formed, and if interactions are significant the real optimum may never be found.
With respect to the aza-Michael addition, we decided in this case to apply the DoE from the beginning of the optimi-zation process. The application of experimental design re-quires a step-by-step approach composed of four different phases: planning, design, reaction performance and data
analysis. The planning phase is one of the most important
phases to assess before starting a DoE experiment. At the outset of the study, in fact, it is crucial to formulate the problem in a clear way, defining the project targets and se-lecting the responses to optimize. As our project target, we decided to simultaneously maximize the enantiomeric ex-cess of the starting material, the conversion and the enan-tiomeric excess of the product, knowing that in a kinetic resolution the enantioselectivity, obtainable simultaneous-ly for both the substrate and the product, is strongsimultaneous-ly cor-related with the conversion. Once the responses to optimize have been planned, we run some pre-DoE experiments in order to choose the variables to investigate and determine their suitable ranges. First of all, we selected our model sub-Figure 6 Results obtained for the reaction scope. The reactions were
run employing 1b (0.595 mmol), imide 2b–f (1.5 equiv) in toluene (0.198 M) with catalyst 8a (20 mol%) and (+)-CSA (10 mol%) at –20 °C. Compound 4a was obtained from complete reduction of the maleimide moiety. Cleavage of the amide and ester functionalities of product 4b afforded cyclic hydroxy amino acid 9
OH CO2Et N O O OH CO2Et N O O OH CO2Et N O O H H OH CO2Et N N N O 4a OH CO2Et N O O OH CO2Et NH2 9 4c 4b 4d 4e 41% yield >20:1 d.r. 91% ee 3 days 30% yield 5:1 d.r. 82% ee 5 days 36% yield 7:1d.r. 80% ee 4 days 56% yield >20:1d.r. 75% ee 7 days 77% yield >20:1 d.r. 89% ee 2 days N O O R OH N O O R R N H O O O + + Thioureas 6, 8a,b Squaramides 13–16 solvent, r.t. rac-10 (S)-10 11 (R)-12
strate, racemic 2,4-diphenyl-4,5-dihydro-1,3-oxazin-6-one (rac-10a) (R = Ph), and we run on it a control reaction and a preliminary catalyst screening in which four chiral squara-mide (13–16) (Figure 7) and three thiourea-derived cata-lysts were tested (6, 8a,b) (Figure 3). The organocatalyst structure was chosen on the basis of the results previously described in the literature for the dynamic kinetic resolu-tion of azlactones and for the kinetic resoluresolu-tion of
oxazi-nones.19 The screening was performed by employing
rac-10a as the substrate, allyl alcohol 11 (1 equiv) as the
nucle-ophile and 12.5 mol% of the catalyst in anhydrous toluene as the solvent (Scheme 2). It is to be underlined that it is possible to tune the catalyst selectivity toward the resolu-tion of the R or S enantiomer of the starting material by changing the cinchona alkaloid on which the multifunc-tional catalyst is based. The pre-DoE experiments allowed us to identify four potential catalysts (thioureas 6, 8a,b) (Figure 3), and squaramide16 (Figure 7), and to obtain some insight about the catalyst loading, the equivalents of allyl alcohol 11 and the concentration range values to be used in the first DoE.
Figure 7 The squaramides tested in the kinetic resolution of oxazi-nones rac-10
So after the preliminary catalyst screening, we were able to plan the first DoE experiment. To understand which variables were having the major impact and influence on the kinetic resolution we employed a screening experi-ment. The variables selected were: (i) The type of catalyst, a categorical (discrete) variable which was studied at four different levels: compounds 6, 8a,b and 16 selected by the preliminary catalyst screening. (ii) The catalyst loading, a continuous variable which was explored in the range 5–20 mol%. (iii) The type of solvent, a categorical variable for which five different levels were chosen: toluene,
dichloro-methane, acetonitrile, tert-butyl methyl ether and ethyl ac-etate. (iv) The solution concentration, a continuous variable studied in the range 0.05–0.5 M. (v) The equivalents of allyl alcohol 11, a continuous variable in the range of 0.5–1 equivalents. (vi) The temperature, a continuous variable in the range 0–20 °C.
Once we had decided the variables and the responses to study, the planning phase was concluded and the designing
phase started. We had to choose the most appropriate
ex-perimental design for the case in hand. A full factorial de-sign, exploring all possible combinations of the six factors, would have required (4 × 2 × 5 × 2 × 2 × 2) = 320 experi-ments with a significant investment of time and resources. Since we were interested in studying the presence of possi-ble interactions between the type of catalyst and the sol-vent, we used the custom design function in the JMP® soft-ware to generate a design specifically tailored to our
prob-lem.20 The custom design we generated required 24
experiments, the order of execution of which was random-ized (reaction performance). The results obtained were then
analyzed by the software to generate models and
predic-tions for the responses of interest. When approaching the DoE data analysis, each result obtained has not to be re-garded as the outcome of a single experiment, but they have to be studied in their complex since they are points lo-cated on the same multidimensional surface. Good models were found for the enantiomeric excess of the starting ma-terial and for the conversion at all-time points. The most important factors affecting the conversion and the enantio-meric excess of the starting material were the solution con-centration, followed by the type of solvent and the tem-perature. In all cases, toluene and dichloromethane ap-peared to be the best solvents in order to optimize these two responses. The type of catalyst did not appear to be im-portant, neither for the enantiomeric excess of compound
10a nor for the conversion, but it was the most important
factor affecting the enantiomeric excess of the product 12a followed by the interactions between the solution concen-tration and the type of catalyst or its loading. The highest enantiomeric excess for the chiral allyl ester 12a could be obtained by employing the dimeric squaramide 16 or the thiourea-derived catalyst 8a. At the end of this custom de-sign, we were able to select two catalysts (8a and 16), two solvents (toluene and dichloromethane) and to set the amount of allyl alcohol 11 at one equivalent. In order to fur-ther understand which solvent and which catalyst worked best, these parameters were selected to perform a new DoE screening. The step-by-step approach of DoE, by running succeeding screening designs, allows the number of vari-ables and their levels to be narrowed, detecting after each design, in a rational way, the most important factors influ-encing the responses under study. In our case, to obtain op-timized reaction conditions, a second design was per-formed by applying a definitive screening design (DSD).21
N H N H O O N MeO N N H N OMe H N H H N O O NH Ph N OMe H 13 N H H N O O NH N OMe H 15 CF3 F3C O O N H N H OH CF3 F3C 14 16 H H H
The variables we investigated were: (i) The type of cata-lyst, a categorical (discrete) variable which was studied at two levels: compounds 8a or 16 selected by the custom de-sign. (ii) The catalyst loading, a continuous variable which was explored at three levels: 5, 12.5 and 20 mol%. (iii) The type of solvent, a categorical (discrete) variable studied at two levels: toluene and dichloromethane, selected by the custom design. (iv) The solution concentration, a continu-ous variable with three levels: 0.5, 0.25 and 0.05 M. (v) The temperature, a continuous variable with three levels: 0, 10 and 20 °C.
The JMP® software selected 14 experiments, which
were run in a random order. Analyzing the table of results for the definitive screening design, there are different sets of conditions which allow the starting material 10a to be obtained with enantiomeric excess values of more than 90%, and the allyl ester 12a with enantiomeric excess val-ues greater than 80%. With squaramide-derived catalyst 16 at the highest catalyst loading in dichloromethane at 0 °C, the reaction was very fast; chiral compound (S)-10a was obtained with 92% ee at 52% conversion after only 30 min-utes. By employing toluene as the solvent, it was possible to lower the catalyst loading running the reaction for a longer reaction time. After five hours at room temperature, chiral oxazinone (S)-10a and allyl ester (R)-12a were obtained, re-spectively, with 99% and 82% ee at 55% conversion in the presence of 12.5 mol% of thiourea 8a. If in place of the thiourea 8a, 5 mol% of squaramide 16 was employed, at 51% conversion the starting material 10a was obtained with 94%
ee and the chiral allyl ester 12a with 89% ee. A statistical
treatment of the error was also included to evaluate the ex-perimental variation and reproducibility of the reaction by performing repeated experiments. As this point, our inves-tigation was concluded with optimized reaction conditions (5 mol% of catalyst 16, a solution concentration of 0.5 M, anhydrous toluene as the solvent, room temperature, 1 equivalent of allyl alcohol 2), which allowed chiral oxazi-none 10a to be obtained with 99% ee and the allyl ester (R)-12a with 87% ee after five hours of reaction. The selec-tivity factor (S) measured under these conditions for cata-lyst 16 was 98. It was calculated by applying the Kagan equation assuming a first-order reaction and neglecting possible non-linear effects.22 The value of S was improved from 35 to 98 by application of a rational approach employ-ing DoE, which by usemploy-ing a suitable design enabled optimum conditions to be identified within the reactivity space stud-ied.
We finally extended the scope of the alcoholytic ring-opening of 4-substituted oxazinones rac-10 by applying the optimum reaction conditions found for the model substrate
rac-10a to a series of substituted oxazinones rac-10b–g,
which possess either aromatic or aliphatic groups at the 4-position. To obtain optimized conditions for all the sub-strates tested, a small screening was necessary, the results of which are summarized in Figure 8.
Figure 8 Scope of the kinetic resolution of substituted oxazinones rac-10b–g (0.1 mmol) with allyl alcohol (1 equiv) and catalyst 8a or 16 (5 mol%) in toluene (0.5 M solution with catalyst 16 and 0.4 M with cata-lyst 8a) at room temperature
5 Conclusions and Outlook
The use of DoE as a tool to optimize the enantioselectiv-ity of reactions has also been exploited by other research groups. While DoE is an extremely useful tool largely em-ployed in process chemistry, its application in the academic
N O O Ph N O O Ph N O O Ph N O O Ph N O O Ph MeO Cl Br Ph N H O O O Ph N H O O O Ph N H O O O Ph N H O O O Ph N H O O O OMe Cl Br Catalyst 16, 52% Conversion, 97% ee 10b, 90% ee 12b
8 h reaction time, 76 Selectivity Factor (S)-10b (R)-12b (S)-10c (R)-12c (S)-10d (S)-10e (R)-12d (R)-12e (S)-10f (R)-12f N O O Ph Ph N H O O O (S)-10g (R)-12g Catalyst 16, 55% Conversion, 99% ee 10c, 80% ee 12c
5 h reaction time, 46 Selectivity Factor
Catalyst 8a, 56% Conversion, 98% ee 10d, 71% ee 12d 3 h reaction time, 34 Selectivity Factor
Catalyst 8a, 54% Conversion, 99% ee 10e, 83% ee 12e 27 h reaction time, 61 Selectivity Factor
Catalyst 8a, 51% Conversion, 92% ee 10f, 90% ee 12f 48 h reaction time, 53 Selectivity Factor
Catalyst 16, 51% Conversion, 88% ee 10g, 85% ee 12g 24 h reaction time, 35 Selectivity Factor
world is still limited today. This might be due to its appar-ently ‘scary’ mathematical background. Then again, to use Windows 10 it is not necessary to have a deep knowledge of a programmers’ language such as C++, so in a similar way it is not required (although it might be useful) to understand in great detail how the commercial programs operate in or-der to use them. DoE is useful in all situations where there are continuous variables and a rational approach toward reaction optimization is not viable. It should also not be for-gotten that DoE does not only save time and resources, but most importantly, it allows the reactivity space to be ex-plored in a rational way, in contrast with a trial-and-error approach. For some types of transformation, specifically the asymmetric organocatalyzed non-covalent reactions, it is probably the ideal tool for optimization. We do hope that other readers will find DoE as useful as it was for us.
Acknowledgment
The research of Marco Bella has been supported by Sapienza Universi-ty of Roma through ‘Progetto di Ateneo’, 2013–2015.
References
(1) Current address: Institut für Organische Chemie, Universität Regensburg, Universitätsstraße 31, 93040 Regensburg, Ger-many.
(2) (a) List, B. Synlett 2001, 1675. (b) Comprehensive Asymmetric
Catalysis; Jacobsen, E. N.; Pfaltz, A.; Yamamoto, H., Eds.;
Springer-Verlag: Berlin, 1999. (c) Catalytic Asymmetric
Synthe-sis, 3rd ed.; Ojima, I., Ed.; John Wiley & Sons: Hoboken, 2010.
(d) Bertelsen, S.; Jørgensen, K. A. Chem. Soc. Rev. 2009, 38, 2178. (e) Melchiorre, P.; Marigo, M.; Carlone, A.; Bartoli, G. Angew.
Chem. Int. Ed. 2008, 47, 6138.
(3) Franzén, J.; Marigo, M.; Fielenbach, D.; Wabnitz, T. C.; Kjærsgaard, A.; Jørgensen, K. A. J. Am. Chem. Soc. 2005, 127, 18296.
(4) (a) Schmid, M. B.; Zeitler, K.; Gschwind, R. M. Angew. Chem.
Int. Ed. 2010, 49, 4997. (b) Schmid, M. B.; Zeitler, K.; Gschwind,
R. M. Chem. Eur. J. 2012, 18, 3362.
(5) In a black hole, the event horizon indicates a boundary beyond which events cannot affect an outside observer. Not even light can go out because it would require a hypothetical speed greater than that of light. No observational data can currently be obtained from the inside of a black hole. In analogy, no evi-dence can be gained at the moment to support the mechanism of some specific asymmetric reactions.
(6) Silvi, M.; Renzi, P.; Rosato, D.; Margarita, C.; Vecchioni, A.; Bordacchini, I.; Morra, D.; Nicolosi, A.; Cari, R.; Sciubba, F.; Scarpino Schietroma, D. M.; Bella, M. Chem. Eur. J. 2013, 19, 9973.
(7) Breugst, M.; Tokuyasu, T.; Mayr, H. J. Org. Chem. 2010, 75, 5250. (8) For a review on cyclic β-ketoesters, see: Schotes, C.; Mezzetti, A.
ACS Catal. 2012, 2, 528.
(9) (a) Jiang, H.; Nielsen, J. B.; Nielsen, M.; Jørgensen, K. A. Chem.
Eur. J. 2007, 13, 9068. (b) Vesely, J.; Ibrahem, I.; Rios, R.; Zhao,
G.-L.; Xu, Y.; Córdova, A. Tetrahedron Lett. 2007, 48, 2193.
(c) Ma, S.; Wu, L.; Liu, M.; Huang, Y.; Wang, Y. Tetrahedron 2013,
69, 2613.
(10) For pioneering work on cinchona-alkaloid-derived thiourea cat-alysts, see: (a) Vakulya, B.; Varga, S.; Csámpai, A.; Sóos, T. Org.
Lett. 2005, 7, 1967. (b) McCooey, S. H.; Connon, S. J. Angew. Chem. Int. Ed. 2005, 44, 6367. (c) Li, B.-J.; Jiang, L.; Liu, M.; Chen,
Y.-C.; Ding, L.-S.; Wu, Y. Synlett 2005, 603. (d) Ye, J.; Dixon, D. J.; Hynes, P. S. Chem. Commun. 2005, 4481.
(11) For a combination of cinchona-alkaloid-derived thioureas and acids, see for example: Chen, X.; Zhu, W.; Qian, W.; Feng, E.; Zhou, Y.; Wang, J.; Jiang, H.; Yao, Z.-J.; Liu, H. Adv. Synth. Catal. 2012, 354, 2151.
(12) Sigman, M. S. Angew. Chem. Int. Ed. 2007, 46, 4748.
(13) The Fourth European Workshop in Drug Synthesis (IV EWDSy), University of Siena Certosa di Pontignano, Siena, Italy, May 27th–31st, 2012.
(14) For books and reviews on DoE, see: (a) Carlson, R.; Carlson, J. E.
Design and Optimization in Organic Synthesis, 2nd ed.; Elsevier:
Amsterdam, 2005. (b) Antony, J. Design of Experiments for
Engi-neers and Scientists; Elsevier: Amsterdam, 2003. (c) Box, G. E. P.;
Draper, N. R. Empirical Model-Building and Response Surfaces; John Wiley & Sons: New York, 1986. (d) Murray, P. M.; Tyler, S. N. G.; Moseley, J. D. Org. Process Res. Dev. 2013, 17, 40. (e) Leardi, R. Anal. Chim. Acta 2009, 652, 161. (f) Tye, H. Drug Discovery
Today 2004, 9, 485.
(15) Murray, P. M.; Bellany, F.; Benhamou, L.; Bucar, D.-K.; Tabor, A. B.; Sheppard, T. Org. Biomol. Chem. 2016, 14, 2373.
(16) In order to estimate the reproducibility, the best results found (Table 1, entry 2) were replicated. Compound 4a was originally obtained in an isolated yield of 46%, with 95% ee. Replicated run 1, 4a: 51% yield, 95% ee. Replicated run 2, 4a: 56% yield, 94% ee. (17) (a) Ideta, R.; Nakazawa, Y.; Iwaki, H.; Ishino, A.; Tajima, M.
EP1806121(A1), 2007. (b) Larsen, S. D.; Connell, M. A.; Cudahy, M. M.; Evans, B. R.; May, P. D.; Meglasson, M. D.; Sullivan, T. J. O.; Schostarez, H. J.; Sih, J. C.; Stevens, F. C.; Tanis, S. P.; Tegley, C. M.; Tucker, J. A.; Vaillancourt, V. A.; Vidmar, T. J.; Watt, W.; Yu, J. H. J. Med. Chem. 2001, 44, 1217. (c) Mittendorf, J.; Kunisch, F.; Matzke, M.; Militzer, H.-C.; Schmidt, A.; Schönfeld, W. Bioorg.
Med. Chem. Lett. 2003, 13, 433.
(18) Renzi, P.; Kronig, C.; Carlone, A.; Eröksüz, S.; Berkessel, A.; Bella, M. Chem. Eur. J. 2014, 20, 11768.
(19) (a) Berkessel, A.; Cleemann, F.; Mukherjee, S.; Müller, T. N.; Lex, J. Angew. Chem. Int. Ed. 2005, 44, 807. (b) Berkessel, A.; Mukherjee, S.; Cleemann, F.; Müller, T. N.; Lex, J. Chem.
Commun. 2005, 1898. (c) Berkessel, A.; Mukherjee, S.; Müller, T.
N.; Cleemann, F.; Roland, K.; Brandenburg, M.; Neudörfl, J.-M.; Lex, J. Org. Biomol. Chem. 2006, 4, 4319. (d) Lee, J. W.; Ryu, T. H.; Oh, J. S.; Bae, H. J.; Janga, H. B.; Song, C. E. Chem. Commun. 2009, 7224. (e) Berkessel, A.; Cleemann, F.; Mukherjee, S. Angew.
Chem. Int. Ed. 2005, 44, 7466.
(20) JMP 11® is a statistical software package available from SAS.
(21) DSD is a three-level designs for screening quantitative factors in the presence of active first- and second-order effects. It is an efficient screening design which allows evaluation of continu-ous parameters at three levels (the extreme values of the range under study plus a central point), also allowing the evaluation of curvature and generating non-linear models, whilst keeping the number of experiments low compared with optimization designs.
(22) Kagan, H. B.; Fiaud, J. C. Kinetic Resolution, In Topics in
Stereo-chemistry; Vol. 18; Eliel, E. L.; Wilen, S. H., Eds.; John Wiley &
Sons: New York, 1988, 249.
