A pragmatic procedure for predicting regioselectivity in nucleophilic substitution of aromatic fluorides

Tetrahedron Letters 52 (2011) 3150–3153

Contents lists available at ScienceDirect

Tetrahedron Letters journal homepage: www.elsevier.com/locate/tetlet

A pragmatic procedure for predicting regioselectivity in nucleophilic substitution of aromatic fluorides Magnus Liljenberg a, Tore Brinck b,⇑, Björn Herschend c, Tobias Rein c, Glen Rockwell c, Mats Svensson d,⇑ a

AstraZeneca, Sweden Operations, S-151 85 Södertälje, Sweden Physical Chemistry, School of Chemical Science and Engineering, KTH—Royal Institute of Technology, S-100 44 Stockholm, Sweden c AstraZeneca R&D, Pharmaceutical Development, S-151 85 Södertälje, Sweden d AstraZeneca R&D, Medicinal Chemistry, S-151 85 Södertälje, Sweden b

a r t i c l e

i n f o

Article history: Received 1 March 2011 Revised 29 March 2011 Accepted 8 April 2011 Available online 15 April 2011 Keywords: Nucleophilic substitution Regioselectivity Computational r-Complex DFT

a b s t r a c t The scope and limitations of a method for predicting the regioisomer distribution in kinetically controlled nucleophilic substitution reactions of aromatic fluorides have been investigated. This method is based on calculating the relative stabilities of the isomeric r-complex intermediates using DFT. A wide set of substrates and anionic nucleophiles have been investigated. Predictions from this method can be used quantitatively—these agree to an average accuracy of ±0.5 kcal/mol with experimental observations in eleven of the twelve investigated reactions. Ó 2011 Elsevier Ltd. All rights reserved.

When novel synthetic routes to a target molecule are designed, one key part of the work-flow is the prioritization of which route alternatives (‘paper routes’) to first evaluate experimentally. Predictive computational methods can aid this prioritization by providing an estimate of the levels of product selectivity that can be expected in key transformations. For such methods to be useful practically there has to be a suitable balance between accuracy and throughput—sufficient accuracy has to be obtainable with a reasonable amount of computational resource and minimal manual input. We are investigating the possibilities for using computational chemistry for semi-quantitative predictions of different useful synthetic reaction types. Our computational approach is designed to assist the synthetic planning, on a daily basis, by answering questions such as: will this reaction give predominantly the desired product, or an undesired product isomer, or is the outcome uncertain? We recently reported a method for predicting product isomer ratios in electrophilic aromatic substitution reactions.1 In this Letter, we report how a similar approach can be applied to predict the outcome of certain types of nucleophilic aromatic substitution reactions. There are several mechanisms by which net nucleophilic aromatic substitution can occur. It can proceed through a unimolecular SN1 reaction via aryl diazonium ions,2 or via an elimination–addition mechanism that involves the ⇑ Corresponding authors. E-mail addresses: [email protected] (T. Brinck), [email protected] (M. Svensson). 0040-4039/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tetlet.2011.04.032

formation of a benzyne (dehydrobenzene) intermediate.3 Another possible mechanism is via transition metal catalyzed substitution, for example with copper4 or palladium catalysts.5,6 The mechanism can also proceed via a two-step addition–elimination mechanism, where the active nucleophile is added to a substituted aromatic carbon atom, followed by departure of the leaving group. The intermediate containing both the nucleophile and the leaving group is known as a Meisenheimer complex or r-complex. This is the type of nucleophilic aromatic substitution reaction, SNAr, we have focused on in this study.7,8 A number of theoretical studies have been carried out over the years to give reactivity indices for the different positions in substrates of SNAr reactions. Among the earlier ones is the Ip-repulsion theory based on calculating the fractional charge with Hückel theory,9,10 and an approach based on the Frontier Molecular Orbital method.11 More recent attempts include calculations based on Fukui indices,12 local softness and hardness reactivity descriptors,13 and dual descriptors for both electrophilicity and nucleophilicity.14 Many of these methods are quite successful in making qualitatively correct predictions of the selectivity pattern in SNAr reactions. However, none of these methods has proven suitable for quantitative predictions of isomer distributions since the structure and solvation of the transition state are not taken into account. One way to make quantitative predictions of the selectivity pattern in SNAr reactions is obviously to calculate the potential energy profile in each case, including the transition states, and theoretical investigations of the potential energy profile in vacuo have recently been

3151

M. Liljenberg et al. / Tetrahedron Letters 52 (2011) 3150–3153

performed within the DFT framework.15,16 At present, however, modeling of the potential energy profile is far too laborious to be routinely used in synthetic route design. Therefore, a simplified method for the quantitative prediction of selectivity in SNAr reactions would be highly valuable. In this work we use an alternative approach: we calculate the relative thermodynamic stability of each isomeric r-complex and predict the regioisomeric distribution on the basis of these data. A similar approach to modeling SNAr reactions has been used in three other reports. In two of them, the approach was used as a qualitative predictive model for substitution in aromatic perfluorocarbons, in other words, to predict the main site for substitution in such systems.17 In the third report, it was used in order to study regioselectivity in reactions with difluoroquinolines at different temperatures.18 The purpose of the work presented in this Letter is to evaluate the scope and limitations of this approach, and, in addition to earlier work,17 to consider the effect of the actual nucleophile and solvent used and evaluate if the approach can be used for quantitative predictions. This approach involves the following assumptions: first, that the reaction is kinetically controlled. Second that the energy differences between the isomeric transition states of the ratedetermining step can be approximated with the energy differences between the corresponding intermediate r-complexes, in accordance with the Bell–Evans–Polanyi principle or Hammond postulate. Third, that relaxation of the structure upon solvation is similar for the different regioisomers, and that this effect, therefore, will cancel out. Fourth, that the entropy terms (TDS) for the reactions forming the different regioisomers will be very similar, and these terms will also, therefore, cancel out (DDE ffi DDG). Calculations were carried out on all possible r-complexes. All substrates studied have at least two distinct substitution sites and some have as many as seven. The method can be described as follows: first, the geometry of each r-complex, without coordinated catalysts, promoters, or counterions, is optimized in vacuo using the DFT functional B3LYP with a DZP quality basis set.19,20 Second, the solvation free energy is taken into account by an a posteriori energy correction from single point calculations using the previously optimized structures and the Poisson–Boltzmann finite element solvation model (PBF) within the same software.19 In the single point calculations, additional diffuse basis functions were used for non-hydrogen atoms, in order to better take into

F F

N

F

4

4

F

F

N

3 6

F 4

5

F

F

F 3

3

6

F

2

N

F

F

N

2

F

F

2

N

F

F

F

2

CN

1

2

4

3 Me

F F

4

F

4

F

F

4

2

3

Me

F

3 6

5

2

N

F

F

5

6

F

2

F

(1) Reaction of 2,3,6-trifluoropyrazine (1) with methoxide ion, Scheme 1.21 Experimentally 100% of the 2-isomer found predicted 100% of the 2-isomer (>6 kcal/mol to the isomer closest in energy). (2) Reaction of tetrafluoropyrimidine (2) with azide ion, Scheme 2.22 Experimentally 100% of the 4-isomer found predicted 100% of the 4-isomer (>4 kcal/mol to the isomer closest in energy). (3) Reaction of pentafluoropyridine (3) with methoxide ion, Scheme 3.23 Experimentally 100% of the 4-isomer found predicted 100% of the 4-isomer (>6 kcal/mol to the isomer closest in energy). (4) Reaction of pentafluoro-cyanobenzene (4) with azide ion, Scheme 4.24 Experimentally >99% of the 4-isomer found (with ‘practically no ortho isomer’) predicted 99% of the 4isomer (2.8 kcal/mol to the isomer closest in energy, the ortho isomer). The results for the eight cases in which two or more isomers have been experimentally determined are shown in Table 1. Energies (relative to the lowest energy r-complex) and regioisomer ratios are shown for all calculated r-complexes that are within 4.0 kcal/mol of the lowest. The accuracy of the method in these cases, measured as an average absolute deviation, is 0.8 kcal/ mol.25 Such a deviation in energy corresponds (at rt) to a deviation of a factor of four in regioisomer distribution. If, for example, the predicted ratio between the two main isomers is 50:50, then an absolute deviation of 0.8 kcal/mol in the computed model corresponds to an experimentally observed ratio of 80:20 or 20:80. In a synthetic planning situation this prediction is sufficiently accurate to tell the chemist that the reaction is likely to yield a mixture, with both isomers in substantial amounts. In an example where the predicted ratio between the two main isomers is 98:2, an absolute deviation of 0.8 kcal/mol in the computed model corresponds to an experimentally observed ratio of 93:7 or 99.5:0.5, in which

F

F

4

N 8

7

F

F 6

F

6

5

F

3

Br

Br

F

N

6

account the electron-rich nature of the anionic r-complexes. Third, the distribution of isomers is calculated from the Boltzmann distribution of the r-complexes, at the temperature used experimentally in each specific reaction. Different conformations of the r-complexes were not considered in this study, but comparable starting conformers were used in each case. In order to highlight the ability of this method to give quantitative predictions, we have chosen to present separately the reactions in which only one isomer has been experimentally determined on the one hand, and the cases in which several isomers have been experimentally determined on the other. The molecules used in our investigation are all taken from the literature, and the structures and the labeling of their positions is shown schematically in Figure 1. The selection of experimental cases was guided both by the theoretical interest of a given case and by the intention of spanning the widest possible synthetically interesting space. These cases encompass a variety of anionic oxygen, nitrogen, and sulfur nucleophiles, and carbocyclic and heterocyclic substrates. We have investigated four cases in which only one isomer was experimentally determined, and they all gave correct predictions. The four cases are (the compound numbers in bold refer to Fig. 1):

4

F 3

7

8

2

1

F

F 9

F

F

N 8

1

F

F 10

Figure 1. The structures studied in this work.

F

N

MeO-

3

1

6

F

N

2

OMe

Scheme 1. Methoxylation of 1.

3152

M. Liljenberg et al. / Tetrahedron Letters 52 (2011) 3150–3153 N3 2

N3-

4

5

F

N 6

F

2

N

F

Scheme 2. Azidation of 2.

OMe MeO-

4

F

F 3

3 F

2

N

F

Scheme 3. Methoxylation of 3.

N3 N3-

F

4

F 3

4 2

F

F

CN

Scheme 4. Azidation of 4.

Table 1 Modeling of anionic SNAra Entry 1

2

3

4

5

6

Isomer

In vacuob

Solventb

Reaction of 5 with the anion of benzyl alcohol 2 0.0 (89) 0.0 (95) 4 1.2 (11) 1.7 (5)

Experimentalc 29

0.0 (95) 1.7 (5)

Reaction of 6 with the anion of benzyl alcohol29 3 0.0 (95) 0.0 (96) 4 1.7 (5) 1.8 (4)

0.0 (91) 1.3 (9)

Reaction of 7 with the anion of methanold,23 4 0.0 (98) 0.0 (96) 2 2.5 (1.5) 1.8 (4)

0.0 (95) 1.8 (5)

Reaction of 8 with the anion of methanold,21 6 0.0 (37) 0.0 (58) 2 -0.3 (63) 0.2 (42)

0.0 (90) 1.3 (10)

Reaction of 9 with the anion of methanol30 2 0.0 (75) 0.0 (88) 4 1.9 (3) 1.7 (5) 1 0.7 (22) 1.5 (7)

0.0 (70) 0.7 (20) 1.1 (10)

Reaction of 10 with the anion of methanold,e,31 1 0.0 (83) 0.0 (99) 6 0.6 (17) 1.7 (1) 8 3.1 (0.0) 4.3 (0.0)

0.0 (93) 1.0 (7) (0)

7

Reaction of 10 with the hydrogen sulfide aniond,e,31 1 4.2 (0.0) 1.7 (4) 1.3 (8) 6 0.0 (100) 0.0 (96) 0.0 (92) 8 4.1 (0.0) 3.6 (0.1) (0)

8

Reaction of 10 with the anion of methanethiold,e,31 1 4.2 (0.0) 3.1 (0.0) 0.3 (30) 6 0.0 (100) 0.0 (100) 0.0 (70)

a The compound numbers in bold refer to Figure 1. The calculated isomer distributions have been adjusted for degenerate positions. b Energies for the r-complexes, in kcal/mol, relative to the lowest energy rcomplex. The isomer distributions corresponding to these energy differences have been calculated, and are given in %, in parentheses. c The experimental isomer distributions (given in % in parentheses) have been determined in different ways, for example, isolated yields and 19F NMR, see references for the exact procedure in each case. The energy difference, in kcal/mol, corresponding to this distribution has been calculated by us. d No other isomers were reported experimentally. The isomers not included in this table had computed energies >4.0 kcal/mol, or had unreasonable structures. e The solvent calculation was performed without diffuse functions, due to technical convergence problems.

case the model with sufficient accuracy will predict that the main isomer is likely to be formed with high selectivity.26

The largest prediction error is 2.8 kcal/mol (Table 1, entry 8).27 We have also calculated the relative stabilities of some of the final products, and they do not correlate at all with the experimentally determined isomer distributions, which support our assumption of kinetic control. It is interesting to note that the proposed method correctly reproduces the experimental observation that the regioisomeric outcome changes from predominantly position 1 to position 6 when the nucleophile is changed from methoxide to hydrosulfide in reactions with hepta-fluoroisoquinoline (10) as the substrate (Table 1, entries 6 and 7). This would not have been possible to predict with a reactivity index model that uses only the substrate itself. The solvation calculation used in the method gives a slight improvement of the prediction. The average absolute deviation of the in vacuo calculations is 1.4 kcal/mol,25 compared to the already mentioned 0.8 kcal/mol after the solvation energy corrections. One cannot, however, expect this type of dielectric continuum calculations to predict adequately changes in regioisomeric outcome caused by a change of solvent. In a recent paper the SNAr reaction between azide anions and 4-fluoronitrobenzene was studied.28 The experimentally observed rate increase in going from protic to dipolar aprotic solvents could not be modeled by DFT/PCM calculations, but QM/MM Monte Carlo simulations gave useful results. In the case of electrophilic aromatic substitution (SEAr) the intermediate r-complex is, in general, much more reactive than the starting material, and is thus a good model for the TS, in accordance with the Bell–Evans–Polanyi principle.1 In contrast, this is not generally the case for its r-complex counterpart in SNAr reactions, however, and an approach similar to ours17 has been considered to be of limited value.32 The authors of the latter paper pointed out that an approach such as that proposed is less suited for activated substrates in particular, because it incorrectly assumes that the r-complex is a good model for the rate-determining TS and that coulombic effects (the ‘ortho-effect’) are not important.32 The objective of our approach, however, is limited to predictions of regioisomeric outcome, and is not aimed at reactivity comparisons between different substrates. We have included a number of substrates that are activated compared to the fluorinated carbocyclic substrates in our investigation (Fig. 1, compounds 1–4, 7–8 and 10), and our results indicate that the accuracy of this approach is sufficiently good to make it useful for quantitative predictions. The crucial requirement, for our purpose, is not that the r-complex is a good model of the ratedetermining TS, but that the energy differences between the r-complexes mimic those between the transition states of the rate-determining step, and this seems to be the case for the substrates in this study. In summary, we have demonstrated that the regioisomer distribution in kinetically-controlled nucleophilic substitution reactions of aromatic fluorides can be computed with an average accuracy of 0.8 kcal/mol. This limited scope is of considerable synthetic preparative interest. In terms of throughput, we can in general obtain an answer for a problem of average complexity in one working day. Furthermore, the simplicity of our protocol also makes it useful for a non-expert. Acknowledgment We thank Simone Tomasi and Ian Ashworth (AstraZeneca R&D Macclesfield) for stimulating discussions.

Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.tetlet.2011.04.032.

M. Liljenberg et al. / Tetrahedron Letters 52 (2011) 3150–3153

References and notes 1. Liljenberg, M.; Brinck, T.; Herschend, B.; Rein, T.; Rockwell, G.; Svensson, M. J. Org. Chem. 2010, 14, 4696–4705. 2. March, J. Advanced Organic Chemistry, second ed.; McGraw-Hill International Book Company, 1984. 3. Hoffman, R. W. Dehydrobenzene and Cycloalkynes; Academic Press: New York, 1967. 4. Frlan, R.; Kikelj, D. Synthesis 2006, 2271–2285. 5. Yang, B. H.; Buchwald, S. L. J. Organomet. Chem. 1999, 576, 125–146. 6. Jiang, L.; Buchwald, S. L. In Metal-Catalysed Cross-Coupling Reactions; 2nd ed.; de Meijere, A., Diederich, F., Eds., Wiley-VCH Verlag GmbH: Weinheim, Germany, 2004; pp 699–760. 7. Zoltewicz, J. A. Top. Curr. Chem. 1975, 59, 33–64. 8. Miller, J. Aromatic Nucleophilic Substitution; Elsevier: Amsterdam, 1968. 9. Burdon, J. Tetrahedron 1965, 21, 3373–3380. 10. Burdon, J.; Parsons, I. W. J. Am. Chem. Soc. 1977, 99, 7445–7447. 11. Epiotis, N. J.; Cherry, W. J. Am. Chem. Soc. 1976, 98, 5432–5435. 12. Bulat, F. A.; Chamorro, E.; Fuentealba, P.; Toro-Labbe, A. J. Phys. Chem. A 2004, 108, 342–349. 13. Roy, R. K.; Krishnamurti, S.; Geerlings, P.; Pal, S. J. Phys. Chem. A 1998, 102, 3746–3755. 14. Morell, C.; Grand, A.; Toro-Labbe, A. J. Phys. Chem. A 2005, 109, 205–212. 15. Giroldo, T.; Xavier, L. A.; Riveros, J. M. Angew. Chem., Int. Ed. 2004, 43, 3588– 3590. 16. Glukhovtsev, M. N.; Bach, R. D.; Laiter, S. J. Org. Chem. 1997, 62, 4036–4046. 17. Muir, M.; Baker, J. J. Fluorine Chem. 2005, 126, 727–738; Baker, J.; Muir, M. Can. J. Chem. 2010, 88, 588–597. 18. Politanskaya, L. V.; Malysheva, L. A.; Beregovaya, I. V.; Bagryanskaya, I. Y.; Gatilov, Y. V.; Malykhin, E. V.; Shteingarts, V. D. J. Fluorine Chem. 2005, 126, 1502–1509. 19. Jaguar, version 7.0; Schrödinger, LLC: New York, 2007.

3153

20. The basis set used is 6-31G⁄⁄. 21. Sodium methoxide in methanol at rt. Chambers, R. D.; Musgrave, W. K. R.; Urben, P. G. J. Chem. Soc., Perkin Trans. 1 1974, 2580–2584. 22. NaN3 in acetonitrile at 0 °C. Banks, R. E.; Prakash, A.; Venayak, N. D. J. Fluorine Chem. 1980, 16, 325–338. 23. Sodium methoxide in methanol at rt. Chambers, D. R.; Seabury, M. J.; Williams, D. L. H.; Hughes, N. J. Chem. Soc., Perkin Trans. 1 1988, 255–257. 24. NaN3 in acetone/water at reflux Keana, J. F. W. J. Org. Chem. 1990, 55, 3640– 3647. 25. In order to measure the accuracy, we calculated the energy differences between the two experimentally determined major isomers and the corresponding differences for the proposed method. For each reaction, we then calculated the deviation between the experimental energy difference and the calculated energy difference, and so obtained the average absolute deviation. 26. Provided that these reactions will proceed at all—this model has nothing to say about absolute reaction rates. 27. This could be an indication that for some reactions involving sulfur nucleophiles, TS modeling is required to achieve high accuracy. We will return to this question in forthcoming work. 28. Acevedo, O.; Jorgensen, W. L. Org. Lett. 2004, 6, 2881–2884. 29. Benzyl alcohol with excess NaH in THF at rt overnight. Dirr, R.; Anthaume, C.; Desaubry, L. Tetrahedron Lett. 2008, 49, 4588–4590. 30. Sodium methoxide in methanol at rt. Chambers, D. R.; Seabury, M. J.; Williams, D. L. H. J. Chem. Soc., Perkin Trans. 1 1988, 251–254. 31. Sodium methoxide in methanol at 82 to 84 °C (entry 6). Sodium hydrosulfide (NaHS) in DMF and ethylene glycol at 5 to 2 °C (entry 7). Sodium methanethiolate in ethanol at 85 to 90 °C (entry 8). Brooke, G. M.; Chambers, R. D.; Drury, C. J.; Bower, M. J. J. Chem. Soc., Perkin Trans. 1, 1993, 2201-2209. 32. Chambers, R. D.; Martin, P. A.; Sandford, G.; Williams, D. L. H. J. Fluorine Chem. 2008, 129, 998–1002.