Report

Generating Diverse Skeletons of Small Molecules Combinatorially

See allHide authors and affiliations

Science  24 Oct 2003:
Vol. 302, Issue 5645, pp. 613-618
DOI: 10.1126/science.1089946

Abstract

Lack of efficient access to collections of synthetic compounds that have skeletal diversity is a key bottleneck in the small-molecule discovery process. We report a synthesis strategy that involves transforming substrates with different appendages that pre-encode skeletal information, named σ elements, into products that have different skeletons with the use of common reaction conditions. With this approach, split-pool synthesis can be used to pre-encode skeletal diversity combinatorially and thereby generate such small molecules very efficiently. A split-pool synthesis of more than 1000 compounds produced overlapping, combinatorial matrices of molecular skeletons and appended building blocks in both enantiomeric and diastereomeric forms.

The macromolecules that carry out the many functions required for life have enormous structural diversity, and this suggests that complementary levels of structural diversity will be needed in collections of candidate small molecules in order to find specific modulators for each of those functions. Diversity-oriented synthesis (DOS) is being used by organic chemists with the aim of populating chemical space efficiently with small molecules that have complex and diverse molecular skeletons (1, 2). Efficient access to skeletal complexity can be achieved in DOS using pairs of complexitygenerating reactions, where the product of one is the substrate for another (13). Gaining efficient access to skeletal diversity, however, has proven to be much more challenging (46). Achieving this goal in a format amenable to screening in biological assays is likely to benefit the field of chemical genetics, where small molecules are used in a systematic way to perturb and thereby understand protein function (2), and may also find use in the pharmaceutical industry, where small molecule–mediated modulation of protein function is used to promote and restore human health.

The synthesis strategy most commonly used to access diverse collections of small molecules involves appending different sets of building blocks to a common molecular skeleton (7, 8). If this molecular skeleton has multiple reactive sites with potential for orthogonal functionalization, the technique of split-pool synthesis (911) can be used to harness the power of combinatorics (a multiplicative increase in the number of products with an additive increase in the number of reaction conditions) and thereby generate all possible combinations of building blocks (i.e., the complete matrix) very efficiently (Fig. 1A). This one synthesis–one skeleton strategy has proven to be general and capable of generating hundreds, thousands, or even millions of distinct small molecules in just three to five steps (12, 13). However, although this approach is highly efficient, its impact in the academic and pharmaceutical realms has been limited (14), likely because compounds that have a common molecular skeleton display chemical information similarly in three-dimensional (3D) space, thus limiting the pool of potential binding partners to only those macromolecules with a complementary 3D binding surface.

Fig. 1.

Synthesis strategies for generating building block and skeletal diversity in DOS. (A) The one synthesis–one skeleton approach for generating building block diversity combinatorially (the diamond-filled arrow is used to represent a split-pool step). (B) The σ element–based synthesis strategy for generating skeletal diversity combinatorially. (C) A hybrid synthesis strategy for generating overlapping, combinatorial matrices of building block and skeletal diversity elements. (D) Forward-synthetic plan for a σ element–based DOS pathway. (E) Transformation of substrates having different σ elements into products with different skeletons using a common set of reagents. Conditions (22) are as follows: (a) Tf OH, DCM, room temperature (rt); 5-hexen-1-ol, 2,6-lutidine, DCM, rt. (b) 9-BBN, THF, rt; 5-bromofuraldehyde, PdCl2dppf, NaOH, THF-H2O, 65°C, loading 0.679 meq./g. (c) allyldiethylphosphonoacetate, LiOH, THF, rt. (d) Pd(PPh3)4, thiosalicylic acid, THF, rt. (e) isobutylchloroformate, NMM, iPr2NEt, THF, 0°C; LiBH4, iPr2NEt, THF, 4°C, 76% (three steps), purity 68%. (f) PhNCO, pyridine, DCM, rt. (g) OsO4, (DHQD)2PHAL, NMO, TEAAT, Acetone-H2O, 4°C, 65% (two steps), purity >90%. (h) (S)-(+)-4-benzyl-3-propionyl-2-oxazolidinone, n-Bu2BOTf, Et3N, DCM, –78° to 0°C; 30% aqueous (aq.) H2O2, pH 7 buffer, MeOH, 4°C, >95%, purity >90%. (i) Ac2O, iPr2NEt, DMAP, DCM, rt, 88%, purity >90%. (j) NBS, NaHCO3, NaOAc, THF-H2O, rt; PPTS, DCM, 40° to 45°C, 20 hours. 8: 33%, purity 64% (31); 9: 35%, purity 86%; 10: 81%, purity >90%. Purities determined by LCMS (UV detection, λ214).

Therefore, we set as our aim the development of an alternative synthesis strategy with the potential to generate collections of compounds representing many different molecular skeletons as efficiently as the one synthesis–one skeleton approach generates collections of compounds representing a single molecular skeleton decorated with many combinations of building blocks. Achieving this goal requires the ability to generate skeletal diversity, rather than building block diversity, combinatorially. Toward this end, we envisioned replacing building blocks with skeletal information elements (σ elements), which we define as appendages that preencode skeletal information and thereby cause substrates with a common skeleton to be transformed into products with different skeletons under a common set of reaction conditions (Fig. 1B). As demonstrated in this report, sets of σ elements can be identified that act in combination, i.e., a complete matrix of σ elements can pre-encode a complete matrix of skeletal outcomes, thus making it possible to generate skeletal diversity combinatorially. Moreover, we demonstrate the use of encoded split-pool synthesis to generate a collection of ∼1260 compounds representing both a matrix of σ elements and a matrix of building blocks appended to the same skeleton, followed by the transformation of these pooled substrates into ∼1260 products representing a complete, combinatorial matrix of molecular skeletons, each derivatized with a complete, combinatorial matrix of building blocks in both enantiomeric and diastereomeric forms (Fig. 1C).

To realize the potential of this σ element–based strategy for generating diverse skeletons combinatorially in the context of split-pool synthesis, the transformation of substrates with different σ elements into products with different skeletons using common reaction conditions is a requirement (15), and it was unclear at the outset of our studies whether this would be possible. Such a transformation can be planned by first identifying a relatively unreactive core structure that can be transformed under mild conditions into a more reactive intermediate. If different σ elements that have complementary reactivity with this latent intermediate are appended to this common core, then, in theory, these mild conditions can be used to liberate the latent reactive intermediate and allow this complementary reactivity to be realized, resulting in the formation of different skeletons (i.e., diverse displays of chemical information in 3D space).

For example, as shown in Fig. 1D, the aromatic furan ring is a relatively unreactive core structure that can be transformed into a more reactive, electrophilic cis-enedione intermediate under mild oxidative reaction conditions (1618). We anticipated that a collection of three substrates with distinct two-carbon side chains (i.e., σ elements) containing two, one, or zero nucleophilic hydroxyl groups could be transformed into a collection of products with distinct molecular skeletons (1621) using an identical set of oxidative and acidic reaction conditions (22). To explore this possibility, we developed the reaction pathway shown in Fig. 1E. A palladium-catalyzed B-alkyl Suzuki reaction was used to generate macrobead-bound (23) furaldehyde 3 from the terminal olefin 2. This furaldehyde precursor was then converted into three different products 5, 6, and 7 containing two-carbon side-chains with two, one, and zero nucleophilic hydroxyl groups, respectively. Specifically, a sequence of Horner-Wadsworth-Emmons olefination, deallylation, reduction, carbamate formation, and Sharpless asymmetric dihydroxylation converted 3 into the diol 5. Alternatively, the Evans aldol reaction (24), with or without subsequent acetylation of the hydroxyl group of the resulting aldol adduct, transformed the same furaldehyde precursor 3 into the two products 6 and 7.

After screening a variety of oxidative and acidic reaction conditions, we were successful in identifying a common set, N-bromosuccinimide (NBS) in THF:H2O 4:1 and pyridinium p-toluene sulfonate (PPTS) in CH2Cl2, that were effective in transforming these three substrates 5, 6, and 7 with distinct σ elements appended to a common furan core into the three products 8, 9, and 10, each with a distinct molecular skeleton. The diol 5 underwent NBS-mediated oxidative ring expansion and subsequent bicycloketalization (18, 19) to yield the [3.2.1] bicycle 8. The aldol adduct 6 containing one flanking hydroxyl group underwent initial NBS-mediated oxidative ring expansion to yield an isolable, intermediate cyclic hemiketal (18, 20) followed by an unanticipated, PPTS-catalyzed dehydration to yield the alkylidene pyran-3-one 9 as a single geometric isomer. Finally, the acetylated aldol adduct 7 underwent oxidative furan ring opening followed by olefin isomerization (17, 21) to yield the trans-enedione 10.

Having demonstrated the transformation of substrates having different σ elements into products with different skeletons using common reaction conditions, we set out to determine whether this approach could generate skeletal diversity combinatorially. To do so requires the identification of at least two sets of σ elements that can be appended at different sites and can function in combination to pre-encode a matrix of distinct skeletal outcomes (Fig. 1B). Toward this end, it was determined during the course of our studies that different appendages at the 4 position of the furan core can also effect a variety of distinct skeletal outcomes; i.e., appendages at this position function as a second σ element. Specifically, as shown in Fig. 2A, we varied the 5-bromofuraldehyde unit used in the Suzuki reaction to generate both the 4-bromo (25) and 4-aryl derivatives 11 and 12. These 4-substituted furaldehydes were excellent substrates for the Evans aldol reaction and subsequent acetylation resulting in formation of products 13 to 16. When exposed to the same oxidative (NBS) and acidic (PPTS) conditions used previously, each substrate 13 to 16 was transformed into a product with a distinct molecular skeleton. The 4-bromo-α-hydroxy furan 13 underwent oxidative ring expansion without subsequent acid-catalyzed dehydration to yield the cyclic hemiketal 17 as a >9:1 mixture of epimers (26). The 4-bromo-α-acetoxy furan 14, having two electron-withdrawing appendages, was completely resistant to oxidation and remained unchanged upon exposure to these reaction conditions. Treatment of the 4-aryl-α-hydroxy furan 15 with NBS resulted in initial oxidative ring expansion to yield an isolable, aryl-substituted cyclic hemiketal similar to 17, which upon exposure to PPTS underwent a ring contraction, dehydration, and rearomatization reaction to yield the α-keto furan 18 (27). Finally, upon exposure to the same NBS or PPTS conditions, the 4-aryl-α-acetoxy furan 16 underwent oxidative furan cleavage without subsequent olefin isomerization to yield the cis-enedione 19.

Fig. 2.

Generating skeletal diversity combinatorially. (A) X = (S)-4-benzyl-2-oxazolidinone, Ar = m-methylphenyl. Conditions (22) are as follows: (a) 9-BBN, THF, rt; 4,5-dibromofuraldehyde, PdCl2dppf, NaOH, THF-H2O, 65°C, loading 0.188 meq./g. (b) 9-BBN, THF, rt; 4-m-MePh-5-bromofuraldehyde, PdCl2dppf, NaOH, THF-H2O, 65°C, 0.545 meq./g. (c) (S)-(+)-4-benzyl-3-propionyl-2-oxazolidinone, n-Bu2BOTf, Et3N, DCM, –78° to 0°C; 30% aq. H2O2, pH 7 buffer, MeOH, 4°C. 13: >95%, purity >90%; 15: 95%, purity >90%. (d) Ac2O, iPr2NEt, DMAP, DCM, rt. 14: 90%, purity >90%; 16: 84%, purity >90%. (e) NBS, NaHCO3, NaOAc, THF-H2O, rt; PPTS, DCM, 40° to 45°C. 17: 82%, purity 90%; 14': 88%, purity >90%; 18: 74%, purity 72%; 19: 72%, purity 66%. (B) A complete, combinatorial matrix of σ elements resulted in a complete matrix of distinct skeletal outcomes under a common set of reaction conditions (22). (C) To provide a metric for the skeletal diversity (28) generated with this σ element–based transformation, we generated abbreviated structures that share in common the seven contiguous carbon atoms labeled C1 to C7 by replacing each of the missing bonds in the 12 structures shown in Fig. 2B with methyl groups (or a methylene group for the “left side” of structure 9). Determination of equilibrium conformer and equilibrium geometry using semiempirical (AM1) and Hartree-Fock (6-31G*) calculations, respectively, produced 3D structures from which the following geometric parameters were derived (each parameter provides unique information regarding the relative positions of the building block attachment sites, C1 and C7, in 3D space): (i) the distance in Å between C1 and C7, (ii) the angle C1–(the midpoint between C3 and C5)–C7, and (iii) the dihedral angle comprising C1, C3, C5, and C7 (every other carbon). As shown in (C), when these three parameters are plotted, the six substrates (representative of the compounds typically derived from the one synthesis–one skeleton approach) create a dense cluster (left). In contrast, the six products (having distinct skeletons generated using the σ element–based approach) distribute much more broadly (right), consistent with a diverse display of chemical information in 3D space.

Combining results from Fig. 1E and Fig. 2A, it was possible to assemble a collection of six macrobead-bound substrates 6, 7, and 13 to 16 [one substrate per macrobead (Fig. 2B)] representing a complete, combinatorial matrix of σ elements (–H, –Br, or –Ar at the 4-position of furan combined with –OH or –OAc on the α-carbon). These six individual macrobeads were placed in the same reaction vessel and collectively exposed to the same oxidative and acidic reaction conditions described above, resulting in a complete, non-redundant, combinatorial (3 × 2 = 6) matrix of distinct skeletal outcomes, i.e., a multiplicative increase in skeletons with an additive increase in σ elements, in the form of products 9, 10, 14', and 17 to 19 (Fig. 2B). A metric for evaluating the diversity of the display of chemical information in 3D space generated in this collective transformation is presented in Fig. 2C (28).

These results demonstrate the potential of this σ element–based strategy to harness the power of combinatorics and thereby generate a complete matrix of distinct molecular skeletons (as opposed to a complete matrix of building blocks) very efficiently. We next set out to determine whether this combinatorial matrix of σ elements could prove to be general and effectively pre-encode the same matrix of distinct skeletal outcomes when a complete, combinatorial matrix of building blocks was also appended to the same common core (Fig. 1C). If successful, this strategy would provide a highly efficient mechanism to access a collection of compounds representing a set of complete, overlapping matrices of these diversity elements, i.e., a complete combinatorial matrix of molecular skeletons, each derivatized with a complete combinatorial matrix of building blocks (the equivalent of several different collections of compounds synthesized individually using the one synthesis–one skeleton approach).

To explore this potential, we first identified two sets of candidate building blocks, BB1 and BB2, by determining that structurally diverse coupling partners could be used in the Suzuki and Evans aldol reactions (Fig. 3A). We then prepared a collection of 36 compounds 20a to 20jj representing all possible combinations of a (2 × 3) matrix of these candidate building blocks and the (3 × 2) matrix of σ elements described above. After cleaving each of these 36 substrates from macrobeads (∼5 mg of macrobeads were cleaved yielding ∼0.5 mg of each compound), 36 of 36 (100%) of the predicted structures were confirmed by 1H NMR (nuclear magnetic resonance) and HRMS (high-resolution mass spectroscopy) (error < 5 ppm), and 35 of 36 (97%) of these compounds were determined to be ≥ 70% pure by LCMS (liquid chromatography–mass spectrometry) analysis [ultraviolet (UV) detection at λ214]. We then exposed these 36 substrates to the same set of oxidative and acidic reaction conditions described above and characterized all of the resulting products (after cleaving from macrobeads) using 1H NMR, LCMS, and HRMS. As shown in Fig. 3B, all three forms of characterization were consistent with the formation of the anticipated functionalized skeleton in 36 of 36 cases (100%, HRMS error < 5 ppm), and 26 of 36 (72%) of these products were determined to be ≥70% pure by LCMS analysis. These results demonstrate that a common set of reaction conditions were effective in transforming these 36 substrates, representing all possible combinations of σ elements and building blocks appended to a common α-alkoxy furan core, into a collection of 36 products representing a complete, combinatorial (3 × 2 = 6) matrix of molecular skeletons, each derivatized with a complete, combinatorial (2 × 3 = 6) matrix of building blocks.

Fig. 3.

Parallel synthesis of a complete, combinatorial matrix of molecular skeletons, each derivatized with a complete, combinatorial matrix of building blocks (22). (A) Four sets of appendages attached to a common α-alkoxy furan skeleton, two that do (σ elements σ1 and σ2), and two that do not (building blocks BB1 and BB2) influence the formation of distinct molecular skeletons upon exposure of a collection of substrates to a common set of reagents. (B) Transformation of this collection of 36 substrates 20a to 20jj into a collection of 36 products representing complete, overlapping matrices of molecular skeletons and appended building blocks.

Finally, we demonstrated the potential of this σ element–based strategy to generate overlapping, combinatorial matrices of molecular skeletons and appended building blocks in the context of a highly efficient, five-step, fully encoded split-pool synthesis (Fig. 4). We first expanded our collections of candidate building blocks to include seven commercially available, terminal olefin-containing primary alcohols (BB1A to BB1G) and 15 acyl oxazolidinone coupling partners (BB2AS to BB2OS: a complete matrix of five commercially available, nonracemic, chiral oxazolidinones and three different acyl side chains) shown in Fig. 4A. The 15 enantiomeric acyl oxazolidinones (BB2AR to BB2OR) were also prepared, allowing us to take advantage of reagent-based stereocontrol to generate both sets of possible enantiomeric and diastereomeric (when BB1 is chiral) aldol adducts. We then confirmed that our synthesis pathway was compatible with the Still chemical encoding technology (23, 29), and we carried out a fully encoded split-pool synthesis (Fig. 4B).

Fig. 4.

(A) Building blocks. (B) Split-pool synthesis of ∼1260 compounds 55 representing a complete, combinatorial (3 × 2 = 6) matrix of molecular skeletons, each derivatized with a complete, combinatorial (7 × 15 = 105) matrix of building blocks in both enantiomeric and diastereomeric forms (6 × 105 × 2 = 1260) (22).

Four consecutive split-pool steps were used to generate efficiently a collection of ∼1260 compounds 54 representing a set of overlapping matrices of σ elements (σ1 × σ2) and building blocks (BB1 × BB2) appended to a common α-alkoxy furan skeleton in both enantiomeric and diastereomeric forms. The compound and chemical tags were cleaved (23) from 60 individual macrobeads 54 and analyzed by LCMS and gas chromatography (GC), respectively. These data were found to be consistent for 60 of 60 (100%) of these macrobeads, and the compounds cleaved from 55 of 60 (92%) of these macrobeads were determined to be ≥70% pure by LCMS analysis. We then placed this pooled collection of ∼1260 macrobead-bound substrates [∼4410 macrobeads; multiplicative factor, 3.5 (30)] in a single reaction flask and exposed them to the same oxidative and acidic reaction conditions described above to yield a collection of ∼1260 products 55 representing a complete, combinatorial (3 × 2 = 6) matrix of molecular skeletons, each derivatized with a complete, combinatorial (7 × 15 = 105) matrix of building blocks in both enantiomeric and diastereomeric forms (6 × 105 × 2 = 1260). LCMS analysis of compounds cleaved from 120 of these macrobeads was consistent with formation of the functionalized skeleton encoded by the corresponding chemical tags in 120 out of 120 cases (100%). Moreover, 84 of 120 (70%) of these final products were determined to be ≥70% pure by LCMS analysis.

Forming diverse skeletons late in the synthesis pathway (in contrast to forming a skeleton first, as is typical with the one synthesis–one skeleton approach) facilitates the generation of functionalized skeletons that might otherwise be difficult to access, such as those that have building blocks coupled via carbon-carbon bonds at stereogenic quaternary carbon centers (e.g., 17 and related products) and those that have potentially unstable structural elements (e.g., enediones 10 and 19 and related products). Also, the maintenance of structural similarity and, therefore, common reactivity until late in the synthesis pathway facilitates the realization of this strategy using the efficient split-pool technique. Moreover, this approach can be used to generate a collection of compounds representing overlapping matrices of molecular skeletons and appended building blocks in both enantiomeric and diastereomeric forms (e.g., 55). To the best of our knowledge, such a collection of nonoligomeric small molecules potentially representing all possible combinations of building block, stereochemical, and skeletal diversity elements is unprecedented. Systematic screening of this collection of compounds should advance our fundamental understanding of the roles these three diversity elements play in small molecule-protein interactions.

Supporting Online Material

www.sciencemag.org/cgi/content/full/302/5645/613/DC1

Materials and Methods

Tables S1 to S7

References and Notes

View Abstract

Navigate This Article