Library design and synthesis. The synthetic route of sub-library B (SL-B) is shown in Fig. 1C. We designed two 33-nt oligonucleotides that were partially complementary with 6 plus 12 base pairs. Various bi-functional building blocks were conjugated to the oligonucleotides at the 3' or 5' termini, resulting in 3' and 5' conjugates, respectively (Figure S1). Then, the conjugates underwent different DNA-templated reactions between matching functional groups to generate DNA-compound-DNA conjugates. The reactions were monitored by denaturing urea PAGE, and the reaction products were purified from the gel. The molecular weight of the reaction products was confirmed by LC-ESI-MS (Fig. 1D and Supporting Information 6). By employing a variety of reaction types, such as amide bond formation, reductive amination, azide-alkyne cycloaddition, Michael addition, and Diels-Alder reaction, we have generated 30 conjugates covering four structural categories (Figure S2). Each DNA-compound-DNA conjugate was then encoded by splint ligation. The encoding process was also monitored by denaturing PAGE, and only the successfully encoded conjugate was purified from the gel to ensure the high purity of the library members of SL-B (Fig. 1D).
An 883-member fragment sub-library (SL-A) and an 890-member fragment sub-library (SL-C) were synthesized to form a dynamic dual-pharmacophore DEL15,18, which are partially complementary by 6 bp. SL-A and SL-B share a 33 bp complementary region, and SL-B and SL-C share a 13 bp complementary region. We examined whether the SL-B can assemble with the SL-A and SL-C to form a stable T-DEL using native DNA PAGE. As shown in Fig. 1E, when SL-A and SL-B (lane 4) or SL-B and SL-C (lane 6) were mixed and allowed to anneal, the bands indicative of the assembled duplexes were observed. As expected, the mixture of SL-A and SL-C did not form a larger complex (lane 5). When all three sub-libraries were mixed and allowed to anneal (lane 7), the highest band corresponding to the assembled trimeric complex was observed.
T-DEL for linker optimization. To investigate the use of T-DEL to optimize linkage between fragment pairs, we performed affinity maturation selections against the model proteins bovine carbonic anhydrase II (CAII) and bovine trypsin with their known ligand pairs. As depicted in Figure S3A, we utilized the reported fragment pair of CAII15, aryl sulfonamide and 3-{5-[3-(trifluoromethyl) phenyl]-2-furyl} acrylic acid (compound A) as single-member SL-A and SL-C, respectively. After assembling with the 30-member SL-B, the T-DEL library was selected against CAII immobilized on solid support (Fig. 2A, target selection). Selection against blank solid support served as a negative control. Selection with SL-B assembled with non-modified SL-A and SL-C was also performed (Figures S3A and 2A, no-ligand target selection). qPCR was used to quantify the amount of each member of SL-B in the three selections with code-specific primers (Supporting information 7 and Figure S49). In Fig. 2A, the enrichment was calculated by normalizing the enrichment profile against no-target selection. As expected, the enrichment of the entire SL-B was higher in the target selection than in the no-ligand target selection, demonstrating that the ligand pair facilitates the interaction of SL-B members with the target.
We have chosen two compounds with the highest enrichment (c1 and c2), two with moderate enrichment (c3 and c4), and one compound, c5, with low enrichment for further off-DNA synthesis and validation. These selected structures from SL-B were used to connect sulfanilamide and compound A, resulting in compounds C-1 to C-5 (Fig. 2B). We also synthesized compound C-0 by connecting sulfanilamide with compound A without a linker moiety. The compounds C-0 to C-5 were validated in an enzyme inhibition assay to measure the IC50 values. Sulfanilamide showed an IC50 value of 13.36 µM, and compound A exhibited moderate inhibition at 100 µM (Fig. 2C). The compound with the highest enrichment (C-2) displayed a 20-fold improvement in the IC50 value (0.67 µM). Compounds with moderate and low enrichment, C-3, C-4, and C-5, exhibited lower inhibitory effects than C1 and C2, agreeing with the selection outcome. Interestingly, C-0 showed the second-highest inhibitory effect (IC50 0.83 µM).
We implemented molecular docking studies to gain more insights into the compounds' binding mechanism and compared the docking poses among the compounds (Figs. 2, S4, S5, and supporting information 5.1). As reported previously38,39, the sulfonamide moiety binds deeply in the catalytic site via coordinating with Zn2+, and forming two hydrogen bonds with Thr198, and one with Pro200 (Figure S4B). Conjugation of compound A to sulfanilamide contributed predominantly to the hydrophobic interactions with the protein, as shown in Figure S4A. The sulfanilamide moiety remained well-positioned in the active site in all re-synthesized compounds (C-0 to C-5) (Figs. 2 and S4). We then investigated the docking pose of each compound to understand the different inhibitory effects associated with the linker moieties. The binding pose of C-0 resembled the ligand in the reported crystal structure (PDB:6skv) (Figs. 2D and S5). C-2 adopted a compact conformation in the catalytic pocket, forming five hydrogen bonds with the surrounding residues (Fig. 2E). Also, the large hydrophobic effect and low binding energy may support the highest inhibitory effect of C-2 (Figure S4A). On the contrary, the linker moiety of C-3 and C-4 protruded out of the catalytic pocket (Fig. 2F), which may explain their lower inhibitory effects.
Next, we tested the use of T-DEL for linker optimization with bovine trypsin and its ligand pair, 4-aminomethyl benzamidine and 2-iodophenyl isothiocyanate (compound B), reported by the Neri group in their DEL selection with a dual-pharmacophore library40. The selection and decoding strategies are identical to CAII (Figure S3B). We have chosen the four highly enriched linker fragments, t1 to t4, and one with low enrichment, t5 (Fig. 3A). The linkers were used to tether the fragment pair to generate small molecules T-1 to T-5 (Fig. 3B). Again, the two fragments were directly conjugated without a linker, resulting in compound T-0. The compounds were evaluated by an enzyme inhibition assay. 4-aminomethyl benzamidine showed an IC50 value of 147.23 µM, in agreement with the previous report40, while compound B alone did not display any detectable inhibitory effect. T-1 to T-4 showed remarkable enhancement in the inhibitory effect, especially T-2 and T-3, displaying 70-fold and 30-fold improvement, respectively. T-0 showed an approximately 9-fold improvement (Fig. 3C).
We further studied the binding modes of compounds to trypsin by molecular docking (Figs. 3, S6, S7, and supporting information 5.2). As previously reported41–43, 4-aminomethyl benzamidine binds to the substrate recognition site and forms hydrogen bonds with the key residue Asp189, the neighboring Ser190, Gly219, and Ser195 (Fig. 3D). In analogy to CAII, conjugating compound B to 4-aminomethyl benzamidine by linkers largely increased the hydrophobic contacts with trypsin, maintaining the binding mode of the benzamidine moiety (Figure S6). T-2 and T-4 displayed similarity in terms of the binding site and pose of both fragments, agreeing with their observed inhibitory effects (Figs. 3, S6A, and S7A). Notably, T-3 preserved the binding pose of the single compound B best in all conjugates (Figs. 3 and S7A), and T-5 displayed the worst docking score compared to other resynthesized small molecules (Figure S6A).
Together, the affinity maturation selections have demonstrated the capability of T-DEL to guide linker optimization for known fragment pairs.
T-DEL for de novo selections. A 23.576 million-member T-DEL (883 x 30 x 890) was constructed to test its utility in de novo selections. Matrix metalloproteinases (MMPs) are zinc-dependent endopeptidases capable of degrading and remodeling extracellular matrix components44,45. They are attractive therapeutic targets as high expression levels were detected in various diseases, such as inflammatory diseases, and at different stages of cancers, including metastasis, invasion, and angiogenesis45–51. As shown in Figure S3C, we performed selections against the two gelatinases (human MMP-2 and human MMP-9) to identify binding fragments for later design and synthesis of small molecule inhibitors. After selection, the three sub-libraries were decoded, and the enrichment was calculated by dividing the post-selection fraction (count/total counts) by the pre-selection fraction (Figs. 4A and S8). Selections using a dynamic dual-pharmacophore DEL with the same members (883x890) were also performed against MMP-2 and MMP-9 to select relevant hits that enrich through different DEL formats (Figures S3D and S8).
The enrichment profiles of all three sub-libraries have shown similarities between MMP-2 and MMP-9, presumably due to the high structural homology of the two proteins52,53 (Fig. 4B). To validate the selection outcome, we chose three fragments from SL-A (66, 182, and 693), three fragments from SL-C (787, 826, and 828), and three linker fragments with the highest enrichment (12, 24, and 10), and two linker fragments with moderate enrichment (1 and 4) from SL-B for further off-DNA synthesis (Fig. 4). We deployed enzyme inhibition assays of MMP-2 and MMP-9 using a fluorogenic peptide substrate for hit validation. We first measured the inhibition of two enzymes by the fragments. As shown in Figs. 5A and 5B, fragment 182 exhibited the highest inhibitory effect with IC50 values of 95.8 µM and 48.1 µM against MMP-2 and MMP-9, respectively. Fragments 693 and 828 displayed IC50 values in the high µM range against both targets, while fragment 787 showed a high µM IC50 value only against MMP-9. No inhibitory effects could be measured for fragments and 826.
Since the three sub-libraries were independently decoded and analyzed, in order to identify the best combination of the selected fragments, we synthesized 45 (3x5x3) small molecules covering all possible combinations (Figure S9). Figure 5C shows the MW distribution of the 45 compounds. They comply with the current drug-likeness criteria regarding MW, e.g., showing an average MW of 503 Da and 90th percentile of 606 Da, similar to the analysis of all approved drug molecules in the past 20 years30. We then assayed the compounds against MMP-2 and MMP-9 (Figures S10-S12), and the resulting IC50 values are shown in Figs. 5D and 5E. The compounds are grouped by the fragment combinations, and each group has five compounds differing by the linker fragments. Like the enrichment profiles (Fig. 4B), the inhibitory effects of the small molecules showed similar patterns on MMP-2 and MMP-9. For both enzymes, the combinations 182 + 828, 182 + 787, 693 + 828, and 693 + 787 displayed higher inhibitory effects than the other pairs, suggesting a synergistic effect from the combinations of these fragments. 66 and 826, showing the weakest inhibition as fragments, resulted in weak binders after connecting them with various linkers. In addition to the fragment pairing, the linking moiety also impacts the inhibitory effect. When the fragments from SL-A and SL-C were linked by the linker fragments enriched from selections (12, 24, and 10), they often showed lower IC50 values than those with the controls (1 and 4). Compound 693_12_828 displayed IC50 values of 10 µM and 15 µM against MMP-2 and MMP-9, respectively, tens-of-fold improvements compared to the two fragments. Compounds 182_12_828 and 182_24_828 also exhibited enhanced inhibitory effects compared to the starting fragments (Fig. 5F).
Since the fragment combinations 182 + 828 and 693 + 828 displayed higher potency than the other combinations, we applied molecular docking to shed light on their binding modes. The study suggested that the compounds bind to the catalytic domain, accommodating the hydrophobic subsite 1′ (S1′) of the substrate-binding cleft in both enzymes via fragment 182 or 693. Moreover, the 828 moiety displayed interaction with the catalytic Zn2+ in all compounds (Supporting information 5.3 and Figures S13-S17).