Considerations for achieving maximized DNA recovery in solid-phase DNA-encoded library synthesis

Abstract

DNA-encoded library (DEL) technology enables rapid, economical synthesis and exploration of novel chemical space. Reaction development for DEL synthesis has recently accelerated in pace with a specific emphasis on ensuring that the reaction does not compromise the integrity of the encoding DNA. However, the factors that contribute to a reaction’s “DNA compatibility” remain relatively unknown. We investigated several solid-phase reactions and encoding conditions and determined their impact on DNA compatibility. Conditions that minimized the accessibility of reactive groups on the DNA encoding tag (switching solvent, low temperature, double-stranded encoding tag) significantly improved compatibility. We showcased this approach in the multi-step synthesis of an acyldepsipeptide (ADEP1) fragment, which preserved 73% of DNA for a >100-fold improvement over canonical conditions. These results are particularly encouraging in the context of multi-step reaction sequences to access natural product-like scaffolds, and more broadly underscore the importance of reconciling the biophysical properties and reactivity of DNA with chemistry development to yield high-quality libraries of those scaffolds.

Graphical Abstract

Introduction

DNA-encoded library (DEL) technology^1–3 has gained popularity in the last decade as a tool for finding leads from large collections of novel chemical matter. DEL synthesis employs split-and-pool combinatorial chemistry to generate immense chemical diversity using only a few simple and robust chemical reactions. Recording the synthesis history of each library member in an attached DNA sequence has dramatically expanded the chemical space of combinatorial libraries, which now routinely yield attractive drug leads and sometimes even clinical candidates.^4–8 In the current state of the art, accessing new chemical space is a matter of devising reactions that preserve the integrity of the DNA.

Over the last 5 years especially, DNA-compatible reaction development has expanded the structural diversity of DELs. Despite the extreme constraints of DEL synthesis (reactions must occur in aqueous media and at low reactant concentration), a surprisingly wide range of reactions is amenable to on-DNA DEL synthesis⁹ and is steadily growing. Recent notable examples include [3 + 2] nitrone–olefin cycloaddition,¹⁰ imidazole synthesis,¹¹ C-C bond formation via Ni/photoredox dual catalysis^12,13 and even hydrogen atom transfer.¹⁴ DEL reaction development has continued at a rapid pace, but DNA solubility and reactivity constraints have left several classes of chemical transformations out of reach.

The constraints of conventional on-DNA reaction development can be circumvented by using solid-phase synthesis strategies. Our laboratory disclosed DNA-encoded solid-phase synthesis (DESPS),¹⁵ which integrated the enzymatic DNA-encoding advantages of DEL with the automation and reaction development advantages of solid-phase chemistry. Foremost among those was the ability to conduct chemical reactions directly in neat organic solvent. DESPS, however, is limited to building solid-phase DELs, which are not common. Similar results were achieved for on-DNA synthesis in emergent pseudo-solid-phase strategies, where DNA heteroconjugates are reversibly immobilized to modified ion exchange resin, dehydrated in nonpolar solvent, and then subjected to chemical reaction conditions.^16,17 These approaches have rendered highly water-sensitive reactions, such as the stannylamine protocol (SnAP) reaction, amenable to on-DNA synthesis.¹⁷ Other solid-phase-inspired reactions on controlled porous glass have shown that removing DNA from the aqueous milieu can unlock, for example, acid-catalyzed transformations among others.^18,19

Here, we describe efforts to develop the widely used and water-sensitive Steglich esterification,²⁰ which precipitated the discovery of solid-phase reaction conditions that dramatically and unexpectedly preserved the integrity of encoding DNA during synthesis. Esterification is a common prodrug strategy for increasing bioavailability and cell penetration²¹ and ester bonds occur frequently in natural products, ranging from small molecule metabolites to complex macrocycles.²² In pilot studies of the Steglich esterification en route to DNA-encoded macrocyclic natural products, we discovered that reactions conducted in anhydrous, nonpolar solvent at room temperature lead to near-quantitative recovery of DNA post synthesis. Applying these conditions to amide formation made feasible an 8-step DESPS reaction sequence to afford a DNA-encoded fragment of acyldepsipeptide 1 (ADEP1)²³ and, more importantly, deepened our understanding of how some parameters of solid-phase reactions confer synthesis resilience on DNA.

Results and Discussion

Previous work in our lab described the assessment of several common bond-forming reactions and protecting-group removal conditions for their compatibility with DNA.^15,24 Our preferred method for this type of analysis is reaction rehearsal, in which magnetic beads displaying DNA encoding tags are mixed with solid-phase synthesis resin during chemical transformations. Afterwards, the magnetic beads are recovered and the number of viable DNA tags remaining are determined via qPCR and compared to magnetic beads that did not experience reaction conditions. Molecular-biology-based approaches such as this and recent ligation-based study by Ratnayake and coworkers²⁵ now complement chromatography-based methods²⁶ as analytical tools to determine comprehensive and robust reaction conditions for DEL synthesis.

Previous DNA-compatibility studies in our lab did not investigate ester bond formation. Initial development of Steglich conditions on resin resulted in ~80% conversion of the on-bead hydroxyl starting material to ester product upon incubation of the resin with 67/67/7 mM carboxylic acid/DIC/4-dimethylaminopyridine (DMAP) in 9:1 dry DCM/DMA for 2 h at RT. Subsequently, the DNA compatibility assay for these reaction conditions revealed that 95% of the DNA was recovered. Using reaction conditions similar to those of the Steglich esterification, we were able to make amide bonds in high yield while losing only 5% of DNA recovery (Table 1, entry 1).

Table 1.

Reaction conditions for improved DNA compatibility.

This result is remarkable considering that the first step in this esterification mechanism – formation of the O-acylisourea activated ester – is the same as that used to form amide bonds, with the exception that the alcohol is a much poorer nucleophile than the primary amine. However, our previously-optimized conditions for amide bond formation resulted in a 50% loss of DNA recovery.²⁴ Upon first inspection, the major discrepancy between our new acylation conditions and those used previously is the solvent environment. DCM – present at 90% by volume – is significantly less polar than DMF. This affects the reaction environment and dictates the physical state of the highly-polar DNA encoding tag. While DMF is known to denature the helical DNA complex,²⁷ it is not empirically known if the duplex is also disrupted in nonpolar solvents. Preservation of the DNA duplex in a nonpolar solvent may attenuate DNA’s reactivity, and therefore make it less susceptible to chemical modifications. The nonpolar environment would also promote DNA adsorption onto the bead surface.

With this positive result in hand, we revisited many of the less DNA-compatible chemical transformations that we had previously examined in the context of DESPS. As a common companion to amide bond formation, Fmoc-deprotection can be accomplished using a solution of 20% piperidine in DCM without any significant loss in DNA recovery (Table 1, entry 2). Secondary amine acylation, which requires “harsher” reaction conditions due to it being a poorer nucleophile than the primary amine, can be performed quantitatively in DCM at RT. This reaction causes only a 15% loss in DNA recovery, which is a significant improvement over the 83% loss in DNA recovery obtained previously (Table 1, entry 7). Similarly, substituted anilines also have poor reactivity and acylation requires much stronger electrophiles that can be incompatible with DNA. In DCM, however, we discovered that the reaction of an aromatic amine with an acid chloride proceeds quantitatively with only a 5% loss of DNA recovery (Table 1, entry 9).

These results demonstrated that more attention should be given to DNA-encoded reaction development. Since even a modest operational change (e.g., switching solvents) can greatly increase the DNA compatibility of a reaction without sacrificing yield or purity, perhaps other subtle changes can have a similar impact. For instance, previous work in our lab demonstrated that acetylation of free amines could be accomplished using acetic anhydride in DMF, which preserved 70% of DNA recovery.²⁴ Upon closer inspection, we hypothesized that most of this damage is caused by the hydrolysis product of the anhydride, acetic acid. By utilizing dry DCM and adding base to the reaction cocktail, we were able to both reduce the amount of acetic anhydride employed (5% vs 20% v/v) and preserve 90% of DNA (Table 1, entry 5). Unlike the damage propagated by acylation reactions, the mechanism of DNA damage caused by H⁺ is well known²⁸ – protonation of the N7 nitrogen causes depurination and subsequent scission of the phosphodiester backbone upon pentose ring-opening caused by addition of water at C1’ – and can be mitigated in some circumstances. Instead of treating 4-methyltrityl-protected amines with 1% TFA in DCM, deprotection was facilitated by treatment with 75:22:3 HFIP/DCM/TIPS, which improved DNA compatibility from 50% to 90% DNA recovered (Table 1, entry 10). HFIP (pK_a = 9.3) is much less acidic than TFA (pK_a = 0.2), but still sufficient to remove the highly labile Mtt group. Similarly, palladium is also known to cause DNA strand scission via depurination of adenine residues.²⁹ Addition of washing steps with 100 mM aqueous DEDTC – a “soft” metal ion chelator – after Pd(PPh₃)₄-mediated allyl ether deprotection increased the percentage of DNA recovered from 11% to 99% (Table 1, entry 8).³⁰

Through the course of our experiments, we discovered that switching to a less polar solvent is not guaranteed to significantly increase DNA compatibility and other factors should be considered. For instance, nucleophilic substitution of an alkyl bromide with a primary amine to produce the secondary amine proceeded quantitatively in DCM with a 13% loss in DNA recovery, compared to a 20% loss for an alkyl chloride in DMF (Table 1, entry 4). This isn’t a direct comparison because – unlike DMF – DCM isn’t polar enough to solvate the chloride ion properly (no reaction). Using DCM facilitates a small gain in DNA compatibility, but starting reagents are limited to alkyl bromides and iodides. Additionally, we previously determined that silyl ether deprotection using 1 M TBAF in THF preserved only 3% of DNA, so a strategy using TEA·3HF³¹ was employed that yielded ~75% of the deprotected alcohol and preserved 70% of DNA. Upon reexamination of this deprotection step, we determined that 2.5 mM TBAF in THF (a 400-fold reduction) yielded quantitative deprotection of the TBDMS group with superior reaction kinetics (1 h vs. 16 h) and comparable DNA compatibility to TEA·3HF treatment (Table 1, entry 6).

Given that substituting DMF with DCM during acylation produced encouraging results, we further investigated the impact of solvent polarity on DNA compatibility. We performed the secondary amine acylation using a panel of six solvents: CPME, DCM, THF, DMF, NMP, and HFIP. This reaction was selected because previous conditions optimized for DMF were dramatically less DNA-compatible than conditions optimized for DCM. Using reaction rehearsal (Figure 1A), those latter conditions optimized for 90% DCM (Table 1, entry 7) were replicated with the other five solvents. We found that the DNA compatibility of each reaction varies widely across the solvent panel (black bars, Figure 1B) while the solvents themselves are benign (gray bars, Figure 1B). Less DNA was recovered from the reactions using polar solvents DMF and NMP (76% and 68%, respectively) than from the DCM reaction. Interestingly, CPME and THF are much closer in polarity to DCM but are the two least DNA-compatible solvents in the panel (53% and 39% DNA recovered, respectively). HFIP did not facilitate any conversion of starting material to product. Agnostic of reaction yield, the results obtained with this small solvent panel do not reveal any clear trend between solvent polarity and DNA compatibility.

Figure 1.

Effect of solvent identity on DNA damage for secondary amine coupling. (A) TentaGel beads (gray) displaying a secondary amine and magnetic beads (purple) displaying a dsDNA tag were subjected to acylation conditions using a panel of solvents. (B) The percentage of DNA tags recovered after acylation (black bars) varies significantly across the solvent panel, but not distinctly as a function of polarity, designated here by dipole moment (D) and dielectric constant (ε). Solvent alone is shown to have little effect on DNA compatibility (gray bars). Error bars represent the standard deviation of six technical replicates of either three (full reaction) or one (solvent only) biological replicate(s).

Solvent-mediated modifications to the state of the DNA encoding tag may only account for a small fraction of “DNA incompatibility.” The reaction conditions used in the DCM-optimal secondary amine acylation led to incomplete reactions when performed in more polar solvents (Figure S13). Driving these reactions to completion will require more “aggressive” conditions, which usually means more reactive reagents, increased [reagent], or increased reaction temperature. We investigated DNA integrity as a function of temperature during secondary amine acylation in both 9:1 DCM/DMA and 9:1 DMF/DMA. In the absence of reagents, double-stranded DNA (dsDNA) features excellent recovery (88–99%), regardless of solvent or incubation temperature (blue bars, Figure 2A). Upon addition of DIC and Oxyma, DNA recovery is preserved in both solvents at RT, but experiences a smaller drop in DCM (−14%) and a larger drop in DMF (−43%) when the temperature is increased to 55 ℃ (blue bars, Figure 2B). When the complete reaction is performed (blue bars, Figure 2C), DNA recovery is best preserved using DCM at RT (85%), followed by DMF at RT (76%), DCM at 55 ℃ (55%), and then DMF at 55 ℃ (1%).

Figure 2.

Effect of DNA strandedness and reaction temperature on DNA compatibility. The percentage of DNA tags recovered for beads displaying either dsDNA tags (blue bars, n = 6) or ssDNA tags (gray bars, n = 6) is determined after incubation at either RT or 55 ℃ in (A) solvent alone (90% DCM or DMF), (B) with only acylation reagents DIC and Oxyma, or (C) full secondary acylation conditions. Error bars represent the standard deviation of technical replicates.

Elevating temperature is a common strategy to increase reaction yield, but the results obtained here suggest that this should be done with extreme caution during the construction of DELs. Increasing the temperature by 30 ℃ had a much more pronounced effect on DNA compatibility than changing the polarity of the solvent. Moreover, these two effects are additive, as the reaction in 9:1 DMF/DMA at 55 ℃ led to the most severe loss in DNA recovery. A larger loss of DNA recovery occurred with the full complement of reagents rather than just the coupling reagents, which suggests that – for this reaction – intermediates contribute more to DNA damage than direct action of the starting materials. Interestingly, a water-soluble carbodiimide is known to form adducts with uracil, thymine, and guanine nucleobases^32,33 with a large preference for single-stranded nucleotides. If a similar reaction occurs with DIC in our system, it would explain why a loss of recovery is observed in 60/60 mM Oxyma/DIC only with elevated temperature. Subsequently, these carbodiimide adducts may serve as placeholders for essentially irreversible nucleobase modification during amidation.

Increasing the reaction temperature increases the rate of base-pair dissociation and disruption of base-stacking via base-flipping,³⁴ and ultimately induces duplex denaturation. To investigate the effect of strandedness on DNA integrity during chemical synthesis, we employed sensor beads displaying a single-stranded DNA oligonucleotide (ssDNA) to evaluate the same conditions that were used to probe DNA compatibility of dsDNA. As with dsDNA, ssDNA does not lose any significant recovery when incubated in solvent alone (89–94% DNA recovered, gray bars, Figure 2A). Furthermore, we observed that DIC and Oxyma produce a decrease in DNA recovery for DCM at 55 ℃ (−19%), DMF at RT (−30%), and DMF at 55 ℃ (−61%) whereas ssDNA templates in DCM at RT were not affected (gray bars, Figure 2B). All ssDNA templates followed the same trend, but faired poorer than their dsDNA counterparts in the complete reaction environment (gray bars, Figure 2C), since the reaction in DCM at RT was most DNA compatible (60%) and the reaction in DMF at 55 ℃ was least compatible (0.4%). These observations align with previous studies of post-synthesis DNA recovery in reactions on ss- and ds-DNA species; recovery from dsDNA samples uniformly outperformed recovery from ssDNA samples.³⁵

Our investigations that probed the effect of solvent, temperature, and DNA-strandedness all point toward DNA compatibility for a given reaction being dependent upon the accessible surface area³⁶ of the oligonucleotide. Although directly probing the structure of the DNA encoding tags during synthesis is not possible, we hypothesize that increasing the temperature of the reaction transitions the DNA from a duplexed state that is potentially adsorbed onto the bead surface to a better-solvated conformation with more single-stranded character. This increased exposure to the solvent provides more opportunities for reactive species to interact with the DNA nucleobases. For example, the amount of hydrogen abstraction at specific positions on deoxyribose in the presence of hydroxyl radicals scales with the solvent accessibility of that site,³⁷ and solvent accessibility is a well-known strategy for probing nucleic acid secondary structure.³⁸

If our hypothesis is correct, optimized reaction conditions that minimize the accessible surface area of the DNA tag can also expand the chemical space accessible to DEL by increasing the number of possible reaction cycles. We demonstrated this by synthesizing a fragment of the antimicrobial acyldepsipeptide ADEP1²³ using both the prior reaction conditions and our newly-optimized reaction conditions. Bifunctional 160-μm TentaGel beads displaying Fmoc-protected amines and full DNA-encoding tags were subjected to iterative rounds of Fmoc deprotection and amino acid coupling. Overall, eight chemical transformations were performed during the multi-step synthesis (Figure 3A): four Fmoc-deprotection steps, three primary amine acylations, and one secondary amine acylation. The cumulative percent DNA recovered for the prior reaction conditions was 0.4% (red, Figure 3B), which is very close to the theoretical value of (0.6)⁴ × (0.51)³ × 0.17 = 0.3%. When these same reactions were performed using the improved conditions detailed in Table 1, 73% of the DNA tags were recovered (blue, Figure 3B), which agrees exactly with the theoretical value of (1)⁴ × (0.95)³ × 0.85 = 73%. This increase in DNA compatibility has significant ramifications for solid-phase DELs since a single 10-μm TentaGel bead may not be decodable after sustaining a 250-fold decrease in viable DNA, leading to increased false negative rates. This may become particularly problematic for larger DELs where library member representation is lower.³⁹

Figure 3.

Comparison of new conditions to prior conditions during the multi-step synthesis of an ADEP1 fragment. (A) 160-μm TentaGel beads displaying both an Fmoc-protected amine and a dsDNA tag were subjected to the following processes in order: Fmoc deprotection, primary amine acylation, Fmoc deprotection, secondary amine acylation, Fmoc deprotection, primary amine acylation, Fmoc deprotection, and primary amine acylation. These reactions were performed using both the conditions described herein or previously-published conditions, and compared to beads that were excluded from reaction conditions (buffer control). A TFA control was used demonstrate damage. (B) After all the reaction steps were completed, the number of tags recovered per bead was determined in single-bead qPCR assays.

In pursuit of conditions for performing the water-sensitive Steglich esterification, we serendipitously discovered generalized acylation reaction conditions that preserved DNA in near quantitative yield. These conditions were uniquely the product of conducting synthesis on solid supports, which enabled the use of nonpolar solvents, coupled with a systematic investigation of other more obvious parameters, such as temperature and DNA strandedness. Optimization of conditions suggested that a combination of parameters favoring duplex formation and thereby protecting reactive moieties on the DNA yielded a general framework for understanding the mechanism by which the conditions were more protective of the encoding DNA’s fidelity.

DNA compatibility is an emergent concept in the field of DEL reaction development, but one that has received increased attention as reactions that invoke conditions and reagents with unknown reactivity toward DNA are implemented.^14,40,41 Initial investigations only focused on the constraints that DNA solubility placed on the selection of solvent and limitations in reaction kinetics with respect to conversion yield.⁹ Subsequently, we began quantitative explorations of a given reaction’s effect on the encoding DNA’s PCR amplifiability; DNA-functionalized “sensor” beads were subjected to various reaction conditions and then analyzed by qPCR to determine the recovery.²⁴ The sensor approach was further adapted to probe on-DNA synthesis conditions with the use of a modified headpiece that could be used to analyze yield by LC-MS, and both ligation yield and qPCR recovery.²⁵ These approaches were critical in establishing the dramatic improvement in compatibility for the conditions of this study. Analysis of the pseudo-solid-phase SnAP reaction (in 4:1 DCM/HFIP) showed that their reaction was also essentially innocent to DNA.¹⁷ Although it would not be possible to conduct the reaction without the anhydrous conditions afforded by DNA adsorption, it would be interesting to see if other pseudo-solid-phase reactions experience similar compatibility enhancement.^16,18

Conclusions

In conclusion, we hypothesize that DEL construction should seek to minimize the accessible surface area of the encoding DNA tag during synthesis steps. Changing the reaction solvent to induce DNA adsorption is one possible approach. It might also be unnecessary to replace one solvent completely with another; mixtures of miscible solvents could balance reagent/intermediate solubility, DNA solubility, and reaction kinetics. Other routes to decrease accessible surface area include the use of double-stranded encoding tags, which mitigates nucleobase amine reactivity via hydrogen bonding and sterics. Maintaining a low reaction temperature produces a similar effect by favoring duplex secondary structure and reducing the base-flipping rate. Overall, a balance between reaction conditions that optimize both chemical yield and DNA compatibility is best determined by the practitioner. We balanced yield and DNA compatibility for the esterification reaction by choosing conditions that prioritized quantitative synthesis yield (Table 1, entry 3).

These studies demonstrate the possibility of performing reactions that cannot well tolerate the polar, aqueous environments required for “traditional” on-DNA DEL synthesis by thoughtfully modifying the reaction format. Such modifications could include screening alternate solvents, temperatures, and coupling reagents, or adopting solid-phase synthesis technologies. Conventional solid-phase synthesis strategies have the added benefit of enabling activity-based DEL screening capabilities. For instance, DESPS produces beads that each display myriad copies of a single DEL member on a photocleavable linker separate from the encoding tag. The polyvalent DEL bead products of DESPS can be compartmentalized in picoliter-scale droplets of either enzymatic activity assay reagent^42–45 or competition binding assay reagent,⁴⁶ reconstituting high-throughput screening-type functional assay capabilities.

Adopting these revised bead-based approaches to optimize DNA compatibility will require retooling DEL synthesis automation infrastructure to address bead handling, solvent compatibility, and aqueous/organic phase transitions. Further investment in automation seems advantageous, since bead-facilitated chemistry generally appears to unlock reactions that are otherwise inaccessible in the current on-DNA synthesis paradigm. Established reaction conditions may initially prove too destructive to DNA, but our studies show that undertaking mechanistic studies of DNA damage to inform synthesis technology development can greatly improve reaction compatibility with DNA.