Multiplexed Small Molecule Ligand Binding Assays by Affinity Labeling and DNA Sequence Analysis

Abstract

Small molecule binding assays to target proteins are a core component of drug discovery and development. While a number of assay formats are available, significant drawbacks still remain in cost, sensitivity, and throughput. To improve assays by capitalizing on the power of DNA sequence analysis, we have developed an assay method that combines DNA encoding with split-and-pool sample handling. The approach involves affinity labeling of DNA-linked ligands to a protein target. Critically, the labeling event assesses ligand binding and enables subsequent pooling of several samples. Application of a purifying selection on the pool for protein-labeled DNAs allows detection of ligand binding by quantification of DNA barcodes. We demonstrate the approach in both ligand displacement and direct binding formats and demonstrate its utility in determination of relative ligand affinity, profiling ligand specificity, and high-throughput small molecule screening.

Graphical Abstract

The affinity labeling of DNA-barcoded ligands to target proteins allows ligand binding assays using DNA as a label. Pooled sample manipulation and DNA sequence detection enabled by DNA encoding yields quantitative binding assays amenable to both ligand displacement and direct binding formats.

The identification and characterization of small molecule ligands to proteins are central to the development of drugs and chemical probes.^1–4 While a variety of assay approaches are available for detecting protein-ligand interactions,^5–7 significant limitations exist, particularly in the areas of throughput, cost, and sensitivity.^8,9 Labeled-ligand binding assays are commonplace and have a particularly long history in the quantitative determination of ligand receptor affinity. Typical labels include radioactive atoms,¹⁰ fluorophores for detection by fluorescence polarization (FP),¹¹ fluorescence energy transfer (FRET),¹² bioluminescence energy transfer (BRET),¹³ or solid phase beads (Alphascreen, e.g.).¹⁴ The use of DNA as a quantitative label in such assays has not been explored and stands to benefit from the several advantages of DNA detection (Figure 1).

Figure 1.

Use of DNA as a detection label in ligand binding assays offers advantages over conventional radioactive and fluorophore labels.

The in vitro selection of DNA-encoded chemical libraries (DELs), as well as phage and mRNA display libraries, can be considered as a parallel ligand binding assay that uses DNA tags as detection labels. While there are a few examples of effective affinity ranking of ligands using selection approaches,¹⁵ enrichment values from selections often correlate modestly to affinity. A number of complicating factors, such as the ratio of the ligand affinity constant to target protein concentration, ligand purity, and sequencing depth can affect the enrichment-affinity relationship.¹⁶ As these approaches are most often used for discovery, a strong correlation is typically not required.

DEL-based ligand discovery is a dramatically more accessible and lower cost alternative to high throughput screening (HTS) and illustrates the many benefits of DNA detection.¹⁷ DNA sequence analysis as a detection platform allows for remarkable sensitivity, low sample requirements, and use of widely available instrumentation. Using DNA as a label readily allows for multiplexing by using unique labels as barcodes. Combining DNA encoding with split-and-pool approaches not only facilitates the synthesis of a large number of unique molecules in a DEL but also allows for a collective assay of molecules by in vitro selection. The ability to assay compounds within a pool, rather than individually (as in traditional HTS), is a significant benefit that dramatically decreases assay complexity, variability, and cost.¹⁸

We recently reported a DNA-based assay approach, which we call selection-based sensing, that uses DNA-linked molecules as sample probes and also exploits split-and-pool techniques with DNA-encoding. We have used this method for assaying enzymatic activity with DNA-linked substrates or active site-reactive probes.¹⁹ In this report, we extend the use of selection-based sensing as an assay platform with a DNA-based ligand binding assay for the characterization of protein-ligand interactions. We implement the approach in both a ligand displacement assay (Figure 3 and Figure S1a) suitable for quantitative measurement of free ligands and high throughput screening of traditional small molecule libraries and also in a direct binding format to determine relative affinity of DNA-linked ligands (Figure 6 and Figure S1b) through protein titration. Both applications rely upon use of DNA-encoding to record sample history and covalent crosslinking of a DNA-linked ligand to a target protein using a dual display format, which has previously shown to be effective for protein affinity labeling with DNA-linked ligands (Figure 2).²⁰

Figure 3.

A DNA-based ligand displacement assay for quantifying affinity of non-labelled ligands by DNA sequence analysis. a) Schematic illustration of multiplexed ligand binding assays by displacement of affinity labelling of a DNA-linked probe ligand with free ligands and subsequent DNA sequence analysis. The assay was applied with a broad specificity ligand for the Polycomb (Pc) CBX Chromodomains (ChDs)³⁴ and 3 competing ligands (see Figure S2 for structures) to the CBX7-ChD. b) Concurrent determination of IC₅₀ values for ligand 1 to five protein targets within crude E. coli cell lysates. Labeling HaloTag (HT) fusion proteins with 20-mer oligos enabled orthogonal purification using complementary oligo beads. In all panels, error bars indicate one standard deviation of the signal mean for three unique DNA constructs at each free ligand concentration.

Figure 6.

DNA-based direct binding assay for the determination of relative binding affinity of DNA-linked CBX7-ChD ligands. a) Schematic illustration of multiplexed ligand binding assays by affinity labeling of DNA-linked ligands and subsequent DNA sequence analysis. Application of the assay with ligands 4–7 to the CBX7-ChD. Red boxes indicate structural changes from ligand 4. (b) Measurement of the EC₅₀ values of pure and “contaminated” compound 7 to CBX7-ChD. (c) Direct binding assays to determine relative affinity of 96 DNA-linked ligands concurrently were conducted using photo-crosslinking, and binding curves of representative 7 compounds are shown. See Figs. S24–25 for additional curves. d) Correlation plot between off-DNA K_d of purified compounds and on-DNA EC₅₀ values of crude library members determined by photocrosslinking. In panels a and b, error bars indicate one standard deviation of the signal mean for three unique DNA constructs at each protein concentration. In panel d, error bars indicate one standard deviation of the signal mean for two off-DNA K_d values

Figure 2.

Selection-based sensing approach to ligand binding assays by covalent crosslinking of DNA-linked ligand to target proteins.

Our approach to a ligand displacement assay to quantify the affinity of unlabeled ligands by selection-based sensing involves incubation of a protein target with a known ligand appended to DNA along with an electrophilic crosslinker, which serves as a binding probe (Figures 2, S1a). Co-incubation with a competing free ligand (not linked to DNA) will displace the DNA-linked probe, resulting in decreased affinity labeling and lower DNA recovery in a subsequent purification of DNA’s linked to the protein. As an initial demonstration of the technique, we determined half maximal displacement (IC₅₀) values for three compounds of known affinity to the Chromobox Homolog 7 Chromodomain (CBX7-ChD). DNA-linked probes were prepared by conjugating a peptidic ligand (ligand 1 (BrBA) (Figure S2), K_d ≈ 26 nM)²¹ to different encoding DNAs. In this application, we used the DNA barcodes within the probes to encode the sample identity. Each sample contains a particular competing free ligand at various concentrations in a titration series (Figures 3a, S1a). Following incubation, excess free ligand (1000x relative to the DNA-linked probe) was added to suppress additional crosslinking, and components from all samples were pooled. Then, the protein-labeled probes were purified via a HaloTag (HT)²² on the target protein using chloroalkane beads. The recovery of each barcoded probe determined by next generation DNA sequencing of the pool. The recovery together with the decoded ligand identity and concentration information yields IC₅₀ curves for each of the 3 ligands (Figure 3a). To verify these results, a fluorescence polarization (FP) assay was performed using identical concentrations of both protein and a fluorescein-labeled probe ligand, which provided consistent IC₅₀ values (Figure S3).

In drug discovery and development, biochemical assays are generally considered low content in terms of the information provided.²³ The ability to assay compounds concurrently against a pool of multiple protein targets would improve this significantly by providing information-rich biochemical profiles while keeping costs low. Most assay platforms are limited to a single protein target, however. Such multiplexing offers the potential to reduce costs of HTS by splitting over multiple targets and can also allow up front identification of selectivity by concurrent screening of similar proteins. This approach does allow for multiple proteins to be assayed concurrently (provided they contain orthogonal purification tags). In addition, it is adequately robust and specific to allow assays within crude protein mixtures. To demonstrate these features, we determined the IC₅₀ of free ligand 1 (Figure S2) to 5 protein targets labeled with 20-mer oligonucleotides as orthogonal purification tags within E. coli cell lysates (Figure 3b and Figure S4a). Crude lysates containing the labeled proteins were pooled and then split for displacement assays, as in Figure 3a. After incubation with free ligand, probes crosslinked to target proteins were selected by purification using complementary oligonucleotides on magnetic beads. The relative recoveries of the DNA probes were determined by qPCR and gave IC₅₀ values of the ligand to each of these five targets, which are consistent with reported values.²⁴

To evaluate robustness for high throughput applications, we prepared 96 probes composed of the DNA-linked ligand 1 with unique barcodes. Half of the probes were incubated with protein and crosslinker, while the other half were additionally incubated with excess free ligand. Enrichment of the probes from the two groups were well distinguished (Z’ factor = 0.77, Figure S6). This indicates little variation in DNA barcode detection and a robustness suitable for HTS. We then developed a binding assay for E. coli dihydrofolate reductase (eDHFR) using a DNA-linked trimethoprim (TMP) probe, a commonly used model ligand receptor pair.²⁵ This assay gave a similarly robust Z’ factor (0.54, Figure 4a).

Figure 4.

Application of DNA-based ligand displacement assay for screening. a) Assessment of assay robustness with a trimethoprim (TMP) probe and E. coli DHFR (eDHFR) in a 96-well plate. b) Screening of the LOPAC library against eDHFR. Samples exhibiting read numbers 1 or less (40 of 1280) were excluded as failed wells due to low sampling. c) Multiplexed screening of 1000 compounds against 5 bromodomain proteins.

Next, we screened a small molecule library of 1280 compounds (LOPAC®1280, Sigma-Aldrich) against HaloTag-eDHFR (Figure 4b). The samples containing the three known eDHFR inhibitors in the library, as well as a novel compound, GW1929²⁶ (Figure S7), showed low recovery of their DNA-linked probes. Inhibition assays of eDHFR with GW1929 and TMP validated the results of the screen (Figures S8–S10, K_i of GW1929 = 410 nM, K_i of TMP = 15 nM). Similarly, GW1929 was able to displace a fluorescein-labeled methotrexate (MTX) from eDHFR in an FP assay (Figure S11, IC₅₀ = 7.9 μM). In addition, GW1929 showed modest inhibition of human DHFR (Figure S9b), which may have implications for its use as a chemical probe for activation of peroxisome proliferator-activated receptor γ.²⁷

We then sought to test this approach in a concurrent HTS against multiple protein targets using multiple DNA probes. We selected five bromodomains as targets: Polybromo 1-bromodomain 2 (PB1-BD2), PB1-BD4, PB1-BD5, BRD7 and BRD9, which were expressed and purified as HaloTag fusions (Figure S5). We prepared two DNA-linked binding probes using the previously reported inhibitors MW99²⁸ (modest affinity (low μM K_d’s) to PB1-BDs 2/4/5) and bromosporine²⁹ (high affinity (mid nM K_d’s) to BRD7/9) (Scheme S11). Initially, we optimized crosslinking efficiency of the DNA-linked ligands to each of the five proteins using various reactive groups and settled on the aryltrifluorodiazirine photocrosslinker (Figure S12). As a first pass test, we performed a concurrent assay with the protein pool and the two binding probes and demonstrated significantly decreased recovery of the probes with addition of competing free inhibitors (Figure S13). A subsequent screen of a collection of 33 commercially available bromodomain inhibitors against the protein pool showed binding profiles consistent with reported specificities and identified some previously uncharacterized ligands (Figure S14). We validated these results for BRD9 with an FP assay (Figure S15) and with individual DNA-based displacement assays for BRD7 and PB1-BDs 2/4/5 (Figure S16).

We then performed concurrent screening of 1000 compounds from the TargetMol^® Bioactive Compound Library against the five bromodomain proteins (Figure 4C). All compounds in the collection annotated as inhibitors to these bromodomains were identified as hits, as were several novel compounds (see Figure S17 and Table S6–S10 for the detailed screening results and hit compound structures). Representative screening results for BRD9 and PB1-BD2 are shown in Figure 4C. For BRD7/9, several previously reported kinase inhibitors (e.g. selonsertib³⁰ and XMU-MP-1³¹) significantly displaced the DNA-bromosporine probe. As anticipated, the concurrent screen directly indicated many selective hit compounds. XMU-MP-1, a protein kinase inhibitor, demonstrated selectivity towards BRD7/9 over PB1-BD2/4/5; while UNC-0379,³² a methyl transferase inhibitor, showed selectivity for PB1-BD2/4/5 over BRD7/9 (Figure 4c, 5a, 5b). A HIV reverse transcriptase inhibitor, dapivirine,³³ showed selectivity for PB1-BD4 (Figure 5c). Many of the novel inhibitor compounds contained diaminopyrimidine or similar moieties, which have been observed in many dual kinase-bromodomain inhibitors.³⁵

Figure 5.

Select screening results for representative hits and hit validation. (a-c) Hit compounds XMU-MP-1 (a), UNC0379 (b), and dapivirine (c) showed binding selectivity in the primary screen, which was validated by IC₅₀ determination in subsequent assays (Figures S19, S20, S21, S22). (d) Validation of hit compounds to BRD9 by ligand displacement assay (LDA). (e) Validation of hit compounds to PB1-BD4 by LDA. Photocrosslinking-based LDA was performed against bromodomain proteins in the presence of DMSO (negative control) or 20 μM inhibitor. The recovery of DNA-bromoprobe was quantified by qPCR.

We validated the screening results by performing selections against each protein individually in the presence or absence of hit compounds (Figure 5d, 5e, S18). Thermal shift assays (TSA) and FP assays further validated hits for BRD7 and BRD9 (Figure S19, S20). For PB1-BD2/4/5, TSA or FP assays were unsuccessful. Therefore, we determined the IC₅₀ values of hit compounds using the DNA-based ligand displacement assay (as shown in Figure 3a). Both UNC-0379 and dapivirine demonstrated potent binding to their respective bromodomains with selectivity that matched that of the primary screen (Figure S21 and S22). Future studies with orthogonal assays will be conducted to further confirm and explore the selectivity profile of these hits.

This technique can also be used in a direct binding, rather than competitive, format to determine relative affinity of numerous DNA-linked ligands concurrently (Figures 6a, S1b). In this approach, a ligand to be tested is conjugated to DNA in a way that allows tethering of a crosslinking moiety on the opposite strand, as with the probe ligand for the displacement assay. The barcodes within the DNA-linked molecules serve to encode the identity of the ligand and also the protein concentration of a given sample within a titration series. After crosslinking, samples are pooled, and DNAs linked to protein are purified. DNA sequencing then determines the relative crosslinking yield of each DNA to the target protein, which is dependent upon the fraction of the DNA-ligand bound to the protein. As an initial test, four compounds were prepared on encoding DNA scaffolds (Figure 6a) with a range of affinities to CBX7-ChD. After incubation with various protein concentrations, further crosslinking was stopped by addition of excess ligand, all samples were pooled, and the ligands linked to the target protein were purified. DNA sequencing of the pool gave enrichment curves, and the four ligands showed differential affinity to CBX7-ChD as expected.²¹

This approach enables the affinity of DNA-linked ligand to be measured as crude compounds (provided contaminants are not ligands), which is a key benefit to improve throughput. Most ligand binding assays are conducted with the small molecule ligand in molar excess over the protein target. Thus, a high purity of a tested ligand is required, as this purity would affect the perceived potency. Because this approach uses the protein in excess, EC₅₀ values are not dependent on the concentration of the DNA-linked ligand. To demonstrate this, we compared the values obtained for the DNA-conjugated compound 7 that was intentionally “contaminated” with a non-ligand DNA. As expected, the recovery yield of impure compound 7 was lower than the pure compound as assessed by qPCR, yet the EC₅₀ values were within error (Figure 6b, 700 ± 200 nM vs 920 ± 90 nM).

To further evaluate assay performance in a complex mixture, we used a collection of 96 DNA-encoded crude compounds containing single monomer substitutions of the BrBA ligand 1 (Table S11–S14). In this case, DNA barcodes were appended to the compounds to indicate both the compound identity and the concentration of HaloTag-CBX7-Chd (0.2 nM – 50 μM) used in the crosslinking reaction. Using this approach, 53 of the 96 compounds gave curves sufficient to determine EC₅₀’s (Figure S24, Table S4). The remaining compounds gave incomplete curves and are presumed to have low affinity. Because the labeling rate using the sulfonyl fluoride crosslinker is slow relative to ligand binding (t_1/2 ≈ 1 hour, Figure S23), dose response curves do not represent true binding curves. Thus, in these tests, we have expressed affinities as effective protein concentrations that give half-maximal enrichment (EC₅₀ values). The use of fast-reacting photocrosslinking groups, as previously demonstrated,^20c are expected to give EC₅₀ values equal to dissociation constants. Therefore, we repeated these binding assays using a fast-reacting photocrosslinking group, aryltrifluorodiazirine^20d, to determine the EC₅₀ values of the 96 compounds (Figure S25, Table S5). 7 exemplary curves are shown in Figure 6c. To verify the values obtained, we synthesized 12 compounds off-DNA and tested their K_d values in a fluorescent displacement assay (Figure S26). A good correlation (R² = 0.84, Figure 6d) was observed in the comparison between off-DNA K_d values of the pure compounds and EC₅₀ values of the crude on-DNA compounds for 12 selected compounds among the collection of the 96.

In summary, the affinity labeling of DNA-linked ligands enables high-level multiplexing capability in labeled-ligand binding assays. These assays capitalize on DNA encoding to record the sample history of a DNA-linked molecule to enable pooled manipulation and downstream analysis. This approach allows concurrent screening of small molecules against multiple protein targets and rank ordering of the affinity of DNA-linked ligand libraries. As DNA sequence analysis continues to increase in accessibility and decrease in cost, we expect this method to accelerate numerous aspects of the small molecule ligand discovery and development process.