Cell-Based Selection Expands the Utility of DNA-Encoded Small-Molecule Library Technology to Cell Surface Drug Targets: Identification of Novel Antagonists of the NK3 Tachykinin Receptor

Article abstract of DOI:10.1021/acscombsci.5b00124

DNA-encoded small-molecule library technology has recently emerged as a new paradigm for identifying ligands against drug targets. To date, this technology has been used with soluble protein targets that are produced and used in a purified state. Here, we describe a cell-based method for identifying small-molecule ligands from DNA-encoded libraries against integral membrane protein targets. We use this method to identify novel, potent, and specific inhibitors of NK3, a member of the tachykinin family of G-protein coupled receptors (GPCRs). The method is simple and broadly applicable to other GPCRs and integral membrane proteins. We have extended the application of DNA-encoded library technology to membrane-associated targets and demonstrate the feasibility of selecting DNA-tagged, small-molecule ligands from complex combinatorial libraries against targets in a heterogeneous milieu, such as the surface of a cell.

Full text of DOI:10.1021/acscombsci.5b00124

Subscriber access provided by UNIV OF LETHBRIDGE
Article
Cell-based selection expands the utility of DNA-encoded small-
molecule library technology to cell surface drug targets:
Identification of novel antagonists of the NK3 tachykinin receptor
Zining Wu, Todd L. Graybill, Xin X Zeng, Michael Platchek, Jean Zhang, Vera Q Bodmer, David D
Wisnoski, Jianghe Deng, Frank T. Coppo, Gang X Yao, Alex Tamburino, Genaro S Scavello, G Joseph
Franklin, Sibongile Mataruse, Katie L Bedard, Jeffrey A Messer, Andrew J Pope, and David I Israel
ACS Comb. Sci., Just Accepted Manuscript • DOI: 10.1021/acscombsci.5b00124 • Publication Date (Web): 12 Nov 2015
Downloaded from http://pubs.acs.org on November 17, 2015
Just Accepted
“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
ACS Combinatorial Science is published by the American Chemical Society. 1155
Sixteenth Street N.W., Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.

Page 1 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
9
Cell-based selection expands the utility of DNA-encoded small-
molecule library technology to cell surface drug targets: Identification
of novel antagonists of the NK3 tachykinin receptor
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Zining Wu1, Todd L Graybill1, Xin Zeng1, Michael Platchek1, Jean Zhang2, Vera Q.
Bodmer1, David D. Wisnoski1, Jianghe Deng1, Frank T. Coppo1, Gang Yao2, Alex
Tamburino2, Genaro Scavello1, G. Joseph Franklin2, Sibongile Mataruse2, Katie L.
Bedard2, Jeffrey A Messer2, Andrew J Pope1 and David I Israel2
1Molecular Discovery Research, GlaxoSmithKline, Collegeville, Pennsylvania;
2Molecular Discovery Research, GlaxoSmithKline, Waltham, Massachusetts;
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 2 of 36
1
2
3
4
Keywords
5
6
7
8
9
DNA encoded chemical library
Affinity selection
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Drug discovery
Combinatorial chemistry
Gꢀprotein coupled receptor
Cellꢀassociated target protein
Abstract
DNAꢀencoded smallꢀmolecule library technology has recently emerged as a new
paradigm for identifying ligands against drug targets. To date, this technology has been
used with soluble protein targets that are produced and used in a purified state. Here,
we describe a cellꢀbased method for identifying smallꢀmolecule ligands from DNAꢀ
encoded libraries against integral membrane protein targets. We use this method to
identify novel, potent, and specific inhibitors of NK3, a member of the tachykinin family
of Gꢀprotein coupled receptors (GPCRs). The method is simple and broadly applicable
to other GPCRs and integral membrane proteins. We have extended the application of
DNAꢀencoded library technology to membraneꢀassociated targets and demonstrate the
feasibility of selecting DNAꢀtagged, smallꢀmolecule ligands from complex combinatorial
libraries against targets in a heterogeneous milieu such as the surface of a cell.
ACS Paragon Plus Environment

Page 3 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
Introduction
8
9
With the recent advent and description of DNAꢀencoded smallꢀmolecule library
technology (ELT) (1ꢀ5), library collections containing billions of unique compounds have
been used to identify potent and selective ligands against biological targets. These
libraries are synthesized using combinatorial methods and exist as mixtures containing
millions to billions of different compounds per library. Ligands that bind to a target
protein are identified using affinity based selection methods from the mixture of
compounds that comprise the library, and the chemical structures of selected
compounds are deduced from the sequence of the attached DNA (1). While massively
parallel DNA sequence analysis generates a large data set of selected molecules, only
a small number of the more enriched and chemically interesting smallꢀmolecule
structures are synthesized discretely and tested for activity against the target. Rapid
screening of mixtures (selection) coupled with focused synthesis greatly reduces the
time and cost of interrogating the DNAꢀencoded library collection compared to many
conventional screening methods.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Previously, encoded smallꢀmolecule libraries have been used with soluble,
tagged forms of target proteins used at high concentrations and in a purified state (1ꢀ8).
Many drug targets, however, are integral membrane proteins that reside on the cell
surface, such as Gꢀprotein coupled receptors (GPCR) and ion channels. The
conventional means to discover ligands against these types of drug targets is high
throughput screening (HTS), where at most several million compounds can be
interrogated for activity against a target. Adapting ELT methods developed for use with
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 4 of 36
1
2
3
4
5
6
7
8
9
soluble protein targets to integral membrane protein targets has been extremely
challenging. Detergent solubilization of integral membrane proteins often results in
denaturation and/or aggregation of the protein, making their purification in a native and
functional conformation nearly impossible. Our attempts using detergent solubilized
GPCRs as ELT targets have all yielded negative results. Furthermore, it is difficult to
express integral membrane proteins at a sufficiently high level and in a native state that
they comprise the majority of protein in a cell membrane. Svensen et al. have described
selections of a PNAꢀencoded peptide library against cells overexpressing αvβ5 and αvβ3
integrins (9) and the chemokine receptor CCR6 (10). This library, however, contained
only 10,000 different peptidic compounds and is unlikely to generate starting points for
small molecule drug discovery. Several examples of ligand discovery from small
molecule DEL libraries against soluble domains of cell surface proteins have also been
reported (11, 12). This strategy might be successful for a subset of membrane proteins,
particularly those with large extracellular domains and a single transmembrane domain.
However, it is unlikely that integral membrane proteins containing multiple
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
transmembrane domains can be successfully interrogated using this strategy.
Here, we describe a cellꢀbased selection method for identifying high affinity and
selective ligands against cell surface targets from smallꢀmolecule DNAꢀEncoded
Libraries (DELs). We used this method to interrogate a library collection containing
approximately 15 billion compounds and identified multiple chemotypes of antagonists
with high affinities and specificity to the NK3 receptor, a member of the tachykinin family
of GPCR. We provide detailed pharmacological characterization of several novel
chemotypes that are potent and specific antagonists of the NK3 receptor.
ACS Paragon Plus Environment

Page 5 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
Results
8
9
Summary of the method. ELT is a robust methodology for identification of biologically
active low molecular weight compounds that bind to purified biochemical targets (1ꢀ8).
One objective of our continuing development of the technology has been to apply it to
membraneꢀassociated targets. We modified the previously described methods for
soluble protein targets to integral membrane targets expressed on the cell surface.
Briefly, after evaluation of various host cell/vector expression systems and selection
protocols, we developed an optimized protocol that uses HEK293ꢀderived cells
transduced with recombinant BacMam viruses encoding the targets of interest. Intact,
viable cells were used in several rounds of selections to purify molecules in the DELs
that bind to the cell surface. Cells transduced with targetꢀnegative control viruses were
used as nonꢀtarget controls to identify DEL molecules that bind to endogenous cell
surface molecules. Informatics tools were used to subtract cellꢀbinding molecules
present in the targetꢀnegative sample from the molecules obtained against cells
expressing the recombinant GPCR target, revealing a population of targetꢀspecific
ligands. Compounds lacking the encoding DNA tags were then synthesized and tested
for biological activities in biochemical or cellꢀbased functional assays. This method is
rapid, simple and broadly applicable to various cell surface proteins.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Methods development using a DNA-tagged tool compound. We chose the
tachykinin receptor neurokininꢀ3 (NK3) to develop and optimize a cellꢀbased ELT
selection method for integral membrane proteins because of its relevance as a
therapeutic target, and the availability of smallꢀmolecule tool compounds, expression
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 6 of 36
1
2
3
4
5
6
7
8
9
vectors, and quantitative binding assays. The tachykinin receptor family and their
natural neuropeptide ligands substance P (SP), neurokinin A (NKA) and neurokinin B
(NKB), are described in detail elsewhere (13ꢀ16). A number of nonꢀpeptide tachykinin
receptor antagonists have been identified as research tools and clinical candidates (17ꢀ
20).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
We first transduced HEK293 cells with a recombinant BacMam virus encoding
NK3 at various virus to cells ratios. Receptor expression levels on the cell surface were
measured with a filter binding assay using 125IꢀNKB. Cells expressing approximately
500,000 receptors per cell, the highest level achieved with this expression system for
NK3, were used in protocol optimization and selection experiments. HEK293 cells
transduced with a NK3ꢀnegative BacMan virus were used as nonꢀtarget control (NTC) in
parallel selections.
Using HEK293 cells expressing the maximum obtainable levels of NK3, we
optimized and validated a general cellꢀbased ELT selection method with a tool
compound synthesized by conjugating a known NK3 antagonist to DNA. SB-235375-
DNA (Fig. 1B) is a modified, DNAꢀtagged derivative of a quinoline NK3 antagonist
previously used in preꢀclinical studies of disease modulation by NK3 antagonists (21).
SAR for this antagonist series (22) suggested the carboxylic acid group is a rational
point of attachment to the starting DNA molecule of ELT libraries (1). While SB-235375-
DNA had slightly lower affinity to NK3 than the starting compound (Ki of 20 nM in
competition binding assay and compared to 3 nM of unmodified SB-235375), it
nevertheless remained a high affinity NK3 ligand. The DNA portion of SB-235375-DNA
was further modified by ligation with a fluorescenceꢀlabeled oligonucleotide or
ACS Paragon Plus Environment

Page 7 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
9
alternatively, with a doubleꢀstrand DNA (dsDNA) fragment similar in size to the DNA
tags used in DEL library molecules. These DNAꢀtagged tool compounds provided
sensitive reagents for use in development and optimization of cellꢀbased ELT selection
methods. The resulting SB-235375-DNA-Fluor (Fig. 1C) and SB-235375-DNA-Seq
(Fig. 1D) were used in fluorescence binding assays or as a sequencing probe in
selection experiments, respectively. SB-235375-DNA-Fluor was added to 1x107 cells
expressing NK3 or NTC, and cellꢀbased selections were performed as described in the
Methods section. Fluorescence intensities were measured on samples after exhaustive
washing. Typically, about 10ꢀ15 fold higher fluorescence intensities were obtained from
cells expressing NK3 than that with targetꢀnegative control cells, suggesting an
enrichment of compound SB-235375-DNA-Fluor that is NK3 specific.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 8 of 36
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 1. Positive target specific enrichment of DNAꢀtagged tool compounds with a
cellꢀbased ELT selection protocol. (A) a quinoline NK3 antagonist. (B) a modified, DNAꢀ
tagged derivative of SBꢀ235375. The DNA portion of SB-235375-DNA was further
modified by ligation with a fluorescenceꢀlabeled oligonucleotide (C) or with a doubleꢀ
strand DNA (dsDNA) fragment similar in size to the DNA tags used in DEL library
molecules (D). SB-235375-DNA-Seq was spiked into a DEL library at different ratios
and subjected to three rounds of selection. The recovered library DNA molecules were
PCR amplified and sequenced to decode the dsDNA tags. The enrichment factors for
ACS Paragon Plus Environment

Page 9 of 36
ACS Combinatorial Science
1
2
3
4
SB-235375-DNA-Seq, i.e. fold increase of copy counts relative to all DEL molecules in
5
6
the sample, was calculated and summarized in the table (E).
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
SB-235375-DNA-Seq was next spiked into a DEL library at different ratios
(chosen to mimic the concentration of an individual member in DELs), and subjected to
three rounds of selection. Typically, a total of 1.5x1015 DEL molecules (2 nMoles of
library) were used as input, and approximately 0.1ꢀ1% of library molecules were
recovered after each round of selection. After three rounds of selection, between 1x107
and 1x109 DEL molecules were typically recovered. The frequencies of DEL library
molecules and SB-235375-DNA-Seq were measured after PCR amplification and next
generation DNA sequence analysis. The ratios of SB-235375-DNA-Seq to DEL
molecules allowed us to calculate enrichment of the DNAꢀtagged tool compound relative
to naïve library. The enrichment factors for SB-235375-DNA-Seq, i.e. fold increase in
copy counts relative to all DEL molecules in the sample, were calculated. Fig. 1E
summarizes the outcome of 5 independent selection experiments with SB-235375-
DNA-Seq spiked into DELs, and demonstrates reproducible enrichment of the tool
compound that is independent of the concentration of the tool in the starting DELs. The
average enrichment factor was approximately 3000ꢀfold after three rounds of selections.
This is equivalent to an enrichment factor of approximately 15 per round for each of the
three rounds of selection, a value similar to the 10ꢀ15 fold signal observed with SB-
235375-DNA-Fluor on NK3ꢀpositive versus NK3ꢀnegative cells using a fluorescence
binding assay. The finding is also consistent with the principle that the ELT selection
was driven by the concentrations of target, i.e. NK3 receptor in this case, but not the
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 10 of 36
1
2
3
4
5
6
concentration of ligand, as the enrichment factor is independent of the amount of SB-
235375-DNA-Seq spiked into the sample.
7
8
9
Selection of novel NK3 ligands. We next conducted cellꢀbased ELT selections with a
number of DELs, ranging in diversity from 5x105 to 4x109 unique smallꢀmolecule
compounds per library. After three rounds of selection, DNA was amplified by PCR,
sequenced, and decoded as described previously (1). DEL35, a threeꢀcycle library
based on a triazine scaffold (Fig 2A), was among the set of DELs employed. We
visualize a selection outcome as a cubic scatter plot (Fig 2B). Each axis of the plot
represents one cycle of synthesis, and contains a number of points equal to the number
of building blocks used in that cycle of library synthesis. Thus, each unique molecule of
the threeꢀcycle DEL35 library is represented at a unique location in the cubic space.
This visualization allows families of structurallyꢀrelated molecules to appear as lines
(two building blocks in common) or planes (one building block in common). NK3ꢀ
specific patterns (called “features” and highlighted in RED color) became visible in the
DEL35 selection cube after single copy molecules are removed from the visualization.
The cubic scatter plots from affinity selections against NK3 or NTC are shown in Fig. 3A
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
ACS Paragon Plus Environment

Page 11 of 36
ACS Combinatorial Science
1
2
3
4
and 3B, respectively.
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 2. (A) DEL35 is a library consisting of 64, 854, and 758 building blocks (BBs) at
cycle 1, 2 and 3, respectively, yielding a total of 41 million unique compounds on DNA.
Methods for library synthesis are described in detail in Supplemental Data. (B)
Interpretation of ELT selection outcomes for DEL35 as selection cube where a dot
represents a unique compound, with the building block for individual cycle mapped to
the corresponding axis. Dots are scaled by the frequency the specific encoding tag
sequence is found in the DNA sequencing output.
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 12 of 36
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 3. Results of DEL35 whole cell selection experiments with cell expressing NK3
or NTC. A total of 34,746 and 28,637 unique DEL35 molecules were identified and
visualized in the cubes for selection against NK3 (A, left) or NTC (B, left), respectively.
Some library molecules are more abundant than others in the selected population, as
reflected by their corresponding DNA sequence appearing more often in the selection
outcome. In this dataset the probability due to random chance of having 2 or more
copies of the same warhead is less than 1 in one million (23). In order to more easily
ACS Paragon Plus Environment

Page 13 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
9
visualize selected molecules with a copy count greater than one, the single copy
molecules were removed from visualization, revealing several prominent lines
(“features”) in the NK3 selection (A, right), but absent in NTC selection (B, right).
DEL35 features reproduced and competed away by a known NK3 antagonist (C).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
The features consisting of a set of parallel lines (Fig. 3A, right) demonstrate that
related molecules are more abundant in the population of molecules selected against
NK3ꢀexpressing cells. These putative NK3ꢀbinding pharmacophores are composed of
several specific combinations of library building blocks at synthesis cycle 1 and 3 with
many building blocks tolerated at cycle 2.
To elucidate the binding site of the selected DEL molecules, the experiment was
repeated in the presence or absence of a known NK3 antagonist from the same
structural series (SBꢀ735204, Ki of 10 nM). The selection results against NK3 were
highly reproducible, with the same features observed again when no inhibitor was
added (compare Fig. 3A right and 3C left). In contrast, SBꢀ735204 completely
eliminated the NK3ꢀspecific features in the selected DEL35 population (Fig. 3C, cube on
right). The competition of these lines by SBꢀ735204 suggested that the library
molecules that comprise these selected features may directly interact with NK3 at the
same site as the known inhibitor.
The DEL35 NK3 selection was repeated with more inꢀdepth sequencing (to a
total of about 114,000 unique compounds identified). The resulting DEL35 selection
cube (Fig. 4) shows additional clearlyꢀdefined features and allowed us to formulate
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 14 of 36
1
2
3
4
5
6
7
8
selection SAR hypotheses. The most prominent features in each cubic scatterꢀplot are
highlighted in shades of RED with GREEN, YELLOW, and BLUE highlight indicating
decreasing levels of copy count of each feature.
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 4. Features from cellꢀbased ELT selection provide testable SAR hypotheses for
NK3ꢀbinding pharmacophores. Each individual dot in the cube represents a DEL35
molecule, while the size of the dot corresponds to the total copy number identified for
that library molecule. Three types of features are seen in the cube: a plane, prominent
lines (exemplified as 1, 2 & 3), and highꢀcopyꢀnumber dots (exemplified by 4). The
plane and lines highlight shared building blocks. For example, dots in the light BLUE
plane share the same cycle 3 building block 5. In a colored line, all dots share the same
cycle 1 and cycle 3 building blocks.
ACS Paragon Plus Environment

Page 15 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
9
The line features (RED, GREEN, YELLOW) suggested only a small number of
cycleꢀ1 and cycleꢀ3 building blocks were selected with NK3ꢀpositive cells, while
variability in cycleꢀ2 was more widely tolerated. The plane feature (BLUE highlight)
defined by cycleꢀ3 building block 5 suggests this building block is important for NK3
affinity, and two of the most prominent feature lines (RED highlight) lie on this plane.
The encoded structure 1 of the most prominent line predicts the combination of the
highly preferred cycleꢀ3 building block 5 and the cycleꢀ1 building block 8 may be
especially important for binding. Structure 2, which corresponds to the other feature line
on the same plane, differed from structure 1 only by a single methyl group in cycleꢀ1
building blocks (compare 8 with 9). For cycleꢀ3, in addition to building block 5, building
blocks 6 (GREEN line) and 7 (YELLOW line) were the only other prominent building
blocks in the NK3 selection among a total of 758 cycleꢀ3 amines. Interestingly, these
cycleꢀ3 building blocks are also structurally related. Several cycleꢀ2 building blocks,
such as building blocks 10 and 11, occurred more often than the rest of the cycleꢀ2
building blocks found in the line encoding structure 1, suggesting certain cycleꢀ2
building blocks could further enhance the affinity to NK3. Structure 4 is a compound
representing a combination of highest copy count building blocks from all three cycles,
building block 8, 10 & 5 from cycleꢀ1, ꢀ2 and ꢀ3, respectively. In contrast to cycles 1 & 3,
almost all 854 cycleꢀ2 amines were found in the line with encoded structure 1, allowing
us to predict that cycleꢀ2 building blocks are not required for binding to NK3.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Synthesis and characterization of selected small-molecule compounds. A small
set of compounds were synthesized, without the DNA tags and subjected to
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 16 of 36
1
2
3
4
5
6
7
8
9
biochemical and functional assays. These compounds have biological activity and the
assay results support the SAR hypotheses. Figure 5 summarizes the potency data for a
representative set of compounds tested in a 125IꢀNKB competition binding assay and
cellꢀbased NK3, NK2, and NK1 functional FLIPR assays. There were consistent potency
trends amongst the compounds in both binding and functional assays. All of the
compounds listed in Figure 5 were characterized as functional antagonists. No agonist
or positive modulator activities were observed with any of these compounds. Compound
12, which corresponded to the onꢀDNA compound 4 and had the highest copy count in
the NK3 selection, demonstrated low nanoꢀmolar potency in both the NK3 binding and
functional cellꢀbased assays. In addition, greater than 1000ꢀfold selectivity for NK3 was
seen versus the related NK1 and NK2 receptors. Furthermore, the compound that
corresponds to the library molecule with the second highest copy count, compound 13,
was nearly as potent and selective. Consistent with the hypothesis based on selection
SAR, a cycleꢀ2 building block was not required for NK3 activity in a functional assay
(compound 14; fKi 25 nM). Furthermore, the group that mimics the attachment point to
the DNA construct (compound 15; fKi 160 nM) was dispensable for activity. These
truncations provided lower molecular weight compounds that remained potent NK3
antagonists and retained selectivity versus NK1 and NK2. An additional loss of 10ꢀfold
potency was seen with compound 16 (a simple combination of strongly selected
building blocks 8 and 5 and lacking the library scaffold). The SAR trends predicted by
the selection cube are largely consistent with rank order potency of the “offꢀDNA”
compounds, i.e. compound 14 vs 17 corresponding to cycleꢀ3 building block 5 vs 6 and
compound 14 vs 18 corresponding to cycleꢀ3 building block 8 vs 9, respectively.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
ACS Paragon Plus Environment

Page 17 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 5. Activity of offꢀDNA exemplars support SAR hypotheses derived from
selection cube. Ki values for each compound were determined in competition binding
assays, while fKi values were calculated from IC50 values determined in FLIPR
functional assays as described in Experimental Procedures.
Finally, a FACS assay was used to demonstrate direct binding of a fluorescentlyꢀ
labeled DNA conjugate of compound 12 to NK3ꢀpositive cells (Figure 6). The DNA
conjugate of compound 12-DNA was made fluorescent by ligation of a 5’ꢀFAMꢀlabeled
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 18 of 36
1
2
3
4
5
6
7
8
9
DNA oligonucleotide, providing 12-DNA-Fluor (Figure 6A). NK3ꢀexpressing cells
probed with 12-DNA-Fluor showed greater than 10ꢀtimes the fluorescent signal than
the parental NK3ꢀnegative cell line probed with the same molecule (Figure 6B), similar
to that observed with the tool compound SB-235375-DNA-Fluor. This result
demonstrates that binding of 12-DNA-Fluor requires the presence of NK3 on the cell
surface, and recapitulates enrichment of this molecule from the library with a pure
molecule using a whole cell fluorescence binding assay.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 6. FACS assay demonstrates direct binding of a DNAꢀtagged DEL35 molecule
to NK3 expressing cells. A fluorescent probe 12-DNA-Fluor (A) was used in the binding
experiment on NK3 positive (red histogram) or NK3 negative (blue histogram) cells (B).
ACS Paragon Plus Environment

Page 19 of 36
ACS Combinatorial Science
1
2
3
4
5
6
Cells were analyzed by FACS. The mean channel fluorescence of NK3 positive cells
was 12ꢀfold greater than control NK3 negative cells (2921 versus 244, respectively).
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Identification of additional NK3 antagonist chemotypes. To determine if ligands
derived from other DNAꢀencoded libraries could be identified against NK3, additional
selection experiments were performed. Approximately 30 libraries containing greater
than 15 billion compounds in total were selected against NK3. NK3ꢀspecific features
were observed from multiple libraries with various scaffolds. Representative smallꢀ
molecules were synthesized offꢀDNA and tested for NK3 receptor activities. Fig. 7
highlights representative features and active molecules found from three additional
libraries. These chemotypes are structurallyꢀdistinctive from each other and from the
original DEL35 feature. Full descriptions of these additional DNAꢀencoded libraries will
be communicated at a later time.
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 20 of 36
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Figure 7. Additional NK3ꢀspecific features from cellꢀbased ELT selection provide novel
NK3 antagonist pharmacophores. From libraries DEL39 (A), DEL47 (B) and DEL50 (C),
representative selection features and smallꢀmolecule feature exemplars are shown.
NK3 antagonist and tachykinin receptor selectivity data are also summarized.
Selection with the 2ꢀcycle library DEL39 (600,000 unique compounds) provided
the 2ꢀD plot shown in Fig. 7A, indicating a single cycleꢀ1 building block was strongly
preferred among approximately 2,000 building blocks used in the library synthesis.
Compounds containing this cycleꢀ1 building block and a few selected cycleꢀ2 building
ACS Paragon Plus Environment

Page 21 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
9
blocks were prepared to confirm that the 3,4,5ꢀsubstituted pyrazole (cycleꢀ1 building
block) was the critical pharmacophore driving NK3 receptor binding. Potency and
selectivity data for compound 19, is also shown in Fig. 7A. Similarly, features were
identified from DEL47, a 3ꢀcycle, 320 million member library based on the
benzimidazole core. Guided by selection SAR (Fig. 7B), compound 20 was prepared.
Low nanomolar potency, drugꢀlike ligand efficiency (24), and excellent selectivity
against NK1 and NK2 were found for this compound.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Additionally, a set of features from DEL50, a 3ꢀcycle library with 1.3 million
unique compounds, yielded compound 21 (Fig. 7C). This compound represents a
distinct chemotype and is also a ligand efficient NK3 antagonist.
Summary
The DNAꢀtagged tool compounds and proof of concept experiments described
herein enabled us to establish a simple protocol for cellꢀbased affinity selection using
DNAꢀencoded combinatorial libraries. Selection on cells expressing NK3 yielded
enriched features encoding multiple series of novel chemotypes. Structure activity
relationships, deduced from the selection outputs, effectively guided decisionꢀmaking
regarding which compounds to synthesize and test for activity. Multiple series of potent,
ligandꢀefficient, and selective NK3 antagonists were identified that are structurallyꢀ
distinct from known NK3 antagonists (25, 26), and to our knowledge have not been
reported in the literature.
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 22 of 36
1
2
3
4
5
6
7
8
9
Initial selection experiments with DEL35 enabled proof of concept and provided
potent and NK3ꢀselective triazineꢀbased antagonists 12ꢀ18. Thereafter, hit confirmation
efforts focused on features found from structurallyꢀdiverse, nonꢀtriazine 2ꢀ and 3ꢀcycle
libraries. The lower molecular weight and higher ligand efficiency (24) make these
chemotypes more attractive starting points for medicinal chemistry. The majority of
these new chemotype exemplars (i.e. 19, 20, 21) have ligand efficiency values that are
in excess of 0.3 and thus consistent with quality leads and drugs (see Fig. S8 in
Supplemental Data). The potency/molecular weight profile of the DEL47 exemplar,
compound 20, compares well to that of Talnetant 22 and Osanetant 23, two NK3
antagonists with the most clinical studies for disease interventions (18).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
The described method is based on affinity enrichment of DNAꢀtagged
compounds on a target of interest. Since the concentration of any individual compound
in the DEL library is very low, often in the pM or fM range, binding of a DNAꢀlinked
compound to a target protein is driven by target concentration, cell surface receptor
density in this case. As such, a critical factor for successful enrichment is the receptor
expression level. Using a DNAꢀtagged NK3 antagonist as a tool compound, we
determined the minimum level of NK3 expression for successful enrichment was around
25,000 receptors per cell, an expression level that is achievable for most GPCRs. The
enrichment of targetꢀspecific DNAꢀtagged compounds was directly correlated to the
surface receptor density (see Fig.S9 in Supplemental Data).
Conclusions and Implications
ACS Paragon Plus Environment

Page 23 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
9
Identification of high quality and novel chemical templates as lead compounds is
essential for the success of small molecule drug discovery. Most lead discovery efforts
are carried out with traditional HTS campaigns that utilize biochemical or cellꢀbased
functional assays, in microꢀtiter plates with one compound per well. In this paper, we
describe a cellꢀbased selection methodology and representative screening results of
DNAꢀencoded smallꢀmolecule library technology against a Gꢀprotein coupled receptor,
NK3. To our knowledge, this is the first reported application of small molecule DNA
encoded library technology to integral membrane protein targets, such as GPCRs, and
represents a new paradigm for discovery of low molecular weight ligands that bind to
integral membrane proteins.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Our results illustrate the exciting potential of ELT for membrane based targets,
such as GPCRs (27). Combined with the very simple BacMam expression technology
(28, 29), the selection with multiple DEL libraries can be done quickly with target
expression on intact cells and without the requirement of protein purification. The
selection method is simple and easily amendable for modification to include additional
selection conditions, such as coꢀfactors or competitors. Libraries with combined
diversity of billions or more can be selected simultaneously.
Since the original proofꢀofꢀconcept experiment with NK3, a number of other
GPCRs have gone through the selection process. Novel chemotypes were successfully
identified for some of these receptors, ranging from chemokine receptors,
peptides/protein hormone receptors to lipid sensing receptors. Each receptor yielded a
unique set of enriched features that were specific to that receptor. Off DNA resynthesis
and testing demonstrated that all compounds were antagonists. Interestingly, to date we
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 24 of 36
1
2
3
4
5
6
7
8
9
have had little success with receptors whose natural ligands are low molecular weight
compounds, such as the aminergic receptor subꢀfamilies in the family A of GPCR. This
is in contrast with our experience with traditional plateꢀbased HTS, in which high hit
rates were often found with aminergic receptors. The cause of this discrepancy is not
fully understood, although compounds with structures similar to antagonists of amine
receptors are expected to be widely present in the DEL libraries.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Not surprisingly, all compounds we identified here were antagonists and none
were active as agonists or positive modulators. This is analogous to our experience with
other lead identification strategies and is probably due to the difficulty of mimicking
biological activity of a peptide ligand with low molecular weight synthetic compounds.
However, encoded library technology should be able to identify agonist ligands if
contained in the library and suitably presented to cells. For example, we were able to
significantly enrich a DNAꢀtagged tool compound prepared from a known agonist
peptide ligand of NK3 using the same cellꢀbased selection methodology described here
(unpublished data), demonstrating that an agonistꢀDNA conjugate can bind to the
receptor.
The NK3 antagonists discovered from DEL35, DEL39, DEL47 and DEL50
(compounds 12, 19, 20, 21, respectively) are structurally distinct from one another. A
retrospective analysis revealed that no closely related structural analogs of these
compounds were present in the approximately 2 million member screening collection in
use at that time (entire HTS set < 0.85 Tanimoto similarity using Chemaxon chemical
fingerprints) (30). Our results with the onꢀDNA tool compound SBꢀ235375ꢀDNAꢀSeq
(Figure 1) demonstrate that a DNA conjugate derived from Talnetant is capable of
ACS Paragon Plus Environment

Page 25 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
9
binding to NK3. However, we did not discover analogues of Talnetant (or Osanetant) in
our ELT screen. In this regard, ELT and traditional plateꢀbased HTS methods are
complementary hit identification approaches because different chemical templates were
found with each platform.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
All of the chemotypes described here represent primary hits directly selected
from the DEL library collection (or are lower molecular weight, truncated derivatives)
and are unꢀoptimized by medicinal chemistry. Several of these compounds have
potency, specificity and ligand efficiency properties comparable to Talnetant and
Osanetant (i.e. DEL35ꢀderived 12 and DEL47ꢀderived 20). These results demonstrate
the capability of ELT to efficiently provide high quality, leadꢀ and drugꢀlike molecules
against integral membrane protein targets.
Experimental Procedures
Preparation of DNA-encoded tool and library molecules. For the synthesis of all
DNAꢀencoded tool compounds and library molecules, a common AOPꢀheadpiece
construct was employed as starting material. The structure, synthesis, characterization,
and several common synthetic protocols used to modify this construct were reported
previously (1). Briefly, DNAꢀtagged NK3 antagonist tool compound SB-235375-DNA
was obtained by DIC/HOAt mediated coupling in water of the known NK3 antagonist
SBꢀ235375 to the AOPꢀheadpiece. Synthesis of a 4ꢀcycle triazineꢀbased library has
already been described in our earlier work (see DELꢀB in reference 1). The simpler 3ꢀ
cycle DEL35 library (Fig. 2A) used here was prepared using only the last three of the
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 26 of 36
1
2
3
4
5
6
7
8
9
four encoded diversity cycles of DELꢀB. Triazine was linked to the DNA headpiece
through a short PEGꢀbased spacer. Cycleꢀ1 and cycleꢀ2 building blocks were directly
linked to the other two coupling sites on triazine. Cycle 1 contains 64 amino acid or
phenolic acid building blocks, while cycle 2 contains 854 amines. Cycle 3, which was
linked to cycle 1, contains 758 amines, giving a library with a theoretical total of
41,429,248 components. Validation of library building blocks and installation of the
DNAꢀencoding of DEL35 were done with only minor modification to the published
methods and protocols (for DEL35 reaction scheme and more details, see
Supplementary Methods).
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Preparation of cells. HEK293 cells were grown as a monolayer culture and
transduced with recombinant BacMam virus (31) encoding NK3 receptors at various
virus to cell ratios. After 24 hours, cells were harvested with Versene® treatment and
receptor expression levels on the cell surface were determined with a filter binding
assay using radioactiveꢀlabeled NKB as described below. For cellꢀbased DNAꢀencoded
smallꢀmolecule library selection and competition binding assays, HEK293 cells
transduced with the appropriate virus to cell ratios were prepared as described above.
Cell-based DNA-encoded small-molecule library selection. Selection experiments
were started with 1.5x1015 of input DEL molecules with or without the onꢀDNA tool
compound spikeꢀin. DEL library (either a single library or a pool of libraries) was
incubated with 107 cells at 37oC for 30 minutes in 1 mL PBS containing 0.1% (g/100 mL)
sodium azide to inhibit receptor internalization. A blocking agent, such as sheared
salmon sperm DNA, was also included at a final concentration 1 mg/mL to reduce
nonspecific binding. Cells were then washed with ice cold PBS (1 mL) ten times by
ACS Paragon Plus Environment

Page 27 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
9
gentle centrifugation and reꢀsuspension to remove unbound library molecules. Bound
molecules were eluted by heating cells in PBS to 95oC for 10 minutes, followed by
centrifugation at 14,000 rpm to remove cells. The recovered DEL molecules in the
supernatant were used for the next round of selection with freshly prepared cells using
the same protocol. Typically, this process was repeated one more time for a total of
three rounds of selection. For competitive selections, known NK3 antagonist SBꢀ735204
was added to the selection buffer to a final concentration of 10 ꢀM at each selection
cycle, and the selection was performed using the same method.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
DNA sequencing, tag decoding, and data visualization. The amount of total DEL
library recovered after each round of selection was measured by quantitative
polymeraseꢀchainꢀreaction (qPCR) using a DEL specific primer pair. Selection outputs
after the final round of selection were also amplified using preparative PCR, which was
then used for highꢀthroughput sequencing using either the Roche 454 Genome
Sequencer or Illumina Genome Analyzer. Sequences were parsed into individual
libraries using a DNA library tag that was unique to each library. Each selected
molecule was “translated” into the chemical building blocks used during library
synthesis, encoded binding motifs structures virtually enumerated, and analyzed as
described previously (1). All PCR and sequencing reactions were conducted using
standard molecular biology methods as recommended by the reagent suppliers. The
resulting data were visualized as a cubic scatter plot using Spotfire DecisionSite
(Spotfire, Inc., Somerville, Massachusetts).
Calculation of Enrichment Factor, i.e. fold increase of copyꢀnumber relative to all
DEL molecules in the sample, is calculated as: Enrichment Factor = (Copy-number A
ACS Paragon Plus Environment

ACS Combinatorial Science
Page 28 of 36
1
2
3
4
5
6
7
8
9
/ Total) after selection / (Copy-number A / Total) before selection, where Copy-number A is
the copyꢀnumber of sequence that represent DEL molecule of interest, such as SB-
235375-DNA, and Total is the sum of copy numbers for all variable sequences present
in the samples.
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Filter binding assays with radioactive-labeled NK receptor ligands. Radioactiveꢀ
labeled tachykinin receptor ligands, 125IꢀSubstance P, 125IꢀNKA and 125Iꢀ[MePhe7]ꢀNKB
for NK1, NK2 and NK3 receptor, respectively, were used in the filter binding assays as
described previously (32ꢀ34).
Measurement of NK functional activity. Compounds were characterized in a
functional assay for calcium release in response to NK receptor ligands. Changes in
intracellular Ca2+ concentrations were monitored using a fluorescence imaging plate
reader (FLIPR, Molecular Devices Corp., Sunnyvale, California) by a modified method
described previously (32, 33). The functional affinities of antagonist compounds fKi
were calculated using the following equation, where the concentration of agonist used is
[A], the concentration of agonist producing 50% maximal response is EC50 and the n is
the Hill coefficient of the agonist doseꢀresponse curve (35).
fKi = IC50 / ( 2 + ([A]/EC50)n)1/n – 1
Acknowledgements
The authors wish to acknowledge Matthew Clark and Steven Skinner for
guidance and help during synthesis of the DEL libraries and DNAꢀtagged tool
ACS Paragon Plus Environment

Page 29 of 36
ACS Combinatorial Science
1
2
3
4
5
6
7
8
compounds and Ming Jiang for testing the compounds in competitive binding and FLIPR
functional assays. The authors want to also acknowledge Barry A Morgan and Robert P
Hertzberg for their sponsorship of this project.
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
Supporting Information (summary)
The Supporting Information section describes in detail syntheses of the DEL35 library,
onꢀDNA tool compounds, and molecules derived from the libraries that were reꢀ
synthesized (off DNA) and tested for biological activity. There is also more detail on
ligand efficiency and selection enrichment calculations as it relates to receptor
expression levels, topics that are only briefly alluded to in the manuscript.
Abbreviations
NK – neurokinin
GPCR ꢀ Gꢀprotein coupled receptor
ELT – encoded library technology
HTS – high throughput screening
DEL – DNA encoded library
SAR – structure activity relationship
NTC – no target control
FACS – fluorescence activated cell sorter
FAM – fluorescein
PBS – phosphate buffered saline
ACS Paragon Plus Environment