Nanoscale synthesis and affinity ranking

Article abstract of DOI:10.1038/s41586-018-0056-8

Most drugs are developed through iterative rounds of chemical synthesis and biochemical testing to optimize the affinity of a particular compound for a protein target of therapeutic interest. This process is challenging because candidate molecules must be

Full text of DOI:10.1038/s41586-018-0056-8

Letter
https://doi.org/10.1038/s41586-018-0056-8
Nanoscale synthesis and affinity ranking
Nathan J. Gesmundo1,3, Bérengère Sauvagnat2, Patrick J. Curran2, Matthew P. richards2, Christine L. Andrews2,
Peter J. Dandliker2 & tim Cernak1,4
*
Most drugs are developed through iterative rounds of chemical protein concentration. With this advance, potent protein inhibitors
synthesis and biochemical testing to optimize the affinity of a may be identified at a nanomole-scale while forgoing time-consuming
particular compound for a protein target of therapeutic interest. This reaction purification.
process is challenging because candidate molecules must be selected
We validated NanoSAR by synthesizing libraries based on three of
from a chemical space of more than 1060 drug-like possibilities1, and the most popular transformations used in pharmaceutical research22,23
.
a single reaction used to synthesize each molecule has more than Each library contains about 20 compounds and each reaction was run
107 plausible permutations of catalysts, ligands, additives and other on a nanomole scale (less than 0.05mg). Every analogue was resyn-
parameters2. The merger of a method for high-throughput chemical thesized on a traditional scale (about 20mg) so that bioassay results
synthesis with a biochemical assay would facilitate the exploration of of unpurified nanoscale reactions could be compared directly to data
this enormous search space and streamline the hunt for new drugs obtained from purified products using a conventional approach. The
and chemical probes. Miniaturized high-throughput chemical crude nanoscale reaction mixtures were analysed by using ultra-
synthesis3–7 has enabled rapid evaluation of reaction space, but so high-performance liquid chromatography–mass spectrometry (UPLC–
far the merger of such syntheses with bioassays has been achieved MS) to evaluate the reaction efficiency, pooled, and submitted to the
with only low-density reaction arrays, which analyse only a handful ASMS affinity-ranking assay directly (Extended Data Fig. 1). Next, each
of analogues prepared under a single reaction condition8–13. High- product was resynthesized on an approximately 20-mg scale using the
density chemical synthesis approaches that have been coupled to productive reaction conditions identified on the nanoscale and puri-
bioassays, including on-bead14, on-surface15, on-DNA16 and mass- fied. These purified products were likewise submitted to the ASMS
encoding technologies17, greatly reduce material requirements, affinity-ranking assay, but were also either submitted to a traditional
but they require the covalent linkage of substrates to a potentially assay of biochemical function or compared to reported binding affini-
reactive support, must be performed under high dilution and must ties. Potent molecules could be identified regardless of the scale of the
operate in a mixture format. These reaction attributes limit the reaction.
application of transition-metal catalysts, which are easily poisoned
The bioassay results from the crude nanoscale reactions matched
by the many functional groups present in a complex mixture, and of those obtained from the purified products. For example, a library of
transformations for which the kinetics require a high concentration 19 known inhibitors24 of the kinase ERK2 was prepared by allowing
of reactant. Here we couple high-throughput nanomole-scale carboxylic acid 1 to react with diverse amines in the presence of HATU
synthesis with a label-free affinity-selection mass spectrometry (hexafluorophosphate azabenzotriazole tetramethyl uronium) and
bioassay. Each reaction is performed at a 0.1-molar concentration in iPr2NEt (19 reactions, Fig. 2a) and submitted to affinity ranking. High-
a discrete well to enable transition-metal catalysis while consuming affinity amide product 2 (inhibitory constant Ki =35nM), which was
less than 0.05 milligrams of substrate per reaction. The affinity-
selection mass spectrometry bioassay is then used to rank the
a
O
O
Simultaneous optimization of
affinity of the reaction products to target proteins, removing the
need for time-intensive reaction purification. This method enables
the primary synthesis and testing steps that are critical to the
invention of protein inhibitors to be performed rapidly and with
minimal consumption of starting materials.
N
N
S
S
Nu
H
H
H
N
N
N
N
reaction
conditions
building
blocks
protein
afꢀnity
Br
Nu
N
H
N
H
The merger of nanoscale synthesis with affinity ranking, which we call
NanoSAR, could enable data-rich interrogations of reaction–chemical–
bioactivity space (Fig. 1). Nanoscale synthesis consumes less than
0.05mg of substrate per reaction3, so precious starting materials such
as complex drug-like intermediates can be used. Affinity-selection
b
mass spectrometry (ASMS) is a sensitive bioassay18,19 that relates pro
-
Traditional
synthesis
1 condition
~20 mg
Nanoscale
synthesis
Many conditions
~0.05 mg
per reaction
ASMS
Affinity
ranking
Decreasing the
protein concentration
induces competition
tein binding to mass detection of a candidate molecule. ASMS was
chosen to assay crude nanoscale reaction mixtures directly because
misleading readouts related to catalysts or other reaction components
can be omitted as long as the molecular masses of the desired prod-
ucts and residual reagents can be resolved by high-resolution mass
spectrometry. However, most existing ASMS protocols provide only
a binary readout of affinity20. We envisioned that affinity could be
ranked by titrating down the concentration of protein while main-
taining a constant concentration of compound. As the protein con-
centration decreases, competition for binding is induced21, and only
compounds with high affinity to the target are observed at the lowest
Separates
compounds
on the basis
of afꢀnity
per reaction
Fig. 1 | Nanoscale synthesis and affinity ranking (NanoSAR).
a, Reaction space and chemical space are navigated at the nanomole-scale
and bioassayed directly to locate potent protein inhibitors and productive
reaction conditions simultaneously. b, Nanoscale synthesis consumes
a fraction of the material that is required for traditional synthesis.
Crude reactions are assayed by ASMS, with affinity ranking achieved by
decreasing the concentration of the target protein to induce competition
among compounds.
1Department of Discovery Chemistry, Merck & Co., Inc., Boston, MA, USA. 2Department of Pharmacology, Merck & Co., Inc., Boston, MA, USA. 3Present address: Research and Development, AbbVie,
North Chicago, IL, USA. 4Present address: Department of Medicinal Chemistry, Department of Chemistry, Program in Chemical Biology, University of Michigan, Ann Arbor, MI, USA. *e-mail:
tcernak@umich.edu
N A t U r e | www.nature.com/nature
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

reSeArCH Letter
purified and assayed for inhibition of the function of the kinase
MK226. NanoSAR can readily distinguish potent inhibitors, such
as 5 (half-maximal inhibitory concentration IC50 = 59 nM), from
impotent inhibitors, such as 6 (IC50 3,400 nM). IC50 values for all
18 analogues were determined, using purified compounds produced
on a 50-μmol scale, and could be predicted from the NanoSAR data
(Extended Data Fig. 3). Similarly, bromide 7 was used to generate 20
distinct inhibitors27 of the kinase CHK1 via C–C and C–N coupling
(170 reactions, Fig. 2c). Potent inhibitors, such as 8 (IC50 23nM), were
readily distinguished from an array of less-potent, lower-affinity com-
pounds, such as 9 (IC50 >10,000 nM). Once again, affinity-ranking
results from crude nanoscale reactions predicted the CHK1 func-
tional activity of purified samples produced on a micromole scale
(Extended Data Fig. 4). Samples treated with a metal scavenging resin
had low levels of residual palladium; however, simple dilution with
water reduced palladium levels to less than 5p.p.m. (Extended Data
Fig. 5).
Yield (%)
[Protein]
ID High1 Low
a
H
[ERK2]
(μM)
HATU,
HATU,
O
1
N
iPr2NEt, NMP
iPr2NEt, NMP
0
100
A1
A2
A3
A4
A5
A6
A7
A8
A9
A10
A11
A12
A13
22
49% yield
38% yield
N
OH
HN
HO
O
OH
HN
HN
1
Cl
19 reactions
+ afnity ranking
33
A14
A15
A16
A17
2
3
0.7 mg consumed
Ki = 35 nM
Ki > 5,000 nM
A
NB 5 1.2 0.2
XPhos Pd G3,
aq. K3PO4, DMSO aq. K3PO4, NMP
tBu3P Pd G2,
8 reaction
conditions
[MK2]
(μM)
b
Br
N
ID
B1
55
B2
B3
B4
B5
B6
B7
B8
B9
B10
B11
B12
B13
B14
66
89% yield
87% yield
OH
NH
O
HN
O
Me
4
NH
The information density produced by NanoSAR permits explora-
tions that were previously impractical or impossible, such as the rapid
survey of reaction conditions, building blocks and protein affinity
(Fig. 3). For example, 7 was coupled to various functionalized nucleo-
philes via multiple distinct chemical transformations, with affinity
information for CHK1 generated for each analogue of the form of 10
produced. As shown in the outer reaction-condition heat map in Fig. 3,
we explored combinations of catalysts (11–26), bases (27–34) and sol-
vents (N-methyl pyrrolidinone (NMP), dimethylsulfoxide (DMSO) and
N,N-dimethylformamide (DMF)) with 17 representative nucleophiles
(D1–D17, Extended Data Fig. 6). For each class of nucleophile, four of
the most productive reaction conditions were selected so that 7 could
be effectively coupled to 384 diverse building blocks, as shown in the
inner heat map of building-block reactivity (Fig. 3). By using four reac-
144 reactions
+ afŒnity ranking
5.1 mg consumed
5
6
B15
B16
IC50 = 59 nM
IC50 = 3,400 nM
ABCDEFGH NB 4.8 1.2 0.3
tBuBrettPhos Pd G3, RuPhos Pd G3,
8 or 9 reaction
conditions ID
[CHK1]
(μM)
c
MTBD, NMP
aq. K3PO4, DMF
NH
N
C1
C2
C3
C4
C5
C6
C7
C8
88
85% yield
79% yield
H
N
HN
NH
N
O
S
C9
N
N
C10
C11
C12
C13
C14
C15
C16
C17
99
N
7
Br
170 reactions
+ afŒnity ranking
8.8 mg consumed
OMe
8
9
IC50 = 23 nM IC50 > 10,000 nM
C18
ABCDEFGH I
NB 10 5 2.51.2
tion conditions, we identified productive C–N, C–O, C–S or C–C cou
-
Fig. 2 | NanoSAR identification of ERK2, MK2 and CHK1 inhibitors.
Inhibitors were produced using nanoscale synthesis and submitted directly
to an ASMS affinity-ranking assay, then resynthesized on a 50-μmol
scale, purified, and reassayed by ASMS for comparison to the functional
bioactivity (IC50) or reported binding affinity (Ki) of each compound.
a, A library of 19 ERK2 inhibitors prepared by amide coupling. b, A library
of 18 MK2 inhibitors prepared by Suzuki coupling. c, A library of 20 CHK1
inhibitors prepared by Suzuki and C–N coupling. In each library, every
compound has a unique identifier (ID) and was formed under one or
more reaction conditions, with the yield, determined by UPLC, for each
reaction depicted in the heat map and the affinity-ranking data depicted
in the scatter plot. One high-affinity compound (2, 5 and 8) and one
low-affinity compound (3, 6 and 9) from each library are shown. The
reaction conditions identified on the nanoscale (red box) were then used
for the 50-μmol-scale resynthesis of each exemplary compound, and the
corresponding affinity-ranking data for that compound are highlighted
(red circle). NB indicates no binding was observed in the affinity-ranking
assay at the highest protein concentration tested. See Extended Data
Figs. 2–4 for reaction conditions (A–I), structures, yields and Ki or IC50
values of all compounds.
pling conditions for 345 of 384 products (90%) (Extended Data Figs. 7,
10). The 345 crude reaction mixtures, outlined in black in the inner
heat map, were submitted directly to affinity ranking by titrating the
CHK1 concentration from 10μM to 0.2μM: products that bind tightly
to CHK1 appear closer to the centre of the circle in the affinity-ranking
plot (Fig. 3). Key compounds were identified for scale-up, purification
and analysis in an assay of CHK1 functional activity (Extended Data
Fig. 8). Exemplars (35–41) are shown in Fig. 3 (see Extended Data Fig. 9
for dose titration curves), with each class of functional group revealing
unique structure–reactivity and structure–bioactivity relationships.
Compounds 38, 39 and 41 were isolated in 98%, 52% and 24% yield,
respectively, from reactions run on a 50-μmol scale, and were potent
CHK1 inhibitors, as predicted by NanoSAR.
Subtle reactivity trends were illuminated in the data analysis. For
instance, one of the thiol nucleophiles that we tested in the initial survey
of reaction conditions (3-phenylpropane-1-thiol, D15) couples to 7 in
high yield regardless of the catalyst, base or solvent used (see Extended
Data Fig. 6 for reaction-condition mapping). This finding suggests that
alkyl thiols couple to 7 via a nucleophilic aromatic substitution (SNAr)
mechanism rather than via palladium catalysis. Guided by this infor-
mation, two of the four reaction conditions selected to couple thiols
to 7 used no catalyst. Compound 35 was isolated in 98% yield on a
50-μmol scale using no catalyst, with P2Et in DMF; however, it was
inactive against CHK1, as predicted by NanoSAR.
Higher-order trends also emerge from the data. For example, com-
paring the reaction performance of boronate to amine building blocks
suggests that, at least with 7, the Suzuki coupling is a more robust
reaction than is the C–N coupling. Years of anecdotal experience
with these two transformations led us to hypothesize this disparity
in reaction robustness7, and here the phenomenon can be readily
visualized. Patterns of bioactivity are also apparent in the central
plot of affinity ranking in Fig. 3. For instance, many of the boronate-,
amide- and amine-derived products were bioactive, exhibiting affi-
nity to CHK1 and appearing as points closer to the centre of the circle;
these compounds exhibited functional inhibition of CHK1 function
observed at the lowest ERK2 concentration tested in the affinity-ranking
assay, was easily distinguished from low-affinity amide product 3
(Ki >5,000nM), which was not observed at any of the ERK2 concen-
trations tested. Therefore, with NanoSAR it was possible to differentiate
potent from impotent analogues with less than 0.05mg of material and
with no reaction purification (Extended Data Fig. 2).
For this library of ERK2 inhibitors produced by amide coupling, a
single reaction condition was sufficient. For subsequent transition-
metal-catalysed Suzuki or C–N coupling reactions, we surveyed
eight and nine reaction conditions, respectively, to ensure that every
product was synthesized successfully. Nanoscale Suzuki coupling
of heterocyclic bromide 4 with 18 diverse boronates was achieved
with 8 combinations of Buchwald precatalysts25, bases and solvents
(144 reactions, Fig. 2b). Using conditions identified at the nanos-
cale, all 18 products were prepared on a traditional 50-μmol scale,
N A t U r e | www.nature.com/nature
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

Letter reSeArCH
O
O
Nu
H
reaction
building
blocks
protein
afꢀnity
N
N
S
S
H
conditions
H
N
N
N
N
17 BBs ×
24–54 RCs
384 BBs ×
4 RCs
345 products
Br
Nu
Yield (%)
Conversion (%)
[CHK1] (μM)
N
H
N
H
7
10
0
100
0
100
10
0.1
3,114 reactions
+ afꢀnity ranking
123 mg consumed
2
286
Bases
Buchwald
precatalysts
27 P2Et
28 BTMG
29 BTTP
30 MTBD
31 BEMP
32 DBU
11 tBuXPhos
12 DTBPF
13 XantPhos
14 tBu3P
15 APhos
16 XPhos
17 RuPhos
18 CPhos
33 Cs2CO3
34 K3PO4
19 JackiePhos
20 DTBPF
0
21 BrettPhos
22 BINAP
23 tBuBrettPhos
24 RockPhos
25 AdBrettPhos
26 SPhos
No catalyst,
P2Et, DMF
98% yield
No catalyst,
P2Et, DMF
77% yield
RuPhos,
BTTP, NMP
20% yield
RuPhos,
K3PO4 (aq.), NMP
98% yield
tBuXPhos,
MTBD, DMF
52% yield
tBuBrettPhos,
P2Et, DMF
62% yield
tBuXPhos,
P2Et, NMP
24% yield
H
O
N
N
H
N
S
H2N
Me
Nu = Me
Me
N
O
N
N
N
H
S
S
NH
O
O
O
Me
MeO
35 Thiol
36 Alcohol
37 Alkyne
38 Boronate
39 Amide
40 Sulfonamide
41 Aryl amine
IC50 > 10,000 nM
IC50 > 10,000 nM
IC50 = 6,570 nM
IC50 = 51 nM
IC50 = 5 nM
IC50 > 10,000 nM
IC50 = 6 nM
Fig. 3 | NanoSAR mapping of reactivity and bioactivity for diverse
submitted directly to the ASMS affinity-ranking assay against CHK1, as
CHK1 inhibitors. 3,114 reactions were run, using only 123 mg of 7, to map shown in the central plot. Points closer to the centre of the circle represent
structure–reactivity and structure–bioactivity relationships. The outer heat compounds whose ASMS binding signals persisted at lower CHK1
map shows a survey of reaction conditions (RCs) with 17 representative
building blocks (BB; D1–D17, Extended Data Fig. 6). The inner heat map
shows library synthesis of 384 diverse building blocks (see Supplementary
Information for structures) using four productive reaction conditions
(A–D) selected from the reaction-condition survey. Using the reaction
conditions outlined in black in the inner heat map, 345 compounds were
concentrations, corresponding to higher CHK1 affinity. NT, not tested
owing to 0% yield. NB, no binding observed at 10 μM CHK1. Exemplary
compounds (35–41), the NanoSAR data for which are outlined in red,
were resynthesized on a 50-μmol scale and purified to obtain IC50 values
from a functional assay of CHK1 activity. See Extended Data Figs. 6–9
for details.
N A t U r e | www.nature.com/nature
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

reSeArCH Letter
Online content
a
b
Single condition
Best of 4 conditions
Any Methods, including any statements of data availability and Nature Research
reporting summaries, along with any additional references and Source Data files,
are available in the online version of the paper at https://doi.org/10.1038/s41586-
018-0056-8.
Received: 2 October 2017;Accepted: 14 February 2018;
Published online xx xx xxxx.
1. Reymond, J. L. The chemical space project. Acc. Chem. Res. 48, 722–730
(2015).
2. Murray, P. M., Tyler, S. N. G. & Moseley, J. D. Beyond the numbers: charting
chemical reaction space. Org. Process Res. Dev. 17, 40–46 (2013).
3. Buitrago Santanilla, A. et al. Nanomole-scale high-throughput chemistry for the
synthesis of complex molecules. Science 347, 49–53 (2015).
4. Shevlin, M. Practical high-throughput experimentation for chemists. ACS Med.
Chem. Lett. 8, 601–607 (2017).
5. Collins, K. D., Gensch, T. & Glorius, F. Contemporary screening approaches to
reaction discovery and development. Nat. Chem. 6, 859–871 (2014).
6. Troshin, K. & Hartwig, J. F. Snap deconvolution: an informatics approach to
high-throughput discovery of catalytic reactions. Science 357, 175–181 (2017).
7. Kutchukian, P. S. et al. Chemistry informer libraries: a chemoinformatics
enabled approach to evaluate and advance synthetic methods. Chem. Sci. 7,
2604–2613 (2016).
PC2
PC2
t
c BuBrettPhos,
MTBD, DMF
72% yield
tBuBrettPhos,
MTBD, NMP
36% yield
tBuBrettPhos,
MTBD, NMP
27% yield
RuPhos,
K3PO4 (aq.), NMP
31% yield
O
F
HN
S
Me
O
N
N
N
H
O
Me
8. Werner, M. et al. Seamless integration of dose-response screening and fow
chemistry: efcient generation of structure–activity relationship data of
β-secretase (BACE1) inhibitors. Angew. Chem. Int. Ed. 53, 1704–1708 (2014).
9. Desai, B. et al. Rapid discovery of a novel series of Abl kinase inhibitors by
application of an integrated microfuidic synthesis and screening platform.
J. Med. Chem. 56, 3033–3047 (2013).
10. Guetzoyan, L., Nikbin, N., Baxendale, I. R. & Ley, S. V. Flow chemistry synthesis of
zolpidem, alpidem and other GABAA agonists and their biological evaluation
through the use of in-line frontal afnity chromatography. Chem. Sci. 4,
764–769 (2013).
11. Karageorgis, G., Dow, M., Aimon, A., Warriner, S. & Nelson, A. Activity-directed
synthesis with intermolecular reactions: development of a fragment into a
range of androgen receptor agonists. Angew. Chem. Int. Ed. 54, 13538–13544
(2015).
12. Murray, J. B., Roughley, S. D., Matassova, N. & Brough, P. A. Of-rate screening
(ORS) by surface plasmon resonance. An efcient method to kinetically sample
hit to lead chemical space from unpurifed reaction products. J. Med. Chem. 57,
2845–2850 (2014).
42 Amide
IC50 = 17 nM
43 Alkyl amine
IC50 = 17 nM
44 Aryl amine
IC50 = 18 nM
45 Boronate
IC50 = 52 nM
Fig. 4 | NanoSAR enhances coverage of chemical space. a, When coupling
7 to diverse nucleophiles (D18–D402), a single reaction condition (11, 27,
NMP, room temperature) produced 158 (yellow circles) out of 384 (41%;
see inset pie chart) designed products, with 226 failures (0% yield; purple
crosses). The purple highlighted regions indicate ‘dark’ regions of principal
component (PC) chemical space, where reactions failed. b, Matching each
nucleophile to the best of four reaction conditions produced 345 out of 384
(90%) products. e, 21 compounds from the ‘dark’ space that were made
accessible by NanoSAR bound to CHK1, such as 42–45. CHK1 IC50 values
are from a functional assay of purified compounds.
(Extended Data Fig. 9). By contrast, few of the sulfonamide-, thiol-, 13. Baranczak, A. et al. Integrated platform for expedited synthesis–purifcation–
testing of small molecule libraries. ACS Med. Chem. Lett. 8, 461–465 (2017).
14. Price, A. K., MacConnell, A. B. & Paegel, B. M. hνSABR: photochemical
dose–response bead screening in droplets. Anal. Chem. 88, 2904–2911 (2016).
15. Vastl, J., Wang, T., Trinh, T. B. & Spiegel, D. A. Encoded silicon-chip-based
platform for combinatorial synthesis and screening. ACS Comb. Sci. 19,
255–261 (2017).
16. Goodnow, R. A. Jr, Dumelin, C. E. & Keefe, A. D. DNA-encoded chemistry:
alkyne- or alcohol-derived products bound to CHK1, and these link-
ages could reasonably be deprioritized from future design cycles. In
total, only 123mg of 7 were consumed to run 3,114 reactions, to match
345 of the 384 diverse nucleophiles with coupling conditions, to assay
the products and to paint a landscape of reactivity and bioactivity.
For comparison, coupling of bromide 7 to the same 384 diverse
enabling the deeper sampling of chemical space. Nat. Rev. Drug Discov. 16,
131–147 (2017).
nucleophiles under a single reaction condition—the recently reported
17. Annis, D. A. et al. An afnity selection–mass spectrometry method for the
combination of tBuXPhos Pd G3 (11)25 with P2Et (27) as a base28, which
identifcation of small molecule ligands from self-encoded combinatorial
is among the most substrate-tolerant palladium-catalysed coupling
libraries. Discovery of a novel antagonist of E. coli dihydrofolate reductase. Int. J.
Mass Spectrom. 238, 77–83 (2004).
conditions known7—yielded detectable product (10) for only 158 of
18. Andrews, C. L., Ziebell, M. R., Nickbarg, E. & Yang, X. in Protein and Peptide Mass
384 reactions (41%). Therefore, screening reaction conditions by using
Spectrometry in Drug Discovery (eds Gross, M. L. et al.) 253−286 (John Wiley &
Sons, Hoboken, 2012).
19. O’Connell, T. N., Ramsay, J., Rieth, S. F., Shapiro, M. J. & Stroh, J. G. Solution-
nanoscale synthesis enabled deeper exploration of chemical space,
with an additional 187 targeted compounds synthesized and assayed
based indirect afnity selection mass spectrometry—a general tool for
high-throughput screening of pharmaceutical compound libraries. Anal. Chem.
86, 7413–7420 (2014).
successfully (Fig. 4). Under a single reaction condition, synthesis was
unsuccessful for large swaths of chemical space and for compounds on
the fringes of the space that we studied (Fig. 4a). NanoSAR identified 20. Annis, D. A. et al. A general technique to rank protein-ligand binding afnities
and determine allosteric versus direct binding site competition in compound
21 compounds from this ‘dark’ space that bound to CHK1 (Extended
mixtures. J. Am. Chem. Soc. 126, 15495–15503 (2004).
Data Fig. 10). Exemplary compounds were resynthesized at a 50-μmol
21. Cuozzo, J. W. et al. Discovery of a potent BTK inhibitor with a novel binding
scale, purified and tested in a functional assay, whereupon several prod-
mode by using parallel selections with a DNA-encoded chemical library.
ChemBioChem 18, 864–871 (2017).
22. Schneider, M. et al. Big data from pharmaceutical patents: a computational
analysis of medicinal chemists’ bread and butter. J. Med. Chem. 59, 4385–4402
(2016).
23. Brown, D. G. & Boström, J. Analysis of past and present synthetic
methodologies on medicinal chemistry: where have all the new reactions gone?
J. Med. Chem. 59, 4443–4458 (2016).
24. Aronov, A. M. et al. Flipped out: structure-guided design of selective
pyrazolylpyrrole ERK inhibitors. J. Med. Chem. 50, 1280–1287 (2007).
25. Bruno, N. C., Tudge, M. T. & Buchwald, S. L. Design and preparation of new
ucts (such as 42–45) displayed functional inhibition of CHK1 at low,
nanomolar concentrations (Fig. 4c). As can be seen from Fig. 3, a large
number of reactions produce no observable product and a large num-
ber of compounds have no affinity to the protein target, which, in our
experience, is typical of drug discovery. Machine-learning algorithms
may one day facilitate the navigation of these problems29,30; however,
locating bioactive molecules and conditions for their synthesis for now
remains an experimental science, which NanoSAR is well positioned
palladium precatalysts for C–C and C–N cross-coupling reactions. Chem. Sci. 4,
916–920 (2013).
26. Anderson, D. R. et al. Pyrrolopyridine inhibitors of mitogen-activated protein
to accelerate.
kinase-activated protein kinase 2 (MK-2). J. Med. Chem. 50, 2647–2654
Data availability
(2007).
The data that support the findings of this study are available within the paper
27. Huang, X. et al. Structure-based design and optimization of 2-aminothiazole-4-
and its Supplementary Information. Source Data for Figs. 2–4 and Extended Data
Figs. 2–6, 8–10 are available with the online version of the paper.
carboxamide as a new class of CHK1 inhibitors. Bioorg. Med. Chem. Lett. 23,
2590–2594 (2013).
N A t U r e | www.nature.com/nature
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

Letter reSeArCH
28. Buitrago Santanilla, A. et al. P2Et phosphazene: a mild, functional group tolerant
base for soluble, room temperature Pd-catalyzed C–N, C–O, and C–C
cross-coupling reactions. Org. Lett. 17, 3370–3373 (2015).
29. Schneider, P. & Schneider, G. De novo design at the edge of chaos. J. Med. Chem.
59, 4077–4086 (2016).
Author contributions N.J.G., P.J.D. and T.C. conceived the study and wrote
the manuscript. N.J.G., B.S., P.J.C., P.J.D. and T. C. designed and executed the
experiments and analysed the data. C.L.A. and M.P.R. developed the protein
titration method. T.C. supervised the work.
30. Ahneman, D. T., Estrada, J. G., Lin, S., Dreher, S. D. & Doyle, A. G. Predicting
reaction performance in C–N cross coupling using machine learning. Science
360, 186–190 (2018).
Competing interests The authors declare no competing interests.
Additional information
Extended data is available for this paper at https://doi.org/10.1038/s41586-
018-0056-8.
Supplementary information is available for this paper at https://doi.
org/10.1038/s41586-018-0056-8.
Reprints and permissions information is available at http://www.nature.com/
reprints.
Acknowledgements We are grateful to L. Nogle, D. Smith and M. Pietrafitta
(MSD) for assistance with compound purification, S. Bano and X. Bu (MSD)
for measuring residual palladium concentrations, A. M. Norris (MSD) for
assistance with graphics and Z. Gu (DFKZ German Cancer Research Center) for
suggestions for data visualization. N.J.G. was supported by an MRL Postdoctoral
Research Fellowship.
Correspondence and requests for materials should be addressed to T.C.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
Reviewer information Nature thanks J. Janey, D. Young and the other
anonymous reviewer(s) for their contribution to the peer review of this work.
N A t U r e | www.nature.com/nature
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

reSeArCH Letter
Extended Data Fig. 1 | Additional details of the nanoscale synthesis
and affinity ranking workflow. a, Reaction solutions are dosed using a
TTP mosquito liquid-handling robot that handles increments of 20 nl.
b, c, UPLC–MS analysis confirms the presence of products from crude
reactions (b), which is visualized in heat maps (c). d, For the ASMS assay,
productive reactions are mass-encoded, pooled and incubated with a
protein of interest. e, f, Elution through a size-exclusion column separates
protein-bound compounds from unbound compounds (e), and binders
can be observed by high-performance liquid chromatography–mass
spectrometry (HPLC–MS) following denaturation of the protein–ligand
complex (f). g, Decreasing the concentration of protein in the ASMS assay
increases the competition for binding so that ligands can be categorized as
high, medium or low affinity. h, Scale-up of compounds and purification
by using a traditional approach. i, Measurement of IC50 in a functional
assay enables comparison of the new method to existing assay technology.
See Supplementary Information for additional details.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

Letter reSeArCH
Extended Data Fig. 2 | Structures, yields, Ki values and ASMS
affinity (magenta) or high affinity (yellow), as determined by the ASMS
assay. rxn, reaction. c, Comparison of the results of the ASMS affinity-
ranking assay (abscissa) to the reported ERK2 Ki values (ordinate). Both
datasets are from purified product samples produced on a 50-μmol scale.
d, Comparison of results of ASMS affinity ranking from crude reaction
mixtures generated on a 100-nmol scale with those from purified product
samples generated on a 50-μmol scale. Points are coloured by Ki, and
jittering was applied to this categorical data to reveal overlapping data.
e, Product structures with reaction conditions used, isolated yields and Ki
values. See Supplementary Information for additional details.
comparison plots for the ERK2 library. a, A library of 19 unique ERK2
inhibitors (2, 3, A1–A17) was produced by coupling 1 to diverse amines
under a single reaction condition (HATU, iPr2NEt, NMP; see Fig. 2a).
b, Crude nanoscale reactions were submitted to the ASMS assay, with
affinity ranking determined by reducing the concentration of ERK2 to
increase the competition for binding. The lowest ERK2 concentration
at which the mass signal of the compound was still observed is shown
on the abscissa. The Ki value reported24 from purified product samples
generated on a 500-times-larger reaction scale is shown on the ordinate.
The colour shading groups compounds into low affinity (purple), modest
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

reSeArCH Letter
Extended Data Fig. 3 | Structures, yields, IC50 values and ASMS
comparison plots for the MK2 Library. a, A library of 18 unique MK2
inhibitors (5, 6, B1–B16) was produced by coupling 4 to diverse boronates
under eight reaction conditions (see Fig. 2b). b–e, As in Extended Data
Fig. 2, but for MK2 and IC50 values instead of ERK2 and Ki values.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

Letter reSeArCH
Extended Data Fig. 4 | Structures, yields, IC50 values and ASMS
comparison plots for the CHK1 Library. a, A library of 20 unique CHK1
inhibitors (8, 9, C1–C18) was produced by coupling 7 to diverse boronates
and amines under eight or nine reaction conditions (see Fig. 2c). b–e, As
in Extended Data Fig. 2, but for CHK1 and IC50 values instead of ERK2
and Ki values.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

reSeArCH Letter
Extended Data Fig. 5 | ASMS comparison plots for different reaction
workup protocols and residual levels of palladium from the CHK1
library. Reactions with different workup protocols were submitted
to the ASMS assay, with affinity ranking determined by reducing the
concentration of CHK1 to increase the competition for binding. The
axes of the plots are as in Extended Data Fig. 4b, c. The reaction and
workup protocols are as follows: a, nanoscale reactions submitted to
the affinity-ranking assay with no purification; b, nanoscale reactions
submitted to the affinity-ranking assay following treatment with SiliaMetS
dimercaptotriazine (DMT) resin; and c, 50-mmol-scale reactions in
which the product samples were purified by reverse-phase HPLC before
submission to the affinity-ranking assay. See Supplementary Information
for additional details.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

Letter reSeArCH
Extended Data Fig. 6 | Catalyst–base survey. See Fig. 3. a, Diverse model
nucleophiles (D1–D17) were screened against combinations of catalysts
(11–26) and bases (27–34) in NMP, DMSO or DMF solvent. b, The details
beside the heat maps show the mapping of nucleophiles, solvents, catalysts
and bases. NA, no catalyst used. Reactions were run on a 100-nmol scale
and analysed by UPLC–MS for conversion to products of the form of 10
compared to a biphenyl internal standard. Productive reaction conditions
from this screen were selected for use in the subsequent library synthesis
campaign. aD1 is compound C2 in Fig. 2c; bD2 is compound C3 in Fig. 2c
and 43 in Fig. 4; cD3 is compound C5 in Fig. 2c; dD4 is compound C6 in
Fig. 2c; eD5 is compound 8 in Fig. 2c; fprepared from the boronic acid.
See Supplementary Information for additional details.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

reSeArCH Letter
Extended Data Fig. 7 | Exemplary compounds from the synthesis of
the library. See Fig. 3. Structures for 62 diverse coupling products (of
the form of 10) are shown, which were selected from the 345 productive
reactions identified in a library synthesis campaign targeting 384 products
(Fig. 3). Diverse nucleophiles were coupled to 7 using the four reaction
conditions selected in Fig. 3 and Extended Data Fig. 6. See Supplementary
Information for additional details.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

Letter reSeArCH
Extended Data Fig. 8 | Comparison of the results of affinity ranking
with IC50 values for inhibition of CHK1 functional activity. The 62
exemplary compounds (Extended Data Fig. 6) were selected from 345 that
were submitted to affinity ranking (Fig. 3). Crude nanoscale reactions
were submitted to the ASMS assay, with affinity ranking determined
by reducing the concentration of CHK1 to increase the competition for
binding. The axes of the plots are as in Extended Data Fig. 4b, c. Points
are coloured by the IC50 value for inhibition of CHK1 function (as in
Extended Data Fig. 4d). See Supplementary Information for additional
details.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

reSeArCH Letter
Extended Data Fig. 9 | Dose response curves for exemplary compounds.
The compounds shown are 35–41 and 43. The inhibition of CHK1
functional activity by diverse coupling products (10) from the reaction
of 7 with nucleophiles under various reaction conditions is shown.
Isolated yields were achieved when the reaction conditions shown were
used. The dose response curves and IC50 values shown were measured
on purified product samples generated on a 50-μmol scale and could be
predicted from affinity-ranking results of crude nanoscale reactions as
shown in Extended Data Fig. 8. See Supplementary Information for dose
response curves of additional compounds.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

Letter reSeArCH
Extended Data Fig. 10 | Reaction metrics and chemical ‘dark’ space.
a, Histogram of reaction performance for 384 diverse coupling reactions
with bromide 7 using a single reaction condition (10 mol% tBuXPhos
Pd G3, P2Et, NMP). Only 158 of the 384 targeted products were
observed by mass spectrometry and the majority of the reactions failed
(0% conversion to product). b, Histogram of reaction performance for
384 diverse coupling reactions with bromide 7 using the best of four
reaction conditions, as described in Extended Data Figs. 6 and 7. Of the
384 targeted products, 345 were observed by mass spectrometry, with a
more even distribution of reaction yields and the majority of reactions
succeeding (100% conversion to product). c, Principal-component (PC)
analysis of chemical ‘dark’ space, with each dot representing a compound
that was not formed under a single reaction condition (0% yield) but that
had affinity to CHK1. By using the best of four reaction conditions, 187
additional products were produced and assayed that bound to CHK1;
the colour of the dots reflects the affinity ranking of the compounds.
The boundaries of this space are identical to those depicted in Fig. 4a, in
which purple shading highlights additional regions where the majority of
reactions failed (0% yield). Areas where dots are shown in purple shaded
regions depict products with affinity to CHK1 that were formed in 0%
yield in a but in more than 1% yield in b. d, Potent CHK1 inhibitors that
were produced in 0% yield under a single condition (10 mol% tBuXPhos
Pd G3, P2Et, NMP), but in modest to good yields following reaction-
condition screening.
© 2018 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.