ARTICLE

https://doi.org/10.1038/s42003-023-05334-8 OPEN

Deep learning driven de novo drug design based on

gastric proton pump structures
Kazuhiro Abe 1,2,3 ✉, Mami Ozako2, Miki Inukai2, Yoe Matsuyuki2, Shinnosuke Kitayama2, Chisato Kanai4,
Chiaki Nagai2, Chai C. Gopalasingam 5, Christoph Gerle 5, Hideki Shigematsu 6, Nariyoshi Umekubo 2,
Satoshi Yokoshima 2 ✉ & Atsushi Yoshimori 7 ✉

Existing drugs often suffer in their effectiveness due to detrimental side effects, low binding
affinity or pharmacokinetic problems. This may be overcome by the development of distinct
1234567890():,;

compounds. Here, we exploit the rich structural basis of drug-bound gastric proton pump to
develop compounds with strong inhibitory potency, employing a combinatorial approach
utilizing deep generative models for de novo drug design with organic synthesis and cryo-EM
structural analysis. Candidate compounds that satisfy pharmacophores defined in the drug-
bound proton pump structures, were designed in silico utilizing our deep generative models, a
workflow termed Deep Quartet. Several candidates were synthesized and screened according
to their inhibition potencies in vitro, and their binding poses were in turn identified by cryo-
EM. Structures reaching up to 2.10 Å resolution allowed us to evaluate and re-design com-
pound structures, heralding the most potent compound in this study, DQ-18 (N-methyl-4-
((2-(benzyloxy)-5-chlorobenzyl)oxy)benzylamine), which shows a Ki value of 47.6 nM.
Further high-resolution cryo-EM analysis at 2.08 Å resolution unambiguously determined the
DQ-18 binding pose. Our integrated approach offers a framework for structure-based de novo
drug development based on the desired pharmacophores within the protein structure.

1 Cellular and Structural Physiology Institute, Nagoya University, Nagoya, Aichi 464-8601, Japan. 2 Graduate School of Pharmaceutical Sciences, Nagoya

University, Nagoya, Aichi 464-8601, Japan. 3 Center for One Medicine Innovative Translational Research (COMIT), Nagoya University, Nagoya, Aichi 464-8601,
Japan. 4 INTAGE Healthcare, Inc., 3-5-7, Kawaramachi Chuo-ku, Osaka 541-0048, Japan. 5 RIKEN SPring-8 Center, Kouto, Sayo-gun, Hyogo 679-5148, Japan.
6 Japan Synchrotron Radiation Research Institute (JASRI), SPring-8, 1-1-1 Kouto, Sayo, Hyogo 679-5148, Japan. 7 Institute for Theoretical Medicine, Inc., 26-1,

Muraoka-Higashi 2-chome, Fujisawa, Kanagawa 251-0012, Japan. ✉email: [email protected]; [email protected]; [email protected]

COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio 1

ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8

T
o better meet medical needs, improvement of existing for clinical use in 2015 in Japan), have been developed (Supple-
drugs’ efficacy is highly desired due to problems caused by mentary Fig. 1). These P-CABs are proving to be successful,
harmful side effects, poor pharmacokinetics or low binding providing rapid and reliable cures for acid-related gastrointestinal
affinity. Designing new compounds by derivation, sometimes diseases. On the other hand, the relationship between long-term
prevented by patent restriction, and phenotype screening often administration of PPI or P-CAB and their effects on intestinal
leads to unsatisfactory results. Recent advances in structural microflora21, interstitial nephritis22 and gastric cancer23 have also
biology eases access to drug bound protein structures. However, received attention. Therefore, further pharmacokinetic improve-
even when available, structural data on the drug-target interaction ment and increased binding affinity is desired to reduce the
remains hard to exploit for the design of novel compounds. To required drug dosage. Furthermore, the availability of alternative
this end, establishing a framework that enables the production of compounds is expected to broaden the treatment options and
a compound with a radically different chemical skeleton and expand its clinical use. Because all the above drugs have been
effectiveness is required, which we attempt to address here by developed based on phenotypic screening, there is room for
exploiting the accumulation of structural and functional infor- further improvement, specifically, the development of novel drugs
mation of drug-bound gastric proton pump1,2. using structure-based rational design.
Painful symptoms of acid-related gastric diseases such as peptic So far we have reported X-ray crystal and cryo-EM derived
ulcers or gastroesophageal reflux disease are associated with dis- structures of H+,K+-ATPase complexed with seven different P-
orders of the gastrointestinal tract3. Gastric mucosal injury due to CABs; SCH280802, vonoprazan2, BYK9924, tegoprazan1,
continuous use of nonsteroidal anti-inflammatory drugs PF037165561, soraprazan1 and revaprazan1 (Supplementary
(NSAIDs)4 or gastrin-producing tumors may also cause peptic Fig. 1). All P-CABs bind to the luminal-facing cavity leading to
ulcers5. Current therapies to treat these conditions either prevent the cation-binding site in the luminal-open E2P state, and phy-
the stimulation of parietal cells by antagonizing histamine H2 sically block K+-entry to the cation-binding site, hence prevent-
receptors6 or inhibit the final step in acid production by targeting ing luminal gate closure which is induced by K+-binding to the
the gastric proton pump, H+,K+-ATPase. Eradication of Heli- cation-binding site. The binding mode of some of P-CABs are,
cobacter pylori, the main cause of gastric ulcers and gastric cancer, however, significantly distinct; while SCH28080 (Fig. 1b, c) and
has been accomplished by the suppression of gastric acid secre- its related compounds bind around the entrance of the luminal
tion in combination with antibiotic treatment7. The molecular cavity, vonoprazan (Fig. 1d) and to some extent revaprazan binds
targets of acid suppressants, H+,K+-ATPase, is a gastric proton more deeply toward the cation-binding site, indicating there are
pump that mediates H+ export in exchange for luminal K+ across multiple pharmacophores in the P-CAB binding site. To date,
the parietal cell membrane accompanied by ATP hydrolysis8,9. however, not a single P-CAB satisfies all these pharmacophores.
Like other P-type ATPases10,11, vectorial cation transport is Therefore, if a compound that satisfies the greatest common
accomplished by the cyclical conversion of the pump consisting denominator of these pharmacophores can be created, it is
of four cornerstone states; E1 − E1P − E2P − E2, according to the expected to be highly potent in gastric acid suppression.
so-called Post-Albers scheme12, which describes alternating Deep generative models (DGMs) have been successfully
access and affinity for the H+ and its counterion K+. applied in various fields, including image processing25, language
Cytoplasmic-facing E1 and luminal-facing E2P states show high translation26 and the generation of chemical structures for drug
affinity for H+ and K+, respectively13. H+,K+-ATPase consists of development27–30. Despite its remarkable advances, most of the
two subunits. The catalytic α-subunit (100 kDa) comprises 10 previous studies utilizing DGMs for compound generation
transmembrane (TM) helices in which the cation and inhibitor focused on optimizing molecular properties such as logP (lipo-
binding sites are located and three cytoplasmic domains (the philicity), QED (Quantitative Estimate Drug-likeness score)31
nucleotide-binding (N), phosphorylation (P), and actuator (A) and SA (Synthetic Accessibility) score calculated on a 2D repre-
domains) executing ATP hydrolysis and autophosphorylation. An sentation of the molecule32, and the application of DGMs to the
accessory β-subunit (35 kDa) has a single TM helix with a short actual drug discovery for the specific target is surprisingly limited
N-terminal tail and an ectodomain with six N-linked glycosyla- so far28,30. Given the potential of DGMs to optimize compound
tion sites, and is required for the folding of the complex and chemical structures based on a biological target derived phar-
membrane trafficking14. macophore 3D coordinates, the appropriate application of DGMs
Proton pump inhibitors (PPIs) such as omeprazole have been is expected to significantly accelerate drug discovery.
utilized for acid suppression. The PPI drug itself is a prodrug, Here, we demonstrate the approach by focusing on improving
requiring acid activation to irreversibly inhibit H+,K+-ATPase by gastric proton pump blockers, where our deep learning-driven
forming a covalent bond with its Cys813 residue15. However, drug design based on the desired pharmacophores in the target
given its relatively short plasma half-life and requirement for an protein structure (Fig. 1a) was capable of suggesting useful
acidic pH to convert the prodrug to the active compound, con- compounds with distinct chemical skeletons. Subsequent rounds
siderable effort has been expended to develop different types of of candidate selection, synthesis, in vitro drug screening and
H+,K+-ATPase inhibitors. The K+-competitive acid blockers high-resolution cryo-EM structure analysis allows for successful
(P-CABs) differ from the PPIs in that they are not dependent on de novo drug design resulting in a greatly facilitated process of
acid activation, are rather stable in the acidic canaliculus, and drug discovery.
bind directly to the proton pump, thereby providing a more rapid
onset and better inhibition of acid secretion3. The binding of
omeprazole and P-CABs is mutually exclusive, indicating these Results and discussion
drugs share an overlapping binding site16. They are currently in De novo design of P-CABs based on the desired pharmaco-
clinical use in some Asian countries. Although the prototypic P- phores. Based on the reported structures of the H+,K+-ATPase
CAB, the imidazo[1,2-a]pyridine derivative SCH2808017, is complexed with different P-CABs, we tried to define the vital
unsuitable for clinical use due to its hepatotoxicity, its benzimi- pharmacophores for P-CAB binding. SCH28080, known as an
dazole derivative tegoprazan18 was approved in 2018 in South ancestor compound of P-CABs, and its related compounds
Korea. Besides SCH28080-related compounds, chemically distinct (BYK99, PF-03716556 and soraprazan) or a drug (tegoprazan)
pyrimidine-based revaprazan19 (approved for clinical use in 2007 (Supplementary Fig. 1), share a similar binding site at the
in South Korea) and pyrrole derivative vonoprazan20 (approved entrance of the luminal cavity leading to the cation-binding site

2 COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio

COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8 ARTICLE

Fig. 1 De novo drug design based on the desired pharmacophores. a Workflow of “Deep Quartet”. See main text and “Methods” for details. b Crystal
structure of the gastric H+,K+-ATPase complexed with SCH28080 (PDB ID: 5YLV, gray) in ribbon representation, viewed parallel to the membrane plane
with its luminal-side facing upwards. c, d Close-up view of the P-CAB-binding site (indicated as a red box in b). Clipped cross section of the luminal-facing
conduit (surface) of SCH28080- (gray) and vonoprazan- (PDB ID: 5YLU, wheat) bound forms from the viewpoint similar to (b). Positions of P-CAB binding
are indicated as purple or pink surface for SCH28080 or vonoprazan, respectively. e Defined pharmacophore features (I–IV, see text for details) on the
luminal-facing cavity of the H+,K+-ATPase structure. Navy and yellow ovals, and blue triangle indicates pharmacophore features with aromatic,
hydrophobic or cationic properties, respectively. f Binding poses of candidate compounds in the template structure (5YLV) calculated by Deep Quartet.
g Same as (f) but with vonoprazan-bound form used as a template (5YLU). The most constrained positions of the lumina-facing conduit in each structure
are indicated as dotted lines and blue arrows, with accompanying distance in Å. Two distinct binding modes of candidates from vonoprazan-bound form
are displayed separately in (h) and (i). In (c, d, f–i), TM2 is omitted for clarity.

where Rb+ is bound as a congener of K+ in the SCH28080-bound vonoprazan binds close to the K+-binding site where three
form (Fig. 1b, c, e). The imidazo[1,2-a]pyridine ring of SCH28080 glutamates (Glu343, Glu795 and Glu820) are located.
is bound to Tyr799 via π-π interaction33, and its benzyl group (II) A hydrophobic pharmacophore feature between TM1 and
makes hydrophobic contacts with residues in TM1 (Ala123), TM4, near the residue Ala123: the terminal hydrophobic
TM2 (Asn138) and TM4 (Val331, Ala335)34. On the other hand, group of SCH28080-related P-CABs binds at this position;
vonoprazan has a unique chemical structure and harbors thus far, the Ala123Val mutant shows significantly reduced affinity
to the best of our knowledge, the most potent in vitro inhibition for SCH28080 and related compounds, but almost no effect
activity amongst P-CAB. Its methylamino group reaches near the for vonoprazan2.
cation-binding site located deep in the conduit thus preventing (III) An aromatic pharmacophore feature close to Tyr799: the
K+-binding, while its pyridine ring keeps Tyr799 at the luminal aromatic group of all P-CABs interacts with the side chain
entrance (Fig. 1d, e)2. A recently reported revaprazan bound of Tyr799, and mutation of this residue (Tyr799Ala)
structure shows the intermediate binding pose between severely reduces the apparent affinity of all P-CABs1,2,34.
SCH28080-type compounds and vonoprazan; while the pyr- (IV) A hydrophobic pharmacophore feature connecting I and
imidine and fluorophenyl rings of revaprazan overlap the binding III: the conduit connecting cation-binding site and luminal
pose of SCH28080, its tetrahydroisoquinoline moiety is accom- solution is mostly hydrophobic; the aromatic five-
modated in the middle of conduit and faces toward the cation- membered ring of vonoprazan, and the hydrophobic
binding site1. tetrahydroisoquinoline moiety of revaprazan occupy this
In order to design unique P-CABs by “Deep Quartet” (DQ), a position.
de novo design workflow for generating molecules with a desired These four pharmacophore features (I–IV) were set alongside
pharmacophore (Fig. 1a), we defined multiple features for the crystal structure of SCH28080- (PDB ID: 5YLV) or
pharmacophores, based on the above-described structural vonoprazan-bound (PDB ID: 5YLU) forms, and we subsequently
information (Fig. 1e and Supplementary Fig. 2). generated candidate compound structures by DQ. After iterative
(I) A cationic and hydrogen bond donor pharmacophore processing, compounds having a methylamino group were
feature near the cation-binding site: the amino group of extracted. Although the two crystal structures used as templates

COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio 3

ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8

are almost identical (RMSD for Cα atoms = 0.755 Å), at least DQ-02. Because alkyne backbone structures were generated using
same conformation, to our surprise, generated candidate the SCH28080-bound structure as a template, we compared the
compounds from each template structures are distinct. Char- dimension of its P-CAB binding pocket with another template,
acteristic compound structures with an alkyne moiety were the vonoprazan-bound form (Fig. 1f, g). The comparison revealed
generated when SCH28080-bound structure is used as a template that the width of the most constricted portion of the luminal
(Fig. 1f and Supplementary Fig. 2 Type A). Likewise, when we set conduit in SCH28080-bound form (5.6 Å) is narrower than that
two aromatic rings at pharmacophore feature III position (akin to in the vonoprazan-bound form (7.2 Å), which may be one reason
the heterocyclic ring structure of the imidazo[1,2-a]pyridine ring for the production of the alkyne-connecting candidates by DQ.
of SCH28080 is in mind) of the SCH28080-bound form, similarly Given the relatively low affinity and instable binding mode of
alkyne-connecting, but incorporating heteroaromatic ring struc- DQ-02, we halted further improvement of DQ-02 and other
tures were generated (Fig. 1f and Supplementary Fig. 2 Type B). alkyne-backbone compounds.
In contrast, aromatic-rich compounds were generated from the
vonoprazan-bound structure (Fig. 1g–i and Supplementary Fig. 2
Type C). DQ generated diverse compounds for each pharmaco- DQ-06: satisfying the pharmacophore of H+,K+-ATPase Deep
phore type, among which 71, 10 and 181 compounds were Quartet assisted drug design. In contrast to DQ-02 and other
selected as candidates with pharmacophore scores (see “Meth- alkyne-backbone compounds, another variety of candidates were
ods”) above 0.90 (Supplementary Fig. 3). Synthetic accessibility suggested by DQ, when using the vonoprazan-bound template
scores32 suggest that most of the DQ-generated candidates show structure which has a much wider hydrophobic conduit (Fig. 1g and
mean values of around 3, which is similar to the score for most Supplementary Fig. 2). With reference to these candidates, we
drug-like compounds available for clinical use, thus capable to designed and synthesized a simple compound that has three benzene
synthesize (Supplementary Fig. 3). From the suggested candidates rings connected via ether linkers and has a terminal secondary amine
by DQ, several compounds were selected and synthesized moiety (DQ-06, Supplementary Data 2). To our surprise, DQ-06
(Supplementary Fig. 4), and their inhibition potencies and shows an apparent affinity (IC50,DQ06 = 0.70 ± 0.04 μM) significantly
binding poses are characterized in the following sections. higher than SCH28080 (Fig. 3a), and its inhibition mode is pure-
ly competitive with K+ (Supplementary Fig. 5, Ki,DQ06 = 93.8 nM).
To evaluate the importance of this terminal secondary amine,
Alkyne-backbone of DQ-02 and related compounds. From the systematic modifications were made on the secondary amine
candidate compound structures with an alkyne moiety (Fig. 1f and moiety of DQ-06, generating DQ-10 ~ 12. Comparison of the inhi-
Supplementary Fig. 2) as a reference, we chemically synthesized sev- bition potencies of DQ-06-related compounds, which have a
eral compounds taking into account their synthetic feasibility. Because hydroxy (IC50,DQ10 = 292 ± 67 μM), dimethylamino (IC50,DQ11 =
most of the suggested compounds have various modifications, and 5.57 ± 0.45 μM), or amino (IC50,DQ12 = 1.54 ± 0.07 μM) group,
different functional groups were employed even for similar chemical shows clear structure-activity relationships (SAR), indicating that the
skeletons (Supplementary Fig. 2), we synthesized simple and repre- compounds with a secondary amine moiety showed the best inhi-
sentative compounds after examining these candidates (Supplemen- bition potency (Fig. 3a).
tary Figs. 3 and 4, see also Supplementary Data 2 for details), and then The binding pose of DQ-06 to the H+,K+-ATPase was
evaluated their potency by measuring dose-dependence of ATPase analyzed by using a 2.19 Å resolution cryo-EM structure (Fig. 3b,
activity inhibition using H+,K+-ATPase-enriched membrane frac- Supplementary Fig. 6, Supplementary Movie 2 and Table 1). In
tions (Fig. 2a). Among them, although exhibiting lower apparent contrast to the blurred EM density observed for the DQ-02 bound
affinity than SCH28080 (IC50,SCH = 1.97 ± 0.12 μM), DQ-02 and DQ- to H+,K+-ATPase (Fig. 2c), the EM density clearly defines the
04 inhibit H+,K+-ATPase activity in a dose-dependent manner with binding pose of DQ-06, with several bound water molecules
similar apparent affinities (IC50,DQ02 = 39.9 ± 3.3 μM, IC50,DQ04 = visualized. The initial DQ calculation suggests roughly two
33.9 ± 3.8 μM), suggesting that halogen modification to a benzene different binding poses for DQ-06 related candidates (Fig. 1h, i
ring does not make a large difference for the apparent inhibition and Supplementary Fig. 2); starting from the cationic secondary
affinity. Others showed lower apparent affinity than DQ-02 (IC50s for amine (pharmacophore feature I), ~60% of DQ-06 related
DQ-09: >2000 μM, DQ-14: >120 μM, DQ-15: 49.5 ± 4.9 μM, DQ-16: candidates took a binding pose that connected the pharmaco-
89.4 ± 13.5 μM), and some compounds gave scattered values due to phore features in the order (I)→(IV)→(III)→(II) (Fig. 1h), and
low water solubility. Regardless of the low apparent affinity of DQ-02, others took a different binding pose (I)→(IV)→(II)→(III)
its double reciprocal plot analysis (Supplementary Fig. 5) shows a (Fig. 1i). The binding pose of DQ-06 in the cryo-EM structure
typical competitive inhibition pattern (estimated inhibition constant is unambiguously determined as a single conformation (Fig. 3b),
(K i,DQ02) is 4.65 μM), suggesting that DQ-02 binds to the luminal- which agrees with the former binding pose
facing cavity as other P-CABs do. [(I)→(IV)→(III)→(II)] (Fig. 3c, d, g for simplified schematics).
To determine the DQ-02 binding pose, we performed cryo-EM The cationic secondary amine reaches the cation-binding site, and
analysis of H+,K+-ATPase, bound with DQ-02 and obtained a is located within hydrogen bond distance to Glu795 side chain
structure at 2.10 Å overall resolution (Fig. 2b, Supplementary oxygen (3.0 Å), main chain carbonyl from the Ala339 located on
Fig. 6 and Table 1). Counterintuitively, however, the EM density the unwound portion of TM4 (3.2 Å) and an accompanied water
map corresponding to DQ-02 was poorly defined, and does not molecule (3.1 Å) (Fig. 3c, d), help fulfill the requirement of
account for a single binding pose of the compound (Fig. 2c and “pharmacophore feature I” (Fig. 1e). The SAR of DQ-06-related
Supplementary Movie 1). We therefore modeled two possible compounds shows the importance of the methylamino group
conformations of DQ-02 in the binding cavity. In both binding (Fig. 3a), which is now explained from the structural viewpoint.
poses, the terminal secondary amine moiety reaches close to the The terminal methyl group is in van der Waals contact with the
cation-binding site, as seen in the vonoprazan binding mode. Two surrounding side chains including Glu343 (3.4 Å), Asn792 (3.4 Å)
benzene rings are likely bound to the Tyr799 and Ala123 similar and Glu820 (3.3 Å). The benzene ring III (see Fig. 3a for the
to SCH28080 and related P-CABs. The structure of the alkyne nomenclature of the benzene rings in DQ-06) is located close to
backbone may be too thin to stably bind in a single conformation Tyr799 (3.5 Å) with a nearly parallel relationship, supposedly
to the hydrophobic conduit, and the nearly symmetrical structure interacting via their π electron systems (“pharmacophore
of two benzene rings also allows for multiple binding poses of feature III”). These two portions (pharmacophore features I and

4 COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio

COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8 ARTICLE

Fig. 2 Inhibition potency and the binding pose of DQ-02. a Dose-dependent inhibition of H+,K+-ATPase activity by indicated synthesized compounds
(SCH28080 as a control, gray circles and a dashed line). Data plotted represent each data point from triplicate of three independent measurement at 12
different concentrations of P-CABs, using membrane fractions purified from pig stomach. b EM potential map (colored surface) and cartoon model of the
gastric H+,K+-ATPase complexed with DQ-02 (α-subunit, skyblue; β-subunit, gray; lipids, orange; waters, red). c Close-up view of the DQ-02 binding site
indicated by the red box in (b). Transparent blue surface and black mesh represent EM potential maps in high and low-contour levels, respectively. Only
the EM density around DQ-02 is displayed for low-contour map (mesh). Two possible conformations of DQ-02 (gold and yellow sticks) are shown.

III) are connected by the benzene ring IV, which now occupies derivatives (Fig. 4a and Supplementary Data 2). Compared to the
the space at the hydrophobic conduit (pharmacophore feature original DQ-06, introduction of three chloro groups (DQ-07)
IV) surrounded by side chains of Leu141 (4.2 and 4.3 Å), Ala339 significantly reduced its inhibition potency (IC50,DQ07 = 3.20 ±
(3.5 Å) and Ile816 (3.6 and 3.7 Å) and Cβ of Glu795 (3.6 Å). The 0.20 μM). As for compounds with a single modification on ring
terminal benzene ring II is located near Ala123 (3.6 Å) and C, the apparent affinity of 5-chloro derivative (DQ-18) is
making further hydrophobic contacts with Cys120 (3.9 Å), improved by a factor of two (IC50,DQ18 = 0.31 ± 0.02 μM) relative
Asn138 (3.4 Å) and Ala335 (3.5 Å), thus fixing its azimuthal to DQ-06, while that of 6-chloro derivative (DQ-19) is sig-
position is fixed in a thin, restricted pocket formed between TM1 nificantly reduced (IC50,DQ19 = 3.52 ± 0.22 μM), indicating that
and TM4 (pharmacophore feature III) (Fig. 3e, f). In contrast to the effect is position-specific (Fig. 4a). We also confirmed the K+-
above-described portions that match the defined pharmacophore competitive inhibition mode of DQ-18 (Supplementary Fig. 5,
features, EM density corresponding to the ether linker connecting Ki,DQ18 = 47.6 nM).
benzene rings II and III is relatively weak (Fig. 3b), indicating its The binding pose of DQ-18 was determined by a cryo-EM
flexibility, also seen in the SCH28080 binding mode1,35. reconstruction at 2.08 Å resolution (Supplementary Fig. 6 and
Because of the simple chemical structure of two benzene rings II Table 1), in which the chloro group is clearly seen as a protruded
and III in DQ-06, which are bound to the luminal entrance of the density at the C5 position of benzene ring III (Fig. 4a, b). Due to
binding cavity, there is some vacant space in the binding site (Fig. 3e, the modification by the chloro group, the binding position of the
f). This observation let us consider the possibility that a simple terminal (II) and central (III) benzene rings are offset by 0.6 Å
modification in this portion (ring II or III) of the compound may compared to DQ-06 (Fig. 4c and Supplementary Movie 3), which
improve the apparent affinity, given that the much tighter binding gives DQ-18 a much closer contact to the surface of the binding
mode may enhance van der Waals interaction. pocket (Fig. 4d–h). The high-resolution EM map unambiguously
determined the 5-chloro moiety (Fig. 4b, d, e), which is now in
close contact with the main chain oxygen of Leu811 (2.9 Å)
Chloro derivatives of DQ-06. For further improvement of the presumably interacting via a halogen bond, and the main chain
binding affinity, we synthesized a series of DQ-06 chloro nitrogen of Cys813 (3.3 Å), thus enhanced van der Waals

COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio 5

6
Table 1 Cryo-EM data collection, refinement and validation statistics.

DQ-02 (EMDB-35500) (PDB 8IJV) DQ-06 (EMDB-35501) (PDB 8IJW) DQ-18 (EMDB-35502) (PDB 8IJX) DQ-21 (EMDB-36424) (PDB 8JMN)
Data collection and processing
ARTICLE

Magnification 60,000 60,000 60,000 60,000

Voltage (kV) 300 300 300 300
Electron exposure (e–/Å2) 60 60 60 60
Defocus range (μm) 0.8–1.8 0.8–1.8 0.8–1.8 0.8–1.8
Pixel size (Å) 0.752 0.580 0.752 0.752
Symmetry imposed C1 C1 C1 C1
Initial particle images (no.) 2,832,257 1,564,257 3,387,269 2,230,909
Final particle images (no.) 354,147 168,758 288,130 599,919
Map resolution (Å) 2.10 2.19 2.08 2.26
FSC threshold 0.143 0.143 0.143 0.143
Refinement
Initial model used (PDB) 5ylu 5ylu 5ylu 8IJX
Model resolution (Å) 2.16 2.27 2.15 2.30
FSC threshold 0.5 0.5 0.5 0.5
Map sharpening B factor (Å2) −49.6 −46.5 −47.1 −79.4
Model composition
Non-hydrogen atoms 10,533 10,543 10,545 10,491
Protein residues 1251 1251 1251 1251
Waters 370 379 373 373
Ligands DQ02,Mg2+,5PCW,2CLR,3NAG DQ06,Mg2+,5PCW,2CLR,3NAG DQ18,Mg2+,6PCW,CLR,3NAG DQ21,Mg2+,5PCW,CLR,3NAG
B factors (Å2)
Protein 49.75 52.98 48.68 94.43
Ligand 54.70 58.78 53.09 84.62
Waters 46.13 44.69 44.86 66.10
R.m.s. deviations
Bond lengths (Å) 0.003 0.003 0.003 0.003
Bond angles (°) 0.945 0.796 0.778 0.779
Validation
MolProbity score 1.69 1.78 1.35 1.76
Clashscore 5.79 7.00 4.84 5.76
Poor rotamers (%) 2.92 2.44 1.32 3.29
Ramachandran plot
Favored (%) 97.9 97.5 98.3 97.8
Allowed (%) 2.1 2.5 1.7 2.2
Disallowed (%) 0.0 0.0 0.0 0.0

COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio

COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8
COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8 ARTICLE

interactions can be expected. Due to the 0.6 Å offset of the leading to the water-filled pocket, may instead exclude this water
binding position of DQ-18 relative to that of DQ-06, the benzene molecule. Alternatively, it would be possible to try different
ring II is in much closer contact to amino acids surrounding compound skeletons; e.g., replacement of the benzene ring III to a
pharmacophore feature II position near Ala123, and thus the 5-membered aromatic ring or a heterocyclic ring like imi-
number of amino acids located within 3.9 Å is increased at this dazo[1,2-a]pyridine of SCH28080 (Supplementary Fig. 1) that
position (Fig. 4f, five residues for DQ-06 → eight for DQ-18). In could change the direction of the side chain, whilst preserving the
contrast to the relatively large difference in the binding pose π electron donation system bound to Tyr799.
between DQ-06 and DQ-18 at the luminal side of the cavity, In contrast to the clear EM densities for the most of the
the effect for cationic secondary amine (I) and connecting functional groups of DQ-06 and DQ-18, the EM densities
benzene ring (IV) is limited (Fig. 4c). Similar to the case for DQ- corresponding to their ether linkers that connecting ring II and
06, the secondary amine moiety of DQ-18 also reaches the cation- III are weak, indicating this portion is mutually flexible for both
binding site, and is surrounded by three oxygen atoms within compounds. This is in good agreement with the conclusion from
3.5 Å distance. Likewise, the terminal methyl group on the amino a previously reported systematic SAR of SCH28080 derivatives
group is accommodated with amino acids located at the cation- with fixed ring analogs (BYK99 and soraprazan, Supplementary
binding site, including main chain oxygen of Val341 (3.6 Å) and Fig. 1)35, which show more than 25 and 7 times higher apparent
side chains of Asn792 (3.3 Å), Glu795 (3.4 Å) and Glu820 (3.6 Å), affinity than SCH28080, respectively1,34. We also reported that
within van der Waals distance (Fig. 4e, f). Therefore, based on the the dihedral angle at the connecting oxygen atom of SCH28080 in
high-resolution cryo-EM structure, we conclude the rationale for solution distributed three metastable positions in the molecular
the improved affinity of DQ-18 to be most likely due to tighter dynamics simulation, indicating its conformational freedom1.
packing of the terminal and central benzene rings (ring II and III) Such flexibility in the relative orientation of two benzene rings II
into the binding pocket, enhancing hydrophobic interaction and III connected by an ether linker is also expected for the DQ-
between DQ-18 and H+,K+-ATPase. 06-related compounds. By fixing the linker structure and thereby
This hypothesis is confirmed by structural and functional restricting the conformation that DQ-06-related compounds can
analysis of DQ-21, a chloro-modified DQ-06 at para-position of adopt, a dramatic improvement in their apparent affinities could
ring II (Fig. 5 and Supplementary Data 2). DQ-18 and DQ-21 have be anticipated. However, we cannot exclude the possibility that
a chloro group on just opposite sides of ring II and ring III, the flexibility of the linker region helps DQ-06 and DQ-18
respectively, and the spatial volume in this part of the compound is binding, as chemical backbone of these compounds is different
the same. Therefore, if the improvement of apparent affinity of DQ- from that of SCH28080.
18 would be due to the tighter packing of DQ-18 compared to DQ- In this manuscript, all the ATPase measurements for the
06, similar improvement would be observed for DQ-21. As shown determination of apparent affinities of the compounds were
in Fig. 5a, DQ-21 shows an apparent affinity (IC50,DQ21 = 0.28 ± performed at neutral (pH 7) condition. In the case of SCH28080,
0.02 μM) similar to that of DQ-18 (IC50,DQ18 = 0.31 ± 0.02 μM). As because of its pKa value of 5.6 for imidazo[1,2-a]pyridine amine,
we expected, cryo-EM analysis at 2.26 Å reveals that the binding its apparent affinity differs around neutral (Ki = 66 nM at pH
pose of DQ-21 is 1.2 Å offset toward the TM6 side (Fig. 5b, c), and 7.35) and weakly acidic (Ki = 20 nM at pH 6.11) conditions36.
the chloro group at ring II is embedded in the cleft formed between However, expected pKa value of secondary amine groups of DQ-
TM1 and TM4, close to Ala123 (Fig. 5d, e). We have thus succeeded related compounds are in the range of 9-11, indicating they are
in further increasing the affinity of the DQ-suggested compound by mostly protonated at neutral pH, and also at highly acidic
its tailored modification based on the high-resolution structural condition in the stomach. According to Henderson-Hasselbalch
information. equation, ratio of protonated : free amine is expected to be 1:25 at
pH 7 and 1:10 at pH 6.6 for SCH28080 with pKa of 5.6. In
contrast, it is expected to be 100:1 ~ 10,000:1 for the secondary
Potential further improvements of compound binding. High- amine of DQ-related compounds with pKa of 9–11, and this ratio
resolution cryo-EM structures of DQ-06, DQ-18 and DQ-21 does not largely change around neutral (pH 7), weakly acidic (for
bound forms allowed us to visualize several water molecules in example, pH 6.0) and even strongly acidic condition in the
their direct vicinity. This includes a water located in the pocket stomach, thus the apparent affinities of DQ-related compounds
surrounded by Pro798, Pro810 and Leu811, stabilized by the unlikely affected by the in vitro solution pH.
hydrogen-bond network connecting main chain carbonyl of
Glu795 and main chain amide of Tyr799 in DQ-06, DQ-18 and
DQ-21 bound forms (Figs. 3c–e, 4d, e, g and 5b–d). In the A unique framework for de novo drug generation. Here, a
SCH28080-bound structure, this water-filled pocket is occupied combination of AI-driven compound design, chemical synthesis
with its cyanomethyl group and thus excluded from the binding and cryo-EM analysis of drug bound structures underpinned the
pocket (c.f., Fig. 1d, see ref. 2), which would give a favorable development of P-CAB candidates that have de novo chemical
increase in the entropy of the whole system. In fact, an SCH28080 structures (Fig. 1a). Recent advances in structural analysis by
derivative without a cyanomethyl group shows more than 20- single particle cryo-EM37,38 renders drug-bound protein struc-
times reduced affinity34. Given the C6 position of the benzene tures significantly more accessible and reliable than previously.
ring III of DQ-06 (Fig. 3a, c, e) is the closest to the aforemen- This enables the determination of pharmacophores, relevant
tioned water molecule in the pocket, it was expected that the structure-based modifications, and iterative repetition of these
6-chlorobenzyl derivative (DQ-19) may exclude it, and resulting cycles to improve affinity, potencies and other pharmacokinetic
in an improvement of its apparent affinity. Thus, a question arises parameters. To address unforeseen side effects and expanding
as to why DQ-19 does not improve, rather reduces, the binding medical needs, the development of alternative drugs with distinct
affinity compared to DQ-06 (Fig. 4a). We speculate that a bulky chemical skeletons is highly desired. However, when creating new
chlorine modification at C6 position of the ring III may interfere compounds based on the original drug-bound structure, it is
with the oxygen atom of the ether linker connecting two benzene often difficult to generate a truly different chemical skeleton
rings III and IV, and prevents the adoption of this particular derived far from the original drug structures. As highlighted in
binding pose. Modification of C6 position of ring III with a our case, even if several pharmacophore features have been
smaller atom, or a much longer functional group at C5 position determined on the protein structure, it is not easy to manually

COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio 7

ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8

create the optimal chemical structure to link them with. Deep like” chemical structures (Fig. 1a and Supplementary Fig. 2).
Quartet, a deep learning based de novo design workflow, gen- However, in some cases DQ generates too many candidate
erates chemical structures that satisfy the desired pharmaco- compounds, and some of them are not trivial to synthesize, or,
phores on the protein structure29,30. Since the software “learns” compounds themselves are instable in aqueous solution. We
chemical structures from the database ChEMBL39 (https://www. therefore selected candidates “heuristically”, based on the
ebi.ac.uk/chembl/) in which more than 2-million bioactive knowledge of organic chemistry and previous functional analysis
molecules are stored, generated candidates have mostly “drug- of H+,K+-ATPase with P-CABs, and started with simple

8 COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio

COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8 ARTICLE

Fig. 3 Inhibition potency and the binding pose of DQ-06. a Dose-dependent inhibition of H+,K+-ATPase activity by indicated compounds as shown in
Fig. 2a. Data plotted represent each data point from triplicate of three independent measurement at 12 different concentrations of P-CABs. Three benzene
rings in DQ-06 are attributed as illustrated in the figure, which corresponds to the defined pharmacophore features (II)–(IV). b EM potential map (mesh)
and the atomic model of the gastric H+,K+-ATPase complexed with DQ-06 (green sticks). Shown is a close-up view of the DQ-06 binding site as in Fig. 2c.
c, d Molecular interactions between H+,K+-ATPase and DQ-06 in stick representation. Hydrophilic and hydrophobic atoms of DQ-06 within 3.5 and 4.0 Å
distance from amino acids of H+,K+-ATPase are connected by orange and gray dotted lines, respectively. Panels are viewed from luminal side (c), or
parallel to the membrane plane with luminal-side up (d). e, f Clipped cross sections of the DQ-06 binding site from the viewpoints approximately similar to
(c) and (d). Molecular surface (gray) of H+,K+-ATPase shows the dimension of the binding site in which DQ-06 is accommodated (green stick with
transparent van der Waals spheres). Except for TM2, which is removed from the figure for clarity, TM helices and some of the key amino acids are shown
in ribbon and stick representations. g A schematic 2D representation of DQ-06 binding pose. Hydrophobic residues that are located within 3.9 Å from DQ-
06 are shown, and those within 3.5 Å are highlighted as red. Expected polar interactions within 3.5 Å are indicated as orange dotted lines.

compounds, like DQ-02 (Fig. 2) or DQ-06 (Fig. 3), for stream- Agent network were stored. In tabu list approach, scaffolds of the
lined structure-activity studies. This strategy may be particularly structures generated from each Agent network are appended to
useful to determine the direction of which compounds are pro- the tabu list. The scaffolds were calculated using MurckoScaffold
mising in the very early stages of drug development. function implemented in RDKit (https://www.rdkit.org)42. When
Pharmacophore-based de novo design of the chemical the Agent network generated a chemical structure with the same
structure may also be applied for the replacement of large scaffold as the scaffolds included in the tabu list, the score of
molecular weight drugs such as polypeptides, nucleic acids and the compound was set to zero. The tabu list was updated each
antibodies by small molecule compounds. Recent technological time the Agent network was initialized. In REINVENT, training
innovations enabled the identification of high-affinity cyclic of the Agent network was done with sigma (a parameter related
peptides through DNA-encoded library screening40. Many to learning rate) of 120, set to experience_replay and total
antibody-based medicines are also available for the clinical training steps of 10,000. All other parameters were set to default
treatment. These large molecules often suffer from pharmacoki- values and pre-trained Prior network provided from REINVENT
netic issues including their low permeability to the cell membrane is used. To generate chemical structure with desired pharmaco-
or expensive production cost. When the target structures phore, the Agent networks were trained using the scoring func-
complexed with these large molecules are available, our strategy tion (Relative Pharmacophore-Fit) of LigandScout 4.443. The
may offer a promising option to develop small compounds that Relative Pharmacophore-Fit outputs a pharmacophore score that
satisfy the greatest common denominator of pharmacophores on is normalized to [0, 1] range based on the number of matching
the protein structure. pharmacophore features and the RMSD of the pharmacophore
“Deep Quartet” (Fig. 1a) consists of a series of flows including alignment. Based on the gastric proton pump complexed with its
(1) deep reinforcement learning, (2) LigandScout, a software blockers derived from X-ray crystallography (PDB ID: 5ylv and
using pharmacophore models, and (3) Sub-structure filtering to 5ylu), three different types of pharmacophore models (Type A, B,
select candidates that match the desired target. In addition to and C) were defined as shown in Supplementary Fig. 2. A DQ run
(1)–(3), (4) knowledge of medicinal or organic chemistry, is also using Type A pharmacophore generated 550 chemical structures.
included in the flow, to create an AI-platform for the drug Among them, compounds with methylamino group and phar-
discovery achieved by the “quartet” (Fig. 1a). Now Deep Quartet macophore score of 0.9 or higher were selected, and the resulting
has been added a fifth flow (5) high-resolution cryo-EM 71 compounds were considered as candidates for synthesis. In the
structural analysis. Thus, this platform now is “Deep Quintet”, same manner, a DQ run using Type B pharmacophore generated
providing a distinct and more powerful framework for the drug 600 chemical structures from which 10 compounds were con-
discovery. sidered as candidates for synthesis. A DQ run using Type C
pharmacophore generated 650 chemical structures where 181
compounds were considered as candidates for synthesis.
Methods
De novo drug design by Deep Quartet. Deep Quartet (DQ) is a
workflow for generating chemical structures with desired phar- Selection and synthesis of candidate compounds. DQ-02, DQ-
macophore (Fig. 1a)29,30. REINVENT41, which is an open-source 04 (Type A) and DQ-07 (Type C) were selected from the list of
Python application, is used as DGMs in DQ. The DGMs consist the above candidates. The structures of DQ-06 (Type C) and DQ-
of two recurrent neural network models named as Prior and 09 (Type A) were generated by removing halogen substituents of
Agent networks. The Prior network is trained using SMILES the compounds in the list. The structure of DQ-10, DQ-11 or
representation of molecules obtained from ChEMBL39. The DQ-12 were systematically generated by replacing the methyla-
Agent network is initialized by Prior network and then trained mino group of DQ-06 with a hydroxy, dimethylamino or amino
using reinforcement learning41. During the training, the prob- group, respectively. The structures of DQ-18, DQ-19 and DQ-21
ability distribution of the Agent network shifts toward a dis- were designed by adding a chloro group onto DQ-06. The list
tribution modulated by a scoring function. After the training, generated by using Type B pharmacophore included bicyclic
Agent network generates SMILES with a desired property heteroaromatic compounds having two alkyne units. Based on
obtained from a scoring function. In this study, to generate these structural features, DQ-14, DQ-15 and DQ-16 were newly
diverse chemical structure, two approaches, early stop & refresh designed from the viewpoint of synthetic accessibility, and were
and tabu list, were implemented in REINVENT. In early stop & evaluated by the pharmacophore scores before the synthesis. All
refresh, when averaged scores exceeded a pre-defined score (set to the compounds were synthesized according to the synthetic
0.8) threshold or training steps exceed pre-defined steps (set to procedures provided in the Supplementary Data 2.
1000) during the training, the training was stopped. Then, the
Agent network was initialized and the training was re-started. ATPase activity measurement. H+,K+-ATPase-enriched mem-
This process was done up to a pre-defined number of total brane fractions were purified from pig stomach according to the
training steps. The top 50 scoring structures generated from each previously reported procedures44. These membrane fractions

COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio 9

ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8

Fig. 4 Inhibition potency and the binding pose of DQ-18. a Dose-dependent inhibition of H+,K+-ATPase activity by indicated P-CABs (DQ-06, 07, 18, 19
and SCH28080) as shown in Fig. 2a. Data plotted represent each data point from triplicate of three independent measurement at 12 different
concentrations of P-CABs. b EM potential map (mesh) and atomic model of the gastric H+,K+-ATPase complexed with DQ-18 (blue sticks). c Comparison
of the binding poses between DQ-06 (green) and DQ-18 (blue). Arrow indicates the displacement of the binding position from DQ-06 to DQ-18 (0.6 Å).
d, e Molecular interactions between H+,K+-ATPase and DQ-18 in stick representation as shown in Fig. 3c, d. f A schematic 2D representation of DQ-18
binding pose as shown in Fig. 3g. g, h Clipped cross sections of the DQ-18 binding site as in Fig. 3e, f.

10 COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio

COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8 ARTICLE

Fig. 5 Inhibition potency and the binding pose of DQ-21. a Dose-dependent inhibition of H+,K+-ATPase activity by indicated P-CABs (DQ-06, 18 and 21)
as shown in Fig. 2a. Data plotted represent each data point from triplicate of three independent measurement at 12 different concentrations of P-CABs.
b EM potential map (mesh) and atomic model of the gastric H+,K+-ATPase complexed with DQ-21 (purple sticks). c Comparison of the binding poses
between DQ-18 (blue) and DQ-21 (purple). Arrow indicates the displacement of the binding position from DQ-18 to DQ-21 (1.2 Å). d, e Clipped cross
sections of the DQ-21 binding site as in Fig. 3e, f.

COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio 11

ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8

(~0.05 μg protein/40 μl solution in each well) were suspended in of 4 s under 99% humidity, and then plunge-frozen in liquid
buffer containing 40 mM PIPES/Tris (pH 7.0), 2 mM MgCl2, ethane. The prepared grids were transferred to a CRYO ARM 300
2 mM ATP di-tris salt and 10 mM KCl in the presence of different microscope (JEOL), operated at 300 kV, with a cold-field emis-
concentrations of SCH28080 or synthesized compounds in 96- sion gun as the electron source, an in-column Ω filter and
well microtubes34. Reactions were initiated by incubating the equipped with a Gatan K3 direct electron detector in the electron
fractions at 37 °C using a thermal cycler and maintained for 1 h. counting mode. Imaging was performed at a nominal magnifi-
Reactions were terminated by adding 2 M HCl, and the amount cation of ×60,000 for DQ-02 and DQ-18 bound forms, and
of released inorganic phosphate was determined ×80,000 for DQ-06 bound one, corresponding to a calibrated
colorimetrically45 using a microplate reader (TECAN). The spe- pixel size of 0.753 and 0.580 Å/pix, respectively (EM01CT at
cific H+,K+-ATPase activity was calculated by subtracting the SPring-8). Each movie was recorded in correlated-double sam-
activities in the presence of 0.5 mM SCH28080. The IC50 value pling (CDS) mode for 2.6 s and subdivided into 60 frames. The
was estimated by the sigmoidal curve fitting using software electron flux was set to 8.46 e−/pix/s at the detector, resulting in
PRISM 9. an accumulated exposure of 60 e−/Å2 at the specimen. The data
Data measured from 96-well plate contained triplicates of four were automatically acquired by the image shift method using
different sets of K+-dependent ATPase assays in the absence or SerialEM software52, with a defocus range of −0.8 to −1.8 μm.
presence of three different concentrations of synthesized The dose-fractionated movies were subjected to beam-induced
compounds. Data were corrected for background values in the motion correction, using RELION 3.153, and the contrast transfer
absence of K+ at each compound concentration, and plot them function (CTF) parameters were estimated using patch CTF
against the double reciprocal axes in Supplementary Fig. 5. Data estimation in cryoSPARC (v4, Structura Biotechnology)54.
were also fitted by using simultaneous nonlinear regression as For each dataset, particles were initially picked by blob picker
described previously to estimate their Ki values34. using cryoSPARC (v4), and extracted with down-sampling to a
Raw data for the ATPase measurement can be found in pixel size of 3.24 Å/pix. These particles were subjected to several
Supplementary Data 1. rounds of 2D classifications. Good-looking classes were then
subjected to ab initio reconstruction in three models, and refined by
non-uniform refinement55. The particles from the best class was
Expression and purification of recombinant H+,K+-ATPase.
then re-extracted to the full pixel size and subjected to non-uniform
Procedures for protein expression are essentially the same as
refinement with per-particle defocus refinement, beam-tilt refine-
those reported previously2,46. Briefly, Flag epitope tag
ment in cryoSPARC (v4). The particle stack was then transferred to
(DYKDDDDK), a hexa-histidine tag and the enhanced green
RELION 3.1, and subjected to Bayesian polishing56. Polished
fluorescence protein (EGFP) were inserted in the amino terminal
particles were re-imported to cryoSPARC (v4), and subjected to
side of Met48 of the pig gastric H+,K+-ATPase α-subunit, fol-
non-uniform refinement. The resolution of the analyzed map was
lowed by a tobacco etch virus (TEV) protease recognition
defined according to the FSC = 0.143 criterion57 (Supplementary
sequence and subcloned into a hand-made vector2. The wild type
Fig. 6). The local resolution and angular distributions for each
pig gastric H+,K+-ATPase β-subunit was also cloned. The αβ-
structure were estimated by cryoSPARC (v4). All the models were
complex of H+,K+-ATPase were expressed in the plasma mem-
manually built in Coot58 using the crystal structure of SCH28080-
brane using baculovirus-mediated transduction of mammalian
bound H+,K+-ATPase (5ylu) as a starting template2. Phenix59
HEK293S GnT1- cells (BacMam)47 purchased from ATCC.
(version 20) was used for refinement.
For cryo-EM analysis, cells were directly solubilized with 1%
lauryl maltose neopentyl glycol (LMNG)48 in the presence of
40 mM MES/Tris (pH 6.5), 10% glycerol, 5 mM dithiothreitol, Reporting summary. Further information on research design is
1 mM MgCl2, in the presence of 1 mM BeSO4, 3 mM NaF and available in the Nature Portfolio Reporting Summary linked to
protease inhibitor cocktail (Roche) on ice for 20 min. After this article.
removing insoluble material by ultracentrifugation, the super-
natant was mixed with anti-GFP nanobody resin49 at 4 °C for 2 h,
which was followed by washing with buffer containing 40 mM Data availability
The structural data generated in this study have been deposited in the Protein Data Bank
MES/Tris (pH 6.5), 5% glycerol, 1 mM MgCl2, 1 mM BeSO4,
and EM Data Bank under accession codes 8IJV and EMD-35500; Cryo-EM structure of
3 mM NaF, 50 mM NaCl and 0.06% glyco-diosgenin (GDN)50. the gastric proton pump with bound DQ-02, 8IJW and EMD-35501; Cryo-EM structure
After addition of TEV protease and endoglycosidase, anti-GFP of the gastric proton pump with bound DQ-06, 8IJX and EMD-35502; Cryo-EM
nanobody was incubated at 4 °C overnight. Digested peptide structure of the gastric proton pump with bound DQ-18; Cryo-EM structure of the
fragments containing EGFP and endoglycosidase were removed gastric proton pump with bound DQ-21, 8JMN and EMD-36424. Source data are
provided as Supplementary Data.
by passing the fractions through Ni-NTA resin (Qiagen). Flow-
through fractions were concentrated and subjected to size-
exclusion column chromatography using a Superose6 Increase
column (Cytiva) equilibrated in buffer comprising 20 mM MES/ Code availability
Deep Quartet is a software developed by Intage Healthcare Inc., is provided as
Tris (pH 6.5), 1 mM MgCl2, 1 mM BeSO4, 3 mM NaF, 50 mM Supplementary_Software.zip.
NaCl and 0.06% GDN. Peak fractions were collected and
concentrated to 8 mg/ml. A final concentration of 0.1 mM
Received: 11 August 2023; Accepted: 8 September 2023;
synthesized compound (DQ-02, DQ-06 or DQ-18) was added
to the protein sample.

Cryo-EM structural analysis. Preparation of sample and cryo-

EM grids were done according to a previous report1,51. The References
purified protein samples (at 8 mg/ml) containing 0.1 mM syn- 1. Tanaka, S. et al. Structural basis for binding of potassium-competitive acid
thesized compound were applied to a freshly glow-discharged blockers to the gastric proton pump. J. Med. Chem. 65, 7843–7853 (2022).
Quantifoil holey carbon grids (R1.2/1.3, Cu/Rh, 200 mesh), using 2. Abe, K., Irie, K., Nakanishi, H., Suzuki, H. & Fujiyoshi, Y. Crystal structures of
a Vitrobot Mark IV (Thermo Fisher) at 4 °C with a blotting time the gastric proton pump. Nature 556, 214–229 (2018).

12 COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio

COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8 ARTICLE

3. Shin, J. M. & Sachs, G. Pharmacology of proton pump inhibitors. Curr. 32. Ertl, P. & Schuffenhauer, A. Estimation of synthetic accessibility score of drug-
Gastroenterol. Rep. 10, 528–534 (2008). like molecules based on molecular complexity and fragment contributions. J.
4. Wallace, J. L. How do NSAIDs cause ulcer disease? Best Pract. Res. Clin. Cheminform. 1, 1–11 (2009).
Gastroenterol. 14, 147–159 (2000). 33. Asano, S., Io, T., Kimura, T., Sakamoto, S. & Takeguchi, N. Alanine-scanning
5. Hain, E., Coriat, R., Dousset, B. & Gaujoux, S. Prise en charge du gastrinome. mutagenesis of the sixth transmembrane segment of gastric H+,K+-ATPase
Presse Med. 45, 986–991 (2016). α-subunit. J. Biol. Chem. 276, 31265–31273 (2001).
6. Black, J. W., Duncan, W. A. M., Durant, C. J., Ganellin, C. R. & Parsons, E. M. 34. Abe, K. et al. The cryo-EM structure of gastric H+,K+-ATPase with bound
Definition and antagonism of histamine H2-receptors. Nature 236, 385–390 BYK99, a high-affinity member of K+-competitive, imidazo[1,2-a]pyridine
(1972). inhibitors. Sci. Rep. 7, 6632 (2017).
7. Wroblewski, L. E., Peek, R. M. & Wilson, K. T. Helicobacter pylori and gastric 35. Kaminski, J. J., Wallmark, B., Briving, C. & Andersson, B. M. Antiulcer agents.
cancer: factors that modulate disease risk. Clin. Microbiol. Rev. 23, 713–739 5. Inhibition of gastric H+/K(+)-ATPase by substituted imidazo[1,2-a]
(2010). pyridines and related analogues and its implication in modeling the high
8. Ganser, A. L. & Forte, J. G. K+-stimulated ATPase in purified microsomes of affinity potassium ion binding site of the gastric proton pump enzyme. J. Med.
bullfrog oxyntic cells. BBA Biomembr. 307, 169–180 (1973). Chem. 34, 533–541 (1991).
9. Chang, H., Saccomani, G., Rabon, E., Schackmann, R. & Sachs, G. Proton 36. Keeling, D. J., Laing, S. M. & Senn-Bilfinger, J. SCH 28080 is a lumenally
transport by gastric membrane vesicles. BBA Biomembr. 464, 313–327 (1977). acting, K+-site inhibitor of the gastric (H+ + K+)-ATPase. Biochem.
10. Palmgren, M. G. & Axelsen, K. B. Evolution of P-type ATPases. Biochim. Pharmacol. 37, 2231–2236 (1988).
Biophys. Acta Bioenerg. 1365, 37–45 (1998). 37. Maki-Yonekura, S., Hamaguchi, T., Naitow, H., Takaba, K. & Yonekura, K.
11. Toyoshima, C., Nakasako, M., Nomura, H. & Ogawa, H. Crystal structure of Advances in cryo-EM and ED with a cold-field emission beam and energy
the calcium pump of sarcoplasmic reticulum Ê resolution. Nature 405, filtration—refinements of the CRYO ARM 300 system in RIKEN SPring-8
647–655 (2000). center—. Microscopy 70, 232–240 (2021).
12. Post, R. L., Kume, S., Tobin, T., Orcutt, B. & Sen, A. K. Flexibility of an active 38. Nakane, T. et al. Single-particle cryo-EM at atomic resolution. Nature 587,
center in sodium-plus-potassium adenosine triphosphatase. J. Gen. Physiol. 152–156 (2020).
54, 306–326 (1969). 39. Gaulton, A. et al. ChEMBL: a large-scale bioactivity database for drug
13. Rabon, E. C. & Reuben, M. A. The mechanism and structure of the gastric discovery. Nucleic Acids Res. 40, D1100–D1107 (2012).
H,K-ATPase. Annu. Rev. Physiol. 52, 321–344 (1990). 40. Yoshida, S. et al. Peptide-to-small molecule: a pharmacophore-guided small
14. Vagin, O., Turdikulova, S. & Tokhtaeva, E. Polarized membrane distribution molecule lead generation strategy from high-affinity macrocyclic peptides. J.
of potassium-dependent ion pumps in epithelial cells: different roles of the Med. Chem. 65, 10655–10673 (2022).
N-glycans of their beta subunits. Cell Biochem. Biophys. 47, 376–391 (2007). 41. Olivecrona, M., Blaschke, T., Engkvist, O. & Chen, H. Molecular de-novo
15. Shin, J. M., Cho, Y. M. & Sachs, G. Chemistry of covalent inhibition of the design through deep reinforcement learning. J. Cheminform. 9, 1–14 (2017).
gastric (H+, K +)-ATPase by proton pump inhibitors. J. Am. Chem. Soc. 126, 42. Bemis, G. W. & Murcko, M. A. The properties of known drugs. 1. Molecular
7800–7811 (2004). frameworks. J. Med. Chem. 39, 2887–2893 (1996).
16. Hersey, S. J., Steiner, L., Mendlein, J., Rabon, E. & Sachs, G. SCH28080 43. Wolber, G., Dornhofer, A. A. & Langer, T. Efficient overlay of small organic
prevents omeprazole inhibition of the gastric H + / K +-ATPase. Biochim. molecules using 3D pharmacophores. J. Comput. Aided Mol. Des. 20, 773–788
Biophys. Acta 956, 49–57 (1988). (2006).
17. Wallmark, B. et al. Inhibition of gastric H+,K+-ATPase and acid secretion by 44. Abe, K. & Olesen, C. Isolation of H+,K+-ATPase-enriched membrane fraction
SCH 28080, a substituted pyridyl(1,2a)imidazole. J. Biol. Chem. 262, from pig stomachs. Methods Mol. Biol. 1377, 19–27 (2016).
2077–2084 (1987). 45. Chifflet, S., Torriglia, A., Chiesa, R. & Tolosa, S. A method for the
18. Kim, D. K. et al. Effects of tegoprazan, a novel potassium-competitive acid determination of inorganic phosphate in the presence of labile organic
blocker, on rat models of gastric acid-related disease. J. Pharmacol. Exp. Ther. phosphate and high concentrations of protein: application to lens ATPases.
369, 318–327 (2019). Anal. Biochem. 168, 1–4 (1988).
19. Han, K. S., Kim, Y. G., Yoo, J. K., Lee, J. W. & Lee, M. G. Pharmacokinetics of 46. Young, V. C. et al. Structure and function of H+/K+ pump mutants reveal Na
a new reversible proton pump inhibitor, YH1885, after intravenous and oral +/K+ pump mechanisms. Nat. Commun. 13, 5270 (2022).
administrations to rats and dogs: hepatic first-pass effect in rats. Biopharm. 47. Dukkipati, A., Park, H. H., Waghray, D., Fischer, S. & Garcia, K. C. BacMam
Drug Dispos. 19, 493–500 (1998). system for high-level expression of recombinant soluble and membrane
20. Otake, K. et al. Characteristics of the novel potassium-competitive acid glycoproteins for structural studies. Protein Expr. Purif. 62, 160–170 (2008).
blocker vonoprazan fumarate (TAK-438). Adv. Ther. 33, 1140–1157 (2016). 48. Chae, P. S. et al. Maltose–neopentyl glycol (MNG) amphiphiles for
21. Nagata, N. et al. Population-level metagenomics uncovers distinct effects of solubilization, stabilization and crystallization of membrane proteins. Nat.
multiple medications on the human gut microbiome. Gastroenterology 163, Methods 7, 1003–1008 (2010).
1038–1052 (2022). 49. Kubala, M. H., Kovtun, O., Alexandrov, K. & Collins, B. M. Structural and
22. Nochaiwong, S. et al. The association between proton pump inhibitor use and thermodynamic analysis of the GFP:GFP-nanobody complex. Protein Sci. 19,
the risk of adverse kidney outcomes: a systematic review and meta-analysis. 2389–2401 (2010).
Nephrol. Dial. Transplant. 33, 331–342 (2018). 50. Chae, P. S. et al. A new class of amphiphiles bearing rigid hydrophobic groups
23. Hagiwara, T., Mukaisho, K.-I., Nakayama, T., Sugihara, H. & Hattori, T. Long- for solubilization and stabilization of membrane proteins. Chem. Eur. J. 18,
term proton pump inhibitor administration worsens atrophic corpus gastritis 9485–9490 (2012).
and promotes adenocarcinoma development in Mongolian gerbils infected 51. Nakanishi, H. et al. Transport cycle of plasma membrane flippase ATP11C by
with Helicobacter pylori. Gut 60, 624–630 (2011). Cryo-EM. Cell Rep. 32, 108208 (2020).
24. Abe, K., Yamamoto, K., Irie, K., Nishizawa, T. & Oshima, A. Gastric proton 52. Mastronarde, D. N. Automated electron microscope tomography using robust
pump with two occluded K+ engineered with sodium pump-mimetic prediction of specimen movements. J. Struct. Biol. 152, 36–51 (2005).
mutations. Nat. Commun. 12, 1–11 (2021). 53. Zivanov, J. et al. New tools for automated high-resolution cryo-EM structure
25. Huang, H., Yu, P. S. & Wang, C. An introduction to image synthesis with determination in RELION-3. Elife 7, e42166 (2018).
generative adversarial nets. https://doi.org/10.48550/arxiv.1803.04469 (2018). 54. Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC:
26. Johnson, M. et al. Google’s multilingual neural machine translation system: algorithms for rapid unsupervised cryo-EM structure determination. Nat.
enabling zero-shot translation. Trans. Assoc. Comput. Linguist. 5, 339–351 Methods 14, 290–296 (2017).
(2017). 55. Punjani, A., Zhang, H. & Fleet, D. J. Non-uniform refinement: adaptive
27. Bilodeau, C., Jin, W., Jaakkola, T., Barzilay, R. & Jensen, K. F. Generative regularization improves single-particle cryo-EM reconstruction. Nat. Methods
models for molecular discovery: recent advances and challenges. WIREs 17, 1214–1221 (2020).
Comput. Mol. Sci. 12, e1608 (2022). 56. Zivanov, J., Nakane, T. & Scheres, S. H. W. A Bayesian approach to beam-
28. Zhavoronkov, A. et al. Deep learning enables rapid identification of potent induced motion correction in cryo-EM single-particle analysis. IUCrJ 6, 5–17
DDR1 kinase inhibitors. Nat. Biotechnol. 37, 1038–1040 (2019). (2019).
29. Yoshimori, A., Kawasaki, E., Kanai, C. & Tasaka, T. Strategies for design of 57. Rosenthal, P. B. & Henderson, R. Optimal determination of particle
molecular structures with a desired pharmacophore using deep reinforcement orientation, absolute hand, and contrast loss in single-particle electron
learning. Chem. Pharm. Bull. 68, 227–233 (2020). cryomicroscopy. J. Mol. Biol. 333, 721–745 (2003).
30. Yoshimori, A. et al. Design and synthesis of DDR1 inhibitors with a desired 58. Emsley, P. et al. Coot: model-building tools for molecular graphics. Acta
pharmacophore using deep generative models. ChemMedChem 16, 955–958 Crystallogr. Sect. D Biol. Crystallogr. 60, 2126–2132 (2004).
(2021). 59. Adams, P. D. et al. PHENIX: a comprehensive Python-based system for
31. Bickerton, G. R., Paolini, G. V., Besnard, J., Muresan, S. & Hopkins, A. L. macromolecular structure solution. Acta Crystallogr. Sect. D Biol. Crystallogr.
Quantifying the chemical beauty of drugs. Nat. Chem. 4, 90–98 (2012). 66, 213–221 (2010).

COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio 13

ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-023-05334-8

Acknowledgements Correspondence and requests for materials should be addressed to Kazuhiro Abe,
K.A. thanks Drs. A. Oshima and K. Tanaka for their help in grid screening and image Satoshi Yokoshima or Atsushi Yoshimori.
processing, and Dr. P. Artigas for discussion. Cryo-EM experiments were performed at
EM01CT at SPring-8 with the approval of the Japan Synchrotron Radiation Research Peer review information This manuscript has been previously reviewed at another
Institute (JASRI proposal number 2021B2530). The authors thank the ACISS Academia Nature Portfolio journal. The manuscript was considered suitable for publication without
Support Service for the in silico experiments. This work has been funded by Grant-in-Aid further review at Communications Biology. Primary Handling Editor: Gene Chong.
for Scientific Research (21H02426), Novartis Foundation, Daiichi Sankyo Foundation of
Life Science, JST CREST Grant Number JPMJCR22E4 (to K.A.). This work was finan- Reprints and permission information is available at http://www.nature.com/reprints
cially supported by the Research Support Project for Life Science and Drug Discovery
[Basis for Supporting Innovative Drug Discovery and Life Science Research (BINDS)] Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
from AMED under Grant Number JP23ama121044 (to S.Y.) [support number 4085 published maps and institutional affiliations.
(to K.A.)].

Open Access This article is licensed under a Creative Commons

Author contributions
K.A. designed the study with S.Y. and A.Y. C.K. and A.Y. performed DQ calculation. Attribution 4.0 International License, which permits use, sharing,
M.O., M.I., Y.M., S.K., N.U. and S.Y. synthesized the compounds. K.A. and C.N. per- adaptation, distribution and reproduction in any medium or format, as long as you give
formed ATPase assay. K.A. expressed and purified protein. K.A. performed cryo-EM appropriate credit to the original author(s) and the source, provide a link to the Creative
analysis with C.C.G., C.G. and H.S. K.A. performed image processing, model building Commons license, and indicate if changes were made. The images or other third party
and structural interpretations. K.A., S.Y. and A.Y. wrote the manuscript. C.C.G. and C.G. material in this article are included in the article’s Creative Commons license, unless
commented on and edited the manuscript. indicated otherwise in a credit line to the material. If material is not included in the
article’s Creative Commons license and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from
Competing interests the copyright holder. To view a copy of this license, visit http://creativecommons.org/
A.Y. is a CEO of Institute for Theoretical Medicine, Inc. C.K. is an employee of INTAGE licenses/by/4.0/.
Healthcare Inc. Other authors declare no competing interests.

© The Author(s) 2023

Additional information
Supplementary information The online version contains supplementary material
available at https://doi.org/10.1038/s42003-023-05334-8.

14 COMMUNICATIONS BIOLOGY | (2023)6:956 | https://doi.org/10.1038/s42003-023-05334-8 | www.nature.com/commsbio