Introduction

Helicobacter pylori (H. pylori) is a spiral-shaped gram-negative microaerophilic bacterium which can survive and colonize in the human stomach. H. pylori infection is a major risk factor for chronic gastritis, dyspepsia, and peptic ulcers, and is even believed to be a crucial initiating factor leading to gastric cancer and its precancerous lesions.1,2,3 Furthermore, it was the primary carcinogenic pathogen associated with gastric tumorigenesis and classified as a Class I carcinogen by the International Agency for Research on Cancer (IARC) of the World Health Organization (WHO) in 1994.4,5 The infection rate of H. pylori exceeds 50% worldwide and is even higher in develo** countries where sanitation and hygiene practices may be inadequate.6

Currently, the first-line eradication regimens for H. pylori infection primarily consist of antibiotic-based triple therapy and quadruple therapy, which involve two types of antibacterial drugs, along with a proton pump inhibitor (omeprazole (OPZ)/vonoprazan/lansoprazole, etc.) or/and bismuth agent for synergistic treatment.7,8 However, several large clinical trials and meta-analyses have shown that the eradication rate of the standard triple therapy has generally declined to less than 80%.9 The main cause is the global increasing drug resistance rate of H. pylori to clinically recommended antibiotics, including amoxicillin (AMX), clarithromycin (CLA), metronidazole (MTZ), levofloxacin (LEV), and tetracycline (TC) over the past two decades.10,11 The resistance rate of MTZ ranges from 29% to 60% in different WHO regions during 2012 to 2016, whereas the resistance rate of CLA ranges from 21% to 35%.12 Additionally, the dual resistance rate of CLA and MTZ is greater than 25%, which makes recurrent infections extremely difficult to be cured.13,14 It is worth noting that CLA-resistant H. pylori was listed as one of the high-priority category pathogens by WHO, highlighting the urgent need for new antibacterials to combat this issue.15,16 Current H. pylori eradication regimens are also associated with side effects and gastrointestinal flora imbalances, further emphasizing the need for new treatment options.

Berberine (BBR) has long been used as traditional Chinese medicine for treating gastrointestinal disorders. An increasing number of clinical studies have suggested that the use of BBR in combination with triple therapy could improve the eradication rates of H. pylori,17,18 with the advantages of proper pharmacokinetic properties, low incidences of adverse reactions, and improved internal environment and pathological repair, etc. It was found that BBR could inhibit the growth, respiration, and sugar metabolism of H. pylori.19 In addition, BBR can exert anti-H. pylori effects by inhibiting the activity of the arylamine N-acetyltransferase20 and urease.21,22 Meanwhile, another protoberberine alkaloid palmatine (PMT) also exhibits antibacterial activity against H. pylori both in vitro and in vivo,23,24 as well as its potential therapeutic effects on gastritis and peptic ulcer caused by H. pylori infection. However, both BBR and PMT only exhibited moderate inhibitory activity against H. pylori (minimum inhibitory concentrations, MIC = 16–256 μg/mL).21,24 Thus, there is a potential to obtain candidates from BBR or PMT with improved activity, new mechanism of action, and favorable safety profiles.

Deep learning has emerged as a pivotal tool in drug discovery, revolutionizing the field by enabling the rapid and efficient screening of compounds for potential therapeutic applications. Its significance lies in the ability to analyze large datasets with multiple features, extracting intricate patterns that may not be discernible through traditional methods. These models can predict the potency, toxicity, and mode of action of compounds with remarkable accuracy, expediting the identification of promising candidates. Furthermore, deep learning facilitates the exploration of chemical space, allowing for the generation of novel compounds with desired properties.25 In the context of antibacterial development, deep learning models have been successfully used in the discoveries of novel antimicrobial agents, such as anti-Acinetobacter baumannii26,27 and methicillin-resistant Staphylococcus aureus28 candidates. Therefore, the objective of this study was to establish a dependable deep learning model for the exploration of novel candidates against H. pylori.

Firstly, the molecular fingerprint features and molecular graph embeddings were concatenated to form the compound feature vectors through pre-training and fine-tuning. After a rigorous iterative learning process, a deep learning model was constructed based on different training sets of 13,638 molecules. To validate the effectiveness of the deep learning model, a series of novel BBR derivatives were strategically designed, predicted, synthesized, and biologically validated. This endeavor led to the successful prediction of four candidates which exhibited remarkable anti-H. pylori activity. Among these predicted candidates, 3-,13-disubstituted analog 8 exhibited the most promising activity against all tested antibacterial-susceptible and resistant H. pylori strains, with the MIC values ranging from 0.25–0.5 μg/mL, indicating strong inhibitory activity and a potential different mechanism of action compared to the first-line anti-H. pylori antibiotics. Given its potential as a novel anti-H. pylori agent, 8 was selected as a representative compound for the further investigation of its potency, safety, and mechanisms of action.

This study showcases a comprehensive and integrative methodology in anti-H. pylori agent discovery, leveraging both traditional experimental techniques and cutting-edge machine learning algorithms.

Results

A deep learning training set is established for novel anti-H. pylori agents exploration

First, the dataset was curated from reputable sources and ensured the diversity in chemical structures and activity levels. A sizable collection of 938 compounds with known anti-H. pylori properties was established. This dataset included 801 reported anti-H. pylori compounds with structural diversity from ChEMBL database,29 as well as 137 self-established BBR derivatives.30,31,32 An MIC value of 16 μg/mL was set as the critical value. The compounds with MICs ≤ 16 μg/mL were defined as ‘active’ (label 1) and MICs > 16 μg/mL as ‘inactive’ (label 0). The proposed deep learning framework firstly represented compounds with molecular graph, and extracted the molecular extended-connectivity fingerprints (ECFP)33 which preserve rich functional group information. Feature engineering was performed to extract the ECFP that captured essential functional group information, and leveraged message passing deep neural network to extract properties directly from molecular structure.

Since the significant interactions between atomic pairs with topologically distant could also affect the overall molecular properties (Fig. 1a), a deep graph neural network (Attentive FP)49 BamD facilitates the delivery of OMP substrates to BamA β-barrel and the subsequent assembly. To investigate if the function of BamD was affected by 8, a Co-Immunoprecipitation (Co-IP) test was performed using GST-tagged BamD and His-tagged BamA. As depicted in Fig. 6f, BamD exhibited a strong interaction with BamA, and this effect was suppressed by 8, indicating that 8 might inhibit the function of the BAM machinery by affecting the BamA-BamD interaction.

Discussion

The prevalence of H. pylori infection remains significant worldwide. In the context of increasing rates of drug-resistant H. pylori strains globally, it is of great importance to develop new treatment strategies that can overcome the limitations of current regimens. The use of agents with novel mechanism for combating H. pylori is essential to potentially avoid cross-resistance with existing antibacterials.

In this study, a deep learning model was successfully established for the prediction of new anti-H. pylori agents, and a novel candidate 8 specifically acting on H. pylori was predicted. The results of the proposed deep learning model demonstrate a promising accuracy in identifying potential anti-H. pylori agents, thereby offering a possible tool in the pursuit of novel therapeutic interventions against this medically significant pathogen. The integration of deep learning and traditional drug discovery holds immense promise for addressing the growing threat of this pathogen, advancing the possible process towards more effective and sustainable therapeutic solutions. This approach may not only accelerate the anti-H. pylori candidate discovery but also provide a more targeted and informed approach for compound selection.25 Building upon the accurate prediction and validation on a small-scale dataset in this study, we intend to further conduct predictions and validations on larger datasets. Also, leveraging this deep learning model, structure-based rational fragment growth might be an efficient approach to establish an effective compound library against H. pylori including drug-resistant strains, which can be utilized for the initial or recurrent eradication therapy.

Due to the efficacy of 8 against multidrug-resistant H. pylori strains (those resistant to first-line agents such as MET, CLA, and AMX) and its synergistic or additive effects when combined with CLA or AMX we observed in this study, 8 may offer a potential alternative to one or two antibiotics to tackle the severe drug resistance. The poor compliance of the patients with the current eradication therapies resulting from the prolonged high-dose and the pronounced impact of antibiotics on intestinal microbiota are factors that limit the efficacy of the current regimens used clinically. The unique pharmacokinetic profile of 8, characterized by its predominant gastric retention, may reduce its systemic side effects, dosing frequency, and course of treatment. Besides, less gastrointestinal flora disturbance by 8 enables its promising compliance. In addition, the favorable acid stability contributes to the efficacy of the mono-therapy of 8, exhibiting comparable eradicative effects compared with both triple-therapy (OPZ + AMX + CLA) and the quadruple-therapy (OPZ + AMX + CLA + bismuth citrate) groups. The promising anti-H. pylori activity both in vitro and in vivo, favorable acid stability and gastric accumulation, low possibility of resistance evolution, and less perturbance on intestinal flora enable 8 an ideal candidate for H. pylori eradication.

ABPP is a technique in the field of proteomics that possesses several distinct advantages, contributing to its widespread adoption and application in diverse areas of biological research. ABPP facilitates the identification of potential therapeutic targets by profiling the active proteome in disease-relevant tissues or cells. These verified targets may be crucial for understanding the key players in disease pathways and for designing targeted therapeutic interventions. To demonstrate that probe 8-O based target exploration was reliable, several verifications were conducted. Besides, to verify the reliability of the above-mentioned target exploration results, the same interactions between 8-O and the specific binding sites of SecA/BamD were predicted by molecular docking, as depicted in Supplementary Fig. S9. Also, 8-O could dose dependently inhibit SecA, with an IC50 value of 13.92 μg/mL (Supplementary Fig. S10). And a similar inhibitory effect on BamA-BamD interaction was observed in compound 8-O treatment group (Supplementary Fig. S11). Thus, 8-O could be regarded as an appropriate tool for ABPP-based direct target fishing, and verification results were highly consistent with the mechanisms speculated in morphologic observation and transcriptomic investigation in this study.

SecA and BamD, the validated targets of compound 8, as indispensable component of the Sec complex and Bam machinery in OMPs transport, respectively, are essential for bacterial viability, making them promising targets for antibacterial therapy. The Sec pathway is involved in the translocation of unfolded preproteins and the insertion of membrane proteins across or into the cytoplasmic membrane, while the Bam machinery mediates the folding and insertion of β-barrel OMPs into the outer membrane of gram-negative bacteria. Currently, limited studies have been conducted on inhibitors of H. pylori Sec and Bam systems. Compound 8 was predicted and found to act on both systems, potentially blocking the entire transport process of OMPs. Consequently, the OMPs cannot be adequately assembled into the outer membrane. Therefore, the inhibition of SecA and BamD would lead to the deficiency in OMPs of the gram-negative bacteria. In order to verify 8 can lead to the OMPs deficiency of H. pylori by targeting SecA and BamD, the outer membrane of H. pylori from the untreated control group and 1/2 MIC (0.25 μg/mL) of 8-treated group was isolated by super-centrifugation to quantify the change in the total OMPs amount after the treatment of 8. As shown in Fig. 6g, a significant reduction in the amount of OMPs in the 8-treated group was observed, which further confirmed the effect of 8 on transport and assembly system of OMPs. The reduction of outer membrane proteins was also observed after 8-O treatment (Supplementary Fig. S12).

OMPs also play a critical role in the adhesion of H. pylori to the host cells as the important first step in persistent colonization.50 The absence of OMPs, especially adhesion-related proteins (BabA, SabA, OipA, etc.), could lead to a decline in the ability of H. pylori to adhere to gastric host cells, and this effect was observed for the H. pylori after the treatment of 8. As shown in Fig. 6h, i, the nuclei, cell membrane, and H. pylori are stained with blue, red, and green fluorescence, respectively. After treatment with 8, the green fluorescence was markedly weakened (Fig. 6i), suggesting a significant reduction of the adhesive or intracellular H. pylori.

Increasing attention has been paid to the OMP secretion and transport system of gram-negative bacteria in recent years.51 OMPs have been shown to be involved in bacterial adhesion and immune stimulation/evasion, and are essential for colonization and/or pathogenesis.52 As OMP transport and assembly system of H. pylori is relatively conserved and species-specific,53 agents targeting this system can be regarded as attacking the Achilles’ Heel of H. pylori. Therefore, agents targeting OMP transport system might be used alone or considered as an ideal component against H. pylori.

To sum up, a deep learning model was successfully established through correlating a set of over 10,000 molecules with anti-H. pylori potencies. This approach showcases a comprehensive and integrative methodology in anti-H. pylori drug discovery, leveraging both traditional experimental techniques and cutting-edge machine learning algorithms for enhanced candidate selection and prioritization. A novel BBR derivative 8 was positively predicted and found to exhibit satisfactory potencies against all tested H. pylori strains, including MDR strains, with an MIC range of 0.25-0.5 μg/mL, and demonstrated a good safety profile both in vitro and in vivo. The pharmacokinetic study revealed that the gastric concentration of 8 maintained higher than its MIC value up to 24 h, indicating an ideal gastric retention performance of 8 for gastrically colonized H. pylori eradications. Mono-therapy of 8, as well as 8-based dual (OPZ + 8) and quadruple-therapy treatment (OPZ + AC + 8) significantly decreased stomach bacteria load in vivo. Using ABPP technique, SecA and BamD belonging to OMP secretion and transport system were identified as direct targets of 8, and further confirmed by pull-down, CESTA, and single mutation assays (Fig. 7). By down-regulating OMPs, 8 could also inhibit the adhesion of H. pylori to GES-1 cells. Based on its novel mechanism, the risk of cross-resistance of 8 with existing antibacterial agents was relatively low. In conclusion, compound 8, with safety and unique pharmacokinetic properties, could exert anti-H. pylori activity through targeting SecA and BamD simultaneously, making it a potential new chemical entity for eradicating H. pylori including drug-resistant strains.

Fig. 7
figure 7

Cartoon of the mechanism of action of 8 (By Figdraw)

Full size image

Materials and methods

Chemicals, reagents, bacterial strains and experimental animals

All the chemical reagents and anhydrous solvents were purchased from commercial sources (J&K scientific) and used without further purification. CLA, MTZ, LEV, and AMX were purchased from National Institutes for Food and Drug Control, Bei**g, China. All H. pylori strains used in this study were obtained from the Collection Center of Pathogen Microorganism of Chinese Academy of Medical Sciences (CAMS-CCPM-A) in China, including the American Type Culture Collection (ATCC) reference strains. Two standard strains (H. pylori ATCC 43504 and ATCC 700392) and 31 clinically isolated strains, including CLA, MTZ, LEV, AMX resistant strains, were included in the current study.

All the animals were purchased from SPF Biotechnology Co. (Bei**g, China). In vivo acute toxicity and pharmacokinetic studies were performed under the permission and supervision of the Ethics Committee of Institute of Medicinal Biotechnology, Chinese Academy of Medical Sciences and Peking Union Medical College, China (Approval No.: IMB-20210224D602). In vivo anti-H. pylori activity evaluations were performed under the permission and supervision of the Ethics Committee of Bei**g Viewsolid Biotechnology Co. LTD, China (Approval No.: VS2126A00691 & VS2126A00720).

Deep learning model construction

A training set of 938 compounds with known anti-H. pylori properties were first collected (801 reported anti-H. pylori compounds and 137 BBR derivatives). Compounds with MICs ≤16 μg/mL were defined as ‘active’ (label 1) and the compounds with MICs >16 μg/mL were defined as ‘inactive’ (label 0). The proposed deep learning framework represented compounds with a molecular graph, and extracted the molecular ECFP fingerprints preserving rich functional group information. Then, a deep graph neural network (Attentive FP) was applied to learn the embeddings of molecular graphs. The pre-train-then-fine-tuning paradigm was utilized to pre-train the deep learning model on some large-scale bioassays (12,700 compounds) from PubChem database and fine-tune the pre-trained deep learning model on the collection of 938 compounds. Finally, the molecular fingerprint features and molecular graph embeddings were concatenated to form the compound feature vectors and then an MLP layer was leveraged to predict their activity against H. pylori.

Antimicrobial susceptibility tests in vitro

The MICs were determined by agar dilution method following the Clinical and Laboratory Standards Institute (CLSI) guidelines (M45) as follows. Briefly, the bacterial suspension adjusted to 2.0 McFarland density (about 108 CFU/mL) was prepared from a Columbia agar plate containing 5% defibrinated sheep blood. Then, 1 μL of the bacterial suspension was directly inoculated onto the MH agar plates with 5% defibrinated sheep blood containing 2-fold serial dilutions of test compounds. The plates were incubated at 37 °C for 72 h under microaerobic conditions (10% CO2). H. pylori ATCC 43504 was used as a control strain. After incubation, the plates were examined visually, and the MICs were defined as the lowest concentration of a drug showing no growth. The MIC results were obtained from two independent experiments. AMX, CLA, MTZ, and LEV were tested as quality controls. The resistance breakpoints were defined as >0.125, >0.25, >8, and >1 μg/mL for AMX, CLA, MTZ, and LEV, respectively, according to EUCAST.

The agar dilution checkerboard assay was used to assess the combination effect of compound 8 and AMX/CLA on 25 H. pylori strains. MH blood agar (including 5% defibrated sheep blood) plates containing different concentration combinations of 8 and AMX/CLA were inoculated with bacterial suspension (adjusted to 2.0 McFarland density) spots of 1 μL. After incubation at 37 °C for 72 h under microaerobic conditions, the effect of the combination was evaluated using the FICI equation: FICI = (MIC of drug A in the combination/MIC of drug A alone) + (MIC of drug B in the combination/MIC of drug B alone). The combination was defined to be synergistic when the FICI was ≤0.5; additive when 0.5 < FICI ≤ 4; antagonistic when FICI > 4. The experiments were performed in triplicate on different days.

Resistance development by serial passage

Bacteria were induced to develop resistance to drugs by continuously passaging bacteria grown in sub-MIC levels of antimicrobial agents. H. pylori ATCC 43504 was serially passaged in sub-MIC concentration of 8, for 12 consecutive passages (36 days) to assess the potential resistance induction. AMX and CLA were tested under identical conditions as controls. Briefly, bacteria suspension was adjusted to a density of 2.0 McFarland and diluted 20-fold in Brain & Heart Infusion (BHI broth) supplemented with 10% (v/v) fetal bovine serum (FBS) containing 8, AMX or CLA at the concentration of their 1/4 MIC, respectively. The inocula were incubated for 72 h at 37 °C with shaking under microaerobic conditions. The process was repeated for 12 consecutive passages. The MICs of each passage were determined together with the un-induced strain.

Cell culture

GES-1 cells, HepG2 cells and 293T cells were cultured with DMEM complete medium (10% FBS, 1% penicillin/streptomycin was added to the medium), and H460 cells were cultured with RPMI-1640 complete medium (10% FBS, 1% penicillin/streptomycin was added to the medium) and placed in a constant temperature incubator at 37 °C and 5% CO2 nourish. When cells have grown to approximately 80–90% confluence, subculture was conducted at a ratio of 1:3.

In vitro cytotoxicity assay (MTT assay)

Serial dilutions of test compounds, starting from 100 µM were incubated with GES-1, HepG2, H460 or 293T cells in 96-well plates for 24 h at 37 °C. Subsequently, 20 μL MTT (5 mg/mL) was added to each well and cultured for 4 h. After the removal of the culture medium, 150 μL DMSO was added to dissolve formazan crystals and the absorbance at 570 nm was measured on a fluorescent microplate reader (Synergy H1 model, BioTek, Agilent, USA). The TC50 values of test compounds were calculated by GraphPad Prism 8 (San Diego, CA).

In vivo acute toxicity test

Kunming mice weigh 18-20 g, 6 in each group, half males and half females were used in acute toxicity evaluation. Compound 8 was administered orally once at the dose of 125, 250, or 500 mg/kg, and the animals were monitored and death recorded for 2 weeks. After 14 days post administration, blood was taken after fasting. The liver and kidney function of the mice was evaluated by detecting the GOT, GPT, BUN, and CRE.

Pharmacokinetic study

C57BL/6 mice (20 ± 2 g, half males and half females) were housed in a temperature (22 ± 2 °C) and humidity (55 ± 6%)-controlled facility. The mice had free access to standard laboratory food and water, and food was withdrawn overnight before experiments. Thirty-two mice were randomly divided into eight groups (4 mice each time point), and blood samples and stomachs were collected at 0.5, 1, 2, 4, 6, 8, 12 and 24 h after a single oral dose of 8 (30 mg/kg). Plasma or stomach homogenate (50 μL) was mixed with 200 μL acetonitrile, vortexed for 2 min and centrifuged at 15,000 rpm for 15 min at 4 °C. The supernatant was directly injected into the LC-MS/MS system for quantitative analysis of 8, with BBR as an internal standard. Quantification was performed in the selected reaction monitoring mode with (m/z)+ 336 → 320 (collision energy 20 eV) and 446 → 390 (collision energy 20 eV) for 8 and BBR, respectively. Concentration-versus-time profiles of 8 in plasma and stomach were obtained.

Anti-H. pylori efficacy in vivo

Specific-pathogen-free female C57BL/6 mice (18 ± 2 g) were used for this study. Each mouse received 0.3 mL of 109 CFU/mL H. pylori CCPM(A)-P-3722159 through oral gavage every 48 h, repeated four times (on day 1, 3, 5, and 7), and the infection was allowed to develop for 2 weeks. The infected mice were randomly assigned to four treatment groups (n = 5) to receive CMC, dual therapy (OPZ + 8), triple therapy (OPZ + AC) and quadruple therapy (OPZ + AC + 8), respectively. Mice were administered OPZ through oral gavage at a dose of 200 μg/kg 30 min before the administration of the assigned treatments. Compound 8 (30 mg/kg), AC (15 mg/kg AMX and 15 mg/kg CLA), and 8 + AC were administered once daily for a consecutive 5 days by oral gavage. The control group received an equivalent volume of CMC. Forty eight hours after the last treatment, mice were sacrificed, and the stomachs were harvested. The stomach was cut along the greater curvature, and the gastric content was removed and rinsed with PBS. The stomachs were cut into two longitudinal sections, and each section was weighed. The two sections were used for the assessment of bacterial colonization and histology (HE stain), respectively. For bacterial colonization, a gastric tissue section was suspended in 1 mL PBS and homogenized for H. pylori recovery. The homogenate was serially diluted and spotted onto a Columbia blood agar plate containing vancomycin (10 μg/mL), amphotericin (5 μg/mL), cefsulodin sodium (5 μg/mL), trimethoprim (5 μg/mL), colistin (0.3 μg/mL) and bacitracin (30 μg/mL). The plates were then incubated at 37 °C under microaerobic conditions for 3 to 5 days, and bacterial colonies were counted, and expressed as CFU per gram of stomach.

In addition to the above in vivo efficacy evaluation, the mono therapy of compound 8 was also performed compared with triple- and quadruple-therapy in the second batch of animal experiment with the same dosage of each drug as in the first batch.

Scanning electron microscopy and transmission electron microscopy

The effects of 8 on the structure of H. pylori were examined via SEM and TEM. One hundred microliters bacterial suspension (~108 CFU/mL) of H. pylori ATCC 43504 was inoculated onto Columbia agar plates with 5% defibrinated sheep blood containing 8 at 1/2MIC (0.25 μg/mL) for 48 h, collected and fixed with 2.5% glutaraldehyde in PBS for 24 h. The fixed bacteria were rinsed in fresh PBS, passed through an ethanol gradient for dehydration, dried, coated with gold, and observed by scanning electron microscope (Hitachi SU8020, Tokyo, Japan). The fixed organisms were washed and postfixed with 1% osmium tetroxide. Then the samples were washed, dehydrated in a graded series of ethanol, and embedded in Epon Araldite. Ultrathin sections containing the cells were placed on copper grids, stained with uranyl acetate and lead citrate, observed, and photographed with a TEM microscope (Hitachi, Tokyo, Japan).

Experimental procedure for ABPP-based direct target exploration

After the protein concentration of H. pylori ATCC 43504 lysate was determined using Bradford protein assay (Bio-Rad USA), it was prepared into several 1 μg/μL aliquots using PBS buffer containing 1% SDS. Then they were treated with probe (8-O) DMSO solution with or without 8 (500 µM), and the mixtures were incubated at room temperature for 2 h followed by UV irradiation (~365 nm) for 20 min. Equal amount sample with different treatment was used for fluorescent tagging or biotin tagging. For each click reaction, SDS (0.1%), TCEP (1 mM), CuSO4 (1 mM), TBTA (100 µM) and Cy3-azide or biotin-azide (100 µM) was sequentially added to the lysates, incubated for 2 h, and then the labeled proteins were precipitated with acetone. For fluorescence detection, the samples were dissolved with 1×SDS loading buffer and boiled at 95 °C for 10 min. For LC-MS/MS and immunoblot detection, the samples were dissolved in PBS buffer with 1% SDS, and incubated with washed Pierce Streptavidin Magnetic Beads for 2 h. After washing 5 times with TBS-T, the beads were mixed with 1×SDS loading buffer, followed by boiling at 95 °C for 10 min. Samples were detected by gel fluorescence or immunoblot blot assay (Tanon 5200 system, Tanon, Shanghai, China).

The proteins were released from beads after 95 °C boiling and identified with a Label-free based LC-MS/MS approach (Bei**g Qinglian Biotech Co., Ltd., China). MS was analyzed using Maxquant software (version 1.5.3.30) with the UniProtKB/Swiss-Prot Helicobacter Pylori ATCC 43504 database.

Protein labeling and pull-down assay

H. pylori was pretreated with 8 or DMSO for 1.5 h and then treated with 8-O for 0.5 h at room temperature, and lysed. After being 365 nm ultraviolet irradiated, click chemical reagents were used to introduce biotin, then the samples were enriched and purified with Pierce streptavidin magnetic beads. The immobilized beads were then washed three times with TBS-T buffer and mixed with 1 × loading buffer, followed by boiling at 95 °C for 10 min. The amount of target protein in the pull-down after enrichment of lysate and magnetic beads was detected by immunoblot analysis.

SPR analysis

The SPR assay was performed on a BIAcore T200 biosensor system (GE Healthcare Life Sciences, Piscataway, NJ, USA) at 25 °C using a CM5 chip. The bindings of SecA or BamD at different concentrations of 8 were performed in 1 × PBS-P (GE Healthcare Life Sciences) at a flow rate of 30 μL/min for 120 s. After each binding reaction, a further dissociation time of 60 s was applied after each injection to allow the signal to return to the baseline.

CESTA assay

For the temperature-dependent thermal shift assay, 50 µL of lysates (3 mg/mL) from SecA-overexpressing or BamD-overexpressing E. coli strains were incubated with DMSO or 500 µM of 8 at each temperature point from 44 to 76 °C for 4 min. The samples were centrifuged at 20,000 × g for 10 min at 4 °C to separate the supernatant and pellet. The supernatant (16 µL) was mixed with 4 µL of 5 × loading buffer and then separated on a 10% SDS-PAGE for immunoblotting analysis of SecA or BamD.

Binding-site exploration between 8 and SecA/BamD

E. coli lysates (100 µL) with recombinant SecA or BamD (3 mg/mL in 10 mM PBS buffer) were incubated with 8 (400 µM) or DMSO respectively, and photo-labeled with 25 µM of the probe. The labeled proteins were washed with cold acetone and methanol, and then solubilized with 100 µL of 1% SDS/PBS. Then the samples were incubated with washed beads for 2 h at room temperature. After washing five times with TBS-T, the beads were reconstituted in 500 µL of water for the pretreatment and LC-MS/MS analysis.

Molecular modeling analysis

An automated docking was carried out using the 3D structure of SecA (Alphafold code: AF-I0ZGP5-F1), BamD (Alphafold code: AF-I0ZHZ3-F1) and Discovery Studio 4.5 software. All conformers of 8 were interactively docked to active binding pockets selected based on previous experimental results using the Libdock protocol.

Site-directed mutagenesis

The lysates of E. coli expressing SecA-, BamD-wild type (WT), SecA-E225A, R239A, E733A, D773A and BamD-F125A, E171A, S209A, D219A were preincubated with 8-O (25 μM) at 25 °C for 0.5 h, and then incubated with biotin-streptavidin complex at 25 °C for 2 h. Following TBS-T buffer wash, the proteins that had attached to the beads were analyzed by immunoblotting using antibody to SecA or BamD. Site-mutated plasmids of SecA and BamD were obtained by seamless cloning and confirmed by sequencing (DIA-UP Biotech, Bei**g, China).

RNA extraction and transcriptome analysis

Briefly, H. pylori cells were incubated with/without 0.25 μg/mL 8 for 2 days, then collected. Extraction of RNA was carried out using an RNAsimple total RNA Kit (TIANGEN, China) according to the manufacturer’s instructions. RNA quantity measurement and transcriptome analysis was conducted by Allwegene Tech (Bei**g, China). Then sequencing was performed in an Illumina Hiseq platform with paired ends mode. The RNA-Seq raw data was filtrated and employed to assemble the bacterial transcriptome using Cutadapt software and RNA-Seq reads were mapped using the Bowtie2 software against the genome of H. pylori ATCC 43504. The experiment was designed and performed with three biological replicates.

Quantitative PCR with reverse transcription (RT-qPCR)

Total RNA was extracted as mentioned above. An aliquot (500 ng) of the total RNA from each sample was subjected to cDNA synthesis using HiScript® III All-in-one RT SuperMix (Vazyme, China). The housekee** gene 16S rRNA was used as control. The cDNAs were quantified by qPCR using Taq Pro Universal SYBR qPCR Master Mix (Vazyme) on a qTOWER®3 Real-Time PCR Systems (Analytik Jena) according to the manufacturers’ instructions. Primers used in this study are listed in Supplementary Table S11. All assays were carried out in three independent experiments in triplicates. The relative RNA expression levels were calculated according to ΔΔCt with normalization to the 16S rRNA levels.

SecA ATPase activity

Purified recombinant SecA was incubated at 37 °C with different concentrations of 8 for 1 h in a reaction buffer. ATP was then added, and the mixture was further incubated at 37 °C for another 40 min. The amount of ATP converted into adenosine diphosphate (ADP) was determined by luminescent ADP detection with an ADP Glo Kinase Assay kit (Promega) according to the manufacturer’s protocol.

Co-IP assay

To prove that 8 influences the interaction between BamD and BamA, the solution of purified His-tagged BamA and GST-tagged BamD was treated with or without 8 for 2 h, immunoprecipitated with anti-His beads (Abbkine), and then assessed by immunoblot analysis.

H. pylori OMPs extraction and quantification

H. pylori cells were harvested, washed and suspended in 20 mM Tris-HCl (pH 7.5). DNase and protease inhibitors were added to the cell suspension, and the bacteria cells were disrupted by repeated ultrasonication. Cell debris was removed by centrifugation (9000 × g, 20 min). Total membrane pellet was collected by centrifugation (50,000 × g, 40 min), resuspended in 20 mM Tris-HCl containing 2.0% sodium lauryl sarcosine, and incubated at room temperature for 30 min. The sarcosine-insoluble constituent (OMPs) was collected by centrifugation (50,000 × g, 30 min), washed with 20 mM Tris-HCl (pH 7.5), and dissolved in XT Sample Buffer (Bio-Rad) with 50 mM DTT.

Fluorescence staining and H. pylori adhesion assays

Briefly, GES-1 cells were seeded and cultured on cover glass bottom dishes, and then infected with FITC-labeled H. pylori. After a 3-day incubation, H. pylori cells were harvested from agar plates with/without 8 exposures, re-suspended in 0.15 M NaCl and 0.1 M Na2CO3, and adjusted to 1.0 McFarland. Ten microliters of freshly prepared 1% FITC in DMSO were added to the suspension and incubated for 1 h in the dark. Bacteria were recovered by centrifugation at 3000 × g for 5 min, and resuspended in 1.0 mL PBS supplemented with 5% inactivated fetal bovine serum, 0.2% BSA and 0.05% Tween-20. The FITC-labeled H. pylori cells were added into the dishes and incubated 4 h at 37 °C. After three washes with PBS to remove nonadherent bacteria, paraformaldehyde PBS solution (4%) was applied to immobilize for 10 min. The cell membrane and nucleus were stained with DiD (5 μM, 20 min) and DAPI (10 μg/mL, 5–10 min) at room temperature sequentially, respectively. After fully removing the stains, the H. pylori adhesion to GES-1 cells were observed and imaged under confocal laser microscope (LSM710, Zeiss, Germany).

Statistical analyses

Statistical analysis was performed with one-way ANOVA by SPSS 16.0 with all data points showing a normal distribution, *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001. No exclusion of data points was used.