Genomic Investigation of Salmonella Isolates Recovered From a Pig Slaughtering Process in Hangzhou, China

The pig industry is the principal source of meat products in China, and the presence of pathogens in pig-borne meat is a crucial threat to public health. Salmonella is the major pathogen associated with pig-borne diseases. However, route surveillance by genomic platforms along the food chain is still limited in China. Here, we conducted a study to evaluate the dynamic prevalence of Salmonella in a pig slaughtering process in Hangzhou, Zhejiang Province, China. Fifty-five of 226 (24.37%) samples were positive for Salmonella; from them, 78 different isolates were selected and subjected to whole genome sequencing followed by bioinformatics analyses to determine serovar distribution, MLST patterns, antimicrobial resistance genes, plasmid replicons, and virulence factors. Moreover, phenotypic antimicrobial resistance was performed using the broth dilution method against 14 antimicrobial agents belonging to 10 antimicrobial classes. Our results showed that samples collected from the dehairing area (66.66%) and the splitting area (57.14%) were the most contaminated. Phenotypic antimicrobial resistance classified 67 of 78 isolates (85.90%) as having multidrug resistance (MDR), while the highest resistance was observed in tetracycline (85.90%; 67/78) followed by ampicillin (84.62%; 66/78), chloramphenicol (71.80%; 56/78), and nalidixic acid (61.54%; 48/78). Additionally, serovar prediction showed the dominance of Salmonella Typhimurium ST19 (51.28%; 40/78) among the 78 studied isolates, while plasmid prediction reported the dominance of IncHI2A_1 (20.51%; 16/78), followed by IncX1_1 (17.95%; 14/78) and IncHI2_1 (11.54%; 9/78). Virulence factor prediction showed the detection of cdtB gene encoding typhoid toxins in two Salmonella Goldcoast ST358 and one Salmonella Typhimurium ST19, while one isolate of Salmonella London ST155 was positive for genes encoding for the siderophore “yersiniabactin” and the gene senB encoding for enterotoxin production. From this study, we conclude that pig slaughterhouses are critical points for the dissemination of virulent and multidrug-resistant Salmonella isolates along the food chain which require the implementation of management systems to control the critical points. Moreover, there is an urgent need for the implementation of the whole genome sequencing platform to monitor the emergence of virulent and multidrug-resistant clones along the food chain.

The pig industry is the principal source of meat products in China, and the presence of pathogens in pig-borne meat is a crucial threat to public health. Salmonella is the major pathogen associated with pig-borne diseases. However, route surveillance by genomic platforms along the food chain is still limited in China. Here, we conducted a study to evaluate the dynamic prevalence of Salmonella in a pig slaughtering process in Hangzhou, Zhejiang Province, China. Fifty-five of 226 (24.37%) samples were positive for Salmonella; from them, 78 different isolates were selected and subjected to whole genome sequencing followed by bioinformatics analyses to determine serovar distribution, MLST patterns, antimicrobial resistance genes, plasmid replicons, and virulence factors. Moreover, phenotypic antimicrobial resistance was performed using the broth dilution method against 14 antimicrobial agents belonging to 10 antimicrobial classes. Our results showed that samples collected from the dehairing area (66.66%) and the splitting area (57.14%) were the most contaminated. Phenotypic antimicrobial resistance classified 67 of 78 isolates (85.90%) as having multidrug resistance (MDR), while the highest resistance was observed in tetracycline (85.90%; 67/78) followed by ampicillin (84.62%; 66/78), chloramphenicol (71.80%; 56/78), and nalidixic acid (61.54%; 48/78). Additionally, serovar prediction showed the dominance of Salmonella Typhimurium ST19 (51.28%; 40/78) among the 78 studied isolates, while plasmid prediction reported the dominance of IncHI2A_1 (20.51%; 16/78), followed by IncX1_1 (17.95%; 14/78) and IncHI2_1 (11.54%; 9/78). Virulence factor prediction showed the detection of cdtB gene encoding typhoid toxins in two Salmonella Goldcoast ST358 and one Salmonella Typhimurium ST19, while one isolate of Salmonella London ST155 was positive for genes encoding for the siderophore "yersiniabactin" and the gene senB encoding for enterotoxin production. From this study, we conclude that pig slaughterhouses are critical points for the dissemination of virulent

INTRODUCTION
Salmonellosis is a global zoonotic disease, caused by Salmonella and characterized by self-limited gastroenteritis in immunocompetent adults, in which typical symptoms like diarrhea, fever, abdominal cramps, and vomiting occur between 6 and 72 h (usually 12-36 h) after ingestion of bacteria and the illness lasts from 2 to 7 days [World Health Organisation (WHO), 2018]. It might also cause severe invasive infection, particularly in immunocompromised patients (Deen et al., 2012;. Recently, it was estimated that Salmonella was responsible for about 180 million (9%) of the diarrheal illnesses that occur globally each year, causing about 298,000 deaths (41%) of all diarrheal disease-associated deaths (Besser, 2018). In China, a study based on the literature review estimated that the incidence of nontyphoidal salmonellosis was 626.5 cases per 100,000 persons (Mao et al., 2011;. Moreover, it has been reported that Salmonella was responsible for approximately 70∼80% of foodborne pathogenic outbreaks in China (Jun et al., 2007).
Salmonella spp. are Gram-negative rod-shaped bacteria, facultatively anaerobic, and belong to the family Enterobacteriaceae. To date, more than 2,600 serovars have been described among Salmonella species; among them, only a few serovars were mostly linked to human and/or animal infections, including Typhimurium and Enteritidis (so-called majority serovars) for human infections , Gallinarum and Pullorum for poultry infections , Dublin for cattle infections , and Choleraesuis and Typhisuis for pig infections (Boyen et al., 2008;Asai et al., 2010). Generally, animal farms are considered natural reservoirs of Salmonella, especially poultry and pigs (Li et al., 2013;Zhou et al., 2017;. Salmonella could colonize the digestive tract of animals and are excreted in feces and spread into the environment (Kagambèga et al., 2013;Bonardi, 2017;Jiang et al., 2019), then transmitted to humans via the food chain (Ed-Dra et al., 2018;Wang et al., 2019;Wilson et al., 2020;Liu et al., 2021). Therefore, several studies have reported the presence of Salmonella in foods of animal origin, especially meat products (Ed-Dra et al., 2017;Jiang et al., 2021;Liu et al., 2021).
Pork meat is considered the most frequently contaminated food and the major source of Salmonella infections in humans (Bonardi, 2017;Wilson et al., 2020). In fact, pig farms seem to be a suitable environment for the replication and the persistence of Salmonella (Lettini et al., 2016;Bonardi, 2017;Vico et al., 2020). However, the slaughtering process which is located downstream of the pig-breeding process and upstream of pork sales is a critical step in determining the contamination/decontamination of animal carcasses and thus the meat products (Zhou et al., 2018). Moreover, the application of good hygienic practices in slaughterhouses has great importance and could participate in reducing the prevalence of Salmonella in the final meat products (Rahkio and Korkeala, 1996;Biasino et al., 2018). During the slaughtering process, animals pass through different processing stages with complicated manipulations (Zhou et al., 2018). However, since pigs are considered reservoirs of pathogens, they could contaminate/cross-contaminate the carcasses or muscle tissues during the slaughtering process. In fact, it has been demonstrated that slaughter practices, such as splitting the head and incising tonsils, were associated with higher levels of hygiene indicator bacteria and Salmonella in pig carcasses (Biasino et al., 2018). Therefore, the surveillance of Salmonella along the slaughtering process and its environment is with a high priority to determine the key points that are responsible for the contamination of carcasses and the final meats.
Recently, whole genome sequencing followed by bioinformatics analysis was considered as a cost-effective method for the diagnosis and characterization of foodborne pathogens (Biswas et al., 2020;Liu et al., 2020Liu et al., , 2021Yu et al., 2020). As proof-of-concept, we conduct a study in a pig slaughterhouse in Hangzhou (Zhejiang Province, China), to obtain Salmonella isolates from different sources. The recovered strains were subjected to whole genome sequencing followed by in silico analysis to determine serovar distribution, multilocus sequence types, plasmid replicons, antimicrobial resistance, and virulence genes. Moreover, phenotypical antimicrobial resistance was investigated by the broth dilution method and compared with genotypical resistance.

Sample Collection and Characterization of Salmonella
The present study was conducted in Linpu Pig Slaughterhouse in Xiaoshan, Hangzhou (China). The capacity of the studied slaughterhouse was approximately 1,000 pigs per day. A sampling visit was organized during December 2018 allowing the collection of 226 samples from different origins (pig carcasses, swab samples, environmental samples, equipment samples, operator samples, intestinal content samples, hepatobiliary samples, and sewer samples) along the slaughtering process of pigs ( Table 1). The sampling method was in accordance with those described in previous studies Zhou et al., 2017). The isolation of Salmonella was performed from different samples according to the protocols described previously (Jiang et al., 2019;Liu et al., 2021). Then, molecular confirmation of presumptive isolates was carried out by the amplification of invA gene according to the protocol previously described (Zhu et al., 2015;Liu et al., 2021). Among the 226 collected samples, 55 were positive for the presence of Salmonella. However, since some samples present more than one presumptive isolate, we decided to select 78 different Salmonella isolates that show differences in morphological and biochemical criteria for genome sequencing and analysis ( Table 2). Serotyping of the PCR confirmed Salmonella isolates were performed according to White-Kauffmann-Le Minor scheme by slide agglutination method to define O and H antigens using commercial antisera (SSI Diagnostica, Hillerød, Denmark).

Genomic DNA Extraction
All the obtained Salmonella isolates (n = 78) were selected for genomic DNA extraction according to the protocol described previously (Liu et al., 2021). Briefly, a broth culture of each Salmonella isolate was prepared by inoculation of a pure colony in Luria-Bertani broth followed by incubation at 37 • C under 180 rpm shaking conditions. Then, DNA extraction was conducted by using TIANamp bacteria DNA kit (Tiangen Biotech, Beijing, China) according to the instructions of the manufacturer. The quantification of the extracted DNA was performed by the Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, CA, United States), as per the instructions of the manufacturer.

Genomic Sequencing and Bioinformatic Analysis
The genomic DNA library was constructed using NovaSeq XT DNA library construction kit (Illumina, San Diego, CA, United States, No: FC-131-1024), followed by genomic sequencing using Illumina NovaSeq Platform with NovaSeq 6000 SP Reagent Kit (300 cycles). The raw sequence reads were checked for quality and assembled using SPAdes v3.12.0 (Bankevich et al., 2012). Virulence gene prediction was conducted based on the virulence factors database (VFDB) (Chen et al., 2005). Moreover, in silico serotyping of Salmonella strains was performed by the SISTR web tool, whereas sequence types, antimicrobial resistance genes (ARG), and plasmid replicons were detected using the assemblies of the samples on the in-house Galaxy platform (Afgan et al., 2016), in combination with mlst v2.16.1 1 and abricate v0.8 (Zankari et al., 2012), including the CGE ResFinder database (updated on February 19, 2021) with a similarity cutoff of 90% for ARG and PlasmidFinder database (updated on February 19, 2021) with a similarity cutoff of 95% (Carattoli et al., 2014).

Phenotypic Antimicrobial Resistance Testing
The antimicrobial resistance of the isolated Salmonella strains was evaluated phenotypically by the broth dilution method to determine the minimum inhibitory concentration (MIC) of a panel of 14 antimicrobial agents belonging to 10 antimicrobial classes according to the protocol described previously Liu et al., 2021

Salmonella Prevalence, MLST Pattern, and Serovar Distribution
The results obtained in this study showed that 55 of 226 (24.37%) samples were contaminated by Salmonella (Table 1). According to the sampling points along the pig slaughtering process, our results showed that the samples collected from the dehairing area were the most contaminated (66.66%), followed by those collected from the splitting area (57.14%). However, samples collected from the washing area and scalding area were not contaminated (Table 1). Additionally, from the 55 samples, 78 different Salmonella isolates were obtained, purified, and subjected to whole genome sequencing. The genomic prediction of serovars and MLST patterns showed the distribution of five different serovars and six MLST patterns, namely, Typhimurium ST19 (n = 40), Typhimurium ST34 (n = 14), London ST155 (n = 14), Rissen ST469 (n = 7), Goldcoast ST358 (n = 2), and Derby ST40 (n = 1) ( Table 3). Additionally, serotyping performed by in silico analysis and slide agglutination methods provided the same results.

Phenotypic Antimicrobial Resistance
The antimicrobial resistance of the isolated Salmonella strains was evaluated against 14 antimicrobial agents belonging to 10 classes or categories. The phenotypic antimicrobial profiles were classified as resistant, susceptible, and intermediate according to the criteria of the Clinical Laboratory Standard Institute guidelines and the results are presented in Table 4 and  Figure 1A). According to the sources, it appears that Salmonella isolates recovered from sewer samples (SS) and hepatobiliary samples (HS) were more resistant to the tested antimicrobial agents compared with those collected from other sources ( Figure 2B). Moreover, among different serovars identified in this study, Salmonella serovars Derby and Goldcoast appear to be the most resistant to the tested antimicrobial agents (Figure 2A). However, it should be noted that this conclusion cannot be generalized since only one strain of Salmonella Derby and two strains of Salmonella Goldcoast have been identified in this study. Additionally, Salmonella Typhimurium isolates from ST34 appear to be more resistant than isolates from ST19 (Figure 2A).

Antimicrobial Resistance Gene Prediction
The whole genome sequences of the 78 isolated Salmonella strains were subjected to in silico detection of antimicrobial resistance genes. The results obtained showed the detection of 35 different genes encoding resistance to nine antimicrobial classes (Figure 3 and Supplementary Material 2). The most detected genes were bla TEM−1B encoding resistance to penicillins (74.36%; 58/78), sul2 encoding resistance to sulfonamides (87.93%; 51/58), tet(A) encoding resistance to tetracyclines (64.10%; 50/78), floR encoding resistance to phenicols (64.10%; 50/78), and qnrS1 encoding resistance to fluoroquinolones (60.26%; 47/78) ( Figure 1C). Moreover, 64 of 78 isolates (82.05%) harbor the resistance genes of more than two classes ( Figure 1B). However, regarding the serovar distribution, it appears that Salmonella Typhimurium ST34 harbors more diversified antimicrobial resistance genes while Salmonella London ST155 appears to be poor in resistance genes (only one strain that harbors the genes cat and tet(J) encoding resistance to phenicols and tetracyclines classes, respectively) ( Figure 4A). Moreover, our results showed that Salmonella isolates obtained from carcass swabs after splitting (CSAS) and HS harbor more resistance genes compared with those isolated from other sources ( Figure 4B).
The relation between phenotypical antimicrobial resistance and the presence/absence of corresponding resistance gene obtained by in silico analysis was evaluated and the results are presented in Table 5. Our results showed that ciprofloxacin has the higher incoherence percentage (55.13%; 43/78) for which several isolates were genotypically positive but phenotypically negative, while cefoxitin presents the lower incoherence percentage (1.28%; 1/78).

Virulence Gene Prediction
In this study, the presence of 117 genes that are implicated in virulence and pathogenicity mechanisms of Salmonella was evaluated among the genomes of the 78 Salmonella isolates. The results are summarized in Supplementary Material 3. Our results showed that the number of detected genes ranged from 88 to 113 per isolate. Among the 78 isolates, three isolates (two Salmonella Goldcoast ST358 and one Salmonella Typhimurium ST19) were positive for the gene cdtB encoding typhoid toxin production, and these isolates were all isolated from fecal samples (FS). Additionally, only one Salmonella isolate (Salmonella London ST155) isolated from CSBS sample was positive for the genes encoding for the siderophore "yersiniabactin" (fyuA, ybtA, ybtE, ybtP, ybtQ, ybtS, ybtT, ybtU, ybtX, irp1, and irp2) and for the gene senB encoding for enterotoxin production. However, the typical virulence factors carried on Salmonella Pathogenicity Island 1 and 2 (SPI-1 and SPI-2) were detected in all the studied isolates.

Plasmid Profiles
The results of the prediction of plasmid replicons in the 78 Salmonella isolates are presented in Figure 5 and

DISCUSSION
Pig slaughterhouses are critical points of the meat processing chain; they are situated downstream of the pig-breeding process and upstream of pork sales. Since reception, animals undergo different complicated manipulations and are in contact with slaughterhouse facilities, workers, etc., which favors the contamination/cross-contamination of animal carcasses and, thus, the meat products (Arguello et al., 2013;Zhou et al., 2018). However, comparison between the contamination rate of Salmonella in pigs in the preslaughter stage and in the postslaughter stage revealed that the prevalence in the preslaughter stage often seems to be lower (Jiang et al., 2019). In this regard, Colello and his group conducted a study along the production chain of pig farms and showed that the prevalence of Salmonella in farms (2.6%) and slaughterhouses (2.0%) was lower than that observed in boning rooms (8.8%) and retail markets (8.0%) (Colello et al., 2018). Additionally, Jiang et al. reported that the prevalence of Salmonella in pigs at the farm stage was 11.77%, lower than that observed in the slaughtered pigs (45.23%) (Jiang et al., 2019), demonstrating the criticality of the slaughtering process in determining the quality and safety of derived pig food products.
In this regard, we conducted a study to evaluate the prevalence of Salmonella during the pig slaughtering process. Our results showed that 55 of the 226 samples (24.37%) were contaminated by Salmonella. These results were lower than those reported previously in pig slaughterhouses in other Chinese regions (ranged between 29.2 and 46.6%) (Bai et al., 2015;Li et al., 2016; , 2018). According to the slaughtering process, samples recovered from the dehairing area and splitting area were the most contaminated samples. In the dehairing area, the frequently used knife for carcass modification FIGURE 4 | Heatmap of antimicrobial resistance genes according to serovars (A) and sampling sources (B). The isolates of Salmonella Typhimurium ST34 harbor the most diversified antimicrobial resistance genes. However, Salmonella isolates recovered from HS and CSAS contain more resistance genes compared with those isolated from other sources. CSBS, carcass swabs before splitting; CSAS, carcass swabs after splitting; HS, hepatobiliary samples; FS, fecal samples; SS, sewer samples. was considered as the risk factor for the observed carcass crosscontamination. However, the splitting step located at the next step after evisceration has been confirmed as the other step with a higher risk of Salmonella contamination Zhou et al., 2017). After evisceration, the intestinal content can contaminate a part of animal carcasses; however, during splitting, the splitter could be contaminated and then cross-contaminate other carcasses, resulting in the increase of Salmonella prevalence in the splitting area Li et al., 2016). Therefore, the implementation of good hygienic practices and management systems to control critical points during the slaughtering process is of high priority to reduce the prevalence of Salmonella. Among the 55 positive samples, 78 different Salmonella isolates were identified in this study. These isolates belong to five different serovars and six MLST patterns, namely, with importance degree, Salmonella Typhimurium ST19, Salmonella Typhimurium ST34, Salmonella London ST155, Salmonella Rissen ST469, Salmonella Goldcoast ST358, and Salmonella Derby ST40. In China, Salmonella Derby was identified as the most isolated serovar from pig slaughterhouse samples Zhou et al., 2017Zhou et al., , 2018Liu et al., 2020). However, Salmonella Typhimurium has been reported previously as the dominant serovar in Salmonella isolates recovered from pig slaughterhouses in Henan Province (Bai et al., 2015). In fact, it is well known that Salmonella Typhimurium was classified among the major serovars causing human salmonellosis worldwide (CDC, 2018;EFSA and ECDC, 2018), especially those with multilocus sequence types ST19 and ST34, which were reported in several cases of human infections (Wong et al., 2013;Carden et al., 2015;Panzenhagen et al., 2018;Luo et al., 2020;Monte et al., 2020). Therefore, the transmission of these isolates to the final meat products along the food chain is of high risk for consumers and may cause severe cases of foodborne diseases.
The infections caused by Salmonella are treated with different antimicrobial drugs. However, in the last decades, development of Salmonella resistance to many antimicrobials has been observed worldwide, either for the isolates provided from clinical, food, and environmental samples. In this study, the phenotypical and genotypical antimicrobial resistance profiles of the 78 isolated Salmonella strains were evaluated. Phenotypical results classified tetracycline and ampicillin as the less effective antimicrobial agents. In fact, the high resistance of Salmonella isolates to tetracycline and ampicillin has been reported over the world in samples collected along the animal food chain (Ed-Dra et al., 2017;Jiang et al., 2019Jiang et al., , 2021Wang et al., 2019;Liu et al., 2020Liu et al., , 2021, since they were frequently used in animal farms (Lekagul et al., 2019). In fact, the abuse and the misuse of antimicrobial drugs in animal livestock for therapeutic, prophylaxis, and growth promotions have led to the development of antimicrobial resistance. Moreover, our results showed that 85.90% of isolates/strains were resistant to more than two antimicrobial classes (MDR), which is considered a serious threat to public health that leads to therapeutic failure after a simple infection by MDR isolates.
Genotypical antimicrobial resistance prediction showed the detection of 35 resistance genes encoding resistance to nine antimicrobial classes, with a high prevalence of bla TEM−1B gene encoding resistance to penicillins, sul2 gene encoding resistance to sulfonamides, tet(A) gene encoding resistance to tetracyclines, floR gene encoding resistance to phenicols, and qnrS1 gene encoding resistance to fluoroquinolone. The presence of these genes in bacterial genomes could be responsible for the acquisition of resistance to the corresponding antimicrobial classes. However, the analysis of coherence between genotypic and phenotypic antimicrobial resistance showed that phenotypic resistance cannot always be linked to the presence of resistance genes. Our results are in agreement with those reported previously in Salmonella isolates, showing a difference between phenotypic and genotypic resistance profiles (Liu et al., 2020(Liu et al., , 2021. Hence, the phenotypic test remains the gold method for the assessment of bacterial behavior toward antimicrobial agents. The prediction of virulence genes implicated in virulence and pathogenicity mechanisms reveals the detection of 117 different genes, particularly the detection of cdtB gene encoding typhoid toxins in two isolates of Salmonella Goldcoast ST358 and one isolate of Salmonella Typhimurium ST19 and the detection of genes encoding for the siderophore "yersiniabactin" in one isolate of Salmonella London ST155, and this isolate also harbors the gene encoding for the enterotoxin TieB (senB). In fact, it has been reported that the presence of cdtB in the Salmonella genome was linked to isolates implicated in human bloodstream and invasive infections (Miller et al., 2018;. Additionally, yersiniabactin siderophore that was initially described in Yersinia spp. is required for iron uptake and growth of the bacteria in an iron-restricted environment (Perry and Fetherston, 2011;Khan et al., 2018). However, the enterotoxin TieB was initially described in enteroinvasive E. coli (EIEC) (Nataro et al., 1995) and has been suggested to play a key role in bacteria virulence in humans (Meza-Segura et al., 2020). Indeed, the presence of these virulence genes in the genome of Salmonella isolated from the pig slaughtering process may lead to severe disease outcomes in humans.
In this study, nine different plasmid replicons were detected among the 78 Salmonella isolates. The most abundant plasmids were IncHI2A_1, IncX1_1, and IncHI2_1. IncHI2A_1 and IncHI2_1 were predominant in Salmonella Typhimurium ST34, while IncX1_1 was detected only in Salmonella London ST155. These plasmids were identified previously in Salmonella isolates recovered from the animal food chain, especially pork production chains (Liu et al., 2020(Liu et al., , 2021Viana et al., 2020). Interestingly, it has been demonstrated that these plasmids were associated with resistance to different antimicrobial classes, including β-lactams, aminoglycosides, sulfonamides, tetracyclines, and polymyxins (Elbediwi et al., 2020b,a;Gu et al., 2020;McMillan et al., 2020). Consequently, these plasmids may mediate the horizontal transmission of antimicrobial resistance genes during this slaughtering process.

CONCLUSION
In this study, we provided the dynamic prevalence of Salmonella during the pig slaughtering process. Additionally, we demonstrated the use of whole genome sequencing as a cost-effective approach for routine surveillance of foodborne pathogens, especially Salmonella. The prediction of serovar distribution, MLST patterns, antimicrobial resistance genes, plasmid replicons, and virulence factors in Salmonella isolates recovered from the pig slaughtering process showed the isolation of MDR isolates harboring different antimicrobial resistant determinants and virulence factors like cdtB gene encoding typhoid toxins, senB gene encoding for the enterotoxin production, and several genes encoding for the siderophore "yersiniabactin." Therefore, it is time to prevent the use of antimicrobials in animal livestock in order to avoid the dissemination of antimicrobial resistance determinants along the food chain and to implement management systems to control critical points in order to avoid the transmission of foodborne pathogens to humans.

DATA AVAILABILITY STATEMENT
The datasets generated for this study can be found in the NCBI Bioproject with the accession number no. PRJNA686895.

AUTHOR CONTRIBUTIONS
BW, AE-D, and HP contributed equally to this work. AE-D and HP analyzed the data and finalized the figures. AE-D and MY were wrote the manuscript. BW, HP, CD, and CJ did the experiment and data collection. MY conceived the idea and assisted with data analysis and writing. All authors read, revised, and approved the final manuscript.