Metagenomic analysis of isolation methods of a targeted microbe, Campylobacter jejuni, from chicken feces with high microbial contamination

Background Originating from poultry, particularly chickens, Campylobacter jejuni is the leading foodborne pathogen worldwide and a major cause of campylobacteriosis. Isolating C. jejuni is difficult due to its specific growth requirements, the presence of viable but non-culturable bacteria, and because it is often masked by competing flora. Currently, there is no optimized method for isolating C. jejuni from chicken feces. Here, we evaluated the method for isolating C. jejuni from chicken feces using culture-independent sequence-based metagenomics and culture-dependent tools. Further, we assessed changes in microbial communities during microbe isolation to determine how the process can be improved. Results Fourteen different variations of C. jejuni isolation procedures were applied to all 35 chicken fecal samples. These variations included using different enrichment broths (without enrichment or enrichment in Bolton or Preston broth), different ratios of sample-to-enrichment broth (1:101, 1:102, and 1:103), and different selective agars (modified charcoal-cefoperazone-deoxycholate agar (mCCDA) or Preston agar). Enrichment during isolation of C. jejuni was evaluated on the basis of microbial diversity and taxonomic composition using metagenomics tools. The effect of selective media was evaluated using a combination of metagenomics and culture-dependent tools. Microbial diversity significantly decreased during the enrichment process, regardless of the type of enrichment broth, with the most significant decrease observed at a feces-to-broth ratio of 1:103. Particularly, in 103-Preston broth, the relative abundance of Campylobacter increased, while extended-spectrum beta-lactamase-producing Escherichia coli, which interfere with Campylobacter isolation, decreased. Metagenomics results were validated by quantitative PCR and culture-dependent analysis. Additionally, selective media affected the isolation results, although microbes with high relative abundance during enrichment were also frequently isolated using culture-dependent methods. Significantly more C. jejuni was isolated from mCCDA than from Preston agar enriched in 103 Preston broth. Conclusions Enrichment in Preston broth at a ratio of 1:103 followed by spreading onto mCCDA was the most effective method for isolating C. jejuni. This is the first study to apply metagenomics to evaluate a method for isolating a targeted microbe, C. jejuni, from chicken feces, a source with high microbial contamination. Thus, metagenomics can be applied to improve methods for isolating bacteria that are difficult to separate. Electronic supplementary material The online version of this article (10.1186/s40168-019-0680-z) contains supplementary material, which is available to authorized users.


Background
Campylobacter spp., particularly Campylobacter jejuni, is a leading foodborne pathogen, worldwide [1]. Approximately nine million people in the European Union and 1.3 million people in the USA suffer from the Campylobacter-borne disease, and the number of infected people increases each year [2,3]. A commensal organism that naturally colonizes and is asymptomatic in the gastrointestinal tract of poultry, particularly chickens, C. jejuni is a major cause of human campylobacteriosis [4,5]. Symptoms of C. jejuni infections in humans include gastro-intestinal distress such as abdominal pain and diarrhea, as well as neurological issues including Guillain-Barre and Miller-Fisher syndrome [6]. Therefore, for the prevention and management of campylobacteriosis, isolation of C. jejuni from chicken should be prioritized. However, unlike other common foodborne pathogens, including pathogenic Escherichia coli and Salmonella spp., C. jejuni is difficult to isolate because of its specific growth conditions (i.e., requires a microaerophilic environment) [7]. The presence of viable but nonculturable (VBNC) C. jejuni and the tendency of C. jejuni to be masked by competing microbes, such as extended-spectrum beta-lactamase (ESBL)-producing E. coli, make it challenging to isolate C. jejuni from chickens [8,9]. The most common methods for isolating C. jejuni from chicken meat are pre-enrichment (4-5 h, microaerophilic conditions, 37°C) and enrichment (48 h, microaerophilic conditions, 42°C) procedures in Bolton broth or Preston broth, followed by selection (48 h, microaerophilic conditions, 42°C) on modified charcoalcefoperazone-deoxycholate agar (mCCDA) or Preston agar. However, the prevalence of C. jejuni infections varies between studies, as there is no standard method, and outcomes differ depending on the type of enrichment broth and selective agar used [1,10]. Additionally, most studies were performed using chicken meat. Because Campylobacter is a commensal flora of the gastrointestinal tract in poultry, it is important to isolate Campylobacter from the farm stage (feces of chicken). However, a method for isolating C. jejuni from chicken feces has not been established, and few studies have been performed [11].
The need for an effective method of isolating C. jejuni from chicken meat and feces has increased. Several studies were conducted comparing currently used methods for isolating C. jejuni and improving isolation by adding several antibiotics, such as polymyxin B and triclosan, to the enrichment broth or selective agar [7][8][9]12]. However, most studies were carried out using culture-based techniques, which rely on the isolation results of a specific microbe, such as Campylobacter spp. and ESBL-producing E. coli. Changes in the overall microbial community and composition of specific microbes affecting the isolation of C. jejuni during isolation procedure have not been identified. Additionally, evaluation of the enrichment procedure using culture-based techniques, which greatly affects microbe isolation compared to selective procedures, is not possible [13]. Therefore, to accurately compare isolation methods of C. jejuni, studies based on metagenomics using next-generation sequencing are needed.
Currently, various next-generation sequencing (NGS) systems, such as Roche 454 pyrosequencing, Illumina Genome Analyzer (HiSeq, MiSeq), Applied Biosystems SOLiD System, Life Technologies Ion Torrent, and the PacBio RX system, are available in microbiological research fields [14]. Advances in these NGS technologies have enabled comprehensive analyses in microbiome analysis and whole-genome-sequencing analysis [15,16]. However, most microbiological studies using these approaches have been conducted on the gut microbiome, which is closely linked to host health status [17][18][19]. Studies have been conducted to identify specific microorganisms based on sequence analysis but only for disease diagnosis [20]. However, disease diagnosis, alone, is not suitable for epidemiologic investigations or analyzing pathogenic factors, such as susceptibility to antibiotics, serotypes, subtypes, and virulence factors. These studies are possible only if the target microbes are isolated. Application of metagenomics approaches can improve the methods used to isolate specific microbes, particularly pathogens, by evaluating changes in microbial communities during pathogen isolation procedures.
The current study was carried out to evaluate the method for isolating C. jejuni from chicken feces through culture-independent sequence-based metagenomics tools and culture-dependent tools. First, the microbial community change, including alpha diversity, beta diversity, and taxonomic composition of chicken feces during the isolation procedures of C. jejuni, was analyzed for a total of seven procedures with different combinations of enrichment broth (enriched in Bolton or Preston broth or a no-enrichment process) and the ratio of sample-to-enrichment media (1:10 1 , 1:10 2 , and 1:10 3 ). Second, the isolation results of C. jejuni and competing colonies in a total of 14 procedures, according to the combinations of the above seven procedures and type of selective media (mCCDA and Preston agar), were compared. Finally, the metagenomics results were validated by comparing the sequencing results to the culture-dependent results. mCCDA and 57.1% (20/35) from Preston agar (Fig. 1). In samples enriched in Bolton broth, C. jejuni was not isolated, regardless of the ratio of sample-to-enrichment broth and types of selective agars. In Preston broth-mCCDA procedures, C. jejuni was not isolated from the 1:10 1 ratio of sample-to-enrichment broth, while 31.4% (11/35) and 97.1% (34/35) of C. jejuni was isolated at ratios of 1:10 2 and 1:10 3 , respectively. In the Preston broth-Preston agar procedures, C. jejuni was not isolated from the 1:10 1 and 1:10 2 ratios of sample-tobroth, while 5.7% (2/35) of C. jejuni was isolated from the 1:10 3 ratio. The isolation rate of C. jejuni was significantly higher in the 10 3 -Preston broth-mCCDA procedure than in the other procedures (chi-square test, p < 0.05) and significantly higher in the 10 2 -Preston broth-mCCDA procedure than in the other procedures except for the 10 3 -Preston broth-mCCDA procedure (chi-square test, p < 0.05, Table 1). Fig. 1 A schematic diagram of the current study. All chicken fecal samples were analyzed to compare isolation methods of C. jejuni using cultureindependent sequence-based metagenomics and culture-dependent tools. Metagenomic analysis was performed to investigate the effects of seven enrichment procedures. Six procedures involved enrichment processes (Bolton and Preston broth at 1:10 1 , 1:10 2 , and 1:10 3 , respectively), and one procedure did not. In culture-dependent analysis, a total of 14 procedures were applied for all samples, which were combinations of different enrichment broths (without enrichment process or enriched in Bolton broth or Preston broth), ratio of sample-to-enrichment broth (1:10 1 , 1:10 2 , or 1:10 3 ), and selective media (mCCDA or Preston agar). a The ratio of fecal sample-to-enrichment broth  Additionally, ESBL-producing E. coli (all isolated E. coli were identified as ESBL-producing), Proteus mirabilis, and Enterococcus spp. were found in each procedure as colonies competing with C. jejuni (Table 1). For ESBL-producing E. coli, the isolation rate was significantly higher in mCCDA than in Preston agar (same enrichment broth type and ratio of sample-to-enrichment broth) and significantly higher in Bolton broth than in Preston broth (same ratio of sample-to-enrichment broth and selective media type) (chi-square test, p < 0.05, respectively). The isolation rate of ESBL-producing E. coli according to the ratio of sample-to-enrichment broth (in the same type of enrichment broth and selective agar) was significantly higher in 10 3 -Bolton broth/ Preston agar and significantly lower in 10 3 -Preston broth/mCCDA (chi-square test, p < 0.05, respectively). For P. mirabilis, the isolation rate was significantly higher in Preston agar than in mCCDA (same enrichment broth type and the ratio of sample-to-enrichment broth) (chi-square test, p < 0.05). Enterococcus spp. was not isolated using the procedures enriched in Bolton broth, and the isolation rate was significantly higher in mCCDA than in Preston agar (same enrichment broth type and the ratio of sample-to-enrichment broth) (chi-square test, p < 0.05).
Proportion of competing colonies and its relationship with isolation result of C. jejuni according to each procedure determined using culture-dependent tools The ratio of the number of each colony to the total number of isolated colonies in the plate except for C. jejuni was used as a criterion for determining the proportion of competing colonies. Comparison of the proportion of each microbe according to enrichment type revealed that the proportions of ESBL-producing E. coli were significantly different (no enrichment process 22.9 ± 39.3%, Bolton broth 38.1 ± 40.9%, Preston broth 10.7 ± 26.1%, one-way ANOVA, p < 0.01), while P. mirabilis was not different (Fig. 2a, b). Comparison of the proportions of each microbe according to the selective media used showed that the proportion of ESBL-producing E. coli (mCCDA 41.8 ± 40.8%, Preston agar 6.9 ± 22.2%, t test, p < 0.01) and P. mirabilis (mCCDA 36.0 ± 39.5%, Preston agar 86.5 ± 31.7%, t test, p < 0.01) was significantly different. Comparison of the proportion of each microbe according to the ratio of sample-to-enrichment broth indicated that the proportion of ESBL-producing E. coli was significantly higher in 10 3 -Bolton broth, while the proportion of P. mirabilis was significantly lower in 10 3 -Bolton broth.
Additionally, the proportion of ESBL-producing E. coli and P. mirabilis significantly differed according to the isolation of C. jejuni in a total of 490 samples by applying 14 methods to each of 35 samples (Fig. 3a, b). In 83 C. jejuni-positive plates and 407 C. jejuni-negative plates, based on culture-dependent tools, the proportions of ESBL-producing E. coli were 15.9 ± 32.8% and 25.9 ± 22.2% ( t test, p < 0.05), respectively, while the proportions of P. mirabilis were 40.5 ± 46.7% and 65.0 ± 42.1%, respectively (t test, p < 0.01).
Diversity of microbial communities in chicken feces of each enrichment process determined using metagenomics tools A total of 8,326,000 reads were obtained from 54 selected samples (all seven procedures for seven fecal samples and additional four fecal samples enriched in 10 3 -Preston broth and one additional fecal sample without enrichment) (Additional file 1: Table S1), of which 1,680,254 were valid after being filtered, de-replicated, and de-noised. As enrichment of the fecal samples progressed, the number of operational taxonomic units (OTUs) tended to decrease in the other processes ( Fig.  4a). Additionally, the number of OTUs of the fecal samples in the non-enriched process was significantly higher than those in the 10 3 -Bolton (Mann-Whitney U test, p < 0.05) and Preston broth (Mann-Whitney U test, p < 0.01) processes. There was no significant difference in the number of OTUs between each enrichment broth type (same ratio of sample-to-enrichment broth). The Shannon index showed no significant difference between each enrichment broth type (same ratio of sample-to-enrichment broth). However, the Shannon index tended to decrease as the ratio of sample-to-enrichment broth decreased ( Fig. 4b), particularly in 10 3 -Bolton and Preston broth (Mann-Whitney U test, p < 0.05 for each). Additionally, there was no difference between each type of enrichment broth in Faith's phylogenetic diversity, while this value tended to decrease as the ratio of sample-to-enrichment broth decreased (Fig. 4c). Particularly, Faith's phylogenetic diversity of fecal samples enriched in 10 3 -Preston broth was significantly decreased (Mann-Whitney U test, p < 0.05).
The microbial communities in fecal samples in each of the seven enrichment processes were compared using a principal coordinates analysis (PCoA) plot based on a weighted UniFrac distance metric and heat map. The chicken feces that had not undergone enrichment, and fecal samples enriched in the Bolton broth process (regardless of the sample-to-broth ratio), were not clustered together. In 10 1 -/10 2 -Preston broth, the enriched fecal samples were not clustered together. However, fecal samples enriched in 10 3 -Preston broth were strongly clustered together (Fig. 5a). In the heat map, as well as the PCoA plot, fecal samples enriched in 10 3 -Preston broth were clustered together, unlike those enriched in Bolton broth or other ratios of sample-to-Preston broth (Fig. 5b). In permutational multivariate analysis (PERMANOVA) and analysis of similarities (ANOSIM) to detect differences between groups, the 10 3 -Preston broth process was significantly different from the other groups (p < 0.001).
Composition of core microbial communities in feces of each enrichment process determined using metagenomics tools The core taxonomic composition at the genus level (with an average of ≥ 1% in at least one process) is shown in Fig. 6a. At the genus level, 61 genera were identified; among these, 18 genera comprised the core microbiota in all processes and accounted for more than 98% of the total microbiota. In feces without the enrichment process, the predominant core microbiota included Escherichia-Shigella (35.8%), followed by Lactobacillus (27.6%), Clostridium sensu stricto 1 (10.8%), and Peptostreptococcus (10.3%). In the 10 1 -Bolton broth procedure, Proteus (30.1%), followed Fig. 2 Proportion of competing microbes according to isolation procedures of C. jejuni based on culture-dependent tools. Proportion of a ESBL-producing E. coli and b P. mirabilis among competing microbes according to the isolation procedure of C. jejuni (mean ± SEM). Significance was determined by t test and one-way ANOVA. The proportion of ESBL-producing E. coli was significantly different according to the isolation procedure including the type of enrichment broth, selective media, and the combination of different enrichment broths, the ratio of sample-to-enrichment broth, and selective media. The proportion of P. mirabilis was significantly different according to the isolation procedure including the type of selective media and combination of different enrichment broths, the ratio of sample to enrichment broth, and selective agars. *p < 0.05, **p < 0.01 Fig. 3 Evaluating the effect of competing flora on the isolation of C. jejuni using culture-dependent tools. The proportion of a ESBL-producing E. coli and b P. mirabilis according to the isolation results of C. jejuni (mean ± SEM). Significance was determined by t test. The proportion of ESBL-producing E. coli and P. mirabilis was significantly higher in C. jejuni-negative fecal samples. *p < 0.05, **p < 0.01 by Escherichia-Shigella (17.2%) and Bilophila (11.9%), were the predominant core microbiota; in the 10 2 -Bolton broth procedure, Escherichia-Shigella (24.5%) followed by Proteus (22.9%) were the predominant core microbiota; in the 10 3 -Bolton broth process, Escherichia-Shigella (64.6%) was the predominant core microbiota. In the 10 1 -Preston broth process, Proteus (35.9%), followed by Peptostreptococcus (25.9%), and Escherichia-Shigella (11.1%) were the predominant core microbiota; in the 10 2 -Preston broth process, Proteus (38.9%) followed by Campylobacter (21.0%) were the predominant core microbiota; in the 10 3 -Preston broth process, Campylobacter (43.7%), followed by Proteus (38.7%), were the predominant core microbiota.
Relative abundance of Campylobacter and competing flora in feces of each enrichment process determined using metagenomics tools The relative abundance of Campylobacter in feces that had not been subjected to the enrichment process and fecal samples enriched in Bolton broth was low, while that in fecal samples enriched in Preston broth tended to increase as the ratio of sample-to-broth decreased (Fig. 6b). In particular, the relative abundance of Campylobacter in fecal samples in 10 3 -Preston broth (43.7%) and 10 2 -Preston broth (21.0%) were significantly higher than in other procedures (Mann-Whitney U test, p < 0.01). The relative abundance of Campylobacter in fecal samples in 10 3 -Preston broth was significantly higher than that in 10 2 -Preston broth (Mann-Whitney U test, p < 0.05). The relative abundance of Escherichia-Shigella in fecal samples in Preston broth was reduced compared to in feces (Fig. 6b). The abundance of these bacteria tended to decrease as the ratio of sample-to-broth decreased during Preston broth enrichment, particularly in 10 3 -Preston broth (Mann-Whitney U test, p < 0.01). However, a substantial increase was observed in fecal samples enriched in 10 3 -Bolton broth (Mann-Whitney U test, p < 0.05) compared to other procedures. In the case of Proteus, the relative abundance was only 0.8% in feces not subjected to the enrichment process, while it was significantly increased in enrichment broth at all ratios of sample-to-broth (Mann-Whitney U test, p < 0.01), except for in 10 3 -Bolton broth (Fig. 6b). The abundance of Proteus in fecal samples in 10 3 -Bolton broth was significantly lower than in other procedures (Mann-Whitney U test, p < 0.01), except for feces that had not been subjected to the enrichment process. The results of linear discriminant analysis effect size (LEfSe) performed to compare 10 3 -Preston broth with other processes (no enrichment and 10 3 -Bolton broth) were also consistent (Additional file 3: Figure S1).
In correlation analysis, Campylobacter was negatively correlated with Proteus (Pearson correlation coefficient (r) − 0.39), and Escherichia-Shigella was also negatively correlated with Proteus (r − 0.82) in Bolton broth, regardless of the ratio of sample-to-broth (Additional file 4: Figure S2A). In Preston broth, regardless of the ratio of sample-to-broth, Campylobacter was negatively correlated with nearly all microbes including Escherichia-Shigella (r − 0.54) (Additional file 4: Figure S2B).
Relationship between competing flora and C. jejuni determined using metagenomics tools and culturedependent tools A total of 54 samples analyzed by metagenomics tools were divided into two groups according to the culture results; 17 samples were positive for C. jejuni, and 37 were negative. The relative abundance of Campylobacter was significantly higher in fecal samples from which C. jejuni was detected based on the culture result (t test, p < 0.01, Fig. 7a). However, the relative abundance of Escherichia was significantly lower in fecal samples from which C. jejuni was detected based on the culture result (t test, p < 0.05, Fig. 7b). For Proteus, there was no difference in the relative abundance in the fecal sample according to isolation result of C. jejuni based on the culture result (Fig. 7c). Validation of metagenomics data using quantitative PCR Quantitative PCR (qPCR) of C. jejuni was conducted on all 54 samples to validate the metagenomics data and estimate the levels of C. jejuni in each procedure. First, qPCR was performed using standard strains (NCTC 11168 and ATCC 33560) to confirm the correlation between colony-forming units (CFU) and cycle threshold (Ct) values of C. jejuni (y = − 3.3631x + 44.547, R 2 = 0.9118, p < 0.001) (Additional file 5: Figure   S3). Next, qPCR was conducted on the DNA used in the current study. The log CFU/ml of C. jejuni in each sample was inferred through the standard curve: 0.2 ± 0.2 (no enrichment, mean ± SEM), 0.7 ± 0.7 (10 1 -Bolton broth), 0.9 ± 0.9 (10 2 -Bolton broth), 3.9 ± 1.5 (10 3 -Bolton broth), 2.8 ± 1.5 (10 1 -Preston broth), 7.3 ± 1.3 (10 2 -Preston broth), and 10.2 ± 0.2 (10 3 -Preston broth) (Fig. 8a). The amount of C. jejuni was significantly higher in fecal samples in 10 3 -Preston broth than in the Fig. 6 Microbial community of fecal samples in each procedure at the genus level. a Taxonomic composition and b relative abundance (min to max) of Campylobacter, Escherichia, and Proteus of fecal samples in each procedure. The relative abundance of Campylobacter in fecal samples in 10 3 -Preston broth was significantly higher than in other procedures. In the case of Escherichia-Shigella, the relative abundance was reduced, compared to feces, but was still present in large proportions in the process of 10 1 -/10 2 -Bolton broth enrichment. In the case of Proteus, the relative abundance increased during the enrichment process in both enrichment broths, at all ratios of sample-to-broth, with the exception of 10 3 -Bolton broth other procedures (Mann-Whitney U test, p < 0.05). Additionally, the amount of C. jejuni was significantly higher in fecal samples in 10 2 -Preston broth than in the other procedures (Mann-Whitney U test, p < 0.05), except for the 10 3 -Preston/Bolton broth. Furthermore, the correlation between the Ct values from the qPCR results and number of Campylobacter reads from metagenomics results was examined (y = − 0.0009x + 25.416, R 2 = 0.664, p < 0.001, Fig. 8b).

Discussion
This is the first study to apply culture-independent sequence-based metagenomics to evaluate C. jejuni isolation methods in chicken feces. Isolation of Campylobacter spp. is difficult because of the specific growth conditions of the bacteria, presence of VBNC bacteria, and masking by competing flora on selective media. Therefore, culture-independent diagnostic tests, which show high sensitivity and rapid diagnosis, were recently applied to detect Campylobacter-infections, particularly by the Centers for Disease Control Foodborne Diseases Active Surveillance Network [21]. However, applying culture-independent diagnostic tests alone without obtaining isolates from culture-dependent methods makes further analyses, such as epidemiological investigation and characterization of strains, impossible. Given Fig. 7 The relative abundance of each microbe according to the isolation results of C. jejuni. The relative abundance of a Campylobacter, b Escherichia, and c Proteus according to the isolation results of C. jejuni based on culture-dependent tools (mean ± SEM). A total of 54 samples analyzed by metagenomics tools were divided into two groups according to the culture results; 17 samples were positive for C. jejuni, and 37 were negative. Significance was determined by t test. The relative abundance of Campylobacter was significantly higher in the fecal sample from which C. jejuni was isolated, while the relative abundance of Escherichia was significantly lower. There was no difference in the relative abundance of Proteus in the fecal sample according to isolation result of C. jejuni. *p < 0.05, **p < 0.01 Fig. 8 Quantitative PCR of C. jejuni applied to all 54 samples. a Log CFU/ml of C. jejuni in each procedure (mean ± SEM) inferred from the standard curve and b correlation of metagenomics data and quantitative polymerase chain reaction (qPCR) results. Significance was determined by Mann-Whitney U test. The amount of C. jejuni was significantly higher in fecal samples in 10 3 -Preston broth, 1.0 × 10 10.2 ± 0.2 , followed by 10 2 -Preston broth (1.0 × 10 7.3 ± 1.3 ) and 10 3 -Bolton broth (1.0 × 10 3.9 ± 1.5 ). In addition, the cycle threshold (Ct) values from qPCR results and the number of Campylobacter reads from metagenomics results had a high correlation. *p < 0.05, **p < 0.01 this, effective methods for isolating C. jejuni are needed. Numerous studies have been conducted to evaluate and improve the methods of isolating C. jejuni from various food sources using culture-dependent tools [7,8,22]. Although food sources are the main cause of human campylobacteriosis, isolation of C. jejuni at the farm level is primarily necessary for managing campylobacteriosis. However, most studies have been carried out using chicken meat with low microbial contamination rather than chicken feces, which is a high microbial contamination source [23,24]. Therefore, to isolate Campylobacter from feces, a specific method, which differs from that used for food, should be applied. Here, the methods for isolating C. jejuni from chicken feces were evaluated using metagenomics tools and culture-dependent tools.
Based on metagenomics analysis, the current study indicated that feces undergoing an enrichment process in Preston broth at a 1:10 3 ratio of feces-to-broth were the most effective for isolating C. jejuni. In metagenomics applications to evaluate the isolation method of Campylobacter, microbial diversity and taxonomic composition were calculated. First, the effect of the enrichment process during the isolation of C. jejuni was assessed based on microbial diversity. Based on the alpha diversity, including the number of OTUs, Shannon diversity index which measures species abundance and evenness, and Faith's phylogenetic diversity index which measures biodiversity, the diversity of the microbial community in the feces was decreased during enrichment compared to fecal samples that were not enriched [25,26]. The microbial diversity decreased to a greater extent as the ratio of feces-to-broth decreased, particularly when the level of enrichment broth was high compared to the fecal samples. However, at the same ratio of fecal samples-to-each enrichment broth, there was no significant difference in microbial diversity. This suggests that the ratio of feces-to-broth affects microbial diversity, regardless of the type of enrichment broth. This was because during the enrichment process for C. jejuni, only the target microbe selectively grew, and other microbes were inhibited by antibiotics and specific enrichment conditions (microaerophilic conditions and a high incubation temperature of 42°C). Additionally, the difference in microbial diversity according to the ratio of fecal samples-tobroth may be related to differences in the relative amounts of antibiotics-to-fecal samples. A greater relative amount of antibiotics-to-fecal samples led to greater inhibitory effects of antibiotics on the fecal sample.
Second, the effect of the enrichment process was evaluated based on taxonomic composition. In non-enriched feces, Escherichia-Shigella and Lactobacillus were the most abundant genera. This is similar to results of previous studies showing that Lactobacillus (14.9%) and Escherichia-Shigella (9.0%) are the most abundant bacteria in feces from healthy chickens [27]. However, the original taxonomic composition in the feces changed dramatically during the enrichment process. The relative abundance of Campylobacter (identified as C. jejuni at the species level using metagenomics and culture-based tools) was significantly higher in Preston broth, especially in 10 2 -and 10 3 -Preston broth, than in Bolton broth. In Preston broth, the relative abundance of Campylobacter increased as the ratio of feces-to-broth decreased from 1:10 1 to 1:10 3 . Furthermore, qPCR was performed to validate the metagenomics data, which varied depending on the type of sequencer used, type of database, and DNA extraction method [14,28,29]. The number of reads of C. jejuni from metagenomics data and qPCR results of C. jejuni was highly correlated, suggesting that the metagenomics data in this study was reliable, and the amount of C. jejuni in each enrichment process was deduced. The amount of C. jejuni was high in fecal samples in 10 3 -Preston broth at 1.0 × 10 10.2 ± 0.2 . This was similar to that for C. jejuni in brain heart infusion broth at 48 h (1.0 × 10 8 -1.0 × 10 9 ) [30]. The negative qPCR results for C. jejuni may be due to DNA contamination from other bacteria, or that transcription was below the detection limit.
Differences in the levels of C. jejuni, according to enrichment procedures, were observed because of competing flora, particularly ESBL-producing E. coli, which were influenced by the type of antibiotics in each enrichment broth and by the relative amount of antibiotics-tofeces in each ratio of broth. In general, the amount of background flora affects the isolation result of C. jejuni, often producing false-negative results [31]. Comparison of the two enrichment broth types revealed that polymyxin B effectively controlled ESBL-producing E. coli; this antibiotic is present in Preston broth but not in Bolton broth [9]. Recent studies also showed that ESBLproducing E. coli is poorly resistant to the antibiotics cefoperazone, vancomycin, and trimethoprim, which are present in Bolton broth [8]. Additionally, the growth of C. jejuni in co-cultures of C. jejuni and ESBL-producing E.coli was significantly lower than in the single culture of C. jejuni, indicating that ESBL-producing E. coli acts as a competitive inhibitor leading to a reduction of C. jejuni levels [32]. In our study, the relative abundance of Escherichia (identified as ESBL-producing E. coli using culture-based tools), which was highest in non-enriched feces, showed a more substantial decrease in Preston broth than in Bolton broth at the same ratio of fecal samples-to-enrichment broth and tended to decrease as the ratio of feces-to-Preston broth decreased. Campylobacter was negatively correlated with Escherichia when fecal samples were enriched in Preston broth. Therefore, in the enrichment process of Campylobacter, Preston broth more effectively inhibited ESBL-producing E. coli compared to Bolton broth. Furthermore, a comparison of the ratio of fecal samples-to-enrichment broth of 1:10 1 to 1:10 3 in Preston broth revealed that the differences depended on the amount of antibiotics used. The ratio of sample-to-enrichment broth of 1:10 1 , which is the most commonly used ratio for isolating Campylobacter from food sources, did not effectively enrich target microbes in the fecal samples. This means that antibiotics in the enrichment broth may not work effectively in the enrichment process unless the ratio of feces-to-enrichment broth is optimized. Therefore, the relative amount of antibiotics should be high in order to inhibit competing microbes when isolating C. jejuni from feces with higher microbial contamination than in food. Moreover, fecal samples enriched in 10 3 -Preston broth were strongly clustered together in the PCoA plot and heat map, while fecal samples in the other procedures were not clustered with each other. These results suggest that 10 3 -Preston broth has the most constant effect on fecal samples during the enrichment process, including decreasing microbial diversity and changing the microbial composition.
P. mirabilis was a major competing microbe observed during the process of isolating Campylobacter in previous studies [33]. The current study also showed that the relative abundance of Proteus (identified as P. mirabilis at the species level using metagenomics tools and culture-based tools) in the enrichment process increased, compared to fecal samples that were not enriched. This may be because of their high ability to differentiate and survive [34,35]. However, the relative abundance of Proteus did not affect the relative abundance of Campylobacter, unlike ESBL-producing E. coli. Instead, the relative abundance of Proteus was high in Preston broth at 1:10 2 and 1:10 3 ratios, but low in Bolton broth at a 1:10 3 ratio. This was most likely due to resistance to polymyxin B in Preston broth, but susceptibility to vancomycin in Bolton broth [36]. We also predict that competition occurred between Proteus and ESBL-producing E. coli based on the negative correlation between these two microbes in Bolton broth, although further studies are needed to confirm this.
Analysis of C. jejuni isolation procedures using metagenomics was confirmed using culture-dependent tools. Campylobacter, Escherichia-Shigella, and Proteus sequence reads based on metagenomics tools matched those of isolates obtained by culture-dependent tools. Campylobacter and Proteus were identified as C. jejuni and P. mirabilis at the species level, respectively, and only C. jejuni and P. mirabilis, without other species, were isolated based on culture-dependent tools. For Escherichia-Shigella at the genus level, although species-level identification was not possible, only E. coli was isolated, all of which were identified as ESBL-producing E. coli using culture-dependent methods. The microbes that, using metagenomics, were identified to be present at a relatively higher abundance in each enrichment process were also frequently isolated using culture-dependent methods, regardless of the type of selective media used. The isolation rates of C. jejuni, ESBL-producing E. coli, and P. mirabilis were significantly higher in enriched fecal samples, with a higher relative abundance of all three. Additionally, the inhibitory effect of competing flora in the isolation of C. jejuni, which was determined using metagenomics tools, was confirmed using culture-dependent tools; the proportion of ESBL-producing E. coli and P. mirabilis was significantly higher in C. jejuni-negative fecal samples than in C. jejuni-positive fecal samples. The consistency between the culture-based results and metagenomics results suggests that our evaluation of the isolation method using metagenomics tools was reliable.
The effect of selective media was evaluated based on the use of culture-dependent tools. Using metagenomics alone, only the effects of the enrichment process on the isolation of Campylobacter can be evaluated; however, using a combination of metagenomics and culturedependent tools enabled evaluation of the enrichment broth and selective media in isolating Campylobacter. Generally, the enrichment step had a greater effect on the isolation of microbes than did the selective media [13]. However, we found that selective media significantly affected target microbe isolation as well. Even if the fecal sample was subjected to the same enrichment process (same type of enrichment broth and ratio of feces-to-enrichment broth), the isolation rate and proportion of microbes differed depending on the selective media used. ESBL-producing E. coli was isolated from mCCDA significantly more frequently than from Preston agar, while P. mirabilis isolation was greater from Preston agar than from mCCDA. This was most likely due to the antibiotics present in the selective media. The antibiotics in mCCDA were similar those in Bolton broth, while the antibiotics in Preston agar were consistent with those in Preston broth. Because of this, Preston agar and broth showed similar inhibitory effects on ESBLs, while mCCDA inhibited Proteus as in Bolton broth. Particularly, Preston agar, which does not inhibit P. mirabilis, made it difficult to isolate C. jejuni because of the swarming differentiation characteristics of P. mirabilis, which colonized the entire plate surface causing a masking phenomenon [37]. Additionally, in the enrichment process in 10 2 -and 10 3 -Preston broth, while the same type of enrichment broth was used, the isolation rate of C. jejuni was significantly higher in Preston agar than in mCCDA. This may be due to a cross-selective inhibition effect of antibiotics in enrichment broth and selective media. The relative abundance of ESBL-producing E. coli decreased during the enrichment process in Preston broth, and P. mirabilis was inhibited in mCCDA, facilitating the isolation of C. jejuni.

Conclusions
This is the first study to apply culture-independent sequence-based metagenomics and culture-dependent tools to evaluate C. jejuni isolation methods from feces, a high-level microbial matrix. The application of metagenomics revealed changes in the microbial community during the microbial isolation procedure, which cannot be detected using previous culture-based analyses. Particularly, changes in the target microbe and competing flora during the enrichment process, and their correlation, were observed. Our results suggest that (1) Preston broth is more effective for isolating C. jejuni from chicken feces compared to Bolton broth, which is widely used to isolate this bacteria from chicken meat; (2) the effectiveness of the enrichment process depends on the ratio of samples to enrichment broth, even in enrichment broth containing effective antibiotics; (3) when isolating microorganisms, although the type and ratio of enrichment broth should be considered, the type of selective agars is also essential. Taken together, we optimized the method for isolating C. jejuni from chicken feces, enriched in Preston broth at a ratio of 1:10 3 (sample-to-broth) followed by spreading onto mCCDA. This procedure effectively enriched C. jejuni and controlled competing flora, particularly ESBL-producing E. coli and P. mirabilis. This isolation method aids in obtaining Campylobacter isolates effectively, enabling further analyses, which include epidemiological investigation and molecular characterization of strains. This study provides a new perspective and possibilities for applying metagenomics in microbiological research, such as improving diagnostic methods for fastidious bacteria that are difficult to isolate.

Sampling of chicken feces and applying different isolation methods of C. jejuni
In December 2017, 35 fecal samples from the gastrointestinal tract of chickens at a slaughterhouse were collected. All fecal samples were analyzed to compare isolation methods of C. jejuni ( Fig. 1 and Additional file 1: Table  S1). A total of 14 procedures were applied for all samples, which were combinations of different enrichment broths (without enrichment process or enriched in Bolton broth or Preston broth), ratio of sample-to-enrichment broth (1:10 1 , 1:10 2 , or 1:10 3 ), and selective media (mCCDA or Preston agar). First, one loop (10 μL) of all 35 fecal samples was directly spread onto two different types of selective media, mCCDA (Oxoid Ltd., Hampshire, UK) with a CCDA selective supplement (Oxoid Ltd.) and Preston agar (Oxoid Ltd.) with a Preston Campylobacter selective supplement (Oxoid Ltd.) and lysed horse blood (Oxoid Ltd.) without enrichment. The selective media were microaerobically incubated at 42°C for 48 h. Second, all 35 fecal samples were enriched in Bolton broth (Oxoid Ltd.) with a Bolton broth selective supplement (Oxoid Ltd.) at different ratios of sample-to-enrichment broth (1:10, 1:10 2 , and 1:10 3 , respectively). Next, the enrichment broths were micro-aerobically incubated at 42°C for 48 h, followed by plating onto mCCDA and Preston agar at 42°C for 48 h under microaerobic conditions. Third, all 35 fecal samples were enriched in Preston broth (Oxoid Ltd.) with a Preston broth selective supplement (Oxoid Ltd, Hampshire, UK) at different ratios of sample-to-enrichment broth (1:10, 1:10 2 , and 1:10 3 , respectively). The enrichment broths were microaerobically incubated at 42°C for 48 h, followed by plating onto mCCDA and Preston agar at 42°C for 48 h microaerobically.

Identification of bacterial colonies obtained from culturedependent tools
In 14 different procedures, at least four colonies suspected as Campylobacter spp. were transferred to Müller-Hinton agar (MHA, Oxoid Ltd.) and microaerobically incubated at 42°C for 48 h. Next, C. jejuni was confirmed by PCR for each colony (Additional file 2: Table S2) [38]. For colonies that did not show typical morphology of Campylobacter spp., an average of 2.47 (range 1-9) colonies with different shapes was transferred to MHA, according to the proportion of colonies constituting the selective agar. The agars were incubated at 42°C for 48 h under microaerobic conditions. Colonies suspected to be Escherichia coli and Enterococcus faecium/faecalis were confirmed by PCR (Additional file 2: Table 2) [39,40]. For colonies identified as E. coli, an antibiotic resistance test was performed according to the Clinical and Laboratory Standards Institute guidelines to identify ESBL-producing E. coli. Additionally, colonies not confirmed by PCR (C. jejuni, E. coli, E. faecium, E. faecalis) were identified by 16 s rRNA sequencing using the universal 16S rRNA primers 518F and 805R [41].

DNA extraction and Illumina MiSeq sequencing
Microbial community analysis of feces and enriched fecal samples was performed to investigate the effects of seven enrichment procedures using different types of enrichment broth and ratios of sample-to-enrichment broth. Six procedures involved enrichment processes (Bolton and Preston broth at 1:10 1 , 1:10 2 , and 1:10 3 , respectively), and one procedure did not. At least seven samples per each procedure were selected according to the selection criteria (isolation result of C. jejuni based on the culture-dependent tool), and a total of 54 samples (8 feces and 46 enriched fecal samples) were examined in microbial community analysis (Additional file 1: Table  S1). Metagenomic DNA was extracted using a FastDNA SPIN extraction kit (MP Biomedicals, Santa Ana, CA, USA) according to the manufacturer's instructions. DNA amounts and quality were quantified using PicoGreen (Thermo Scientific, Waltham, MA, USA) and a Nanodrop spectrometer (Thermo Scientific). Bacterial DNA amplification was carried out, targeting the V3-V4 regions of the 16S rRNA gene (Additional file 2: Table S2) [42]. Next, to add multiplexing indices and the sequencing adapter, a subsequent amplification was performed. The amplified products were pooled and normalized using PicoGreen. Sequencing was conducted at Macrogen, Inc. (Seoul, Korea), using an Illumina MiSeq (Illumina, San Diego, CA, USA).

Processing of sequencing analysis
Bioinformatic analysis was performed using QIIME2 software (version 2018.2) pipeline [43]. Raw sequence reads were filtered, de-replicated, and de-noised using DADA2's recommended parameters [44]. Phylogenetic diversity analysis was performed using MAFFT, and a phylogenetic tree was created using the midpoint rooting method with the FastTree plugin [45,46]. Alpha diversity was measured in QIIME2 with a subsampling depth of 12,001 sequences, and beta diversity was measured in QIIME2 based on the weighted UniFrac distance [47,48]. Taxonomy assignments were conducted using the SILVA database version 119 [29]. Pearson's correlation coefficient was calculated in MicrobiomeAnalyst, and a correlation matrix plot was generated using R version 3.5.0 [49]. To identify the bacterial taxa that are differentially abundant in each group's microbial community, LEfSe analysis was performed using the online Galaxy interface (http://huttenhower.sph.harvard.edu/galaxy/), and the logarithmic linear discriminant analysis score cutoff was set to 2.0.
Quantitative PCR of C. jejuni Quantitative PCR of C. jejuni was performed using DNA extracted from ATTC 33560 and NCTC 11168 (DNA concentration was 0.0000237-237 ng/μL) and 54 DNA samples subjected to microbial community analysis. hipO was selected as a target gene for distinguishing C. jejuni from other microbes. Primers and probes for qPCR were used with slight modifications from previous studies (Additional file 1: Table S1) [50]. All reactions were performed in a total volume of 25 μL containing 5 μL of DNA, 2.5 μL of primers (0.8 μM of each primer), 0.5 μL of probe (0.2 μM), 12.5 μL of 2x TaqMan Universal Master Mix II (Life Technologies, Carlsbad, CA, USA), and 4.5 μL of distilled water. The amplification conditions were 95°C for 10 min, followed by 45 cycles at 95°C for 15 s and 58°C for 1 min. All reactions were run in triplicate using an ABI PRISM 7500 fast real-time PCR system (Life Technologies).