Campylobacter Colonization of the Turkey Intestine in the Context of Microbial Community Development

Abstract

Patterns of microbial community dynamics in the turkey intestine were examined. Every week of the 18-week production cycle, cecal bacterial communities and Campylobacter loads were examined from five birds for each of two flocks. Molecular fingerprinting via the automated ribosomal intergenic spacer analysis (ARISA) and terminal restriction fragment length polymorphism (T-RFLP) of the cecal samples revealed that microbial communities changed in a time-dependent manner, and during both trials they developed via transition through three phases during the production cycle. A core component of the microbiota consisting of 11 Bacteroidetes types was present throughout both trials. In contrast, constant succession was detected in the Clostridiales populations until week 10 or 11, with few shared sequences between the flocks. Changes in Campylobacter jejuni and Campylobacter coli loads were correlated to, but not dependent on, the two acute transitions delimiting the three developmental phases.

Campylobacter is the prevailing cause of bacterial enteritis in developed countries, resulting in roughly two million cases in the United States annually. The consumption of poultry is a significant risk factor, and Campylobacter has been found to contaminate 81% of fresh, whole broiler chicken carcasses and a third of turkey carcasses (3, 14, 25, 29). A commensal in poultry, Campylobacter is present in the crop at 10⁴ and in the ceca at 10⁷ CFU/g contents; thus, a single rupture early in processing can contaminate carcasses from several flocks (31). The estimated Campylobacter infectious dose for humans is 500 cells, and in the United States, broiler chickens routinely are contaminated with 10³ CFU per carcass (4, 43). Other risk factors for campylobacteriosis include the consumption of fresh vegetables and bottled water (11). Campylobacter has been found in poultry manure used to fertilize crops as well as runoff from these farms (18, 20, 45). Thus, it is evident that there is a need for preharvest strategies to reduce Campylobacter levels in poultry.

To effectively develop Campylobacter intervention strategies, it is necessary to understand the ecology of this pathogen in its animal host. Very little is known regarding the ecology of this pathogen in vivo, but it is evidently highly complex. It was demonstrated recently that bacteriophage populations alter the intestinal Campylobacter contingent (37). It is known that Campylobacter spp. can use amoeba and protozoa as Trojan horses to colonize the poultry intestine, but the role of these relationships in vivo is unknown (40, 41). Skanseng et al. examined Campylobacter colonization in chickens with two microbiota backgrounds (39). Their results indicated a shift in dominant Campylobacter jejuni strains at 4 weeks after hatching irrespective of the extant microbiota, suggesting a host-mediated effect on Campylobacter colonization. Finally, a previous report identified a mid-grow-out microbiota community shift in the cecal communities of turkeys and a possible correlation with Campylobacter colonization (38).

Previous poultry studies of the intestinal bacteria examined microbiota community development during the 7-week broiler chicken production cycle (19, 26, 44, 47). Results from these studies indicate the succession from hatch through grow-out, starting with communities consisting predominantly of enterococci, coliforms, and clostridia (8). The communities culminate in clostridia, sporomusa, enterics, and bacteroides (47). Throughout community development, lactobacilli and proteobacteria decrease in numbers, while bifidobacteria numbers increase (2, 26). In addition, a period of community stabilization exists during rapid skeletal growth, followed by further changes during grow-out. In addition to age, factors such as host diet, stress, and antibiotic use cause microbial populations to change (9, 27, 46).

Research in the current manuscript expands a previous examination of Campylobacter colonization in relation to intestinal bacterial population dynamics (38). A series of molecular methods were used to describe cecal bacterial dynamics. Automated ribosomal intergenic spacer analysis (ARISA) was used to generate an overview of community dynamics in two commercial turkey flocks through the entire 18-week production cycle. Terminal restriction fragment length polymorphism (T-RFLP) then was used to identify species involved in the community shifts. A Bacteroidetes core community was maintained throughout, while a complex succession of Clostridiales developed during the first 11 weeks. Campylobacter communities also were in flux, with changes in the prevalence of different species linked to, but not dependent upon, microbiota population dynamics.

MATERIALS AND METHODS

Animals.

Two trials were performed in which five hybrid converter turkeys per week were killed throughout the 18-week commercial growing period. Turkeys from week 6 of trial 2 were not obtainable, thus a total of 175 animals were sampled. Trials were sampled semiconcurrently, from July to December 2005. Birds were raised in flocks of 5,000 animals, in standard two-phase, all-in-all-out facilities in which the animals were moved from brooders to screened grow-out houses at day 28. The brooder and grow-out facilities for the trials were roughly 2.5 miles apart. Birds received vaccines for Newcastle virus on day 0, hemorrhagic enteritis live vaccine on day 30, and pneumovirus on day 7, with boosting at week 5 or 6 (Table 1). Further medications included sulfadimethoxine treatment for Bordetella spp., penicillin G-potassium and 3-nitro roxarsone for enteritis, and amprolium for coccidiosis. Citric acid was diluted 1:4 in the drinking water for intestinal acidification. Feed from a local mill was supplemented with 10 to 20 g/ton of the growth promotant virginiamycin.

TABLE 1.

Therapeutic treatments for two flocks

Time point	Trial 1	Trial 2
Day 0	Newcastle vaccine	Newcastle vaccine
Wk 1	Penicillin G, pneumovirus vaccine
Wk 2	Penicillin G	Pneumovirus vaccine
Wk 3	3-Nitro roxarsone, amprolium	Newcastle boost
Wk 4	Hemorrhagic enteritis live vaccine	Penicillin G
Wk 5	Sulfadimethoxine	Amprolium, hemorrhagic enteritis live vaccine, pneumovirus boost
Wk 6	3-Nitro roxarsone, pneumovirus boost
Wk 7	Sulfadimethoxine

DNA extraction.

DNA was extracted from 0.2 g of intestinal contents according to the Qbiogene Fast Prep method using lysing matrix A (QBiogene, Carlsbad, CA). Microbial cells were lysed by shaking at 5 m/s in a Fast Prep FP120 for 30 s. DNA from the lysates was purified using phenol-chloroform extraction and ethanol precipitation.

DNA from Campylobacter cultures was isolated by boiling cells in 50 μl H₂O for 5 min and centrifuged to pellet cell debris.

Campylobacter quantification.

Campylobacter jejuni and C. coli loads were quantified in DNAs pooled for five birds at each time point (22). C. jejuni hipO genes and C. coli glyA genes were quantified using TaqMan chemistry, forward primers Cj-F1 (5′-TGCTAGTGAGGTTGCAAAAGAATT-3′) and Cc-F1 (5′-CATATTGTAAAACCAAAGCTTATCG-3′), reverse primers Cj-R1 (5′-TCATTTCGCAAAAAAATCCAAA-3′) and CcR1 (5′-AGTCCAGCAATGTGTGCAATG-3′), and probes CjFAM (5′-6-carboxyfluorescein [FAM]-ACGATGATTAAATTCACAATTTTTTTCGCCAAA-black hole quencher 1a-3′) and CcFAM (5′-6-FAM-TAAGCTCCAACTTCATCCGCAATCTCTCTAAATTT-black hole quencher 1a-3′) (22). Amplifications for both species were performed using an iCycler IQ5 according to the published protocol, except reactions were not multiplexed (Bio-Rad Laboratories, Hercules, CA) (22). Campylobacter quantifications were normalized against universal 16S quantities generated via SYBR PCR amplification. SYBR reaction mixtures were composed of 1× iQ SYBR green Supermix (Bio-Rad Laboratories, Hercules, CA) and 400 nM each primer 342R (5′-CTGCTGCSYCCCGTAC-3′) and 27F (5′-AGRRTTTGATYBTGGYTCAG-3′) (23). Both the hipO and glyA indicator genes are believed to exist in single copies on the pathogen genome, thus normalization against total 16S copy number allowed the results to be reported as Campylobacter sp. genomes per 100,000 16S genes (22).

ARISA.

ARISA was performed on 175 individual samples, 147 of which yielded usable fingerprints. Bacterial ARISA was performed as described previously using the primer set ITSF (5′-GTCGTAACAAGGTAGCCGTA-3′) and ITSReub (5′-GCCAAGGCATCCACC-3′) (5, 7, 12). Primer ITSReub was 5′-end labeled with the FAM fluorochrome (Operon, Valencia, CA). Each 20-μl PCR mixture contained 4 ng template DNA. After mixing with MapMarker 1000 size standard (BioVentures, Murfreesboro, TN), PCR products were separated on an ABI3100 capillary sequencer for 167 min at 12.2 kV, using POP6 polymer and a 50-cm array (Applied Biosystems, Foster City, CA). Amplicon sizes were identified using the Local Southern size-calling method with the GeneMapper v3.5 software. Peaks in the 50- to 500-bp range were normalized such that any peak representing ≥10% of the average MapMarker 1000 size standard peak area was considered different from the background value. Similarity matrices were generated using the Dice coefficient. Principle coordinate analyses (PCoA) were performed using the Paleontological Statistics Software package (PAST) to visualize community changes over time (17). Dendrograms illustrating the clustering of similar communities over time were constructed using the unweighted-pair group method using average linkages (UPGMA) algorithm via the Mega 3.0 software (21, 42). Finally, nonparametric multivariate analysis of variance (NP-MANOVA) calculations were performed using the PAST software to examine the statistical likelihood of the existence of successional phases that were detected by the PCoA and UPGMA cluster analyses. A Mantel test was used to investigate the similarity of community dynamics between the two trials (28).

T-RFLP.

Bacterial community analysis was performed by T-RFLP (http://rdp8.cme.msu.edu/html/t-rflp_jul02.html). Pooled samples for all five birds per time point per trial were examined. Thermocycling reaction mixes contained 50 mM Tris, pH 8.3, 500 μg bovine serum albumin per ml, 2.5 mM MgCl₂, 250 μM each deoxynucleoside triphosphate, 400 nM primers, 8.75 U Taq, and 40 ng target DNA per 100-μl reaction. Reaction parameters included a 5-min initial denaturation at 95°C. Cycling consisted of 45 s of 94°C denaturation, 1 min of 48°C annealing, and 2 min of 72°C elongation. Reactions were finished with a 7-min elongation at 72°C. Universal HEX-labeled forward primer 8-27f (5′-AGAGTTTGATCMTGGCTCAG-3′) and reverse primer 1392-1407r (5′-ACGGGCGGTGTGTACA-3′) were used to amplify 16S genes from the fecal DNA (1, 15). Amplicons were purified using QIAquick PCR purification (Qiagen, Valencia, CA) and then restricted using enzyme HhaI, RsaI, or MspI. Digests were cleaned using QIAquick PCR purification. Fragments were separated and fluorescently detected on an ABI3100 capillary sequencer for 167 min at 12.2 kV, using POP6 polymer and a 50-cm array (Applied Biosystems, Foster City, CA). Amplicon sizes were identified using the MapMarker 1000 internal size standard (BioVentures, Murfreesboro, TN) and the Local Southern size-calling method with the GeneMapper v3.5 software (Applied Biosystems, Foster City, CA). Profiles were normalized and analyzed using FRAGSORT (30), and a restriction size database was generated using previously published poultry intestinal microbiota 16S sequences (6, 16, 38, 47). Only restriction fragments (RF) representing ≥2% of the total peak area and identified by at least two of the three restriction profiles were included in the analysis. PCoA and Mantel statistical analyses were performed as described above.

RESULTS

ARISA.

The succession of the composition of the intestinal microbiota over time is presented for two different flocks of commercially raised turkeys (Fig. 1 and 2; also see Fig. S1 in the supplemental material). Pooled data for five birds tested per week per flock are presented in principle coordinate plots (Fig. 1). UPGMA dendrograms indicate the similarity between the fingerprints of individual birds over time (Fig. 2). Taken together, the principle coordinate analysis and the UPGMA clustering method suggest a progression through three phases in both trials: an initial phase encompassing weeks 1 to 3, a median phase of weeks 4 to 10, and a late phase including weeks 11 to 18. One-way NP-MANOVA confirmed the presence of the three phases (Table 2). Bonferroni-corrected and -uncorrected P values are reported in Table 2. Progression into the median phase at week 4 and the late phase at week 11 was statistically repeatable in both trials (Table 3).

FIG. 1.

PCoA of ARISA profiles. Profiles for five turkeys were pooled for each week of the 18-week trials, and data points are labeled by week. (A) Trial 1; (B) trial 2. For both trials, the first and second principle coordinates indicate the transitions between the initial phase (weeks 1 to 3) and median phase (weeks 4 to 10) and from the median phase into the late phase (weeks 11 to 18). The arched shape of the data points is indicative of community succession through time.

FIG. 2.

Dendrogram of ARISA microbial fingerprints for individual birds through 18 weeks posthatch. (A) Trial 1; (B) trial 2. Numbers indicate the weeks of the trial, while letters indicate individuals sampled at each week. Three major developmental phases (initial, median, and late) are highlighted. Fingerprint pattern differences were determined using the Dice coefficient, and the dendrogram representation of those differences was produced using UPGMA clustering in Mega2. The scale bar represents percent homogeneity.

TABLE 2.

One-way NP-MANOVA pairwise comparisons of putative microbiota community phases^a

Data and time point (wk)	Result for wk:
	1-3	4-10	11-18
ARISA fingerprint data
Trial 1
1-3		0.0078	0.0056
4-10	0.0234		<0.0001
11-18	0.0168	<0.0001
Trial 2
1-3		0.0126	0.0061
4-10	0.0378		0.002
11-18	0.0183	0.006
CFTb
1-3		0.0082	0.0066
4-10	0.0246		0.0002
11-18	0.0198	0.0006
T-RFLP fingerprint data
Trial 1
1-3		0.2425	0.0202
4-10	0.7275		<0.0001
11-18	0.0606	<0.0001
Trial 2
1-3		0.6901	0.0438
4-10	1		0.0016
11-18	0.1314	0.0048

^aThe presence of three putative phases was supported for trial 1 by an F statistic of 2.342 and for trial 2 by an F statistic of 1.654. Bonferroni-corrected P values are reported above the diagonal, and uncorrected values are reported below the diagonal. Statistically significant results are reported in boldface.

^bCFT indicates ARISA data for six cohabiting turkeys sampled during 18 weeks in an antibiotic-free environment (38).

TABLE 3.

Mantel correlations of the similarities between microbial community dynamics in trials 1 and 2^a

Data set	Correlation R	P
C. jejuni	0.158	0.0770
C. coli	−0.0997	0.6184
ARISA	0.5971	<0.0001
T-RFLP	0.399	0.0310
Therapeutics	−0.1217	0.4094

^aStatistically significant correlations are reported in boldface.

T-RFLP.

Fingerprinting via T-RFLP was performed in an effort to accurately describe the cecal microbial community composition. Taxonomic assignments were developed using previously published, poultry-derived 16S sequences available through GenBank. A FRAGSORT analysis of the profiles indicated that a subset of RF types correlating to sequences identified as Bacteroidales was present throughout both trials (see Tables S1 and S2 in the supplemental material). RF types correlating to the Clostridiales fraction of trial 1 showed successive development during weeks 1 through 10. Thereafter, previously observed RF types randomly reemerged. At week 18, the Clostridiales RF types were replaced by Firmicutes RF types. RF types representing Clostridiales in trial 2 were different from those in trial 1 and showed successive development during weeks 1 through 11, with random reemergence thereafter. During weeks 1, 7, 12, and 17, a single RF correlating to a Mollicutes sequence predominated in the trial 2 restriction profiles.

PCoA of the T-RFLP fingerprints mirrored the ARISA results, indicating community succession over time (see Fig. S1 in the supplemental material). However, NP-MANOVA analyses of the T-RFLP fingerprints identified the week 11, but not the week 4, community conversion (Table 2), and this was reflected by the lack of Mantel correlation between the ARISA and T-RFLP results (Table 4). Community dynamics detected by T-RFLP, including the presence of a conversion at week 11, were statistically repeatable across both trials (Table 3).

TABLE 4.

Mantel comparisons of patterns between data sets^a

Trial and data set	Data set
	C. jejuni	C. coli	ARISA	T-RFLP	Therapeutics	CFT
Trial 1
C. jejuni		−0.01101	0.3007	0.1389	0.06812
C. coli	0.3332		−0.1213	0.3198	−0.1388
ARISA	0.0626	0.7548		−0.07791	−0.1836	0.6922
T-RFLP	0.1836	0.1048	0.6372		−0.211
Therapeutics	0.3284	0.6526	0.8010	0.8076
CFTb			<0.0001
Trial 2
C. jejuni		0.6144	0.4924	−0.1914	−0.2251
C. coli	0.0012		0.3967	−0.1917	−0.3873
ARISA	0.0162	0.0356		−0.1084	−0.314	0.6730
T-RFLP	0.8656	0.8890	0.7072		0.1833
Therapeutics	0.8044	0.9850	0.9060	0.3008
CFT			<0.0001

^aCorrelation R values are reported above the diagonal, and P values are reported below the diagonal. Statistically significant correlations are reported in boldface.

^bCFT indicates pooled ARISA data for six cohabiting turkeys sampled during 18 weeks in an antibiotic-free environment (38).

Campylobacter quantification.

Campylobacter was first detected in trial 1 at week 2 and in trial 2 at week 3, and initial colonization was dominated by C. coli in both cases. The quantification of Campylobacter jejuni and C. coli over time for the two trials indicated that pathogen loads generally were low, with 33 C. jejuni hipO genomes per 100,000 16S genes (statistical error, 23) and 13 C. coli glyA genomes per 100,000 16S genes (statistical error, 7) (Fig. 3). Real-time results indicated transient surges in Campylobacter loads, most notably C. jejuni at week 9 in trial 1 (142 genomes) and weeks 5 and 7 in trial 2 (780 and 147 genomes, respectively) and C. coli at week 14 in trial 1 (238 genomes) (Fig. 2A; also see Table S3 in the supplemental material). In trial 2, Campylobacter blooms coincided with the first microbiota conversion at week 4, and Campylobacter reductions coincided with the second conversion at week 11 (Table 4; also see Table S1 in the supplemental material). Excluding the early phase from analysis, C. jejuni reduction also coincided with the week 11 conversion in trial 1 (data not shown).

FIG. 3.

Campylobacter jejuni and Campylobacter coli quantification throughout two time course trials. (A) Trial 1; (B) trial 2. Real-time PCR quantifications of the hipO (C. jejuni; black bars) and glyA (C. coli; gray bars) genes were performed on pooled samples from five birds at each time point (22). The detection of Campylobacter gene copy numbers is reported per 100,000 16S gene numbers.

DISCUSSION

This work reports the development of cecal bacterial communities of turkeys throughout the commercial production cycle, including an examination of Campylobacter populations over time. Results described here complement a previous study in which five to six individual animals raised under laboratory conditions were examined over time in the absence of growth promotants or therapeutic antibiotics (38). The previous experiment allowed the description of community development within individual hosts, thus reducing interindividual artifacts. The main findings of that study were the presence of a substantial community conversion in the ceca around week 11 posthatch with concurrent Campylobacter colonization in one trial. The current study used turkeys raised in a commercial farm setting in which animals were exposed to different antibiotic regimens, feed compositions, and weather conditions.

Fingerprinting methods were used to examine community structural changes over time and to analyze the components of the observed transitions. All fingerprinting methods, including ARISA and T-RFLP, have limited sensitivity, detecting the most abundant 1% of the microbiota (32). Therefore, fingerprinting results reported here potentially miss functionally important low-density populations.

Intestinal microbiota succession is well documented in the ceca of young chickens (19, 26, 44, 47). In the current study, microbiota community dynamics, measured as the fingerprint dissimilarities between all pairs of time points, were described by two different fingerprinting methods. The PCoA of the ARISA fingerprints from both trials revealed arches through principle coordinate space, with a continuum of early-phase fingerprints present at one end through late-phase fingerprints at the other (Fig. 1). The distribution of the fingerprints along the arch is indicative of the unimodal distribution of species along a gradient, in this case time, indicating that various species have one optimal environment along the gradient (24). In both trials, the first and second principle axes detected two community conversions delineating three developmental phases, initial (weeks 1 to 3), median (weeks 4 to 10), and late (weeks 11 to 18) (Fig. 1 and 2). UPGMA cluster analysis also was able to detect the three phases (Fig. 2). The PCoA of the T-RFLP profiles, derived from a single animal at each time point, detected only the week 11 conversion between the median and late phases (see Fig. S1 in the supplemental material). Microbiota community dynamics were compared between the two trials via the Mantel test and found not to be statistically different by either fingerprinting method (Table 3).

Statistical comparisons of the microbial community dynamics in both trials to those of a previously reported trial indicated that in all three cases the community dynamics were not significantly different (Table 4). Therefore, commercial production practices, including the use of antimicrobials, do not affect large-scale microbial community dynamics.

In addition to community dynamics, the T-RFLP analyses revealed the presence of 11 distinct Bacteroidales RF types throughout both trials (see Tables S1 and S2 in the supplemental material). Although the day-old birds from the two trials were placed in the brooder houses 3 weeks apart, poults for both trials originated from a single hatchery. A recent report indicated that the chicken intestine is colonized in ovo; thus, the Bacteroidales RF types may have derived from the hatchery (34). It is unlikely that these 11 RF types derived from the farm environment, as the brooder houses were 2.5 miles distant. Nevertheless, the presence of the Bacteroidales RF types in this experiment, correlating to published poultry microbiota sequences, suggests that these organisms play an important role in the cecum. Bacteroidales sequences thus identified had very low (order-level) similarity to cultured isolates.

In contrast to the Bacteroidetes communities, the development of the Firmicutes communities was very different between the two trials. In both cases, the dominant Firmicutes RF types were novel at each time point until week 10 or 11, indicating succession. However, very few Firmicutes RF types were shared between the two trials. It therefore is likely that Firmicutes colonizers originate from the environment and that much of the succession observed in the fingerprinting profiles derives from that phylum. Differences in microbial composition between trials, combined with the similarity in community dynamics, indicate that (i) large-scale community dynamics likely are driven by host factors, and (ii) functional redundancy is common in the microbiota.

Campylobacter quantification during trials 1 and 2 indicated pathogen loads at levels of less than 1,000 genomes per 100,000 16S genes, and at most time points less than 60 genomes per 100,000 16S genes (Fig. 3; also see Table S3 in the supplemental material). However, the pathogen load generally was greater in trial 1, including both C. jejuni and C. coli blooms, than the load in trial 2. In trial 2, with the exception of a C. jejuni bloom detected in weeks 5 and 7, both C. jejuni and C. coli averaged 1.3 to 1.5 genomes per 100,000 16S genes.

Mantel tests indicated that changing Campylobacter loads in trial 2 occurred concurrently with one another and with the microbiota conversions (Table 4). A lack of correlation between the two trials suggests that Campylobacter loads are not directly affected by host signals but rather respond to the presence of subsets of the microbiota (Table 3). In addition, the discordant Campylobacter species population dynamics apparent in trial 1 indicate that C. jejuni and C. coli have dissimilar responses to given environments (Table 4). More in-depth studies need to be done to identify microbes present at the conversion points as well as predominant community functions at these times.

The effect upon the intestinal microbiota is unknown for most antibiotics. A recent metagenomic analysis of human microbiota disrupted by ciprofloxacin treatment indicated that, while antibiotic treatment reduced diversity in all subjects, individual communities responded uniquely (10). Of great note, however, was the finding that during a 6-month period, some taxa did not recover pretreatment levels.

In the current turkey time course study, antibiotic use and feed regimens were at the discretion of the turkey producer and not experimentally determined, and no statistical correlation was detected between the treatment programs for the two flocks (Table 3). Thus, the direct causation of community changes by these factors should not be inferred. However, as reported above, the community dynamics similarities between the on-farm trials and the previously reported antibiotic-free trial suggest that the ARISA fingerprinting method was not sensitive enough to detect community changes due to these therapeutic treatments (Table 4) (38). In addition, statistical examination did not identify correlations between therapeutic treatments and either microbiota changes or Campylobacter loads (Table 4). The lack of correlation between therapeutic treatments and Campylobacter loads was not surprising, as Campylobacter itself is not sensitive to penicillins or sulfadimethoxine and has demonstrated resistance to roxarsone (13, 33, 36).

The results reported here indicate that despite the vagaries of commercial production conditions, significant microbiota dynamics are maintained, including significant community transitions around weeks 4 and 11, and an increased C. jejuni presence is seen mid-grow-out. The author believes that the combined results from the reported trials support a hypothesis that host signals affect intestinal community development and contribute to Campylobacter ecology. Further work is needed to identify the in vivo environmental conditions that collaborate to enhance Campylobacter loads in the poultry cecum.