Postinfective bowel dysfunction following Campylobacter enteritis is characterised by reduced microbiota diversity and impaired microbiota recovery

Abstract

Objectives

Persistent bowel dysfunction following gastroenteritis (postinfectious (PI)-BD) is well recognised, but the associated changes in microbiota remain unclear. Our aim was to define these changes after gastroenteritis caused by a single organism, Campylobacter jejuni, examining the dynamic changes in the microbiota and the impact of antibiotics.

Design

A single-centre cohort study of 155 patients infected with Campylobacter jejuni. Features of the initial illness as well as current bowel symptoms and the intestinal microbiota composition were recorded soon after infection (visit 1, <40 days) as well as 40–60 days and >80 days later (visits 2 and 3). Microbiota were assessed using 16S rRNA sequencing.

Results

PI-BD was found in 22 of the 99 patients who completed the trial. The cases reported significantly looser stools, with more somatic and gastrointestinal symptoms. Microbiota were assessed in 22 cases who had significantly lower diversity and altered microbiota composition compared with the 44 age-matched and sex-matched controls. Moreover 60 days after infection, cases showed a significantly lower abundance of 23 taxa including phylum Firmicutes, particularly in the order Clostridiales and the family Ruminoccocaceae, increased Proteobacteria abundance and increased levels of Fusobacteria and Gammaproteobacteria. The microbiota changes were linked with diet; higher fibre consumption being associated with lower levels of Gammaproteobacteria.

Conclusion

The microbiota of PI-BD patients appeared more disturbed by the initial infection compared with the microbiota of those who recovered. The prebiotic effect of high fibre diets may inhibit some of the disturbances seen in PI-BD.

Trial registration number

NCT02040922.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Changes in the microbiota including reduction in some Clostridial taxa and increases in Proteobacteria have been variably reported in irritable bowel syndrome (IBS).

• Approximately 13% of IBS patients report a postinfectious origin (PI-IBS).

• Campylobacter enteritis alters the microbiota and 14% of cases develop PI-bowel dysfunction (PI-BD) but how these are linked is unclear.

WHAT THIS STUDY ADDS

• Recovery of the microbiota in PI-BD differed significantly from those whose bowel habit had returned to normal.

• PI-BD was associated with a reduction in Firmicutes and increase in Proteobacterial taxa (including taxa from class Gammaproteobacteria) which persisted for >12 weeks.

• Low consumption of fibre was associated with increased levels of Gammaproteobacteria.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• By indicating specific changes in microbiota in PI-BD, it will facilitate targeted manipulation of microbiota (eg, dietary fibre, probiotics or faecal microbiota transplants) to restore normal function.

Background

Postinfectious irritable bowel syndrome (PI-IBS) is a well-recognised symptom complex occurring in about 1 in 10 of cases of enteritis¹ and may account for up to 13% of all IBS cases.² The risk of developing PI-IBS appears to be greater in protozoan and bacterial enteritis as compared with viral gastroenteritis.¹ The associated activation of the immune system is an important strategy for pathogens infecting the gut since it suppresses the resident microbiota, particularly anaerobes, allowing overgrowth of the infecting pathogen³ as well as other potentially pathogenic taxa. The reduction in anaerobic metabolites, including short chain fatty acids (SCFAs) and secondary bile acids, raises the colonic luminal pH^{4 5} and reduces colonisation resistance, typically allowing an overgrowth of both the pathogen and Proteobacteriacae, including facultative anaerobes such as Enterobacteriacae.⁶

The definition of a healthy microbiota is complicated due to the large compositional variation between subjects.⁷ Nonetheless, parameters such as high diversity and gene richness, abundance of SCFA production and resilience are considered to be relevant markers of health.^{7 8} A resilient microbiota is able to return to its original composition after facing a perturbation, such an infection, whereas non-resilient microbiota may shift its composition permanently to a new altered state of dysbiosis.^{8 9} It has been shown that in healthy subjects, the gut microbiota recovers rapidly after a non-inflammatory diarrhoea such as that induced by osmotic laxatives like macrogols, when the colonic lumen is alkalinised.¹⁰ This has been associated with a profound depletion of anaerobes and an increase in Proteobacteria but the observed dysbiosis was largely reversed after 14 days.¹¹ However, what host or dietary factors determine the recovery of the microbiota after an inflammatory diarrhoea is unclear, while the potential lack of resilience has not yet been characterised in patients developing PI-IBS. Most studies of PI-IBS combine patients infected by varying pathogens, which introduces considerable variability since each pathogen has unique features. Our work has attempted to reduce this source of variability by focusing on a single pathogen, Campylobacter jejuni, ¹² one of the most common causes of bacterial gastroenteritis in Europe.¹³

Previous pilot studies have shown that PI-IBS following Campylobacter enteritis could be characterised by an index of microbial dysbiosis based on 27 taxa, which distinguished PI-IBS from controls. It was characterised by a 12-fold increase of Bacteroidetes taxa in patients, and a 35-fold reduction in the strict anaerobes characterised as uncultured Clostridia compared with healthy controls.¹⁴ These findings were replicated in a meta-analysis including an additional PI-IBS group.¹⁵ Furthermore, similar findings were seen in those who had persistent bowel dysfunction (BD) after C. jejuni enteritis but who did not meet Rome criteria (postinfectious BD, PI-BD).¹⁴

The aim of this study was to define in more detail and with greater patient numbers the serial changes in microbiota recovery over the 3 months following a culture-proven infectious gastroenteritis due to C. jejuni. We compared the microbiota composition, bowel symptoms, stool form and dietary habits and potential predisposing factors of PI-BD patients with controls whose bowels had returned to normal within 3 months of infection. Previous studies indicated that PI-BD is more common than PI-IBS but has similar bowel disturbance, namely persistent diarrhoea, the main difference being lack of pain.¹⁶ We hypothesised that there would be a difference between those with PI-BD both in their response to infection and during the recovery period. More specifically, we expected to see an initial loss of microbial diversity for all patients, with a greater disturbance in those who went on to develop persistent BD. We aimed to identify these indicators of non-resilience leading towards PI-IBS-associated microbiota.

Materials and methods

Subjects and study design

This was a single-centre cohort study of patients who tested positive for Campylobacter spp. in the Public Health England Laboratory in Nottingham. The General Data Protection Regulations and heavy workload meant that potential participants were informed of their diagnosis and invited to participate by weekly mail out. Only once the subject had made contact could we then negotiate a date for a visit. This meant that the first visit was several weeks after the initial diagnosis. Figure 1 shows recruitment details. The clinical study included all 155 eligible subjects who provided clinical details of their illness, psychological parameters and bowel function.

Figure 1.

CONSORT diagram. The 48 mechanistic controls were chosen because they provided the most complete set of stool samples. The mechanistic study was confirmed to be unbiased from the larger clinical study by demonstrating there were no significant differences in demographics, psychological scores nor markers of initial illness severity (online supplemental tables S6–7). CONSORT, Consolidated Standards of Reporting Trials; IBD, inflammatory bowel disease; IBS, irritable bowel syndrome.

The first stool sample was collected as early as possible following microbiological diagnosis and further samples were collected 6 and 12 weeks after diagnosis. Our previous study indicated that symptoms persisting at 12 weeks would be long-lasting (ie, >6 months).¹⁶ However, administrative delays meant that the first faecal sample was collected at visit 1 which was a mean of 46 days (range 17–93) and the final sample at visit 3 was collected at mean 97 days (range 57–160) from the start of symptoms.

At visit 1, eligibility was confirmed and written informed consent obtained. Demographics and current bowel habits were recorded, and all completed the Hospital Anxiety and Depression Scale¹⁷ and the Patient Health Questionnaire-12 Somatic Symptom Scale (PHQ-12 SS).¹⁸ They were also asked about features of the acute illness with markers of severity including rectal bleeding, vomiting, weight loss, duration of time off normal activities and any antibiotic treatment.

Patients were asked to collect stool samples for each visit (see online supplemental methods for more details). If visit 1 occurred within 5 weeks of diagnosis, patients were asked to return for visit 2 at 6 weeks (typically 1 week later) to provide a further stool sample. At visit 3, 12 weeks after diagnosis, patients were asked to complete a questionnaire on their bowel symptoms from the past week and provide a further stool sample.

Dietary data

We analysed 7-day completed food diaries at visit 2 and visit 3 of 19 cases who returned an adequate food diary and age-matched and sex-matched them to 31 controls. Dietary data from each recording was manually entered into a dietary software package: Dietplan 7 (Forestfield Software V.7.00.64) for nutrient analysis. Macronutrient and micronutrient analysis was based on McCance and Widdowson’s food composition data, UK. A cut-off for energy intake was set for energy levels of ≤800 kcal or ≥4500kcal/day to remove implausible reported intake.

Stool measurements

Stool SCFA concentrations and dry weights were measured in visit 2 and visit 3 samples from 14 cases and 23 controls who provided adequate additional faecal samples. Samples were analysed using gas chromatography-mass spectrometry as described previously.¹⁹

Microbiota analysis

Faecal DNA was extracted using a validated method.^{20 21} In short, cells were lysed using a bead beater (MagNA lyser, Roche diagnostics, Indianapolis, USA). Ammonium acetate, isopropanol and centrifugation were used to precipitate the proteins and nucleic acid. A commercially available kit (QIAamp DNA Mini Kit, Qiagen, Venlo, Netherlands) was used to clean the DNA by removing the RNA and proteins. The DNA was eluted in 200 μL nuclease-free water.

Microbiota composition was analysed with the Illumina MiSeq platfrom amplifying the V3-V4 hypervariable region of the 16S rRNA gene.²² The obtained sequence reads (on average 88 213 per sample) were prepossessed with the Mare R package²³ ProcessReads and TaxonomicTable functions the use of these is detailed in online supplemental methods. We used the SILVA 16S rRNA reference database (version 115) for taxonomic assignment. After preprocessing, there were on average 64 385 reads per sample (ranging from 28 680 to 351 004). The reads have been deposited to ENA (PRJEB52306).

Outcome measures

Clinical study

The primary outcome was the proportion of patients with BD 12 weeks after laboratory report of infection, hereafter described as PI-BD. This was defined by answering ‘no’ to the question ‘have your bowels returned to normal since your Campylobacter infection?’ at visit 3. We used this simpler measure rather than the Rome definition since we knew from previous studies¹⁶ that a substantial number of those who complained of persistently altered bowel habit did not meet Rome criteria, mainly because they did not experience significant pain, despite having all the other key symptoms. Secondary outcomes included number of patients meeting Rome III criteria for IBS (other than 6-month duration) to allow easier comparison with other studies. We also examined the influence of age, gender, psychological factors and severity of initial illness on the risk of developing PI-BD.

Microbiota analysis

The primary outcome was microbiota recovery as assessed from diversity, richness and the abundance of key bacterial taxa. Secondary outcomes were associations between dietary components and SCFA concentrations and stool water content.

Statistical analysis and sample size calculation

Clinical study

Data are represented by mean (SD) and non-symmetrical data by median (IQR). All statistical analyses were performed by using R (V.3.6.1) and GraphPad Prism (V.8.2.1). Normality was tested with D’Agostino’s K² test. Statistical differences of markers of disease severity were tested using Fisher’s exact test or unpaired t-test, depending on normality.

We originally planned to recruit 450 participants aiming for 80% power to detect an increase in PI-BD to 39% in those taking antibiotics from 25% in those not taking antibiotics, assuming that 30% took antibiotics. However, the end of funding was reached with only 155 subjects recruited so we were substantially underpowered for this endpoint. However, the mechanistic study was larger than expected, being one of the largest in-depth study of the changes in microbiota following Campylobacter enteritis.

Microbiota analysis

To exclude biases due to antibiotics consumption, we excluded all samples collected from those subjects who consumed antibiotics (n=18, 9 each in cases and controls) until 60 days after reported infection. All taxonomic ranks from phylum down to genus level taxa were used for statistical testing. Microbial alpha-diversity was assessed using inverse Simpson diversity index using amplicon sequence variance (ASV)-level data. There was no significant correlation between alpha diversity and sample read counts (see online supplemental methods). Principal co-ordinate analysis (PCoA) with Bray-Curtis dissimilarities was used to visualise microbial beta-diversity using ASV-level data. The statistical difference between groups in the PCoA was tested using permanova and using vegan package function adonis. To test differences in the bacterial abundance between cases and controls and the associations between the bacterial taxa and nutritional components and SCFA amounts, generalised linear mixed models were used (detailed in online supplemental methods). Here, the read number for each sample was used as an offset and subject’s age was used as a confounding factor. This was also supported by Spearman correlation testing. The obtained p values were adjusted for multiple testing with the falce discovery rate (FDR) approach, and FDR-adjusted p values (q-values) below 0.05 were considered to be significant.

Results

Clinical study

There were 22 of the 99 subjects who completed the trial, who reported that their bowels had not returned to normal after the infection (cases) and 77 subjects whose bowels had normalised (controls). As table 1 shows, cases were significantly more likely to be younger, female and scored significantly higher on the assessment of somatisation. The main features recorded in the PHQ-12 SS distinguishing cases from controls were trouble sleeping, headaches, back and limb pain and lethargy (figure 2A). The main features of the BD included more bloating, more frequent episodes of pain associated with loose stools, more urgency and stools being more often loose or watery (see table 1).

Table 1.

Patient demographics at baseline

	Cases	Controls	P value
Subjects	22	77	–
Age, median (IQR)	57 (41–64)	62 (48–71)	0.05
Female, n (%)	18 (82)	33 (65)	0.002
PHQ-12 SS, median (IQR)	5 (3–6)	2 (1–4)	0.002
HADS-A, median (IQR)	5 (4–10)	5 (3–7)	0.22
HADS-D, median (IQR)	4 (1–6)	3 (1–5)	0.67
Weekly stool frequency preinfection, median (IQR)	7 (7–7)	7 (7–14)	0.31
Weekly stool frequency postinfection, median (IQR)	9 (6–14)	7 (7–14)	0.55
Recurrent pain in last 14 days	57%	21%	<0.001
Pain associated with loose stools	71%	48%	<0.001
Reported bloating	57%	17%	<0.001
Reported urgency	52%	30%	0.08
Stools often loose or watery	59%	13%	<0.001

HADS-A, Hospital Anxiety and Depression Scale-Anxiety subscale; HADS-D, HADS–Depression subscale; PHQ-12 SS, Patient Health Questionaire-12 Somatic Symptom Scale.

Figure 2.

Differences in patients’ symptoms 3 months after gastroenteritis (A) average PHQ-12S scores for cases and controls, showing the increased prevalence of trouble sleeping (p<0.0001), headaches (p=0.034), back pain (p=0.015) and limb pain (p=0.0248) in cases. Statistical significanse indicated with asterisk. (B) Proportion of loose and watery stools and water content. Cases were significantly more likely to report loose/watery stools which was confirmed with the significant difference in stool water content (p=0.04). (C) GI-symptoms. The cases also reported significantly more sensations of bloating (p<0.001) and urgency (p<0.001). GI, gastrointestinal; PHQ-12S, Patient Health Questionnaire-12 Somatic.

Characterising PI-BD

Cases were characterised by significantly looser stools 3 months after infection (figure 2B and table 2). Stool water content of cases was significantly greater than controls (cases, n=14, mean (SD) 77.95 (6.70)%; controls, n=23, mean (SD) 71.97 (7.83)%, Fisher’s exact test p=0.04, figure 2C). In addition, cases more often reported a sensation of urgency and bloating, and visible swelling of the abdomen (table 2). Rome III criteria for IBS were fulfilled in 10 (45%) cases who were very similar to the remaining 12 that did not meet the criteria (PI-BD) with no significant difference in age, anxiety, depression nor PHQ-12 SS. In addition, markers of severity of gastroenteritis did not differ significantly between PI-BD or PI-IBS, including fever, blood in stool, vomiting nor antibiotic consumption (see online supplemental table S1).

Table 2.

Features of postinfective bowel dysfunction 3 months after Campylobacter infection comparing cases versus controls

	Cases (n=22)	Controls (n=77)	RR (95% CI)	P value
Stools often loose or watery?	12 (55%)	12 (16%)	3.8 (1.9 to 7.4)	<0.001
Stools often hard or lumpy?	2 (9%)	15 (19%)	0.5 (0.1 to 1.5)	0.347
<3 bowel movements per week	0 (0%)	3 (4%)	0 (0 to 2.6)	>0.999
>3 bowel movements per day	6 (27%)	10 (13%)	1.9 (0.9 to 3.9)	0.185
Presence of mucus	1 (5%)	1 (1%)	2.3 (0.4 to 5.1)	0.397
Straining on defecation	4 (18%)	9 (12%)	1.5 (0.6 to 3.2)	0.477
Sensation of incomplete evacuation	10 (45%)	18 (23%)	2.1 (1.0 to 4.2)	0.06
Sensation of abdominal bloating	13 (59%)	14 (18%)	3.9 (1.9 to 7.9)	<0.001
Abdominal swelling	8 (36%)	6 (8%)	3.5 (1.7 to, 6.4)	0.002
Urgency	11 (50%)	18 (23%)	2.4 (1.2 to 4.8)	0.031
IBS by Rome III criteria?	10 (45%)	0	N/A	N/A

IBS, irritable bowel syndrome; N/A, not available.

Markers of gastroenteritis severity

We found that cases were significantly more likely to report a fever during gastroenteritis (82% cases and 55% controls, p=0.02) but other markers of severity such as blood in stool, vomiting, days off work or weight loss were not significantly different between cases and controls (see online supplemental table S2).

Effect of antibiotics and concomitant medication on disease recovery

There was no significant difference in the proportion of cases versus controls who received antibiotic prescription (41% and 32%, respectively, Fisher’s exact test p=0.45). Patients who received antibiotics did not appear to have any worse symptoms during the initial illness and had no clinical features significantly different from those who did not (see online supplemental table S3), however, they were significantly more likely to attend their general practitioner (GP) more than once for this illness (50% vs 28%, Fisher’s exact test p=0.05). Most of our patients were healthy and taking no medication, which can of course affect the microbiota. A small number of both patients and controls took a range of medications with no consistent difference between the groups (online supplemental table S4).

Dietary habits

A subset of the subjects’ (19 cases and 31 controls) dietary habits as well as faecal SCFA concentrations (14 cases and 23 controls) were assessed from visits 2 and 3. There were no significant differences in any of the nutrition components or faecal SCFAs between cases and controls or between either of the time points (online supplemental table S5).

Microbiota study

The demographics and disease severity of both the cases and controls in the mechanistic study did not differ significantly from those of the larger cohort (see online supplemental table S6 and S7, respectively).

Microbiota composition in samples collected less than 40 days after gastroenteritis is impacted by infection

The largest influence on the microbiota composition was the time since the initial infection, with a gradual recovery over the 12 weeks of study. The early samples, collected less than 40 days after reported infection, were significantly different from the later samples (MANOVA, p=0.001, figure 3A). The differences in microbiota recovery are characterised in online supplemental table S8-S10. In addition, there were significant differences in microbiota recovery in cases as compared with the controls (MANOVA, p=0.045, figure 3B, C). These significant changes were due to increased levels of the genera Collinsella (mean relative abundance 10.7% in cases vs 4.31% in controls, negative binomial generalised linear model q≤0.001) and Eggerthella (1.82% in cases vs 0.18% in controls, negative binomial generalised linear model, q=0.06, (online supplemental table S8). In addition, there was a significant decrease among cases in many taxa belonging to Firmicutes phyla, these included reduced levels of genera Faecalibacterium (6.06% in cases vs 8.45% in controls, negative binomial generalised linear model, q<0.001), Enterococcus (0.05% in cases vs 0.39% in controls, negative binomial generalised linear model, q=0.003) and taxa from the Ruminococcaceae family (11.66% in cases vs 18.22% in controls, negative binomial generalised linear model, q<0.001) (online supplemental table S8).

Figure 3.

Microbiota recovery after infection in cases and controls. (A) PCoA plot with Bray-Curtis dissimilarity from all subjects. The largest variation in microbiota composition is due to time since infection, samples obtained early after infection being significantly different from the later ones (MANOVA multivariate analysis of variance, p=0.001). The coloured circles represent 50% of the data. (B) Inverse Simpson diversity. Microbial recovery during the follow-up period was different between cases and controls. The inverse Simpson diversity shows that cases fail to recover to normal levels in samples collected more than 80 days after infection. (C) Proportion of total of Clostridia, Coriobacteriia and Fusobacteria. There were also significant class level differences including lower clostridia, but higher Coriobacteriia and Fusobacteria (for details, see online supplemental table S8-S10). SE of mean is shown as whiskers and statistically significant difference (p<0.05) is shown with asterisk. PCoA, principal co-ordinate analysis.

Microbiota recovery

We aimed to focus on the difference in microbiota recovery between cases and controls and concentrated on the late samples collected more than 60 days after the reported infection when we had the most samples since those who had taken antibiotics were no longer excluded. In these samples, alpha diversity (mean 12.1 in cases vs 15.8 in controls, ANOVA, p=0.015) and richness (mean 132.1 in cases vs 149.2 in controls, ANOVA, p=0.017, online supplemental figure S1) were significantly decreased in cases when compared with controls. Furthermore, the abundance of several taxa were significantly different between cases and controls in samples collected >60 days after reported infection (table 3). There was a significant decrease in the abundance of bacteria from the phylum Firmicutes, especially taxa from the order Clostridiales, which were reduced by 20.1% when compared with controls. More specifically, taxa belonging to Clostridiales such as Ruminococcaeceae and Christensenella were both significantly reduced in cases. Moreover, there were two genera of Coriobacteria (Eggerthella and Goronibacter) that were more abundant in cases and the abundance of the family Coriobacteriaceae was increased by 32.2% in those with persistent BD in samples collected more than 60 days after reported infection. In addition, Fusobacteria and several taxa from the phylum Proteobacteria were increased in cases, these included a 35.4-fold increase of Klebsiella (a member of the Gammaproteobacteria class).

Table 3.

The significantly different taxa between cases and controls in samples collected more than 60 days after reported infection

Phylum	Class	Order	Family	Genus	Cases (n=18)	Controls (n=48)	Fold change
Actinobacteria	Actinobacteria	Actinomycetales	Actinomycetaceae	Actinomyces	0.16%	0.06%	2.75
Actinobacteria	Coriobacteriia	Coriobacteriales	Coriobacteriaceae		13.61%	9.23%	1.47
Actinobacteria	Coriobacteriia	Coriobacteriales	Coriobacteriaceae	Eggerthella	2.59%	0.79%	3.28
Actinobacteria	Coriobacteriia	Coriobacteriales	Coriobacteriaceae	Gordonibacter	0.35%	0.05%	6.38
Bacteroidetes	Bacteroidia	Bacteroidales	Porphyromonadaceae	Butyricimonas	0.03%	0.10%	0.27
Bacteroidetes	Bacteroidia	Bacteroidales	Porphyromonadaceae	Porphyromonas	0.02%	0.12%	0.12
Firmicutes					55.57%	64.85%	0.86
Firmicutes	Clostridia	Clostridiales			44.60%	53.56%	0.83
Firmicutes	Clostridia	Clostridiales	Christensenellaceae	Christensenella	0.35%	0.71%	0.50
Firmicutes	Clostridia	Clostridiales	FamilyXIIIIncertaeSedis		0.07%	0.19%	0.38
Firmicutes	Clostridia	Clostridiales	Ruminococcaceae		17.76%	23.69%	0.75
Firmicutes	Clostridia	Clostridiales	Ruminococcaceae	Anaerofilum	0.13%	0.28%	0.46
Firmicutes	Erysipelotrichia	Erysipelotrichales	Erysipelotrichaceae	Kandleria	0.15%	0.10%	1.51
Firmicutes	Erysipelotrichia	Erysipelotrichales	Erysipelotrichaceae	Solobacterium	0.00%	0.21%	0.01
Firmicutes	Negativicutes	Selenomonadales	Acidaminococcaceae	Phascolarctobacterium	0.06%	0.41%	0.16
Firmicutes	Negativicutes	Selenomonadales	Veillonellaceae	Dialister	1.84%	1.16%	1.58
Firmicutes	Negativicutes	Selenomonadales	Veillonellaceae	Veillonella	0.09%	0.27%	0.34
Fusobacteria					0.26%	0.01%	19.32
Fusobacteria	Fusobacteriia	Fusobacteriales	Fusobacteriaceae	Fusobacterium	0.26%	0.01%	19.32
Proteobacteria	Betaproteobacteria	Burkholderiales	Burkholderiaceae	Burkholderia	0.04%	0.00%	16.42
Proteobacteria	Deltaproteobacteria	Desulfovibrionales	Desulfovibrionaceae	Desulfovibrio	0.02%	0.22%	0.11
Proteobacteria	Gammaproteobacteria	Enterobacteriales	Enterobacteriaceae	Klebsiella	0.73%	0.02%	35.42
Proteobacteria	Gammaproteobacteria	Pasteurellales	Pasteurellaceae	Haemophilus	0.02%	0.08%	0.22

The mean relative abundance of each taxa is shown along with the fold change in cases versus controls.

Associations between microbiota and dietary components, SCFA and stool water content

Although the cases and controls did not differ in their dietary habits (54 records in total) or SCFA concentrations (52 records in total) we found several associations with their microbiota profiles. There were 38 significant associations computed with linear models between the microbiota composition and measured SCFAs (online supplemental table S11) all values were also supported with significant spearman correlation. These included the positive association between butyric acid and the genus Faecalibacterium (linear mixed effects (log), q=0.09, r=0.384) and negative association between the total SCFA concentration and Gammaproteobacteria (generalised linear mixed models, q=0.01, r=−0.36). In addition, there were 23 associations to food components. Most strikingly, there was a strong negative association between levels of Gammaproteobacteria and the consumption of fibre (generalised linear mixed models, q=0.03, r=−0.46, figure 4), non-starch polysaccharides (generalised linear mixed models, q=0.05, r=−0.47) and starch (generalised linear mixed models, q=0.003, r=−0.43, online supplemental Table S12). Increased stool water content was associated with increased levels of the class betaproteobacteria (generalised linear mixed models, q=0.03, r=−0.23).

Figure 4.

Association between fibre consumption and gammaproteobacterial abundance. The association was statistically significant (q=0.032), where low consumption of fibre was associated with high Gammaprotebacteria abundance. Light area indicates SE of mean.

Discussion

We confirmed previous findings that PI-BD followed by Campylobacter infection is characterised by loose stools, bloating and urgency suggesting faster overall transit.12 16 24 25 What determines this change in function is unclear, but we now report that the microbiota recovery from gastroenteritis was slower and less complete in PI-BD cases than controls. A key feature which could be relevant to the ongoing new symptoms includes a significantly lower diversity, which we found in the early samples. This was significantly greater in cases as compared with controls and this persisted more than 60 days after the reported infection, regardless of antibiotic use. This is likely to be due to inflammation since similar loss of diversity has been reported in association with Crohn’s disease^{26 27} and after norovirus infection,²⁸ which, as we found, were also associated with increased Proteobacteria. In our study, cases did not differ from controls in antibiotic use nor disease severity except a much greater proportion (94% vs 55%) reported fever. The changes in microbiota are likely therefore to reflect the combined effect of the resilience of the original microbiota together with the patient’s inflammatory response to C. jejuni. This depletes normal commensal bacteria and, by reducing colonisation resistance, allows the pathogen to proliferate.²⁹

The adult gut microbiome characteristically exists in a steady state requiring a major disturbance, such as a bout of gastroenteritis, to alter that state permanently. Indicative of such a shift in the cases of this cohort is the large and persistent changes in the major bacterial classes including the decreased levels of Clostridia, a taxon often associated with health benefits such as SCFA production. We found that the decrease in Clostridia was mirrored by the increase in classes such as Gammaproteobacteria in the cases as compared with controls more than 60 days after infection. Interestingly the levels of Gammaproteobacteria were inversely associated with total SCFAs and more specifically butyrate and propionate acids. In addition, the patient’s consumption of fibre, non-digestible polysaccharides and starch were negatively associated with Gammaprotebacteria abundance. There is substantial evidence that the health benefits of high fibre consumption are mediated in part via increased SCFA production which decreases pH in the colon, inhibiting the growth of Gammaproteobacteria.^30–32 Taken together, this suggests high fibre diets could contribute to correcting the microbiota disturbance and preventing PI-BD, something which should be further evaluated in randomised controlled clinical trials.

We showed here cases have a significant reduction in microbial diversity and the total Firmicutes, especially taxa from Clostridiales and Ruminococcaceae groups. This may reflect continuing disturbance of transit as reflected by increased stool water content and reporting loose or watery stools. This is in line with previous findings where, even in healthy subjects, soft stools were associated with reduced diversity.³³ Most individuals with firmer stools in that study had the Ruminococcaeae-Bacteroides enterotype showing how different consistency favours different species. Both fast transit and mucosal inflammation disturb the anaerobicity of the colonic environment, which depletes the strict anaerobes and allows facultative anaerobes and those with rapid replication such as Gammaproteobacteria and Fusobacteria, to proliferate and occupy the vacant ecological niche. Similar reductions in Firmicutes have been recently reported in children from Peru who were hospitalised with gastroenteritis, particularly those with bacterial infections like Campylobacter, Shigella and Salmonella.³⁴ Similarly, the persistent reduction of Firmicutes and increased Proteobacteria seen in IBD is thought to represent increased availability of small molecules created by the inflammatory process such as nitric oxide and reactive oxygen species that can act as electron acceptors for facultative anaerobes like Proteobaceria.³⁵

Several members of the Coriobacteriea family were increased in our PI-BD cases very early after the infection and this increased abundance persisted throughout the study. Previous studies have also associated this family with IBS.^36–38 Vich Vila et al who studied a cohort of 412 patients with IBS with shotgun metagenomics showed that IBS patients had increased levels of Coriobacteria, especially the genus Eggerthella. This was complemented with the decreased abundance of several important clostridial species including Ruminococcaceae, ³⁷ a pattern also detected in our cases. A similarly increased abundance of Coriobacteriaceae, Proteobacteria and Fusobacteria has been reported after Roux-en-Y surgery for obesity.^{39 40} The common aspect shared with previous findings and our PI-BD patients may be faster transit through the gut, which alters the colonic milieu in multiple ways including reducing secondary bile acids, raising pH and reducing SCFAs.

Although Fusobacterium accounts for only a small percentage of total bacteria it was markedly higher in our cases through-out the study. Fusobacterium has also been noted to be part of a characteristic cluster of organisms that bloom immediately after V. cholera infection.⁶ The pattern of low Firmicutes and increased Fusobacterium is of special interest since in stressed maternally separated rats the same pattern is seen and the severity of hypersensitivity to rectal distension in maternally deprived rats was correlated with Fusobacterium numbers.⁴¹ Furthermore, when gavaged into rats, Fusobacterium induces visceral hypersensitivity.⁴²

Previous studies suggested that the risk of PI-BD increased proportionate to the severity of the initial insult.¹⁶ We found that fever was an important risk factor in developing PI-BD, possibly a marker of severity reflecting the increased permeability due to C. jejuni infection⁴³ allowing systemic access of pyrogens such as lipopolysaccharide. Our findings differ from a recent meta-analysis where receiving antibiotics was deemed a risk factor for developing PI-IBS.¹ We did, however, find those receiving antibiotics were more likely to make more than one visit to their GP despite having similar markers of illness severity so it may reflect underlying differences in healthcare seeking behaviour rather than a direct effect of antibiotics. This is supported by our finding that cases had a significantly elevated PHQ12-SS, confirming other studies which have indicated that adverse psychological features such as neuroticism,⁴⁴ depression¹² and multiple non-gastrointestinal somatic symptoms² increase the risk of postinfective IBS. As the recent meta-analysis⁴⁵ reported females have an increased relative risk compared with males of developing PI-IBS, mean (95% CI) 2.2 (1.6 to 3.1). Relative risk in our study at 4.2 was higher despite an equal number of males and females taking part but why is unclear and gender did not appear to affect the microbiota.

Only a small proportion of the total 1286 infected patients chose to take part which raises the question of bias. However, the proportion of subjects developing PI-BD, 22% was in fact very close to the 25% reported in our less demanding survey previously reported in which response rate was much higher at 72%.¹⁶ This suggests that the severity of bowel disturbance is not a major factor in determining participation, but a multitude of other factors like altruism, proximity to study site and ability to take time of work. Those with PI-BD did show greater somatisation which has been found in other studies^{1 2} but the underlying mechanisms are unclear. By choosing those whose bowels returned to normal as controls we aimed to control for the many factors which influence both getting infectious gastroenteritis, attending a doctor and sending a stool sample to the Public Health laboratory which include age, gender, severity and most importantly the GPs beliefs, which vary widely.⁴⁶ The samples at 3 months of those who report bowel function back to normal would seem to be the best estimate of what is normal for the controls. Sampling was also limited by administrative obstacles which mean we could not get samples as early as we would have wished when the changes might have been more substantial, however, since our main focus is the long-term effects this is perhaps not such a limitation. Our attempt to avoid the effects of antibiotics by analysing samples taken at least 60 days after antibiotic consumption represents a compromise since excluding all 9/22 cases who took antibiotics would have seriously underpowered our study.

An important limitation of a descriptive study such as ours is that it does not allow one to distinguish cause from effect. An alternative interpretation of the lower diversity in cases is that those with lower initial diversity are less resilient and hence predisposed to a more severe infection and disturbance of gut function. Interestingly in a prospective study of Campylobacter infection among abattoir workers, a pre-existing higher abundance of Bacteroides and E. coli increased the risk of developing Campylobacter enteritis⁴⁷ suggesting that this profile leads to lower colonisation resistance. Our study adds to the existing information and invites further studies both to confirm the findings but also to include interventions such as high fibre/prebiotics or drugs to slow transit that might normalise the microbiota and improve symptoms.

Data availability statement

Data are available on reasonable request.

Ethics statements

Patient consent for publication

Consent obtained directly from patient(s).

Ethics approval

All patients gave written informed consent, and the study was approved on 26 September 2013 by the National Research Ethics Committee East Midlands – Nottingham 1 (reference code:13/EM/0310).