For stable expression of the pgl operon in the chromosome of avian pathogenic E. coli O:78(aroA) (Table 1), we created an operon that contains the minimal set of genes required for N-glycan expression. First, the pglB gene was removed from plasmid pACYC184pgl (ppgl); using ppgl as a template, one PCR product (gne-pglH) was amplified with oligonucleotides gne-SalI-F and pglJ-SbfI-R (Table 2) using the plasmid-borne SalI site upstream of gne and introducing a unique SbfI site after the stop codon of pglH. A second PCR product was generated with oligonucleotides pglA-SbfI-F and pglG-BamHI-R to amplify pglA to pglG, taking advantage of the BamHI site on the plasmid downstream of pglG and introducing a SbfI sequence upstream of pglA. Digested PCR products were inserted into plasmid pACYC184 cut with BamHI and SalI in a 3-arm ligation reaction, resulting in plasmid ppglΔpglB. For the insertion of the removable selection marker [kanamycin (kan) cassette flanked by FRT sites], a unique XhoI restriction site was generated between pglK and pglH: PCR products with oligonucleotide combinations pglK-XhoI-R, gne-SalI-F, and pglH-XhoI-F, pglJ-SbfI-R were used to amplify gne-pglK and pglH from template plasmid ppgl ΔpglB. SalI-XhoI digested gne-pglK and XhoI-SbfI digested pglH PCR products were inserted (3-arm ligation reaction) into SalI-SbfI digested plasmid ppglΔpglB resulting in plasmid ppglΔpglB-XhoI. Next, the kan cassette flanked by FRT sites was PCR amplified from plasmid pKD4 using oligonucleotides P1-pKD4-XhoI and P2-star-pKD4-XhoI introducing XhoI sites in the 3' and 5'-prime ends. The obtained product was digested with XhoI and inserted into the XhoI-linearized ppglΔpglB-XhoI plasmid. One positive candidate plasmid where the kan cassette is transcribed in the same orientation as the pgl genes was named ppgl-kan. To remove the upstream region of gne and genes pglC to pglG, a PCR product encompassing gne-pglK-kan-pglHIJA was generated with oligonucleotides gne-BamHI-F1 and pglA-BamHI-R and inserted into BamHI digested plasmid pACYC184 (Supplementary Figure S1A).

The targeted insertion locus the wzy (rfc) gene of strain APEC O:78 was amplified from chromosomal DNA with oligonucleotides O78pol-BamHI-F and O78pol-EcoRI-R (Table 2) and ligated into BamHI-EcoRI digested pBluescript II KS(+) (Table 1). For the insertion of the minimal kan cassette containing pgl operon, a unique BamHI compatible BglII site was generated by inverse PCR with oligonucleotides O78 inverse BglII-F and O78 inverse BglII-R using plasmid pBSK-wzyO78 as template. The obtained construct after re-ligation of the inverse PCR product was digested with BglII and PCR amplified gne-pglK-kan-pglHIJA was inserted. After ligation and selection for Kan and Amp resistant colonies, one candidate in which the pgl genes are transcribed in the same orientation as the wzy^O78 gene was (after linearization with EcoRV) used for integration into APEC O:78(aroA) following the method of Datsenko and Wanner (2000; Supplementary Figure S1B). The correct insertion of the construct was verified by PCR with pgl-gene-specific oligonucleotides and primers that hybridize outside of the recombination event. Expression of the lipid A-core N-glycan in the vaccine strain, APEC O:78(aroA)/wzy::pgl::kan (Table 2), was verified by Western blotting with Cj-N-glycan-specific (R1) antiserum as described (Nothaft et al., 2016; Supplementary Figure S1C).

Total DNA was extracted from chicken cecum samples as described (Nothaft et al., 2016). Composition of the bacterial community in cecal samples was characterized using 16S rRNA gene amplicon sequencing. PCR targeting the V6-V8 region of the 16S rRNA gene with primers 926F (5'-AAACTYAAAKGAATWGRCGG-3') and 1392R (5'-ACGGGCGGTGWGTRC-3'), and subsequent amplicon sequencing (Illumina MiSeq platform producing 300-bp paired-end sequences) was performed. Amplicon sequencing produced a total of 27,819,932 raw sequences (average=137,035; minimum=76,068 and maximum=208,755). For each sequencing run, raw reads were trimmed to 270 bases long using FASTX-toolkit (v.0.0.14).² R1 and R2 ends were quality filtered and paired using the merge-illumina-pairs application from Illumina utils v.2.0.1 (Eren et al., 2013). Sequences that did not meet the quality criteria (value of p of 0.03, enforced Q30 check, perfect matching to primers, and no ambiguous nucleotides allowed) were discarded. After merging and quality control, a total 13,124,620 sequences were obtained (average=71,376; minimum=10,795 and maximum=129,875). To make the dataset manageable, samples were randomly subsampled to obtain 20,000 reads using mother v.1.39.2 (Schloss et al., 2009). All reads were kept for samples that had less than 20,000 sequences (total of three samples; range of 10,795–17,532 sequences). Subsequently, reads were compiled and dereplicated, singletons were discarded, and chimeras removed using usearch v.10 (Edgar, 2013). After dereplication and chimera removal, all reads from all sequencing runs were combined and OTUs were clustered at 98% identity, representative sequences for each OTU were selected, and the OTU table was generated also using usearch v.10 (Edgar, 2013). Non-chimeric sequences were binned by sample and submitted to Ribosomal Database Project Classifier (Wang et al., 2007) for taxonomic assignment from phyla to genera. OTUs were assigned taxonomy using Silva database (Westram et al., 2011) and sequence identity was confirmed using NCBI blastn (Altschul et al., 1990), EZ biocloud (Yoon et al., 2017), and Ribosomal Database Project Seqmatch (Michigan State University, 2016). Counts were transformed to relative abundance. Taxa with a mean relative abundance less than 0.01% were removed from the dataset prior to statistical analyses. Diversity analyses were performed using Qiime v.1.9.1 (Caporaso et al., 2010).

To determine differences between responders and non-responders for the different taxonomic levels, multiple t-tests followed by FDR test (determined using the two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli) were performed using GraphPad Prism version 8, GraphPad Software, La Jolla, California, United States.³ To determine significance between responders and non-responders for diversity metrics, Mann-Whitney tests were performed also using GraphPad Prism version 8. To determine significance of non-metric multidimensional scaling based on Bray-Curtis distances, PERMANOVA test was performed using the vegan v2.5-7 package (Oksanen et al., 2017) in R version 3.5.1 (R-Core-Team, 2018). The vegan package was also used to determine the goodness of fit statistic (squared correlation coefficient r²) of unconstrained ordination through the application of the envfit function. PCA plots were generated using Euclidean distances using basic statistic functions in R. Random forest analysis were performed using the randomForest v4.6-14 package in R. Abundance values were transformed to Z-scores prior to analyses to normalize data. To decrease the error rate of the forest, the best mtry (random variables used in each tree) was calculated and used to fine-tune the forest. ROC plots and AUC were determined using the RCOR v1.0 package in R (Sing et al., 2005).

Three independent replicates of a vaccination and challenge experiment (Experiments A, B, and C) were conducted to verify the efficacy of the avian pathogenic E. coli (APEC) O78-based vaccine strain that expresses the C. jejuni N-glycan on its surface (Figure 1 and Supplementary Figure S1). Each replicate consisted of a vaccine group (vaccinated and challenged), a positive control group (not vaccinated, challenged), and a negative control group (not vaccinated, not challenged). In all three experiments, the C. jejuni colonization levels in cecal samples, taken on day 35, showed a similar bimodal distribution that has been observed in previous experiments (Nothaft et al., 2016, 2017). A statistically significant proportion (determined by Fisher’s exact test, p<0.001) of birds (averaging 65%) were found to be fully protected after vaccination (= responders), while an average of 35% of birds was still colonized after C. jejuni challenge (= non-responders) as determined by colony counts obtained after serial diluting the cecal contents of each bird and culturing on selective plates for 48 h (Figure 1) and subsequently confirmed by 16S sequencing (see below). In particular, in Experiment A, 26 out of 44 (= 60%); in Experiment B, 26 out of 46 (= 57%), and in Experiment C, 36 out of 46 (= 78%) had non-detectable levels of C. jejuni, by both methods, in their ceca. It is worth mentioning that minor deviations from the original group sizes were due to unrelated mortalities or morbidities (e.g., birds with splayed leg/unable to move were euthanized according to the approved animal protocol). In contrast, all birds (100%) in the positive control group were colonized with Campylobacter (statistical mean of 1.1×10¹⁰CFU/gram cecal content) and no Campylobacter was detected in birds in the negative control groups. The live vaccine strain persisted in at least 60% of the birds up to day 10 (3days after the first vaccine dose) and in 12–40% 4days after of the second vaccine dose (day 25; Supplementary Figure S2). In experiments A and B, we could still detect the live vaccine strain in nine (out of 44) and four (out of 45) of the vaccinated birds at the day of euthanasia (day 35) and in experiment C, the vaccine strain could not be detected on day 35. However, no correlation between vaccine persistence and the responder/non-responder effect could be established (Fisher’s exact test, p=0.8).

First, we determined the N-glycan-specific immune responses using our established ELISA format (Nothaft et al., 2016). In all three vaccination experiments, N-glycan-specific IgY responses were observed in sera prepared from blood taken 1day prior to challenge (day 27) as well as in the sera prepared from blood taken on day 35 (Figure 2A). In addition, we were able to detect N-glycan-specific IgY and IgA in cecal content supernatants individually pooled from vaccinated and challenged birds (Figure 2B). Whereas cecal IgY levels were significantly increased compared to the control groups, the observed increase in N-glycan-specific IgA levels in the vaccinated and challenged groups was not statistically significant when compared to the control groups. However, and as seen in our previous studies (Nothaft et al., 2016, 2017), no correlation between the antibody titers and the level of protection could be established and no statistically significant difference was observed in the IgY level between responders and non-responders.

Pooled sera from all responder and non-responder birds identified in Experiments A, B, and C were tested for their ability to induce antigen-specific killing of C. jejuni 81-176 (Figure 3A). In comparison to bacteria incubated with pooled serum from the negative control groups (naïve chickens), an average of 21% bacterial killing was observed with pooled sera from non-responder birds, whereas an average of 54% bacterial killing was observed with pooled sera from responder birds. This result indicates that chicken antibodies targeting the C. jejuni N-glycan are indeed opsonizing, and further demonstrated that sera from responder birds display a significantly higher Campylobacter-killing activity when compared to sera from non-responder birds.

Different glycan profiles might result in different antibody avidities, i.e., in a stronger binding to the antigen and even though antibody levels could not be correlated with protection between responder and non-responders, variations in the glycosylation profile could increase antibody potency and could explain the observed differences in bacterial killing in the opsonophagocytosis assay. Therefore, we analyzed the total chicken serum and the IgY N-glycomes to evaluate potential differences in glycosylation between responders and non-responders by two orthogonal glycomics technologies (PGC nano LC-ESI MS/MS and MALDI-TOF MS).

To investigate if the responder/non-responder effect is manifested through specific properties of the host itself, a GWAS of the response to vaccination was conducted. To identify single-nucleotide polymorphisms (SNPs) and genomic regions potentially associated with vaccine phenotypes, the GWAS was done using SNP data (n=475,058) from all vaccinated birds (n=134; 88 responders and 46 non-responders) and was analyzed following the SNP set approach described by Wu et al. (2010). Population stratification depicted by MDS plots indicate that for the top three dimensions, the 134 chickens can be divided into two sub-populations (Supplementary Figure S9A). However, no obvious association between the phenotype (responders or non-responders) and the population structure could be found (chi-square test; p=1). The results of PCA also showed that the top principal components accounted for only 2.5% of the total variation, suggesting that the population structure did not need to be integrated into the association analysis. Association analysis after FDR adjustment for multiple testing also showed no significant (FDR<0.1) association between SNPs and the responder versus non-responder phenotypes (Supplementary Figure S10). Further analysis revealed that the population structure resulted from the differences between the single experiments, as the smaller sub-population was enriched with chickens from Experiment A, while the larger sub-population was enriched with chickens from Experiments B and C (Supplementary Figure S9B). This could be due to a change in the flock generation and/or be due to a change in supplier of the birds. However, the exact reason for the observed differences could not be determined and GWAS data strongly suggest that the differences in vaccine responses are not due to genetic differences or variations in the chicken host.

In addition to the genotype, we also investigated if the gender of the birds might have an influence on colonization and vaccine responses. Analysis of 26,502 SNPs on the Z chromosome and 18 SNPs on the W chromosome from each bird did not reveal any evidence for an association between sex and the vaccine response phenotype [chi-square test (p=0.74)] (Supplementary Figure S11).

To analyze a potential role of the gut microbiota in the variable responses toward vaccination, we compared the cecal bacterial community between responder and non-responder birds. The composition of the cecal bacterial community was determined from each bird from the three vaccine experiments (total n=134; 88 responders and 46 non-responders) by Illumina sequencing of the V6-V8 region of the 16S rRNA gene. Considering that the composition of microbial communities could be affected by environmental, stochastic, and host -related factors, random forest analysis was performed using transformed OTU abundances to determine the out of bag (OOB) estimate of error rate for the main environmental and host variables that could potentially impact the microbial communities. These include experimental group (environmental factor), sex, and response (host factors). The OOB predicted error was estimated to be 1.5% if the analysis of the microbial data was done by experiment, but it increased to 13.53% if all samples were analyzed together by response (Supplementary Figure S12A). These results were supported by the goodness of fit test to determine significance of variables in unconstrained ordination (envfit function of the vegan package), which identified the “experiment” as the only significant variable impacting communities (r²=0.4702 implying moderate correlation, p<0.001; Supplementary Figure S12B). Clustering of all samples by PCA of Bray-Curtis dissimilarities also revealed clustering of samples by experiment (Supplementary Figure S12C) rather than by response (Supplementary Figure S12D). Overall, this analysis indicated that the samples from the three experiments should not be analyzed together as confounded by the experiment. The microbiota of birds from experiment A was vastly different to that of experiments B and C. Envfit analysis of abundance at the level of genera (Supplementary Figure S12B) showed that differences were especially dominant in the occurrence of Bacteroides spp., which were present at an average of 45.9% in experiment A, but with less than 0.02% on average in birds of experiments B and C.

PERMANOVA analysis on Bray-Curtis distances was performed for each individual experiment to determine differences in microbial community composition between responders and non-responders. Significant differences were only found for experiment A (Supplementary Figure S13A). Analysis of Bray-Curtis dissimilarities between individual birds as a measurement of β-diversity revealed a higher level of inter-individual variation between responder and non-responder birds for experiments A and C, but not for experiment B (Supplementary Figure S13B). Interestingly, within group inter-individual β-diversity is higher for responders in experiment A, but seems lower in experiment C, implying that these findings are not consistent and might be due to stochastic differences between experiments. Analysis of α-diversity showed statistically significant differences between responders and non-responders for birds in experiment C (p=0.0023). Noteworthy, there is a tendency for α-diversity to be lower in responders than in non-responders in all experiments (Supplementary Figure S13C).

Comparisons of cecal microbiota composition between responder and non-responder birds at different taxonomic levels showed significant differences. The clearest difference was detected for the family Campylobacteraceae, the genus Campylobacter, and an OTU assigned to Campylobacter jejuni, all of which were much lower in responders, confirming the effect of the vaccine in this subgroup of birds. Effects in other taxa were more subtle showing differences in genera within the order Enterobacterales, such as Escherichia/Shigella/Salmonella, in addition to differences observed in genera and species belonging, for example, to the Clostridium IV and XIVb groups within the order Clostridiales (Figure 4A). While Enterobacterales seem to be higher in responder than non-responder, the observed trend for Clostridiales is the opposite. Although we found significant differences among responder and non-responder, the findings were not consistent among all experiments, as it also has been observed in our previous work (Nothaft et al., 2017). However, there is some consistency in that Clostridiales or Clostridiales-related taxa (Ruminococcus and Lachnospiraceae) were again detected as being significantly different between responders and non-responders, also in agreement with findings from our previous studies (Nothaft et al., 2016, 2017).

In order to determine if the microbiota can predict the response to the vaccine, random forest analyses, using OTUs (Figure 4) and higher-level taxa (Supplementary Figure S14), were used to establish classifiers. Campylobacter was removed from the equation since the responder/non-responder groupings were determined by whether or not they were colonized with Campylobacter. Utility of models was examined using receiver operating characteristics (ROC) plots. The area under the curve serves to summarize the performance of the classifier into a single measure, with bigger values indicating improved accuracy of prediction. The area under the curves for experiments A, B, and C were estimated to be 0.639, 0.535, and 0.654 when using genera (Supplementary Figure S14A) and 0.689, 0.619, and 0.537 when using OTUs (Figure 4B). The random forest classifier for the combined data from all three experiments showed an area under the curve of 0.628 (using genera) and a slightly higher value of 0.664 when using OTUs. This finding indicated that there were major differences between experiments and that microbiota composition might not be a good predictor of the vaccine response. However, even though identified taxa had low predictive values, several OTUs that were significantly different between responders and non-responders (Figure 4A) were also identified to have a higher predictive value in the variable importance test (Figure 4C), and several fell within the same genera (e.g., Enterocloster and Enterococcus). In addition, genera within the Clostridium spp., Ruminococcaceae and Lachnospiraceae showed albeit low predictive potential (Supplementary Figure S14B), indicating that Clostridiales and Clostridiales-related taxa could be a variable in the response equation, supporting our previous findings (Nothaft et al., 2016, 2017).

To determine the causal role of differences in the cecal microbiota in vaccine response, we conducted smaller experiments that explored the effect of cecal microbiota transplantation (CMT) on vaccine performance (n=9 for each study, n=18 combined). First, we investigated if using a cocktail of antibiotics reduces the native microbiota to subsequently increase engraftment. In an independent study, we then applied the same procedure to birds that received the cecal transplant. Subsequent analyses (by 16S rRNA sequencing) of the microbiota of birds that received the antibiotic cocktail resulted in a significant reduction of both inter-individual β-diversity (PERMANOVA of Bray-Curtis dissimilarities) and α-diversity (Chao 1 index), respectively (p<0.001 for each of the three analyses; Figures 5A–C, left columns), indicating that the antibiotic treatment was effective in reducing microbial diversity in the birds.

We then compared efficacy of the APEC O:78-based N-glycan vaccine in birds that received the cecal microbiota isolated from responder birds versus birds that did not receive a CMT. In birds that received the CMT in combination with the vaccine, a lower percentage of Campylobacter colonization (after vaccination and challenge) was observed when compared to birds that received the vaccine alone (Figure 6A). In particular, we could demonstrate that the standard vaccination regimen on day 7 and day 21 with the vaccine alone resulted in 46.7% of birds still being colonized, whereas only 27.8% colonization was observed in birds that received the responder bird CMT in combination with the vaccine after antibiotic treatment (Figure 6A); however, this difference in colonization was not statistically significant (Fisher’s exact test, p=0.29). All birds in the positive control group (not vaccinated, but challenged) were found to be colonized and all except one bird in the group that received the CMT, but did not get vaccinated, was colonized with Campylobacter at similar levels. However, we observed statistically significant higher vaccine antigen-specific antibody titers (determined by two-tailed t-test) in the vaccine CMT group when compared to the vaccine alone group in serum prepared from blood taken before challenge (day 28, Figure 6B upper panel) and at the day of euthanasia, day 35 (Figure 6B, lower panel), suggesting that the microbiota of previously protected birds has a positive effect on the vaccine-induced immune response.

We further compared the microbial composition (by 16S sequencing) of the cecal contents at the trends suggesting that OTU classified as Clostridium spp., Ruminococcaceae and Lachnospiraceae are involved in the separation of both groups day of euthanasia (day 35) in responder and non-responder birds from the group that received the standard vaccination regimen (vaccine alone) as well as the group that received antibiotic treatment and CMT in combination with the vaccine. The results indicate that the alpha and beta diversities were different in responder versus non-responder birds (Figures 5A–C, middle and right columns, respectively) with (Figure 5D) consistent with what has been observed in the above-described experiments A, B, and C.

Our previous studies did not detect any changes in phase-variable gene expression in our challenge strain, C. jejuni 81-176 (Nothaft et al., 2016, 2017; Wanford et al., 2018) capable of persisting in non-responder birds; however, the limited analyses did not test for other SNPs in the genome of Cj-81-176 that could potentially result in the evolution of a C. jejuni super-colonizer or persister strain. We therefore collected C. jejuni isolates from non-responder birds and used those pooled isolates in a subsequent challenge study. Similar to Experiments A, B, and C, we observed a bimodal distribution in colonization after a second dose vaccination and challenge regimen. Challenge with the “standard” C. jejuni 81-176 (group three, n=8) or the “persister” C. jejuni pool (group four, n=8) did not result in statistical differences in protection. All birds in both positive control groups (not vaccinated, but challenged with either the standard or the persister pool of C. jejuni) were colonized at similar levels (Figure 7A). In addition, comparable levels of N-glycan-specific IgYs (determined by ELISA) were be found in the sera of vaccinated birds from groups three and four isolated on day 28 and day 35 (Figure 7B).

Similar immune responses to vaccination were induced in all our studies and the N-glycan-specific serum and cecal content supernatant IgY levels clearly distinguished vaccinated versus unvaccinated birds with increased antigen responses in the former compared to C. jejuni infection alone. In addition, in day 35 cecal samples, N-glycan-specific IgA could be detected at low levels, but there was no significant difference between challenged only and vaccinated birds which could be due to the shorter half-life of IgA versus IgY (Davis et al., 2020). Therefore, IgA may still contribute, but since samples were collected 7days after challenge (14days after the vaccine boost) our assay might not be fully representative for IgA levels present at the day of challenge. Intestinal IgY (in addition to IgA) has also been observed in other vaccination studies (Kulkarni et al., 2010; Annamalai et al., 2013; Wilde et al., 2019) and has been suggested to contribute to the reduction of the pathogen in the gut. However, the intestinal response to our oral E. coli-based vaccine would not explain the responder/non-responder effect (similar to the serum IgY response) but could in combination with other/unknown factors contribute to the observed reduction of Campylobacter in responder birds; similar to other studies that have shown that vaccine response does not necessarily correlate with protection (Chintoan-Uta et al., 2015). In general, B-cell responses to C. jejuni colonization in chickens show that IgY levels correlate with the presence of the pathogen in the intestine, but natural antibody levels in the cecum are not sufficient to clear the bacteria (Lacharme-Lora et al., 2017). However, it has also been shown that IgY antibodies are protective in the first 2weeks of life and that this effect can be prolonged by oral administration of pathogen-specific IgY (Vandeputte et al., 2019). We also demonstrated that antisera from responder birds showed a higher Campylobacter-killing activity, indicating that sera from those birds might possess a higher affinity for the antigen or the antibody pool is enriched with IgYs with desired effector functions. It has been shown that human antibody N-glycosylation (IgG glycan at N297) and the structure of the N-glycan are important for antibody functionality (Krapp et al., 2003; Jefferis, 2009). For example, changes in IgG glycosylation play a role in autoimmune disease and cancer in humans and impact viral and bacterial colonization (Chen et al., 2020; Irvine and Alter, 2020). But little is known about IgY glycosylation and its contribution to antibody functionality in chickens. Despite the differences in killing activities, we found that glycosylation (at least in serum) is very similar in responders and non-responders, although whole IgY preparations showed a trend, albeit insignificant, toward reduced fucosylation in responders. This indicates that vaccination does not change the global N-glycosylation profile. Although one might speculate that N-glycosylation profiles of antigen-specific IgYs could be different, anti-C. jejuni (N-glycan)-specific IgY constitutes only a minor subfraction of the total IgY and changes (if present) would not result in a significant difference in the global N-glycosylation profile. It has been shown that effector function of staphylococcal protein A antibodies in mice is indeed glycosylation-dependent. Here, galactosylation favors recruitment of the complement component C1q and is indispensable for killing of staphylococci (Chen et al., 2020). Similarly, afucosylated Abs protected against infection through enhanced Fcγ receptor engagement and phagocytosis (Chen et al., 2020). While a glycosylation-dependent impact on IgG function is undisputed in mammals, the functional role IgY glycosylation has in chickens is still unclear. However, from a producer/consumer point of view, the fact that vaccination does not induce any detectable impact on the qualitative and quantitative composition of the serum and total IgY N-glycome, our E. coli vaccine does not induce any health risk by generating potentially immunogenic glyco-epitopes that could be harmful for humans.

In general, the chicken gut microbiota is part of a complex ecosystem influenced by many factors, such as stocking density, stress, feed type, and other additives, that can either increase or decrease microbial diversity (Ijaz et al., 2018; Hankel et al., 2019; Kridtayopas et al., 2019). The composition of the microbiota has been reported to affect the ability of Campylobacter to reside in the chicken cecum (Han et al., 2017; Sakaridis et al., 2018; Ocejo et al., 2019). Increased diversity and increased environmental pressure on the microbial community favor the presence of Campylobacter (McKenna et al., 2020); however, feed additives did not change the alpha diversity of the cecal microbiota but decreased the C. jejuni cecal counts by 0.7 logs and were associated with an increase of Bifidobacterium species (Thibodeau et al., 2015). Studies in humans and mice indicate that the microbiome can also influence vaccine-induced immune responses (Jamieson, 2015; De Jong et al., 2020), but very limited information is available for other species. In our studies, OTU classified as Clostridium spp., Ruminococcaceae and Lachnospiraceae seem to be involved in the responder/non-responder effect after vaccination and challenge. Specifically, Clostridiales, but also other bacteria (Zhang et al., 2007), present in the natural microbiota contribute to the control of C. jejuni colonization, an effect postulated to occur by exclusion or immune-priming (Han et al., 2017). Interestingly, the Clostridiales effect is not restricted to the chicken model. Clostridiales and also higher relative abundances of Lachnospiraceae seem to have protective properties against Campylobacter infections in humans (Dicksved et al., 2014; Kampmann et al., 2016), and members of Lachnospiraceae were also enriched in our responder birds. In mice, the presence/enrichment of Clostridium cluster XI, but also of Enterococcus faecalis (Gram-positive lactic acid bacteria, Firmicute), Bifidobacterium, and Lactobacillus have been correlated with reduced Campylobacter infections (O’Loughlin et al., 2015; Sun et al., 2018), whereas antibiotic treatment (removal of those species) increases severity of colitis (Sun et al., 2018). On the other hand, numbers of enterobacteria were reported to potentially facilitate C. jejuni infection (Haag et al., 2012). Although the latter observation was made in infant mice, an effect of these bacteria in the chicken model cannot be ruled out since members of Enterobacterales were also affected in our studies.

Competitive exclusion of Campylobacter by potentially beneficial bacteria present in the cecal transplant obtained from responder birds can most likely be eliminated since the fecal transplant alone had no effect on Campylobacter colonization. Interestingly, it has been shown that the transfer of microbiota from a line of birds with a higher resistance to Campylobacter cannot exert the same protective effect in another line of birds that is naturally more susceptible to colonization. In fact, the transfer of the microbiota from the more resistant line further increased Campylobacter susceptibility of the less resistant line (Chintoan-Uta et al., 2020). However, the authors also reported that, among others, three specific OTUs from the Ruminococcaceae family differed between susceptible and more resistant lines (Chintoan-Uta et al., 2020). In line with this, we found members of the Ruminococcaceae family as a potential factor in distinguishing responder and non-responders. Another study showed that the introduction of the Clostridiales-dominated microbiota from eight-week-old birds into newly hatched chicks had a protective effect against Campylobacter colonization and altered the gut microbiota. However, no effect was observed when 7-day-old chicks received the transplant (Gilroy et al., 2018). This is in accordance with the “window of opportunity” for colonization that has been described in mice and suggested for human infants for the protective/immune stimulating effect of some bacteria (Torow and Hornef, 2017).

Moreover, Campylobacter infection was associated with decreases in the abundance of Clostridium cluster XIVa and Lactobacillaceae OTUs and also with an age-dependent shift in OTUs of members of the Lachnospiraceae and Ruminococcaceae families (Connerton et al., 2018). Based on the observation that transplanted bacteria only exist for a limited time (Chintoan-Uta et al., 2020), one might speculate that our live vaccine potentially creates a more favorable environment for the more beneficial bacteria to thrive in certain (responder) birds resulting in the complete elimination of Campylobacter or that Campylobacter changes the niche environment in the gut in its favor [a strategy described for other pathogens (Baumler and Sperandio, 2016)] and that the vaccine removes or reduces Campylobacter, which then redresses some of the shifts.

This study indicated that microbial composition and antibody function, but not host genetics or the serum glycome, may contribute toward the efficacy of the glycoconjugate vaccine in chickens. Although the gut microbiota has a low predictive value, this is likely due to the substantial variations between experiments that are not linked to the vaccine. However, certain OTUs that fell within the same genera (Enterocloster and Enterococcus) and within the Clostridium spp., Ruminococcaceae and Lachnospiraceae were still associated with responders and non-responders suggesting there may be functional redundancy among microbiotas as different taxa appear in different experiments. There are likely stochastic factors that influence the assembly of the microbial community resulting in inter-individual differences and subsequently in changes in the levels of certain microbial components that could function as adjuvants or impact immune responses.

Despite best efforts to hold environmental factors constant, a change in supplier (through the same distributer), time of the year during which the study, was conducted influencing temperature levels during transport, or even differences in animal handling during vaccination/challenge, as well as other minimal variations (e.g., changes in feed intake), were beyond the control of the study design. This could result in an overall different microbial composition not related to the vaccine, and that most likely generates noise. Such variations in the composition of the microbiota even between experimental repeats has also been observed in previous studies (Stanley et al., 2013) and could further influence vaccine efficacy and Campylobacter colonization potential and could make it difficult to increase vaccine efficiency. Our experiments exemplify the challenges experienced in many other areas of microbiome research. Microbiomes are clearly causal to many host phenotypes, but the identification of causal components is confounded by the high level of unexplained variation (Velazquez et al., 2019).

Future efforts could include whole microbiome metagenomics and metabolomics in combination with ex vivo immune assays to determine which fraction/compound/metabolite of the microbiota could impact immune cell functions in relation to vaccine activity. In addition, data could be collected at different time points of the study (especially prior to challenge) since having sequence data only at the end of the experiment limits correlations to the microbiota identified after the Campylobacter challenge and the effect of Campylobacter or the immune response in response to vaccination might alter the microbiome differently. However, our E. coli lipid A-core N-glycan expression system has been repeatedly shown to reduce C. jejuni CFU after vaccination and challenge, while other carriers do not provide this protective effect (Vohra et al., 2020). Therefore, optimization of the antigen carrier in combination with microbiome manipulations will potentially result in further improvement of vaccine efficacy and help to completely eliminate Campylobacter from the food chain.