The experimental setting was previously described in detail by Valečková et al. (2020); a brief description follows. Two experiments were conducted concurrently at the Swedish Livestock Research Centre of the Swedish University of Agricultural Sciences, with approval from the Uppsala region’s animal ethics committee (approval number 5.8.18-16271/2017). The experiments were conducted using fast-growing broilers Ross 308 (R-308), used in conventional production in Sweden, and the Rowan Ranger broilers (RR), with a slower growth preferred in, e.g., organic broiler production. In Experiment 1, a total of 160 unsexed day-old R-308 broiler chickens were used for the 42-day (6-week) experiment, which is considered a normal period of growth for fast-growing strains in the EU. In Experiment 2 a total of 160 unsexed day-old RR broiler chickens (also referred to as “slow-growing broiler”) were used in the 63-d experiment (9 weeks) corresponding to the age at which this broiler type is generally slaughtered in organic production systems in Sweden. In each study, broilers were randomly distributed in groups of eight individuals in 20 raised pens with four dietary treatments and five pen replicates for each treatment, arranged in a randomized block design. In both studies, two random broilers per pen were chosen as focal birds, representatives of the entire pen population. Focals were later on used for the collection of fecal droppings for C. jejuni quantification by agar plate culture and at the end of the experiment for cecal content sampling for a quantitative real-time polymerase chain reaction (qPCR) assay to quantify C. jejuni loads and for microbiota analysis done by 16S rRNA sequencing.

The experiments were performed in parallel during the winter in an insulated stable equipped with the facilities for automatic control of temperature and light. Each pen had a floor covered by fresh wood shavings and was equipped with a metal feeder and a 3-liter bell drinker.

Detailed diet specification and forage preservation are stated in Valečková et al. (2020). In brief, fresh feed and water (including treatments) were provided directly after the broilers’ arrival and supplied daily. All experimental diets were based on organic compound feed (13 MJ/kg metabolizable energy and 230 g/kg DM crude protein) and the daily requirement of pellets in all treatment groups was estimated (based on production performance objectives) to ensure ad libitum provision. Broilers were assigned to four different treatment groups: silage, haylage, LP256, or control. Haylage treatment was included in the study as a forage similar to silage but without any inoculum. Silage and haylage experimental diets were composed as total mixed rations (TMR) containing 85% of pellets and 15% of respective forage (on a DM basis). Additionally, the LP256 and the control groups received the organic pelleted compound feed (no forage provided). The LP256 group had drinking water inoculated with L. plantarum 256 (10⁷ CFU/mL).

Second-cut grass, with a seeding composition of 70% timothy and 30% meadow fescue, was used for the production of forages. Silage was inoculated with L. plantarum strain 256 during baling, providing an inoculum concentration of 10⁸ CFU per gram fresh matter. Haylage bales were made without inoculum. After 3 months of storage, bales were separately opened, chopped and thereafter ground to 0.5–1 cm particles. Forage was afterward vacuum-packed (1 kg per bag) and bags were stored at a temperature below 0°C to maintain a similar feed quality throughout the experiments. Enumeration of epiphytic LAB on silage was performed in duplicates monthly (January, February, and March) during the trial period and the pH of silage juice was measured prior to the trials.

Bacterial strains used in this study include L. plantarum 256 and Campylobacter jejuni #65. The L. plantarum strain (also known as L. plantarum NC7, Cosby et al., 1989) was previously used in our in vitro experiments and proved among other LABs to elicit the best inhibitory effect against C. jejuni #65 (unpublished data). The L. plantarum strain 256 isolate was stored at – 80°C in Luria Bertani (LB) broth with 20% glycerol. It was propagated in De Man, Rogosa and Sharpe broth for 24 h at 37°C for silage preparation and as a prophylactic probiotic in the study. C. jejuni #65 (ST-104, in ST-21 CC; isolated from a broiler chicken in the UK 2006) was cultured in Brucella broth at 42°C under microaerobic (85% N2, 10% CO2, 5% O2) conditions for 24 h. After 24 h incubation, optical density at 405 nm and plate counts were used to determine the infection dose used in the C. jejuni challenge.

To investigate the effects of the L. plantarum treatments on C. jejuni colonization of the broilers’ ceca, all birds were orally challenged (also referred to as “infected”) at 22 d of age in Experiment 1 and 29 d of age (corresponding to the 4 weeks of age at which organic broilers in Sweden must have access to an outdoor environment) in Experiment 2 (Figure 1). On the day of the challenge, 0.5 L of water with 10⁶ CFU/mL of the C. jejuni strain #65 was provided in the bell drinker of each pen. The inoculated water was administered for 3 h and C. jejuni viability in the water was determined by colony counts on blood agar plates at the start and end of the challenge.

The colonization pattern of C. jejuni was monitored during 19 days and 32 days for R-308 and RR, respectively, by fecal culture and colony counts on modified Charcole Cefoperazone Deoxycholate (mCCDA) agar plates. For fresh fecal sampling, two focal birds from each pen were placed individually in plastic boxes. Sterile plastic loops were used for the collection of droppings from the box bottom. Fecal samples were collected from all pens 1 day before the challenge with C. jejuni, to verify that the broilers were Campylobacter negative before the challenge. One hundred mg of fresh fecal droppings from the focal birds in each pen were collected in 1 mL LB medium supplemented with 20% glycerol on 1, 3, 5, 7, 14, and 19 days post-infection (dpi) in Experiment 1 and 1, 3, 5, 7, 14, 21, 28, and 32 dpi in Experiment 2 (Figure 1). Tubes were directly transported on ice to the laboratory for analysis. Samples were vortexed and centrifuged (100 × g for 15 s) to pellet crude fecal matter. Next, 100 μL was withdrawn and serially diluted in 10-fold dilution series. Afterward, 100 μL was plated on mCCDA and incubated for 26 h at 42°C under microaerobic conditions (Campygen, Thermo Fisher). After incubation, colonies were counted on the plate corresponding to the dilution that gave approximately 100 CFU per plate. Raw plate counts data are provided in Supplementary Figure 1 (Experiment 1) and Supplementary Figure 2 (Experiment 2).

In both experiments, one random bird per cage was sacrificed 1 day before infection (−1 dpi) and 3 days after infection (3 dpi) by an intravenous injection of sodium pentobarbital through the wing vein; the birds age corresponding to mentioned dpi is stated in Figure 1. The cecal content was sampled with an aseptic procedure into 2.0 mL screw cap microtubes (Sarstedt AG & Co, Germany) and placed in liquid nitrogen (followed by storing at – 80°C until analyzed). At 42 days of age, all focal birds in Experiment 1 (20 dpi), and one random bird in each replicate in Experiment 2 (13 dpi) were sacrificed and sampled. Experiment 2 focal birds were sacrificed at 63 days of age (34 dpi) followed by cecal sampling as described above. Samples were analyzed by a qPCR assay to assess C. jejuni colonization and the microbial composition of cecal content was investigated by 16S rRNA amplicon sequencing.

For quantification of C. jejuni load in the broiler cecum using qPCR, a standard curve was developed as a reference for the proceeding analysis (Supplementary Table 1). Bacterial colonies of C. jejuni #65 from a 36-h incubated mCCDA culture were suspended in 300 μL of phosphate-buffered saline (PBS). The suspension was briefly vortexed and divided into 100 μL duplicates, and was diluted in a 10-fold series of PBS for CFU counting on mCCDA plates and incubated under microaerobic conditions for 24 h at 42°C.

Simultaneously, another 100 μL replicate was extracted for DNA and qPCR analysis. The sample was mixed with 200 mg of 0.1 mm zirconia/silica beads (Biospec products, Bartlesville, USA) and 900 μL of ASL lysis buffer (Qiagen, Germany), briefly vortexed, and incubated at 95°C for 5 min to lyse cells, followed by immediate placement on ice for 10 min. The sample was then bead-beaten on Precellys24 sample homogenizer (Bertin Technologies, Montigny-le-Bretonneux, France) at 8000 rpm for 2 × 60 s with 30 s pause to disrupt bacterial cell walls mechanically. Centrifugation of the sample at 2500 x g for 1 min followed, and 200 μL of the supernatant was withdrawn into the sample tube together with 20 μL of proteinase K for DNA extraction. Extraction was performed on an EZ1 Advanced XL instrument (Qiagen, Germany) according to the manufacturer’s instructions. The extract was diluted in a 10-fold series of nuclease-free water and used as a template in the real-time PCR for generating a standard curve.

A real-time PCR targeting the d65_1178 gene, specific to C. jejuni Strain #65 and its ST type ST-104 (ST-21 CC) was conducted using a primer pair adapted from Atterby et al. (2018). The PCR was performed on a CFX96 Optics Module C1000 Touch Thermal Cycler (Bio-Rad Laboratories, USA). The reaction mixture contained: 1 x SsoAdvanced Universal SYBR Green Supermix, 0.3 μL of forward and reverse primer each, and 1 μL of the template. Reactions were run in triplicates. The amplification parameters were as follows; 98°C for 3 min, 40 cycles of 98°C for 15 s and 63°C for 60 s and followed by a melt curve ranging from 65 to 95°C as a check for assay specificity. Generated qPCR data was analyzed on Bio-Rad CFX Manager 3.1 software (Bio-Rad Laboratories, USA) and Microsoft Excel. Raw qPCR data are provided in Supplementary Figure 3 (Experiment 1) and Supplementary Figure 4 (Experiment 2). The amplification efficiency of the PCR reaction was 76% with R² of 0.9996.

All the cecal samples followed the same pre-treatment, DNA extraction procedure as the C. jejuni #65 suspension, with minor pre-treatment modifications. Modifications included; 400 μL of ASL lysis buffer, added to the sample and vortexed briefly to homogenize. Then, 120 μL of the sample was used in downstream steps. Sample DNA extracts were analyzed by qPCR and sequencing. For quantification, DNA extracts were run on qPCR along with a standard (C. jejuni #65) DNA extract. The CT values obtained from sample runs were compared to that of the standard and transformed into CFU using the generated standard curve equation. The generated CFU was multiplied by five to compensate for a five times dilution of the sample performed during pre-treatment, a dilution not performed on the standard suspension. The ultimate quantification was expressed in CFU/ml by multiplication of 10.

One hundred forty cecal sample DNA extracts were sequenced using the Illumina Miseq PE 250 sequencing platform at Novogene Bioinformatics Technology Co., Ltd. (Beijing, China). The 16S rRNA gene V3–V4 regions were amplified using Illumina primer set 341F (CCTAYGGGRBGCASCAG) and 806R (GGACTACNNGGGTATCTAAT) with a barcode. All template DNAs were normalized to the same concentration. PCR reactions were performed with Phusion® High-Fidelity PCR Master Mix (New England Biolabs, USA). PCR products were separated by electrophoresis on 2% agarose gel, purified with a Qiagen Gel Extraction Kit (Qiagen, Germany) and pooled at equal concentrations. Sequencing libraries were generated using NEBNext Ultra DNA Library Prep Kit (Illumina, USA) following the manufacturer’s recommendations, and index codes were added. Library quality was assessed on the Qubit 2.0 Fluorometer (Thermo Fisher Scientific, USA) and the Bioanalyzer 2100 system (Agilent Technologies, USA).

The raw sequencing data were uploaded to the National Center for Biotechnology Information database (NCBI) with accession number PRJNA876811. The bioinformatics data processing was performed by Quantitative Insights into Microbial Ecology 2 – QIIME2 (version 2020.2.0) (Bolyen et al., 2019). The barcode and primer sequence of raw demultiplexed reads were trimmed off and further processed by DADA2 to denoise and dereplicate reads, merge pair-end reads and remove chimeras (Callahan et al., 2016). The truncation length of 221 bp was used for both forward and reverse reads. The phylogenetic tree was built using FastTree and MAFFT alignment (Katoh et al., 2002; Price et al., 2010). The SILVA SSU Ref NR 99132 dataset was first trimmed to the corresponding primer region and trained as a classify-sklearn taxonomy classifier (Pedregosa et al., 2012; Quast et al., 2013; Bokulich et al., 2018). Subsequently, the amplicon sequence variants (ASV) were assigned taxonomy using the resulting classifier. After trimming and quality filtering, the sequencing of 16S rRNA gene yielded a total of 4,539,867 sequences from 140 samples. The ASV table was rarefied according to the minimum reads per sample (i.e., 21,377 reads) (Weiss et al., 2017). The generalized UniFrac distance matrix (alpha = 0.5) and alpha rarefaction was generated using the QIIME2 diversity plugin (Chen et al., 2012; Bolyen et al., 2019).

The data generated from plate counts and qPCR from both experiments were organized in Microsoft Corporation (2018) and statistically analyzed in SAS version 9.4 (SAS Institute Inc, 2013). Plating data are graphically represented as scatter panel plots showing bacterial counts as log (CFU/ml) and qPCR data as a box plot; plots were generated with R and the package ggplot2 (Wickham, 2016; R Core Team, 2021). Data from Experiments 1 and 2 were treated by the same pattern.

Statistical analyses of fecal plate count data were performed with a mixed effect linear model (Proc Mixed procedure in SAS) due to the repeated measure structure of the data. The model included treatment, days post-infection (dpi), and their interactions as fixed factors and pen as a random factor. To account for the repeated structure when several observations were made on the same birds (focals) at different dpi we included an error term with an unstructured covariance matrix. Post-hoc tests were conducted to compare C. jejuni load (log CFU/ml) at individual dpi among treatment groups as well as to compare all dpi within each treatment. In order to compare C. jejuni loads through the whole challenge period between four treatment groups, plate count data were expressed as the mean of all observed dpi samples within one treatment (colonization mean). qPCR data were analyzed with the same mixed-effect linear model and in the same pattern as the plating data. However, the pen as a random factor and repeated structure were removed from the model since only one cecal sample per pen was analyzed. Residual plots were inspected to ensure that residual were approximately normally distributed with equal variances for all models. Results are considered significant if p < 0.05.

For cecal microbiota, diversity analyses were performed with the q2-diversity plugin. The rarefied ASV table was used to calculate the number of observed ASV. Kruskal–Wallis rank test with Benjamini & Hochberg (B-H) correction was used to observe statistical differences in a number of observed ASV between groups (i.e., dpi and treatment, Kruskal and Wallis, 1952; Benjamini and Hochberg, 1995). Principal coordinate analysis was used to visualize the difference in the microbial composition based on the generalized UniFrac distances. Permutational multivariate analysis of variance (PERMANOVA) test of generalized UniFrac distance matrix with (B-H) correction was conducted to evaluate the difference among groups (Anderson, 2001). To identify bacterial taxa that differed in abundance between groups, we performed an analysis of composition of microbiomes (ANCOM) (Mandal et al., 2015).

Monthly enumeration of LAB during the experiment period revealed that silage contained 8.0, 7.4, and 7.2 log CFU/g of LAB, respectively, while haylage levels were 5.0, 3.8, and 3.0 log CFU/g. Consequently, silage contained ≥3 × 10-log (CFU/g) higher LAB concentrations than haylage and a gradual decrease in LAB concentrations was observed in both matters. The silage pH measurement just prior to the experiment showed pH 4.4 while haylage displayed pH 6.2.

Culture results revealed Campylobacter jejuni negativity in all birds prior to the infection and successful C. jejuni colonization in both broiler types after the challenge. In both experiments, C. jejuni loads peaked within 3 days after the challenge and thereafter colonization intensity had decreasing tendency with time.

In R-308, there was an overall significant treatment effect (p = 0.023) observed. Specifically, a significantly lower C. jejuni colonization mean was observed in the silage (p = 0.010) and haylage (p = 0.013) groups in comparison to the control (Figure 2). At 1 dpi, colonization in the silage group was significantly lower (2.01 logs) in comparison to the control group (p = 0.039). However, at the end of the experiment (19 dpi), there was no significant difference in the colonization between any of the treatments. No significant effect of LP256 (directly provided via the drinking water) treatment on C. jejuni loads was observed. A comparison of C. jejuni colonization within each treatment at 1 and 19 dpi (start and end of colonization period) revealed no significant difference in bacterial load (CFU/ml). The same was true for colonization comparison between the 3 dpi (supposed peak of C. jejuni load) and 19 dpi.

As determined by qPCR (Figure 3), C. jejuni loads in ceca at 3 dpi (25 days of age) or 20 dpi (42 days of age) were not significantly affected by dietary treatments in R-308. A comparison of C. jejuni colonization within treatment revealed a decreasing pattern between 3 dpi and 20 dpi. However, the differences in C. jejuni CFU/ml were not statistically significant.

No significant effect of treatments on C. jejuni loads was observed in RR as determined by culture (Figure 4). However, colonization comparison between the start of the infection period (1 dpi) and end of the trial (32 dpi) within each treatment revealed significant changes with mean reductions in C. jejuni of 2.65 and 2.46 10-log (CFU/ml) for LP256 and haylage, respectively (p = 0.006 and p = 0.017, respectively). Colonization comparison between the supposed peak of bacterial load (3 dpi) and end of the trial (32 dpi) displayed significant C. jejuni reduction in all treatment groups; p = 0.002, p = 0.001, p = 0.001, and p = 0.001 for control, LP256, haylage and silage group, respectively.

At 3, 13, or 34 dpi, no significant treatment effects on the ceca C. jejuni loads were observed in RR as determined by qPRC (Figure 5). Comparing C. jejuni loads between the supposed peak of bacterial colonization (3 dpi; 34 days of age) and end of the trial (34 dpi; 63 days of age) revealed significant reductions in control, LP256, and haylage group (p = 0.001, p = 0.001, and p = 0.024, respectively). A comparison between cecal C. jejuni colonization at 13 dpi (42 days of age) and 34 dpi for each treatment displayed significant C. jejuni reductions in the control, LP256, and silage groups (p = 0.001, p = 0.032, and p = 0.014, respectively).

Characterization of the cecal microbiota composition before and after the C. jejuni challenge in Experiments 1 and 2 was performed by 16S rRNA amplicon sequencing. A total of 140 samples were analyzed and altogether 675 amplicon sequence variants (ASVs) were identified, representing 122 taxonomic genera, 52 families, 33 orders, 13 classes, and 5 phyla. The rarefaction curves of observed ASVs revealed sufficient sequencing depth to capture species richness at all time points tested in Experiment 1 (Supplementary Figure 5) and Experiment 2 (Supplementary Figure 6).

In detail, 619 ASVs, representing 109 taxonomic genera and 5 phyla were observed in Experiment 1. Results of the principal coordinate analysis (PCoA) are shown in Figure 6 to visualize the variation of cecal microbiota between different days post-infection. Significant differences in the gut microbiota composition between different dpi were observed (p = 0.001), while no clear effect of the feed treatments on cecal microbiota composition was found; therefore, the treatments were pooled for further analysis.

At the phylum level, Firmicutes and Bacteroidota comprised more than 97.5% of the bacteria’s relative abundance (RA) in all treatment groups at −1, 3, and 20 dpi, suggesting that they were the major components of the cecal microbiota (Figure 7). Since the different dietary treatments had no influence on the gut microbiota composition, treatments were pooled together and changes between different days post-infection were investigated. Changes in the RA at phylum level was observed after the Campylobacter challenge; a temporary decrease in the RA of phylum Firmicutes (mean RA at −1, 3, and 20 dpi was 84.2, 74.7, and 85.4% respectively) was substituted by a corresponding significant increase in Bacteroidota (mean RA at −1, 3, and 20 dpi was 14.5, 23.7, and 13.4%, respectively). The RA of phylum Proteobacteria decreased at 3 dpi and returned to a similar level as before the C. jejuni challenge at 20 dpi (mean RA at −1, 3, and 20 dpi was 1.3, 0.7, and 1.1% respectively). A significant increase in the RA of phyla Campilobacterota was observed at 3 dpi (mean RA of Campilobacterota at −1, 3, and 20 dpi was 0.01, 0.9, and 0.1% respectively). The RA of Actinobacteriota ranged from 0.02 to 0.04%.

At the genus level, the top 25 genera (Table 1) constituted 89% of the total sequencing read pool. Genus Campylobacter was added as the 26th genus due to the interest of this study. A description of major changes in the RA at the genus level between different dpi follows: genus Bacteroides was the most dominant in the cecal microbiota at −1 dpi and clearly most dominant after the C. jejuni challenge (3 dpi). Thereafter, a considerable decrease was observed to the advantage of other genera at 20 dpi (Table 1). Clostridia UCG-014 and uncultured Ruminococcaceae continuously increased in RA throughout the sampling points; where Clostridia UCG-014 became the most abundant genus at 20 dpi. A significant decrease in the RA of the twelfth most abundant genus Lactobacillus was observed after the C. jejuni challenge at 3 dpi. However, at 20 dpi the RA of this genus reached higher levels than before C. jejuni challenge. In the genera Clostridia vadinBB60 group and unclassified Lachnospiraceae, a decrease in RA appeared at 3 dpi and remained at a similar level at 20 dpi. The RA of the genera Faecalibacterium (the second most dominant bacterial genus in broilers’ ceca), Eisenbergiella, Subdoligranulum, and Escherichia-Shigella decreased after C. jejuni challenge (3 dpi) but eventually, an increase was observed in all four genera at 20 dpi. As expected, the RA of genus Campylobacter significantly increased after the challenge (3 dpi) but had diminished at 20 dpi. Although clear general trends in the mean RA of different genera could be observed over the experiment as described above, there were high individual variations in the birds’ cecal microbiota composition within the same treatment group at each infection time point (dpi).

In Experiment 2, sequencing results comprised 671 ASVs, representing 121 taxonomic genera and 5 phyla. The results of the principal coordinate analysis are presented in the PCoA plot (Figure 8), where significant differences in the microbiota composition were observed (p = 0.001), while no clear effect of the feed treatments was found.

As seen in Experiment 1, cecal microbiota composition at the phylum level was dominated by Firmicutes and Bacteroidota also in Experiment 2; representing together at least 92.2% of the bacteria in all treatment groups (Figure 9) at −1, 3, 13, and 34 dpi. Proportional changes in RA were observed after the C. jejuni challenge (dpi 3) by an increase of phylum Firmicutes (mean RA at −1, 3, 13, and 34 dpi was 70.6, 84.5, 80.1, and 81.8% respectively) with a concomitant decrease of the taxonomic group Bacteroidota (mean RA at −1, 3, 13, and 34 dpi was 26.9, 13.6, 18.5, and 17.0% respectively). This was in contrast to the reverse pattern observed in Experiment 1. The RA of phylum Proteobacteria decreased after the infection (3 dpi), remained on a similar level at 13 dpi, and eventually increased again at 34 dpi; mean RA at −1, 3, 13, and 34 dpi was 2.4, 1.6, 1.3, and 2.1%, respectively. A significant increase in RA of the phylum Campilobacterota was observed at 3 dpi; mean RA at −1, 3, 13, and 34 dpi was 0.01, 0.3, 0.1, and 0.01%, respectively. The RA of phylum Actinobacteriota increased continuously throughout the experiment, ranging from 0.03–0.09%.

The top 25 genera (Table 2) constituted 91% of the total sequencing read pool. Genus Campylobacter was added as the 26th genus due to the interest of the study. A description of the trends in relative abundance at genus level during the experiment period follows: Faecalibacterium was the second most abundant genus at −1 dpi, the most abundant after the C. jejuni challenge (3 dpi), and clearly the most abundant at 13 and 34 dpi. Bacteroides dominated the cecal microbiota at −1 dpi. Despite a considerable decrease after C. jejuni challenge, the genus was the second most dominant at 3, 13, and 34 dpi. The RA of the genera Clostridia UCG-014, Ruminococcus torques group, and uncultured Ruminococcaceae peaked at 3 dpi, and thereafter gradually decreased at 13 and 34 dpi. Genus Lactobacillus was the ninth most abundant bacteria present in the ceca and its RA increased throughout all sampling points. Relative abundance of genera unclassified Lachnospiraceae and Subdoligranulum increased at 3 dpi, decreased at 13 dpi, and was maintained at a similar level at 34 dpi. In the genus Clostridia vadinBB60 group, a continuous decrease in RA was seen throughout the sampling points. Escherichia-Shigella decreased after the challenge and remained on the same level at 13 dpi; thereafter an increase was observed at the end of the trial. The RA of Campylobacter significantly increased after the infection (3 dpi) and declined at 13 and 34 dpi. As observed in Experiment 1, a high individual variation in birds’ cecal microbiota composition was observed also in Experiment 2.