Genomic analysis on broiler-associated Clostridium perfringens strains and caecal microbiome profiling reveals key factors linked to poultry Necrotic Enteritis

Background Clostridium perfringens is a key pathogen in poultry-associated necrotic enteritis (NE). To date there are limited Whole Genome Sequencing based studies describing broiler-associated C. perfringens in healthy and diseased birds. Moreover, changes in the caecal microbiome during NE is currently not well characterised. Thus, the aim of this present study was to investigate C. perfringens virulence factors linked to health and diseased chickens, including identifying caecal microbiota signatures associated with NE. Results We analysed 88 broiler chicken C. perfringens genomes (representing 66 publicly available genomes and 22 newly sequenced genomes) using different phylogenomics approaches and identified a potential hypervirulent and globally-distributed clone spanning 20-year time-frame (1993-2013). These isolates harbored a greater number of virulence genes (including toxin and collagen adhesin genes) when compared to other isolates. Further genomic analysis indicated exclusive and overabundant presence of important NE-linked toxin genes including netB and tpeL in NE-associated broiler isolates. Secondary virulence genes including pfoA, cpb2, and collagen adhesin genes cna, cnaA and cnaD were also enriched in the NE-linked C. perfringens genomes. Moreover, an environmental isolate obtained from farm animal feeds was found to encode netB, suggesting potential reservoirs of NetB-positive C. perfringens strains (toxinotype G). We also analysed caecal samples from a sub-set of 11 diseased and healthy broilers using 16S rRNA amplicon sequencing, which indicated a significant and positive correlation in genus Clostridium within the wider microbiota of those broilers diagnosed with NE, alongside reductions in beneficial microbiota members. Conclusions These data indicate a positive association of virulence genes including netB, pfoA, cpb2, tpeL and cna variants linked to NE-linked isolates. Potential global dissemination of specific hypervirulent lineage, coupled with distinctive microbiome profiles, highlights the need for further investigations, which will require a large worldwide sample collection from healthy and NE-associated birds.

identifying caecal microbiota signatures associated with NE. 23

Results 24
We analysed 88 broiler chicken C. perfringens genomes (representing 66 publicly available 25 genomes and 22 newly sequenced genomes) using different phylogenomics approaches and 26 identified a potential hypervirulent and globally-distributed clone spanning 20-year time-27 frame . These isolates harbored a greater number of virulence genes (including 28 toxin and collagen adhesin genes) when compared to other isolates. Further genomic analysis 29 indicated exclusive and overabundant presence of important NE-linked toxin genes including 30 netB and tpeL in NE-associated broiler isolates. Secondary virulence genes including pfoA, 31 cpb2, and collagen adhesin genes cna, cnaA and cnaD were also enriched in the NE-linked C. 32 perfringens genomes. Moreover, an environmental isolate obtained from farm animal feeds 33 was found to encode netB, suggesting potential reservoirs of NetB-positive C. perfringens 34 strains (toxinotype G). We also analysed caecal samples from a sub-set of 11 diseased and 35 healthy broilers using 16S rRNA amplicon sequencing, which indicated a significant and 36 positive correlation in genus Clostridium within the wider microbiota of those broilers 37 diagnosed with NE, alongside reductions in beneficial microbiota members. 38 Conclusions 39 8 healthy-broiler isolates (8.0 ± 1.6; P<0.0001). Notably, isolates in lineage Vf encode the most 157 virulence genes (12.0 ± 1.0 genes). 158 To explore potential enrichment and correlation of NE-related toxin genes including netB, 159 tpeL and other secondary toxin genes pfoA and cpb2, an association statistical analysis (Chi-160 square test) was performed (Additional file 2: Genome-wide association analysis highlighted additional factors that may correlate with 168 widespread nature of lineage Vf isolates. Aside from the associations of toxin genes tpeL 169 (sensitivity: 64.2%; specificity: 97.3%; 9/14 isolates) and netB (sensitivity: 85.7%; specificity: 170 72.9%; 12/14 isolates) as described above, collagen adhesin cnaA (sensitivity: 100%; 171 specificity: 85.1%; 14/14 isolates) and a pilin-associated gene (group_5443; sensitivity: 172 100%; specificity: 86.5%; all 14/14 isolates) were specifically associated with this lineage of 173 isolates (Additional file 1: Table S10). When we further compared the representative pilin-174 associated gene group_5443 using NCBI non-redundant (nr) nucleotide database via 175 BLASTn, this gene was detected in both reference chicken isolates EHE-NE18 and Del1 176 complete genomes at 100% identity, which was suggested as a hypothetical protein in the 177 annotated file (Additional file 1: Table S10). Other lineage Vf-associated genes including 178 ABC transporter-related genes (n=4; group_1636, group_3194, group_2785, group_3195; 179 sensitivity: 100%; specificity: 82.4%) and phage-associated genes including capsid protein 180 (group_1646), phage-regulatory protein (group_6371) and endolysin (group_4126) were also 181 identified. 182 Adhesin is an important virulence factor in broiler-linked NE [14,18], and in this study we 183 found that adhesin genes (at least one variant) were overall enriched (P<0.0001) in NE-linked 184 isolates (58/62; 93.5%) vs healthy isolates (9/20; 45%). Among all related adhesin variants, 185 cna, cnaA and cnaD genes were significantly enriched in NE-associated C. perfringens 186 isolates (P<0.05), linking these genes to NE-disease development (Additional file 2: Fig. S6). 187 Importantly, environmental isolates encoded comparable virulence gene profiles, suggesting 188 potential reservoir including soil and feeds. Indeed, environmental isolate C. perfringens 189 FR063 was found to encode the  Previous studies [20,23] have indicated that acquired AMR genes are not widespread, and a 191 total of 7 AMR genes were detected across 88 isolates (Fig. 3). Tetracycline resistance genes 192 tetA(P) and tetB(P) were encoded in the greatest number of genomes (44 and 32 isolates 193 respectively), with erythromycin-resistance genes ermB and ermQ encoded in 2 and 4 isolates 194 respectively. Macrolide-resistant efflux-pump gene mefA was detected in two sub-lineage Vf 195 isolates, while multidrug-resistant gene emeA was detected in one isolate SYD-NE41 [24]. 196 Notably, approximately half (47.7%) of healthy-broiler and NE-linked isolates (n=42) did not 197 carry any acquired AMR genes. 198 The presence of plasmid(s) was predicted in all genomes using a reference-based sequence-199 search approach. Overall, 43 out of 88 (~48.8%) isolates carried at least 1 plasmid (18  200 isolates carried 1 plasmid, 15 isolates harboured 2 plasmids, 4 isolates harboured 3 plasmids, 201 6 isolates carried 4 plasmids; detailed in Additional file 1: Table S11). Geographical 202 association analysis indicated that two specific plasmids were present in birds from Europe, 203 Australia and North America -plasmids pCP15_1 and pCP15_2 which were first identified in 204 isolate CP15 from an NE-linked chicken in USA (Additional file 2: Fig. S7). These two 205 reference plasmids did not carry any well-studied virulence-related genes, nevertheless, the 206 re-annotation of plasmid genes using genus-specific database indicated that this small 14-kb 207 plasmid pCP15_2 encoded a number of genes associated with sugar metabolism including 208 phosphotransferase system (PTS), and sugar transporters sub-units (n=5; Additional file 1: 209 Table S12). In terms of plasmid types, European isolates carried 13 different types of isolates, 210 Australian 5 types and USA 6 types. Australian isolates were not found to encode a 211 'continent-specific', plasmid type, while Europe had 9 unique plasmid types. Plasmids 212 pDel1_4 and pCW3 were the common plasmids detected in isolates from England, Finland 213 and Denmark (Additional file 2: Fig. S8). Importantly, both plasmids pDel1_4 (Additional 214 file 1: Table S13) and pCW3 (Additional file 1: Table S14) belonged to conjugative plasmid 215 pCW3 family, carrying AMR genes tetA(P) and tetB(P) and adhesin gene cnaC; sharing 216 highly similar genomic characteristics including plasmid size (47-49 kb) and CDS number 217 (50-55; Additional file 1: Notably, disease-specific profiles might be masked by the fact that a probiotic mix was given 227 to these broilers as a preventative measure against NE development. Therefore, raw reads 228 from these genera were removed and another PCA was performed to understand the impact 229 of other secondary or low abundance microbiota members. Again, health vs. disease-status 230 clustering was not observed, however secondary NE-associated profiles did appear to 231 positively correlate with genus Clostridium. Clustering at family level also did not indicate 232 specific health-status signatures. Diversity analyses (including Inverse Simpson index, 233 Shannon-weaver index and Fisher index) indicated there was no significant difference 234 (P>0.05; ANOVA) in genus diversity between groups (Additional file 2: Fig. S10). 235 Relative bacterial genus abundance in each caecal sample was constructed to visualise 236 microbiota profiles (Fig. 5). Thirty-seven genera were represented, with Bifidobacterium and 237 Lactobacillus most abundant, which likely reflected the probiotic supplementation in the 238 chicken feed. A number of secondary genera, which are usual intestinal microbiota members, 239 were detected in these samples (relative abundance <10% in each sample) including Blautia, 240 Coprococcus, Dorea and Oscillospira. Blautia was more abundant in health-associated caecal 241 microbiomes (mean abundance: 3.06 ± 2.84 %) compared to diseased-associated caecal 242 microbiomes NE (0.72 ± 0.5%) and SNE (0.14 ± 0.89%; Fig. 5B An additional paired-end BLASTn approach, to assign species level 16S rRNA sequences, 257 indicated several important genera were present (Additional file 2: Fig. S12) [26]. reuteri 258 (common broiler gut member, also widely used as probiotic supplement), Lactobacillus 259 salivarius (common swine gut microbiota member used as broiler probiotic that improves 260 production and general health) and Lactobacillus vaginalis (frequently found in broiler gut 261 and a persistent gut coloniser) were the main species within the Lactobacillus genus [27][28][29][30]. 262 Enterococcus genus primarily consisted of species Enterococcus faecium, a widely-used 263 probiotic reported to promote broiler growth and suppress C. jejuni and C. perfringens 264 infections, while stimulating the growth of Lactobacillus and Bifidobacterium [31,32]. 265 Importantly, Clostridium genus was mainly assigned to C. perfringens sequences, denoting 266 the potential NE-link of C. perfringens origins in NE-broilers particularly NE3, where C. 267 perfringens strains C036 and J36 were isolated from the same NE bird. 268 269

270
Clostridium, particularly C. perfringens, is consistently described as the primary infectious 271 agent to cause chicken NE. As C. perfringens thrives at ambient bird body temperature (i.e. 272 40-42ºC), with a doubling time <8 mins in vitro (the shortest generation time known for a 273 microorganism), this may link to its ability to rapidly overgrow and cause disease pathology 274 [5,33]. In this study, we profiled the genomes of C. perfringens isolates, including newly 275 sequenced strains, across a geographically diverse and varied health status sample collection. 276 Genome-wide analysis revealed positive associations of important toxin genes with broiler-277 NE, and we identified a globally-disseminated potentially hypervirulent lineage Vf, which 278 comprised isolates encoding important toxin genes netB and tpeL [13,34]. 279 In silico toxin profiling indicated that the NetB toxin, which has been identified as an 280 essential toxin in NE development, [13,34], was present in ~50% of the NE isolates, with 281 environmental samples also encoding this toxin, which may act as potential reservoirs, linked 282 to NE outbreaks [35]. The fact that netB gene was exclusively encoded in NE-linked broiler 283 isolates, when compared to healthy isolates, further supported the strong association of this 284 toxin and NE pathogenesis. 285 Other virulence factors have also been implicated in NE pathogenesis. Several studies have 286 indicated that collagen adhesin (encoded by cna) [14,18,36,37] may facilitate bacterial 287 colonisation within the chicken gut. Our analysis indicated this gene (including its variants 288 cnaA and cnaC) was overabundant in NE-associated isolates when compared to healthy-289 broiler isolates (P<0.05), which also suggests a positive association with NE outcomes [37]. 290 C. perfringens encodes a diverse array of toxins, and interestingly we also observed that 291 several other accessory toxins were enriched in NE isolates, indicating these may also play an 292 underrated role in broiler NE [38]; PFO, a pore-forming toxin which has been linked with 293 bovine haemorrhagic enteritis [39], and CPB2, or beta2-toxin, another pore-forming cytolytic 294 toxin associated with NE in piglets and enterocolitis in foals [40]. 295 This genomic study indicates a potentially prevalent hypervirulent lineage Vf (comprised 14 296 isogenic strains; pairwise mean SNPs: 65 in 1 810 core genes), with strains obtained from 297 Australia, Canada, Denmark and USA, spanning a period of 20 years . Previous 298 analysis with 9 isolates (out of 14) also indicated these (isogenic) strains grouped within the 299 same lineage [18]. Notably, lineage Vf isolates carried significantly more virulence genes 300 (toxins, including netB and tpeL and collagen adhesin) than isolates in other NE-linked 301 isolates, toxin genes, supports that this lineage may be hypervirulent. TpeL toxin is not 302 typically considered essential for pathogenesis due to its low carriage rate among NE-linked 303 C. perfringens isolates (in this study tpeL was exclusively detected in lineage Vf) [41]. 304 Nevertheless, in a broiler-NE infection model, infection with tpeL-positive (also netB-305 positive) strains induced disease symptoms more rapidly, and with a higher fatality rate, in 306 contrast to tpeL-negative strains encoding only netB, highlighting a role for TpeL in more 307 severe chicken-NE pathogenesis [42]. These data also indicate a potential global 308 dissemination of NE-associated virulent genotypes, which is in agreement with a previous 309 study that indicated clonal expansion of C. perfringens via multiple-locus variable-number 310 tandem repeat analysis (n=328) [43]. However, significantly larger sample sizes from various 311 geographical origins will be required for in-depth WGS population structure analysis, if we 312 are to understand the spread of C. perfringens in chicken farms worldwide, which will be 313 vital in the context of disease control. 314 Key C. perfringens virulence factors including toxin and AMR genes are known to be carried 315 on plasmids [44], including the poultry-NE-related toxin netB [13,34]. The universal tcp 316 conjugative system in majority of plasmids may facilitate horizontal gene transfer and 317 enhance the virulence of C. perfringens strains [45,46]. As almost half of the genomes 318 carried plasmids (~48.8%) this implies widespread plasmid transfer within broiler-associated 319 C. perfringens strains. However, as our analysis was carried out using reference-based 320 approaches, in some cases, fragmented short-read sequenced genome assemblies from public 321 databases this may not readily identify plasmid sequences. Indeed, within lineage Vf we did 322 not observe the expected high carriage rates of plasmids encoding netB [47]. The availability 323 of long-read sequencing (e.g. PacBio and Nanopore) will improve investigations into C. 324 perfringens, as plasmids can be sequenced and predicted more accurately despite encoding 325 numerous tandem repeats [48,49]. 326 In this study, we also analysed C. perfringens isolates obtained from healthy or asymptomatic 327 birds, with several isolates (LLY_N11 and T18) encoding comparable numbers of virulence 328 genes when compared to broiler-NE isolates (n>10). Importantly, healthy-broiler isolate 329 LLY_N11 (netB-negative strain, encoded pfoA and cpb2) has previously been shown to 330 successfully induced NE in an experimental model [4,50]. These data highlight the important 331 role other host factors that may play a role in prevention of overt disease e.g. the chicken gut 332 microbiota. Gut-associated microbial ecosystems are known to play a key colonisation 333 resistance role, preventing overgrowth of so called pathobionts, or infection by known enteric 334 pathogens (e.g. Salmonella). 335 In this small-scale broiler microbiome study, healthy broiler caecal microbiomes appeared to 336 have enhanced abundance of the genera Blautia. Members of the Blautia genus are known to 337 be butyrate producers, and reductions in this genus have previously been associated with a C. as key beneficial microbiota members, serving to enhance intestinal health of chickens by 342 strengthening the epithelial barrier, thus preventing pathogenic microbes successfully 343 colonising and initiating disease. 344 In NE caecal samples we observed appearance of the Clostridium genus, which was 345 significantly enriched, albeit at low reads in NE individuals. Further species-level assignment 346 analysis indicated that most Clostridium sequences mapped to C. perfringens, indicating that 347 even a small proportion (mean relative abundance: 0.44%) of C. perfringens could potentially 348 be lethal to broiler hosts. Therefore, microbiota profiling of Clostridium may be useful as a 349 potential biomarker for NE-onset, however larger studies would be required to verify these 350 findings. 351 Probiotics, including Bifidobacterium and Lactobacillus, and also Enterococcus, have been 352 frequently used in broiler farming primarily for growth-promotion and prevention of bacterial 353 infections [29,54,55]. These taxa of beneficial bacteria have been reflected in caecal 354 microbiome analysis, with predominant OTU proportions been assigned to Bifidobacterium 355 and Lactobacillus across all samples. A previous study identified specific antibacterial 356 peptides produced by Bifidobacterium longum that may correlate with the proposed 357 probiotic/pathogen-inhibitory effect against C. perfringens [56]. Nevertheless, our analysis 358 does not definitively verify the colonisation potential of these probiotic-associated genera in 359 broilers' intestines, or whether the high levels were more transient in nature, as it is common 360 practice in poultry farms to administrate these strains in large amounts within the feed. 361 Moreover, although there were no significant changes in the OTU proportions of these two 362 probiotics across three groups, several birds did present with SNE and NE suggesting these 363 strains may not be effective in reducing the disease burden associated with C. perfringens. 364 However, large scale-controlled supplementation trails would need to be completed to 365 provide robust evidence for health promotion using probiotics in poultry.

Sample collection and bacterial isolation 376
Birds (culled as part of routine farm surveillance) were collected from sites reporting both 377 healthy flocks and flocks that had been diagnosed with NE. Birds were necropsied and 378 putative disease identification performed, followed by caecum content collection. Isolation of 379 C. perfringens was carried out by isolating organs and submerging 0.1% peptone water 380 (Oxoid, UK) in a 1:10 ratio of organ to peptone. Samples were streaked onto egg yolk agar 381 supplemented with cycloserine (Oxoid, UK) and incubated overnight anaerobically at 37°C 382 [57]. Single black colonies were re-streaked on brain heart infusion agar (Oxoid, UK) and 383 incubated anaerobically at 37°C overnight. Several colonies were collected and subjected to 384 identification of the plc gene by PCR, followed by 16S rRNA full-length amplicon 385 sequencing as described previously for species verification [58,59]. 386

Bacterial isolates and DNA sequencing 387
We isolated 22 novel C. perfringens strains from broilers and environmental samples from 388 farms in Oxford, UK. Genomic DNA of these bacterial isolates was extracted using phenol-389 chloroform method as described previously [59]. Details of these isolates are given in 390 Additional file 1: Table S1. Sequencing was performed at the Wellcome Trust Sanger 391 Institute using Illumina HiSeq 2500 to generate 125 bp paired-end reads. Illumina reads are 392 available in the European Nucleotide Archive under project PRJEB32760. 393

Phylogenetic analysis, SNP detection, in silico virulence gene and plasmid detection 408
Annotated gff files were used as input for Roary v3.12.0 to construct pangenome with option 409 -e -n to generate a core gene alignment via MAFFT, -s do not split paralogs, -i to define a 410 gene at BLASTp 90% identity and -y to obtain gene inference [61] Sequencing reads were analysed using OTU clustering methods via QIIME v1.9.1 using 439 SILVA_132 as reference database to assign OTU by clustering at 97% similarity [69,70]. 440 Briefly, paired-end sequences were merged using PEAR, followed by quality filtering using 441 split_libraries_fastq.py, chimera identification using identify_chimeric_seqs.py and chimera 442 removal using filter_fasta.py [71]