Metagenomic evidence for co-occurrence of antibiotic, biocide and metal resistance genes in pigs

Antibiotic-resistant pathogens constitute an escalating public health concern. Hence a better understanding of the underlying processes responsible for this expansion is urgently needed. Co-selection of heavy metals/biocides and antibiotic resistance genes (ARGs) has been suggested as one potential mechanism promoting the proliferation of antimicrobial resistance (AMR). This paper aims to elucidate this interplay and exploit differences in antibiotic usage to infer patterns of co-selection by the non-antibiotic factors metals and biocides in the context of pig farming. We examined 278 gut metagenomes from pigs with continuous antibiotic exposure, only at weaning and at no exposure. Metals as growth promoters and biocides as disinfectants are currently used with little restrictions in stock farming. The pigs under continuous antibiotic exposure displayed the highest co-occurrence of ARGs and other genetic elements while the pigs under limited use of antibiotics still showed abundant co-occurrences. Pathogens belonging to Enterobacteriaceae displayed increased co-occurrence phenomena, suggesting that this maintenance is not a random selection process from a mobilized pool but pertains to specific phylogenetic clades. These results suggest that metals and biocides displayed strong selective pressures on ARGs exerted by intensive farming, regardless of the current use of antibiotics. Highlights A comprehensive gut microbiome metagenomics analysis of 278 pigs Co-selection phenomena were investigated via co-occurrence patterns as a proxy Twenty-seven types of co-occurrences involving 131 resistance genes were detected Regardless of use of antibiotics, AMR can be maintained by co-occurrence with MRGs/BRGs Maintenance of AMR is not a random selection process but pertains to specific phylogenetic clades


Introduction
7 | Page mapped reads to the gene (t), gene length (t) is the length of gene (t), n is the number of all the predicted 146

ORFs. 147
Between-individual diversity (β-diversity) of gene coverage/abundance was evaluated by Bray-Curtis 151 dissimilarity matrices through R function "vegdist" in the package "vegan" (Oksanen et al., 2008). β-152 diversity matrices were ordinated by PCoA plot (R function "plot_ordination" in R package "phyloseq") 153 to compare the differences in β-diversity between groups when FDR correction was required (R function 155 "pairwise.adonis" in R) (Arbizu, 2017). Within-individual diversity (α-diversity) was measured by 156 observed richness of gene coverage/abundance. One-way ANOVA (R package "stats") was used to 157 compare the differences in α-diversity among groups. TukeyHSD test (R package "stats") was used to do 158 pairwise comparisons of α-diversity. The pairwise comparison of gene abundance between three groups 159 was done using the Wilcoxon rank sum test (R function "pairwise.wilcox.test"). Venn diagram was 160 plotted using R function "draw.triple.venn" in R package "VennDiagram" (Chen and Boutros, 2011). 161 Fisher's exact test was used to test the statistical significance of the number of contigs between groups 162 (R function "fisher.test" in R package "stats"). 163 164

Co-localization of ARGs and BRGs/MRGs/MGEs in assembled contigs 254
To further verify the co-occurrence of ARGs and BRGs/MRGs/MGEs on the contigs, we visualized 255 the gene organization on representative contigs (Fig. 3A). According to resistances carried in each 256 representative contig, there were six major co-occurrence combinations and 27 co-occurrence subtypes. 257 We selected the contig carrying the most complete gene distribution in each of 27 subgroups as the 258 representative contig. All the co-occurrences involved 131 resistance genes to 17 metals, 28 biocides and 259 25 antibiotics. Of these, Cu, Ni and Zn resistance genes, fluoroquinolone, penam and tetracycline 260 resistance genes, acid and QAC resistance genes were the most co-occurring genes. The information 261 about these genes has been summarized in Table S2.
both the number of contigs carrying co-occurrences and the type of co-occurrence (Fig. 3A). We 265 identified all 27 co-occurrence subtypes in the Chinese farm, 20 subtypes in French farms, and 9 subtypes 266 in Danish farms. Some co-occurrences affiliated with the same subtype were only partly present in the 267 presentative contig. Therefore, to further investigate the abundance of co-occurrences, we plotted a 268 network to show the frequency of co-occurrences between resistance genes in all the contigs (Fig. S1). 269 The detailed information for the network has been summarized in Table S3. As shown in Fig. S1, 270 ARGs in the pigs from Danish farms mostly co-occurred with BRGs, which mostly tended to be 271 functionally associated, such as multidrug efflux transporter genes acrAB and its regulator gene acrR. genes encoding the two-component regulatory system CpxAR co-occurred with Fe-Zn resistance gene 301 fieF, Se-H2O2 resistance gene sodA and Sb-As-glycerol uptake and resistance gene glpF; Cu resistance gene 302 copA, cueR and Fe-H2O2 resistance genes fetA, fetB were frequently detected to co-occur with genes 303 encoding the AcrR/A/B multidrug efflux pump operon; And acid resistance genes including gadE, hdeA, 304 hdeB, mdtE, mdtF, gadW, gadX, gadA were located in the same contig with Zn-Te resistance gene pitA and 305 As resistance gene arsA/B/R; Beta-lactam resistance gene PBP3 was found to co-occur with two Cu 306 resistance gene corD, cueO, Fe resistance gene ygjH (only in Chinese pigs), one BRG ostA and one 307 transposase gene; A Cd-H2O2-HCl resistance gene yhcN co-occurred with the efflux pump system AcrE/F 308 and its repressor AcrS; A broad spectrum MDR ABRG mdfA was detected in the same contig with Cr-309 nitrofuran resistance gene nfsA. IIIa/APH(6)-Id/aadA/aadA2) were also frequently detected to co-locate with MGEs in the same contigs. 320 The polymyxin resistance gene pmrF was found to be in the vicinity of genes encoding transposases 321 in the same contig. Colistin resistance-related genes mexB and oprM, and polymyxin B resistance gene 322 eptA were also detected to co-occur with MGEs in Chinese and Danish pigs. The detailed information 323 for the network has been summarized in Table S6. node label size, and edge weight were proportional to mapped frequency. All the bacteria taxa that had a 350 successful match have been shown in Table S7. Zmantar et al., 2011). Unfortunately, we cannot provide an effects size estimate using our study, since no 371 quantitative or qualitative data on the use of these compounds was collected. In addition to the 372 widespread use of QACs, the spread of these resistance genes via HGT among prokaryotes may be 373 another reason for their widespread presence since a significant correlation between QAC resistance 374 genes and MGEs was detected in this study. The spread of resistance determinants through HGT is a 375 favored way for microbes to adapt to complex environmental pressures (Ye et al., 2017). In this study, 376 Notably, we found some polymyxin resistance genes and their co-occurrences with other resistance genes. 406 As we know, polymyxins have reemerged as a final line of defense against Gram-negative 'superbugs' 407 (Sun et al., 2018). The co-occurrences of polymyxin resistance genes and other resistance genes would 408 aggravate their spread and maintenance. In this study, although we found integrons carrying ARG 409 cassette in pigs only from the Chinese farm, the plasmids carrying these integrons have been found 410 around the world (Fig. S2-b/d) and mostly in pathogens. The co-occurrence between QAC resistance 411 genes qacF/qacEΔ1 and ARG cassettes in the integrons could help AMR maintenance in pathogens.
can cause a variety of communicable bacterial dysenteries in their hosts (Jennison and Verma, 2004). The 418 era of plentiful antibiotics is forcing more bacteria to develop ARGs to survive and grow in the newly 419 established toxic environment, while non-degradable metals and easily accessible biocides probably 420 further promote the maintenance and spread of AMR in the form of co-occurrence investigated in this 421 study, which represents an increased public health risk. 422 While this study can distinguish patterns of co-selection that align with countries and farms that are 423 known to have large differences in antibiotics usage, it is a limitation that there are no observations on 424 many other aspects of farming practices. Cleaning agents and frequency would be particularly pertinent 425 given the observed co-selection of biocide resistance genes, but other differences in cultural and 426 regulatory practices could impact both the microbiome and the resistome this can harbor. Ideally, 427 observations of fully factorial designs could be used to separate the variance of these, but ultimately 428 intervention designs are needed to establish practices that can minimize the horizontal spread of ARGs. 429 430

431
In this study, we have used publicly available data to map the mobilizable resistance genes in pig gut 432 microbiomes and exploited the patterns of co-selection by non-antibiotic factors. The genetic evidence 433 presented here clearly suggests these factors contribute to the maintenance of AMR in pig farming. We 434 demonstrate the commonness of this co-selection with genetic evidence and augment this with an 435 overview of mobilizing elements. We clarify that this maintenance is not a random selection from a 436 mobilized pool but pertains to specific phylogenetic clades. We hope this work will give further insights