Metagenomic sequencing and analysis

Sixty representative DNA extracts were selected to provide a cross section among all manure types and the small-scale and externally heated compost conditions for metagenomic sequencing. DNA extracts were sequenced by the Biocomplexity Institute of Virginia Tech, Blacksburg, VA, on an Illumina HiSeq 2500 in High Output mode with 2 × 100 paired-end reads, with 12 samples pooled per lane across 5 lanes. Paired-end reads were annotated in MetaStorm using default parameters, with the amino acid identity (80%) aimed at preventing false positive annotations [47] and the e-value cutoff (1e−10) utilized to ensure lower quality matches were filtered out prior to assessment [48]. ARGs were annotated against the Comprehensive Antibiotic Resistance Database (CARD v1.0.6) [49] and plasmid-associated genes against the ACLAME database [50]. Given that differing annotation parameters and databases could produce different trends [51], resistomes of all samples were also annotated with DeepARG, which incorporates several publicly available databases and uses a deep learning algorithm to maximize ARG detection [47]. Relative abundances of total ARGs predicted by DeepARG were found to be strongly and significantly correlated with those annotated using CARD via MetaStorm, as described above (Fig. S1; Spearman’s r = 0.8, p <0.01). ARG richness was determined by enumerating each unique ARG detected by CARD and normalizing to the total million of reads for the sample.

Contigs were assembled in MetaStorm using IDBA-UD [52]. “Resistome risk,” defined as the cumulative potential for ARGs to occur on MGEs and in human pathogens, was calculated from the assembled contigs and compared among the samples using MetaCompare(v2 )[53]. Resistome risk is intended as a relative comparison among a similar sample set and is calculated from assembled metagenomic data as the product of the number of contigs containing an ARG, the number of contigs containing an ARG and MGE, and the number of contigs containing an ARG, MGE, and pathogen. Resistome risk is determined by annotating to an integrated ARG databases (CAR D[54], ARD B[55], MEGARe s[56], SAR G[57], and DeepARG-D B[47]), an integrated MGE databases (NCBI search for “integron” and I-VI P[58]), and a human pathogen database (WHO priority pathogens for ARG s[59]) with an e-value < 1e−10 and amino acid identity > 60%. To assess the potential influence of the assembly method on the observed resistome risk trends, all small-scale samples were also assembled using MEGAHIT [60] (Fig. S2). No significant differences were observed in resulting resistome risk scores (Wilcoxon, p = 0.5). On average, 48% of metagenomic reads per sample were successfully assembled into contigs using IDBA-UD. Contigs that occurred more than 3 times across all samples were further examined. Relative abundances were calculated by normalizing gene counts to abundance of 16S rRNA genes, annotated against the Greengenes database [61], factoring in target gene and 16S rRNA gene length as proposed by Li et al. [62].