Resistome identification

Gene prediction for the 17 metagenomes was performed using Prodigal v2.6.3 [37] with the meta option specified. Genes predicted by Prodigal were compared against the local noradab database by means of BLASTp with an E value threshold of 1 × 10^− 6. Results were filtered for hits with a minimum percentage identity of 75% and alignment length of at least 25 amino acids. For each predicted gene adhering to these parameters, only the hit with the highest score was annotated as an ARG. We used the Markov Cluster (MCL) algorithm to resolve ARG family redundancy [38]. All ARGs were compared against each other in an all-against-all BLASTp approach [39] with a cutoff E value of 1 × 10^− 10. The resulting E values were used in the MCL algorithm to cluster ARGs into families with an inflation parameter set to 1.1 and log-transformed E values with a limit set at 100. ARG families were annotated according to the corresponding ARDB or CARD database descriptions for all members of the cluster [35, 36].

The relative abundances of ARGs were calculated as the total number of ARGs per sample divided by the total number of Prodigal predicted genes. The relative ARG family abundance was inferred by calculating the number of different ARG families represented by the number of different ARGs in the sample. The local non-redundant ARG database contained 4485 unique AR protein sequences.