12The gut microbiome harbors a 'silent reservoir' of antibiotic resistance (AR) genes that is thought to 13 contribute to the emergence of multidrug-resistant pathogens through the process of horizontal gene 14 transfer (HGT). To counteract the spread of AR genes, it is paramount to know which organisms harbor 15 mobile AR genes and with which organisms they engage in HGT. Despite methods to characterize the 16 bulk presence 1 , abundance 2 and function 3 of AR genes in the gut, technological limitations of short-read 17sequencing have precluded linking bacterial taxa to specific mobile genetic elements (MGEs) and their 18 concomitant AR genes. Here, we apply and evaluate a high-throughput, culture-independent method for 19surveilling the bacterial carriage of MGEs, based on bacterial Hi-C protocols. We compare two healthy 20individuals with a cohort of seven neutropenic patients undergoing hematopoietic stem cell 21 transplantation, who receive multiple courses of antibiotics throughout their prolonged hospitalizations, 22and are thus acutely vulnerable to the threat of multidrug-resistant infections 4 . We find that the networks 23 of HGT are surprisingly distinct between individuals, yet AR and mobile genes are more dispersed across 24 taxa within the neutropenic patients than the healthy subjects. Our data further suggest that HGT is 25occurring throughout the course of treatment in the microbiomes of neutropenic patients and within the 26 guts of healthy individuals over a similar timeframe. Whereas most efforts to understand the spread of AR 27genes have focused on pathogenic species, our findings shed light on the role of the human gut 28 microbiome in this process. 29 preferentially recruited to contigs that are longer and more abundant, but to a lesser degree than expected, 81reducing potential bias in our dataset toward highly abundant organisms (Extended Data Figure 4). We 82 binned contigs using several tools (Maxbin 22 , MetaBat and Concoct), and applied a binning aggregation 83 strategy, DAS Tool 23 , to obtain a set of draft genomic assemblies. As misassembly can resemble HGT, we 84removed assemblies with greater than 10% contamination, as determined by CheckM, resulting in 85 taxonomically coherent assemblies (Extended Data Figure 5), albeit a greater number of unbinned contigs 86(24.6% of the total) (Extended Data Table 3). We then apply conservative criteria to link mobile and 87 mobile AR-containing contigs with the genomic draft assemblies, considering an MGE part of a genome 88 assembly only if it is directly linked to it by at least two uniquely-mapped Hi-C read-pairs. As MGEs are 89 known to recombine, this mitigates the potential for falsely linking contigs that merely share common 90 mobile genes. However, this also potentially reduces our ability for overall detection, especially for larger 91MGEs, since mobile contigs are often fragmented in metagenomic assemblies 24 . Nevertheless, we 92restricted our analysis to those AR-organism and MGE-organism linkages derived from high-confidence ...