Context-Seq: CRISPR-Cas9 Targeted Nanopore Sequencing for Transmission Dynamics of Antimicrobial Resistance

Antimicrobial resistance (AMR) aligns with a One Health framework in that resistant bacteria and antibiotic resistance genes (ARGs) can be transmitted between humans, animals, and the environment. However, there is a critical need to more precisely understand how and to what extent AMR is exchanged between animals and humans. Metagenomic sequencing has low detection for rare targets such as ARGs, while whole genome sequencing of isolates is burdensome and misses exchange between uncultured bacterial species. We developed a novel, targeted sequencing assay using CRISPR-Cas9 to selectively sequence ARGs and their genomic context with long-read sequencing. Using this method, termed Context-Seq, we investigated overlapping AMR elements containing the ARGs blaCTX-M and blaTEM between adults, children, poultry, and dogs in animal-owning households in Nairobi, Kenya. We identified 22 genetically distinct clusters (> 80%ID over ≥ 3000 bp) containing blaTEM and one cluster containing blaCTX-M that were shared within and between households. Half of the clusters were shared between humans and animals, while the other half were shared only between animals (poultry-poultry, dog-dog, and dog-poultry). We identified potentially pathogenic hosts of ARGs including Escherichia coli, Klebsiella pneumonia, and Haemophilus influenzae across sample types. Context-Seq complements conventional methods to obtain an additional view of bacterial and mammalian hosts in the proliferation of AMR.


Introduction
Antimicrobial resistance (AMR) is a global challenge that threatens to undermine modern medicine.It is estimated that in 2019, nearly 5 million deaths were associated with bacterial antimicrobial resistance. 1 The global burden of AMR disproportionately falls on low-income countries, where high rates of illness, unregulated antibiotic usage, and limited access to sanitation infrastructure contribute to the selection and spread of AMR bacteria. 2,3Current approaches to control antibiotic resistance rely on antibiotic stewardship; however, this approach is difficult in low-income countries where unregulated antibiotic usage in humans and animals is common and the burden of infectious diseases is high.
Between 2000 and 2015, antibiotic drug consumption rates increased by 77% in low-and middle-income countries, compared to a decrease of 4% in high income countries.

4
AMR can be shared between humans, animals, and the environment.For this reason, AMR fits within the One Health framework, which integrates human, animal, and environmental health to tackle complex public health problems.AMR can spread through the dissemination of whole bacteria carrying resistance, as well as through horizontal gene transfer of mobile elements, including between benign and pathogenic bacteria.6][7][8] From a molecular perspective, determining the genomic context of ARGs is critical for studying the proliferation of AMR as it can allow for the identification of mobile elements, co-occurring genes, and host bacteria. 9 These additional pieces of information may yield insights on the mechanisms of exchange between reservoirs and the role of zoonotic transmission.In order to curb the spread of AMR, we need to be able to identify the most important transmission pathways (e.g., poultry, dogs, water, soil) in a given setting.
1][12] However, culturing only captures a small fraction of organisms, with a bias towards bacteria more fit for selective conditions.Culture-independent methods, such as metagenomic sequencing, require high sequencing depth to capture low-abundance targets such as ARGs. 13In addition, untargeted metagenomic sequencing can waste millions of reads per sample with very low coverage of medically important ARGs.This abundance of data can be costly to store and require significant computing resources to analyze.
. CC-BY 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made The copyright holder for this preprint this version posted September 12, 2024.; https://doi.org/10.1101/2024.09.12.612745 doi: bioRxiv preprint Targeted sequencing approaches are promising for studying AMR by enrichment of genomic regions of interest.Illumina probe capture is inherently limited by the read length (200-500 bps), which makes investigating ARGs in their genomic context infeasible.Recently, Cas9-based enrichment has been applied to short and long read sequencing for clinical applications. 14,15In brief, extracted DNA is dephosphorylated followed by Cas9 cutting facilitated by guideRNAs.Sequencing adapters are selectively ligated to only the d(A) tailed ends that result from Cas9 cutting.For example, Cas9-guided adapter ligation was used to perform multiplexed detection of ARGs in human blood spots with Illumina short-read sequencing 14 and to investigate human alleles in breast tissue with Oxford Nanopore long reads. 15The selectivity introduced through guideRNAs coupled with long-read sequencing that can capture long DNA fragments make this a promising approach to investigate ARGs within their genomic context.
In this study, we developed and optimized a Cas9 targeted sequencing assay to selectively sequence ARGs and their genomic context, hereby referred to as Context-Seq.We demonstrate the utility of Context-Seq by applying the method to detect the ARGs bla CTX-M and bla TEM in human (adult and child), poultry, and canine fecal samples collected from households in Nairobi, Kenya to investigate overlapping antimicrobial resistance elements.

Results
Samples and ARG Target Selection.In order to select relevant ARG targets for Context-Seq, we first analyzed available human and animal fecal samples collected from seven households in Nairobi, Kenya 16 using a Taqman Array Card to detect 14 ARGs and eight pathogen targets (Table S1 and Figure S1).All samples were positive for tetA, sul1, and bla TEM .Bla NDM and mcr-1 were detected in canine and poultry samples only (mcr-1: 75% canine, 7% poultry; bla NDM : 50% canine, 13% poultry).Bla CTX-M Group 1 , bla CTX-M Group 9 , bla OXA-10 , and bla SHV were detected in at least 50% of samples of each type.We selected two clinically relevant targets in high abundance (bla TEM and bla CTX-M ) and samples from four households that were positive for bla TEM and bla CTX-M group 1 (Table S2).
While many metagenomic approaches capture hundreds to thousands of resistance genes, not all resistance genes are clinically important. 17Extended-spectrum beta-lactamases (ESBLs) genes are of high medical importance as they can confer resistance to most beta-lactams including cephalosporins. 18CTX-M is a globally distributed gene group where all alleles are considered ESBLs. 19Common genotypes include bla CTX-M-15 and bla CTX-M-14 . 19Bla CTX-M alleles have been found in humans, 20 animals, 21 wastewater, 22 and other environmental reservoirs. 23They are also present in clinical isolates of Escherichia coli, Klebsiella pneumoniae, Salmonella species, Pseudomonas aeruginosa, and other bacterial taxa. 246][27][28] However, bla TEM alleles differ in phenotypic resistance conferred, ranging from penicillin resistance (e.g., bla TEM-1 ) to ESBLs (e.g., bla TEM-10 ). 29Design Tool.To design Cas9 guide RNAs, we utilized existing software (CHOPCHOP) 30 and developed a custom script to estimate off-target activity of guides in complex microbial communities, which is publicly available (https://github.com/Shruteek/Optimized-sgRNA-Design).Guides were selected based on high CHOPCHOP predicted efficiency, genomic location near the ends of the genes, and lower predicted off-target activity (Table S3).Notably, predicted efficiency as well as off-target activity are based on empirical data 31 and may not be well representative of real systems.When comparing guides for bla TEM , all pairs resulted in enrichment; there was, however, variation in enrichment based on the combined pair (Figure S2) and best performing guides were not necessarily predicted to have the highest on target activity and lowest off-target activity (Table S3).To facilitate capture of genomic context in both directions of a target ARG, different fractions of the sample DNA were cut by guides on the sense and antisense strands separately and then pooled.Alleles targeted by guides are indicated in the supporting information (Tables S4-S5).
Protocol Optimization.We optimized long-read Cas9 enrichment, originally validated for variant detection in cell culture and human tissue samples, 15 to detect ARGs in fecal and soil samples.We investigated these modifications on a mock community comprised of an Escherichia coli isolate with bla CTX-M-55 and bla TEM-1 genes spiked into a composite sample of extracted DNA from Kenyan soil.
Modifications evaluated included adaptive sequencing, multiple guides per target per strand, longer incubation time for Cas9 cleavage, and the addition of thermolabile Proteinase K. We also evaluated the impact of including two targets in the model system and in a human fecal sample.
Our final protocol enriched for two targets (bla CTX-M and bla TEM ) in two directions and added a thermolabile Proteinase K digestion to previously published methods (Figure 1A).Adaptive sequencing, longer Cas9 digestion, and additional guides per target did not improve the assay performance (Figure 1B) but utilizing thermolabile Proteinase K after the Cas9 digestion did (Figure 1C).Inclusion of guides for both bla CTX-M and bla TEM decreased enrichment for both targets in our mock system (Figure 1B) and in the human fecal sample (Figure 1D and 1E).Since the decrease in enrichment was minor for bla TEM (non-normalized coverage [mean (std)]: two targets [1265 (253)]; one target [1517 (337)]), the target in higher abundance, we processed samples enriching for both targets.All protocol modifications resulted in an enrichment of 7-15X coverage over untargeted methods (Figure 1B).
Using our final protocol, we sequenced 13 fecal samples across four households (five human, three canine, and five poultry) on individual MinION flow cells (Table S6).The percentage of reads that aligned to either bla CTX-M and/or bla TEM out of the total reads that passed quality filtering was 0.4% (range 0.02-2.01%).Of the reads that aligned to bla TEM across all samples, the average length was 4854 [std 1081] base pairs (bps) and of those that aligned to bla CTX-M the average length was 4381 [745] bps.

Context-Seq enabled identification of ARGs and annotation of their surrounding genomic context.
Annotated sequences resulted in the expected cut pattern based on guide design.Sequences begin with an ARG cut on either the sense or antisense strand and contain additional annotated genes across variable long read lengths (Figure 2).Sequences for bla CTX-M ranged from

ARGs were identified on plasmids and primarily in E. coli, K. pneumoniae, and Hemophilus
influenzae.Of the taxonomic identifications that were identified via Kraken2 and confirmed via BLASTn, bla TEM was identified on plasmids and chromosomes (consensus sequences not identified as plasmids with a plasX confidence <0.5) annotated as K. pneumoniae, H. parainfluenzae, H. influenzae, E. coli and Enterobacteriaceae (Figure 3A).Bla CTX-M was identified on plasmids and chromosomes in K.
pneumoniae and E. coli (Figure 3B).Across hosts and genes (bla TEM and bla CTX-M ), the majority of sequences were identified as E. coli followed by K. pneumoniae.Other gammaproteobacteria, H.
parainfluenzae and H. influenzae, were only found in human stool samples and not in animals.
We compared our method (Context-Seq in total DNA extracts) to a parallel study that cultured E.
coli without antibiotics and sequenced up to 5 pooled isolates from the same samples. 16 3C).
In one poultry and one canine sample, no contigs containing bla TEM or bla CTX-M were assembled from the cultured isolates, but both ARGs were identified in the canine and bla TEM was identified in the poultry using Context-Seq.ARGs were shared across hosts through plasmids and chromosomes.The bla CTX-M cluster was identified as K. pneumoniae, shared between three households, and shared between canines and poultry (Figure 3D and Figure 4B).The sequence containing the bla CTX-M cluster was identified as a likely plasmid in one of the four samples (Figure 3D).Bla TEM was shared across households and hosts on consensus sequences identified as E. coli, K. pneumonia, and H. influenzae (Figure 3D).Of the shared bla TEM clusters, approximately half were shared on the same element (plasmid to plasmid or chromosome to chromosome) and half were shared between elements (plasmids and chromosomes).The sole human to human shared cluster (cluster 3 between children in HH 1 and 3) was classified as a H. influenza plasmid in both samples.

Discussion
Here we developed a novel method, Context-Seq, using Cas9 targeted sequencing paired with long-read sequencing to enrich for fragments of DNA containing two clinically relevant ARGs, bla TEM and bla CTX-M .We identified sharing of ARGs and genomic context between humans and animals, as well as between poultry and canines in urban Kenyan households, emphasizing the importance of the One Health approach to combatting AMR.Suspected hosts of ARGs were not just limited to E. coli but also included K. pneumonia, H. influenzae, and H. parainfluenzae.E. coli remains one of the most commonly characterized antimicrobial-resistant organism, 32 but many others are important to investigate in the context of AMR transmission.
An important finding of this work is the occurrence of non-E.coli hosts, primarily K. pneumoniae and H. influenzae.K pneumoniae is a leading cause of antimicrobial resistant infections, 33,34 especially in hospitalized patients 35 and is on the WHO global priority pathogens list. 36While K. pneumoniae can survive in multiple environments including soil, water, and the intestinal tract of humans and animals, 35 few studies have investigated resistant K. pneumoniae from a molecular epidemiology, One Health perspective (i.e., strain sharing in humans, animals, and the environment).A previous study in Kenya collected K. pneumoniae from community fecal samples, healthcare-associated fecal samples, and hospital surfaces across multiple counties.Among extended spectrum beta-lactamase (ESBL) isolates, bla CTX-M-15 and bla TEM-181 were the most common genes. 37We identified bla CTX-M-15 like in K. pneumoniae in a shared cluster, while bla tem-181 like was identified in one sequence, but not in a shared cluster.Most bla TEM genes were highly similar to non-ESBL genes bla TEM-141 and bla TEM-135 (groups 1 and 2 above).Very few studies have been conducted on K. pneumoniae in household animals, although one study investigated K. pneumoniae isolated from raw meat samples in Nairobi. 38They found high resistance to ampicillin but very few isolates had ESBL resistance. 381 This work hypothesized the similar poultry resistomes were the result of similar antimicrobial selective pressure since use of antimicrobials for therapeutic or prophylactic purposes is consistent across Nairobi. 42We observed significant overlap between poultry and canines (7 out of 11 animal-animal clusters), however the UrbanZoo project did not investigate canines.A similar mechanism may exist for canines and poultry in that they may consume similar antibiotics, and their guts could select for similar ARGs. 43Another potential mechanism of AMR acquisition in canines is through scavenging.Previous work has demonstrated that the widespread waste (including human and animal feces as well as garbage) across the urban landscape of Nairobi can serve as a reservoir for AMR. 44Canines could acquire similar AMR as poultry through eating of poultry feces.
Human and animal overlap was observed in the other half of shared clusters (11 out of 23).
Previous evidence for human and animal resistome sharing has been mixed. 10,12,45,46In a related study, where E. coli was isolated from humans, animals, and the environment in the same households as our work, human and animal strain sharing was rare. 16The majority of E. coli strain similarity was observed between humans and stored drinking water, and poultry and soil, implicating the environment as a reservoir between hosts. 16The observed higher relative degree of sharing between humans and animals in our work is likely due to method specific differences.Here, we investigated ARG containing DNA fragments only and our metagenomic-based approach captured additional species outside of E. coli.We did not apply enrichment sequencing to soil and water, thus cannot compare findings related to the environment.Together, these paired efforts highlight that multiple approaches may be needed to obtain a more complete understanding of AMR in a given context.Finally, our results are consistent with the paired study and UrbanZoo 11 in that when human to animal overlap is observed, it may occur between households.
We observed ARG sharing both through plasmids and through chromosomes.We note that chromosome was classified as sequences that were not identified as plasmids, and not through whole genome or strain analysis.Diverse mobile elements carrying ARGs have been observed in humans and animals.Horizontal gene transfer can facilitate transfer of AMR through these reservoirs.Previous work on E. coli in Nairobi concluded organismal spread, rather than transduction or transformation, was the dominant mechanism of highly similar mobile elements between human and animal hosts. 44Similarly, the paired study of E. coli isolates in our same study households found strain sharing was more likely to contribute to resistome sharing than horizontal gene transfer. 16Notably both studies were conducted on a single species, and our work demonstrates shared ARGs and genomic context between species (E. coli and H. influenzae, K. pneumoniae and E. coli) which likely occurred through horizontal gene transfer.
This work has several limitations.As previously mentioned, we processed samples on individual flow cells which is expensive and infeasible for a large study.Multiplexing samples has the potential to significantly reduce costs, though may result in decreased coverage.In addition, Oxford Nanopore Technologies' long-read sequencing is a relatively new technology and is consistently changing to improve the nominally high error rate (≈ 90-95% accuracy).This project was conducted on previous generation flow cells (R9.4.1), which are available from ONT upon request.Additional validation would need to be conducted for R10.4.1 since the duplex chemistry is a significant shift from the previous versions.Similar methods could be applied to alternative long-read technology such as PacBio. 47Finally, we did not process environmental samples in this study; future work to process environmental samples (e.g.soil and water) with Context-Seq is recommended for a complete One Health approach to investigating AMR. 48ext-Seq is a promising approach for enriched long-read sequencing.While we demonstrate the utility of this assay with two ARGs, there is potential for ARG multiplexing, sample multiplexing, and further optimization of the enriched alleles.A previously published method (FLASH) used Cas9 with short-read sequencing to target detection of 127 genes with 5513 guides. 14Since many ARGs are colocated, 49 a guide pool of this size would likely be counterproductive to obtaining long reads but there is possible room for target expansion before compromising read length.One potential area of expansion is including guides for different alleles (e.g.CTX-M group 1 vs. CTX-M group 9) as they would likely be present on different DNA fragments.Further, the greatest potential for improving this method is sample multiplexing to reduce costs.While we ran each sample on a single flow cell, multiplexing on a MinION or promethION would significantly reduce the per sample cost.However, multiplexing is non-trivial and requires careful optimization to reduce off-target reads as it adds an additional step where non-target fragments can shear and introduce phosphorylated ends available for adapter ligation.Lower cost Context-Seq could be transformative to inform transmission dynamics of AMR through human, animal, and environmental reservoirs in diverse settings.

Sample Collection and Processing
Poultry-owning households from Dagoretti South and Kibera subcounties of Nairobi, Kenya were sampled in June-August 2019.Up to three poultry fecal samples and one canine fecal sample were collected during an initial visit.To collect animal feces, a sterile plastic scoop was used to transfer feces from the top, center layer of a fresh fecal pile.Approximately one week after the first visit, households were revisited to collect human stool from one household member in the following three age groups: child aged 0 -4 years, child aged 5 -14 years, and adult aged 15 years or older.A stool collection kit was provided during the first visit, which included a 50 mL plastic pot with a sterile scoop for each member with instructions on how to collect the sample.The primary caretaker of each household was informed by mobile phone one day prior to the revisit to collect stool from the previous night or the morning of the revisit day.All human and animal fecal samples were placed in a cooler filled with ice and transported to KEMRI. 1 g of fresh feces was aliquoted for storage at -80°C without preservatives.DNA was extracted from animal samples at Kenya Medical Research Institute (KEMRI) and stored at -80°C until transport.
Human fecal samples and DNA extracts from animals were shipped to Tufts University on dry ice.For all fecal samples, DNA was extracted from 0.2 g of feces using Qiagen's Powersoil Pro kit according to the manufacturer's instructions.
The study was approved by the KEMRI Scientific and Ethics Review Unit (Protocol number 3823) and the Tuft Health Sciences Institutional Review Board (13205).Additionally, a research permit was granted by the Kenyan National Commission for Science, Technology, and Innovation.

Taqman Array Card
Five adult fecal samples, 15 child fecal samples, four canine, and 15 poultry samples from eight households were prescreened for antibiotic resistance genes 50 using a Taqman Array Card prior to enrichment sequencing.22 targets were run in duplicate, 14 of which were ARGs.Samples positive for CTX-M group 1 and the TEM assay were candidates for enrichment sequencing (Table S2) (see supporting information for more details).

gRNA Design
Candidate guide RNAs (gRNAs) for ARGs were determined by CHOPCHOP. 30A custom program (https://github.com/Shruteek/Optimized-sgRNA-Design) was created to screen off-target effects in representative metagenomes.The program, implemented in Python, used empirical methods for onand off-target effect analysis by taking in a candidate sequence and a sample metagenome, and returning a heuristic representing the overall likelihood of the candidate sequence to experience off-target effects in the metagenome.Based on guide RNA binding behavior, 31 we counted off-target sites as valid only if they had 5 or fewer mismatches, 1 or fewer mismatches in the 10 PAM-proximal base-pairs, and a PAM of the form 5'-NGG-3' or 5'-NRG-3'.Bowtie 51 was used to identify potential off-target sites, then each off-target site was evaluated for its binding likelihood based on the number of PAM sequences in the forward and reverse sequence, 52 the 1-and 2-base-pair nucleotide features in and around the site, 53 the identity of each nucleotide, [53][54][55] the individual nucleotide mismatches between the site and the guide, 31 and the proximity of each mismatch to the PAM. 31,56,57The binding likelihood scores of the on-target sequence and each off-target site were normalized from 0 to 100, the latter corresponding to maximum binding odds for a perfectly stable matching sequence, and all off-target likelihood scores for a single guide were summed and subtracted from the on-target likelihood score to generate the heuristic for offtarget effects.

Library Preparation
We modified a previously published Cas9 enrichment protocol. 15To evaluate performance and test modifications, we made a model system comprised of an E. coli isolate with bla CTX-M-55 and bla TEM-1 genes spiked into a composited DNA extracted from Kenyan soil (see supporting information for additional details).Unless otherwise specified, the protocol was performed as described below.For the evaluated modifications we made the following adjustments 1.)For adaptive sequencing, the MinKNOW software was set up in adaptive mode using the bla CTX-M-55 gene as the reference and aligning up to 200 bps.Adaptive sequencing is a software-based method that allows the MinKNOW software to read the first few hundred base pairs of a fragment and selectively reject the fragment from the pore if it is classified as off-target. 582.) For longer Cas9 cut time, Cas9 digestion proceeded for 2 hours instead of 20 minutes.3.) For two guides per target per strand (sense and antisense), guides were added in an equimolar mix of 0.75 μL each to a 0.5 mL centrifuge tube and 1uL of the mix was complexed with Cas9.The protocol below describes the addition of Proteinase K as that was incorporated in our final procedure.
CrRNAs and tracrRNAs, together forming the gRNA, were resuspended to a final concentration of 100 μM in duplex buffer (IDT).8 μL of nuclease free water ,1 μL of tracR, and 1 μL of crRNA were mixed and heated at 95°C for 5 minutes for duplex formation.To create the ribonucleoprotein complex (RNP), 1X CutSmart Buffer (NEB), 2 μM of gRNA, 0.5 μM of HiFi Cas9 Nuclease V3 (IDT), and nuclease free water were combined to a total reaction volume of 30 μL.The reaction was incubated at room temperature for 20 minutes.Input DNA (approximately 1.5-3.0μg) was dephosphorylated in a 60 μL reaction composed of 6 μL 1X CutSmart buffer, DNA, nuclease free water, and 3 μL of QuickCIP (NEB).
The reaction was incubated at 27°C for 20 minutes followed by inactivation at 80°C for 2 minutes.For Cas9 cleavage and A-tailing, dephosphorylated DNA was split up into two reactions (2 reactions of 30 μL).Input DNA was split to allow for Cas9 cutting on the sense and antisense strands separately using two sets of guide RNAs for the two target ARGs.30 μL of DNA and 10 μL of RNP were mixed and incubated at 37°C for 15 minutes.1μL thermolabile Proteinase K (NEB) was added and incubate at 37 °C for 10 min.Proteinase K was then heat inactivated at 65 °C for 10 minutes. 1 μL dATP (Invitrogen) and 1μL Taq polymerase (NEB) were added to the mixture and incubated at 37 °C for 15 min followed by 72 °C for 5 minutes.To ligate on sequencing adapters, ligation mix was prepared by adding 9 μL nuclease free water, 40 μL ligation buffer, 20 μL T4 quick ligase (NEB), and 7 μL of adapters.38 μL of the adapter mix was added to each reaction and incubated at room temperature for 10 minutes on a tube rotator.
Equal volume of TE buffer was added to each reaction and then the two libraries (one for sense and another for antisense strand) were pooled.0.5X Ampure beads (Beckman Coulter) were added.The reaction was rotated for 5 minutes followed by incubation at room temperature on a bench top for 5 minutes.Magnetic beads were washed with 250 μL of long fragment buffer.After addition of 13 μL of elution buffer, beads were incubated at 37°C for 30 minutes.MinION flow cells were loaded according to the manufacturer's instructions with 12 μL DNA in elution buffer, 25.5 μL loading beads, and 37.5 μL sequencing buffer.

Figure 1 .
Figure 1.Cas9 is used to selectively sequence DNA fragments containing bla CTX-M and bla TEM .A) Schematic of Context-Seq workflow involving dephosphorylation, library splitting for Cas9 cutting on each strand, and adapter ligation.Phosphorylated ends are indicated with a red P. B) Comparison between library preparation modifications (adaptive sequencing, longer Cas9 digestion, and additional guides) against no enrichment in a mock system.Depth normalized coverage is calculated by dividing coverage by the total reads obtained per each sequencing run.C) Comparison between conventional Cas9 enrichment protocol and the inclusion of Proteinase K following Cas9 digestion in a mock system.D) Normalized coverage of bla TEM in a human fecal sample comparing enrichment of TEM alone (yellow) and both TEM and CTX-M (black).E) Normalized coverage of CTX-M in a human fecal sample comparing enrichment of CTX-M alone (green) and both TEM and CTX-M (black).

Figure 2 .
Figure 2. Annotated enriched sequences containing bla CTX-M and bla TEM generated by Context-Seq.A subset of sequences containing A) bla CTX-M, B) bla TEM , and C) bla TEM ≈ 20,000 bp annotated for ARGs and mobile genetic elements.Sample type is indicated by symbol (child, adult, poultry, canine) and household by number.Note these sequences are a subset and some consensuses sequences were obtained from the same sample.In addition, all ARGs were aligned in the same orientation representing how cutting can proceed from the sense or anti-sense strand.

Figure 3 .
Figure 3. Taxonomic identifications of Context-Seq sequences containing bla TEM and bla CTX-M .A) The number of sequences containing bla TEM identified as each taxa by Kraken2 and confirmed by BLASTn.Plasmids (plasX ≥ 0.5) are indicated by a circle.B) The number of sequences containing bla CTX- M identified as each taxa by Kraken2 and confirmed by BLASTn.Plasmids (plasX ≥ 0.5) are indicated by a circle.C) Comparison of Context-Seq to cultured and sequenced E. coli in the same sample.Coverage of the assembled contigs that resulted from 1.) the cultured E. coli using Illumina sequencing of the isolates and 2.) Context-Seq.Box shows interquartile range (25th to 75th percentiles) with the median and whiskers extending to 1.5 times the interquartile range.An open orange circle indicates no bla TEM or bla CTX-M was identified in that sample using Illumina.D) Taxonomic identification (color) and plasmid or chromosome designation (shape) in the 23 clusters (> 80%ID over ≥ 3000 bp) that are shared between hosts (human, poultry, canine) and households.

Figure 4 .
Figure 4. 23 clusters containing bla TEM or bla CTX-M were shared between samples.A) Presence (black)/absence of shared clusters (> 80%ID over ≥ 3000 bp) by sample type and household.ARG and mobile genetic element annotation of shared clusters.B) Shared clusters annotated by household (shape) and host type (color).
We calculated the average coverage of the assembled contigs that resulted from Illumina sequencing and assembly of the cultured E. coli isolates.All instances of bla TEM and bla CTX-M identified in cultured E. coli by Illumina sequencing were also identified by Context-Seq in the same sample.In human fecal samples, the median coverage of the contigs was higher with Illumina sequencing compared to Context-Seq; coverage was 160 with Illumina sequencing (range:[17-735]) and 49[7-1237]with Context-Seq (paired two-sided t-test p=0.82).In animal samples, the median coverage with Context-Seq (171[66-3852]) was greater than Illumina (138[19-1099]), although not statistically significant (paired two-sided t-test p=0.24) (Figure (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.It is made 41mpared to K. pneumoniae and E. coli, there are even fewer studies of resistant H. influenzae, especially in LMICs.One study investigated resistant H. Influenzae in Morocco and found one-third of isolates carried resistance genes to betalactams.41Mostweresusceptible to ESBLs, primarily demonstrating resistance to ampicillin and amoxicillin.41Approximatelyhalf of shared ARGs and their genomic context(11/23 clusters)were shared between animals.An extensive previous study of E. coli across hosts in Nairobi (the UrbanZoo project) found highly similar resistomes among livestock poultry, both within and between households.