Antibiotic-resistant bacterial infections have contributed to 4.95 million deaths worldwide, with 1.27 million of those directly resulting from antimicrobial resistance (AMR; Gigante et al., 2022; Papavarnavas and Mendelson, 2022). It is the leading cause of death worldwide and affects high, low, and middle-income countries (Pokharel et al., 2019; Gupta and Gupta, 2021). Antibiotic-resistant bacterial infections are rising exponentially, making them harder to treat (Dušan et al., 2016; Robinson et al., 2016). This continuation will lead to an estimated 10 million deaths annually by 2050 if the burden of AMR is not alleviated (Van Boeckel et al., 2015).

For most of human history, bacteria have caused infectious diseases and contributed to high mortality rates (MacLean and San, 2019). To counteract this, antibiotics were routinely prescribed to treat and prevent such diseases. This saved millions of lives each year. Due to its success, antibiotics were used in animal husbandry and agricultural practices with the added benefits of growth promotion (Kang et al., 2018). Unfortunately, this success was limited to the emergence of antibiotic-resistant bacteria (ARB) and its effects on human health and longevity (Fair and Tor, 2014; MacLean and San, 2019; Kumar et al., 2020).

The continuous and unwarranted use of antibiotics in humans, animals, and agricultural areas exerts selection pressure on the bacteria found in these sectors. This gives rise to bacteria resistant to multiple antibiotics classes leading to untreatable infections (Davies and Davies, 2010; Cabello and Godfrey, 2016).

The ability of bacteria to resist antibiotics is embedded in their evolutionary history, which has led to various phenotypic and genotypic resistance mechanisms (Pornsukarom and Thakur, 2017; Power, 2019). These include the inactivation, modification, degradation, and expulsion of the antibiotic or its target site, therefore protecting the bacteria (Capita and Alonso-Calleja, 2013; Blair et al., 2015; Munita and Arias, 2016; Kumar, 2017; Kang et al., 2018). It is also possible that bacteria gain resistance through mutations, changing the nature of proteins expressed in the bacterial organism, or by horizontal gene transfer phenomena, which is the exchange of genetic material by the use of mobile genetic elements (MGEs) between bacterial strains (Pornsukarom and Thakur, 2017; MacLean and San, 2019). This allows bacteria to collect numerous resistant traits and ultimately become multi-drug resistant (MDR; Munita and Arias, 2016; von Wintersdorff et al., 2016; Kumar, 2017; Power, 2019).

The ineffectiveness of antibiotics threatens the ability to treat common bacterial infections and minor injuries. Since first-line antibiotics are deemed sub-par, more expensive treatment, alternative therapy and additional care is necessary. This can result in prolonged infections and an increase in morbidity and mortality rates. This puts an economic burden on families, society, and modern medicine (Gandra et al., 2014; Friedman et al., 2016; Aslam et al., 2018; Power, 2019).

For over a decade, antibiotic resistance has been studied using the traditional culture-dependent approach. This involves the use of artificial conditions to culture microorganisms. Unfortunately, some microorganisms cannot be cultured by this approach (Schmieder and Edwards, 2012; Nowrotek et al., 2019). Approximately 80% of bacterial species in the human gut and 99% of environmental bacterial species remain uncultured (Lagier et al., 2012; Nowrotek et al., 2019). Factors such as slow growth, microbial competition, specific growth requirements, and environmental stressors can affect the culturing process. This leaves limited information regarding gene transfer between these bacterial communities (Garmendia et al., 2012; Schmieder and Edwards, 2012; Suenaga, 2012; Nowrotek et al., 2019; Moralez et al., 2021).

Whole metagenome DNA sequencing can describe the genomes of the total microbial community found in nature. This approach enables the study of culturable and non-culturable bacteria by bypassing the need for isolation and laboratory cultivation of microorganisms. DNA directly isolated from the environmental sample can broaden the understanding of the structure, gene/species richness and distribution, and the functional and metabolic potential of a microbial community (Suenaga, 2012).

Applying this approach to antibiotic resistance in different microbial communities can identify known and novel resistance genes and mobile genetic elements, i.e., plasmids, integrons, transposons, and phage’s (Partridge et al., 2018; Stalder et al., 2019; Hall et al., 2022; Yao et al., 2022). This information is the stepping stone to making new policies for infection and prevention control measures, thereby reducing the incidence of infection and optimizing the use of antibiotics by health professionals, in the healthcare, animal, and agricultural industries. In addition, it will provide improved awareness and understanding of antibiotic resistance as a whole.

Unfortunately, AMR monitoring systems focus on clinical and the public health sectors as a representation of AMR as a whole. Organizations such as WHO and the implementation of the Global Antimicrobial Surveillance System (GLASS) have relied on the use of traditional culturable methods for clinical AMR research. This creates a biased outlook on AMR as non-clinical sectors, and non-culturable bacteria are excluded. This review will focus on AMR in non-clinical sectors and the use of bioinformatics as an aid in this research.

Microbiomes are complex and require sequencing technologies to be comprehensive enough to capture the present representative sequences from species. Several sequencing platforms can be used to study AMR in clinical and non-clinical sectors. They are useful in detecting antibiotic resistance genes (ARGs), virulence factors, MGEs, and diversity of the microbial community (Forbes et al., 2017). The advancements in metagenomics have been driven by next-generation sequencing technologies (NGS) i.e., second and third-generation sequencing technologies (Table 1).

Second-generation sequencing has been the most widely used sequencing technology for microbial genomics. Second-generation sequencing technologies, i.e., Illumina and Ion Torrent, produce short reads from 100 up to 300 base pairs with high accuracy (~0.1% error rate; De Maio et al., 2019; Yee and Simner, 2019). These short reads provide information on species population, evolutionary relationships, allelic variations, SNPs, and the detection of specific genes, i.e., ARGs (Boolchandani et al., 2019; De Maio et al., 2019; Yee and Simner, 2019). Although short reads provide information on the genomic content of the bacterial isolates in a sample, it does not provide information on the genomic structure, i.e., which genes are present on chromosomes and MGEs such as plasmids which are crucial in understanding the dissemination of ARGs (George et al., 2017; De Maio et al., 2019; Yee and Simner, 2019; Maguire et al., 2020).

Third-generation technologies overcome this limitation by providing a better view of the genetic structure as longer reads (on average 1–100 Kbp) span most of the repetitive sequences, linking ARGs to MGEs, e.g., plasmids (Boolchandani et al., 2019; Yee and Simner, 2019). This long-read technology is advantageous for studying AMR in bacterial isolates, and metagenomic samples as the complexity of assembly is reduced (De Maio et al., 2019). Unfortunately, the high error rate makes it difficult to identify specific allelic variants or SNPs in chromosomal genes leading to AMR (Yee and Simner, 2019). On the other hand, Pacbio and ONT are easily accessible and able to produce raw data in real-time, making them a good tool for the rapid diagnostics (Che et al., 2019). Long reads are also a practical method to determine the species/gene richness, distribution, and functional potential of a microbial community (Garmendia et al., 2012; Crofts et al., 2017).

While third-generation technologies are successful in AMR research, it is limited by the high error rate and low accuracy (Yee and Simner, 2019). To overcome this, a combination of short-reads (i.e., Illumina) with long-reads (i.e., ONT) is a promising way to generate fully resolved and accurate bacterial genome assemblies which characterize antibiotic resistance genes on plasmids and genomes from environmental samples (Liu et al., 2020; Chen et al., 2021; Moralez et al., 2021). This combination allows for better assembly of complex genomes as long reads provide information on the structure of the genome, and short reads can be used to correct errors in long reads (De Maio et al., 2019).

These sequencing technologies directly characterize the microbiomes in humans, animals, and environmental samples, providing high-throughput sequencing of either whole genomes, targeted amplicons, or whole metagenomes (Di Bella et al., 2013).

Whole-genome sequencing (WGS) uses DNA sequencing technologies to sequence the entire genome of isolated pure organisms to characterize genomic variants (Table 2; Collineau et al., 2019). WGS overcomes the limitations of traditional culture-dependent approaches and phenotypic tests, i.e., disk diffusion for AMR (Balloux et al., 2018). By opting to use WGS, more information on the bacterial genetic determinants conveying resistance and their association with mobile genetic elements can be established (Balloux et al., 2018; Oniciuc et al., 2018). This contributes to outbreak detection, infection control, and epidemiological surveillance, all of which are an essential part of the AMR surveillance (Collineau et al., 2019).

Amplicon sequencing uses marker genes such as 16S (prokaryotes) or 18S rRNA (eukaryotes) genes. Both approaches require the DNA extracted directly from the microbial community and are subjected to either direct sequencing or amplification via polymerase chain reaction (PCR; Table 2). In both instances, the sequences are queried against a reference database (Table 3). This is a reliable process to identify the composition of a microbial community down genus-level within a sample (Ranjan et al., 2016; Forbes et al., 2017; Oniciuc et al., 2018).

Metagenomic sequencing involves the fragmentation, sequencing, assembly, and annotation of the total genomic DNA isolated in a given sample (Oniciuc et al., 2018; Xu et al., 2021). Similar to WGS, whole metagenome sequencing (WMS) can provide information on the entire gene content of either prokaryotic and eukaryotic organisms as well as species or strain-level identification and virulence or resistance potential (Ranjan et al., 2016; Quainoo et al., 2017; Hendriksen et al., 2019). In addition, WMS makes it possible to predict the metabolic potential of the microbial community. This comprehensive view of all gene content allows access to information on AMR that is essential for setting up control and prevention strategies in all sectors (Xu et al., 2021).

Meta-proteomics focuses on functional change in a microbiome. Assisting in the control strategies and plays a key role in identifying molecular mechanisms of bacterial pathogenesis and disease outcome determinants. It can aid in developing pathogen-specific treatment strategies that can lower the spread of AMR. Meta-proteomics is reliable for reviewing bacteria in soil, sludge, food, and the ocean (Wilmes and Bond, 2004; Bastida et al., 2014; Williams and Cavicchioli, 2014; Maier et al., 2017). Unfortunately, there are inconsistent protocols for sample preparation, inefficient bioinformatic tools, and challenges in measuring low-abundance proteins within a complex protein sample (Zhang et al., 2018). This technique, while reliable, should be applied with other sequencing technologies to have an in-depth analysis of microbial communities, antibiotic resistance genes, and host-pathogen interaction (Zhang and Figeys, 2019; Khodadadi et al., 2020).

The constant development of new computational pipelines is advantageous in providing an accurate depiction of the resistome from metagenomic sequencing data. These analyses rely on a variety of algorithms: quality control of DNA sequencing data, genome and metagenome assembly algorithms, read-mapping, variant detection, phylogenetics, taxonomic databases, and visualization (McArthur and Tsang, 2017). AMR databases are discussed in more detail in the next section.

The typical process to identify antibiotic resistance genes from sequenced metagenomes is based on one of two general techniques: (i) read-mapping and (ii) assembly. Each of these could be followed by binning or an annotation step (Table 4). A bioinformatics pipeline for analysing metagenomic data for antibiotic resistance research has been presented in Figure 1.

Read mapping allows for raw read sequences to be directly aligned to reference databases (Table 3) using pairwise tools such as MiniMap2 (Li, 2018), Bowtie2 (Langmead and Salzberg, 2012), Burrows-wheeler aligner (BWA; Langmead and Salzberg, 2012), or by splitting reads into k-mers and mapping to a reference database using k-mer alignment (KMA; Clausen et al., 2016; Figure 2). This provides information on the similarity of one sequence to another by counting the number of alignments to quantify the abundance of similar sequence patterns (Clausen et al., 2016).

In the last decade, read-mapping approaches were deemed superior to assembly approaches for AMR gene detection, i.e., reads are first assembled and mapped to a database using BLAST (Johnson et al., 2008; Hunt et al., 2017; Clausen et al., 2018). The success of an assembly-based approach is highly dependent on the quality of the assembly (Inouye et al., 2014; Clausen et al., 2016). An approach as such is problematic as genes of interest, i.e., ARG, s can be split over two or more contigs and will not be identified if the assembly is poor quality (Clausen et al., 2016, 2018).

Read-mapping based methods overcome this by using the tools above, which allow for fast mapping and alignment of raw reads against large reference genomes and entire databases (Clausen et al., 2016). This can identify genes from low abundance organisms in a complex community with speed and ease of computation. Since epidemiological databases are constantly being updated with new sequences due to natural evolution, the read-mapping approach becomes difficult as results constantly change. There is no guarantee that the read will cover a unique part of the reference sequence and will result in a tie for the best match due to the random selection (Clausen et al., 2016). Fortunately, tools like SRST2 (Inouye et al., 2014) resolve the ties by pre-and post-processing of sequences for read-mapping, which can predict the presence of ARGs in a sample (Clausen et al., 2018; Taş et al., 2021).

Identification and characterization of ARGs can also be achieved using assembly-based approaches (Table 4; Figure 2; Breitwieser et al., 2019). This method allows reads to be assembled into contigs and then queried against reference databases (Table 3). The assembly of WMS data is complicated as there is an uneven or unknown abundance of different genomes (Ghurye et al., 2016).

Unrelated genomes may also contain nearly identical DNA repeats, which could represent MGEs (Boolchandani et al., 2019). In addition, multiple individual organisms could be from the same species but may harbour small genetic differences indicating strain variants (Ghurye et al., 2016). Technical factors, e.g., library preparation, sequencing depth, and sequencing platforms, affect the accuracy in assembling WMS data into larger contigs (Escobar-Zepeda et al., 2015). Metagenomic assemblers such as MEGAHIT (Li et al., 2015), MetaSPades (Nurk et al., 2017), IDBA-UD (Peng et al., 2012), and MetaVelvet (Namiki et al., 2012) have been developed to overcome these challenges by normalizing sequencing depth, correcting read errors, and detecting and reporting genomic variants and repeats (Ghurye et al., 2016; Vosloo et al., 2021). To date, no single assembler is the best at accurately reconstructing known genomes and capturing taxonomic diversity in a metagenomic sample. Multiple assemblers are recommended to be applied to a subset of samples to determine which has the best fit (Escobar-Zepeda et al., 2015; Ghurye et al., 2016; Boolchandani et al., 2019).

Binning classification can predict the taxonomical composition using information contained in reads (Figure 2). Reads are grouped to represent an individual genome or closely related genomes (Table 4). Binning can be done either with reads or assembled sequences and employs two strategies to obtain taxonomic assignment; (i) sequence composition classification and (ii) sequence alignment against references. The first is based on looking at genomic signatures using k-mers to identify evolutionary conservation among species and uses software such as TETRA (Teeling et al., 2004), MetaClusterTA (Wang et al., 2014), and PhylophytiaS (McHardy et al., 2007). Similarly, software like MaxBin (Wu et al., 2016) and Amphora2 (Wu and Scott, 2012) use k-mer signatures but also considers gene markers, GC content, and coverage information for binning on assembled sequences or reads. Another method based on reference read alignment is the Burrows-Wheeler Transform (Burrows and Wheeler, 1994) which indexes like BWA (Li and Durbin, 2009) or Bowtie (Langmead and Salzberg, 2012) and is fast and accurate in assessing species richness and abundance in WMS by aligning them to reference genomes. Genometa (Davenport and Tümmler, 2013) is a software that looks at OTUs in the WMS data and groups according to genomic islands and operons. Taxonomic classification in long reads using the potential coding regions to search in annotated protein databases using BLAST (Johnson et al., 2008) or Megan (Huson et al., 2007). Binning achieved after the assembly of contigs can lead to the generation of partial genomes of unknown or uncultured organisms. This can be used to perform similarity-based binning of other metagenomic datasets. If binning is achieved before assembly, it can reduce the complexity of an assembly-based approach and may reduce the computational demands. However, caution should be taken to ensure the validity of the genome bins as there is a chance of false assignments (Thomas et al., 2012; Bengtsson-Palme, 2018a). In this instance, CheckM (Parks et al., 2015) can be used to evaluate the quality of bins on metagenomic samples and assess the quality of assembled metagenomic genomes by estimation of completeness and contamination based on marker-gene validation (Du and Sun, 2022).

Taxonomic assignments of reads can also be done with genome annotation. This process identifies the coding regions and their location to determine gene function. Annotation of metagenomic sequences can be done in two ways, (i) existing pipelines, e.g., RAST (Aziz et al., 2008) and IMG (Markowitz et al., 2009), can be used for assembled genomes, and (ii) annotation can be performed on an entire community and relies on unassembled reads and short contigs. WMS data is annotated by identifying genes (feature prediction) and assigning putative gene functions and taxonomic neighbours (functional annotation; Thomas et al., 2012). Prediction tools such as MetaGeneAnnotator (Noguchi et al., 2008) or MetaGeneMark use internal information to classify sequences as either coding or non-coding and are done with a low error rate of 2% (Thomas et al., 2012).

The performance of antibiotic resistance gene prediction and taxonomic identification depends on the availability of accurate databases. Two types of public databases exist, generalized AMR databases, which include a wide range of ARGs and mechanistic information, and specialized databases, which provide extensive information on specific gene families, e.g., the β-lactamase family (Hendriksen et al., 2019).

Several bioinformatic tools and databases can be used in AMR research. They focus on fast and reliable predictions of ARB and their genetic determinants in complex communities (Van Camp et al., 2020). It is important to note that the quality and type of sequence play an essential role in AMR research. The bioinformatics tool’s robustness can differ when analysing sequences of low-quality (Hendriksen et al., 2019). Any database’s value depends on maintenance, curation, and continuous updates with detailed metadata. Correctly predicting ARGs from large-scale datasets with improved accuracy depends on input parameters such as e-values, bit scores, identity, and query coverage levels (Lal Gupta et al., 2020).

AMR is a natural phenomenon that predates the use of antibiotics. A wide array of novel and clinically characterized ARGs have been detected in environmental samples ranging from pristine environments to agricultural soil (Yang et al., 2019). The major driving force of AMR is the mis-use of antibiotics while other factors such as poor infrastructure, i.e., hygiene and sanitation play a role in maintaining AMR (Collignon et al., 2018; Essack and Sartorius, 2018). Approximately more than 60% of total antibiotic use occurs outside the field of human medicine (Prestinaci et al., 2015; Kumar et al., 2020; Polianciuc et al., 2020). The main non-clinical sectors that are involved in the development of AMR are animal production (food-animals and aquaculture), agriculture (plants), and the environmental (water and soil) compartment. These sectors are interconnected and facilitate the spread of pathogenic ARB within and between them and ultimately to humans. Table 5 summarizes the findings done on AMR in various non-clinical reservoirs through the use of metagenomics.

The use and misuse of antibiotics in humans, animals, and the environment have been linked to the emergence of ARB in each of these sectors. Antibiotic residues, ARB, and ARGs, are spread to different environments (Miłobedzka et al., 2021). Most antibiotics are excreted in animal urine and faeces (Jayalakshmi et al., 2017). This is introduced to the environment directly by manure amendment, which provides nutrients for crops, or through water which is the primary link between humans, animals, and nature (Miłobedzka et al., 2021).

Since antibiotics are used in the food animal industry, ARB can spread via run-off, which can go into lakes, streams, and rivers. Water from these areas is used for irrigation for agricultural practices. This leads to the spread of ARB from food animals to water, soil, and crops (Figure 1). This interconnectedness facilitates the exchange of AMR and allows for the emergence of new genetic determinants. This emergence and spread can occur by HGT (Schmieder and Edwards, 2012; Pan and Chu, 2017; Sarkar et al., 2018; Mahmood et al., 2019).

Antibiotic resistance affects animals, plants, the environment and human health (Malagón-Rojas and Lagos, 2020). Monitoring and understanding the prevalence, mechanisms and spread of AMR is a priority in non-clinical sectors (Boolchandani et al., 2019). Since measures against AMR are focused on the clinical health sector, the environment is under-represented which allows for significant gaps in data, research and control strategies therefore failing to link the effects of AMR from one sector to the other (Malagón-Rojas and Lagos, 2020).

Metagenomic sequencing allows for the detection of known and novel ARGs, HGT events, and associated MGEs in complex communities providing in-depth information on their prevalence, distribution, and transmission in non-clinical sectors such as the animal industry, agriculture, and the environment (Forbes et al., 2017).

While metagenomics is promising, its implementation in non-clinical sectors is still in its early stages. Metagenomic sequencing creates a vast amount of data and requires specialized bioinformatic expertise, which makes this a costly approach (Forbes et al., 2017; Oniciuc et al., 2018). To date, only a few countries and laboratories have the resources and expertise to use metagenomic approaches as a surveillance system for AMR. In addition, bioinformatics methodologies need to be standardized and constantly updated and curated to allow for an accurate comparison between various samples, as the selection of bioinformatic tools and AMR databases does have a significant impact on the results (Hendriksen et al., 2019; Lal Gupta et al., 2020).

Nevertheless, using bioinformatic tools and databases can assist in identifying risky practices, effects of antibiotic usage, and hotspots of AMR. This can facilitate the design of new policies to control the spread of AMR between clinical and non-clinical sectors (Shen and Zhao, 2018; Nijsingh et al., 2019). Such knowledge and practices are urgent as these sectors, including the public health sector, are faced with a future of untreatable infections, increasingly costly medicinal treatment, a higher cost of living, and an increased mortality rate (Bengtsson and Greko, 2014; Littmann et al., 2015; Nijsingh et al., 2019).

The application of metagenomic and bioinformatic approaches to AMR research can provide fast and reliable predictions of AMR and antibiotic use in various non-clinical sectors. These hold a great promise for understanding AMR molecularly, predicting outbreaks and transmission, and emerging pathogens. This information can lead to better policymaking in each sector and decrease the incidence of infections as conditions for the animal industry, agriculture, and the environmental sectors are improved.

SP and TA conceived, designed and wrote the manuscript. AU and DC-F edited and proof-read the manuscript. All authors contributed to the article and approved the submitted version.

This work is based on the research supported by wholly/in part by the National Research Foundation of South Africa (Grant Numbers: 120192).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.