Conjugation refers to the process wherein donor and recipient bacteria, upon physical contact, establish a stable bridge through pili or channels, facilitating the transfer of MGEs, typically plasmids or transposons, from the donor bacteria to the recipient bacteria (Figure 1A). Conjugative plasmids, equipped with tra gene encoding the complete transfer enzyme, can spontaneously move from one cell to another, as exemplified by the F plasmid in Escherichia coli (E. coli), in the coexistence of non-conjugative plasmids and conjugative plasmids, both can undergo conjugative transfer (Davison, 1999; Qiu et al., 2012). The host range of plasmids is extensive, as conjugative transfer has been observed not only between bacteria of the same genus but also across different genera, and even between different biological species (Li and Zhang, 2022). Compared to bacterial suspension state, biofilms provide bacteria with more stable physical contact conditions, potentially enhancing the occurrence of conjugation. It was found that the conjugative transfer frequency of Staphylococcus aureus in biofilms was 10⁴ times higher than in suspended states (Sandberg et al., 2009). Among the various pathways of HGT, conjugation is considered the most significant and remains the most actively investigated route.

Transformation refers to the process wherein bacteria directly uptake and integrate free DNA fragments from the extracellular environment, acquiring corresponding hereditary traits in the process (Magasanik, 1999) (Figure 1B). Unlike conjugation, transformation does not necessitate physical contact between donor and recipient bacteria; rather, it relies solely on the genetic encoding and regulation within the recipient bacterium (Johnsborg et al., 2007). The occurrence of transformation requires the simultaneous fulfillment of two conditions: the presence of free DNA fragments and competent cells, refer to those that have undergone alterations in cell membrane permeability under specific environmental pressures, with over 80 such species identified to date (Johnston et al., 2014; Liu et al., 2017). Given the widespread occurrence of transformation, researchers speculate the presence of potentially undiscovered bacteria with transformative potential in the environment.

Transduction refers to the process by which phages erroneously package a portion of the host bacterium’s genes (frequency: 10⁻⁵–10⁻⁷) into their heads and transfer these genes to another cell through infection, thereby endowing the recipient cell with the corresponding hereditary traits (Li and Zhang, 2022) (Figure 1C). Transduction is classified into generalized transduction and specialized transduction, with the latter exclusively packaging bacterial DNA near the attachment site of the phage. In contrast, generalized transduction can randomly package any gene from the host into the phage head (Huddleston, 2014). However, due to the specificity of phages, transduction is not suitable for the horizontal transfer of widely distributed genes. Nevertheless, it is crucial to note that the contribution of transduction to HGT in natural environments should not be underestimated (Liu et al., 2020). Furthermore, phages carrying various ARGs have been detected in urban sewage, surface water, animals, and human samples, underscoring phages as potential reservoirs for ARGs (Colomer-Lluch et al., 2014; Quiros et al., 2014).

Recently, a novel gene transfer mechanism mediated by outer membrane vesicles (OMVs) has been identified, termed vesiduction (Figure 1D). Rumbo et al. were the first to discover that OMVs can mediate the transfer of resistance genes, and the transfer occurs rapidly, within a three-hour timeframe (Rumbo et al., 2011). OMVs are double-membrane spherical nanostructures (50–500 nm) generated during bacterial growth (Toyofuku et al., 2019). Several studies have detected plasmids, chromosomal DNA fragments, and phage DNA fragments within OMVs, suggesting their role as carriers for gene transfer (Abe et al., 2020). OMVs can protect DNA from degradation by DNAases or other environmental factors, playing a crucial role in HGT (Liu et al., 2020). However, there are currently significant gaps in the understanding of the exact mechanisms and influencing factors of vesiduction, making it a focal point for future HGT research.

The gut has emerged as a focal point concerning the transfer of ARGs from exogenous bacterial species to indigenous microbiota. Accordingly, the in vitro gut model CoMiniGut has been developed for determining horizontal plasmid transfer under conditions that mimic the human colon environment, with a working volume of only 5 mL (Anjum et al., 2018; Wiese et al., 2018) (Figure 2). It comprises several parallel reactors units connected to a data logger. pH is regulated using injectors containing NaOH, while temperature control is maintained through the water bath device, ensuring stability within the simulated environment. Mehreen et al. utilized this system to regularly sample, with techniques such as the selective plate and Fluorescence-Activated Cell Sorting (FACS) to investigate the conjugative transfer frequency of bla_CMY-2 bearing plasmids in E. coli. Additionally, diversity analysis was conducted on the sorted transconjugants (Anjum et al., 2018, 2019).

Compared to other examining methods, CoMiniGut can simulate realistic intestinal environment, enabling the assessment of public health risks associated with the ARG transfer from exogenous E. coli (Anjum et al., 2018). Compared to animal models, CoMiniGut model exhibits high fidelity, offering more controlled experimental conditions such as temperature, pH, and oxygen concentration. Therefore, the establishment process of this system is relatively complex, necessitating extensive experimental validation to ensure its authenticity and accuracy, at a considerable cost as well.

Building upon conventional cultivation methods, researchers have developed a method that combines cultivation and HGT detection simultaneously, utilizing a microfluidic chip. The microfluidic chip is a miniature experimental platform with microscale channels that can be used to control fluid flow. It can simulate both the physical and biological conditions relevant to microorganisms (Abe et al., 2020). Therefore, as an innovative tool, the microfluidic chip serves as a microreactor for microbial growth and mating assay, opening up new avenues for studying ARG transfer (Karimi et al., 2015; Li et al., 2018) (Figure 2C). Microfluidic chips equipped with delayed imaging, coupled with fluorescence technology, enable real-time observation of community changes and tracking of the transfer dynamics of ARGs, which is frequently employed in studies pertaining to conjugative transfer. In addition to the modified bacteria mentioned in Section 3.1.2, the process of infection and transduction caused by phages can be observed in real time by fluorescent labeling of phage (Doolittle et al., 1996; Tzipilevich et al., 2017), thus microfluidic chips also possess potential applications in the transduction research.

Microscopy enables in situ observation and single-cell imaging of ARG transfer, facilitating the differentiation between HGT and VGT. A microfluidic chip comprising eight parallel shallow micro-chambers was employed for distinguishing HGT and VGT. The chip was utilized for cultivating donor cells with plasmids encoding ARGs (RP4 and PKJK5) and recipient activated sludge bacteria. Plasmids were visualized in situ, and the transfer characteristics were analyzed (Li et al., 2019). In image analysis, the appearance of the first randomly occurring green fluorescence particle was considered as the initial sign of the first HGT. Based on this point, the extension of fluorescence particles due to cell growth and division was considered VGT, allowing for a clear distinction of the ratio between HGT and VGT during the diffusion of ARGs in biofilm structures. Additionally, another microfluidic chip containing a single channel was employed (Li et al., 2018; Qiu et al., 2018). In this chip, nutrients within the flow diffuse through the agarose membrane into the bacterial layer, concomitantly allowing the diffusion of metabolic waste out of the membrane, which is subsequently carried away by the flow. This chip facilitates the determination of dynamic features of the transfer process, encompassing cell growth rate, and kinetic variations in transfer frequency. This on-chip culture enables rapid exchange of substances and allows high density growth, and thus provides a close-to-in vivo model to study real-world biofilms. In a recent study, a microfluidic chip was employed featuring four identical cavities in lieu of channels, ensuring ample nutrient supply, evaluating the effect of heavy metal on ARG transfer between attached bacteria (Lin et al., 2019; Liu et al., 2022).

Furthermore, fluorescence combined with FACS enables high-throughput counting and screening of donors, recipients, and transconjugants, further 16S rRNA gene sequencing of the transconjugants can determine which host the plasmid has spread to (Klumper et al., 2015; Feng et al., 2021). However, this method requires artificial modification or labeling of bacteria, making it not only operationally complex but also prone to false positives or negatives. Additionally, fluorescence can only detect one plasmid in each experiment, and current research is mainly focused on E. coli, with other model strains yet to be developed.

Due to the natural environment is complex, the information obtained from laboratory settings are hardly to reflect the real conditions in the natural environment. Thanks to the evolving bioinformatics methods, we are gradually gaining insights into the HGT within natural microbial communities, allowing for a better characterization of the mechanisms underlying interactions between bacterial hosts and their genomes (Figure 2D). Moreover, the bioinformatics method is suitable for studying most HGT pathways.

The current research on HGT dynamics is predominantly focused on simulating conjugation process, encompassing two categories of model: deterministic models and stochastic models.

Transformation models are gradually developed referring to the establishment process of conjugation models. Lu et al. adapted from the mass action law, allowing the transformation frequency by the proposed model to be influenced by varied DNA or cell concentrations (Lu et al., 2015). This was the first transformation dynamics model, successfully predicting the natural transformation frequencies of tetracycline resistance genes in both motile and non-motile strains of Azotobacter vinelandii. Asgher et al. extended the aforementioned model by incorporating additional process pathways, such as variations in nutrient concentration, alterations in the density of susceptible bacterial population and changes in the extracellular DNA density in response to the population of antibiotic resistant and susceptible bacteria, thereby enhancing the capacity to better capture the underlying dynamics (Ali et al., 2022).

Acinetobacter baylyi (A. baylyi) has become a commonly selected model bacterium in the transformation model-building process. Yue et al. based on the dynamics of A. baylyi ADP1, transformant populations, and the free plasmid pool, proposed and calibrated an ODE model to predict the transformation dynamics exposed to non-antibiotic drugs (Wang et al., 2022). Similarly, Yu et al. used the same model to predict the long-term effects of artificial sweeteners on transformation dynamics (Yu et al., 2022). Additionally, Robert et al. discovered that A. baylyi has a lytic effect on nearby E. coli, acquiring ARGs from neighboring bacteria considered as another way of transformation, based on this process, a population dynamic model with spatially structured microbial communities was established, quantifying transformation frequency on solid surfaces (Cooper et al., 2017).

There are two main types of transduction. Specialized transduction refers to the process in which the prophage erroneously excises DNA adjacent to the integration site (Chen et al., 2018). Generalized transduction involves the “erroneous packaging” of random bacterial DNA fragments, leading to the formation of transducing particles probably containing ARGs (Penades et al., 2015). Therefore, the limited objects of simulation regarding transduction currently focus on generalized transduction.

Volkova et al. initially proposed a mathematical model regarding the risk of antimicrobial resistance (AMR) spread through temperate phage transduction (Volkova et al., 2014). This study assumed a well-mixed and high-density cellular environment in the bovine intestinal tract, estimating the upper limit of transduction in E. coli due to generalized transduction (which is 10⁻³ times the same environmental conjugation rate). Moura et al. used individual-based models suggesting generalized transduction is a powerful mechanism of DNA transfer between strains, allowing the emergence of single and even double resistant variants (Fillol-Salom et al., 2019). The study proposed that the “erroneous packaging” of transduction is not accidental but an evolutionary characteristic of temperate phages in changing environments. Taking into account the no well-mixed host phage systems, Sankalp et al. developed a model with a small volume compartment to consider local effects (Arya et al., 2020). Through sensitivity analysis, it was discovered that transduction frequency was decreased in a more toxic environment, or with higher fitness costs of resistance or phage immunity.

Quentin et al. argued that the parameters of the three models mentioned above mainly rely on assumptions and previous studies, lacking reliable experimental data as evidence, which may limit the reliability of their results. Therefore, an interdisciplinary approach, combining experiments and simulations, was adopted to predict how phage-bacteria dynamics lead to the evolution of multidrug-resistant bacteria. The estimated transduction frequency was approximately 10⁻⁸ (Leclerc et al., 2023). Based on this, the model was further extended to simulate the impact of antibiotics on the phage-bacteria system under the same environmental conditions (Leclerc et al., 2022). The results indicated that the synergistic effect of phages and antibiotics rapidly killed bacteria but also led to faster ARG spread, depending on the interaction duration and antibiotics concentration. This simulation provides important guidance for future experimental and clinical work on the impact of phages on AMR evolution.