One of the earliest methods developed to discriminate between strains of M. tuberculosis involved the use of mycobacteriophages (Table 3), which specifically lyse certain strains, leading to the formation of plaques on solid agar plates (Jones and Greenberg, 1978). Phage typing has also been used to identify cross-contamination, transmission dynamics and MSIs based on the sensitivities of M. tuberculosis isolates to a panel of selective phages (Mankiewicz and Liivak, 1975; Bates et al., 1976; Snider et al., 1984; Jones, 1988), but more modern methods are faster and offer better discriminatory power (Snider et al., 1984). Our primary workflow did not find any reports on the identification of MSIs using phage typing, but secondary searches found two such studies. Mankiewicz and Liivak (1975) sampled 233 patients, of which 33 (14.2%) showed evidence of MSIs due to the presence of multiple M. tuberculosis phage types in a single culture. In the other study, Bates et al. (1976) analyzed samples from 87 different patients and identified three (3.4%) as having mixed phage-typing profiles. While the possibility that the presence of multiple M. tuberculosis phage types within the same patient could indicate an MSI, the limited discriminatory power of the method prevents definitive confirmation and cannot completely rule out intra-host microevolution.

Until recently, RFLP based on the insertion sequence IS6110 was the standard method used for comparing the genetic relatedness of M. tuberculosis isolates (Van Soolingen et al., 1993, 1995). IS6110 (sometimes referred to as IS986) belongs to the IS3 family, members of which are only present in the MTBC (Hermans et al., 1990; Thierry et al., 1990a,b). M. tuberculosis and M. bovis strains can contain 0–25 and 0–3 copies each of IS6110, respectively (Cave et al., 1991, 1994; Van Soolingen et al., 1991; Yuen et al., 1993; Fomukong et al., 1994; Brosch et al., 2000; Lok et al., 2002; Singh et al., 2004; Steensels et al., 2013). Variations in the copy number of IS6110 elements within different M. tuberculosis strains make it an attractive target for epidemio-typing isolates containing multiple copies of the insertion sequence, but not in low copy number strains. Therefore, IS6110-typing has led to the development of extensively used standardized protocols (Table 3) (Van Embden et al., 1993).

In total, 26 (21.5%) of the publications reported herein included the use of IS6110-typing methods for detecting MSIs involving MTBC members (Chaves et al., 1999; Pavlic et al., 1999; Yeh et al., 1999; Braden et al., 2001; Du Plessis et al., 2001; Richardson et al., 2002; García De Viedma et al., 2003, 2005; Allix et al., 2004; Das et al., 2004; Shamputa et al., 2004, 2006; Baldeviano-Vidalón et al., 2005; Cox et al., 2005, 2007; Van Rie et al., 2005; Van Der Zanden et al., 2007; Andrade et al., 2009; Mokrousov et al., 2009; Gardy et al., 2011; Navarro et al., 2011; Von Reyn et al., 2011; Adams et al., 2012; Cerezo et al., 2012; Huyen et al., 2012; De Almeida et al., 2019), 11 of which exclusively used IS6110-RFLP (Supplementary Table 2) and one study used IS6110-inverse polymerase chain reaction (PCR) (Mokrousov et al., 2009). All the studies employed cultures in their analysis, and IS6110-inverse PCR was primarily used as a tool to identify strains belonging to the Beijing evolutionary lineage (Mokrousov et al., 2003, 2009). The first report on MSI detection using IS6110-RFLP was by Yeh et al. (1999), who noticed multiple bands of varying intensities in a sample from a patient due to the presence of two separate strains. Another study found an MSI with two drug-susceptible isolates in an immunosuppressed patient (Pavlic et al., 1999), and similar infections with drug-susceptible and drug-resistant M. tuberculosis have also been detected by IS6110-RFLP in separate reports (Braden et al., 2001; Hofmann-Thiel et al., 2009). Overall, MSIs were identified in 0.4–100% of cases using IS6110-RFLP in these studies (Supplementary Table 2), where most reports with 100% MSI detection rates involved a single patient (Pavlic et al., 1999; Yeh et al., 1999; Andrade et al., 2009). In addition, a recent study detected MSIs in 3 out of 17 samples from patients using two probes for IS6110-RFLP instead of a single probe used conventionally (De Almeida et al., 2019).

Spacer oligotyping (spoligotyping) is a commonly used method for the simultaneous detection and identification of MTBC members (Table 3) (Kamerbeek et al., 1997). Spoligotyping exploits the nucleotide sequence diversity of clustered regularly interspaced palindromic repeats (CRISPRs) (Haft et al., 2005), which are present in many bacteria and archaea (Barrangou et al., 2007). The chromosomal locus specifically used in this assay is known as direct repeat (DR) in mycobacteria (Hermans et al., 1991). Spoligotyping is traditionally performed by amplifying the entire DR region using PCR with a pair of oligonucleotide primers, one of which is labeled with biotin to aid in the detection of products by hybridization. Membranes containing a unique set of 43 covalently bound synthetic oligonucleotide spacer sequences derived from M. tuberculosis and M. bovis Bacille Calmette-Guérin (BCG) are used in the hybridization (Kamerbeek et al., 1997), and can differentiate between MTBC isolates based on the presence or absence of spacers.

Spoligotyping has been widely used in epidemiological studies to investigate the cause of recurrent TB (defined as endogenous reactivation of an initially infecting strain or exogenous reinfection with a different strain) (Small et al., 1994; Warren et al., 2002; Van Rie et al., 2005; Andrews et al., 2008), tracking epidemics (Goyal et al., 1999; Källenius et al., 1999; Caminero et al., 2001), and investigating laboratory cross-contamination (Nivin et al., 2000). Twenty (16.5%) publications reported using spoligotyping as one of the methods for detecting MSIs, with a majority employing one (Pavlic et al., 1999; Cox et al., 2005; García De Viedma et al., 2005; Umubyeyi et al., 2007; Furphy et al., 2012; Biffa et al., 2014; Chaoui et al., 2014; Lamine-Khemiri et al., 2014; Ssengooba et al., 2015; Wang et al., 2015; Egbe et al., 2017; Baffoe-Bonnie et al., 2019) or two (García De Viedma et al., 2003; Mokrousov et al., 2009; Huyen et al., 2012; Navarro et al., 2015; Silva-Pereira et al., 2019) additional methods (Supplementary Table 2). Only three reports used spoligotyping as the sole genotyping method and reported MSIs at frequencies ranging from 11.8 to 57.1% (Andrews et al., 2008; Kamakoli et al., 2018; Guernier-Cambert et al., 2019).

One significant limitation of spoligotyping is that it can underestimate MSIs as hybridization signals from multiple strains in a sample can overlap and appear as a single pattern (Kamakoli et al., 2018). For this reason, when spoligotyping is used to investigate MSIs, subculturing is usually performed to obtain single isolated colonies for testing (García De Viedma et al., 2003; Shamputa et al., 2004; Huyen et al., 2012). The ability to detect MSIs in the latter case is dependent upon the proportion of different strains in the original sample and the number of colonies picked for analysis. To help resolve this problem, Lazzarini et al. (2012) developed a computational method that can predict if individual spoligotypes contained signatures from more than one of four major global lineages, which would indicate an MSI. In most cases, a secondary typing method like MIRU-VNTR, IS6110-RFLP or WGS may be required to verify results that may appear to contain a single spoligotyping pattern. It is worth noting that although spoligotyping can be applied directly to clinical specimens, all the studies reported herein used cultures, possibly due to the requirement of purified DNA for other methods employed by the authors (Shamputa et al., 2004, 2010; Hofmann-Thiel et al., 2009). Warren et al. (2004) used a combination of a lineage-specific PCR and spoligotyping to distinguish between M. tuberculosis strains belonging to the Beijing- and non-Beijing evolutionary lineages. They detected MSIs in 19% of samples from patients associated with retreatment in their study. In addition, WGS was used as second technique in three spoligotyping based studies (Witney et al., 2017; Baffoe-Bonnie et al., 2019; Silva-Pereira et al., 2019), one of which focused on animals. In their work, Silva-Pereira et al. (2019) detected an M. pinnipedii MSI in a South American sea lion (Otaria flavescens) using WGS, which was not suggested by in silico spoligotyping or MIRU-VNTR initially. The above-mentioned studies demonstrate the importance of using more discriminatory methods along with traditional screening techniques to ensure the detection of different strains that might be present in a single sample.

Variable number tandem repeats (VNTR) are short DNA sequences (Nakamura et al., 1987), which are dispersed throughout the genomes of many bacteria and eukaryotes (Cox and Mirkin, 1997). They vary in repeat unit length and repeat number depending on the specific organism and locus being analyzed (Supply et al., 1997). Since the repeat unit length at specific VNTR loci is known for each species, determining the numbers of repeats present at the respective loci can be used to discriminate between strains (Table 3). The use of VNTRs for typing M. tuberculosis strains was first reported in 1998 (Frothingham and Meeker-O'Connell, 1998), and since then the discrimination power has been improved by using combinations of mycobacteria-interspersed repetitive units (MIRUs) located at different loci throughout the genome (Smittipat and Palittapongarnpim, 2000; Le Flèche et al., 2002; Roring et al., 2002; Skuce et al., 2002). Initially, 12-locus MIRU-VNTR was used widely (Supply et al., 2001), but the method had some limitations in its ability to discriminate between unrelated isolates (Supply et al., 2006). To overcome this problem, the stability and resolution power of 29 MIRU-VNTR loci was evaluated using predominant global M. tuberculosis lineages, resulting in the standardization of 24-MIRU-VNTR loci for high-resolution epidemiological studies (Supply et al., 2006). In addition, the 15 most discriminatory loci of the 24 were selected for use in routine epidemiological investigations involving M. tuberculosis (Supply et al., 2006).

To generate a MIRU-VNTR profile, several genomic regions known to contain VNTRs are amplified by PCR using specific primer pairs either individually (simplex) or in multiples (multiplex). In this way, the number of repeats at each VNTR locus can be determined using different DNA sizing techniques for comparing isolates. It was found that 50 studies used VNTR-typing to detect M. tuberculosis MSIs, where 28 studies used it as the sole discriminatory method for this purpose (Allix et al., 2004; Shamputa et al., 2004; Umubyeyi et al., 2007; Fang et al., 2008; Hofmann-Thiel et al., 2009; Mokrousov et al., 2009; Stavrum et al., 2009; Dickman et al., 2010; Mulenga et al., 2010; Cohen et al., 2011; Gardy et al., 2011; Navarro et al., 2011, 2015; Cerezo et al., 2012; Furphy et al., 2012; Huyen et al., 2012; Hingley-Wilson et al., 2013; Muwonge et al., 2013; Peng et al., 2013; Biffa et al., 2014; Chaoui et al., 2014; Lamine-Khemiri et al., 2014; Zetola et al., 2014; Barletta et al., 2015; Mei et al., 2015; Pang et al., 2015; Shin et al., 2015, 2018; Ssengooba et al., 2015; Streit et al., 2015; Wang et al., 2015; Zheng et al., 2015; Antusheva et al., 2016; Hajimiri et al., 2016; Hu et al., 2016; Egbe et al., 2017; Farmanfarmaei et al., 2017; Ghielmetti et al., 2017; Kamakoli et al., 2017, 2019, 2020a,b; Khosravi et al., 2017; Kontsevaya et al., 2017; Nathavitharana et al., 2017; Sadegh et al., 2017; Silva-Pereira et al., 2019; Abascal et al., 2020; Baik et al., 2020; Sichewo et al., 2020). In addition, 24-locus MIRU-VNTR was the most commonly used method for detecting M. tuberculosis MSIs, and nearly all (47/50, 94.0%) of the reports used some form of culturing for analysis. Comparatively, MIRU-VNTR detected more MSIs than any other method based in the current review (Table 2). One reason for this could be the use of PCR amplification during MIRU-VNTR, which increases the sensitivity and detection power of the method (Shamputa et al., 2006), especially in instances where different strains are not proportionally present in a single sample.

Among other PCR-based methods used to detect MSIs, the majority were focused on differentiating between M. tuberculosis strains belonging to the Beijing and non-Beijing evolutionary lineages (Warren et al., 2004; Van Rie et al., 2005; Huang et al., 2010; Wang et al., 2011; Mustafa et al., 2016). Of the 121 studies we examined, 13 (10.7%) used region-specific PCR methods and detected MSIs in 3.2–100% of samples (Supplementary Table 2) (Theisen et al., 1995; Lourenço et al., 2000; García De Viedma et al., 2003; Warren et al., 2004; Baldeviano-Vidalón et al., 2005; Van Rie et al., 2005; Huang et al., 2010; Mallard et al., 2010; Wang et al., 2011; Hanekom et al., 2013; Peng et al., 2013; Müller et al., 2014; Mustafa et al., 2016). For example, Warren et al. (2004) detected MSIs in 35 (18.8%) of the 186 sputum cultures from TB patients tested during an epidemiologic study in South Africa. They reported that MSIs were more often associated with retreatment (23%) vs. new cases (17%), and the sensitivity and specificity of their method was comparable to IS6110-RFLP and spoligotyping (Warren et al., 2004). Using the same method, Van Rie et al. (2005) also detected one case of MSI with drug-susceptible and drug-resistant M. tuberculosis isolates. While the method developed by Warren et al. (2004) used simplex PCR, Huang et al. (2010) utilized multiplex PCR to detect MSIs caused by Beijing and non-Beijing lineage strains in 11.3% of the 185 sputum samples from patients without any prior history of TB treatment. Another group used quantitative PCR to detect MSIs based on the presence of both Beijing and non-Beijing lineages in 3% of TB patients <25 years in age using clinical specimens and cultures (Wang et al., 2011), whereas M. tuberculosis isolates belonging to the two lineages were also detected together in 14.7% cases by Mustafa et al. (2016). In contrast, a Latin American and Mediterranean (LAM) and non-LAM lineage-based PCR found MSIs in 4 out of 160 (2.5%) culture-positive sputa analyzed (Mallard et al., 2010), suggesting that such methods are useful in identifying MSIs under settings where the occurrence of M. tuberculosis isolates from mixed lineages is high. Therefore, by using an in-house PCR to identify isolates from the Beijing, Haarlem, S-family, and LAM evolutionary lineages, Hanekom et al. (2013) were able to detect MSIs in 31 (15%) of the 206 samples analyzed in their study.

M. tuberculosis MSIs have also been reported in patients with discordant drug susceptibility profiles on more than one occasion. Isolates from 10 such TB patients out of 89 (11.2%) were confirmed to have MSIs using Beijing lineage-specific PCR (Chen et al., 2007) and 16-locus MIRU-VNTR (Peng et al., 2013). Other techniques such as linker PCR (Theisen et al., 1995) and gyrA PCR/sequencing (Sreevatsan et al., 1997; Müller et al., 2014) have also detected M. tuberculosis MSIs in cases involving discordant drug susceptibility profiles and also in archaeological samples. In addition, two studies employed double repetitive element PCR (Lourenço et al., 2000; Baldeviano-Vidalón et al., 2005) based on the IS6110 and a GC-rich repetitive sequence described by Friedman et al. (1995). Baldeviano-Vidalón et al. (2005) also observed multiple discrepancies in drug susceptibility testing results among follow up samples from patients, which were attributed to MSIs based on IS6110-RFLP analysis. Such studies emphasize the importance of considering MSIs during TB drug susceptibility testing and while devising appropriate treatment regimens.

Methods based on WGS provide the ability to examine strain diversity at very high resolution, which cannot be achieved by other techniques such as RFLP or MIRU-VNTR. Heterogeneous SNPs are predominantly used for strain discrimination, and the presence of many different SNPs in isolates from a single sample is suggestive of MSIs (Sobkowiak et al., 2018). While the concept is simple–a SNP is confirmed within several sequencing reads used to assemble the locus being examined (Table 3), the technical criteria for identifying bona fide SNPs varies between studies. Factors that can affect SNP detection include the quality and depth of sequencing, experimental design, sample preparation and pathogen species, to name a few (Hatherell et al., 2016). Some reports require that the frequency of the alternative base at a specific locus be found from anywhere between 5 and 30% or more of the reads for SNP calling (Bryant et al., 2013a; Guerra-Assuncaõ et al., 2015; Dippenaar et al., 2019), or even just two reads in some deep sequencing studies (Gan et al., 2016). Some studies also include threshold nucleotide base quality scores to minimize artifacts (Bryant et al., 2013a; Gan et al., 2016; Nimmo et al., 2020). Additionally, several loci with heterogeneous bases must be identified between isolates to qualify them as MSIs. However, the minimum number of SNPs used to qualify the presence of an MSI using isolates also varies, as anywhere over 16 to 80 have been used for the purpose depending on the sequencing technology (Bryant et al., 2013a; Gan et al., 2016). In addition, WGS analysis cannot be performed directly on clinical samples in most cases because the genomic complexity of the sample limits the confidence at which SNPs are called. Therefore, WGS often requires axenic cultures for strain typing and MSI identification (Van Den Berg et al., 2005; Döpfer et al., 2008; Davidson et al., 2016).

Of the 97 studies identified in Covidence, eight (8.2%) primarily used WGS to detect MSIs in humans (Bryant et al., 2013a; Gan et al., 2016; Witney et al., 2017; Baffoe-Bonnie et al., 2019; Dippenaar et al., 2019; O'Donnell et al., 2019; Anyansi et al., 2020; Wollenberg et al., 2020), while three others (3.1%) used the method to detect MSIs in various animal species (Davidson et al., 2016; Pfeiffer et al., 2017; Silva-Pereira et al., 2019), only one of which focused on a MTBC bacteria (Silva-Pereira et al., 2019). Secondary searches found four more human studies on WGS and mycobacterial MSIs (Chan et al., 2013; Kay et al., 2015; Dheda et al., 2017; Nimmo et al., 2020), in addition to reports where the method was used to confirm such infection that were initially identified using other means (Supplementary Table 2). For example, six isolates from 47 paired patient samples taken before and after treatment during the REMoxTB clinical trial were initially classified as relapses or re-infections by MIRU-VNTR but were later determined to be MSIs by WGS analysis (Bryant et al., 2013a). A follow-up study re-examined the same data using QuantTB (Anyansi et al., 2020), a tool developed to identify MSIs through the iterative comparison of SNPs and suggested that only four of the six MSI cases could be classified as such. O'Donnell et al. (2019) used isolates from a patient where drug susceptibility testing alluded toward an MSI involving susceptible and resistant M. tuberculosis, which was confirmed using WGS. They showed that the patient was initially infected with a drug-susceptible strain followed by an extensively drug-resistant M. tuberculosis isolate, which was selected for during antibiotic therapy. The high resolution of WGS underscores its importance in strain typing for devising individualized treatments for TB infections, although its widespread use may be limited in many high TB burden settings due to insufficient resources or technical capabilities. The ability to use traditional culture based drug susceptibility testing methods have limitations for many slow growing pathogenic mycobacteria, but genomics based technologies are more rapid and allow for the detection of resistance based on the presence of conferring mutations (Nicol and Cox, 2019). While PCR methods targeting specific genes can detect important drug resistance mutations for early diagnosis, WGS can additionally infer potential resistance, allowing for individualized treatment regimens (Nicol and Cox, 2019).

The use of WGS also enhances the detection of potential MSIs. For example, while examining pre-and post-treatment isolates from a TB patient, Witney et al. (2017) noticed 57 SNP differences between two of them, indicating a re-infection by a second strain. More detailed analysis of the WGS data indicated a potential MSI in the pre-treatment sample at a 3:1 genotypic ratio by two strains, where the minor genotype was closely related to the post-treatment isolate. This suggested that recurrent disease was caused by a relapse, where one of the two strains from the MSI was eliminated during the initial course of therapy. Interestingly, 24-locus MIRU-VNTR typing did not detect the genotype of the post-treatment isolate in the pre-treatment sample, which is intriguing, as the method was previously shown to detect MSIs in proportions as low as 1:99 (García De Viedma et al., 2005). Such reports suggest that many MSIs might have gone undetected due to technical limitations and could have potentially affected disease outcomes.

Although more TB MSIs are now being reported, evidence from archaeological studies indicate that the phenomenon has been around for a long time. Through metagenomics analysis of ancient DNA, one study identified an MSI (difference of 398 SNPs) within an eighteenth-century Hungarian mummy (Chan et al., 2013). A follow up study by a related group reported five MSIs in eight mummified bodies examined from the same archaeological site using similar metagenomics-based methods (Kay et al., 2015). In addition, a separate study employed gyrA PCR to successfully identify an MSI using ancient DNA from a variety of archaeological samples across Britain and France (Müller et al., 2014). The detection of MSIs using ancient DNA and complex samples is intriguing. Such reports also provide precedence for using metagenomics and other technologies to examine the prevalence and impact of such infections in future prospective and retrospective studies.

Our review showed that MSIs involving NTM have not been investigated to the same extent as compared to those caused by M. tuberculosis (Table 2). Amongst the 121 mycobacterial studies identified, 26 (21.5%) examined NTM, of which 17 (14.0%) and eight (6.6%) found MSIs in humans and animals, respectively, whereas one report identified MSIs in both human (Arbeit et al., 1993; Slutsky et al., 1994; Von Reyn et al., 1995; Devallois and Rastogi, 1997; Picardeau et al., 1997; Wallace et al., 1998; Legrand et al., 2000a,b; Oliveira et al., 2000; Saad et al., 2000; Dvorska et al., 2002; Panunto et al., 2003; Ohkusu et al., 2004; De Sequeira et al., 2005; Fujita et al., 2014; García-Pedrazuela et al., 2015; Kimizuka et al., 2019) and animal (Dvorska et al., 2007; Shitaye et al., 2008; Furphy et al., 2012; Gerritsmann et al., 2014; Johansen et al., 2014; Gioffré et al., 2015; Podder et al., 2015; Davidson et al., 2016; Pfeiffer et al., 2017) populations simultaneously (Pate et al., 2008). Many early studies used pulsed field gel electrophoresis (PFGE) to identify MSIs in 14.3–100% of patients infected with MAC bacteria (Arbeit et al., 1993; Slutsky et al., 1994; Von Reyn et al., 1995; Wallace et al., 1998; Ohkusu et al., 2004). PFGE was also used as the sole method to identify MSIs associated with other NTMs. For example, Legrand et al. (2000a) found 33.3% M. simiae MSI prevalence (3 of 9 hosts) in their study involving AIDS patients. In addition, PFGE was used in combination with other methods in studies on NTM MSIs. A report by Picardeau et al. (1997), detected 3 M. avium MSIs while examining 93 samples from AIDS patients using simple double repetitive element PCR (MaDRE-PCR), which amplifies a region of the M. avium chromosome between IS1245 and IS1311 (Picardeau and Véronique, 1996). They confirmed these results using IS1245-RFLP, which showed the presence of multiple low-intensity bands in the same sample. Pulsed field gel electrophoresis was also able to pick up multiple banding patterns, including in some samples from other patients (Picardeau et al., 1997), but the criteria for attributing them to different strains (≥3 differences) was not surpassed (Tenover et al., 1995). A similar study on 31 AIDS patients from the Caribbean islands initially identified three potential polyclonal M. avium infections based on IS1245-RFLP, but PFGE analysis showed that two isolates had identical banding patterns (Legrand et al., 2000b). Therefore, results from PFGE analysis do not corroborate those obtained by other methods on multiple occasions, which might allude to differences in their discriminatory powers.

IS1245 and IS1311 have also been used as general fingerprinting targets to identify an MSI (25%) involving MAC members from 25 patients (Devallois and Rastogi, 1997). In a separate study, Oliveira et al. (2000) used RFLP analysis of hsp65 PCR products, IS1245 and IS1311, respectively, to detect an M. avium MSI. Though the presence of IS1245-RFLP may be supplemented using additional methods, studies have shown it alone is capable of identifying 2.6–100% of NTM MSIs in samples from human subjects, some with HIV-AIDS (Supplementary Table 2) (Picardeau et al., 1997; Saad et al., 2000; Dvorska et al., 2002; Panunto et al., 2003; Pate et al., 2008). In addition, a study examined 41 samples from 14 AIDS patients using MaDRE-PCR, which found four cases with multiple banding patterns, though re-evaluation using IS1245-RFLP only confirmed two as M. avium MSIs (De Sequeira et al., 2005). Based on the reports mentioned above, the feasibility of using IS1245-RFLP as a screening method for detecting NTM MSIs warrants further evaluation.

Kimizuka et al. (2019) have also used a variety of VNTR loci, including 16 M. avium tandem repeats (MATR) and 5 Higashi Nagoya tandem repeats (HNTR) to identify M. avium MSIs in samples from nine patients (13.8%) out of 65 examined. Another study examined samples from 120 patients with pulmonary MAC infections (94 M. avium and 26 M. intracellulare) (Fujita et al., 2014). MIRU-VNTR analysis using 15 loci for M. avium and 16 loci for M. intracellulare successfully identified 20 and 7 cases of M. avium and M. intracellulare MSIs, respectively. Other methods that have detected NTM MSIs in humans include random amplified polymorphic DNA (RAPD) analysis (Zhang et al., 1997). García-Pedrazuela et al. (2015) used the method and found MSIs involving several species including M. abscessus, Mycobacterium chelonae, Mycobacterium fortuitum and Mycobacterium mucogenicum while examining 64 isolates from Spanish patients.

Systematic screening using the Covidence platform also identified some reports on NTM MSIs in animals, including a few involving M. avium subsp. paratuberculosis (MAP). Gerritsmann et al. (2014) examined 39 MAP containing samples from a variety wild and domestic ruminant species, five of which were classified as MSIs using 8-locus MIRU-VNTR by them (Supplementary Table 2). In another study, Gioffré et al. (2015) used 8-locus MIRU-VNTR to identify a MAP MSI from a group to 97 cattle, sheep and goats based on differences in two loci. MAP MSIs have also been identified using multi-locus short sequence repeat (MLSSR or SSR) typing (Podder et al., 2015), which examines sequences that vary between isolates and allows for their discrimination (Amonsin et al., 2004). In addition, the use of DNA detection and sizing techniques such as fragment analysis of labeled PCR products further improves the resolution or SSR typing (Oakey et al., 2014; Podder et al., 2015). Using this strategy, Podder et al. (2015) identified MAP MSIs in all 18 animals from their study, which was subsequently confirmed by WGS analysis using some of their isolates (Davidson et al., 2016). MAP strains from the same animal had significantly different SNPs at high frequencies, which ruled out microevolution based evolutionary rates (Bryant et al., 2016).

In addition to MAP, there were reports on MSIs due to other NTM. M. avium subsp. avium MSIs were identified in domestic chickens using IS901-RFLP, where multiple banding patterns were detected in 7 of the 16 (43.8%) tissue samples tested (Shitaye et al., 2008). Eight-locus VNTR was used to detect M. avium subsp. hominissuis MSIs in pigs using isolates from multiple organs of a single animal (Johansen et al., 2014). Another study that previously reported a single M. avium subsp. hominissuis MSI in a human (out of 26 patients, 3.8%) using IS1245-RFLP also detected similar infections in 33 pigs (13.5%) (Pate et al., 2008). WGS was also used to detect non-MAP MSIs from 113 birds by Pfeiffer et al. (2017). They reported 12 cases of MSIs (2 involving M. avium and 10 involving M. genavense) based on differences in at least 12 SNPs as suggested by Walker et al. (2013). Therefore, NTM MSIs seem to occur in a variety of animal species, but studies examining their prevalence and impact are few and far between.

Our review did not reveal the occurrence of MSIs within leprosy causing mycobacteria. However, using artificially co-infected armadillos, Shin et al. (2015) demonstrated that distinct strains of M. leprae could simultaneously exist in the same host. They noted that the in vivo growth rate of the non-armadillo strain was significantly higher in the absence of competing strains, suggesting that pathological variations exist between the different strain types. Due to the challenges associated with culturing M. leprae (Scollard et al., 2006), the same experimental approaches used to study MTBC and NTM members might not be feasible, thereby limiting the potential for detecting M. leprae MSIs. Therefore, with the advent of more sensitive and powerful discriminatory methods, it is possible that the use of culture-independent techniques such as metagenomics will help to shed light on MSIs involving M. leprae in the future.

Many questions are raised regarding the use of standard strain typing methods for delineating true MSIs from microevolution. When using IS6110-RFLP and MIRU-VNTR on M. tuberculosis isolates derived from the same host, MSIs are identified based on differences in specific DNA/PCR fragment profiles (Table 4). IS6110-RFLP profiles that differ by at least 2–3 bands indicate the occurrence of concomitant infection by distinct strains (Shamputa et al., 2004, 2006; Marx et al., 2014). Conversely, minor variations, i.e., anything <2 discriminatory bands, is considered as microevolution. In the case of MIRU-VNTR, MSIs are predicted based on heterogeneity at two or more loci (Shamputa et al., 2004, 2006; Cohen et al., 2011; Maghradze et al., 2019), whereas differences in a single locus suggests microevolution (Shamputa et al., 2004, 2006; Maghradze et al., 2019). Therefore, allelic diversity at more than one locus is a criterion for differentiating microevolution from true MSI when using MIRU-VNTR.

Defining MSIs using WGS based methods also has inherent challenges. However, the high discriminatory power of WGS allows for more definitive explanations for TB recurrence post-treatment (Bryant et al., 2013a; Guerra-Assuncaõ et al., 2015; Dippenaar et al., 2019). The classification of infections based on SNP differences between isolates from a single patient varies, where 5–10 suggest the occurrence of a relapse and anything over 100, a re-infection (Bryant et al., 2013a; Guerra-Assuncaõ et al., 2015; Dippenaar et al., 2019) (Table 4). These differences in SNP numbers are apparently set by taking into account the calculated evolutionary rate of M. tuberculosis, which ranges from 0.3 to 0.5 SNPs per genome per year, and limits the number of SNPs that can accumulate within strains in a defined period of time (Bryant et al., 2013b; Ford et al., 2013; Roetzer et al., 2013; Walker et al., 2013). Varying SNP thresholds have been used to define MSIs, where Bryant et al. (2013a) called them as such if isolates derived from the same patient had 80 or more different SNPs based on manual inspection. In comparison, Guerra-Assuncaõ et al. (2015) used a threshold of 140 heterogeneous SNPs to define MSIs based on an empirical cutoff generated during their data analysis. Furthermore, Dippenaar et al. (2019) did not define a specific threshold, as the numbers of heterogeneous SNPs in their study were either limited (1–2, classified as microevolution), or rather numerous (757–883, classified as MSI). These differences in SNP numbers used to identify M. tuberculosis MSIs shows that the field requires some form of standardization so that results can be compared between studies. It is possible that the lack of a clear definition for MSIs using WGS could hinder the identification of some reports for inclusion in this review. Although this is unlikely as only a few studies have used WGS technology for such purposes to date, and most of them were manually screened for their relevance.

Considering that the criteria necessary for identifying different types of M. tuberculosis polyclonal infections using IS6110-RFLP and MIRU-VNTR are well-defined, several studies were also found in the current review that reported on microevolution (Martín et al., 2007; Streit et al., 2015; Cohen et al., 2016; Jajou et al., 2018). The classification was not evident in some other reports that used IS6110-RFLP or MIRU-VNTR, due to results potentially indicating either an MSI or a microevolution event depending on the applied criteria (Al-Hajoj et al., 2010; Alves et al., 2011), or due to WGS not having a well-defined SNP threshold (Liu et al., 2015). While microevolution may initially appear to have little significance when compared to MSIs in terms of virulence and pathogenicity, one study found that the presence of highly evolvable repeats near genes (VNTR52, QUB26, or MIRU10/27) can influence gene expression in different M. tuberculosis isolates (Pérez-Lago et al., 2013). In addition, microevolution has obvious implication on the emergence of drug resistance due to selective pressures applied during treatment regimens (Fonseca et al., 2015).

Microevolution derived infections may not be as distinct as MSIs, but different clonal variants can still spread in unique patterns. While a strain undergoes a genetic drift, some of the progeny might spread both within and between hosts. For example, WGS identified separate M. tuberculosis clonal variants derived from within host microevolution at both respiratory and extrapulmonary sites in a patient (Pérez-Lago et al., 2014; Ssengooba et al., 2016). A study by Buff et al. (2009) examined several TB cases where IS6110-RFLP and spoligotyping of isolates from community transmission events showed identical patterns but exhibited variations at a single locus during 12-locus MIRU-VNTR analysis. It was shown that M. tuberculosis isolates from the source patient also displayed varying MIRU-VNTR profiles that matched the secondary patients, suggesting that different clonal variants were transmitted individually.

In addition to the MSIs described in our review, there is also evidence for the microevolution of mycobacterial pathogens in animals. For example, microevolution events and MSIs involving M. avium subsp. hominissuis in bongo antelopes were identified using 8-locus MIRU-VNTR and IS1245-RFLP (Moravkova et al., 2013). Another study used 8-locus MIRU-VNTR to examine M. caprae from 55 different animal hosts (including goats, cattle, sheep and wild boar) in Portugal and identified microevolution derived infections in 12 of them (Reis et al., 2020). Therefore, mycobacterial microevolution also occurs in animals, which is expected given the current state of knowledge regarding such infections in humans.