Consequences of genomic diversity in Mycobacterium tuberculosis

Abstract

The causative agent of human tuberculosis, Mycobacterium tuberculosis complex (MTBC), comprises seven phylogenetically distinct lineages associated with different geographical regions. Here we review the latest findings on the nature and amount of genomic diversity within and between MTBC lineages. We then review recent evidence for the effect of this genomic diversity on mycobacterial phenotypes measured experimentally and in clinical settings. We conclude that overall, the most geographically widespread Lineage 2 (includes Beijing) and Lineage 4 (also known as Euro-American) are more virulent than other lineages that are more geographically restricted. This increased virulence is associated with delayed or reduced pro-inflammatory host immune responses, greater severity of disease, and enhanced transmission. Future work should focus on the interaction between MTBC and human genetic diversity, as well as on the environmental factors that modulate these interactions.

1. Introduction

Tuberculosis (TB) in humans is mostly caused by the members of the Mycobacterium tuberculosis complex (MTBC) known as Mycobacterium tuberculosis sensu stricto and Mycobacterium africanum. MTBC are gram-positive acid-fast bacteria transmitted via aerosols generated by patients with pulmonary TB. The outcome of TB infection and disease is highly variable: exposure to MTBC can be followed by rapid clearance through innate immunity, direct development of active disease, or latent infection that may or may not re-activate up to several decades following initial exposure. Active TB disease comprises a range of presentations, including classical pulmonary TB, and various forms of extrapulmonary disease such as TB meningitis and miliary TB. Each of these different forms of TB feature a variety of symptoms that are associated with diverse host responses to the pathogen [1]. Traditionally, the different outcomes of TB infection and disease have been attributed to host and environmental variables [2]. Various human genetic determinants are known to influence the susceptibility to TB [3] (see also contributions by Meyer et al. and Bustamante et al. to this special issue). Environmental factors such as overcrowding and poor ventilation increase exposure to infectious particles [4], and implementing improved ventilation has been shown to reduce MTBC transmission [5]. Increasingly however, it is becoming clear that better knowledge of the bacterial determinants of virulence and their interaction with host and environmental factors will improve our understanding of the pathogenesis of TB [6].

Many experimental studies have provided evidence that clinical strains of MTBC differ in virulence (reviewed in [7–9]). However, unlike many other pathogenic bacteria like Corynebacterium diphtheriae, Escherichia coli O157:H7, Shigella dysenteriae, Vibrio cholerae, and Salmonella typhi, MTBC lacks canonical virulence factors such as, e.g., toxins. Hence in TB, determining the effects of strain-specific variation on infection and disease is challenging. Some of the limitations for discovering general patterns in genotype-phenotype associations has been the lack of phylogenetically robust classification systems for MTBC clinical strains [10]. Furthermore, identifying genetic determinants of virulence requires analytical methods that index genomic diversity in a more comprehensive manner as opposed to mere “genotyping”. During the past five years, advances in DNA sequencing technologies have made available many whole-genome sequences of MTBC clinical strains from around the world [11]. This has led to a better understanding of the global phylogenetic diversity of MTBC. Many genotyping schemes have been developed for MTBC in the past, (reviewed in [12]), but only comparative whole-genome sequencing (WGS) provides the phylogenetically robust framework for strain classification coupled with the sensitivity required to unmask the genetic particularities of different strains in detail. In addition, DNA sequence information can be further exploited using various comparative genomic and population genetic tools to predict the potential phenotypic impact of particular genetic polymorphisms [13,14].

In this review, we start by describing the latest insights from WGS data into the nature of MTBC genomic diversity, and provide a list of new databases and analysis platforms for WGS data relevant to MTBC. We then review recent attempts to predict the impact of MTBC genomic diversity on gene function and host immune recognition. We continue by summarizing the most recent experimental and epidemiological evidence supporting the relevance of MTBC genomic diversity for different virulence phenotypes, and end with an outlook on future research directions.

2. The nature of MTBC genomic diversity

2.1. Genotyping methods to classify MTBC clinical strains

During the last decades, various molecular techniques of DNA fingerprinting have been used to discriminate between clinical strains of MTBC (reviewed in [12]). Restriction fragment length polymorphisms (RFLP) typing is based on differences in copy number and the differential genomic location of the insertion sequence (IS) 6110, and became the first gold standard method for genotyping MTBC [15]. This technique has been used successfully to define chains of ongoing TB transmission, discriminate relapse from re-infection, and to detect laboratory cross-contaminations [16,17]. IS6110 is an IS element of 1,361 bp flanked by 28 bp inverted repeats which is differently inserted in the genome across strains. IS6110- RFLP analysis relies on digestion of genomic DNA with endonucleases followed by electrophoretic separation, blotting onto nylon filter, and probing of the restriction fragments with a IS6110-specific DNA fragment. Some of the limitations of IS6110-RFLP are that this technique is difficult to reproduce between laboratories, and that it requires large amounts of good quality DNA.

Because of these limitations, several PCR-based methods have been developed for MTBC strain typing that require little DNA. These include spoligotyping and Mycobacterial Interspersed Repeat Units (MIRUs) typing, which together have been recently defined as the new gold standard for molecular epidemiological investigation of TB [12]. Spoligotyping patterns are defined based on the presence or absence of 43 unique regions intercalated between direct repeats in the Clustered Regularly Interspaced Short Palindromic Repeats region (CRISPRs) of the MTBC genome. MIRUs profiles classify MTBC strains by the number of repeats at different Variable Number of Tandem Repeats (VNTRs) loci. Spoligo-types and MIRU-types can be compared using the SITVITWEB database that includes thousands of spoligotyping patterns [18] and MIRU-VNTRplus [19]. However, the use of spoligotyping and MIRUs is limited for phylogenetics and strain classification because of the propensity of the corresponding molecular markers for convergent evolution; i.e. because these markers change rapidly, the same or similar patterns can emerge by chance in strains that are phylogenetically unrelated [10].

To get around this problem, genomic deletions, often referred to as Regions of Difference (RDs) or Large Sequence Polymorphism (LSPs) have been used as markers to classify groups of MTBC strains into main phylogenetic lineages [20–24], and sub-lineages [25,26]. Although horizontal gene transfer between MTBC and M. canettii has been detected [27], the population structure within MTBC is largely clonal [2,20]. Because on-going horizontal gene exchange is rare in MTBC, LSPs are essentially irreversible, making them ideal phylogenetic markers for strain classification.

Following the completion of the first MTBC genomes [28–30], comparative genomics identified sets of phylogenetically informative single nucleotide polymorphisms (SNPs) that were used to establish various strain-typing methodologies [29,31–33]. In addition, several groups have developed SNP-typing schemes using multilocus sequencing analyses [34–36]. More recently, accumulating WGS data have led to the development of novel SNP-typing methods that rely on a broader understanding of the global MTBC phylogenetic diversity [11]. Some of these methods have been built into highly multiplexed assays [37–40]. In addition to SNPs, other forms of genetic diversity have been incorporated in high-throughput genotyping schemes, including spoligotyping data [41,42], and drug resistance-conferring mutations [43].

2.2. New software and databases to explore genomic MTBC diversity from WGS data

Despite the usefulness of the various genotyping assays discussed above [11], WGS remains the only tool that can classify MTBC strains robustly and simultaneously index genomic diversity at all levels, be it at the level of whole populations [44,45], within large outbreaks [46,47], during household transmission [48], within single patients [49,50], or during in vitro evolution [51–54]. Moreover, DNA sequence data can be used to measure phylogenetic distances, and hence quantify the amount of genetic diversity within and between groups of strains. Finally, WGS allows discovering new mutations in particular strains, or groups of strains, associated with particular phenotypes. While WGS is becoming cheaper and more widely available, the analysis of WGS data remains often limiting. Hence, rapid and user-friendly analysis methods are required. Recently, several new databases have been developed that make WGS data readily available. These platforms also include tools to visualize and analyse MTBC genomic diversity (Table 1). They contain multiple types of information on multiple MTBC genomes, including DNA and protein sequences, maps, assemblies, annotations, and bibliography, as well as gene expression and protein data associated with different genomic regions. Whilst the Tuberculist database [55] offers diverse information on the reference M. tuberculosis genome of the laboratory strain H37Rv, these new databases incorporate tools for comparative genomics, and include data from various MTBC clinical strains as well as other species within the genus Mycobacterium (Table 1). Most of these databases build on already available genomes. In addition, new tools have been developed recently that help explore unknown genomic diversity based on newly generated WGS data. For example, SpolPred predicts the spoligo-type from short DNA sequencing reads [56]. Similarly, KvarQ provides a robust SNP typing schemes using short DNA sequencing reads from WGS, and offers a robust classification in the main human- and animal-adapted lineages of MTBC. In addition, KvarQ identifies drug resistance-conferring mutations [57]. In summary, recent advances in WGS of MTBC clinical strains are revealing a larger amount of between-strain genomic diversity than generally appreciated (see below). Increasingly, it is also becoming clear that this genomic diversity translates into relevant phenotypic variation.

Table 1.

Databases and bioinformatics tools to explore MTBC genomic diversity

Tool type	Tool name	Input genomic data data	Variation investigated	Reference
Database	Tuberculist	H37Rv complete genome	Protein information, drug and transcriptome data, mutant and operon annotation, bibliography, structural views and comparative genomics	[55]
Database/tools	TBDB	9 complete genomes and 10 WGS	Genomic (SNPs, small indels), expression (RNA-seq and microarrays) and bibliography.	[228]
Database/tools	Patrik	101 complete genomes and 1899 WGS	Comparative genomics of mycobacteria, annotations, access to WGS data, metadata associated with genomes.	[229]
Database/tools	PolyTb	1500 GWS	SNPs, small indels and large deletions.	[230]
Database/tools	GMTV	1084 WGS	Clinical, epidemiological, microbiological and genome variations	[231]
Software	SpolPred	WGS:fastq	Spoligotyping	[56]
Software	SeqSphere	Gene sequences	Sequence types	[232]
Software	KvarQ	WGS:fastq	SNPs: drug resistant and phylogenetic markers	[57]

2.3. Differences between MTBC strains and lineages

The MTBC comprises various closely related bacterial species and sub-species, including M. tuberculosis sensu stricto and M. africanum which are adapted to humans, as well as several animal-adapted forms [58], i.e. M. bovis, M. caprae, M. microti, M. pinnipedii, M. origys, M. mungi, M. suricattae, the dassie bacillus, and the chimpanzee bacillus [59–62]. In addition to these classical members of MTBC, the Complex also comprises more distantly related bacteria known as M. canettii and other so-called “smooth tuberculosis bacilli (STBs)”, which are characterized by a smooth colony morphology (STB) [63–65]. The STBs exhibit several other important features, including strong evidence of horizontal gene transfer, which sets them apart from the other members of MTBC [66]. Only approximately 60 isolates of STB have been described so far, most of which from the Horn of Africa. Increasing epidemiological evidence suggests STBs are environmental organisms that only occasionally infect humans [67].

One of the first evolutionary reconstructions of the genetic population structure of the whole MTBC highlighted a group of strains harbouring a deletion in the genomic region known as TbD1 [68]. TbD1-deleted strains have been referred to as evolutionarily “modern” compared to the strains without this deletion, which collectively have been referred to as evolutionarily “ancestral” or “ancient”. Today, based on WGS analyses, we know that MTBC comprises seven human-adapted lineages (Lineage 1 to Lineage 7 in Figure 1A), where the “modern” clade form a monophyletic group comprising the TbD1-deleted Lineages 2, 3 and 4, as these lineages diversified more recently than the remaining MTBC strains. By contrast, the “ancestral” strains are paraphyletic, meaning they do not comprise a single phylogenetic group. Many studies have shown that the human-adapted MTBC lineages show a strong phylogeographical population structure, with the different lineages associated with distinct geographical regions [20,24,33–35,69,70]. Some of these lineages are also more globally widespread than others (Figure 1). Specifically, the most widely distributed groups are Lineage 2 and Lineage 4 (Figure 1B). Lineage 2 (also known as East-Asian lineage, includes the Beijing family of strains) predominates in East Asia, but is also present in Central Asia, Russia and South-Africa. Lineage 4 (also known as the Euro-American lineage) occurs frequently in populations from Asia, Europe, Africa and America. Lineage 1 and Lineage 3 show a more restricted geographical distribution limited to East Africa, Central-, South- and South-East Asia (Figure 1C). The most geographically restricted lineages are Lineage 5, Lineage 6 and Lineage 7, which are all associated with specific regions of Africa (Figure 1D). Lineage 5 and Lineage 6 are also known as M. africanum West Africa 1 and West Africa 2, respectively, and almost exclusively occur in West Africa or in recent immigrants from those regions [71]. Lineage 6 occurs primarily in the Western part of West Africa, whereas Lineage 5 dominates further to the East in regions bordering the Gulf of Guinea [71,72]. Similarly, the recently discovered Lineage 7 is confined to Ethiopia and recent immigrants from that part of the world. The reasons for why these three lineages are limited to specific regions of Africa are unknown [73,74]. Finally, two other lineages within the classical MTBC are adapted to different wild or domestic animal species (Figure 1A). One of these two lineages comprises the classical animal-associated strains M. bovis (includes the vaccine strains BCG), M. caprae, M. microti, M. pinnipedii and M. orygis. The other animal-adapted lineage includes the chimpanzee bacillus, and, although genome data are not yet available, likely also the dassie bacillus, M. mungi and M. suricatae [59–62]. Recently, a new group of MTBC has been identified in 1000-year old human remains from Peru [75]. These ancient MTBC strains were distinct from any known human-adapted MTBC, but most closely related to contemporary M. pinnipedii which is adapted to seals and sea lions. These findings suggest that marine mammals could have played a role in spreading TB from Africa across the Atlantic Ocean to the New World and transmitting to pre-Columbian human populations [75], Figure 1A). For the remainder of this review, we will focus on the classical human-adapted members of MTBC, i.e. the Lineages 1 to 7 depicted in colour in Figure 1.

Figure 1.

A. Maximum likelihood phylogeny modified from Bos et al. [75]. Node support after 1000 bootstrap replications is shown on branches and the tree is rooted by the outgroup M. canettii. Large Sequence Polymorphisms (LSPs) described in [68] are indicated along branches. Scale bar indicates the number of nucleotide substitutions per site. B, C and D. Dominant MTBC lineages per country. Each dot corresponds to 1 of 80 countries represented in the 875 MTBC strains from the global strain collection analysed by Gagneux et al. [24]. The yellow and an orange dot represent Lineage 7 in Ethiopia [74] and the extinct MTBC strains from Peru, respectively [75]: panel B shows the most geographically widespread lineages, panel C the intermediately distributed lineages, and panel D the most geographically restricted lineages.

Strains of MTBC differ in their content of SNPs, small insertion and deletions (indels), large genomic deletions, large duplications and insertion sequences. Unlike MIRU and spoligotyping patterns, LSP are robust makers for phylogenetic classification, but LSPs are not polymorphic enough to differentiate among closely related strains, e.g. within an outbreak or transmission chain. By contrast, WGS reveals all types of mutations, and provides the best discriminatory power to differentiate between strains. Moreover, WGS allows computing phylogenetic distances and quantifying genomic diversity within and between groups of strains. To obtain an initial estimate of within MTBC diversity, we have used previously published whole genome sequences of 217 globally distributed clinical strains [76] and calculated the number of SNPs between any pair of strains. On average, two human-adapted MTBC strains differed by about 1,200 SNPs, which corresponds to 0.03% of the genome when excluding repetitive sequences (Figure 2). By contrast, the corresponding difference between any classical MTBC strain and M. canettii was about 2.7 % [65], which is 90 times larger than the average SNP-distance among human-adapted members of MTBC. Next, we calculated the average SNP distances within and between human-adapted MTBC lineages (Figure 2). The geographical origin of strains included in this analysis was diverse, and except for Lineage 7 that is only found in Ethiopia and Ethiopian immigrants, every lineage included strains from various countries. We found that Lineage 1 harboured the largest genetic diversity with an average of 730 SNPs between any two strains belonging to this lineage. The corresponding average distance was lowest for Lineage 7 with only 230 SNPs. In terms of between lineage diversity, the strains belonging to the “modern” Lineages 2, 3 and 4 differed by 970 SNPs in average. Strains belonging to the “ancestral” Lineages 1, 5, and 6 were more distantly related with an average of 1,500 SNPs between them. The maximum SNP distance of 1,800 SNPs was observed between strains of Lineage 7 and either Lineage 1, 5, or 6 (Figure 2). Although these estimates might change when more genomes are considered, they provide a first indication of the relative genomic distances within and between the different human-adapted lineages of MTBC.

Figure 2.

Number of pairwise difference between MTBC strains. The alignment of 217 human-adapted MTBC clincial strains published previously [76] were used to calculate the number of SNPs between any two strains (i.e. the SNP-distance). We calculated the SNP distance among each pair of strains includeding 44 clinical strains belonging to Lineage 1, 37 strains of Lineage 2, 36 strains of Lineage 3, 64 strains of Lineage 4, 16 strains of Lineage 5, 16 strains of Lineage 6 and 4 strains of Lineage 7. The results are shown in a box-plot generated with R grouping pairwise SNP-distances within each lineage (number of pairwise comparisons were Lineage 1: N=946, Lineage 2: N=666, Lineage 3: N=630, Lineage 4: N=2,016, Lineage 5: N=120, Lineage 6: N=120, Lineage 7: N=6), within “modern” lineages (6,274 pairwise comparisons), between Lineage 7 and Lineages 1, 5, 6 (75 pairwise comparisons), and 12,825 other inter-lineage comparisons.

3. The consequences of genomic diversity in MTBC

3.1. Predictions of the impact of MTBC genomic diversity

In addition to exploring and quantifying the genomic diversity in MTBC, WGS data has been used together with computational methods to predict the impact of this variation on the bacterial phenotype. One of the striking observations has been that contrary to many other organisms, in MTBC about two thirds of SNPs in coding regions are non-synonymous (i.e. amino acid changing) [29,35,76,77]. Moreover, using sequence data from 89 genes in 107 MTBC strains, Hershberg et al. [35] observed that 58% of the non-synonymous mutations fell in positions that were highly conserved in other mycobacteria, suggesting that most of these mutations in MTBC might have functional consequences. More recently, Rose et al. used WGS data to identify all SNPs specific to the different human-adapted lineages of MTBC; these are SNPs that are fixed in the corresponding bacterial populations [78] (Figure 3A). The number of fixed SNPs (i.e. synonymous and non-synonymous) ranged from 124 in Lineage 2 to 698 in Lineage 5, but in all cases, at least 44% of the non-synonymous SNPs fixed in one or the other lineage were predicted to impact gene function (Figure 3B). Taken together, these findings illustrate that even though the overall genomic diversity of MTBC is low compared to other bacteria [79], a large proportion of the mutations that have accumulated in the different phylogenetic lineages of MTBC are likely to lead to phenotypic differences.

Figure 3.

Predicted functional impact of lineage-specific SNPs. A. Neighbour-joining phylogeny based on 28 globally representative MTBC strains, using 13,086 variable positions [227]. The six main lineages are named and branches coloured as reported previously [24,35]. The number of lineage specific SNPs are indicated along the main braches. B. Percentage predicted functional nonsynonymous SNPs per lineage based on the prediction algorithm SIFT [78].

An alternative way to predict the functional consequences of mutations is to determine the evolutionary conservation of the corresponding genomic regions and the selection pressures in response to which these mutations evolve. The importance of a protein residue is reflected in its evolutionary conservation [80]. For example, loci coding for important general functions will be under purifying selection and thus highly conserved. Alternatively, functional innovations might be under positive selection, and hence detecting positively selected sites in genomes can point to adaptive processes. One of the strongest on-going selection pressures in MTBC has been drug pressure, and detection of positive selection has been used successfully to identify molecular markers of drug resistance [81–83]. One particular signature of positive selection is the independent multiple appearance of the same mutation; a phenomenon referred to as convergent evolution. Convergent evolution leading to so-called homoplasies are generally uncommon in MTBC, except for mutations involved in drug resistance [52]. Recently, signatures of positive selection have also been detected in membrane proteins of MTBC, suggesting a role of the corresponding mutations in the interaction with the host immune system [83].

WGS data can also be used to explore the putative impact of host recognition in MTBC. Many pathogens evade host immunity by means of antigenic variation [84]. Intriguingly, a comparative WGS analysis of 21 MTBC clinical strains from global sources revealed that the majority of 495 experimentally confirmed human T-cell epitopes in these strains were highly conserved [77]. The conservation of T-cell epitopes in human-adapted MTBC might reflect the fact that the host immune response is largely responsible for the lung damage during TB disease (i.e. cavitation), which increases patient coughing and hence contributes the successful transmission of the pathogen. Indeed, TB patients with cavitary disease are more likely to generate secondary cases [85]. These findings have implications for the development of new TB vaccines and diagnostics, as highly conserved antigens might represent ideal diagnostic targets, but not adequate vaccine components if the immune responses they elicit benefit the bacteria rather than the host.

3.2. The phenotypic consequences of genomic variation

Four years ago, we reviewed the available experimental and epidemiological evidence for strain-specific phenotypic diversity [7]. Since then, additional studies have come out providing additional support for the view that strain variation in MTBC has important phenotypic consequences. Different MTBC strains and lineages differ in their growth rates in liquid culture [86,87], in monocyte derived macrophages [87], and in mice [88,89]. MTBC strains and lineages also differ in their gene expression [78,88] and metabolic profiles [90]. A recent clustering analysis derived from the differential use of various substrates separated strains according to three major lineages [90]. Most recently, Portevin et al. reported significant differences in mycolic acid profiles between different MTBC strains and lineages [91]. Mycolic acids constitute the most abundant cell wall lipid in MTBC, and play an important role in the host immune response [92]. They also help the bacteria resist oxidative stress [93].

One of the challenges in studying the effect of MTBC strain variation has been linking genotype to phenotype. As mentioned above, a wide range of mutations occur across MTBC genomes including SNPs, deletion, duplications and mobile elements, all of which can have an impact on bacterial phenotypes. In the next few paragraphs, we review recent studies that have managed to link a given mutation in MTBC clinical strains to a particular phenotype.

Single nucleotide polymorphisms

SNPs are natural polymorphism that can be classified as synonymous, non-synonymous, nonsense, or intergenic. Because non-synonymous SNPs change the amino acid content of the corresponding protein, they are the principal contributors to functional mutations. A special category of SNPs which are better referred to as (de novo) single nucleotide mutations occur in genes associated with resistance to anti-TB drugs. These mutations decrease the susceptibility to a particular drug by either modifying the drug-target, increasing the expression of the gene product targeted by the drug, or by reducing the drug activation in the case of prodrugs [94,95]. Other non-synonymous SNPs have been shown to impact on the MTBC phenotype irrespective of drug resistance, with the PhoPR two-component system providing a particularly illustrative example. The PhoPR components are essential for MTBC virulence in animal models of TB (reviewed in [96]), and several point mutations in either component of this system have been shown to lead to important changes in bacterial phenotype. For example, one amino acid change at position 219 of PhoP in the laboratory strain H37Ra alters the binding capacity of PhoP to its own promotor [97]. As a result, H37Ra is highly attenuated compared to its virulent counterpart H37Rv. Similarly, one amino acid change at position 71 of PhoR occurs naturally in all MTBC strains belonging to Lineage 5 and 6, and in all animal-adapted MTBC strains. This mutations was shown to affect the PhoPR regulatory system, and consequently, the synthesis of important surface lipids and the secretion of the 6-kDa antigenic target ESAT-6 were reduced in the corresponding strains [98]. Non-synonymous SNPs can also lead to epigenetic changes. Recently, Shell et al. described one non-synonymous SNP in the methiltransferase mamA in Lineage 2 and another non-synonymous SNP in the active site of hsdM in Lineage 4 that lead to variation in adenine methylation levels in strains belonging to these lineages [99].

Compared to non-synonymous SNPs, intergenic and synonymous SNPs are often referred to as evolutionary neutral because on average, their effect on phenotype, and hence fitness, is less. However, some synonymous SNPs do have important phenotypic effects. For example, a synonymous SNP within the gene immediately upstream of DosR has been shown to generate an alternative internal transcriptional start site (TSS) in Lineage 2/Beijing strains. This TSS has been associated with increased expression of the dosR regulon [78]. Similarly, a synonymous SNP upstream of malQ in Lineage 1 strains has been found to create a new TSS associated with increased expression of malQ [78]. Finally, a recent study reported a synonymous SNP leading to a new internal TSS in mabA which was demonstrated to confer isoniazid resistance in MTBC clinical strains, representing a novel mechanism of resistance to this important first line anti-TB drug [100]. Additionally, SNPs in non-coding regions can have important functional consequences. Non-coding SNPs in promoter regions can modify the promoters and therefore alter transcriptional regulation. For example, several intergenic SNPs in the promoter of the inhA gene, increase the transcription level of inhA, thereby conferring resistance to isoniazid and ethionomide [101,102].

Finally, nonsense mutations result in a premature stop codon that produce truncated peptides with a likely impact on the functionality of the product encoded; this process is also referred to as pseudogenization. Unlike M. leprae where almost half of the coding DNA sequences are pseudo-genes [103], pseudogenization in MTBC seem to be a more recent process and not as significant as in M. leprae [104]. However, some nonsense mutations are found within MTBC, where Lineage 6 and the M. bovis clades contain more pseudo-genes than other lineages [104]. Many of these nonsense mutations are in genes that have analogous genes or pathways in the MTBC genome and presumably, these mutations will have a minor phenotypic impact due to genomic redundancy [104].

Gene Duplications

Gene duplications and the expansion of gene families are important sources of genetic diversity with the potential to lead to evolutionary innovations [105]. Duplications of the ESAT-6 gene clusters in an ancestor of MTBC resulted in the expansion of the PE/PPE gene family [106]. PE/PPE genes have been postulated to play a role in antigenic variation [107–109]. Indeed, these genes are highly polymorphic [110–114], but the evolutionary forces driving this diversity remain unknown. Nevertheless, differences in nucleotide variation and gene expression in some PE/PPE genes have been associated with virulence [115,116].

Although there is currently no evidence of large duplications being a major source of genomic diversity within MTBC, several instances have been reported. A 350kb genomic duplication that includes the DosR operon has been observed in some Lineage 2 strains. This duplication has been suggested to be partially responsible for increased expression of the dosR regulon in Lineage 2/Beijing strains discussed above [117]. However, this duplication is not seen in all Lineage 2/Beijing strains [78], possibly because the duplicated regions is lost during in vitro cultivation [117]. Interestingly, in addition to some Lineage 2 strains, some strains belonging to Lineage 4 have been shown to carry independent duplications spanning the same genomic region [118]. As outlined before, the presence of convergent evolution events in MTBC is rare [52]. Therefore, the convergent duplication of dosR regulon might indicate some evolutionary advantage of strains harbouring this duplication.

Large scale duplications might also have played a role during the in vitro evolution of BCG strains. Two tandem duplications termed DU1 and DU2, of 29,668 bp and 36,161 bp, respectively, are present in M.bovis BCG Pasteur compared to H37Rv [119]. Intriguingly, DU2 showed four alternative forms evolved in the different laboratories where the different BCG strains were passaged, leading to independent duplications of similar genomic regions [120]. If and how these duplications impact BCG phenotypes, including vaccine efficacy, remains unknown [54].

Repetitive and mobile genetic elements

Repetitive elements such as CRISPs, VNTRs and ISs have been used as molecular markers for MTBC strain genotyping (discussed above and reviewed in [12]). These elements are an important source of genomic variation that could impact bacterial phenotypes.

The CRISPR region in MTBC is known to be polymorphic and shows convergent deletions in phylogenetically unrelated strains [121]. However, the functional role (if any) of polymorphisms affecting to the CRISPR region in MTBC, or the impact of CRISPR polymorphisms on mycobacterial phenotypes have not been determined.

In humans, variation in some VNTR loci is associated with complex diseases such as type-1 diabetes [122]. In bacteria, VNTRs have been implicated in the ON/OFF switching of phase variable genes in Neisseria [123,124]. In mycobacteria, intragenic VNTR loci variation has been shown to modify the structure and function of the proteins affected [125]. Intergenic VNTR variation can alter promoter activity in downstream genes [126–128]. For example, an increased number of repeats in the VNTR3239 locus of MTBC leads to a higher expression of the downstream gene fpg1, a DNA glycosylase shown to be involved in the protection against oxidative DNA damage [126].

Finally, differential insertion of the IS6110 sequences in MTBC clinical strains has been shown to lead to important phenotypic effects linked to changes in gene expression. When inserted in the upstream region of a gene, the insertion sequence can modify the transcription of neighbouring genes by acting as a promotor [129–131]. In addition, insertion of IS6110 can disrupt genes. Indeed, the presence of independent disruptions of the same gene in different clinical strains, i.e. another example of convergent evolution, might be associated with an evolutionary advantage for the corresponding strains [132]. For example, one hyper-virulent M. bovis strains that caused an outbreak of human TB in Spain showed an IS6110 insertion upstream of the phoP regulon [98]. The expression of the phoP regulon is generally diminished in animal strains and Lineage 6 strains due to the non-synonymous SNP in phoP mentioned above. The insertion of IS6110 upstream of phoP regulon in this hyper-virulent M. bovis restores phoP transcription, compensating for the effect of the non-synonymous SNP [98]. M. bovis is generally not an efficient pathogen in humans, among which it rarely transmits. However, this single IS6110 insertion seemed to have transformed an MTBC strain adapted to cattle into a variant capable of sustaining a transmission cycle in a new host (i.e. humans).

Genomic deletions

Large genomic deletions are a substantial source of genomic diversity within MTBC. Genomic deletions can result from transposition of mobile genetic elements, like in the case of the prophages phiRv1 (linked to RD3) and phiRv2 (linked to RD11), and insertion sequences IS1532 (e.g. in the case of RD6) or IS6110 (e.g. in the case of RD5) [68,133]. Deletions can also be mediated by homologous recombination between adjacent IS6110 resulting in the loss of the intervening DNA segment; examples for this mechanism include RvD2, RvD3, RvD4, and RvD5 [68]. In addition, many genomic deletions occur through unknown mechanisms [23].

Deletion of one or several loci can greatly influence the bacterial phenotype. One of the most relevant MTBC phenotype is drug resistance, and deletions have been implicated in drug resistance phenotypes. The presence of repetitive sequences near the katG region of the M. tuberculosis genome makes this gene prone to deletions [134], and deletions of katG have been shown to confer high level resistance to isoniazid [135].

Other deletions have been reported to lead to the modification of the lipid composition of the mycobacterial wall, which can therefore alter the bacterial interaction with host cells. A deletion disrupting the locus of the polyketide synthase (Pks)15/1 in strains belonging to Lineage 4 results in a defective production of phenolic glycolipid (PGL) [136], thereby affecting the bacterial interaction with host immune cells [137]. Indeed, the absence of PGL has been linked to reduced MTBC virulence in infection models [138,139]. Similarly, a deletion specific to all Lineage 3 strains involves Rv1519, and has been linked to a decrease in the production of the anti-inflamatory cytokine IL-10 by in infected hosts [140]. Genomic deletions have also been associated with clinical phenotypes. For example, patients with extra-pulmonary TB were more likely to carry MTBC strains with a deletion in the phospholipase C-encoding gene plcD [141]. Moreover, a deletion of the embR locus has been detected in a particularly successful multidrug-resistant strain of MTBC that has successfully spread between continents (Coscolla et al., manuscript submitted). EmbR is a transcriptional regulator implicated in ethambutol resistance, as well as in the regulation of lipomannan/lipoarabinomannan ratio [142], which serves as immunomodulator crucial for mycobacterial virulence (reviewed in [143]). Finally, deletions in PE_PGRS33 have been associated with reduced induction of tumour necrosis factorα (TNF-α) by the host [144], reduced patient clustering (indicating reduced transmission), and absence of lung cavitation, supporting the view that PE_PGRS33 plays a role in the transmission success of MTBC in clinical settings [145].

3.3. New evidence supporting differences in virulence and immunogenicity between MTBC clinical strains

Following the 100 studies we reviewed four years ago [7], many new studies have explored the effect of MTBC strain variation on virulence and immunogenicity. Human-adapted MTBC is an obligate pathogen in the sense that i) it has no other animal or environmental reservoir, and ii) it has to cause (pulmonary) disease to transmit successfully. In other words, virulence is directly linked to transmission [85,146], which is unlike many other pathogens where transmission occurs independently of disease. “Virulence” in TB can be conceptualized as a composite comprising i) the ability of the bacteria to survive in face of the host immune responses, ii) their capacity to cause lung damage, iii) to survive the aerosolisation process outside of the host, and iv) successfully transmit to and infect a new host. In the following sections, we review recent studies that have generated evidence supporting strain differences with regards to these individual components of “virulence”.

Impact on host immune regulation

Studies in humans and animal models have shown that TNF α, interferon-g (IFN-γ), IL-12 and IL-17 are important mediators of a protective immune response against TB (reviewed in [1,147]). Other mediators such as IL-10 may play a role in limiting MTBC clearance during the early immune response [148]. Considering the link between virulence and transmissibility in MTBC, one could postulate that a more severe inflammatory response causing more lung damage would lead to more efficient transmission (further discussed in [147]). Intriguingly however, many studies have found increased MTBC virulence associated with reduced and/or delayed inflammatory responses, perhaps by allowing a stronger bacterial proliferation early during the infection process, leading to increased virulence at a later stage. For example, studies in different infection models have found strain NH878 (belongs to Lineage 2/Beijing) constantly associated with a delayed inflammatory immune response and increased virulence [149–156].

At the lineage level, the so-called “modern” Lineages 2, 3 and 4, showed a lower early inflammatory response compared to Lineage 1 and Lineage 6 [157]. As discussed under Section 2.3. above, “modern” MTBC strains are more globally widespread than other lineages. Hence, the observation that “modern” strains are associated with a delayed inflammatory response (i.e. higher virulence) might be linked to the global success of these strains [6]. Indeed, a study by Reiling et al. [158] showed that “modern” strains were replicating faster in vitro, in human monocyte-derived macrophages, as well as in aerosol-infected mice. Several other studies have demonstrated the reduced/delayed pro-inflammatory response of the “modern” versus other lineages [159]. By contrast, Krishnan et al. [89] showed increased inflammatory responses in Lineage 1 strains compared to Lineage 4, but these were not significantly higher than in Lineage 2 strains.

Other studies have compared the different lineages within “modern” lineages. One study found that Lineage 3 exhibited a higher anti-inflammatory phenotype compared to Lineage 4 [140,160,161]. Wang et al. [162] reported that different Lineage 2/Beijing strains commonly induced lower levels of TNF α, IL-6, IL-10 and GRO-α compared to strain H37Rv in monocyte-derived macrophages and dendritic cells. Similarly, Sarkar et al. [87] showed that Lineage 2 exhibited a lower pro-inflammatory phenotype compared to Lineage 3 and Lineage 4. Some studies however, showed the contrary, Lineage 4 eliciting more TNF α than Lineage 2; this was particularly true when compared to a sub-lineage of Lineage 2 known as “modern Beijing” [163]. Moreover, Krishnan et al. [89] did not detect any differences between Lineage 2 and Lineage 4. These contradictory findings with respect to the host immune responses elicited by Lineage 2 and Lineage 4 could on the one hand be due to differences in experimental conditions. On the other hand, sub-lineages within the main lineages, as well as individual strains within any given lineage might vary in the inflammatory responses they elicit.

Because the main MTBC lineages are not genetically homogeneous (Figure 2), differences in immune regulation among strains are expected at sub-lineage level [157]. Such differences have been most widely studied in Lineage 2, where “modern Beijing” showed a lower inflammatory response compared to so-called “ancient Beijing” [159,163]. Yet, Wang et al. [162] did not find any differences between Beijing strains. Some of the variable cytokine profiles elicited by the distinct groups of strains within Lineage 2 have been linked to differential toll-like receptor recognition [164]. Only one study has investigated intra-lineage variation within Lineage 4 but found no difference between the so-called H and T families of strains [160].

Increased “virulence” surely entails more than a mere delay in early pro-inflammatory response. Indeed, Reiling et al. [158] reported lineage-specific differences in virulence profiles based on variable bacterial uptake by host cells, differences in cytokine induction, and intracellular growth. In agreement with other studies, the “modern” Lineages 2 and 4 showed high replicative potential compared to “ancestral” Lineages 1 and 6. However, Lineage 2 was characterized by low uptake, and low cytokine induction, whereas Lineage 1 and Lineage 4 exhibited high uptake and higher cytokine induction.

To date, few studies have combined host and bacterial factors when studying the different immune response during MTBC infections in humans. Coussens et al. [165] showed that ethnicity plays an important role in the different inflammatory profiles in Africans versus Eurasians from a population in London, while MTBC lineages did not contributed significantly to the different immune responses in this patient population.

Impact on Disease Severity

An alternative way of measuring “virulence” is to look at the severity of the disease. Stavrum et al. [166] showed that TB patients from Tanzania infected with “modern” Lineage 4 strains showed more α1-acid glycoprotein and C reactive protein, higher neutrophils counts, and a lower body mass index than those infected with Lineage 1. De Jong et al. [167] showed that in the Gambia, individuals infected with modern Lineages 2 and 4 were more likely to progress to active disease compared to individuals infected with Lineage 6. The higher virulence of the modern Lineage 2 compared to “ancient” Lineage 6 was corroborated in a novel marmoset model of infection [168]. Infection with Lineage 2 induced more rapid weight loss, and led to a higher bacterial load in liver, spleen and lymph nodes [168]. Finally, several studies have found Lineage 2 associated with relapse [169–171], treatment failure [172], and fever early during treatment [173].

Within the “modern” lineages, Lineage 2 is generally associated with higher “virulence” than Lineage 4. For example, Orgarkov et al. [174] found that patients carrying a polymorphism previously associated with TB (CD209_336 A/G) who were infected with Lineage 2 were more likely to die of TB compared to patients infected with other strains. Nahid et al. [175] found during a clinical trial that patients infected with Lineage 2 strains and one sub-group of Lineage 4 were more likely to yield a positive culture at week eight after treatment initiation when compared to other Lineage 4 strains.

Similar to the observed differences in immune responses caused by different sub-lineages described above, within-lineage differences in disease presentation have also been reported. Kato-Maeda et al. [176] reported that a group of highly transmissible Lineage 2 strains from San Francisco were more virulent than other Lineage 2 strains in guinea pigs. Likewise, Aguilar et al. [177] demonstrated that highly transmissible Lineage 2/Beijing sub-lineages from South Africa were more virulent in mice than less transmissible strains. Taken together, the observation that highly transmissible strains tend to cause more severe disease in animal models is consistent with intrinsic bacterial features linking “virulence” with pathogenicity, transmission, and overall strain fitness.

Impact on disease presentation

Besides the classical pulmonary presentation, MTBC can cause a wide variety of extra-pulmonary manifestations affecting many organs of the human body [178]. Pulmonary and extra-pulmonary TB can also occur simultaneously. Because efficient MTBC transmission relies on lung damage, bacterial genotypes that are more prone to cause pulmonary and cavitary disease will be able to transmit more efficiently. Experiments in the marmoset model have revealed more extra-pulmonary spread to the lymph node, liver and spleen in animals infected with a Lineage 6 clinical strain compared to animals infected with the Lineage 4 strain CDC1551 [168]. Importantly, these experiments were conducted using genetically identical individuals (i.e. marmosets always give birth to twins or triplets), suggesting that the different disease presentation associated with the different MTBC strains are due to strain rather host genetic differences.

In humans, epidemiological studies have been rather inconsistent with regards to MTBC strain or lineage effects on disease presentations. At least five studies have reported such effects. Kong et al. reported an associating between Lineage 2 and extra-pulmonary TB [179,180]. Another study found a similar association when comparing Lineage 2 to Lineages 3 and 4 [181]. By contrast, Click et al. found Lineages 1 and 3 associated with extra-pulmonary TB when compared to Lineage 2 [182]. A study in Vietnam found Lineage 4 associated with pulmonary rather than meningeal TB [183]. On the other hand, at least five studies failed to find any association between MTBC lineage and disease presentation [74,173,184–186].

Impact on transmission

Rates of MTBC transmission are generally inferred by comparing genotypic clustering between patient isolates from a given epidemiological setting [187]. Another proxy for inferring successful transmission of particular MTBC genotypes is measuring increases in the frequency of these genotypes over time in patient populations. Finally, genotypes associated with younger patient age have also been interpreted as reflecting successful transmission, because TB in young patients are more likely to reflect ongoing transmission as opposed to reactivation [188].

Based on these concepts, several epidemiological studies have supported the view that overall, strains from “modern” lineages are more transmissible than other MTBC strains. Buu et al. [189] reported higher genotypic clustering of Lineage 2 compared Lineage 1 in Vietnam. Similarly, a study in Shanghai found Lineage 2 associated with higher clustering and younger age compared to other strains [190]. Many other studies in various settings have reported a higher fitness of Lineage 2/Beijing strains reflected by increases in their frequency over time [69,159,191–196]. In some cases, the increase of Lineage 2 was associated with drug resistance [189,197]. Yet, several studies observed no such increase and no enhanced transmissibility of Lineage 2 strains [167,198–201]. Some of these contradictory findings could be due to differences in the study populations. For example Marais et al. [200] did not find any association between Lineage 2 and higher transmission among children of South Africa. Alternatively, if sub-lineages differ in transmissibility, the genetic heterogeneity within lineages would lead to inconsistent results among different study settings where sub-lineages differ in prevalence. In support of this notion, Kato-Maeda et al. showed that one sub-lineage within Lineage 2 showed higher genotypic clustering in San Francisco [202]. This sub-lineage was also more virulent in guinea pigs, supporting again an intrinsic bacterial role in increased pathogenicity linked to enhanced transmission [176]. Sub-lineages of other lineages also tend to differ in their transmissibility as observed in San Francisco, USA for Lineage 4 [203] and Cotonu, Benin for Lineage 5 [72]. However, three studies have reported decreasing prevalence of M. africanum (i.e. Lineage 5 and 6) in Cameroon, Guinea-Bissau and Burkina Faso [204–206], supporting of the lower “virulence” of the “ancestral” lineages. Therefore, although sub-lineage differences might obscure general patterns, taken together, these studies support the view that on average “modern” MTBC lineages are more transmissible than other strains. Yet, Albanna et al. [199] observed reduced transmissibility of “modern” Lineage 3 compared to other lineages in Montreal, albeit with no difference between Lineage 1, Lineage 2 and Lineage 4.

4. Conclusions

MTBC has for many years been regarded as a “clone”, harbouring too little genetic diversity to be worth considering. This paradigm has now changed, and many experimental and epidemiological studies have demonstrated the phenotypic impact of MTBC strain diversity. During the last four years alone, at least 52 new studies have explored phenotypic differences among strains and lineages of MTBC, in addition to the 100 studies we reviewed earlier [7]. Although clinical studies have reported discordant results regarding the propensity of MTBC lineages to cause pulmonary as opposed to extra-pulmonary disease, studies focusing on other proxies of “virulence” have detected common patterns showing that “modern” lineages are generally more virulent and more globally successful, compared to other more geographically restricted lineages (Figure 1). Differences in immunogenicity, severity of disease, and transmission consistently indicate that Lineage 2 and Lineage 4 are more virulent than Lineages 1 and 6. More work is needed with respect to Lineage 3, 5 and 7, but given the restricted geographical distribution of these latter lineages (Figure 1), chances are that at least Lineage 5 and Lineage 7 will prove to be low virulence variants. Consistent with previous reports [8,207], many studies have noticed an emergence of Lineage 2/Beijing over time. This increase of Lineage 2/Beijing might be partially driven by environmental factors such as antibiotic treatment (reviewed in [208]) or BCG vaccination [209–211]. In addition, this effect could also be driven by changes in host demography. Specifically, the human population expansions during the Neolithic- and Industrial Revolutions have been hypothetically linked to an increase in virulence of some MTBC lineages [6,75,76].

5. Outlook

In this review, we primarily focused on the genetic diversity of MTBC, i.e. the pathogen. As discussed in the contributions by Meyer et al. and Bustamante et al. to this special issue, human genetic variation plays an important role in TB. Considering the long-term association between MTBC and its human host [212], some degree of co-evolution is likely to have occurred [20,24,33–35,69,70,213,214]. Indeed, several recent studies have shown that the susceptibility to TB is influenced by variation in both the pathogen and the host [183,215–217]. Hence, the interactions between host and pathogen genomic diversity needs to be explored and better understood using novel approaches [218,219]. Moreover, the various environmental variables modulating these interactions will have to be considered [2]. A particular challenge will be predicting phenotypes or evolutionary responses in the context of changing environments [220]. MTBC faces many different environmental conditions during its life cycle. These are encountered when penetrating into host cells, surviving within granuloma and cavities, persisting during aerosol transmission, and resisting drug pressure. Recently, a new environment has emerged to which MTBC has to adapt to: HIV co-infection [221]. Epidemiological studies have reported associations between HIV and Lineage 2 [222–224] and Lineage 6 in co-infected TB patients [225]. But how HIV co-infection as a selective pressure might impact the evolutionary trajectory of MTBC on the long run is unclear [146]. All of these complexities call for integrated approaches based on systems biology [226] and systems epidemiology [6] to improve our understanding of the role of variation in the host, the pathogen, and the environment in TB. Such an improved understanding will likely pave the way for novel tools and strategies to better control this important disease.

Highlights.

• Human-adapted MTBC comprises seven phylogenetic lineages

• MTBC strains belonging to separate lineages differ by 1,200 single nucleotide polymorphisms on average

• MTBC lineages are associated with different geographical regions

• Lineage 2 (includes Beijing) and Lineage 4 are more geographically widespread than other lineages

• Lineage 2 and Lineage 4 are more pathogenic and transmissible than geographically restricted lineages