Global variation in bacterial strains that cause tuberculosis disease: a systematic review and meta-analysis

We identified studies by systematically searching PubMed and Scopus. The first search was run on June 8, 2017, and the final search was run on November 13, 2017. Articles identified in these searches were supplemented by six published studies in Papua New Guinea, India, Botswana, Nepal, Ethiopia, and Kenya, and two unpublished studies in Mexico and Panama, which we were directed to by manually checking the reference lists of studies and by reviewing the conference abstracts for the 48th Union World Conference on Lung Health. Complete details of search strings and dates searched are found in Additional file 1.

We identified 206 studies for inclusion in the study, representing over 200,000 bacterial isolates collected over 27 years in 85 countries. Of these studies, 30 also reported transmission chains and 4 reported treatment failure associated with genotypes. Figure 1 shows the PRISMA flow diagram detailing the study selection process. Additional file 1: Figure S1 shows a map of the numbers of studies per year that were included from each country.

The 206 studies included 42 nationally representative samples and 164 samples representative of smaller geographic units. These included 34 studies that used a cluster or random sampling, 170 that collected samples from all reported or new TB cases in a given geographic location and time period, and 2 studies that used different sampling methods for different time periods. We illustrated how these study designs varied globally in Fig. 2. Sub-Saharan Africa was dominated by subnationally representative studies, while Caribbean Latin America was dominated by nationally representative studies (Fig. 2). In addition, we calculated the proportion of estimated prevalent TB cases that were represented in each country. Proportions ranged from 0.0012% in Nigeria to 5.4% in Greenland (Fig. 2). In general, the proportions were lower in countries where TB burden was highest (Fig. 2). The meta-data linked with each individual study is available in its raw format in Additional file 3.

We mapped the distribution of MTBC genotypes identified in our systematic review across all years and for all locations in each country for which we had data. A striking feature of the map was the widespread global distribution of Euro-American lineage 4 (Figs. 3 and 4). Lineage 4 was identified in every country where genotyping data was available for inclusion, and it was the majority lineage in 52 of the 85 countries (Additional file 1: Table S3). Our map also showed the fairly widespread distribution of East Asian lineage 2 (Figs. 3 and 4), which was identified in 67 of the 85 countries and was the majority lineage in 6 countries (Additional file 1: Table S3). In contrast, West African lineages 5 and 6 were identified in only 30 countries and were the majority lineages in zero countries (Additional file 1: Table S2). In addition, Indo-Oceanic lineage 1 and East African-Indian lineage 3 were identified in 64 and 59 countries, respectively, and each was the majority lineage in 2 countries (Additional file 1: Table S3).

The map also illustrated various regions of distinct mycobacterial distribution that may be independent of geopolitical country boundaries (Fig. 3 and Additional file 1: Figure S2). For example, Eastern Africa from Sudan to Mozambique was distinct from the rest of Africa in that it had a higher prevalence of lineages 1 and 3. Western Africa was distinct in that it had the highest prevalence of lineages 5 and 6, while Southern Africa had the highest prevalence of lineage 2 strains and Central Africa had the highest prevalence of lineage 4 strains. In addition, the Indian subcontinent and Australia had a similar genotype distribution, which was distinct from Russia and Eastern Asia. The UK was distinct from the rest of Europe in that it had a greater prevalence of lineages 1 and 3. Finally, Central America and northern South America had distinct genotype distributions from central and southern South America.

The results described above represent MTBC genotype distributions aggregated across all years from 1990 to 2017. In order to investigate the changes over time in genotype distribution, and to illustrate the time periods that more accurately represented the data in each country, we created maps of genotype distribution for three distinct time periods (Additional file 1: Figure S3). To synthesize these data, we plotted the total prevalence of each lineage in each time period by region (Fig. 5). Figure 5 should be used as a guide and interpreted with some caution as it represents data aggregated across diverse geographic locations. The plots showed that lineage 3 strains have increased in prevalence over time in the UK (Additional file 1: Figure S3A-C) and Europe (Fig. 5). In addition, the plots showed a decline in the prevalence of lineage 1 in West and Central Asia (Fig. 5 and Additional file 1: Figure S3B-C).

The results presented in this study are consistent with previously published maps that showed that MTBC strains that evolved more recently in human history—lineage 2, lineage 3, and lineage 4 strains—tend to be more widely distributed around the world [22, 35, 47, 48]. We also showed that lineage 1, lineage 2, and lineage 3 are more prevalent in Europe and in North and South America than shown in previously published maps [35, 47, 48]. Moreover, we show that lineage 3 strains may be increasing in prevalence in Europe, while lineage 1 strains may be decreasing in prevalence in West Asia. These patterns in genotype distribution likely reflect both historical and recent movement of strains with people from East Asia and the Indian subcontinent to Europe and the American continent. The dominance of lineage 4 globally, and in particular in South American countries, also supports the hypothesis that European colonialists aided in the dispersion of this lineage in the mid-sixteenth to nineteenth centuries [32, 48, 49]. If the first inhabitants of the American continent brought early forms of lineage 2 strains with them when they migrated from north-eastern Asia, these strains may have been eliminated with the arrival of strains from European colonialists.

Human migration is likely not the only determinant of MTBC genotype distribution. Lineages 5 and 6 are prevalent only in West Africa [35, 47, 48]. The reasons for this geographic restriction are largely unknown but may have to do with clinical characteristics of the patients infected with these strains. Patients infected with lineage 6 are more likely than patients infected with other strains to be older, HIV-infected, and severely malnourished [50]. In addition, we showed that lineages 5 and 6 strains may be less likely to cause transmission chains than lineage 4 strains and that these findings were more consistent in Europe and the Americas than in Africa, which may reflect biological differences and/or social mixing which prevents these strains from spreading through non-West African populations. We also found that lineage 3 strains were associated with reduced risk of transmission chains in Europe and the Americas, which is consistent with the findings from a household contact study in Montreal [51]. In contrast, we found that Beijing family strains may be more likely to cause transmission chains, which could reflect the ability of Beijing strains to spread quickly through human populations [46, 52, 53]. These findings are not consistent with previous work that showed no differences between lineages in transmission from household contacts [46, 54, 55]. Thus, further studies would be required to confirm our findings.

Several studies included in our analysis showed that treatment failure was associated with lineage 2 Beijing family strains [43, 44]. Beijing family strains are also associated with drug resistance [56], which has been reviewed previously [12, 22, 23]. Additionally, lineage 1 strains have been associated with more rapid response to treatment in drug-susceptible TB cases in the USA [57]. Thus, there is evidence for a relationship between bacterial genotype and treatment outcome, at least in certain populations or contexts. Future studies that carefully control for potential confounders that may impact treatment failure are required to confirm these findings. This type of information could be particularly important to clinicians if it could inform the development of novel diagnostic tools that test for bacterial genotypes associated with poor response to treatment and development of drug resistance.

There was variation in the sampling methods and representativeness of the studies included in this systematic review. The majority of studies were representative of much smaller geographic locations than the national level, and despite the large number of bacterial isolates included in this study, they represented only a small fraction of the total estimated TB cases. While the goal of this study was to summarize the MTBC genotyping data available, not to make nationally representative estimates, it is important to note that this variation was not distributed evenly throughout the world. There was less information available about MTBC genotype distribution in South America and Sub-Saharan Africa than in other regions, and the data in Central and Eastern Asia represented a smaller proportion of all estimated TB cases than elsewhere. Thus, the genetic diversity shown in the map in Fig. 3 for these regions is likely less representative of the underlying populations.

Another source of variation that may impact representativeness is whether studies were biased towards including either rural or urban populations. There is likely greater MTBC genetic diversity in patients from urban populations than patients from rural areas since urban areas experience higher rates of travel and migration. Most studies included in this analysis did not report the urban/rural composition of their sample, and the bias towards one or the other would likely vary depending on study location. For example, the majority of the studies included in our systematic review used samples collected from public hospitals or reference laboratories. Therefore, in countries such as India, where people in urban areas may be more likely to seek care from private health clinics [58], the urban population may be underrepresented and we may have underestimated genetic diversity. On the other hand, in countries such as Uganda, where the rural population has limited access to public health facilities [59], the rural population may be underrepresented and we may have overestimated genetic diversity. This highlights the importance of data from prevalence surveys that use active surveillance techniques to reach a broader subset of the population.

We also identified a significant amount of heterogeneity between studies in the meta-analysis of genetic clustering associated with genotypes. One source of this heterogeneity is likely methodological differences between the studies, such as genotyping method, sampling method, and study duration, which have been shown to impact genetic clustering [27, 28]. For example, duration of sampling ranged from 2 months to 9 years, and genotyping methods ranged from the use of either spoligotyping or MLVA typing to the use of both methods (Additional file 1: Table S4). Studies that used shorter sampling durations may have missed transmission chains and underestimated clustering, while studies that used spoligotyping only may have overestimated clustering [60]. An additional source of heterogeneity may be confounders that impact genetic clustering and transmission, such as social mixing, immigration, age structure, comorbidities, and underlying TB incidence [27, 28]. These confounders likely also varied between these studies but were often not reported. For example, only 14 of the studies reported HIV prevalence (range 0 to 91%), only 6 reported proportion of immigrants (range 0 to 78%), and only 14 reported mean age of patients (range 25 to 50) included in the sample (Additional file 1: Table S4). If social mixing was high in each of the studies, this could have led us to overestimate the impact of genotype on transmission chains, while if migration was high, this could have led us to underestimate the presence of transmission chains.

A limitation of this study is that we grouped strains into seven lineages, which masks within-lineage variation. Distinct sub-lineages of the Beijing family are associated with differences in transmissibility in human populations [61, 62], and lineage 4 contains both geographically widespread and restricted sub-lineages [49]. However, we propose that this was the best method as it allowed us to (1) include a broad range of studies, including those that did not report sub-lineages, and (2) synthesize studies that used WGS- or PCR-based typing together with studies that used methods more common in resource-limited settings, such as spoligotyping and MLVA typing.

Another limitation is that we did not include data from WGS databases. A challenge of incorporating WGS data is identifying study meta-data, such as sampling methods and demographic characteristics of patients, linked with genomes. In addition, many of the WGS data available are poised for phylogeographic studies and for examining the presence of specific mutations [32, 49, 56], but are less representative of the populations they are isolated from. These data are often from outbreaks or studies of specific sub-populations, which we excluded in this analysis. As WGS data linked with meta-data become more available (through prevalence surveys [63] and endeavors such as ReSeqTB) including this data would be an important extension of our study. Our study supports these future studies by illustrating the importance of using genome sequences to determine phylogenetic lineages or sub-lineages. The dataset we have created could be used to fill geographic gaps in future WGS-based maps, particularly in regions where WGS technology is unavailable, and to verify results from convenience-based samples.