Whole-genome sequencing (WGS) has transformed the way we analyze transmission of tuberculosis (TB) (Nikolayevskyy et al., 2016; Comas, 2017; Satta et al., 2017). Identification of transmission clusters is currently based on determination of the magnitude of genomic diversity among the isolates in a population. An exhaustive analysis of the magnitude of this variability acquired in different clinical/epidemiological situations led to the definition of thresholds to determine whether 2 isolates were part of the same transmission chain, i.e., clustered (<5 SNPs) or unrelated (>12 SNPs) (Walker et al., 2013). Since then, these thresholds have been used as a consensus reference, and it has been accepted that microevolution in Mycobacterium tuberculosis (MTB) infection occurs within those ranges and with no wider deviations expected (Casali et al., 2016; Seto et al., 2017; Witney et al., 2017).

Acquisition of variability in MTB has been studied not only to determine the SNP thresholds to be applied in genomic epidemiology, but also to help us to understand one of the microevolution paths in MTB infection which may lead to more successful variants, in addition to other molecular events such as insertion/deletions or intragenomic recombination.

Different studies have addressed the emergence of clonal variants in MTB infections according to different molecular markers (Warren et al., 2002; Shamputa et al., 2006; Al-Hajoj et al., 2010; Navarro et al., 2011; Black et al., 2015). The magnitude of intrapatient diversity which can be measured for a strain within the same patient has been found to be comparable to the one observed after sequential rounds of multiple hosts sequentially along transmission chains (Pérez-Lago et al., 2014). Sometimes, these variants have been found associated to prolonged infections or to a history of previous TB (Navarro et al., 2011), suggesting that infection time longer than the average and reactivation events could be associated with the accumulation of genetic diversity. The characteristics of the healthcare system, the efficiency of the drug treatment and the socioeconomic status also have a role on the genetic diversity expected. The analysis of diversity in the population of MTB has been found to be associated in certain cases with substandard life conditions, delayed diagnosis and improper therapeutical management and poor adherence to treatment, factors which can be responsible for prolonged infections (Shamputa et al., 2006; Navarro et al., 2011).

Post-mortem genetic analysis of the MTB population within individuals (Cohen et al., 2011) has demonstrated the accumulation of diversity by MTB after a prolonged infection period. It has been shown with special exhaustivity in a recent article by Lieberman et al. (2016), based on post-mortem multiple sampling and WGS analysis in patients who had received only minimal anti-TB treatment.

The identification of genetic diversity is not a meaningless finding and different authors have found an effect of certain subtle genetic modifications and the expression of neighboring genes and even in the infectivity between closely related variants and the potential functional role of microevolution has been explored (Soto et al., 2004; Tantivitayakul et al., 2010; Pérez-Lago et al., 2011, 2013).

Therefore, although the similarity thresholds defined in the genomic epidemiology of TB (Walker et al., 2013) are based on a wide and solid analysis of patients, additional analysis of microevolution in MTB in specific clinical/epidemiological situations that might represent greater opportunities to acquire variability, which were not extensively represented in the studies defining these cutoffs, could help us to evaluate the robustness of the similarity thresholds proposed and to understand the influence of various infection scenarios on microevolution dynamics.

We aimed to select representative examples to illustrate deviations in the standard infective circumstances faced by MTB that could provide MTB with better opportunities to acquire variability. We examined both intrapatient scenarios (single infections in selected cases) and interpatient scenarios (transmission chains involving serial hosts). We selected 2 settings in which we could expect a more marked intrapatient microevolution: prolonged infections due to poor adherence to treatment and cases in which reactivation occurred sometime after the microbiological cure of the first episode. For the interpatient evaluation, extensive active transmission clusters involving a high number of cases and clusters involving both latency and active transmission events were selected to attempt to obtain a selective pressure that was higher than average.

Microevolution is a phenomenon which can be modulated by various factors. We have to consider it at least a bifactorial event in which both bacterial and host factors intervene. The role of bacterial factors is illustrated by the finding that certain strains from the Beijing lineage show a hypermutator phenotype (Ebrahimi-Rad et al., 2003)- Other non-Beijing strains have been also described to be prone to acquire diversity (Navarro et al., 2017). Microevolution events leading to higher than expected accumulation of variability have been partially addressed in a previous study from our group (Pérez-Lago et al., 2014). Some of these events involved sequential host-to-host transmissions, but an equivalent degree of diversity was observed also within individuals. This intrapatient diversity was detected between isolates infecting different organs, in patients with respiratory and extrarespiratory infection, but also in isolates recovered strictly from the respiratory site. These data mean that the intrapatient diversity might impact on the inference of recent transmission clusters (Pérez-Lago et al., 2014).

As our previous study was based on a convenience sample of isolates selected once they had been shown to have microevolved, we now aimed to perform a more systematic evaluation (i.e., with no preselection of microevolved isolates) of the impact of several circumstances on the acquisition of variability that MTB infection could meet in various clinical or epidemiological scenarios. As with any evolutionary process, microevolution in MTB is the combination of the opportunities the microorganism has to acquire variability and the occurrence of selective pressure bottlenecks that force the selection of clonal variants from among those which emerge. Following this rationale, we selected clinical/epidemiological situations that could lead to increased acquisition of variability (both intrapatient and interpatient).

We began with the intrapatient scenario by selecting single case infections, looking for situations with longer periods of active infection (prolonged infections) in order to increase the likelihood of the emergence of variants and/or looking for factors involved in selective pressure (such as intermittent poor adherence to treatment or transitions from latency to active infection in reactivations). In general, our findings highlight the limited variability acquired, which was within the ranges expected. However, in 1 reactivation (case 6, Supplementary Table 1), we observed greater accumulation of variability (5 SNPs), thus reaching the lower threshold for considering 2 cases as not epidemiologically linked (Walker et al., 2013). Previously, we had also observed intrapatient situations leading to a higher accumulation of SNPs (Pérez-Lago et al., 2014; Navarro et al., 2017).

The non-systematic association between a greater accumulation of variability and clinical situations that could theoretically favor it underlines the multifactorial nature of this phenomenon. Observations from other studies also lead to contradictory findings. Whereas no variability is found in circumstances where it is highly likely, such as long-term persistent infection by a Beijing strain with constant intermittent treatment (Pérez-Lago et al., 2015a), marked variability is detected in less extreme situations (Pérez-Lago et al., 2014; Navarro et al., 2017). It is also true that greater diversity is described in studies involving MDR strains in which a more dynamic process of acquisition of variability is expected owing to constant competition between resistant and compensatory mutations (Merker et al., 2013), although competition between clonal variants is also expected in heterogeneous populations involving susceptible strains.

We then moved on to the interpatient analysis, which focused first on extensive clusters to evaluate the acquisition of variability when 2 key factors coincided, namely, prolonged observation time and adaptation to a high number of sequentially infected hosts. However, the maximum genetic distance accumulated never surpassed 9 SNPs, and the distance between sequential isolates ranged from 1 to 6 SNPs, i.e., within the previously established thresholds although approaching in some cases to their limits (Walker et al., 2013).

We then looked for the coincidence of several of the factors that had been evaluated independently, making it possible to observe amplification of effects that could only have a subtle impact when evaluated one by one. It was not easy to find transmission events with the intervention of as many as possible of these factors, namely, extended time to allow microevolution to occur, involvement of a high number of hosts to ensure sequential selective pressures due to the establishment of each infection in different hosts, and overlap of transitions from latency to active infections. These kind of clusters with such a complex epidemiological peculiarities were insufficiently represented in the studies defining the SNP thresholds to be applied as a reference in genomic epidemiology (Walker et al., 2013). We found 3 clusters fulfilling all or most of our requirements, each from a different population and involving 2 countries. However, despite the greater opportunities to acquire diversity that the clinical/epidemiological situations in these complex clusters offered, the global variability identified generally lay within the thresholds established, which limits the impact of this diversity on the inference of recent transmission in genomic epidemiology studies.

Although our main aim was to perform a quantitative analysis of the number of SNPs acquired, we had the opportunity to record interesting qualitative observations. The first was the finding that some SNPs (1 each in case 4 and case 6, Supplementary Table 1; and one each in case 1 and 4-1 in cluster 2, Supplementary Table 5) were only observed in the first isolate of a chronological series. While this observation may at first sight seem incongruent, owing to the negligible degree of homoplasy described for Mycobacterium tuberculosis complex (Hershberg et al., 2008; Comas et al., 2009) and the highly unlikely reversion of mutations once acquired, we believe that it might be explained by undetected coexistence of the variant which acquired the SNP together with the parental variant, which did not have the SNP. Given the unavoidable limitations of sputum sampling with respect to the whole bacterial population in the lung, we likely only detected 1 of the coexisting variants at each sampling point. The observation that isolates recovered from 1 sputum sample do not represent true clonal complexity in the lung has been addressed elsewhere (Black et al., 2015; Pérez-Lago et al., 2015b).

About the functional significance of SNPs (13 synonymous and 40 non-synonymous SNPs), it must be highlighted a lower nonsynonymous/synonimous ratio (0.325) compared with a global database including 220 strains (0.606, p = 0.04 Fisher's exact test). The difference is mainly driven by the accumulation of a lower number of nonsynonymous changes in intrapatient cases compared to inter-patient transmission cluster (dN/dS 0.353 vs. 1.714). This observation is in contrast with the expectation that shorter timespan should be associated to higher dN/dS ratios. The lower number of nonysnonymous changes intrapatient may indicate the action of purifying selection. Purifying selection is expected during treatment course (Trauner et al., 2017) and can remove diversity faster, given the clonal nature of MTBC populations. However, the heterogeneity of the dataset involving different patients and transmission clusters as well as the low number of mutations prevents us to support any general conclusion.

Our global analysis did not enable us to identify a clinical/epidemiological situation that could predict greater accumulation of diversity in MTB. We must admit certain limitations resulting of not having considered the diversity based on structural variants and indels or not having performed a high-depth analysis or minority variants. We have focused on diversity based on SNPs mainly due to the fact that this is the criterion in which genomic epidemiology lies, and thus to evaluate the impact of this variability on the inference of recent transmission. The number of SNPs identified under circumstances that could theoretically favor microevolution was limited, within the ranges expected according to the proposed similarity thresholds for considering 2 isolates as related or unrelated (Walker et al., 2013). These findings support the robustness of the cutoffs. However, we also found asymmetric bursts of variability, as reported elsewhere (Pérez-Lago et al., 2014; Navarro et al., 2017). The coincidence of several such cases in the same transmission event could eventually lead us to overlook some transmission clusters.

We must also accept that despite trying to force the coincidence of factors theoretically favoring microevolution, some of these factors, such as length of the active infection period, serial involvement of sequentially infected hosts due to active recent transmission, and lack of adherence to treatment, could be more frequent and of a greater magnitude in settings with weaker diagnostic and control programs. Studies similar to ours should be replicated in these settings to enrich our knowledge of the circumstances which could favor diversity in TB and to determine more precisely its impact on the tracking of transmission.

Experimental load: all authors; Writing: LP-L, DG-d-V; Design conceptualization: LP-L, DG-d-V; Bioinformatic Analysis and design: ÁC-O, IC.