Genetic diversity, transmission dynamics and drug resistance of Mycobacterium tuberculosis in Angola

Tuberculosis (TB) poses a serious public health problem in Angola. No surveillance data on drug resistance is available and nothing is known regarding the genetic diversity and population structure of circulating Mycobacterium tuberculosis strains. Here, we have genotyped and evaluated drug susceptibility of 89 Mycobacterium tuberculosis clinical isolates from Luanda. Thirty-three different spoligotype profiles corresponding to 24 different Shared International Types (SIT) and 9 orphan profiles were detected. SIT 20 (LAM1) was the most prevalent (n = 16, 18.2%) followed by SIT 42 (LAM9; n = 15, 17.1%). Overall, the M. tuberculosis population structure in this sample was dominated by LAM (64.8%) and T (33.0%) strains. Twenty-four-loci MIRU-VNTR analysis revealed that a total of 13 isolates were grouped in 5 distinct clusters. Drug susceptibility data showed that 22 (24.7%) of the 89 clinical isolates were resistant to one or more antibacillary drugs of which 4 (4.5%) were multidrug resistant. In conclusion, this study demonstrates a high predominance of LAM strains circulating in the Luanda setting and the presence of recent transmission events. The rate and the emergence dynamics of drug resistant TB found in this sample are significant and highlight the need of further studies specifically focused on MDR-TB transmission.


Results
Patients and Clinical Isolates. In the present study we have collected 106 sputum samples, positive for acid-fast bacilli, in HDP in 2014. After culture and identification, five samples were excluded due to contamination; seven samples were duplicated from the same patient; and, five samples showed no growth after a 3 month incubation period. The remaining samples yielded 89 M. tuberculosis complex clinical isolates, each corresponding to a different patient (Fig. 1).
It was only possible to collect partial demographical and clinical data for 60 (67.4%) of the 89 studied patients (see Supplementary Table 1). The majority of the patients resided in the Kilamba-Kiaxi municipality (n = 34/58, 58.6%), where HDP is located, followed by a significant proportion of patients living in Viana (n = 19/58, 32.8%) (Supplementary Table 1).
The majority of the patients comprising this sample were registered as new patients (n = 53/60, 88.3%), approximately half presented with cavitary lung disease (n = 29/50, 58.0%) and eight out of 55 patients (14.5%) were co-infected with HIV. Fifty-one (87.9%) out of 58 patients started the 2HRZE/4HR treatment regimen while the remaining seven started the 2HRZES/4HR treatment regimen. The majority of the patients (n = 39/60, 65.0%) achieved both bacteriological and clinical cure after completing the treatment regimen (Supplementary Table 1).
Drug Resistance. We have detected eight different drug-resistance profiles and overall, 22 (24.7%) isolates showed resistance to one or more antibacillary drugs (Table 1). Four (4.5%) isolates exhibited a multidrug resistant (MDR) profile, i.e. resistance to at least isoniazid (INH) and rifampicin (RIF), whereas 13 of the 22 drug resistant isolates were resistant to INH or RIF and can thus be considered pre-MDR-TB isolates.
Comparatively, the 24-loci MIRU-VNTR analysis showed, as expected, a superior discriminatory power in which only 13 isolates were clustered across 5 clusters (Fig. 2, Table 2). Analysis of the genotypic data using the 12, 15 and 24-loci MIRU-VNTR sets showed that the 15-and 24-loci sets had comparable discriminatory ability as illustrated by the comparable values of the Hunter-Gaston Index of diversity (D) ( Table 3) 20,21 . Both spoligotyping and 12-loci MIRU-VNTR, owing to a lesser capability of genotypic discrimination, yielded a high degree of clustering, which does not necessarily illustrate recent transmission events but rather be associated with an earlier common origin followed by diversification at the genotypic level, in particular at loci with a faster molecular clock 11 . main results from drug susceptibility testing and genotyping. From the 106 sputum samples positive for acid-fast bacilli, 7 were excluded since they were duplicates from the same patient, 5 were contaminated and another 5 showed no growth, leaving a total of 89 M. tuberculosis clinical isolates. Drug susceptibility testing revealed that 22 isolates were resistant to one or more antibacillary drugs, of which four corresponded to MDR-TB isolates. Genotyping analysis allowed the identification of 13 clustered isolates across five different MIRU-VNTR clusters with one isolate excluded due to mixed strain infection. Spoligotyping analysis revealed a population structure dominated mostly by LAM and T strains. Drug resistant isolates were found across distinct clades and MIRU-VNTR clusters although the largest MIRU-VNTR cluster detected (n = 5) harboured 3 drug resistant isolates, including one MDR isolate (Fig. 1). A two-isolate cluster composed by pre-MDR strains (mono-INH resistance) was also detected.
To better understand transmission, a minimum spanning tree was constructed using double-locus variants from 24-loci MIRU-VNTR data as a maximum distance to link and define clonal complexes (Fig. 3). We have identified 12 clonal complexes which ranged in size from two to 17 isolates. The largest clonal complex, CC1, was comprised by 17 LAM isolates, mostly belonging to the LAM1 clade and MIT10 (Supplementary Table 2). A second large LAM9 clonal complex, CC2, was also detected comprised by six isolates. T strains were mostly detected on three clonal complexes, CC3-6, comprised by 6, 5 and 4 isolates, respectively. Furthermore, 32 singleton profiles were identified (Table 4).
Next, we have combined the genotypic data with the geographical data, i.e. patient residency area. Geographical case dispersion analysis was restricted to the Kilamba-Kiaxi and Viana since all but 5 patients resided in these municipalities. This is expected since HDP is located in the eastern zone of Kilamba-Kiaxi, near the Viana municipality and primarily serves these two municipalities. Although not meeting statistical significance, genotypic dispersion through these municipalities appears to exhibit substantial differences for some of the clades or clonal complexes assessed by spoligotyping or MIRU-VNTR, respectively (Table 5). Generally, LAM strains were more prevalent in Kilamba-Kiaxi in comparison to the Viana municipality (LAM: 70.59% vs 42.11%, respectively), which consequently holds a higher percentage of T strains (52.63%). Stratification by clonal complex, clade and SIT allowed picturing the distribution of these genetic groups at a finer resolution: in Kilamba-Kiaxi CC1 stands out as the most prevalent MIRU-VNTR clonal complex, which correlates with a high SIT20/LAM1 prevalence in this municipality (Table 5). CC3 also appears to play an important role in Kilamba-Kiaxi's epidemiological context, consistent with the prevalence of T1 strains. Although no clonal complex appears to be more prevalent in Viana, SIT53/T1 and SIT42/LAM9 strains exhibited a higher prevalence in this setting than in Kilamba-Kiaxi (Table 5).

Discussion
In the present study we made the first characterization of the genetic diversity and drug resistance of M. tuberculosis complex strains circulating in Luanda, Angola's most important setting concerning TB epidemiology 2 . Drug susceptibility testing revealed a worrying situation concerning resistance rates. Approximately one-quarter of the studied isolates were resistant to one or more antibacillary drugs. This situation is particularly notorious for INH resistance which was detected on 18% of the clinical isolates. Pre-MDR-TB strains (i.e. monoresistant to INH or RIF) were found to represent 14.6% of the studied samples highlighting an increased potential for MDR-TB development during the course of treatment by resistance amplification due to standardized treatment regimens and particularly, in cases of poor adherence 22,23 . The potential risk for resistance amplification is well patent in the Lisboa3 and Q1 strains in Portugal, or the KZN strains in South Africa that have gradually acquired resistance to an increasing number of drugs over the years in a stepwise manner 8,24,25 . Also, the MDR-TB rate appears to be moderate (4.5%) but may in fact represent a serious problem from a public health perspective. Assuming that the rate found in this study is the same across the entire province setting, new MDR-TB cases can be estimated to be at least around 958 cases based on the official province data estimate for 2013, of 21 281 new cases/year 2 . This situation clearly contrasts with the data reported in a study conducted in the Ndola district, in Zambia, a neighbouring country, in which from 193 clinical isolates only 8.8% and 0.5% showed resistance to any antibacillary drug or MDR, respectively 26 . A clear example of the consequences of not having an adequate laboratory support on the follow up of TB patients under treatment, a recurrent situation in many countries with unreported high levels of acquired MDR-TB [27][28][29] .
Besides drug resistance we also investigated the M. tuberculosis genetic diversity, population structure and transmission dynamics. Genotypic analysis through spoligotyping and 24-loci MIRU-VNTR revealed a population structure dominated by LAM and the ill-defined T strains. Comparing our data with the recently published global framework for the LAM lineage proposed by Mokrousov et al. 30 and the SNP barcode system for phylogenetic classification proposed by Coll et al. 10 , we find that: (i) the single strain belonging to SIT33 exhibits the VNTR signature characterized by loci 2401_2 and 3171_1 alleles in agreement to what has been proposed previously; (ii) 11 (12.5%) isolates bearing LAM or possibly LAM-derived spoligotype profiles formed a monophyletic clade bearing the locus 154_1 allele, characteristic of the RD115-sublineage/sublineage 4.3.3; 44 (50.0%) isolates bearing LAM profiles were found to exhibit the VNTR locus 802_1 allele proposed to be a marker for the RD174-sublineage/sublineage 4.3.4 10,30 . The data herein, presented therefore contributes to the VNTR data available for sub-Saharan Africa and is consistent with an increasing LAM prevalence when moving southwards on the African continent 30 . It is also important to consider when structuring M. tuberculosis samples by spoligotyping clades that spoligotyping profiles show a tendency to converge and that a given clade does not necessarily share a more recent common ancestor. Such is particularly true, e.g., among the ill-defined T lineage and LAM9 clade 10 . From a public health standpoint such polyphyletic clades may therefore enclose a higher degree of genetic diversity and underpin distinct transmission networks.
Again, when comparing with the neighbouring countries, only Zambia and Namibia have sufficient data on SITVIT WEB database to allow such comparison, as the M. tuberculosis population structure in the Democratic Republic of the Congo is presently unknown 16 . While Zambia exhibits a more diverse population structure dominated by the LAM11-ZWE (not found in the present study) but including other LAM, CAS, MANU and EAI strains, Namibia exhibits a more restricted diversity mainly dominated by SIT 20 (LAM1) strains that account to 78.5% of the strains deposited in SITVIT WEB for this country 16 . Although we cannot rule out epidemiological influence from the Democratic Republic of the Congo in the Angolan population structure, for Zambia there is no data supporting any epidemiological influence in Angola since no LAM11-ZWE strains were found in our sample. On the other hand, Namibia's M. tuberculosis population structure might be or have been of importance in shaping the current Angolan epidemiological context concerning genetic diversity as the most prevalent SIT in Namibia (SIT 20) was also the most prevalent SIT in our study albeit with a significant difference in prevalence (78.5% vs 18.2%, respectively). The relations that both countries have maintained during the Angolan civil war might have been of particular importance in strain circulation between both countries as a peak of 30 881 Angolan citizens took refuge in Namibia in 2001, most of which have later returned to the country of origin 31 . Similarly, the Democratic Republic of the Congo, for which no data on the M. tuberculosis genetic diversity presently exists, hosted up to 186 879 Angolan refugees in the same period, which is likely to have influenced the M. tuberculosis population structure on both countries. Strain spreading from the Democratic Republic of the Congo, and eventually from Republic of Congo, is plausible given the documented spread of HIV-1 variants from these countries to Angola 32 .
However, another factor shaping the population structure found herein might be located further away, pertaining the Community of Portuguese Speaking Countries. In fact a very similar M. tuberculosis population structure is found in Portugal when it comes to the prevalence of LAM strains 16 . All SIT20/LAM1 isolates (n = 16, 18.2%), and derived SITs, bear the VNTR locus 2461_2 allele while Lisboa3 clade strains in Portugal harbour the 2461_1 allele, which on the one hand hampers the establishment of a more direct link but, on the other hand, makes it possible to hypothesize that Lisboa3 strains might have derived from such SIT20/LAM1 strains bearing the 2461_2 allele during its evolutionary trajectory towards XDR-TB 8,30 . Such hypothesis warrants, however, further studies as the SIT20/LAM1 strains found in this study might have originated in Latin America as well 16,30 . Furthermore, 13 (18.8%) isolates were also found to harbour the VNTR locus 1644_1_allele that according to Mokrousov et al. 30 is only observed in few isolates across the RD174-sublineage with the exception of Q1 (SIT1106/LAM4) clade strains. This latter finding also raises the hypothesis of a common evolutionary origin between these strains, and countries, that merits further investigation 8,30 . Historical ties connect the Community of the Portuguese Speaking Countries, which likely favour, at a macroepidemiological level, an enhanced strain circulation between these countries.
There is some paucity of data concerning clustering rates in African countries. The clustering rates obtained in this study, particularly with the 15 or 24-loci MIRU-VNTR sets, are comparable with the ones reported by Homolka et al. 33 in a study involving strains from Sierra Leone but considerably lower than the clustering rates obtained by Mulenga et al. 34 in a study from Ndola district in Zambia 33,34 . It is also likely that the clustering rate in our study might be underestimated due the restricted sampling timeframe 34 . Overall, the population structure provided by MIRU-VNTR typing was consistent with the spoligotyping population structure. Nevertheless, as  reported previously, MIRU-VNTR allows a deeper discrimination between isolates and is therefore more adequate for epidemiological surveillance 11 . Still in this regard, no significant difference exists between the discriminatory ability of 24   Besides demonstrating the presence of recent transmission events, complex transmission chains with a common origin may be underlying local strain diversification. In particular, the genotypic diversification exhibited by CC1, in comparison with the other clonal complexes, and its association with SIT20/LAM1 clades suggest an earlier introduction of LAM1 strains in this setting when compared with other clade-associated clonal complexes of restricted genotypic diversification. Concerning its global importance, SIT20/LAM1/MIT10 strains have already been previously described in Portugal as associated with drug resistance, highlighting, again, the putative epidemiological links between Portuguese-speaking countries 8 .
Additionally, given the high number of singleton profiles and non-clustered isolates, another important aspect to consider is that TB reactivation may be playing an important role in this setting's TB epidemiology. Tackling TB by preventive treatment on specific sub-populations associated with known risk factors for disease reactivation may contribute to slowing down the country's increasing TB incidence 35 .
In conclusion, this first cross-sectional study conducted in Luanda, Angola, provides a framework for future studies and programmatic management of TB in Angola. We provide sufficient evidence for cluster-based transmission with differential geographic dispersion. Furthermore, the moderate rate of MDR-TB found in this sample has major public health implications and highlights the need for further studies specifically focused on MDR-TB transmission.