Prevalence and antimicrobial resistance profiles of respiratory microbial flora in African children with HIV-associated chronic lung disease

Background HIV-associated chronic lung disease (CLD) is common among children living with HIV (CLWH) in sub-Saharan Africa, including those on antiretroviral therapy (ART). However, the pathogenesis of CLD and its possible association with microbial determinants remain poorly understood. We investigated the prevalence, and antibiotic susceptibility of Streptococcus pneumoniae (SP), Staphylococcus aureus (SA), Haemophilus influenzae (HI), and Moraxella catarrhalis (MC) among CLWH (established on ART) who had CLD (CLD+), or not (CLD-) in Zimbabwe and Malawi. Methods Nasopharyngeal swabs (NP) and sputa were collected from CLD+ CLWH (defined as forced-expiratory volume per second z-score < − 1 without reversibility post-bronchodilation with salbutamol), at enrolment as part of a randomised, placebo-controlled trial of azithromycin (BREATHE trial - NCT02426112), and from age- and sex-matched CLD- CLWH. Samples were cultured, and antibiotic susceptibility testing was conducted using disk diffusion. Risk factors for bacterial carriage were identified using questionnaires and analysed using multivariate logistic regression. Results A total of 410 participants (336 CLD+, 74 CLD-) were enrolled (median age, 15 years [IQR = 13–18]). SP and MC carriage in NP were higher in CLD+ than in CLD- children: 46% (154/336) vs. 26% (19/74), p = 0.008; and 14% (49/336) vs. 3% (2/74), p = 0.012, respectively. SP isolates from the NP of CLD+ children were more likely to be non-susceptible to penicillin than those from CLD- children (36% [53/144] vs 11% [2/18], p = 0.036). Methicillin-resistant SA was uncommon [4% (7/195)]. In multivariate analysis, key factors associated with NP bacterial carriage included having CLD (SP: adjusted odds ratio (aOR) 2 [95% CI 1.1–3.9]), younger age (SP: aOR 3.2 [1.8–5.8]), viral load suppression (SP: aOR 0.6 [0.4–1.0], SA: 0.5 [0.3–0.9]), stunting (SP: aOR 1.6 [1.1–2.6]) and male sex (SA: aOR 1.7 [1.0–2.9]). Sputum bacterial carriage was similar in both groups (50%) and was associated with Zimbabwean site (SP: aOR 3.1 [1.4–7.3], SA: 2.1 [1.1–4.2]), being on ART for a longer period (SP: aOR 0.3 [0.1–0.8]), and hot compared to rainy season (SP: aOR 2.3 [1.2–4.4]). Conclusions CLD+ CLWH were more likely to be colonised by MC and SP, including penicillin-non-susceptible SP strains, than CLD- CLWH. The role of these bacteria in CLD pathogenesis, including the risk of acute exacerbations, should be further studied. Supplementary Information The online version contains supplementary material available at 10.1186/s12879-021-05904-3.


Study characteristics
This is a case-control study nested within the BREATHe study; a multi-site, double-blind, placebo-controlled, individually-randomised trial that investigated the efficacy of azithromycin therapy in CLD (ClinicalTrials.gov, NCT02426112) [12]. The study setting, population and procedures of the trial are described elsewhere [10,16]. Briefly, perinatally HIV-infected children aged 6-19 years with CLD (CLD+) who had been receiving ART for a minimum of six months were enrolled from outpatient HIV clinics in Blantyre, Malawi, and Harare, Zimbabwe, from June 2016 through August 2018. CLD was defined as forced expiratory volume in 1 s (FEV 1 ) zscore less than − 1.0 with no reversibility (< 12% improvement in FEV 1 after 200 μg of inhaled salbutamol). The justification for this definition is provided elsewhere [10]. A comparison group (CLD-) of perinatally HIVinfected children without CLD (FEV1 z-score > 0) was recruited at the same time as enrolment of trial participants using frequency matching for site, age (6-12 and 13-19 years) and duration of ART use (6 months to < 2 years and two or more years). These children had no heart or apparent lung disease and reported no respiratory symptoms. Socio-demographic data and clinical history were obtained through an interviewer-administered questionnaire.
Sample collection, transportation and processing NP swabs were obtained at enrolment using nylon flocked swabs (Copan Italia, Brescia, Italy) by study staff. Sputum was subsequently obtained (expectorated or induced where necessary). Samples were immediately stored in 1 ml of skim milk, tryptone, glucose, and glycerine (STGG) medium, placed on ice for a maximum of 1 h and then frozen down to at -80°C. The samples were transported on dry ice to Cape Town, South Africa, for batch processing.
Bacterial culture, identification and antimicrobial susceptibility testing (AST) NP and sputum samples were thawed to 22°C and vortexed for 30 s. A 10 μl volume of each sample was inoculated onto Bacitracin-heated blood agar (BHB), Columbia with Gentamicin agar (Colgent), Mannitol salt agar (MSA) and 2% sheep blood agar (BA)(NHLS Greenpoint Media, Cape Town, South Africa). The BHB, Colgent and BA were incubated in 5% carbon dioxide at 37°C for 48 h whilst the MSA was incubated in ambient air at 37°C for 48 h. Alpha haemolytic colonies on Colgent which were susceptible to optochin were presumptively identified as SP. Colourless, medium-sized colonies on BHB with a strict requirement for factor X (Hemin) and/or V (NAD+) were presumed to be HI or Haemophilus parainfluenzae, respectively. Nonhaemolytic grey to white colonies on BA that tested positive for the push test were identified as MC. DNasepositive colonies from MSA were identified as SA. Colonies, other than MC and HI, from BA and BHB that grew on MacConkey agar were presumptively identified as 'other' gram-negative bacilli (GNB) by colony morphology and Gram staining. The species identities of these GNBs were confirmed using Matrix-Assisted Laser Desorption/Ionization-Time-of-Flight mass spectrometry (MALDI-TOF MS). Bacterial load was semiquantitatively assessed as follows: an aliquot of each sample was streaked for single colonies. Growth in one, two, three or all quadrants of a culture plate was assigned the labels 1, 2, 3 and 4, respectively.
AST was conducted using the Kirby-Bauer disk diffusion method. The antibiotics tested for each pathogen were as follows: SP (oxacillin, cotrimoxazole, azithromycin and tetracycline), SA (cotrimoxazole, azithromycin, tetracycline, clindamycin, penicillin, and cefoxitin), HI (cotrimoxazole, azithromycin, tetracycline, ampicillin, amoxicillin-clavulanate, and cefuroxime) and MC (amoxicillin-clavulanate, cotrimoxazole, azithromycin and tetracycline). SA susceptibility to cefoxitin was tested as a surrogate for methicillin-resistance. AST was conducted and interpreted according to the 2018 Clinical and Laboratory Standards Institute guidelines and breakpoints, respectively [17].

Statistical analysis
R version 3.6.0 was used to conduct statistical analyses. The 1990 British growth reference curves [18] were used to generate height-for-age and weight-for-age z-scores. The lung function z-score were calculated using global lung function initiative GLI/ERS reference equations and the African American module [19]. Our team validated this among children in Zimbabwe and findings are published elsewhere [20]. Comparison of categorical data including semi-quantitative bacterial load distribution was performed by Fisher's exact test or Chi-square test where appropriate. Univariate and multivariate analyses of association with carriage of each individual bacterial species were performed using logistic regression and presented as odds ratios (OR) and adjusted ORs with a 95% confidence interval (CI), respectively. The following variables were selected a priori to investigate the risk factors for carriage of each bacterium: CLD, sex, age category, study site (Zimbabwe or Malawi), the season of sample collection (Rainy season: December to April, Cool season: May to August, Hot season: September to November), HIV viral load, previous TB treatment and ART regimen and duration. Any other variables identified as independent predictors of carriage in univariate analysis (p < 0.25) were also included in the adjusted model for that species. The following were excluded from the multivariate model because of co-linearity: Enrolment BMI-for-age z-score and weight-for-age (colinear with height-for-age), and CD4 count (colinear with viral load suppression).
In total, 400 sputa from 326 CLD+ and 74 CLD-participants were collected. At least one bacterial species was isolated from the sputa of half of the participants in each group, with no difference by CLD status (Table 2). There was no difference in semi-quantitative loads of any bacteria from either NP swabs or sputum between the two groups (Supplementary table: T1).
Thirteen different GNBs from NP and sputa were identified by MALDI-TOF MS platform. More GNBs were recovered from the CLD+ than in CLD-group (Supplementary table: T2). Among bacteria isolated from NP swabs and sputa, there were statistically significant co-carriage relationships between SP, HI, and MC  carriage (Supplementary table: T3). These co-carriage relationships were independent of age or site.

Risk factors associated with carriage of bacteria
In multivariate analysis, participants on ART for two or more years were less likely to carry SP in both NP and sputum (Tables 3 and 4 (Table 5). No other variable tested was associated with sputum MC carriage (Table 6).

Antibiotic and multidrug resistance of isolates
Some SP isolates failed to grow after initial isolation and therefore antibiotic susceptibility was conducted on 147/ 154 and 18/19 isolates from NP, and 75/83 and 15/17 isolates from sputa of CLD+ and CLD-participants, respectively. The proportion of NP isolates nonsusceptible to penicillin was 32% (55/173) and that for sputa was 22% (20/90). A total of 29% (75/263) of all SP isolates regardless of sample type were penicillin nonsusceptible. Penicillin non-susceptibility among SP was more common in the CLD+ participants (36% [ 1). Methicillinresistant SA was uncommon (4%, 7/195). β-lactamase production by MC was widespread (93%, 76/82) but not different between the two groups. Antibiotic susceptibility profiles did not differ between sputum or NP isolates for any bacterial species (Fig. 1).

Discussion
The main finding of our study was that CLD+ CLWH were more likely to carry SP and MC in their NP than their CLD-counterparts whereas, participants with high MRC dyspnoea score (reflecting respiratory disability) were more likely to carry MC and HI. In addition, age, HIV viral load, duration of ART, the season of sample collection, site and nutritional status were associated with bacterial carriage among study participants. Longer period on ART or suppressed viral load were associated with reduced carriage for several bacterial species. The observed differences between the CLD+ and CLDgroups were more striking in NP compared to sputa. Antimicrobial susceptibility patterns were similar between the CLD+ and CLD-groups, apart from SP penicillin non-susceptibility, which was higher in the CLD+ group.
Studies of bacterial carriage among CLWH with CLD, but without acute infection, are limited. Masekela et al., [14] observed that among CLWH aged 6-14 years (mean 6.9 years), diagnosed with HIV-related bronchiectasis in South Africa, the most prevalent bacteria, from cumulative induced sputum samples collected over one year, were HI (30%) and H. parainfluenzae (21%) followed by Pseudomonas aeruginosa (2%) and SA (1%) [14]. Similarly, Verwey et al., [21] also found HI to be the most frequent bacterial species (38%) in respiratory samples (mainly sputum) of 52 CLWH with non-CF bronchiectasis [median age, 11.4 years (interquartile range 7.7-12.5)]. Samples were collected over a 2-year period and the prevalence of the other relevant bacterial species were SP (12%), MC (13%) and SA (11%) [21]. Ferrand et al., [6], also found HI (n = 6) to be the most frequent bacteria recovered from 18 expectorated sputa from Zimbabwean CLWH (mean age, 14 ± 2.6 years) with CLD and without acute respiratory infection [6]. Although, the most prevalent bacterial species that we identified were similar to those detected in previous studies using sputum samples, the prevalences differed substantially. SA (29%) followed by SP (25%) then MC (9%) were the most common bacteria isolated from our CLD+ subjects. HI was infrequently detected (4%). These differences in study results may be explained by a number of factors including sample type (lower respiratory tract samples vs NP swabs), clinical state of participants (acute exacerbation vs. stable) and age. The studies by Masekela [14] and Verwey [21] both included samples collected during acute exacerbations as well as intervening periods.
Differences in age between the studies may also explain the differences in bacterial carriage; participants in the studies by Masekela [14] and Verwey [21] were younger than those included in our study (Masekela: mean 6.9 years (range 6-14 years) [14], Verwey: median 9.1 years (IQR 7.2-12.1) [21], Ferrand: mean 14 (SD ± 2.6) years [6]) and 15(IQR: 13-18) in our study). Moreover, Masekela [14] investigated induced sputum while Verwey [21] included expectorated and induced sputum, bronchoalveolar lavage and tracheal samples. In Ferrand's [6] study and ours, the samples collected were expectorated sputum for CLD+ subjects. Finally, both Masekela [14] and Verwey [21] studies also included multiple samples from the same participantsanalysis did not adjust for multiple sampling from the same participant.  (1), Ever treated for TB (2), Ever repeated a grade (9) and MRC dyspnoea score (1). φ Number of years on ART (n = 167), ART regimen(n = 172), Ever treated for TB (n = 172), Ever repeated a grade (n = 169), MRC dyspnoea score (n = 172). λ n = 83, 85 and 85 for the number of years on ART, ever repeated a grade and MRC dyspnoea score. Variables with two levels where a level is less than 10% of total observations are not tested for associations. These included current school attendance and taking cotrimoxazole. Variables that have values in the adjusted odd ratios column were included in the multivariate model for that bacteria. Variables with p values <0.25 were included in the multivariate model except where they are colinear with another variable within the model. Weight-for-age was colinear with height-for-age and hence excluded from the model. Ever repeated a grade was excluded because of co-linearity with MRC score. CD4 count is colinear with viral load suppression. Clinically relevant variables (a priori-defined) that were included in the multivariate model regardless of significance were group, age, sex, site, season of sample collection, number of years on ART, enrolment viral load, ART regimen, ever treated for TB, enrolment height-for-age z-score and MRC dyspnoea score In our study, CLD+ participants were more stunted and underweight compared to their CLD-counterparts (Table  1). A meta-analysis showed that the prevalence of SP carriage is higher in malnourished children compared to their counterparts who were not malnourished [22]. This study also found that stunted and underweight children were also more likely to carry SP than children with normal weight and height, a finding which is consistent with our results. Malnutrition causes immune changes such as atrophy of the thymus, impairment of complement activity and immunoglobulin responses to encapsulated bacteria and a reduction in immunoglobulin A secretion [22].
While we did not record vaccination history, the majority of our study participants would not have been previously vaccinated against pneumococcal disease as these vaccines were introduced in both countries around 2012 [30,31] when our study participants would have been too old to be eligible for immunisation. Pneumococcal vaccines have been shown to reduce both carriage and disease through individual protection of the vaccinated which in turn disrupts transmission [32]. Hence, the high SP carriage in our settings might also be attributed to the lack of pneumococcal vaccination.  (7), CD4 count (1), Ever treated for TB (2), Ever repeated a grade (9) and MRC dyspnoea score (1). φ n = 99 for the number of years on ART, ever repeated a grade, MRC dyspnoea score. λ n = 113 for the number of years on ART and MRC dyspnoea score. Variables with two levels where a level is less than 10% of total observations were not tested for associations. These included current school attendance and taking cotrimoxazole. Variables that have values in the adjusted odd ratios column were included in the multivariate model for that bacteria. Variables with p values <0.25 were included in the multivariate model except where they were colinear with another variable within the model. Weight-for-age was colinear with height-for-age and hence excluded from the model. Ever repeated a grade was excluded because of co-linearity with MRC score. CD4 count is colinear with viral load suppression. Clinically relevant variables (a priori-defined) that were included in the multivariate model regardless of significance were group, age, sex, site, season of sample collection, number of years on ART, enrolment viral load, ART regimen, ever treated for TB, enrolment height-for-age z-score and MRC dyspnoea score The prevalence of SA in NP of CLD+ CLWH (23%) is comparable to similar studies conducted among CLWH in Ethiopia (29-31%) [26,33] and India (24%) [34]. However, the prevalence in CLD-participants (12%) was lower in our study. These discrepancies in bacterial prevalence between studies may be explained by the differences in sampling sites (NP vs nasal) or age. The anterior nares are the natural niche for SA and may be expected to be more frequently colonised than the NP [35].
Zambian [27] and Indian [34] studies of CLWH reported a much higher prevalence of HI (29 and 26% respectively) than either of our groups [12%(CLD+) and 5% (CLD-)]. This higher prevalence may be due to the younger age of the study participants (median ages were 5.1 years and 6.5 years for the Zambian and Indian studies respectively, compared to median 15 years for our study. NP HI carriage declines with increasing age [29,36]. The prevalence of MC in the NP of CLD-CLWH (3%) was much lower than similar studies conducted among CLWH in Ethiopia (12.3%, median age is 11 years) [26] and Ghana (39.8%, median age is 5.8 years) [37], but comparable to a Cambodian study (6%, median age is 7 years) [25]. Again, differences in age may explain these findings. NP MC carriage declines with increasing age from about 60% at 1-2 years to about 12% at 7-14 years [36]. MC is implicated in acute exacerbations of chronic bronchitis [38] and chronic obstructive pulmonary disease (COPD) [39], and therefore our finding of higher NP carriage in CLD+ participants (14%) vs CLD-participants (3%) is of interest. Sputum MC carriage was also higher in CLD+ (9%) vs CLD-CLWH (5%), but this difference was not statistically significant. Whether MC carriage plays a role in the pathogenesis of CLD or is a consequence of CLD requires further study.
The colonisation of the nasopharynx by multiple bacterial species may have important clinical  (1), Ever treated for TB (2), Ever repeated a grade (9) and MRC dyspnoea score (1). σ n = 41 and 42 for the number of years on ART and Ever repeated a grade respectively ξ n = 50 for the number of years on ART, ever treated for TB & ever repeated a grade. Variables with two levels where a level is less than 10% of total observations were not tested for associations. These included current school attendance and taking cotrimoxazole. Variables that have values in the adjusted odd ratios column were included in the multivariate model for that bacteria. Variables with p values < 0.25 were included in the multivariate model except where they are colinear with another variable within the model. Weight-for-age was colinear with height-forage and hence excluded from the model. Ever repeated a grade was excluded because of co-linearity with MRC score. CD4 count is colinear with viral load suppression. Clinically relevant variables (a priori-defined) that were included in the multivariate model regardless of significance were group, age, sex, site, season of sample collection, number of years on ART, enrolment viral load, ART regimen, ever treated for TB, enrolment height-for-age z-score and MRC dyspnoea score consequences including biofilm formation, polymicrobial infections and antibiotic resistance [40,41]. We found strong positive associations between NP carriage of SP, HI and MC. While some previous reports observed similar positive associations [34,36,[42][43][44][45], others found the opposite [46][47][48]. Possible reasons for the varying observations include differences in age of participants enrolled, vaccination schedules, year of study, hostgenetic background, antibiotic use and socio-economic status of the countries of study (low income compared to high-income status). Interestingly, the differences we observed in NP bacterial carriage between CLD+ and CLD-groups was not mirrored in the sputum, despite the belief that the source of bacteria in the lower respiratory tract is largely from the upper airways through micro-aspiration [49]. Evidence from studies suggests that the oropharyngeal samples rather than the nasopharyngeal mirrors the lung microbiota (sputum) [50]. This may explain the discrepancies in NP and sputa bacterial carriage we observed.
We also found that ART for more than two years reduced the odds of pneumococcal carriage in both NP and sputa. This is consistent with a report by Nicoletti et al. [51] who found that consistent use of the same ART for a year or more was negatively associated with risk of NP pneumococcal colonisation in adults living with HIV. Incomplete recovery of B cell function was noted in children who were on ART for less than a year [52] and was associated with high NP pneumococcal carriage [52].
Increased odds of NP pneumococcal carriage in hot, dry seasons compared to rainy seasons is consistent with previous findings in Malawi [53] and other parts of Africa (Kenya [54] and Gambia [55]). Similar observations have also been made in Thailand [56] and the United States of America [57]. One reason for this observation is increased school attendance by children during the hot season, which may increase transmission. Furthermore, a study in Niger revealed that airborne dust and high temperatures were risk factors for invasive pneumococcal disease [56]. Dust exposure attenuates phagocyte-mediated bacterial  (7), CD4 count (1), Ever treated for TB (2), Ever repeated a grade (9) and MRC dyspnoea score (1). σ n = 11 for the number of years on ART. ξ n = 33 for the number of years on ART & ever repeated a grade. Variables with two levels where a level is less than 10% of total observations were not tested for associations. This included current school attendance and taking cotrimoxazole. Variables that have values in the adjusted odd ratios column were included in the multivariate model for that bacteria. Variables with p values <0.25 were included in the multivariate model except where they are colinear with another variable within the model. Weight-for-age was colinear with Height-for-age and hence excluded from the model. Ever repeated a grade was excluded because of co-linearity with MRC score. CD4 count is colinear with viral load suppression. Clinically relevant variables (a priori-defined) that were included in the multivariate model regardless of significance were group, age, sex, site, season of sample collection, number of years on ART, enrolment viral load, ART regimen, ever treated for TB, enrolment height-for-age z-score and MRC dyspnoea score killing whilst the high temperatures promoted SP autolysis, accompanied by the release of the toxin pneumolysin [58]. Our finding that pneumococcal carriage was higher in the hot, dry seasons may further explain the higher incidence of invasive pneumococcal disease (IPD) among both patients living with and without HIV irrespective of age in Malawi during these seasons [59].
The reason for the association between male sex and SA carriage, also reported previously [60], is incompletely understood. Potential reasons include poor hand hygiene and participation in contact sport by males [60]. Furthermore, physiological factors, including sex hormones, regulation of the immune system and bacterial virulence have also been suggested [60].
Our finding of higher prevalence of penicillin-nonsusceptible SP in CLD+ participants is unsurprising since these participants were more likely to have been previously treated with antibiotics for acute exacerbations and were also more likely to have been treated for tuberculosis. Indeed, previous tuberculosis treatment was associated with increased odds of NP SP carriage in our multivariate model. Recent exposure to antibiotics is the strongest risk factor for NP carriage and spread of resistant SP [61]. Again, pneumococcal vaccination may be beneficial in this population. This is because vaccine serotypes are more likely to be resistant and therefore a reduction in carriage of vaccine serotypes resulting from prior immunisation can reduce antimicrobial resistance [62].
All four bacterial species tested exhibited moderate to high levels of resistance to cotrimoxazole. This was expected as 90% were receiving cotrimoxazole prophylaxis. Cotrimoxazole has been shown to reduce morbidity and mortality not only in patients living with HIV but also in their family members who are living without the infection [63], despite high background resistance in respiratory pathogens [27]. This positive effect may result from a reduction of systemic and intestinal inflammation via the modulation of the gut microbiome and immune and epithelial cell activation [64].
The strengths of our study include the enrolment of age-matched CLD-participants from the same area as the CLD+ group as a comparison group. We also included participants from two sites, which increased generalisability. Limitations of our study include the use of culture alone, without PCR, which may be more sensitive for detection of bacteria and would allow more precise quantitation, and the relatively small sample size for the comparator CLD-group which may have reduced statistical power. Also, the definition of CLD used may inadvertently include participants who may have normal lung function. However, our previous studies have shown that FEV1 z-score is an objective measure that correlates well with CLD based on radiological analysis (high-resolution computed tomography) [5][6][7]. Also, within the main trial under which this study was nested, we anticipated a reduced efficacy of azithromycin in participants with advanced disease, and therefore did not restrict enrolment to more severely ill, symptomatic children. In conclusion, CLD+ CLWH were more likely to be colonised by MC and SP, including penicillin-resistant SP strains. The role of these bacteria in CLD pathogenesis, including the risk of acute exacerbations, should be further investigated.