Bacterial pathogens in Xpert MTB/RIF Ultra-negative sputum samples of patients with presumptive tuberculosis in a high TB burden setting: a 16S rRNA analysis

ABSTRACT In patients with presumptive tuberculosis (TB) in whom the diagnosis of TB was excluded, understanding the bacterial etiology of lower respiratory tract infections (LRTIs) is important for optimal patient management. A secondary analysis was performed on a cohort of 250 hospitalized patients with symptoms of TB. Bacterial DNA was extracted from sputum samples for Illumina 16S rRNA sequencing to identify bacterial species based on amplicon sequence variant level. The bacterial pathogen most likely to be responsible for the patients’ LRTI could only be identified in a minority (6.0%, 13/215) of cases based on 16S rRNA amplicon sequencing: Mycoplasma pneumoniae (n = 7), Bordetella pertussis (n = 2), Acinetobacter baumanii (n = 2), and Pseudomonas aeruginosa (n = 2). Other putative pathogens were present in similar proportions of Xpert Ultra-positive and Xpert Ultra-negative sputum samples. The presence of Streptococcus (pseudo)pneumoniae appeared to increase the odds of radiological abnormalities (aOR 2.5, 95% CI 1.12–6.16) and the presence of S. (pseudo)pneumoniae (aOR 5.31, 95% CI 1.29–26.6) and Moraxella catarrhalis/nonliquefaciens (aOR 12.1, 95% CI 2.67–72.8) increased the odds of 6-month mortality, suggesting that these pathogens might have clinical relevance. M. pneumoniae, B. pertussis, and A. baumanii appeared to be the possible causes of TB-like symptoms. S. (pseudo)pneumoniae and M. catarrhalis/nonliquefaciens also appeared of clinical relevance based on 16S rRNA amplicon sequencing. Further research using tools with higher discriminatory power than 16S rRNA sequencing is required to develop optimal diagnostic and treatment strategies for this population. IMPORTANCE The objective of this study was to identify possible bacterial lower respiratory tract infection (LRTI) pathogens in hospitalized patients who were initially suspected to have TB but later tested negative using the Xpert Ultra test. Although 16S rRNA was able to identify some less common or difficult-to-culture pathogens such as Mycoplasma pneumoniae and Bordetella pertussis, one of the main findings of the study is that, in contrast to what we had hypothesized, 16S rRNA is not a method that can be used to assist in the management of patients with presumptive TB having a negative Xpert Ultra test. Even though this could be considered a negative finding, we believe it is an important finding to report as it highlights the need for further research using different approaches.

occurring annually worldwide (1,2).Globally, LRTIs are the fourth leading cause of death claiming 2.4 million lives in 2019 (2).In Ethiopia, LRTIs are the main reason for hospital admissions and the third leading cause of death, accounting for 8.2% of all deaths in 2019 (3).
In most low-and middle-income countries, people presenting with prolonged cough are first investigated for tuberculosis (TB) using smear microscopy or a rapid molecular test, such as the Xpert MB/RIF assay.When Mtb is not detected, clinicians often prescribe a trial of broad-spectrum antibiotics (10,11).This can be problematic as empiric use of antibiotics does not always result in clinical improvement and may drive the emergence of antibiotic resistance (12).When broad-spectrum antibiotics fail to improve clinical symptoms, empiric TB treatment is often initiated even though studies have shown that empiric TB treatment does not affect survival (13)(14)(15) and may even increase mortality (16).
In 2017, the World Health Organization (WHO) endorsed the Xpert Ultra assay (17) given its excellent performance for the diagnosis of TB, with a pooled sensitivity of 88% (95% CI: 85% to 91%) and specificity of 96% (95% CI: 94% to 97%) compared to liquid culture (18,19).With a high negative predictive value (98.1%), patients presenting with symptoms of TB whose sputum sample is Xpert Ultra negative are thus highly unlikely to suffer from pulmonary TB (20).This raises the question whether empiric TB treatment for patients with negative Xpert Ultra test results is the correct management.A better understanding of the etiological cause of respiratory symptoms in patients presenting with symptoms of TB whose sputum sample is negative on Xpert Ultra is important to develop evidence-based algorithms for the optimal management of this patient population.
16S rRNA gene amplicon sequencing is a culture-free method to identify and compare bacterial diversity and microbial composition of a sputum sample.16S rRNA detects both culturable and non-culturable microorganisms (21) and is a less costly method for studying microbial diversity compared to whole genome sequencing and metagenomic approaches (22).When informative, implementation of a 16S RNA assay could help clinicians make decisions and implement effective therapeutic strategies, as has been done for patients with non-cystic fibrosis and chronic obstructive pulmonary disease (COPD) (23,24).16S rRNA gene amplicon sequencing has not yet been applied to study the prevalence of bacterial pathogens in sputum samples of patients presenting with symptoms of TB in whom the Xpert Ultra test result was negative.
This study aimed to use Illumina MiSeq 16S rRNA V4 amplicon sequencing to determine the putative etiology of LRTI in hospitalized presumptive TB patients in whom Mtb was not detected by the Xpert Ultra assay, as this information could result in the development of an assay to guide the management for this population.In addition, we aimed to compare the prevalence and distribution of respiratory bacterial pathogens in sputum samples that were Xpert Ultra positive and negative to assess whether differences in microbial composition observed by 16S rRNA could be a marker of the etiology of the respiratory symptoms.Finally, the association between the presence of specific bacterial pathogenic taxa in Xpert Ultra negative sputum sample and clinical improvement on an antibiotic trial, chest X-ray findings, and 6-month survival was explored to assess the clinical relevance of the bacterial composition of the sputum sample.

Cohort characteristics
Of the 250 Xpert MTB/RIF-negative participants presenting with symptoms suggestive of pulmonary TB, 35 (14%) were diagnosed with pulmonary TB (Xpert Ultra and culture positive) and 215 (86%) were not diagnosed with pulmonary TB (211 Xpert Ultra negative and culture negative; 4 Xpert Ultra negative and contaminated cultures).Among the 215 Xpert Ultra negative patients, 17.2% (n = 37) had a history of TB treatment, 20.2% (n = 42) were living with HIV, 13.5% (n = 29) were elderly (age ≥65 years), 6.3% (n = 13) were severely ill, 1.9% (n = 4) had diabetes mellitus (DM), and 5.1% (n = 11) had a diagnosis of COPD.Most patients had a normal chest X-ray (n = 150, 70.8%) and about half (n = 117, 54.4%) improved clinically after a trial of broad-spectrum antibiotics.Compared to participants with a positive Xpert Ultra test result, those with a negative Xpert Ultra test were less likely to have prolonged symptoms or comorbidity (diagnosis of DM or COPD) and were more likely to be older, underweight or overweight, have a normal chest X-ray and improve clinically after a trial of antibiotics (Table 1).

Association between bacterial pathogens in sputum sample and response to antibiotic trial
The sputum samples were collected from 215 patients with an Xpert Ultra negative result before they received a seven-day antibiotic trial of ceftriaxone and azithromycin (43.7%, n = 94), amoxicillin (30.2%, n = 65), or vancomycin plus doxycycline (26.1%, n = 56) (Fig. S1).Only 54.4% (n = 117) improved clinically after the antibiotic trial.Of 98 patients failing to respond to antibiotics, 21 started empiric TB treatment, whereas 77 did not.Of these 77 patients, the likely causative pathogen could be identified in the sputum sample of 7 (9%): M. pneumoniae (n = 2) B. pertussis (n = 2), Acinetobacter baumanii (n = 2), and P. aeruginosa (n = 1).In addition, 4 (5%) patients were diagnosed with bacteriologically confirmed TB during the 6-month follow-up period.For most 66 (86%) patients, the cause was their prolonged respiratory symptoms and their failure to respond to antibiotics remained unclear.

Association between bacterial pathogens in sputum sample and baseline chest X-ray findings
Overall, 29.2% of patients with an Xpert Ultra negative sputum sample had an abnormal chest X-ray.When adjusted for patient characteristics associated with the presence of an abnormal chest X-ray (rural residence, presence of prolonged cough, fever, chest pain, or shortness of breath [Table S2]), the odds of an abnormal chest X-ray were higher in the presence of S. pneumoniae/pseudopneumoniae (aOR 2.5, 95% CI 1.12-6.16)and lower in the presence of M. catarrhalis/nonliquefaciens (aOR 0.37, 95% CI 0.19-0.74) in the sputum (Table 4).

Association between bacterial pathogens in sputum sample and survival status at 6 months
Among the 215 patients with a Xpert Ultra negative sputum, nine (4.2%) died: three while hospitalized and six after discharge.When adjusted for patient characteristics

DISCUSSION
In this study, we aimed to investigate the bacterial etiology of LRTI in patients presenting with symptoms of TB who had a very low probability of having TB given their sputum's negative Xpert Ultra result based on 16S rRNA sequencing.The presence of potential bacterial pathogens in the sputum samples was identified and compared with their prevalence in Xpert Ultra positive sputum samples.We could determine the presence of most likely causal pathogen in only 13 of the 215 patients, as described in Overall, one or more (potential) bacterial LRTI pathogens were present in 80% of sputum Xpert Ultra negative samples.The most common pathogenic bacterial ASVs detected were Haemophilus spp., Staphylococcus spp., Streptococcus pneumoniae (pseudo)pneumoniae, and M. catarrhalis/nonliquefaciens, present in >20% of patients.The challenge in attributing LRTI to the presence of these pathogens is further highlighted by the observation that one or more of these (potential) bacterial LRTI pathogens were also present in about 82.8% of Xpert Ultra positive sputum samples and that, except for a higher prevalence of Mtb in Xpert Ultra positive sputum samples, the bacterial populations were almost similar for Xpert Ultra negative and positive samples.
This 80% prevalence of one or more potential bacterial LRTI pathogen is higher than what has been reported in Cameron and Cambodia based on culture methods, where bacterial LRTI pathogens was reported in 44% and 46.8% of presumptive TB cases, respectively (7,25).The high prevalence of (potential) bacterial LRTI pathogens in patients with confirmed TB is in line with prior reports that co-detection with other bacterial pathogens is common in patients diagnosed with pulmonary TB (26,27).In Cambodia, co-detection with another potential bacterial LRTI pathogens was observed in 33% of patients diagnosed with pulmonary TB by sputum culture (7).In Nigeria, 50% of sputum samples collected from patients with TB grew both Mtb and other bacteria implicated in LRTI as the same as in this paper (28).The higher prevalence may be explained by the use of 16S rRNA gene amplicon sequencing, which can identify both culturable and unculturable bacteria, providing a complete picture of the bacterial community of sputum samples (29,30).Among the patients with an Xpert Ultra negative sputum result, 29.2% had an abnormal chest X-ray, which is similar to the findings from a study in South Africa where 27.2% of Xpert Ultra negative patients had abnormal findings on chest X-ray (31).We also found that the presence of Streptococcus pneumoniae/pseudopneumoniae in the Xpert Ultra negative sputum samples increased the odds of an abnormal chest X-ray (aOR 2.5, 95% CI 1.12-6.16),whereas the presence of M. catarrhalis/nonliquefaciens decreased the odds of abnormal chest X-ray findings (aOR 0.37, 95% CI 0.19-0.74).
In our study population, just over half (54.4%) of patients with an Xpert Ultra negative sputum result improved on an antibiotic trial.Patients for whom M. catarrhalis/nonlique faciens, Streptococcus pneumoniae/pseudopneumoniae, M. pneumoniae, and Haemophilus spp. was detected in the sputum sample had higher odds of poor response to an antibiotic trial, even after adjusting for patient characteristics.This may be due to the presence of drug-resistant bacteria (32).Three of the four pathogens associated with failure to improve on an antibiotic trial were also associated with an increased odds of mortality in the 6 months following the initial assessment: S. pneumoniae/pseudopneu moniae (aOR 5.31, 95% CI 1.29-26.6),M. catarrhalis/nonliquefaciens (aOR 12.1, 95% CI 2.67-72.8),and M. pneumoniae (aOR 34.5 95% CI 4.79-292.3).
The main strength of the study was the use of 16S rRNA sequencing for the first time to detect bacterial LRTI pathogens in sputum samples of patients presenting with symptoms of TB who had a very low probability of having TB as Mtb was not detected by the highly sensitive Xpert Ultra assay.Another strength is the prospective collection of comprehensive clinical data.This allowed an assessment of the associations between the bacterial community and patient outcomes.Our study also had some limitations.First, this was a hospital-based study, limiting generalizability to outpatient settings.Second, bacterial sputum culture was not available in our resource-poor study setting, and assessment of sputum quality using Gram staining to determine the extent of oral flora contamination was not performed.A positive result from 16S rRNA gene sequencing may indicate either infection or colonization of the normal respiratory flora (29).Third, despite using the DADA2 algorithm with ASVs to increase the sensitivity and specificity compared to OUT picking methods, the 16S rRNA amplicon sequencing of the V4 region could not always discriminate accurately up to species level.For instance, of the Haemophilus spp., Haemophilus influenza type b and non-typable Haemophilus are causal pathogens for LRTI (33).However, Haemophilus parainfluenza is a common isolate from the healthy nasopharynx as well as H. influenzae type b.Non-typable H. influenzae can be found in sputum cultures of nearly half of adults with chronic bronchitis (34).Finally, because it is unclear which level of abundance a pathogen is clinically relevant, we reported any presence above 1%.This may have resulted in the inclusion of minority populations of pathogenic bacteria that are not of clinically important.Among the overall 30 types of the Staphylococcus spp., S. aureus is a common cause of pneumonia, but it is also frequently isolated in respiratory samples from healthy individuals as a colonizing bacterium (35).Of the Streptococcus spp., S. pneumoniae is a well-established cause of LRTI, but the role of S. (pseudo)pneumoniae is less certain, although it has been reported in COPD (36).M. nonliquefaciens frequently colonizes the upper respiratory tract and is usually non-pathogenic, rarely causing invasive disease (37).M. catarrhalis also commonly colonizes the healthy airways (38), but it can cause pneumonia in children and adults with underlying chronic lung disease (39).Third, although viral and fungal communities can cause LRTI, they cannot be detected in sputum samples when using 16 s rRNA.Finally, as drug susceptibility tests were not performed, the presence of antibiotic resistance as a cause for poor response to an antibiotic trial or mortality could not be assessed.
In conclusion, the study found that 16S rRNA could identify the bacterial pathogen responsible for LRTI in 6.0% of Xpert Ultra negative patients but was not specific enough to differentiate between carriage and disease-causing pathogens in 80% of cases, making this approach not appear to be clinically useful.The presence of M. pneumonia was associated with 34 times greater odds of mortality and the presence of S. pneumoniae (pseudo)pneumoniae or M. catarrhalis/nonliquefaciens increased the odds of mortality rate by 5 to 12 times, respectively, suggesting clinical relevance of these pathogens.Further research using tools with higher discriminatory power that can also detect viruses and fungi is required to guide the management of Xpert Ultra negative patients.

Study site, design, and data collection
We performed a secondary analysis of a cohort study that aimed to determine the impact of empiric TB treatment on mortality among hospitalized patients who tested negative on the Xpert MTB/RIF assay (13).In this cohort study, sputum samples were collected before antibiotic trials were started from 250 adults (age ≥18 years) with symptoms of pulmonary TB (current cough, night sweats, fever, and weight loss) who were hospitalized between December 2018 to July 2019 in the Jimma Medical Center in Ethiopia.At the Jimma University Mycobacteriology Research Center, the TB reference laboratory for Southwest Ethiopia, sputum samples were decontaminated and evaluated for the presence of Mtb by liquid culture using the Mycobacteria Growth Indicator Tube (MGIT) BACTEC MGIT 960 System (Becton Dickinson, Sparks, MD, USA), solid Lowenstein-Jensen (LJ) media culture, and the Xpert Ultra assay (Cepheid, Sunnyvale, CA, USA) (40,41).The Xpert Ultra test was repeated on the same sample in case of an invalid result and repeated on another sample in case of a "trace" result.Ethical clearance was obtained from the Ethical Review Board of Institute of Health, Jimma University, with Ref. No: IHRPGD/397/2018.Written informed consent was obtained from all study participants.A structured questionnaire was used to collect demographic and clinical data; medical records were reviewed for HIV status, chest X-ray findings, and response to antibiotic treatment.All study participants were followed up for 6 months to determine survival status.

16S rRNA gene amplicon sequencing
DNA was extracted from stored unprocessed sputum samples (stored at −80°C for 24 months) at the Mycobacteriology Research Center of Jimma University in Ethiopia using the commercially available PowerFecal DNA Isolation Kit (Qiagen) (42).MiSeq prepara tions were done in the Lab of Applied Microbiology and Biotechnology (Belgium) using an in-house optimized protocol for low-biomass samples (38), and dual-index paired-end Illumina MiSeq 16S rRNA V4 region with an amplicon size of 254-bp sequencing was performed at the Center for Medical Genetics of the University of Antwerp (Belgium), as described (38).

Statistical analysis
Processing and quality control of the sequencing reads were performed using the R package Divisive Amplicon Denoising Algorithm 2 (DADA2), version 1.6.0., to increase the sensitivity and specificity compared to OUT picking methods (38).At the genus level, we processed amplicon sequence variants (ASVs) and aggregated ASV read counts.We annotated ASVs and added metadata to samples using R. Statistical analyses and data visualization was performed using R.
Bacteria were categorized as present in the sputum sample when they were present at ≥1% of the population.Bacteria were then classified as potentially pathogenic, opportunistic (i.e., cause of disease in immunocompromised individuals, including people living with HIV or elderly people), or not LRTI-causing based on literature review using PubMed, ScienceDirect, and Google Scholar and using search terms pathogenic bacteria, opportunistic bacteria, bacterial genera, LRTI, and bacterial classification (Table S4).When comparison of the 16S rRNA amplicon data could not classify the bacteria present to species level, the bacteria present were classified as potential LRTI pathogens.
The difference of bacterial LRTI pathogens detected was compared between Xpert Ultra positive and negative samples using chi-squared test.Logistic regression analysis was performed to determine the association (odds ratio [OR] and its 95% CI) between (potential) bacterial LRTI pathogens and response to an antibiotic trial, findings on chest X-ray, and 6-month survival status.For each (potential) bacterial LRTI pathogen identified and each outcome of interest, a separate model was built.For each model, the adjusted OR was estimated by including patient characteristics that were associated with the outcome of interest at P-value < 0.2 in bivariate analysis.Generalized variance-inflation factor was estimated to check multicollinearity.Backward stepwise model reduction was performed using the likelihood ratio test with a P-value cut-point of 0.1.

TABLE 2
Potential bacterial pathogens identified by 16S rRNA gene amplicon sequencing in sputum samples stratified by Xpert MTB/RIF Ultra results

Table 2
, with 7 cases of M. pneumoniae, 2 cases of B. pertussis, 2 cases of A. baumannii, and 2 cases of P. aeruginosa.

TABLE 3
Association between (potential) bacterial pathogens and response to antibiotic trial among 215 symptomatic hospitalized patients with an Xpert Ultra MTB/RIF-negative sputum result a Adjusted for age ≥65 years, HIV status, history of TB treatment, presence of prolonged cough, fever, or chest pain, NA: infinitive number.b Boldface shows association.

TABLE 4
Association between the presence of (potential) bacterial pathogens and chest X-ray findings among 215 symptomatic hospitalized patients with negative Xpert Ultra result a Adjusted for rural residence, presence of prolonged cough, fever, chest pain, or shortness of breath.b Three patients missing CXR diagnosis; NA: infinitive number.c Boldface shows association.

TABLE 5
Association between the presence of potential bacterial LRTI pathogens in sputum and mortality among 215 symptomatic hospitalized patients with negative Xpert Ultra result a Adjusted for rural residence, body mass index, or HIV status.NA: Infinitive number.b Boldface shows association.