INTRODUCTION
The diagnosis of tuberculosis (TB) has significantly improved since the introduction of the automated molecular test Xpert MTB/RIF (Xpert; Cepheid, Sunnyvale, CA, USA) and its more advanced version, Xpert MTB/RIF Ultra (Xpert Ultra), which detect Mycobacterium tuberculosis (MTB) and rifampin (RIF) resistance. In 2017, the WHO endorsed Xpert Ultra to circumvent the comparatively low sensitivity of Xpert, particularly in smear-negative or paucibacillary specimens [1]. Xpert relies on CT delay to detect RIF resistance, whereas Xpert Ultra utilizes melting curves of sloppy molecular probes to identify mutations within the RIF resistance-determining region (RRDR) in rpoB, reducing false resistance reports [2]. While studies have demonstrated the accuracy and rapid diagnostic capability of Xpert Ultra [2–5], its clinical utility in high-resource, intermediate-incidence settings remains unassessed.
In Korea, the number of new TB cases has steadily declined since 2011, reaching 30.6 per 100,000 population, with 551 confirmed cases of multidrug-resistant TB in 2023 [6]. However, the global rise in nontuberculous mycobacterial infections is also evident in Korea [7], posing additional diagnostic challenges. To our knowledge, the performance of Xpert Ultra and Xpert in such a setting has not yet been assessed. We conducted a prospective study comparing the diagnostic performance of Xpert Ultra for detecting MTB and RIF resistance with that of Xpert.
MATERIALS AND METHODS
Samples and study design
This study was conducted at Samsung Medical Center, a tertiary care hospital in Seoul, Korea, and approved by the Institutional Review Board of Samsung Medical Center (approval No.: 2024-09-068). The requirement for informed consent was waived because we only used anonymized data. In total, 309 consecutive lower-respiratory specimens (263 sputum, one endotracheal aspirate, 37 bronchoalveolar lavage, and eight bronchial lavage fluid specimens) from 309 adult patients were prospectively analyzed using Xpert and Xpert Ultra between August 2023 and May 2024. Only specimens with available culture results were included. Patients who had received anti-tuberculosis treatment within the past 6 months, as well as duplicate specimens, were excluded. Patient medical records and microbiological test results, including acid-fast bacilli (AFB) smear, mycobacterial culture, and drug susceptibility testing (DST), were reviewed.
We used two reference standards: (i) mycobacterial culture and (ii) a composite reference standard (Fig. 1). The composite reference standard defines active pulmonary TB as a clinical diagnosis made by a clinician who initiated a full course of anti-tuberculosis treatment based on bacteriological, radiological, and/or other diagnostic findings. Given the low prevalence of RIF resistance in this setting, a spiking experiment was performed to assess the assays’ ability to detect RIF resistance. Well-characterized MTB isolates with rpoB mutations were spiked into pooled MTB-negative sputum specimens. We used 13 clinical MTB isolates with rpoB mutations identified in our previous study [8].
Xpert and Xpert Ultra procedures
Sample reagent was added to the specimen at a 2:1 ratio, and 2 mL of the mixture was then added to each cartridge. Xpert and Xpert Ultra were performed in parallel, according to the manufacturer’s instructions [5, 9]. Positive semiquantitative results are categorized as very low, low, medium, or high. Xpert Ultra additionally provides a “trace” category, in which RIF resistance cannot be determined because of insufficient signal detection. Results were analyzed using Xpert V6 and Xpert Ultra V4 (Cepheid).
Mycobacterial staining, culture, and detection of RIF resistance
AFB staining was performed using an auramine-rhodamine fluorescent stain, followed by Ziehl–Neelsen staining for confirmation. Staining results were graded according to US Centers for Disease Control recommendations [10]. Specimens with an AFB smear score of ≥1 were considered smear-positive. Decontaminated samples were inoculated into a mycobacterial growth indicator tube (MGIT 960 system; Becton Dickinson, Sparks, MD, USA) with 3% Ogawa agar (Shinyang, Seoul, Korea) and cultured for 6 weeks. All M. tuberculosis isolates were tested for RIF resistance using the MGIT 960 system and sent to the Korean Institute of Tuberculosis, a WHO-designated supranational reference laboratory, for conventional DST using the absolute concentration method with Lowenstein–Jensen medium [11]. The critical concentrations for RIF resistance are 0.5 and 40 μg/mL for the MGIT 960 system and absolute concentration method, respectively [12].
Comparison of Xpert and Xpert Ultra for the detection of MTB
The results from the 309 respiratory specimens were analyzed to compare the diagnostic performance of Xpert and Xpert Ultra. Cohen’s kappa value was calculated to assess agreement, and McNemar’s test was applied to evaluate statistically significant differences between the two assays. We calculated the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of Xpert and Xpert Ultra using two reference standards: (i) mycobacterial culture and (ii) a composite reference standard.
Accuracy of the detection of <italic>rpoB</italic> mutations
An MTB-negative sputum specimen pool was prepared for spiking. Thirteen cultured isolates of MTB with rpoB mutations were heat-inactivated at 80°C for 20 min [13, 14]. Each isolate was pelleted and homogenized in 0.05% Tween-80 solution [15]. The suspensions were transferred to screw-capped glass tubes containing 3-mm glass beads, vortexed vigorously for 3 min, and incubated for 30 min to obtain a supernatant free of visible clumps. The supernatant was collected in a fresh tube and adjusted to a turbidity equivalent to 0.5 McFarland standard. Serial dilutions were prepared to achieve medium-level (104 CFU/mL) and low-level (103 CFU/mL) concentrations [5]. Each spiked sample was tested at both concentrations (medium and low levels) using Xpert and Xpert Ultra simultaneously. The melting temperatures (Tms) for Xpert Ultra were recorded.
Statistical analysis
The statistical significance of the overall agreement was evaluated using Cohen’s kappa with a 95% confidence level. McNemar’s test was used to assess differences in sensitivity and specificity between Xpert and Xpert Ultra. Statistical analyses were performed using MedCalc Statistical Software version 23.0.6 (MedCalc Software Ltd., Ostend, Belgium). Statistical significance was set to P<0.05.
RESULTS
Comparison of Xpert and Xpert Ultra for the detection of MTB
Among the 309 respiratory specimens, one bronchial lavage fluid specimen was excluded because of an invalid culture result, leaving 308 specimens for analysis. Xpert and Xpert Ultra detected MTB in 25 and 33 specimens, respectively (Table 1). Cohen’s kappa coefficient (0.81; 95% confidence interval [CI], 0.70–0.92) indicated strong agreement between the two assays. However, McNemar’s test revealed a significant difference in MTB detection rates (P=0.021).
The diagnostic performance of Xpert and Xpert Ultra was further evaluated using mycobacterial culture and a composite reference standard (Table 2). In smear-positive cases, both assays achieved 100% sensitivity and specificity. In smear-negative cases, Xpert Ultra showed higher, albeit not significantly, sensitivity than Xpert, with comparable specificity; however, Xpert Ultra yielded a few false-positive results (five with culture; three with the composite reference standard), resulting in a lower PPV (Table 2).
Ten cases exhibited discordant results between Xpert and Xpert Ultra (Table 3). Notably, Xpert Ultra produced trace- or very low-positive results in six culture-negative cases, two of which were clinically diagnosed as pulmonary TB.
Detection of RIF resistance
We assessed 13 clinical TB strains with known rpoB mutations to evaluate RIF resistance detection. All but one of the mutations were non-synonymous single-nucleotide polymorphisms, with one non-frameshift indel variant [16] (Table 4). Eleven strains had a mutation within the RRDR, two had a mutation outside the RRDR (I491F or D545Q), and three harbored multiple mutations.
RIF resistance was not detected by either assay for isolates with the I491F or D545E mutation outside the RRDR. Xpert successfully detected all 11 RRDR mutations, whereas Xpert Ultra yielded indeterminate results for two isolates. Specifically, for the Q432P mutation, the rpoB1 probe was amplified, but no Tm was obtained. Similarly, for the isolate with multiple mutations of Q429H/L430P/H445Q, Tms were detected for H445Q for both of the rpoB3 and rpoB4 A probes, whereas no amplification was observed for the rpoB1 probe targeting Q429H/L430P. For the remaining nine isolates, Xpert Ultra accurately detected RIF resistance, with corresponding mutant Tm values at specific probe targets. Both assays produced reproducible results across all isolates at medium and low concentrations, except for two low-level spiked samples: (i) Xpert Ultra yielded a trace-positive result for a sample with the H445D mutation, precluding Tm determination, and (ii) Xpert failed to detect MTB in a sample with the P439L mutation.
DISCUSSION
Korea is a high-income country with an intermediate TB burden, ranking second among the Organization for Economic Cooperation and Development countries in TB incidence [6]. To date, the Xpert Ultra assay has been mainly evaluated in high-burden and, to a lesser extent, low-burden settings [17–21], highlighting the need for assessments in settings with distinct TB epidemiology, such as Korea. Because of delayed regulatory approval and supply chain issues, Xpert Ultra became available in Korea only in 2024. To our knowledge, this was the first prospective study to compare the performance of Xpert and Xpert Ultra in a high-resource and intermediate TB-burden setting [18, 20, 22, 23].
A major strength of our study is the use of fresh clinical specimens, allowing a direct comparison of the two assays and providing real-world data on Xpert Ultra performance. We found that Xpert Ultra yielded more false-positive results than Xpert, consistent with previous findings [18, 19]. False-positive results were observed in cases classified as trace- or very low-positive by Xpert Ultra. Among the seven specimens with trace-positive results, five were culture-positive, one was culture-negative but clinically confirmed as TB, and one was false-positive. Similarly, among the nine specimens classified as very low-positive results, six were culture-positive, one was culture-negative but clinically confirmed as TB, and two were false-positive. When evaluating pulmonary TB, the WHO does not recommend repeat testing for patients with an initial Xpert Ultra trace-positive result [24]. This recommendation is based on limited evidence supporting the diagnostic benefit of repeat testing, as well as concerns regarding interpreting trace results, which may indicate a very low bacterial burden or residual DNA from non-viable MTB [4, 24–26]. Given the inherent challenges in distinguishing true infection from non-specific detection, trace-positive results should be interpreted in the clinical context rather than being considered definitively positive or negative. In our study, trace- and very low-positive results were associated with not only culture-positive TB but also culture-negative cases with a clinical diagnosis of TB, as well as possible false positives. These findings highlight the need for a comprehensive diagnostic approach integrating clinical, radiological, and microbiological evidence. In cases of persistent diagnostic uncertainty, follow-up testing with a new specimen may be warranted to guide appropriate clinical management.
The overall sensitivity of Xpert Ultra based on culture as a reference standard was 73.7% (95% CI, 56.9%–86.6%), which was lower than previously reported values in studies using respiratory specimens [18, 23, 27]. This lower sensitivity may be attributed to the high proportion of smear-negative cases (76.3%, 29/38) among TB-positive specimens. Given that our study exclusively included adults, and no pediatric patients—who typically exhibit a higher proportion of smear-negative cases—this percentage is particularly noteworthy [22, 27–29]. A possible explanation for this observation is that, in high-resource settings, TB is often diagnosed at an earlier stage, when the bacterial load is low, leading to a higher proportion of smear-negative cases. Our findings provided valuable real-world data on the performance of Xpert Ultra in such settings. When considering smear-negative samples alone, the sensitivity of Xpert Ultra in this study was comparable with that in previous studies [4, 18, 23].
Evaluation of the detection performance of Xpert Ultra V4 for various rpoB mutations using 13 MTB strains revealed two indeterminate results for the Q432P and Q429H/L430P mutations, both attributed to the failure to generate a Tm value for the rpoB1 probe. The Q432P mutation has been previously reported to yield an indeterminate result. Omar, et al. reported that Xpert Ultra V2 failed to detect the Q432P mutation, whereas Xpert Ultra V3 reported it as indeterminate [30]. Rigouts, et al. confirmed that Xpert Ultra V3 produced indeterminate results for this mutation [31]. Similarly, Xpert Ultra V4, the latest version, failed to generate a Tm value despite successful amplification. Further, an isolate harboring a Q429H/L430P double mutation in the rpoB1 probe region failed to amplify, leading to an indeterminate result. Our findings confirm that the limitation of the rpoB1 probe observed in Xpert Ultra V3 persists in V4. Given these limitations, indeterminate results should be interpreted with caution and require confirmation to ensure accurate detection of RIF resistance.
Our study has certain limitations. First, the relatively small number of TB-positive specimens may have reduced the statistical power of our analysis. Moreover, the extremely low prevalence of RIF-resistant MTB cases in our study population limited our ability to fully evaluate the assays’ performance in detecting RIF resistance using clinical specimens. Despite these limitations, our study has several notable strengths. To our knowledge, this was the first prospective study conducted in a high-resource, intermediate TB-burden setting to directly compare Xpert and Xpert Ultra using fresh clinical specimens. This approach provided real-world performance data while allowing direct assessment of diagnostic accuracy. Furthermore, the use of mock samples enabled the evaluation of RIF resistance detection, reinforcing the applicability of Xpert Ultra in such settings.
In conclusion, Xpert Ultra tended to have increased sensitivity compared with that of Xpert in a high-resource, intermediate-burden setting, although the difference was not statistically significant. Its specificity was comparable with that of Xpert, although a few false-positive results were observed in trace- and very low-positive cases. Certain mutations located within the rpoB1 probe region led to indeterminate results in RIF resistance detection. Given the potential diagnostic ambiguity associated with Xpert Ultra trace- or very low-positive results, careful clinical interpretation is needed in the context of clinical and radiological findings, particularly in smear-negative cases.