Determinants of PCR performance (Xpert MTB/RIF), including bacterial load and inhibition, for TB diagnosis using specimens from different body compartments

The determinants of Xpert MTB/RIF sensitivity, a widely used PCR test for the diagnosis of tuberculosis (TB) are poorly understood. We compared culture time-to-positivity (TTP; a surrogate of bacterial load), MTB/RIF TB-specific and internal positive control (IPC)-specific CT values, and clinical characteristics in patients with suspected TB who provided expectorated (n = 438) or induced sputum (n = 128), tracheal aspirates (n = 71), bronchoalveolar lavage fluid (n = 152), pleural fluid (n = 76), cerebral spinal fluid (CSF; n = 152), pericardial fluid (n = 131), or urine (n = 173) specimens. Median bacterial load (TTP in days) was the strongest associate of MTB/RIF positivity in each fluid. TTP correlated with CT values in pulmonary specimens but not extrapulmonary specimens (Spearman's coefficient 0.5043 versus 0.1437; p = 0.030). Inhibition affected a greater proportion of pulmonary specimens than extrapulmonary specimens (IPC CT > 34: 6% (47/731) versus 1% (4/381; p < 0.0001). Pulmonary specimens had greater load than extrapulmonary specimens [TTPs (interquartile range) of 11 (7–16) versus 22 (18–33.5) days; p < 0.0001]. HIV-infection was associated with a decreased likelihood of MTB/RIF-positivity in pulmonary specimens but an increased likelihood in extrapulmonary specimens. Mycobacterial load, which displays significant variation across different body compartments, is the main determinant of MTB/RIF-positivity rather than PCR inhibition. MTB/RIF CT is a poor surrogate of load in extrapulmonary specimens.

Information regarding Xpert MTB/RIF's performance on nonsputum specimens is emerging 5,[13][14][15][16][17][18][19][20][21][22][23][24][25][26] ; however, it is not extensive, nor sufficiently validated in HIV-prevalent settings. MTB/RIF has thus been granted a conditional recommendation for the diagnosis of extrapulmonary TB by the WHO, however, the overall body of evidence has been cited as weak 9 . Furthermore, countries which are presently implementing it for the diagnosis of pulmonary TB, such as South Africa, do not currently permit its routine use on extrapulmonary specimens.
While the relationship between sputum bacillary load (measured using smear microscopy, culture, and MTB/RIF) has been previously characterised [27][28][29][30][31] , little is known about the comparative variation in mycobacillary load in fluids from different sites in the body, despite the high burden of extrapulmonary and increased risk of poor outcomes in these patients 32,33 . This is critical for informing the development and application of new tests for extrapulmonary TB (where, in some cases, a biomarker-based approach might be optimal). Furthermore, there is no information regarding how the performance of MTB/RIF is influenced by constituents of extrapulmonary specimens or any associated clinical factors. This is important, because salts, proteins or cellular debris are commonly found in non-sputum specimens and can be enriched after specimen processing (e.g., after centrifugation). These can interfere with the amplification enzyme and thereby inhibit the PCR, leading to inaccurate or unreliable results.
In this study, we first compared mycobacterial load in different fluids from different cohorts of patients with TB recruited from similar settings in South Africa (over 1000 patients overall). We identified clinical factors, including HIV co-infection and CD4 count, and specimen characteristics that may modulate liquid culture time-to-positivity (TTP) and MTB/RIF quantitative information [cycle threshold (C T ) values] in these fluids. We evaluated the degree of MTB/RIF PCR inhibition in each fluid, and how this modified the relationship between MTB/RIF and culture results.

Methods
Study information. We have performed a series of studies at the University of Cape Town and the University of KwaZulu-Natal that assessed the accuracy of MTB/RIF for the diagnosis of TB in different body fluids. These were performed in independent cohorts of patients who were clinically suspected of having pulmonary or extrapulmonary TB. Comparative data from these studies for the following specimens types are presented here: expectorated sputum from patients with suspected pulmonary TB attending primary care TB clinics in Cape Town, South African 34,35 ; induced sputum from sputum-scarce or smear-negative patients attending primary care TB clinics in Cape Town 36 ; tracheal aspirates from mechanically-ventilated patients in the intensive care unit of a tertiary level hospital (Groote Schuur Hospital) in Cape Town (#NCT01530568); bronchoalveolar lavage fluid (BALF) from sputumscarce or smear-negative patients attending the respiratory clinic at the same hospital 23 ; pleural fluid from patients with suspected pleural TB attending the same respiratory clinic; cerebral spinal fluid (CSF) from patients with suspected TB meningitis from Inkosi Albert Luthuli Central Hospital in Durban, South Africa 37,38 ; pericardial fluid from patients suspected of TB pericarditis from four district-and one tertiary-level hospital in South Africa 39 ; and urine from patients suspected of TB who are hospitalised in Groote Schuur Hospital 20 . Patients on anti-TB treatment longer than 48 hours were excluded from the analyses. Only patients with paired liquid culture and MTB/RIF results (i.e., from either the same specimen or specimens collected at the same time) were included.
Ethics statement. Each sub-study was approved by the University of Cape Town or University of Kwa-Zulu Natal research ethics committees, all patients provided written informed consent for participation and the use of their data, and each substudy was conducted in accordance with the relevant approvals.
Smear microscopy, liquid culture and Xpert MTB/RIF. When MTB/RIF was performed on sputum, a paired specimen was NALC-NaOH decontaminated, and the sediment used for concentrated fluorescent smear microscopy and liquid culture using the BACTEC MGIT 960 system (BD Diagnostics, USA) performed at a qualityassured accredited reference laboratory. For studies involving other specimen types (induced sputum, tracheal aspirates, BALF, pleural fluid, CSF, pericardial fluid, and urine), the same specimen used for MTB/RIF testing was used for smear microscopy and liquid culture after decontamination. A ,10 ml volume of urine 20 was first centrifuged and resuspended in 1 ml phosphate buffered saline prior to processing for MTB/RIF. As our study objectives are to compare bacterial load and MTB/RIF-inhibition in different fluids, which would be confounded by different methods of specimen concentration, all other specimens (other than urine) were processed raw and centrifuged, and a volume of 1 ml used. The recommended 2-fold volume of sample buffer was thereafter added and the MTB/RIF procedure started 40 .
Statistical analyses. Statistical analyses were performed using Graphpad Prism (version 6.0; GraphPad Software, USA, www.graphpad.com), the VassarStats online statistical package (www.vassarstats.net/index.html), and STATA SE (version 12; StataCorp, USA). P-values less than ,0.05 were considered significant. A backward elimination strategy was used for multivariate analyses of culture TTP, MTB/RIF Mycobacterium tuberculosis-specific C T values, and MTB/RIF inhibition. Variables with p-values ,0.100 in univariate analyses were included in the final multivariate model. Fisher's exact test with mid-P correction was used for comparisons between proportions. The Mann-Whitney test to compare medians. Fisher's z transformation was used to compare differences in Spearman's correlation coefficient between TTPs and C T values. For some within-specimen type comparisons of TTP and C T values, there were too few HIV-infected culture-positive patients (n # 5) for meaningful comparisons.

532)
Demographic characteristics  Correlates of time-to-positivity. Multivariable linear regression analyses of culture-positive patients showed the following clinical and demographic factors to be associated with increased TTP: younger age (p 5 0.035) and HIV-infection (p 5 0.047) for induced sputum (Table S2), previous TB (p 5 0.022) for tracheal aspirates (Table S3). No significant associations were found for the other fluids or pooled pulmonary data after multivariable adjustments were performed (see supplement).
Xpert MTB/RIF-generated cycle threshold values in different types of specimens. Overall. When median MTB/RIF-generated cycle threshold values (C T values; a smaller C T value indicates greater load) (IQR) were compared across fluids, those from pleural fluid, CSF, pericardial fluid, and urine were greater than expectrated sputum ( Figure 1C and D;    Table S10). When pooled pulmonary data were examined, patients who were HIV-infected (p 5 0.059) and had a longer TTP (p , 0.001) were less likely to be MTB/RIF-positive (Table S5). Extrapulmonary specimens with a longer TTP (p 5 0.003) were also less likely to be MTB/ RIF-positive and HIV-infection (p 5 0.013) was associated with an increased likelihood of MTB/RIF-positivity (Table S9).
Comparative PCR inhibition in different specimen types. Overall. Figure 3. Internal control C T values for expectorated sputum differed to those for induced sputum, tracheal aspirates, BALF, pleural fluid, and CSF, and were similar for pulmonary specimens and extrapulmonary specimens. The proportion of MTB/RIF results with an IPC C T value . 34, which have been shown to be due to inhibition in sputum 30    Correlates of inhibition. When multivariable linear regression analyses were performed (p # 0.100), female gender was associated with decreased internal control C T values for expectorated sputum (p 5 0.007; Table S1), HIV infection for CSF (p 5 0.078; Table S6), older age for pericardial fluid (p 5 0.089; Table S8), and reduced protein concentration in urine (p 5 0.072; Table S10). There was no asso-ciation between MTB/RIF positivity and internal control C T value for each of the other fluids tested (see supplement), however, when pulmonary data and extrapulmonary data were pooled, female gender (p 5 0.008) and younger age (p 5 0.076) were respectively associated with less inhibition (Table S4 and S9) for each specimen type respectively.  values were correlated (Spearman coefficients of 0.5043; p , 0.0001), however, there was no significant correlation amongst extrapulmonary specimens (Spearman coefficient of 0.1437; p 5 0.4032), and the correlation observed amongst pulmonary specimens was stronger (p 5 0.030). C T values correlated less strongly with low or high levels of bacterial load, rather than due to any intrinsic properties of extrapulmonary specimens. For example, although C T values and TTP exhibited a significant correlation overall for pulmonary specimens, this was not present amongst specimens with a TTP in the bottom (Spearman coefficient of 0.1805; p 5 0.1535) or top tertile (Spearman coefficient of 0.0899; p 5 0.5139), but was amongst those in the middle tertile (Spearman coefficient of 0.3758; p 5 0.0017).

Discussion
This study is the first to compare mycobacterial load using culture TTP, MTB/RIF-generated C T values, and MTB/RIF inhibition in specimens from different body compartments. Briefly, our key findings are: (i) compared to expectorated sputum, MTB/RIF is inhibited more in induced sputum, tracheal aspirates, and BALF, but less in pleural fluid; (ii) ''false-negative'' MTB/RIF results (MTB/RIF-negative, culture-positive) from CSF displayed a greater inhibition compared to ''true-positive'' results, and pulmonary specimens inhibited MTB/RIF more than extrapulmonary specimens; (iii) C T values correlate with TTP in pulmonary specimens but not in extrapulmonary specimens, suggesting the assay to be unsuitable for estimation of mycobacterial load amongst patients with extrapulmonary TB; (iv) TTP is the strongest correlate of MTB/RIF-positivity in both pulmonary specimens and extrapulmonary specimens, even after adjusting for inhibition; and (v) extrapulmonary specimens are more paucibacillary than pulmonary specimens and, of the pulmonary specimens, expectorated sputum had the highest bacillary load. We found pulmonary specimens to have a greater proportion of MTB/RIF results with evidence of inhibition [IPC C T value .34 30 ] compared to extrapulmonary specimens. It is likely that this is driven by the viscous nature of sputum which, even after the addition of sample buffer, may not be completed homogenised and thus still interefere with the reaciton. Importantly, the inhibitory effect caused by the viscous nature of some sputum specimens is likely offset by the thick mucous within it, which has been shown to contain over 30fold more bacilli than the watery component, and thus the overall sensitivity remains good 30 .
In our study, we found ''false-negative'' MTB/RIF results to display more inhibition on CSF than those that are ''true-positive'', suggesting that this fluid contains material that interferes significantly with the PCR and thus may be a cause of false-negative results. This is the first description of MTB/RIF inhibition in extrapulmonary specimens. Interestingly, we have shown in a separate study 41 that, if a 3 ml volume of CSF is centrifuged, the pellet washed, and resuspended in buffer prior to testing, the sensitivity of MTB/RIF improves by almost 40%. In addition to concentrating the bacilli in the specimen, this centrifugation and resuspension step likely also removes PCR inhibitors. Such an approach should be considered for other fluids that inhbit PCR-based tests. The other types of extrapulmonary specimens analysed did not display evidence of significant inhibition.
In pulmonary specimens, HIV-infection was associated with a decreased likelihood of a positive MTB/RIF result, however, in extrapulmonary specimens, HIV-infection was associated with an increased likelihood of MTB/RIF-posivity. This is reflective of the lower bacillary load seen in the lungs of HIV-coninfected patients with pulmonary TB (due to the lower frequency of caviation in these patients) compared to those who are HIV-uninfected. In contrast, patients who are HIV-infected displayed a higher TB bacillary load in specimens from extrapulmonary sites than those who were HIVuninfected, and thus those who are HIV-infected are more likely to be MTB/RIF-positive for EPTB. Although EPTB is more frequent in HIV-infected patients, their extrapulmonary bacillary load is lower than HIV-infected patients with pulmonary TB. This means that EPTB specimens with a concetration of bacilli below the limit of detection of MTB/RIF will occur more frequently, and that patients with suspected TB who have a negative MTB/RIF result should still be investigated further. A further rammification of the low load seen in extrapulmonary specimens is that in fluids such as pleural fluid or pericardial fluid a biomarker-based approach using a molecule such as interferon-c might be superior 43 to a nucleic acid amplication assay. Thus, MTB/RIF is not necessarily a ''one size fits all'', although it does universally outperform microscopy (the only alternative rapid test in some settings) 7 .
As has been documented by others 23,28,30,34,44,45 , we found MTB/ RIF-generated C T values to correlate significantly with culture timeto-positivity in pulmonary specimens. Such a conclusion is important because, for pulmonary TB, sputum bacillary load at diagnosis is one of the strongest baseline predictors of long-term outcome 46-49 ,  and could thus be used for the prognostication of patients. We now show there is no correlation with bacterial load in extrapulmonary specimens, and that this appears to be as a result of the C T values-TTP correlation deteriorating at low levels of bacterial load. While a direct association between baseline MTB/RIF C T values and clinic outcome has not yet been demonstrated, it appears that MTB/RIF would not be useful for such a purpose amongst patients with extrapulmonary TB, or for a meaningful estimation of disease severity as approximated by bacterial load.
This study has limitations. Although this is the first study to report on MTB/RIF inhibition in fluids other than sputum, we did not capture data on specimen-specific factors such as viscosity, appearance, or salt concentration, which may interefere with MTB/RIF. Although shown to be useful by us and other 25,41,42 , we did not assess bacterial load and inhibition in centrifuged specimens (other than urine), as these data were not available for all the specimen types included in this study and, when it were, different specimen volumes were used for concentration. Our analyses were also restricted in some instances by the comparatively small number of culture-positive specimens, especially after stratification by MTB/RIF-and/or HIV-status and specimen type; however, the size of the cohort in each of these parent studies is mostly in excess of that reported elsewhere 7 . While we 28,31 and others 30 have described how MTB/ RIF can be used to predict smear-positivty in sputum, we were unable to replicate such an analysis here, due to the small number of nonsputum specimens that were smear-positive. Our specimens of different types were stored for different durations, and this may have influenced some differences, however, recent work has demonstrated that MTB/RIF accuracy is not significantly affected by storage duration 7,18 , suggesting this effect, if any, to be minimal.
In summary, this study has demonstrated that low mycobacillary load in extrapulmonary specimens is, rather than inhibition, primarily responsible for the diminished sensitivity of MTB/RIF in these specimens compared to those from the pulmonary system. While ''false-negative'' CSF displayed more inhibition than ''true-positive'' specimens, pulmonary specimens displayed the most inhibition overall, suggesting that MTB/RIF quantitative information might not be useful in a significant minority of patients with suspected pulmonary TB. Furthermore, the quantitative information generated by MTB/RIF from extrapulmonary specimens does not correlate with bacterial load, and is unlikely to be useful. Future studies on the exact clinical and specimen-specific determinants of MTB/RIF inhibition are important, as well additional specimen preparation steps that may reduce inhibition, especially if MTB/RIF quantitative information will used for patient management.