Xpert Ultra testing of blood in severe HIV-associated tuberculosis to detect and measure Mycobacterium tuberculosis blood stream infection: a diagnostic and disease biomarker cohort study

Summary Background Mycobacterium tuberculosis bloodstream infection is a leading cause of death in people living with HIV and disseminated bacillary load might be a key driver of disease severity. We aimed to assess Xpert MTB/RIF Ultra (Xpert Ultra) testing of blood as a diagnostic for M tuberculosis bloodstream infection and investigate cycle threshold as a quantitative disease biomarker. Methods In this cohort study, we obtained biobanked blood samples from a large and well characterised cohort of adult patients admitted to hospital in Western Cape, South Africa with suspected HIV-associated tuberculosis and a CD4 count less than 350 cells per μL. Patients already receiving antituberculosis therapy were excluded. Samples were obtained on recruitment within 72 h of admission to hospital, and patients were followed up for 12 weeks to determine survival. We tested the biobanked blood samples using the Xpert Ultra platform after lysis and wash processing of the blood. We assessed diagnostic yield (proportion of cases detected, with unavailable test results coded as negative) against a microbiological reference, both as a function of markers of critical-illness and compared with other rapid diagnostics (urine lipoarabinomannan and sputum Xpert). Quantitative blood Xpert Ultra results were evaluated as a disease biomarker by assessing association with disease phenotype defined by principal component analysis of 32 host-response markers. Prognostic value compared to other tuberculosis biomarkers was assessed using likelihood ratio testing of nested models predicting 12-week mortality. Findings Between Jan 16, 2014, and Oct 19, 2016, of the 659 participants recruited to the parent study, 582 had an available biobanked blood sample. 447 (77%) of 582 met the microbiological reference standard for tuberculosis diagnosis. Median CD4 count was 62 (IQR 221–33) cells per μL, and 123 (21%) of participants died by 12-weeks follow-up. Blood Xpert Ultra was positive in 165 (37%) of 447 participants with confirmed tuberculosis by the microbiological reference standard, with a diagnostic yield of 0·37 (95% CI 0·32–0·42). Diagnostic yield increased with lower CD4 count or haemoglobin, and outperformed urine lipoarabinomannan testing in participants with elevated venous lactate. Quantitative blood Xpert Ultra results were more closely associated with mortality than other tuberculosis biomarkers including blood culture, and urine lipoarabinomannan, or urine Xpert (all p<0·05). A principal component of clinical phenotype capturing markers of inflammation, tissue damage, and organ dysfunction was strongly associated with both blood Xpert-Ultra positivity (associated with a SD increase of 1·1 in PC score, p<0·0001) and cycle threshold (r= −0·5; p<0·0001). Interpretation Xpert Ultra testing of pre-processed blood could be used as a rapid diagnostic test in critically ill patients with suspected HIV-associated tuberculosis, while also giving additional prognostic information compared with other available markers. A dose–response relationship between quantitative blood Xpert Ultra results, host-response phenotype, and mortality risk adds to evidence that suggests M tuberculosis bloodstream infection bacillary load is causally related to outcomes. Funding Wellcome Trust, National Institute of Health Fogarty International Center, South African MRC, UK National Institute of Health Research, National Research Foundation of South Africa. Translations For the Xhosa and Afrikaans translations of the abstract see Supplementary Materials section.

numerator and denominator. This allows the differential difficulty in obtaining some samples to be incorporated in the later measure of diagnostic utility.
17 Any analyses of variability in diagnostic accuracy, distinguishing pre-specified from exploratory Analysis of variation in diagnostic yield by patient characteristics was a primary aim of the study and prespecified for patient covariates CD4 count, haemoglobin, lactate and survival, as detailed in "Diagnostic utility analysis" subsection of methods section. 18 Intended sample size and how it was determined Sample size was limited by availability of samples from parent study but was determined to give >90% power to detect a 10% absolute difference in diagnostic yield at the 0.95 confidence level comparing two diagnostics by binomial distribution.

Participants 19
Flow of participants, using a diagram Figure S3 20 Baseline demographic and clinical characteristics of participants Table 1 21a Distribution of severity of disease in those with the target condition Test performance is described as a function of disease severity markers as a primary aim of analysis: Figure 1 and table 2 21b Distribution of alternative diagnoses in those without the target condition Cross tabulation of the index test results (or their distribution) by the results of the reference standard Figure 1 and table 2 main manuscript 24 Estimates of diagnostic accuracy and their precision (such as 95% confidence intervals) Figure 1 and table 2 main manuscript 25 Any adverse events from performing the index test or the reference standard Index test performed after end of study so no ability to negatively influence patient care. There were no occurrences of significant adverse events related to venesection in the parent study.

Inclusion criteria
Exclusion criteria 1. HIV-associated tuberculosis suspected as a cause of admission to hospital by admission clinicians (recorded on differential diagnosis).

Bacterial blood culture SOP
For non-mycobacterial blood stream infection pathogen detection, bacterial blood cultures (5-10mL blood in BacT/ALERT® Culture Media) were performed by the parent study if the patient had not received antibiotics at the time of enrolment, and also by clinical teams managing patients when bacterial BSI was suspected. Automated detection of growth in BacT/ALERT® Microbial Detection System was further assessed for pathogen identification using the Groote Schuur microbiology department SOP. In brief, secondary cultures on selective media directed by Gram staining were performed (including but not limited to 2% bile aesculin, MacConkey, blood agar), use of biochemical testing, and use of Vitek 2 GN-ID and GP-ID cards as required.

Supplementary figures and tables
Index: Figure S1: Pre-clinical blood pre-processing method development Figure S2. Imputation model for "trace" positive blood Xpert-ultra samples Figure S3. Inclusion flow chart Figure S4. Qualitative & quantitative associations between tuberculosis detection modalities Figure S5. 32 variables association with blood Xpert Ultra positivity & Ct value (four sets of 16 panels). Figure S6. Association between blood Xpert-ultra on ordinal scale and 32 clinical/immunological markers Figure S7: Indirect comparison of blood Xpert-Ultra sensitivity in TB blood culture positive patients by sample storage condition. .442251] This method is more accurate than colony forming unit counting, because, unlike CFU counting, the flow cytometry method observes intact non-colony forming bacilli and bacilli aggregates (clumps). A known number of bacilli were then spiked into 1.5ml aliquots of a plasma derived from heparinised healthy volunteer blood samples. These were made up to a volume of 3ml by addition of 1.5ml PBS, and then centrifuged using different centrifuge parameters shown, with subsequent removal of 90% supernatant and resuspension to starting volume by pipetting. Bacilli in these processed samples were then enumerated using the flow cytometry method and compared to a control which did not undergo centrifugation. Three samples suspended in PBS (no plasma) were also processed as a comparator.
Proportion of bacilli recovered from spiked healthy volunteer blood after blood processed by full lysis-wash SOP used for the Xpert Ultra method (C ). M. bovis BCG bacilli grown and quantified as above were heat treated and stained with SYBR-gold (ThermoFisher, S-11494) before being spiked at known concentration into 3ml samples of healthy volunteer blood. Blood samples (n=12) were then processed using the lysis-wash protocol described in main manuscript for pre-processing blood for Xpert-Ultra testing. The number of bacilli in the resulting lysate were quantified using fluorescence microscopy amd compared to the original number spiked to measure the proportion recovered after any losses incurred from the blood lysis-wash protocol.
Estimating blood Xpert Ultra protocol limit of detection (LOD, D) and relationship between number of bacilli present and Ct value (E) using spiked healthy volunteer blood samples. A fully attenuated M. tuberculosis strain with auxotrophic mutations in leucine and pantothenate biosynthesis was grown and quantified using the flow cytometry method, before being spiked at varying concentrations (0, 10, 50, 100, 200, 800 bacilli per ml, each 3-6 reps) into 3ml aliquots of heparinised healthy volunteer blood. These samples were then processed using the blood Xpert Ultra protocol described in main manuscript (omitting freeze-thaw step). To estimate LOD where 90% of samples are expected to be positive, a logistic regression model was fit (D, black sigmoid curve) onto the raw data (D, grey points, where positive = 1 and negative = 0). The value of bacilli per ml with 90% predicted probability for detection by blood Xpert Ultra (D, intersection of dashed lines) was then estimated (= 40 bacilli per ml). 50%CI (30 to 50 bacilli per ml) for this estimate were generated by 1000-fold bootstrapping (D, shaded area). For spiked blood that was positive by Xpert Ultra testing, summary Ct values were extracted as per the method described in main manuscript, and compared to spiked bacilli per ml on log2 scale by linear regression (E, the evidence of higher variance at lower spiked bacilli counts has not been modelled). A doubling in bacilli count is estimated to result in a fall in Ct value of ~1.8, a slightly steeper fall than might have been expected.  Factor analysis describing the main dimensions of variation seen for qualitative test results (positive or negative) including where two samples were sent for the same test. The first dimension of variation (capturing 34% of total variance in test results between patients) separates patients with predominantly positive and negative test results; the second dimension of variation (explaining 22% variance) separates patients by compartment yielding positive results, with blood and urine diagnostics separating from sputum based diagnostics. Categories further from the origin are less frequent (higher variance) than those near the origin. C. Correlation between quantitative read-outs from tests (sputum and blood culture time to positivity, TTP; urine and sputum Xpert Ct values) and Variables were transformed to be approximately normal using log to base 2 (log2), square-root (sqrt), or boxcox (BC) transformations as indicated by suffix. D-dimer values for the cohort were from 2 assays measured on different scales; these were combined by standardising both sets of observed values to have mean 0 and sd 1 (ddimer_scaled variable). ALT = alanine aminotransferase; AST resid = residual variation in asparate transaminase independent of covariance with ALT; BRT = total bilirubin; CD4 = CD4 cell count; CRP = C-reactive protein; lymphocyte, monocyte, neutrophil & platelet variables = peripheral blood counts of these variables; SHCO3 = venous standard bicarbonate; IL = interleukin; IL1ra = interleukin-1 receptor antagonist; FGF = basic fibroblast growth factor; IP10 = interferon gamma-induced protein; MIP1a = macrophage inflammatory protein-1 alpha; MIP1b = macrophage inflammatory protein-1 beta; PDGF = platelet-derived growth factor-BB (PDGF); RANTES = regulated on activation, normal T cell expressed and secreted; tgfb1 = transforming growth factor beta-1.

Figure S6. Association between blood Xpert-ultra on ordinal scale and 32 clinical/immunological markers
Manhattan plots showing strength of association between 16 clinical and 16 immunological variables (same as figure S3) and blood Xpert-ultra result represented on an ordinal scale where negative test = 0 and levels 1-3 are defined by tertiles of Ct values from positive tests. P-values from Spearman's rank correlation test corrected for multiple comparison using Benjamini-Hochberg procedure to limit false discovery rate, indicated by q-values. Horizontal dashed line = q-value 0·05; variables above this line are significant at the 0·05 level after correction for multiple comparisons. ALT = alanine aminotransferase; AST resid = residual variation in asparate transaminase independent of covariance with ALT; BRT = total bilirubin; CD4 = CD4 cell count; CRP = C-reactive protein; lymphocyte, monocyte, neutrophil & platelet variables = peripheral blood counts of these variables; SHCO3 = venous standard bicarbonate; IL = interleukin; IL1ra = interleukin-1 receptor antagonist; FGF = basic fibroblast growth factor; IP10 = interferon gamma-induced protein; MIP1a = macrophage inflammatory protein-1 alpha; MIP1b = macrophage inflammatory protein-1 beta; PDGF = platelet-derived growth factor-BB (PDGF); RANTES = regulated on activation, normal T cell expressed and secreted; tgfb1= transforming growth factor beta-1 Figure S7: Indirect comparison of blood Xpert-Ultra sensitivity in TB blood culture positive patients by sample storage condition.
Post-hoc analysis in response to reviewer request. In the study presented in the manuscript, biobanked blood samples stored at -80oC were used for blood Xpert Ultra testing ("Longterm storage -80 o C"), which may have impacted on sensitivity of detection of M.tb. Three pilot studies separate from the main study were performed using blood processed for Xpert Ultra directed without storage, or in some cases blood stored for up to 14 days at -20 o C for batch processing ("Fresh"). Here the proportion of MTBBSI samples (defined by having a positive TB blood culture from concurrent blood sample) detected by a single blood Xpert Ultra test are compared across the studies. All the pilots using "fresh" blood for Xpert Ultra had higher apparent sensitivity for detecting MTBBSI. 95% CI for the proportion detected are indicated with error bars for each pilot and the main study. In a Bayesian mixed-effects binomial regression model allowing random effects by study on this data, posterior probability of lower sensitivity from longterm storage of blood was estimated to be 90%. The model predicts that on average fresh samples will have +20% absolute higher sensitivity compared to stored samples (95%CrI -17 to +79% difference).

Preface
This document reports two systematic reviews and meta-analyses of literature on use of nucleic acid amplification technology (NAAT) to detect M. tuberculosis in patient blood samples. The first meta-analysis is focused on blood NAAT for TB diagnosis; the second is on bloood NAAT to diagnose M. tuberculosis blood stream infection (MTBBSI).
The motivation for these reviews is to inform the "Evidence before this study" section of the Research in context box for the KDHTB blood Xpert-ultra manuscript submission to Lancet Microbe.
This document is an Rmarkdown knitted as a pdf. This means alll the code for the analysis from raw data to final figures is embedded (so 100% reproducible). The code chunks are suppressed for readability but available at github repository.
2 Systematic Review 1: TB diagnosis using blood NAAT

Aims & objectives
We want to systematically review literature on use of nucleic acid amplification technology (NAAT) to identify M. tuberculosis in patients' blood samples as a diagnostic for tuberculosis. Aims are to summarise: 1. What NAAT methods including blood pre-processing have been used for identifying M. tuberculosis in blood. 2. Reported sensitivity and specificity of blood M. tuberculosis NAAT for tuberculosis.
Systematic review and meta-analysis of studies using NAAT on blood to diagnose TB will be performed.

Inclusions & exclusions
Studies in which investigators used NAAT to identify M. tuberculosis in peripheral blood samples (whole blood or component) from patients identified prospectively with either suspected tuberculosis (e.g. cohort design) or confirmed tuberculosis diagnosis (e.g. case-control design) will be included.
Studies where it is unclear if patients were identified prospectively for blood NAAT testing (e.g. studies where inclusion was based on opportunistic receipt of a blood sample), studies where it is unclear what reference standard for tuberculosis diagnosis was, and studies reporting artificially spiked sample experiments (patient samples spiked with M. tuberculosis ex vivo) will be excluded.

Data for extraction
We will extract data on patient populations (adult versus paediatic, HIV status, TB prevalence, pulmonary versus extra-pulmonary, inpatient versus outpatient) and NAAT method (commercial v in-house, blood pre-processing, blood volume).

Analysis plan
Descriptive summaries using figures and tables. Bivariate random-e ects regression acounting for correlation between sensitivity and specificity will be used to summarised central tendancies and heterogeneity. These will be fit using a Baysian (MCMC) approach implemented with the package brms in R studio. Meta-regression on selected covariates will be performed using bivariate regression to test association between study and method covariates and diagnostic performance.

Bias assessment
Identified studies will be assessed for risk of bias using questions adapted from the QUADRA-2 tool: • Patient selection . Are methods of patient selection adequately described ( If answers to >1 or >3 of these questions are "no" or "unclear" risk of bias will be rated as moderate or high.

PubMed
Terms used in main PubMed search engine:

tuberculosis AND (blood OR mycobacteraemia OR "blood stream infection" OR bacteraemia OR bacillaemia) AND (NAAT OR PCR OR Xpert) AND diagnosis
This is translated by PubMed algorithm into an expanded search query which we have edited to remove irrelevant search terms (e.g. "blood" is linked to a range of haematology terms which are irrelavant and these have been removed by editing in the advanced search editor). This give a final expanded search query of:

Scopus
Scopus was searched with query:

( TITLE-ABS-KEY ( tuberculosis AND diagnosis ) AND TITLE-ABS-KEY ( pcr OR naat OR xpert ) AND TITLE-ABS-KEY ( blood OR mycobacteraemia OR bacteraemia OR bacillaemia OR "blood steam infection" ) )
Returning 537 results on 12/12/2020; these are exported as a .bib file using [select all->BibTeX export, including abstract] saved as scopus.bib in the working directory.

Combining search results, removing duplicates
The .bib files are read in and combined, with duplicates identified by doi and removed.

Screening results
Summary of identified, screened, eligible and included study shown in PRISMA flow diagram in figure 1.

Description of included studies
Characteristics of review 1 included studies are shown in figure 2 (study design, setting, patient population/cases, assessed risk of bias) and figure 3 (NAAT methods). A wide variety of study designs, patient populations and NAAT methods have been reported, with little replication of specific approaches. Initial reports of blood NAAT for TB diagnosis had a peak in the 1990s then a relative hiatus, followed by an increase again in last 10 years.

Descriptive summaries of reported diagnostic performance
Reported diagnostic performance in included studies are shown in figure 4-6. There are several striking findings. First, there is very marked heterogeneity in sensitivity (ranging from 0 to 100%), with very little evidence of correlation between sensitivity and specificity. Heterogeneity in sensitivity seems unrelated to study design, setting, HIV prevalance in study, or NAAT methods (blood volume, method of blood preprocessng) other than the finding that the studies using Xpert Rif/MTB had amongst the lowest reported sensitivities. Average reported sensitivity of blood NAAT seems to have slightly decreased over time since initial reports in 1990s.
However, reported sensitivity does appear to vary somewhat by assessed risk of bias, with studies assessed as lower risk of bias, and larger studies, reporting lower sensitivity.
There is also evidence of reporting bias with an asymetrical funnel plot suggesting smaller studies with high sensitivity are over represented (figure 7).

Bivariate regression modelling
Formal bivariate modelling of the reported sensitivities and specificities largely confirms impressions from descriptive plots in previous section. Heterogeneity in reported sensitivity is extreme, with the 90% prediction interval (in which the model estimates 90% of studies from the "population" of studies will lie) encompassing nearly all possible values of sensitivity (90%PI = 0.074 to 0.97).

Meta-regression: covariates versus diagnostic performance
Association of study-level covariates with reported diagnostic performance of blood NAAT was formally investigated using bivariate random-e ects modelling.
These was some evidence that reported sensitivity was associated with risk of bias assessed through adapted QUADAS-2: studies assessed as low risk of bias reported lower sensitivity on average (figure 9). Posterior probability that low-bias risk studies had lower reported sensitivity than 'high or moderate' bias risk studies was 94%.
There was also a 95% posterior probability that larger sample size studies reported lower sensitivity than smaller sample size studies (figure 10).
There was no significant evidence that reported sensitivity for TB diagnosis was improving over time as NAAT technologies have evolved (rather there was a 'non-significant' weak downward trend, with 73% posterior probability that sensitivity was decreasing by year of publication (figure 10).
Proportion of study participants who were HIV positive was not convincingly related to reported diagnostic performance of NAAT with 68% posterior probability that sensitivity was higher in studies recruiting more patients who were HIV positive (figure 10).   Figure 10: Conditional e ects of 3 study-level covariates on reported performance in blood NAAT studies for TB diagnosis (review 1).

Summary review 1
Since the 1990s dozens of reports describing use of NAAT on patient blood samples for TB diagnosis have been published, with extreme heterogeneity in reported sensitivity and specificity, not obviously related to plausible biological or technical covariates. Most studies are poorly reported and are assessed to have high risk of bias. Mostly, in house PCR methods have been used with a wide variety of specific methodologies. Promising results in smaller reports have not been replicated in larger, low-bias studies. Results from scalable / widely available commercial PCR platforms have been disappointing.

Introduction
Marked heterogeneity in sensitivity of blood NAAT to diagnose tuberculosis could be related to disease spectrum of included TB cases in studies, not captured by gross study-level covariates such as inpatient v outpatient and proportion of patients HIV positive examined in review 1 above. We know that severity of HIV-associated tuberculosis is closely related to presence of M. tuberculosis blood stream infection (MTBBSI) and the presence of MTBBSI is clearly a plausible determinant of probability that TB is detected by blood NAAT testing. Detection of MTBBSI by blood culture is therefore a useful reference standard against which blood NAAT detection of tuberculosis can be assessed. We hypothesised that comparing blood NAAT to blood culture for detection of M.tb could resolve some of the heterogeneity in reported sensitivity by accounting for variance in disease spectrum of recruited cases in di erent studies, and therefore proposed a second systematic review limited to studies which peformed both blood NAAT and blood culture for detection of MTBBSI.

Aims and objectives
Objective is to summarise reported sensitivity of blood NAAT compared to mycobacterial blood culture. Aims: 1. Summarise reported relative sensitivity of blood TB-NAAT and TB blood culture against an external reference standard. 2. Summarise reported sensitivity of blood TB-NAAT against TB blood culture as the reference standard.
Note that specificity estimation is not an objective: this is justified as heterogeneity in sensitivity in systematic review 1 is the problem to be addressed.

Included studies
Included studies will be the subset of studies in review 1 that also performed a TB blood culture (liquid or solid media).
Studies which do not report results such that a 2x2 table cross-tabulating blood culture and NAAT results could be extracted were excluded.
[post hoc protocol edit:] Studies where the TB reference standard was based on TB blood culture were excluded for aim 1 but were retained in aim 2 analysis].

Screening results
From the 42 studies identified for review 1, 16 performed a mycobacterial blood culture; 2 of these did not report the results such that a 2x2 cross-tabulation of blood culture and blood NAAT could be extracted, leaving 14 studies with both blood culture and blood NAAT for inclusion in review 2 (figure 11, PRISMA flow chart). Of these 14 studies, 5 used TB blood culture result as the reference standard for TB diagnosis, meaning n=9 studies were available for aim 1. In one study all TB blood cultures were negative, meaning n=13 studies were available for aim 2 analysis.

Comparing sensitivity of blood TB-NAAT and TB blood culture against an external reference standard
In studies reporting sensitivity of blood NAAT and bood culture for TB diagnosis against an external reference standard there was, as expected, correlation between the sensitivities of the two methods across studies (r = 0.48 estimated from bivariate mixed-e ects regression). Most studies reported higher sensitivity of NAAT compared to culture, but with substantial heterogeneity resulting in uncertainty and 95% credibility intervals encompassing both better and worse sensitiviy for NAAT ( figure 12). Larger sample size and lower risk-of-bias studies reported lower relative sensitivity of NAAT on average, but the number of studies available did not support formal meta-regression.  Blood NAAT sensitivity − blood culture sensitivity Posterior density Figure 12: Bivariate mixed-e ects regression relative sensitivity of TB blood culture and blood NAAT versus an external reference standard for TB diagnosis (review 2, aim 1). Left panel shows individual studies raw data for sensitivities (grey circles) and the 95%CrI for the median population value (red shaded area) and 90% prediction intervals for a new unobserved study (red lines). Right panel shows distribution of posterior estimates from the model for di erence in sensitivity (estimate for sensitivity of NAAT minus sensitivity of culture); most probability was asigned by the model to higher sensitivity of NAAT, to the right of vertical black line.

Assessing sensitivity of blood TB-NAAT against TB blood culture as the reference standard
13 studies have reported TB blood culture and TB blood NAAT results in same patients. Estimated sensitivity of TB blood NAAT for TB blood culture cases (population median across all 13 studies, figure 13) was 0.7, but with substantial uncertainty for this population estimate (95% CrI 0.39 to 0.94, and 90% prediction interval for a new, unobserved study 0.07 to 0.99) due to heterogeneity and the limited amount of published data (in total, blood TB NAAT results have only been reported for 174 TB blood culture positive patients, with median of 7 patients per study).
This means there is limited power to support meta-regression. More recent studies, studies using commercial NAAT kits, and studies assessed to be at lower risk of bais all had lower reported sesnitivity of NAAT for TB blood culture positive disease but none of these associations reached a >95% posterior probability statistical significance level (table 1). Sensitivity for TB blood culture +ve disease Figure 13: Mixed-e ects regression blood NAAT sensitivity for detection of blood culture positive TB (review 2, aim 2). Fit and 95%CrI for individiual studies shown with green dots and whisters; estimated population median and 95%CrI shown with vertical solid and dashed lines respectively; 90% prediction intervals for a new unobserved study indicated by shaded green area.

Summary review 2
Published data on blood NAAT for TB diagnosis where sensitivity can be related to a concomitant TB blood culture is sparse. This data was reviewed because sensitivity of blood TB NAAT relative to TB blood culture allows a degree of adjustment for disease spectrum which we hypothesised might underlie the extreme heterogeneity described in review 1. However, variance in reported sensitivity of TB blood NAAT was still pronounced relative to TB blood culture and within the strata of patients who were TB blood culture positive. Reported sensitivity of TB blood NAAT was again lower on average in low bias rated studies, studies using commercial NAAT kits, and in more recently reported studies compared to initial reports in 1990s; these associations were not statistically significant, which may relate to the limited amount of data available in review 2.