JCM Accepted Manuscript Posted Online 16 September 2015 J. Clin. Microbiol. doi:10.1128/JCM.01568-15 Copyright © 2015, American Society for Microbiology. All Rights Reserved.
1
JCM01568-15 Revised
2
Metabolomic Profiling of Plasma from Patients with Tuberculosis Using
3
Untargeted Mass Spectrometry Reveals Novel Biomarkers for Diagnosis
4 5
Susanna K. P. Lau,a,b,c,d* Kim-Chung Lee,d Shirly O. T. Curreem,d Wang-Ngai Chow,d
6
Kelvin K. W. To,a,b,c,d Ivan F. N. Hung,b,e Deborah T. Y. Ho,d Siddharth Sridhar,d Iris W.
7
S. Li,d Vanessa S. Y. Ding,d Eleanor W. F. Koo,f Chi-Fong Wong,g Sidney Tam,h Ching-
8
Wan Lam,h Kwok-Yung Yuen,a,b,c,d Patrick C. Y. Wooa,b,c,d *
9 10
State Key Laboratory of Emerging Infectious Diseases,a Research Centre of Infection and
11
Immunology,b Carol Yu Centre for Infection,c Department of Microbiology,d Department
12
of Medicine,e The University of Hong Kong, Hong Kong; Department of Pathology,f
13
Tuberculosis & Chest Unit,g Grantham Hospital, Hong Kong; Department of Pathology,h
14
The University of Hong Kong, Hong Kong.
15 16
Running title: Novel biomarkers for tuberculosis
17
Keywords: metabolome, metabolomics, plasma, tuberculosis, mass spectrometry, novel,
18
biomarkers, diagnosis
19 20
*Corresponding author. Mailing address: Department of Microbiology, University
21
Pathology Building, Queen Mary Hospital Compound, Pokfulam Road, Hong Kong.
22
Phone: (852) 22554892. Fax: (852) 28551241. E-mail:
[email protected] (SKP Lau);
23
[email protected] (PCY Woo)
24 1
25
ABSTRACT (246 words)
26
Although tuberculosis (TB) is a re-emerging disease that affects developing
27
countries and immunocompromised populations in developed countries, current
28
diagnostic methods are far from optimal. Metabolomics is being increasingly used for
29
studies on infectious diseases. We performed metabolome profiling of plasma samples to
30
identify potential biomarkers for diagnosis of TB. We compared plasma metabolome
31
profiles of TB patients (n=46) to those of community-acquired pneumonia (CAP) patients
32
(n=30) and controls without active infection (n=30), using ultra-high-performance liquid
33
chromatography-electrospray ionization-quadrupole time-of-flight mass spectrometry
34
(UHPLC-ESI-QTOFMS). Using multi-variate and univariate analyses, four metabolites,
35
12(R)-HETE, ceramide (d18:1/16:0), cholesterol sulphate and 4α-formyl-4β-methyl-5α-
36
cholesta-8-en-3β-ol, with significantly higher levels in TB patients than CAP patients and
37
controls, were identified. When compared between TB patients and controls, the four
38
metabolites demonstrated AUC of 0.914, 0.912, 0.905 and 0.856, sensitivities of 84.8%,
39
84.8%, 87.0% and 89.1%, specificities of 90.0%, 86.7%, 86.7% and 80.0%, and fold-
40
changes of 4.19, 26.15, 6.09 and 1.83 respectively. When compared between TB and
41
CAP patients, they demonstrated AUC of 0.793, 0.717, 0.802 and 0.894, sensitivities of
42
89.1%, 71.7%, 80.4% and 84.8%, specificities of 63.3%, 66.7%, 70.0% and 83.3% and
43
fold-changes of 4.69, 3.82, 3.75 and 2.16 respectively. 4α-formyl-4β-methyl-5α-cholesta-
44
8-en-3β-ol combined with 12(R)-HETE or cholesterol sulphate offered ≥70% sensitivities
45
and ≥90% specificities for differentiating TB patients from controls or CAP patients.
46
These novel plasma biomarkers, especially 12(R)-HETE and 4α-formyl-4β-methyl-5α-
47
cholesta-8-en-3β-ol, or their combinations, are potentially useful for rapid and non-
2
48
invasive diagnosis of TB. The present findings may offer insights to the pathogenesis and
49
host response in TB.
50
3
51
INTRODUCTION
52
Tuberculosis (TB) is a disease caused by the bacterium Mycobacterium tuberculosis
53
(MTB). Although it is a well-known disease that has been around for much of human
54
history, there are still millions of new TB cases per year worldwide and it remains a
55
leading cause of deaths worldwide, especially in developing countries. Since the 1980s,
56
TB has re-emerged as a result of the acquired immunodeficiency syndrome epidemic and
57
increasing use of immunosuppressants. In recent years, a higher incidence of
58
extrapulmonary disease in immunocompromised hosts and emergence of multidrug-
59
resistant strains have further complicated diagnosis and treatment (1-3). Despite its
60
medical importance, diagnosis of TB is still associated with many unresolved problems.
61
The traditional gold standard is smear and culture for acid-fast bacilli from clinical
62
specimens. Although culture offers higher sensitivity and specificity than smear, it is not
63
useful for culture-negative cases especially in early, disseminated or extrapulmonary
64
disease (4, 5). Moreover, it often takes two to six weeks before culture is positive and
65
even longer for definitive species identification. While newer diagnostic modalities, such
66
as adenosine deaminase levels in pleural fluid, lipoarabinomannan in urine, polymerase
67
chain reaction (PCR) and Xpert MTB/RIF assays, have been developed (6-12), there are
68
still limitations in terms of their sensitivities and/or specificities.
69
Metabolomics is an emerging platform for studies of infectious diseases or
70
pathogens (13-19). For TB, the technique has been applied on cultured isolates for
71
differentiation from other Mycobacterium species and studies on the biology and
72
virulence (14, 15, 20-22). For example, lipidomics studies have revealed novel
73
metabolites potentially associated with growth and virulence of MTB (23, 24). We have
74
also recently identified extracellular metabolites specific to MTB (25), supporting the
4
75
potential of metabolomics in exploring novel biomarkers to better understand its biology
76
and pathogenesis. On the other hand, metabolomics applied on direct patient samples
77
may reveal specific diagnostic markers or monitor treatment response (26-29). It has been
78
shown that volatile organic compounds (VOCs) in the urine of TB patients can be
79
distinguished from those of healthy subjects (26). A study using nuclear magnetic
80
resonance spectroscopy-based metabolomics on sera of TB patients demonstrated
81
discrimination between patients and healthy controls (27). Another metabolomics study
82
has identified several abundant metabolites, including two mycobacterium-derived cell
83
wall glycolipids, in EDTA-plasma of TB patients compared to household contacts (28).
84
However, patients with other bacterial infections were not included as controls in these
85
studies, and thus, the potential of such metabolic profile for diagnosis of TB remains to
86
be determined.
87
To identify potential biomarkers for non-invasive diagnosis of TB, we applied
88
state-of-the-art technology for metabolomics profiling of plasma samples from TB
89
patients, using ultra-high-performance liquid chromatography-electrospray ionization-
90
quadrupole time-of-flight mass spectrometry (UHPLC-ESI-QTOFMS). Multi- and
91
univariate statistical analyses were used to identify metabolites with significantly higher
92
levels in plasma of TB patients than in plasma of patients with community-acquired
93
pneumonia (CAP) or controls without active infection. The diagnostic performances of
94
the identified biomarkers were assessed using whole-metabolome receiver operating
95
characteristic curve (ROC) analysis.
96
5
97
MATERIAL AND METHODS
98
Patient and control samples. Clinical samples were collected from hospitalized adult
99
patients (≥18 years of age) in Queen Mary Hospital, Hong Kong. A total of 46 plasma
100
samples from 37 patients with newly-diagnosed TB, and 60 plasma samples from 30
101
patients with CAP and 30 controls without active infection were recruited for UHPLC-
102
QTOFMS analysis. Plasma samples from TB patients were collected before
103
commencement of anti-mycobacterial treatment. Plasma samples collected at admission
104
from CAP patients were used. This study was approved by the Institutional Review
105
Board of the University of Hong Kong/Hospital Authority of Hong Kong West Cluster
106
(reference number UW 13-265).
107
Diagnostic criteria. The diagnosis of TB was based on compatible clinical
108
features, together with the presence of the following microbiological evidence: (i)
109
positive stain for acid-fast bacilli, (ii) positive culture for MTB, and/or (iii) positive PCR
110
for MTB from clinical samples. The diagnosis of CAP was based on compatible clinical
111
features and radiological evidence of lung infiltrates, with disease onset within 48 hours
112
of hospital admission. Causative agents of CAP included Acinetobacter baumannii (n=1),
113
Escherichia coli (n=2), Haemophilus influenzae (n=2), Klebisella pnuemoniae (n=1),
114
Moraxella catarrhalis (n=1), Pseudomonas aeruginosa (n=1), Streptococcus pneumoniae
115
(n=4) and influenza B virus (n=1), while the etiological agent was unknown for the
116
remaining cases. Controls consisted of patients without any clinical evidence of active
117
infection.
118
Chemicals and reagents. LC-MS grade water, methanol and acetonitrile were
119
purchased from J.T. Baker (Center Valley, PA, USA). HPLC-grade ethanol and acetone
120
were purchased from Merck (Darmstadt, Germany). Formic acid was of ACS reagent
6
121
grade from Sigma-Aldrich, (Saint Louis, MO, USA). Ceramide (d18:1/16:0) was
122
purchased from Avanti Polar Lipid (Alabaster, Alabama, USA). 12(R)-HETE and
123
Cholesterol sulphate standards were purchased from Cayman Chemical (Ann Arbor,
124
Michigan, USA).
125
Sample preparation. Blood samples were collected in heparin bottles,
126
transferred immediately to the laboratory, and centrifuged at 3000 rpm at 4°C for 10 min
127
to obtain the plasma fractions. For metabolomics analysis, 100 µl of plasma was thawed
128
at 4oC and plasma proteins were precipitated with 400 µl of methanol/ethanol/acetone
129
mixture at a ratio of 1:1:1 (v/v/v). The sample extract was vigorously vortexed for 1 min,
130
and centrifuged at 14,000 rpm at 4oC for 10 min. The supernatant was collected for
131
UHPLC-ESI-QTOFMS analysis. All specimens were immediately kept at -80oC until
132
analysis and stored within one week. The thawed specimens were analyzed within 48 h in
133
a random manner to prevent the batch effect.
134
Untargeted metabolomics profiling of patient plasma using UHPLC-ESI-
135
QTOFMS. The metabolomic profiling of plasma supernatants was performed using
136
Agilent 1290 Infinity UHPLC (Agilent Technologies, Waldbronn, Germany) coupled
137
with Agilent 6540 UHD Accurate-Mass QTOF system (Agilent Technologies, Santa
138
Clara, CA, USA) accompanied with a MassHunter Workstation software for QTOF
139
(version B.03.01 for Data Acquisition, Agilent Technologies, USA). Waters Acquity
140
UPLC BEH C18 column (2.1 × 100 mm, 1.7 μm) (Waters, Milford, MA, USA) was used
141
for the separation with the injection volume of 5 μl. The column and autosampler
142
temperature were maintained at 45°C and 10°C, respectively. The separation was
143
performed at a flow rate of 0.4 mL/min under a gradient program in which mobile phase
144
A was composed of LC-MS grade water containing 0.1% formic acid (v/v) and mobile
7
145
phase B was composed of acetonitrile. The gradient program was applied as follows: t =
146
0 min, 5% B; t = 0.5 min, 5% B; t = 7 min, 48% B; t = 20 min, 78% B; t = 27 min, 80%
147
B; t = 31 min, 99.5% B; t = 36.5 min, 99.5% B; t = 36.51 min, 5% B. The stop time was
148
40 min. The ESI mass spectra were acquired in both positive and negative ion modes
149
using Agilent Jet Stream ESI source with capillary voltages at +3800 V and -3500 V,
150
respectively. Other source conditions were kept constant in all the experiments as follow:
151
gas temperature was kept constant at 300°C, drying gas (nitrogen) was set at the rate of 7
152
L/min, and the pressure of nebulizer gas (nitrogen) was 40 psi. The sheath gas was kept at
153
a flow rate of 10 L/min at a temperature of 350oC. The voltages of the Fragmentor,
154
Skimmer 1, and OctopoleRFPeak were 135 V, 50 V and 500 V, respectively. The mass
155
data were collected between m/z 80 and 1700 at the acquisition rate of 2 scans per
156
second. Two reference masses at m/z 121.0509 (protonated molecular ion of C5H4N4) and
157
m/z 922.0098 (protonated molecular ion of C18H18O6N3P3F24) for positive mode, and m/z
158
119.0363 (deprotonated molecular ion of C5H4N4) and m/z 966.0007 (formate adduct of
159
C18H18O6N3P3F24) for negative mode were used as constant mass correction during LC-
160
MS run. Product ion scanning (PIS) experiments were performed with ACQUITY UPLC
161
I-class system coupled with Waters Synapt G2-Si QTOF system(Waters, Milford, MA,
162
USA) operating in both positive and negative electrospray ionization modes. The mass
163
data were collected between m/z 50 and 1000 at the acquisition rate of 3 scans per second.
164
The voltages of the capillary, sampling cone, and source offset were 3 kV, 30 V and 80 V,
165
respectively. Other source conditions were kept as follow: source temperature at 120oC,
166
desolvation temperature at 380oC, cone gas at 10 L/h, desolvation gas at 800 L/h and
167
nebulizer gas at 6.5 bar. Intermittent injections of leucine enkephalin as lock masses with
168
m/z 556.2771 at positive mode and m/z 554.2615 at negative mode at a concentration of
8
169
400 pg/µL in 50% acetonitrile at a flow rate of 6 µl/min were used for accurate mass
170
measurements. MS/MS analysis was performed using ultra-high purity argon at collision
171
energy at 10, 20 or 40 eV to generate the best quality of MS/MS spectra for putative
172
identification and structural elucidation of the significant metabolites.
173
Data processing and statistical analysis. Multivariate and univariate analysis
174
was performed to identify molecular features that discriminate TB patients from CAP
175
patients and controls. Multivariate analysis was applied to a total of 106 LC-MS data of
176
plasma samples from three groups (46, 30 and 30 samples from newly-diagnosed TB
177
patients, CAP patients and controls without active infection respectively). The raw LC-
178
MS data were converted into mzXML format using MSConvert (ProteoWizard) and
179
subsequently
180
(http://www.bioconductor.org/packages/2.8/bioc/html/xcms.html)
181
(http://www.r-project.org/), which adopted different peak detection and alignment as well
182
as data filtering with centWave algorithms. Data was further processed with
183
normalization, scaling, filtering and statistical analysis using MetaboAnalyst 3.0
184
(www.metaboanalyst.ca). The data were mean-centered, square root scaled and
185
normalized such that the sum of squares for each chromatogram equaled on for statistical
186
analysis (30). Insignificant features between TB patients and CAP patients or controls
187
were filtered out using both uni- and multi-variate analyses. For multivariate analysis,
188
principal component analysis (PCA) was performed for unsupervised analysis while
189
partial-least squares discrimination analysis (PLS-DA) was performed for supervised
190
analysis to identify features with discriminative power. A score plot was applied to
191
reduce the dimensionality of the data for grouping of the samples, in which each point in
192
the score point represented the individual sample and similar data sets exhibit clustering
processed
using
open-source
9
XCMS operating
package in
R
193
while different sets separate further apart. PLS-DA models were validated based on
194
accuracy, multiple correlation coefficient (R2) and cross-validated R2 (Q2) in cross-
195
validation. The significance of the biomarkers was ranked using the variable importance
196
in projection (VIP) score (>1) from the PLS-DA model that was responsible for
197
separation of TB patients from non-TB group, i.e. CAP patients and controls. Significant
198
features were further identified by significance analysis of microarray (SAM), in which
199
false discovery rate (FDR) was determined by running multiple tests on high-dimensional
200
data that distinguish between TB and non-TB groups, with FDR < 0.05 considered to be
201
significant (31, 32).
202
For univariate analysis, statistical significance of features was determined among
203
TB patients, CAP patients and controls without active infection using one-way ANOVA
204
with Fisher’s post-hoc test by MetaboAnalyst 3.0. P0.85 for
215
comparison between TB patients and controls without active infection were identified
216
and ROC curve analysis for the identified metabolites was further performed to compare
10
217
between TB and CAP patients. Box-whisker plots were generated and P values were
218
calculated by the Student’s t test using Analyse-it software (Analyse-it Software, Leeds,
219
United Kingdom).
220
Metabolite identification. Features with significant differences were selected for
221
PIS experiments. MS/MS spectra of the potential biomarkers and commercially available
222
reference standards, including ceramide (d18:1/16:0), 12(R)-HETE and cholesterol
223
sulphate, were processed using Waters MassLynx V4.1 software (Waters Corp, Milford,
224
MA). Potential molecular formula based on the accurate mass and isotopic pattern
225
recognitions of parent and fragment ions were generated. All putative identities were
226
confirmed by matching with entries in the METLIN database (http://metlin.scripps.edu/),
227
Human
228
(http://www.massbank.jp/), LipidMaps (http://www.lipidmaps.org/) including Mtb Lipid
229
database, KEGG database (http://www.genome.jp/kegg), MycoMass and MycoMap
230
databases
231
(http://www.brighamandwomens.org/research/depts/medicine/rheumatology/Labs/Moody
232
/default.aspx) using exact molecular weights, nitrogen rule or MS/MS fragmentation
233
pattern data and literature search. Efforts were made to distinguish metabolites from the
234
other isobaric compounds whenever possible by its elution order and virtue of difference
235
in fragmentation pattern corresponding to its structural characteristics. Putative identities
236
of the biomarkers were confirmed by comparing their chromatographic retention times
237
(RTs) and MS/MS spectra with those obtained from commercially available standards of
238
ceramide (d18:1/16:0), 12(R)-HETE and cholesterol sulphate.
Metabolome
Database
(HMDB)
239
11
(http://www.hmdb.ca/),
MassBank
240
RESULTS
241
Metabolomic profiling of plasma samples and omics-based statistical and
242
bioinformatic analysis. The metabolomes of 106 plasma samples from the three groups
243
(46 samples from newly-diagnosed TB patients, 30 samples from CAP patients and 30
244
samples from controls without active infection) were characterized and compared. A total
245
of 1626 and 1598 molecular features in positive and negative mode respectively were
246
obtained and subjected to statistical analysis using MetaboAnalyst 3.0 (Supplementary
247
Table 1 and 2). For multivariate analysis, PCA revealed that samples from TB patients
248
and CAP patients were clustered separately. However, ambiguous separation was
249
observed between TB patients and controls by this unsupervised method. Therefore,
250
supervised method using PLS-DA was performed, which showed that TB group can be
251
distinguished from non-TB group (CAP patients and controls), in both positive and
252
negative mode data (Fig. 1A and 1B), with accuracies of 99% and 97.2%, multiple
253
correlation coefficient of 92.5% and 88.6%, and cross-validated R2 of 82.7% and 77.4%,
254
respectively. Significant features for the separation between TB and non-TB groups were
255
ranked, yielding 242 and 243 features with VIP score >1 in positive and negative
256
ionization modes respectively. SAM methods were also used to select the most
257
discriminative features responsible for the separation between TB and non-TB groups,
258
yielding 233 and 324 significant features with FDR 0.85, two each in positive and negative modes respectively, were
267
identified, which exhibited up-regulation in TB patients compared to controls (Table 1).
268
The first metabolite, with m/z 538.5190 and retention time (RT) at 33.02 min in
269
positive mode, was identified as ceramide (d18:1/16:0) by the molecular formula and
270
MS/MS fragmentation pattern corresponding to its structural characteristics in Human
271
Metabolome Database (HMDB), and confirmed by matching RT and MS/MS spectrum
272
with commercially available authentic chemical standard of ceramide (d18:1/16:0) (Fig.
273
3A). The peaks at m/z 520.5091 and 502.4996 corresponded to fragments after neutral
274
loss of one water molecule and two water molecules from the parent ion, respectively.
275
The peak at m/z 264.2676 referred to the fragment after combined loss of two water
276
molecules and hexadecanoyl group.
277
The second metabolite, with m/z 429.3733 at RT 32.22 min in positive mode, was
278
identified as 4α-formyl-4β-methyl-5α-cholesta-8-en-3β-ol by the molecular formula and
279
MS/MS fragmentation pattern corresponding to its characteristics in HMDB (Fig. 3B).
280
The peak at m/z 411.3598 was the fragment after neutral loss of water molecule at C-3
281
while the peak at m/z 303.2339 represented the fragment after cleavage of bond between
282
C-17 and C-20 and loss of 4-β-methyl group. The peak at m/z 205.1228 was the fragment
283
resulted from cleavage of bond between C-9 and C-11 and between C-8 and C-14, while
284
the peak at m/z 165.0924 was the fragment resulted from cleavage of bond between C-9
285
and C-10 and between C-7 and C-8, and loss of 4-β-methyl group. However, authentic
286
chemical standard of 4α-formyl-4β-methyl-5α-cholesta-8-en-3β-ol was not commercially
287
available for comparison.
13
288
The third metabolite, with m/z 319.2274 at RT 12.73 min in negative mode, was
289
identified as 12-hydroxyeicosatetraenoic acid (12(R)-HETE) by matching the molecular
290
formula and MS/MS spectra provided in METLIN and LipidMaps, and confirmed by
291
matching RT and MS/MS spectrum using commercially available authentic chemical
292
standard of 12(R)-HETE (Fig. 3C). The peaks at m/z 301.2181 and 257.2274 were
293
fragments after neutral loss of one water molecule and combined loss of water molecule
294
at C-12 and formate group respectively. The peak at m/z 179.1069 was the fragment
295
resulted from cleavage of aliphatic group and neutral loss of water molecule at C-12.
296
The fourth metabolite, with m/z 465.3059 at RT 33.0 min in negative mode, was
297
identified as cholesterol sulphate by matching the molecular formula in HMDB, and
298
confirmed by matching RT and MS/MS spectrum using commercially available authentic
299
cholesterol sulphate (Fig. 3D). Two fragments, at m/z 79.9580 and 96.9594, referred to
300
SO3- and HSO4- ions respectively.
301
Diagnostic performance of metabolites. The AUC, sensitivity and specificity
302
for ROC curves calculated at optimal cutoffs for the four metabolites were summarized in
303
Table 2. Box-whisker plots were generated for the three groups (Fig. 4). Three
304
metabolites, ceramide (d18:1/16:0), 12(R)-HETE and cholesterol sulphate, demonstrated
305
AUC >0.9 when compared between TB patients and controls. 12(R)-HETE showed the
306
largest AUC 0.914, 84.8% sensitivity and 90.0% specificity when compared between TB
307
and controls; but showed relatively lower AUC 0.793, 89.1% sensitivity and 63.3%
308
specificity when compared between TB and CAP patients. Ceramide (d18:1/16:0) and
309
cholesterol sulphate demonstrated AUC of 0.912 and 0.905 respectively when compared
310
between TB and controls, but also showed relatively lower AUC of 0.717 and 0.802
311
respectively when compared between TB and CAP patients. In contrast, 4α-formyl-4β-
14
312
methyl-5α-cholesta-8-en-3β-ol demonstrated better discrimination between TB and CAP
313
patients with higher AUC 0.894, 84.8% sensitivity and 83.3% specificity, than between
314
TB and controls.
315
Using the same optimal cutoffs for each metabolite, we calculated the sensitivities
316
and specificities of different combinations of two or three of the four metabolites for
317
diagnosis of TB (Table 3). The specificities of combined metabolites were generally
318
higher than those of individual metabolites, while the sensitivities of combined
319
metabolites were generally lower than those of individual metabolites, for differentiating
320
TB patients and controls or CAP patients. Two combinations, 4α-formyl-4β-methyl-5α-
321
cholesta-8-en-3β-ol plus 12(R)-HETE, and 4α-formyl-4β-methyl-5α-cholesta-8-en-3β-ol
322
plus cholesterol sulphate, offered ≥70% sensitivities and ≥90% specificities for
323
differentiating TB patients from controls and CAP patients.
324
15
325
DISCUSSION
326
Using metabolomics approach, we identified four novel biomarkers with significantly
327
higher levels in plasma of TB patients. Among the four metabolites, 12(R)-HETE showed
328
the largest AUC 0.914, with 84.8% sensitivity, 90.0% specificity and fold-change 4.19,
329
when compared to controls. However, the AUC dropped to 0.793 with only 63.3%
330
specificity when compared to CAP patients. The diagnostic performance of ceramide
331
(d18:1/16:0) was also promising with AUC 0.912, 84.8% sensitivity, 86.7% specificity
332
and fold-change 26.15 when compared to controls. However, the AUC dropped to 0.717
333
with 71.7% sensitivity, 66.7% specificity and fold-change 3.82 when compared to CAP
334
patients. Similarly, cholesterol sulphate showed a relatively high AUC 0.905, 87.0%
335
sensitivity, 86.7% specificity and fold-change 6.09 when compared to controls. However,
336
the AUC dropped to 0.802 with 80.4% sensitivity, 70% specificity and fold-change 3.75
337
when compared to CAP patients. In contrast, 4α-formyl-4β-methyl-5α-cholesta-8-en-3β-
338
ol showed good diagnostic performance in distinguishing TB from both CAP patients and
339
controls, with AUCs >0.85 and >80% sensitivities/specificities, though with fold-change
340
only 2.16 and 1.83 respectively. Therefore, 12(R)-HETE, ceramide (d18:1/16:0) and
341
cholesterol sulphate should offer better diagnostic performance than 4α-formyl-4β-
342
methyl-5α-cholesta-8-en-3β-ol in differentiating between TB and controls. In particular,
343
the >20 fold-change of ceramide (d18:1/16:0) suggests that it may be a promising
344
candidate for diagnosis of TB. However, these biomarkers may also be elevated in CAP
345
patients, though not to levels as high as in TB patients. On the other hand, 4α-formyl-4β-
346
methyl-5α-cholesta-8-en-3β-ol may offer better discrimination between TB and CAP
347
patients. When 4α-formyl-4β-methyl-5α-cholesta-8-en-3β-ol is combined with 12(R)-
348
HETE or cholesterol sulphate, the specificities were ≥90% for differentiating TB from
16
349
both controls and CAP patients, although the sensitivities were