JCM Accepts, published online ahead of print on 9 April 2014 J. Clin. Microbiol. doi:10.1128/JCM.00238-14 Copyright © 2014, American Society for Microbiology. All Rights Reserved.
Revised Manuscript 1
A simplified microarray system for simultaneously detecting rifampin,
2
isoniazid, ethambutol, and streptomycin resistance markers in Mycobacterium
3
tuberculosis.
4 5 6
Yvonne Linger1, Alexander Kukhtin1, Julia Golova1, Alexander Perov1, Amine Lambarqui1, Lexi
7
Bryant1, George B. Rudy1, Kim Dionne2, Stefanie L. Fisher2, Thomas M. Shinnick3, Nicole
8
Parrish2, and Darrell P. Chandler1#
9 10 11
1
Akonni Biosystems, Inc., Frederick MD 21701, 2The Centers for Disease Control and
12
Prevention, Atlanta, GA 30333, and 3The Johns Hopkins School of Medicine, Baltimore, MD
13
21287
14 15
#
To whom correspondence may be addressed: [email protected]
16 17
Running title: MDR-TB microarray
18 19
Submitted to: Journal of Clinical Microbiology
1
Revised Manuscript 20
ABSTRACT
21 22
We developed a simplified microarray test for detecting and identifying mutations in rpoB, katG,
23
inhA, embB and rpsL, and compared the analytical performance of the test relative to phenotypic
24
DST. The analytical sensitivity was estimated to be at least 110 genome copies per amplification
25
reaction. The microarray test correctly detected 95.2% of those mutations for which there was a
26
sequence-specific probe on the microarray, and 100% of 96 wildtype sequences. In a blinded
27
analysis of 153 clinical isolates, microarray sensitivity for first line drugs relative to phenotypic
28
DST (true resistance) was 100% for RIF (14/14), 90.0% for INH (36/40), 70% for EMB (7/10),
29
and 89.1% (57/64) combined. Microarray specificity (true susceptibility) for first line agents
30
was 95.0% for RIF (132/139), 98.2% for INH (111/113) and 98.6% for EMB (141/143). Overall
31
microarray specificity for RIF, INH, and EMB combined was 97.2% (384/395). The overall
32
positive and negative predictive values for RIF, INH and EMB combined were 84.9% and
33
98.3%, respectively. For the second-line drug STR, overall concordance between agar
34
proportion and the microarray was 89.5% (137/153). Sensitivity was 34.8% (8/23) because of
35
limited microarray coverage for STR-conferring mutations, and specificity was 99.2% (129/130).
36
All false susceptible discrepants were a consequence of DNA mutations that are not represented
37
by a specific microarray probe. There were zero invalid results from 220 total tests. The
38
simplified microarray system is suitable for detecting resistance-conferring mutations in clinical
39
MTB isolates, and can now be used for prospective trials or integrated into an all-in-one, closed-
40
amplicon consumable.
41
2
Revised Manuscript 42
INTRODUCTION
43 44
Mycobacterium tuberculosis (TB) infects one-third of the world’s population, with
45
approximately nine million new cases and two million deaths attributable to the disease each
46
year (33). Early case detection and rapid treatment are considered the most effective control
47
strategies to reduce TB transmission (17), especially in those cases involving multi- or
48
extensively drug resistant TB (MDR- and XDR-TB). Culture-based methods remain the gold
49
standard for diagnosing drug resistant TB, but can take several weeks or months to complete.
50
Thus, molecular-based drug susceptibility tests are becoming increasingly attractive as MDR-TB
51
diagnostic tools in order to initiate individualized, patient-appropriate treatment in a timely
52
manner.
53 54
Technologies such as Cepheid’s GeneXpert and Hain MDR-TB line probe assays reduce the
55
time to diagnosis for many TB patients, provide a rapid read-out indicating resistance to rifampin
56
or selected mutations conferring resistance to other first- or second-line drugs, and illustrate the
57
potential to deploy molecular tests closer to the point of need (34). However, the number of
58
known genes and resistance-conferring mutations to first- and second-line drugs greatly exceeds
59
the multiplexing capacity of these platforms, which may limit their clinical efficacy in the
60
treatment and control of multi- or extensively drug resistant TB. Planar and suspension
61
microarrays are well suited to address the multiple gene, multiple mutation challenge of
62
diagnosing drug resistant TB (4, 12, 13, 21, 24, 25, 30, 31, 38), but clinical adoption of
63
microarray technology is hampered by poor reproducibility (9, 20, 39), complex workflows,
64
and/or extensive user subjectivity and involvement in image and data analysis (15). In order to
3
Revised Manuscript 65
translate microarrays into efficacious TB diagnostics at the point of need, it is therefore
66
necessary to simplify user interaction with the technology while retaining the ability to detect
67
multiple genes and multiple mutations in a timely manner. The objectives of this study were to
68
develop a gel element microarray test for MDR-TB at a level of coverage surpassing what is
69
currently available with WHO-endorsed molecular platforms, estimate the analytical specificity
70
of the test on MDR-TB isolates of known genotype and phenotype, and compare the
71
performance of the test to conventional drug susceptibility testing as a precursor to integrating
72
the method into an entirely closed amplicon consumable (5, 7) and sample-to-answer system.
73 74
MATERIALS AND METHODS
75 76
Isolates and positive controls
77 78
The set of M. tuberculosis isolates and nucleic acids utilized for this study represents a global
79
and historical distribution of drug susceptible, drug resistant, and multi-drug resistant strains.
80
Materials from the Centers for Disease Control and Prevention (CDC, Atlanta, Georgia) and the
81
TDR Tuberculosis Strain Bank (29) were provided as heat-killed crude lysates. Available
82
isolates only covered 22 of 39 resistance-conferring mutations that were present on the gel
83
element array. Purified DNA from M. tuberculosis H37Ra was acquired from the American
84
Type Culture Collection (ATCC, Manassas, VA) and quantified on a Nanodrop 3000
85
fluorometer before use. CDC and TDR materials were used to verify the analytical specificity of
86
microarray probes before verifying array performance on blinded samples from the Johns
87
Hopkins Hospital (JHH) clinical isolate archive. M13mp18 single-stranded DNA was purchased
4
Revised Manuscript 88
from New England Biolabs (Ipswitch, MA) and diluted to 250 pg mL-1 in 10 mM Tris, 1 mM
89
Na2EDTA pH 7.5. Crude lysates and purified nucleic acids were stored at -20°C until use.
90 91
The Johns Hopkins Hospital (JHH) M. tuberculosis isolate archive consists of susceptible, DR-,
92
MDR-, and XDR- M. tuberculosis strains from the mid-Atlantic, south, and southwest regions of
93
the United States, Mexico, South America, Cameroon, and Kazakhstan. Isolates used for this
94
study were collected between 1995 and 2011. JHH isolates were cultivated under BSL-3
95
conditions on Lowenstein-Jensen (L-J) agar slants and maintained at 37°C in 5% CO2. Crude
96
lysates were prepared from each strain by harvesting a small amount of cells from L-J slants,
97
resuspending the cells in Middlebrook 7H9 broth, and adjusting the turbidity to a 0.5 McFarland
98
standard (ca. 1.5 x 108 CFU mL-1). Subsequently, 1 ml of each standardized suspension was
99
centrifuged for 5 minutes at 13,000 x g and the supernatant removed. Cell pellets were
100
resuspended in 200 µL of molecular grade water, heat killed at 95°C for 30 minutes, sonicated
101
for 15 minutes, and centrifuged for 5 minutes at 13,000 x g. A 100 μL aliquot of each
102
supernatant was then removed and stored at -20°C until use, representing approximately 7.5 x
103
107 CFU equivalents of bacteria and 3.3 ng DNA μL-1 assuming a genome size of 4 x106 bp and
104
one bacterium per CFU. PCR setup with crude lysates was performed under BSL-2 conditions,
105
and all other manipulations were performed under BSL-1 controls with universal PCR
106
precautions.
107 108
Phenotypic drug susceptibility testing
109 110
Each JHH isolate was tested for phenotypic resistance to rifampin (RIF), isoniazid (INH),
111
ethambutol (EMB), and streptomycin (STR) using the MGIT-960 system as per the 5
Revised Manuscript 112
manufacturer’s recommendations (Becton Dickinson, Sparks, MD), and 1% agar proportion as
113
per the Clinical Laboratory Standards Institute (CLSI) M24-A2 standards (37). The MGIT-960
114
and agar proportion methods both have established breakpoints for drug susceptibility and
115
resistance determinations. For a select subset of strains (n = 78), the TREK Sensititre™
116
MYCOTB assay (TREK Diagnostic Systems, Cleveland, OH) was used to establish a minimum
117
inhibitory concentration (MIC) for each drug.
118 119
Primers, probes, and synthetic DNA standards
120 121
Amplification primers and microarray probes were designed against M. tuberculosis genes and
122
target codons shown in Table 1. Genes and codons represent high-confidence mutations known
123
to confer a drug resistance phenotype against RIF, INH, EMB, or STR. Nine PCR primer pairs
124
were designed to work together in a multiplex, asymmetric master mix, where one of the primers
125
in each pair was synthesized with a Cy3 label and the resulting single-stranded, labeled amplicon
126
captured by one or more immobilized microarray probes. Only one amplification reaction is
127
required per test. PCR primers were synthesized by standard phosphoramidite chemistry at
128
Akonni or Eurofins MWG Operon (Huntsville, AL), HPLC purified, and quantified by UV
129
absorption before use. Resulting Cy3-labeled, asymmetric amplicons ranged from 92 to 139 nt
130
in length.
131 132
Microarray probes were synthesized by Akonni with a custom 3’ linker and purified to >90%
133
purity by HPLC before use. Probe purity was measured by electrospray ionization mass
134
spectrometry. Each microarray contained a universal hybridization probe for each functional
6
Revised Manuscript 135
gene and primer pair to verify that M. tuberculosis gene targets were amplified from each
136
sample. At least one matched pair of microarray probes (wildtype (WT) and single-nucleotide
137
mutant (MU)) was included for each mutation of interest. Control probes included a Cy3 beacon
138
for manufacturing quality control and positional reference, a synthetic hybridization control
139
probe with no known homology to any naturally-occurring sequence, and a probe for the M13
140
internal amplification and inhibition control. A 92-base, Cy3-labeled hybridization control target
141
was synthesized by TriLink Biotechnologies (San Diego, CA), HPLC purified to >90% purity,
142
and diluted to 100 pg mL-1 in 10 mM Tris, 1 mM Na2EDTA pH 7.5.
143 144
Microarray manufacture
145 146
Gel element arrays were manufactured on custom-coated glass substrates by a single-step
147
copolymerization procedure using a 4% 2-hydroxyethyl methylacrylamide copolymer,
148
essentially as described in (11). Cy3 beacons were immobilized at 1 μM and all other probes
149
immobilized at 50 μM concentration. Up to three microarrays were printed on each glass
150
substrate, but only one microarray is required per test. All microarrays were visually inspected
151
for gel element presence and uniform morphology before use. Microarrays were stored at room
152
temperature in a dust-free container until use (< 1 months). Immediately before use, a 50 μL
153
gasket was applied around each microarray.
154 155
Multiplex, asymmetric amplification
156
7
Revised Manuscript 157
Amplification reactions were assembled in 0.2 mL thin-walled PCR tubes and cycled on Quanta
158
Biotech QB-96 thermal cyclers (Surrey, United Kingdom). The nine pairs of forward and
159
reverse primers were combined in 1:10 ratios and prepared as a concentrated master mix to
160
achieve a final primer concentration of 0.04 to 0.4 μM each per reaction. Each 50 μL reaction
161
contained 3 μL of template (crude lysate or purified DNA), 25 μL Qiagen Multiplex PCR Kit
162
all-in-one amplification buffer and enzyme (Valencia, CA), 18 μL asymmetric PCR primer mix,
163
5% dimethyl sulfoxide (Sigma-Aldrich, St. Louis, MO) and 0.6 μg μL-1 non-acetylated bovine
164
serum albumin (Sigma-Aldrich). One microliter of M13 internal positive control (250 fg total)
165
was included in all reactions, unless otherwise noted. The balance of each reaction was made up
166
with sterile, ultra-pure water as required. No template controls containing the M13 internal
167
positive control were processed in parallel for each set of experiments. The thermal cycling
168
program consisted of an initial denaturation at 95°C for 15 min, followed by 40 cycles of [94°C
169
for 30 sec, 60°C for 60 sec, 72°C for 30 sec], and a final extension at 72°C for 3 min. The actual
170
run time for thermal cycling was ~ 2 hrs. Amplified material was either used immediately for
171
microarray hybridization, or stored at -20°C until use.
172 173
Hybridization and detection
174 175
After amplification, 20 μL of unpurified amplicon was combined with 30 μL of a concentrated
176
hybridization master mix to achieve a final reaction mixture consisting of 1M guanidine
177
thiocyanate, 50 mM HEPES pH 7.5, 5 mM Na2EDTA pH 8.0, 3.3 μg μL-1 non-acetylated BSA,
178
and 4 fmol µL-1 Cy3-labeled internal hybridization control. Forty-nine microliters (49 μL) of the
179
resulting mixture was applied directly to the center of each microarray, microarrays covered with
8
Revised Manuscript 180
a Parafilm coverslip, and hybridized for 3 hr at 55ºC on an MJ Research PTC-225 slide tower.
181
After hybridization, cover slips and gaskets were removed and up to 24 microarrays washed
182
simultaneously in 350 mL 1X SSPE (150 mM NaCl, 10 mM NaH2PO4 • H2O, 10 mM
183
Na2EDTA), 0.01% Triton X-100, pH 7.4 for 10 min with gentle agitation. Arrays were then
184
dipped twice in ultra-pure water and dried with compressed air.
185 186
Imaging and data analysis
187 188
Dried microarrays were imaged on an Akonni field-portable microarray analyzer, a very simple
189
imager consisting of a high-intensity green light emitting diode (LED), custom optics, and non-
190
cooled CCD camera. Image acquisition ranged from 0.05 to 0.1 sec per array, and pixel data
191
were acquired from raw .tif images using TIGR Spotfinder software and a fixed-circle algorithm
192
(mask size = 25 pixel diameter) with all background pixels included in the analysis (Top
193
Background Cutoff value in Spotfinder set to zero). Local background was subtracted from the
194
signal of each gel element. Integrated (background-corrected) signal intensity data were
195
exported as tab-delimited text, and imported into custom Excel spreadsheets for data analysis.
196
Signal to noise ratios (SNR values) were calculated using the background corrected signal from
197
gel elements and normalizing against 3 times the standard deviation of the average local
198
background.
199 200
A test was only declared valid if the average Cy3 beacon, internal hybridization control, and
201
M13 amplification/inhibition control probes all generated SNR values ≥ 3. Otherwise, the test
202
was deemed invalid and test results were not reported. The IS6110 and IS1245 probes are
9
Revised Manuscript 203
internal positive controls, a proxy for detecting members of the M. tuberculosis and M. avium
204
complexes, respectively, and are used to condition the report for RIF, INH, EMB and STR
205
resistance-conferring mutations. Positive detection of both the IS6110 and IS1245 probes is
206
considered a mixed sample. For all valid tests, if the IS1245 probe for M. avium complex
207
(MAC) is detected, then no genotyping results are reported for the M. tuberculosis resistance-
208
conferring mutations. Otherwise, discrimination ratios are calculated for valid tests as D =
209
(SNRWT - SNRMU) / (SNRWT + SNRMU). As currently implemented, a zero-copy IS6110 isolate
210
will still generate discrimination ratios for M. tuberculosis resistance-conferring mutations,
211
because the PCR primers for rpoB, inhA, katG, embB, and rpsL are specific for M. tuberculosis.
212
For this study, any value of D ≥ 0 was interpreted as a wild type sequence at the targeted
213
nucleotide position, and any value of D < 0 was interpreted as a mutation at the targeted
214
nucleotide position. Results from this study were intended, in part, to understand and refine
215
values for D that should correspond to an inconclusive test result before automating the image
216
and data analysis algorithms for a clinical user, and prospectively testing the microarray with
217
decontaminated sediments or raw sputum samples.
218 219
If either IS1245 or IS6110 was detected at an SNR value ≥ 20, then the output from the internal
220
control probes is reported as “Not Applicable”. Deferring the interpretation of internal controls
221
in the event of an M. tuberculosis or MAC Detected result is based on the fact that the M13
222
internal positive control is included at a concentration very near its limit of detection. In those
223
cases where there is a very high concentration of TB DNA in the asymmetric PCR, there may be
224
preferential amplification of the TB genes such that asymmetric M13 amplification is limited,
225
and the M13 SNR value is < 3 (i.e., Not Detected).
10
Revised Manuscript 226 227
RESULTS
228 229
Analytical sensitivity
230 231
Analytical sensitivity for correct genotyping was estimated by preparing a dilution series of
232
quantified H37Ra DNA between 1 ng and 10 fg per test, and performing the test in triplicate.
233
Reproducible detection and correct genotyping was obtained between 500 fg and 1 pg DNA per
234
test, or between 110 and 225 genome copies. At 500 fg, discrimination ratios (D) for rpoB
235
mutation 531W were between -0.07, and -0.12, and one of three replicates for rpoB mutation
236
531C2 resulted in D = -0.03. In the absence of an indeterminate threshold for D, these values
237
resulted in a false positive for RIF resistance. With a threshold of D ≤ -0.2 for declaring
238
resistance (see Discussion), then the estimated limit of detection for the test is between 500 fg
239
and 100 fg, or 110 and 25 genome copies. The upper range for correct genotyping has not yet
240
been determined, although the anticipated quantity of genomic DNA per test from positive
241
control and JHH clinical isolates (below) was estimated to be >5 ng per test and we have
242
anecdotal evidence that 100 ng MTB DNA per test still leads to correct MTB genotyping and
243
behavior of the internal positive controls (not shown). Opportunities for improving the lower
244
LoD are described below.
245 246
Analytical specificity
247
11
Revised Manuscript 248
The M13 internal amplification and hybridization control was detected in all tests and all no
249
template controls were consistently negative. The analytical specificity of the microarray test
250
was evaluated with CDC and TDR isolates of known genotype, with results summarized in Table
251
2. The 32 CDC and TDR isolates harbored 68 mutations in rpoB, inhA, katG, embB, and rpsL
252
genes, six of which were not represented by a specific microarray probe. Of the 62 resistance-
253
conferring mutations for which there was a sequence-specific probe on the microarray, the
254
microarray correctly detected 95.2% of them (59/62), with false negative genotyping results
255
associated with rpoB mutations Q513L and D516E. At least one near-neighbor probe was
256
sensitive to three of the six known mutations that were not represented by a specific probe on the
257
microarray, which is similar to the behavior of “sloppy molecular beacon” probes (10). There
258
were also 96 wildtype sequences interrogated by the microarray test, 100% of which were
259
correctly identified as wildtype. Relative to phenotypic drug susceptibility tests, there was only
260
one false susceptible phenotypic result corresponding with rpoB mutation D516E. These data
261
demonstrate that the test can accurately detect resistance-conferring mutations when a mutation-
262
specific probe is present on the microarray, and provided the impetus for testing the system on
263
blinded clinical isolates.
264 265
Blinded clinical isolates
266 267
Of the 153 blinded clinical isolates from the JHH archive, phenotypic resistance to the first-line
268
drugs RIF, INH, and EMB was detected by agar proportion in 40, 14, and 10, strains,
269
respectively. Resistance was confirmed on these strains using the MGIT-960 system. Additional
270
testing was performed on selected strains to determine MIC’s for the individual antibiotics,
12
Revised Manuscript 271
resulting in 459 data points that served as the phenotypic DST comparator for microarray system
272
performance.
273 274
Overall concordance between agar proportion and the microarray test for the first-line drugs was
275
95.4% (146/153) for RIF, 96.1% (147/153) for INH, 96.7% (148/153) for EMB, and 96.1%
276
(441/459) for the three drugs combined. There was no obvious difference in concordance when
277
isolates were segregated by geographical origin or year of collection.
278 279
The percentage of false resistance or false susceptible results for all drugs is shown in Table 3,
280
and discrepant analysis relative to DNA sequencing shown in Table 4. Microarray sensitivity
281
(true resistance) for the first line drugs was 100% for RIF (14/14), 90.0% for INH (36/40) and
282
70% for EMB (7/10). Overall microarray sensitivity for RIF, INH, and EMB combined was
283
89.1% (57/64). Microarray specificity (true susceptibility) was 95.0% for RIF (132/139), 98.2%
284
for INH (111/113) and 98.6% for EMB (141/143). Overall microarray specificity for RIF, INH,
285
and EMB combined was 97.2% (384/395). The overall positive and negative predictive values
286
for RIF, INH and EMB combined were 84.9% and 98.3%, respectively. MIC’s for isolates that
287
had a false susceptible microarray result for INH or EMB were well above established
288
breakpoints for each drug, and ranged from 0.5 μg mL-1 to 4.0 μg mL-1 for INH and > 5.0 μg
289
mL-1 to 16.0 μg mL-1 for EMB (Table 4); there were no false susceptible microarray results for
290
RIF.
291 292
A total of 153 phenotypic data points were collected for STR, which is now considered a second-
293
line drug. Phenotypic STR resistance was detected in 23 of the 153 strains. Overall concordance
13
Revised Manuscript 294
between agar proportion and the microarray was 89.5% (137/153). Sensitivity was 34.8% (8/23)
295
and specificity was 99.2% (129/130). Nine of the 10 isolates correctly identified as STR-
296
resistant by the microarray test had STR MIC’s ≥ 32 μg mL-1, and one strain had an MIC of 4 µg
297
mL-1. There was only one false resistant STR result (0.65%) but 15 (9.8%) false susceptible
298
results. All other STR resistant isolates (n = 15) for which false susceptible errors occurred had
299
MICs ranging from 4 μg mL-1 to 8 μg mL-1.
300 301
There were 33 discrepant phenotypic results between the microarray and the standard phenotypic
302
tests, but there were only two discrepant results based on DNA sequence (Table 4). Two RIF
303
false resistant calls (isolates S15 and CM161) involved a fairly weak microarray discrimination
304
ratio (D) that may be best categorized as indeterminate; identifying an indeterminate range for D
305
was an explicit objective of this study. All false sensitive discrepants were a consequence of
306
DNA mutations that are not represented by a specific microarray probe. These data demonstrate
307
that, from a DNA sequence and hybridization perspective, the microarray is very specific and
308
provides the correct multi-drug genotype 98.7% of the time when the resistance-conferring
309
mutation is present in both the genome and the microarray (i.e., in 151/153 isolates).
310 311
There were 220 total tests (including negative controls) performed during the course of this
312
study, with zero invalid tests. Six amplified samples were re-hybridized to confirm the initial
313
test result due to observed physical damage on one or more gel elements on an array, and one
314
amplified sample was re-hybridized due to a fluorescent artifact. The overall microarray
315
hybridization repeat rate was therefore 3.2% (7/220). Gel element damage normally occurs as
316
the user is removing the Parafilm cover slip before washing microarrays in a bulk solution. It is
14
Revised Manuscript 317
fully anticipated that physical gel element damage will be reduced or eliminated once the method
318
is converted into an entirely “closed” consumable format.
319 320
DISCUSSION
321 322
Analytical sensitivity
323 324
In the absence of any threshold values for D, the lower limit of detection for the microarray test
325
described here is at least 110 gene copies per amplification reaction, which is very near the limits
326
of molecular sampling error in conventional exponential PCR (14). Provided that a user can
327
purify genomic DNA from primary specimen at >10% efficiency, then, there is no reason that
328
the method or test as described cannot be applied to smear positive sputum or decontaminated
329
sediments that are expected to contain at least 103 -104 CFU mL-1.
330 331
Based on this study, we identified several fairly simple ways in which the lower LoD of the test
332
can be improved or stabilized. We have anecdotal evidence, for example, that increasing the
333
hybridization temperature from 55°C to 57°C can resolve the false positive rpoB signals at 100-
334
500 fg input DNA. It is also possible to increase the number of amplification cycles, at the
335
expense of increasing the total analysis time. Neither of these approaches was specifically
336
addressed during the course of this study, because a 110 genome copy limit of detection was
337
deemed sufficiently sensitive to evaluate microarray specificity.
338 339
Analytical specificity
15
Revised Manuscript 340 341
The analytical specificity of the MDR-TB microarray test is defined by the specificity of probe
342
hybridization to cognate DNA sequences (Tables 2 and 4), rather than microarray response
343
relative to phenotypic drug susceptibility results (Table 3). In this context, analytical specificity
344
of the microarray test was excellent for both the control and blinded clinical isolates. The
345
weakest performing probes were those for rpoB mutations Q513L and D516E, for which
346
hybridization signal intensity was lower than neighboring or related probes regardless of probe
347
design (sequence length and SNP position within the probe) or gel element polymer composition
348
(not shown). The rpoB single-stranded amplicon is predicted to have significant stem-loop
349
secondary structure in the 513-516 region using MFold (40), and it is well known that secondary
350
and tertiary structures can negatively influence hybridization sensitivity and specificity (16, 32).
351
Fortunately, both Q513L and D516E are relatively rare mutations conferring resistance to RIF,
352
and the consequence of weak hybridization is manifest primarily in analytical sensitivity, not
353
specificity.
354 355
It is usually assumed that hybridization specificity needs to be perfect in order for microarrays to
356
have diagnostic value. On the other hand, “sloppy molecular beacons” that are responsive to
357
multiple mutations within a 20-mer reporting probe (10) form the basis of the GeneXpert
358
MTB/RIF assay, and enable reasonably sensitive detection (but not identification) of mutations
359
regardless of where that mutation occurs within the reporting probe. Most of the microarray
360
probes were very specific for their corresponding mutation, and insensitive to near neighbor
361
mutations (Table 2), but some probes were not 100% specific. The clinical importance of perfect
362
versus sloppy hybridization specificity ultimately depends on the information needed to provide
16
Revised Manuscript 363
patient-specific antibiotic therapy. For example, if the clinical decision to treat with RIF is based
364
on the presence of any mutation in the rpoB gene, then sloppy hybridization and reporting probes
365
are adequate for diagnostic purposes. On the other hand, if the clinical outcome of RIF treatment
366
ultimately depends on epistatic interactions or compensatory mutations within the rpoB gene (6,
367
8, 19, 23), then it may be important to identify precisely which mutations are present before
368
making a decision to treat or not treat with RIF.
369 370
This situation was clearly evident in the analysis of the RIF phenotypic discrepant isolates in
371
Table 4. Isolate C9, for example, was sensitive to RIF (MIC
Revised Manuscript 1
A simplified microarray system for simultaneously detecting rifampin,
2
isoniazid, ethambutol, and streptomycin resistance markers in Mycobacterium
3
tuberculosis.
4 5 6
Yvonne Linger1, Alexander Kukhtin1, Julia Golova1, Alexander Perov1, Amine Lambarqui1, Lexi
7
Bryant1, George B. Rudy1, Kim Dionne2, Stefanie L. Fisher2, Thomas M. Shinnick3, Nicole
8
Parrish2, and Darrell P. Chandler1#
9 10 11
1
Akonni Biosystems, Inc., Frederick MD 21701, 2The Centers for Disease Control and
12
Prevention, Atlanta, GA 30333, and 3The Johns Hopkins School of Medicine, Baltimore, MD
13
21287
14 15
#
To whom correspondence may be addressed: [email protected]
16 17
Running title: MDR-TB microarray
18 19
Submitted to: Journal of Clinical Microbiology
1
Revised Manuscript 20
ABSTRACT
21 22
We developed a simplified microarray test for detecting and identifying mutations in rpoB, katG,
23
inhA, embB and rpsL, and compared the analytical performance of the test relative to phenotypic
24
DST. The analytical sensitivity was estimated to be at least 110 genome copies per amplification
25
reaction. The microarray test correctly detected 95.2% of those mutations for which there was a
26
sequence-specific probe on the microarray, and 100% of 96 wildtype sequences. In a blinded
27
analysis of 153 clinical isolates, microarray sensitivity for first line drugs relative to phenotypic
28
DST (true resistance) was 100% for RIF (14/14), 90.0% for INH (36/40), 70% for EMB (7/10),
29
and 89.1% (57/64) combined. Microarray specificity (true susceptibility) for first line agents
30
was 95.0% for RIF (132/139), 98.2% for INH (111/113) and 98.6% for EMB (141/143). Overall
31
microarray specificity for RIF, INH, and EMB combined was 97.2% (384/395). The overall
32
positive and negative predictive values for RIF, INH and EMB combined were 84.9% and
33
98.3%, respectively. For the second-line drug STR, overall concordance between agar
34
proportion and the microarray was 89.5% (137/153). Sensitivity was 34.8% (8/23) because of
35
limited microarray coverage for STR-conferring mutations, and specificity was 99.2% (129/130).
36
All false susceptible discrepants were a consequence of DNA mutations that are not represented
37
by a specific microarray probe. There were zero invalid results from 220 total tests. The
38
simplified microarray system is suitable for detecting resistance-conferring mutations in clinical
39
MTB isolates, and can now be used for prospective trials or integrated into an all-in-one, closed-
40
amplicon consumable.
41
2
Revised Manuscript 42
INTRODUCTION
43 44
Mycobacterium tuberculosis (TB) infects one-third of the world’s population, with
45
approximately nine million new cases and two million deaths attributable to the disease each
46
year (33). Early case detection and rapid treatment are considered the most effective control
47
strategies to reduce TB transmission (17), especially in those cases involving multi- or
48
extensively drug resistant TB (MDR- and XDR-TB). Culture-based methods remain the gold
49
standard for diagnosing drug resistant TB, but can take several weeks or months to complete.
50
Thus, molecular-based drug susceptibility tests are becoming increasingly attractive as MDR-TB
51
diagnostic tools in order to initiate individualized, patient-appropriate treatment in a timely
52
manner.
53 54
Technologies such as Cepheid’s GeneXpert and Hain MDR-TB line probe assays reduce the
55
time to diagnosis for many TB patients, provide a rapid read-out indicating resistance to rifampin
56
or selected mutations conferring resistance to other first- or second-line drugs, and illustrate the
57
potential to deploy molecular tests closer to the point of need (34). However, the number of
58
known genes and resistance-conferring mutations to first- and second-line drugs greatly exceeds
59
the multiplexing capacity of these platforms, which may limit their clinical efficacy in the
60
treatment and control of multi- or extensively drug resistant TB. Planar and suspension
61
microarrays are well suited to address the multiple gene, multiple mutation challenge of
62
diagnosing drug resistant TB (4, 12, 13, 21, 24, 25, 30, 31, 38), but clinical adoption of
63
microarray technology is hampered by poor reproducibility (9, 20, 39), complex workflows,
64
and/or extensive user subjectivity and involvement in image and data analysis (15). In order to
3
Revised Manuscript 65
translate microarrays into efficacious TB diagnostics at the point of need, it is therefore
66
necessary to simplify user interaction with the technology while retaining the ability to detect
67
multiple genes and multiple mutations in a timely manner. The objectives of this study were to
68
develop a gel element microarray test for MDR-TB at a level of coverage surpassing what is
69
currently available with WHO-endorsed molecular platforms, estimate the analytical specificity
70
of the test on MDR-TB isolates of known genotype and phenotype, and compare the
71
performance of the test to conventional drug susceptibility testing as a precursor to integrating
72
the method into an entirely closed amplicon consumable (5, 7) and sample-to-answer system.
73 74
MATERIALS AND METHODS
75 76
Isolates and positive controls
77 78
The set of M. tuberculosis isolates and nucleic acids utilized for this study represents a global
79
and historical distribution of drug susceptible, drug resistant, and multi-drug resistant strains.
80
Materials from the Centers for Disease Control and Prevention (CDC, Atlanta, Georgia) and the
81
TDR Tuberculosis Strain Bank (29) were provided as heat-killed crude lysates. Available
82
isolates only covered 22 of 39 resistance-conferring mutations that were present on the gel
83
element array. Purified DNA from M. tuberculosis H37Ra was acquired from the American
84
Type Culture Collection (ATCC, Manassas, VA) and quantified on a Nanodrop 3000
85
fluorometer before use. CDC and TDR materials were used to verify the analytical specificity of
86
microarray probes before verifying array performance on blinded samples from the Johns
87
Hopkins Hospital (JHH) clinical isolate archive. M13mp18 single-stranded DNA was purchased
4
Revised Manuscript 88
from New England Biolabs (Ipswitch, MA) and diluted to 250 pg mL-1 in 10 mM Tris, 1 mM
89
Na2EDTA pH 7.5. Crude lysates and purified nucleic acids were stored at -20°C until use.
90 91
The Johns Hopkins Hospital (JHH) M. tuberculosis isolate archive consists of susceptible, DR-,
92
MDR-, and XDR- M. tuberculosis strains from the mid-Atlantic, south, and southwest regions of
93
the United States, Mexico, South America, Cameroon, and Kazakhstan. Isolates used for this
94
study were collected between 1995 and 2011. JHH isolates were cultivated under BSL-3
95
conditions on Lowenstein-Jensen (L-J) agar slants and maintained at 37°C in 5% CO2. Crude
96
lysates were prepared from each strain by harvesting a small amount of cells from L-J slants,
97
resuspending the cells in Middlebrook 7H9 broth, and adjusting the turbidity to a 0.5 McFarland
98
standard (ca. 1.5 x 108 CFU mL-1). Subsequently, 1 ml of each standardized suspension was
99
centrifuged for 5 minutes at 13,000 x g and the supernatant removed. Cell pellets were
100
resuspended in 200 µL of molecular grade water, heat killed at 95°C for 30 minutes, sonicated
101
for 15 minutes, and centrifuged for 5 minutes at 13,000 x g. A 100 μL aliquot of each
102
supernatant was then removed and stored at -20°C until use, representing approximately 7.5 x
103
107 CFU equivalents of bacteria and 3.3 ng DNA μL-1 assuming a genome size of 4 x106 bp and
104
one bacterium per CFU. PCR setup with crude lysates was performed under BSL-2 conditions,
105
and all other manipulations were performed under BSL-1 controls with universal PCR
106
precautions.
107 108
Phenotypic drug susceptibility testing
109 110
Each JHH isolate was tested for phenotypic resistance to rifampin (RIF), isoniazid (INH),
111
ethambutol (EMB), and streptomycin (STR) using the MGIT-960 system as per the 5
Revised Manuscript 112
manufacturer’s recommendations (Becton Dickinson, Sparks, MD), and 1% agar proportion as
113
per the Clinical Laboratory Standards Institute (CLSI) M24-A2 standards (37). The MGIT-960
114
and agar proportion methods both have established breakpoints for drug susceptibility and
115
resistance determinations. For a select subset of strains (n = 78), the TREK Sensititre™
116
MYCOTB assay (TREK Diagnostic Systems, Cleveland, OH) was used to establish a minimum
117
inhibitory concentration (MIC) for each drug.
118 119
Primers, probes, and synthetic DNA standards
120 121
Amplification primers and microarray probes were designed against M. tuberculosis genes and
122
target codons shown in Table 1. Genes and codons represent high-confidence mutations known
123
to confer a drug resistance phenotype against RIF, INH, EMB, or STR. Nine PCR primer pairs
124
were designed to work together in a multiplex, asymmetric master mix, where one of the primers
125
in each pair was synthesized with a Cy3 label and the resulting single-stranded, labeled amplicon
126
captured by one or more immobilized microarray probes. Only one amplification reaction is
127
required per test. PCR primers were synthesized by standard phosphoramidite chemistry at
128
Akonni or Eurofins MWG Operon (Huntsville, AL), HPLC purified, and quantified by UV
129
absorption before use. Resulting Cy3-labeled, asymmetric amplicons ranged from 92 to 139 nt
130
in length.
131 132
Microarray probes were synthesized by Akonni with a custom 3’ linker and purified to >90%
133
purity by HPLC before use. Probe purity was measured by electrospray ionization mass
134
spectrometry. Each microarray contained a universal hybridization probe for each functional
6
Revised Manuscript 135
gene and primer pair to verify that M. tuberculosis gene targets were amplified from each
136
sample. At least one matched pair of microarray probes (wildtype (WT) and single-nucleotide
137
mutant (MU)) was included for each mutation of interest. Control probes included a Cy3 beacon
138
for manufacturing quality control and positional reference, a synthetic hybridization control
139
probe with no known homology to any naturally-occurring sequence, and a probe for the M13
140
internal amplification and inhibition control. A 92-base, Cy3-labeled hybridization control target
141
was synthesized by TriLink Biotechnologies (San Diego, CA), HPLC purified to >90% purity,
142
and diluted to 100 pg mL-1 in 10 mM Tris, 1 mM Na2EDTA pH 7.5.
143 144
Microarray manufacture
145 146
Gel element arrays were manufactured on custom-coated glass substrates by a single-step
147
copolymerization procedure using a 4% 2-hydroxyethyl methylacrylamide copolymer,
148
essentially as described in (11). Cy3 beacons were immobilized at 1 μM and all other probes
149
immobilized at 50 μM concentration. Up to three microarrays were printed on each glass
150
substrate, but only one microarray is required per test. All microarrays were visually inspected
151
for gel element presence and uniform morphology before use. Microarrays were stored at room
152
temperature in a dust-free container until use (< 1 months). Immediately before use, a 50 μL
153
gasket was applied around each microarray.
154 155
Multiplex, asymmetric amplification
156
7
Revised Manuscript 157
Amplification reactions were assembled in 0.2 mL thin-walled PCR tubes and cycled on Quanta
158
Biotech QB-96 thermal cyclers (Surrey, United Kingdom). The nine pairs of forward and
159
reverse primers were combined in 1:10 ratios and prepared as a concentrated master mix to
160
achieve a final primer concentration of 0.04 to 0.4 μM each per reaction. Each 50 μL reaction
161
contained 3 μL of template (crude lysate or purified DNA), 25 μL Qiagen Multiplex PCR Kit
162
all-in-one amplification buffer and enzyme (Valencia, CA), 18 μL asymmetric PCR primer mix,
163
5% dimethyl sulfoxide (Sigma-Aldrich, St. Louis, MO) and 0.6 μg μL-1 non-acetylated bovine
164
serum albumin (Sigma-Aldrich). One microliter of M13 internal positive control (250 fg total)
165
was included in all reactions, unless otherwise noted. The balance of each reaction was made up
166
with sterile, ultra-pure water as required. No template controls containing the M13 internal
167
positive control were processed in parallel for each set of experiments. The thermal cycling
168
program consisted of an initial denaturation at 95°C for 15 min, followed by 40 cycles of [94°C
169
for 30 sec, 60°C for 60 sec, 72°C for 30 sec], and a final extension at 72°C for 3 min. The actual
170
run time for thermal cycling was ~ 2 hrs. Amplified material was either used immediately for
171
microarray hybridization, or stored at -20°C until use.
172 173
Hybridization and detection
174 175
After amplification, 20 μL of unpurified amplicon was combined with 30 μL of a concentrated
176
hybridization master mix to achieve a final reaction mixture consisting of 1M guanidine
177
thiocyanate, 50 mM HEPES pH 7.5, 5 mM Na2EDTA pH 8.0, 3.3 μg μL-1 non-acetylated BSA,
178
and 4 fmol µL-1 Cy3-labeled internal hybridization control. Forty-nine microliters (49 μL) of the
179
resulting mixture was applied directly to the center of each microarray, microarrays covered with
8
Revised Manuscript 180
a Parafilm coverslip, and hybridized for 3 hr at 55ºC on an MJ Research PTC-225 slide tower.
181
After hybridization, cover slips and gaskets were removed and up to 24 microarrays washed
182
simultaneously in 350 mL 1X SSPE (150 mM NaCl, 10 mM NaH2PO4 • H2O, 10 mM
183
Na2EDTA), 0.01% Triton X-100, pH 7.4 for 10 min with gentle agitation. Arrays were then
184
dipped twice in ultra-pure water and dried with compressed air.
185 186
Imaging and data analysis
187 188
Dried microarrays were imaged on an Akonni field-portable microarray analyzer, a very simple
189
imager consisting of a high-intensity green light emitting diode (LED), custom optics, and non-
190
cooled CCD camera. Image acquisition ranged from 0.05 to 0.1 sec per array, and pixel data
191
were acquired from raw .tif images using TIGR Spotfinder software and a fixed-circle algorithm
192
(mask size = 25 pixel diameter) with all background pixels included in the analysis (Top
193
Background Cutoff value in Spotfinder set to zero). Local background was subtracted from the
194
signal of each gel element. Integrated (background-corrected) signal intensity data were
195
exported as tab-delimited text, and imported into custom Excel spreadsheets for data analysis.
196
Signal to noise ratios (SNR values) were calculated using the background corrected signal from
197
gel elements and normalizing against 3 times the standard deviation of the average local
198
background.
199 200
A test was only declared valid if the average Cy3 beacon, internal hybridization control, and
201
M13 amplification/inhibition control probes all generated SNR values ≥ 3. Otherwise, the test
202
was deemed invalid and test results were not reported. The IS6110 and IS1245 probes are
9
Revised Manuscript 203
internal positive controls, a proxy for detecting members of the M. tuberculosis and M. avium
204
complexes, respectively, and are used to condition the report for RIF, INH, EMB and STR
205
resistance-conferring mutations. Positive detection of both the IS6110 and IS1245 probes is
206
considered a mixed sample. For all valid tests, if the IS1245 probe for M. avium complex
207
(MAC) is detected, then no genotyping results are reported for the M. tuberculosis resistance-
208
conferring mutations. Otherwise, discrimination ratios are calculated for valid tests as D =
209
(SNRWT - SNRMU) / (SNRWT + SNRMU). As currently implemented, a zero-copy IS6110 isolate
210
will still generate discrimination ratios for M. tuberculosis resistance-conferring mutations,
211
because the PCR primers for rpoB, inhA, katG, embB, and rpsL are specific for M. tuberculosis.
212
For this study, any value of D ≥ 0 was interpreted as a wild type sequence at the targeted
213
nucleotide position, and any value of D < 0 was interpreted as a mutation at the targeted
214
nucleotide position. Results from this study were intended, in part, to understand and refine
215
values for D that should correspond to an inconclusive test result before automating the image
216
and data analysis algorithms for a clinical user, and prospectively testing the microarray with
217
decontaminated sediments or raw sputum samples.
218 219
If either IS1245 or IS6110 was detected at an SNR value ≥ 20, then the output from the internal
220
control probes is reported as “Not Applicable”. Deferring the interpretation of internal controls
221
in the event of an M. tuberculosis or MAC Detected result is based on the fact that the M13
222
internal positive control is included at a concentration very near its limit of detection. In those
223
cases where there is a very high concentration of TB DNA in the asymmetric PCR, there may be
224
preferential amplification of the TB genes such that asymmetric M13 amplification is limited,
225
and the M13 SNR value is < 3 (i.e., Not Detected).
10
Revised Manuscript 226 227
RESULTS
228 229
Analytical sensitivity
230 231
Analytical sensitivity for correct genotyping was estimated by preparing a dilution series of
232
quantified H37Ra DNA between 1 ng and 10 fg per test, and performing the test in triplicate.
233
Reproducible detection and correct genotyping was obtained between 500 fg and 1 pg DNA per
234
test, or between 110 and 225 genome copies. At 500 fg, discrimination ratios (D) for rpoB
235
mutation 531W were between -0.07, and -0.12, and one of three replicates for rpoB mutation
236
531C2 resulted in D = -0.03. In the absence of an indeterminate threshold for D, these values
237
resulted in a false positive for RIF resistance. With a threshold of D ≤ -0.2 for declaring
238
resistance (see Discussion), then the estimated limit of detection for the test is between 500 fg
239
and 100 fg, or 110 and 25 genome copies. The upper range for correct genotyping has not yet
240
been determined, although the anticipated quantity of genomic DNA per test from positive
241
control and JHH clinical isolates (below) was estimated to be >5 ng per test and we have
242
anecdotal evidence that 100 ng MTB DNA per test still leads to correct MTB genotyping and
243
behavior of the internal positive controls (not shown). Opportunities for improving the lower
244
LoD are described below.
245 246
Analytical specificity
247
11
Revised Manuscript 248
The M13 internal amplification and hybridization control was detected in all tests and all no
249
template controls were consistently negative. The analytical specificity of the microarray test
250
was evaluated with CDC and TDR isolates of known genotype, with results summarized in Table
251
2. The 32 CDC and TDR isolates harbored 68 mutations in rpoB, inhA, katG, embB, and rpsL
252
genes, six of which were not represented by a specific microarray probe. Of the 62 resistance-
253
conferring mutations for which there was a sequence-specific probe on the microarray, the
254
microarray correctly detected 95.2% of them (59/62), with false negative genotyping results
255
associated with rpoB mutations Q513L and D516E. At least one near-neighbor probe was
256
sensitive to three of the six known mutations that were not represented by a specific probe on the
257
microarray, which is similar to the behavior of “sloppy molecular beacon” probes (10). There
258
were also 96 wildtype sequences interrogated by the microarray test, 100% of which were
259
correctly identified as wildtype. Relative to phenotypic drug susceptibility tests, there was only
260
one false susceptible phenotypic result corresponding with rpoB mutation D516E. These data
261
demonstrate that the test can accurately detect resistance-conferring mutations when a mutation-
262
specific probe is present on the microarray, and provided the impetus for testing the system on
263
blinded clinical isolates.
264 265
Blinded clinical isolates
266 267
Of the 153 blinded clinical isolates from the JHH archive, phenotypic resistance to the first-line
268
drugs RIF, INH, and EMB was detected by agar proportion in 40, 14, and 10, strains,
269
respectively. Resistance was confirmed on these strains using the MGIT-960 system. Additional
270
testing was performed on selected strains to determine MIC’s for the individual antibiotics,
12
Revised Manuscript 271
resulting in 459 data points that served as the phenotypic DST comparator for microarray system
272
performance.
273 274
Overall concordance between agar proportion and the microarray test for the first-line drugs was
275
95.4% (146/153) for RIF, 96.1% (147/153) for INH, 96.7% (148/153) for EMB, and 96.1%
276
(441/459) for the three drugs combined. There was no obvious difference in concordance when
277
isolates were segregated by geographical origin or year of collection.
278 279
The percentage of false resistance or false susceptible results for all drugs is shown in Table 3,
280
and discrepant analysis relative to DNA sequencing shown in Table 4. Microarray sensitivity
281
(true resistance) for the first line drugs was 100% for RIF (14/14), 90.0% for INH (36/40) and
282
70% for EMB (7/10). Overall microarray sensitivity for RIF, INH, and EMB combined was
283
89.1% (57/64). Microarray specificity (true susceptibility) was 95.0% for RIF (132/139), 98.2%
284
for INH (111/113) and 98.6% for EMB (141/143). Overall microarray specificity for RIF, INH,
285
and EMB combined was 97.2% (384/395). The overall positive and negative predictive values
286
for RIF, INH and EMB combined were 84.9% and 98.3%, respectively. MIC’s for isolates that
287
had a false susceptible microarray result for INH or EMB were well above established
288
breakpoints for each drug, and ranged from 0.5 μg mL-1 to 4.0 μg mL-1 for INH and > 5.0 μg
289
mL-1 to 16.0 μg mL-1 for EMB (Table 4); there were no false susceptible microarray results for
290
RIF.
291 292
A total of 153 phenotypic data points were collected for STR, which is now considered a second-
293
line drug. Phenotypic STR resistance was detected in 23 of the 153 strains. Overall concordance
13
Revised Manuscript 294
between agar proportion and the microarray was 89.5% (137/153). Sensitivity was 34.8% (8/23)
295
and specificity was 99.2% (129/130). Nine of the 10 isolates correctly identified as STR-
296
resistant by the microarray test had STR MIC’s ≥ 32 μg mL-1, and one strain had an MIC of 4 µg
297
mL-1. There was only one false resistant STR result (0.65%) but 15 (9.8%) false susceptible
298
results. All other STR resistant isolates (n = 15) for which false susceptible errors occurred had
299
MICs ranging from 4 μg mL-1 to 8 μg mL-1.
300 301
There were 33 discrepant phenotypic results between the microarray and the standard phenotypic
302
tests, but there were only two discrepant results based on DNA sequence (Table 4). Two RIF
303
false resistant calls (isolates S15 and CM161) involved a fairly weak microarray discrimination
304
ratio (D) that may be best categorized as indeterminate; identifying an indeterminate range for D
305
was an explicit objective of this study. All false sensitive discrepants were a consequence of
306
DNA mutations that are not represented by a specific microarray probe. These data demonstrate
307
that, from a DNA sequence and hybridization perspective, the microarray is very specific and
308
provides the correct multi-drug genotype 98.7% of the time when the resistance-conferring
309
mutation is present in both the genome and the microarray (i.e., in 151/153 isolates).
310 311
There were 220 total tests (including negative controls) performed during the course of this
312
study, with zero invalid tests. Six amplified samples were re-hybridized to confirm the initial
313
test result due to observed physical damage on one or more gel elements on an array, and one
314
amplified sample was re-hybridized due to a fluorescent artifact. The overall microarray
315
hybridization repeat rate was therefore 3.2% (7/220). Gel element damage normally occurs as
316
the user is removing the Parafilm cover slip before washing microarrays in a bulk solution. It is
14
Revised Manuscript 317
fully anticipated that physical gel element damage will be reduced or eliminated once the method
318
is converted into an entirely “closed” consumable format.
319 320
DISCUSSION
321 322
Analytical sensitivity
323 324
In the absence of any threshold values for D, the lower limit of detection for the microarray test
325
described here is at least 110 gene copies per amplification reaction, which is very near the limits
326
of molecular sampling error in conventional exponential PCR (14). Provided that a user can
327
purify genomic DNA from primary specimen at >10% efficiency, then, there is no reason that
328
the method or test as described cannot be applied to smear positive sputum or decontaminated
329
sediments that are expected to contain at least 103 -104 CFU mL-1.
330 331
Based on this study, we identified several fairly simple ways in which the lower LoD of the test
332
can be improved or stabilized. We have anecdotal evidence, for example, that increasing the
333
hybridization temperature from 55°C to 57°C can resolve the false positive rpoB signals at 100-
334
500 fg input DNA. It is also possible to increase the number of amplification cycles, at the
335
expense of increasing the total analysis time. Neither of these approaches was specifically
336
addressed during the course of this study, because a 110 genome copy limit of detection was
337
deemed sufficiently sensitive to evaluate microarray specificity.
338 339
Analytical specificity
15
Revised Manuscript 340 341
The analytical specificity of the MDR-TB microarray test is defined by the specificity of probe
342
hybridization to cognate DNA sequences (Tables 2 and 4), rather than microarray response
343
relative to phenotypic drug susceptibility results (Table 3). In this context, analytical specificity
344
of the microarray test was excellent for both the control and blinded clinical isolates. The
345
weakest performing probes were those for rpoB mutations Q513L and D516E, for which
346
hybridization signal intensity was lower than neighboring or related probes regardless of probe
347
design (sequence length and SNP position within the probe) or gel element polymer composition
348
(not shown). The rpoB single-stranded amplicon is predicted to have significant stem-loop
349
secondary structure in the 513-516 region using MFold (40), and it is well known that secondary
350
and tertiary structures can negatively influence hybridization sensitivity and specificity (16, 32).
351
Fortunately, both Q513L and D516E are relatively rare mutations conferring resistance to RIF,
352
and the consequence of weak hybridization is manifest primarily in analytical sensitivity, not
353
specificity.
354 355
It is usually assumed that hybridization specificity needs to be perfect in order for microarrays to
356
have diagnostic value. On the other hand, “sloppy molecular beacons” that are responsive to
357
multiple mutations within a 20-mer reporting probe (10) form the basis of the GeneXpert
358
MTB/RIF assay, and enable reasonably sensitive detection (but not identification) of mutations
359
regardless of where that mutation occurs within the reporting probe. Most of the microarray
360
probes were very specific for their corresponding mutation, and insensitive to near neighbor
361
mutations (Table 2), but some probes were not 100% specific. The clinical importance of perfect
362
versus sloppy hybridization specificity ultimately depends on the information needed to provide
16
Revised Manuscript 363
patient-specific antibiotic therapy. For example, if the clinical decision to treat with RIF is based
364
on the presence of any mutation in the rpoB gene, then sloppy hybridization and reporting probes
365
are adequate for diagnostic purposes. On the other hand, if the clinical outcome of RIF treatment
366
ultimately depends on epistatic interactions or compensatory mutations within the rpoB gene (6,
367
8, 19, 23), then it may be important to identify precisely which mutations are present before
368
making a decision to treat or not treat with RIF.
369 370
This situation was clearly evident in the analysis of the RIF phenotypic discrepant isolates in
371
Table 4. Isolate C9, for example, was sensitive to RIF (MIC
