Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Alasdair Edge1; Grant Stewart2; Kayleigh Arthur1; Lauren Brown1; Chris Hodkinson1;Aditya Malkar1; Paul Nasca1; Marc van der Schee1
In non-targeted ‘omics applications liquid or gas chromatography (LC or GC) is typically combined with mass spectrometry (MS) for analysis of complex biological matrices. Molecular features can be missed by LC-MS due to: • Trace component suppression by chemical noise.• Chromatographically unresolved isomeric species.
The optimised FAIMS scan settings of CF -0.9 to 3.1 Td at DF 240 Td at 2 CF scans s-1 were used to analyse: • n=10 urine samples from renal cancer patients• n=10 urine samples from healthy individualsRaw spectra were used to generate features lists for each CF using XCMS using parameters:• method="centWave",scanrange=c(0,504), peakwidth=c(5,30),
ppm=25, mzdi�=c(-0.001), snthresh=3, integrate=2, fitgauss=FALSE
Produced feature-extracted and feature-aligned data.An unsupervised, pareto scaled principal component analysis (PCA) plot for data obtained at CF 1.1 Td is shown in Figure 9.• Pareto scaling used for the analysis as consistent with published
untargeted metabolomic studies2
The S-plot, a plot of covariance and correlation based on the features making up the OPLS-DA model was also plotted. • Trend plots for features contributing towards class separation
with the most confidence were plotted (Figure 11).
Next step is to expand the data set to a statistically significant # of patients and validate the identified features by correlating multivariate analysis to raw data, as well as cross referencing to patient data to identify potential confounders.
The PCA plot shows partial separation indicating di�erences in the data sets. A supervised o-partial least squares-discriminant analysis (OPLS-DA), highlighting the within class and between class di�erences, also showed separation (Figure 10).
1. Introduction 5. LC-ultraFAIMS-MS Workflow Validation
• The optimisation of untargeted profiling of the urinary metabolome using LC-FAIMS-MS and the data analysis workflow required to produce 3D nested data sets has been described.
• The optimised workflow has been applied to demonstrate proof of principle metabolic assessment of urine samples from a renal cell cancer cohort vs. healthy controls.
• Following further validation, the LC-ultraFAIMS-MS features list will be used to build a predictive model, or classifier, from which a probability of disease can be predicted.
• Combining FAIMS into a metabolomic workflow o�ers increased peak capacity for untargeted metabolite profiling to diagnose disease or stratify an individual’s treatment.
5. Conclusions
Optimisation of nested FAIMS data acquisition focussed on:• # and range of FAIMS CF settings to give optimal FAIMS
separation.• # of data points within the timescale of the chromatographic
peaks.• Optimal sensitivity via chromatographic peak heights and # of
TOF scans s-1.
The e�ect of the di�erent MS scan rates and CF ranges on feature detection was investigated.• Acquiring data at 2 CF scans s-1 increased the # of features
detected, in all cases, despite the decrease in peak intensity associated with higher acquisition speeds (Figure 7b, c).
• Increasing to 3 CF scans s-1 did not further increase the # of features detected.
The # of features detected by scanning the FAIMS at a rate of 1 CF scan s-1 and 2 CF scans s-1 was compared.• At a S:N of 3, more features were detected using a 2 CF scans s-1
scan rate (Figure 8a).
Optimal DF was determined based on selectivity and sensitivity.• The CF vs features plot (Figure 5) shows good coverage across
the analytical space in the range -1 to +3 Td.• More features were detected in the higher CF region at the higher
DFs. 240 Td was selected for further experiments based on widest distribution of detected features across the CF range.
As the CF scan is synchronised with the MS acquisition, the # of data points across a chromatographic peak is dependent on the # of MS scans s-1 and the CF range.• Increased MS scan rate means more data points can be acquired
over a given CF range.• Alternatively, a smaller CF step size could be applied, increasing
the # of data points across the CF peak whilst maintaining the # of CF data points across LC peak.
• Faster ToF scan rates reduced peak intensity (Figure 6).
4. LC-ultraFAIMS-MS optimisation
Figure 1. LC-MS and LC-ultraFAIMS-MS feature
determination
Figure 5. Detected features at DFs of 230, 240 and 250 Td
Figure 9. Unsupervised PCA (Pareto Scaling) plot at 1.1 Td CF
Figure 10. OPLS-DA plot at 1.1 Td CF
Figure 11. Example Trend plots of features contributing towards class separation
Figure 6. (a) Raw EIC for m/z 273 across all CFs (black) 10 spectra s-1 and (red) 20 spectra s-1, showing twice as many CF scans across the LC peak and (b)
deconvoluted LC-MS peak at CF of 1.6 Td at 20 spectra s-1
Figure 7. Comparison of # of features detected in experiments 3 a-f as described in Table 1
To determine if this result was accurate, or a result of increased noise using the faster scan rate, the analysis was repeated at a S:N of 10 (Figure 8b).• Whilst total # of features decreased, the faster scan rate again
produced more features than the slow scan rate. • 2 CF scans s-1 scan rate was used for analysis.
Figure 8. Comparison of features detected at 1 CF scan s-1 and 2 CF scans s-1 at (a)S:N of 3 and (b) S:N of 10
Figure 2. (a) ultraFAIMS chip and system (b) installed on Agilent 6230 ToF-MS
Figure 3. Integrated LC-ultraFAIMS-MS analysis
Figure 4. LC-ultraFAIMS-MS feature extraction workflow
www.owlstonemedical.com/ultraFAIMS
The Development of an LC-FAIMS-MS Metabolomics Workflow and its Applicationto the Discrimination of Benign from Malignant Renal Cell Masses
References1. Arthur, KA; Turner, MA; Reynolds, JC; Creaser, CS.; Anal. Chem.,
2017, 89, 34522. van den Berg, RA; Hoefsloot, HCJ; Westerhuis, JA; Smilde, AK; van
der Werf, MJ; BMC Genomics, 2006, 7, 142
1Owlstone Medical Ltd., 162 Cambridge Science Park, Cambridge, CB4 OGH, UK, 2Academic Urology Group, University of Cambridge, Box 43,Addenbrookes Hospital, Cambridge Biomedical Campus, Hills Road, Cambridge, CB2 0QQ
For further information, email: [email protected]
Analysis was performed on a 6230 TOF-MS and a 1290 series LC (Agilent, Santa Clara, US) combined with an ultraFAIMS device (Owlstone Medical Ltd, Cambridge, UK).
• The key dimensions of the ultraFAIMS device are the 100 µm electrode gap and 700 µm path length
• The small scale is key to the ability to integrate into the LC-MS workflow; an entire CF scan s-1 can be achieved, making the scanning approach compat ib le with chromatographic timescales.
• The CF values are synchronised with MS acquisition so a mass spectrum is acquired for each CF (Figure 3).
• LC was performed using a
2. Methods
The workflow for feature extraction from LC-FAIMS-MS files is detailed in Figure 4.A feature was defined as a unique identifier for each component of a m/z, a retention time (tR) and a compensation field (CF).
3. Data analysis
Table 1: LC-ultraFAIMS-MS optimisation experiments
Orthogonality between field asymmetric ion mobility spectrometry (FAIMS), LC and MS provides additional unique compound identifiers with detection of features based on (Figure 1):• Retention time• FAIMS dispersion and
compensation fields (DF and CF)
• Mass-to-charge (m/z)
RT1IdentifiedFeature
IdentifiedFeature1
IdentifiedFeature2
m/z1+ =
RT1
LC-ultraFAIMS-MS
LC-MS
DF+CF1
m/z1
DF+CF2
m/z2
+ + =
RT1 + + =
ChromatographicPeak
Mass Spectra1 per CF
CompensationField Scan
(nCFs per s)
LC peak width
IncorporateFAIMS into
LC-MS
Mass Spectrometry
Field Asym
metric W
aveform
Liquid Chromatography
Ion Mobility Spectrom
etry
MS scanrate (s-1)
per s m/zstart
m/zend
Index StartCF (Td)
EndCF (Td)
N CFsteps
N CFactual
CF stepsize (Td)
NRepeats
StartDF (Td)
EndDF (Td)
N DFSteps
12 1 80 1500 0 -0.9 4.1 10 12 0.5 503 250 250 0
12 1 80 1500 0 -0.9 4.1 10 12 0.5 503 240 240 0
12 1 80 1500 0 -0.9 4.1 10 12 0.5 503 230 230 0
24 2 80 1500 0 -0.9 4.1 10 12 0.5 1007 Test 1 0Test 1
10 1 80 1500 0 -0.9 3.1 8 10 0.5 503 Test 1 0Test 1
20 2 80 1500 0 -0.9 3.1 8 10 0.5 1007 Test 1 0Test 1
18 1 80 1500 0 -0.9 3.1 16 18 0.25 503 Test 1 0Test 1
6 1 80 1500 0 -0.9 3.1 4 6 1 503 Test 1 0Test 1
12 2 80 1500 0 -0.9 3.1 4 6 1 1007 Test 1 0Test 1
18 3 80 1500 0 0-0.9 3.1 4 6 1 1511 Test 1 Test 1
Test
1a
1b
1c
2
3a
3b
3c
3d
3e
3f
No
. of
feat
ures
3d3e3f
No
. of
feat
ures
3a3b3c
0
500
1000
1500
2000
2500
3000
3500
4000
4500
0
500
1000
1500
2000
2500
3000
3500
4000
4500
0
500
1000
1500
2000
2500
3000
3500
4000
4500
-0.9 -0.4 0.1 0.6 1.1 1.6 2.1 2.6 3.1
No
. of
feat
ures
CF (Td)-0.9 -0.4 0.1 0.6 1.1 1.6 2.1 2.6 3.1
CF (Td)-0.9 -0.4 0.1 0.6 1.1 1.6 2.1 2.6 3.1
CF (Td)
3a3b3c3d3e3f
(a) (b) (c)
A recent study1 reported a threefold increase in features detected in non-targeted profiling of human urine with the addition of FAIMS to LC-MS analysis.Here we describe an optimised workflow to produce three-dimensional metabolomics data sets and its application to the metabolic assessment of indeterminate masses using urine samples from a renal cell cancer (RCC) cohort (DIAMOND study; NRES East Of England 03/018).
(a) (b)
reversed phase Poroshell 120 EC-C18 column, 2.1 x 100 mm, particle size 2.7 µm (Agilent Technologies).
LC-FAIMS-MS(Agilent.d Files)
Convert All ExtractedTIC to .mzML
Intermediate .mz5Files
Segregate Samples asPer CF TIC in Foldets
CF 1 .mzML
Feature List (1)
Extract TICsCorresponding toVarious CF Values
Done via in-housePython Script and
Proteowizard
FeatureExtraction using
XCMS in R
CF 2 .mzML
Feature List (2)
FeatureExtraction using
XCMS in R
500
0
1000
1500
2000
2500
3000
3500
4000
4500
-0.9 -0.4 0.1 0.6 1.1 1.6 2.1 2.6 3.1 3.6 4.1
# o
f fe
atur
es
CF
250
240
230
0
100000
200000
300000
400000
500000
600000
700000
250 260 270 280 290
Retention time (s)
CF 1.6 Td
3a
3b
(b)
Inte
nsit
y
(a)
Inte
nsit
y
259
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
2.4
x105
260 261 262 263 264 265 266 267 268 269 270 271
Counts vs. Acquisition Time (sec)
+ESI EIC(273.1390) MS(all) 1705_FAIMS-Optimisation_MS4_FAIMS3b_014.d
272 273 274 275 276 277 278 279 280 281 282
(a) (b)100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Sum offeatures
No. of 3
937
2
25 64
3
8011
44
88
414
1
104
09 32
704
890
5
44
88
5084
No. of 2 Commonfeatures
No. of 1 Uniquefeatures
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
Sum offeatures
No. of 3
1384
1 62
1257
99
1
328
1470
3
64
1333 99
1
40
6
No. of 2 Commonfeatures
No. of 1 Uniquefeatures
2 CF/s
1 CF/s
Mid
2 CF/s
1 CF/s
Mid
1500
1000
500
0
-500
-1000
-1500
-2000
-2500-2000 -1500 -1000 -500 0 500 1000 1500
t[2]
t[1]] = 0.186 R2X[2] = 0.152 Ellipse: Hotelling’s T2 (95%)
Control
RenalCancer
SIMCA-22Aug_RenalCancer.M2 (PCA-X)Colored according to classes in M2
1000
500
-500
0
-1000
-1500-2000 -1500 -1000 -500 0 500 1000 1500
t[2]
t[1]] = 0.1233 R2Xo[2] = 0.119 Ellipse: Hotelling’s T2 (95%)
Control
RenalCancer
SIMCA-22Aug_RenalCancer.M3 (OPLS-DA)Scaled proportionally to R2X
Colored according to classes in M3
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
-0.3 -0.2 -0.1 0 0.1 0.2
p(c
orr
)[1]
XV
ar(M
167T
10)
p[1]R2X[1] = 0.123
Obs ID ($ClassID)
SIMCA-22Aug_RenalCancer.M3 (OPLS-DA)
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
-0.3 -0.2 -0.1 0 0.1 0.2
p(c
orr
)[1]
p[1]R2X[1] = 0.123
SIMCA-22Aug_RenalCancer.M3 (OPLS-DA)
Co
ntr
ol
10000
20000
30000
40000
50000Average
-2 std. dev.
2 std. dev.
3 std. dev.
-3 std. dev.
60000
70000
80000
90000
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
SIMCA-22Aug_RenalCancer.M3 (OPLS-DA)
Average
-2 std. dev.
2 std. dev.
3 std. dev.
-3 std. dev.
XV
ar(M
167T
10_1
)
Obs ID ($ClassID)
Co
ntr
ol
40000
45000
50000
55000
60000
65000
70000
75000
80000
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
Co
ntr
ol
Ren
alC
ance
r
SIMCA-22Aug_RenalCancer.M3 (OPLS-DA)