Metabolic Profiling and the Metabolome-Wide Association Study: Significance Level For Biomarker...

Use of an Atmospheric Solids Analysis Probe (ASAP) for High

Throughput Screening of Biological Fluids: Preliminary Applications

on Urine and Bile

Marian Twohig,† John P. Shockcor,† Ian D. Wilson,‡ Jeremy K. Nicholson,§ andRobert S. Plumb*,†,§

Pharmaceutical Business Operations, Waters Corporation, Milford, Massachusetts, AstraZeneca, ClinicalPharmacology, Drug Metabolism and Pharmacokinetics, Alderley Park, Macclesfield, Cheshire SK10 4TG, U.K.,

and Division of Oncology and Surgery, Imperial College, London, SW7

Received February 8, 2010

A hybrid quadrupole orthogonal time-of-flight mass spectrometer (QToF) equipped with a solids analysisprobe (atmospheric solids analysis probe-mass spectrometry (ASAP-MS)) has been applied to the highthroughput qualitative analysis of bile (rat and dog) and urine (rat) samples. The metabolic profilesgenerated by ASAP-MS was less comprehensive than that provided by liquid chromatography (LC) orgas chromatography-mass spectrometry (GC-MS) metabonomic profiling, though simple types ofsample preparation were found to increase the range of ions detected for bile (a complex, multicom-partment sample type). While unsuited to biomarker discovery, ASAP-MS of these biofluids generatedsufficiently complex metabolic fingerprints to enable them to be distinguished from each other usingmultivariate statistical methods such as principal components analysis (PCA). This ability to providean effective means of sample classification suggests possible diagnostic applications.

Keywords: Metabolic profiling • metabonomics • mass spectrometry • biofluid screening • urine • bile.

IntroductionPersonalized medicine strategies, delivered via the applica-

tion of “omics” of various sorts (genomics, proteomics, meta-bonomics, etc.) are believed by many to hold the key to thefuture of drug development and human health care.1 The abilityto determine, and understand, the individual biochemicalprofiles of patients (at the gene, protein, or metabolite level)may well allow the selection of the best combinations of drugs,at optimal doses, to be administered to a patient therebyminimizing nonresponders and adverse events. Personalizingthe treatment to the patient via an appropriate diagnostic testshould also allow drugs that might only work on a subset ofthe diseased population to be developed and marketed withobvious benefits to human health.

Arguably, such biochemical profiling at the population levelrequires a rapid, inexpensive, information-rich, analyticaltechnology that can be easily deployed in a clinical setting.Metabonomics, as shown by a number of studies where, forexample, direct profiling of biofluids such as urine or plasmaby nuclear magnetic resonance spectroscopy (NMR), hasalready provided significant insights into human disease (e.g.,refs 2–4). Such studies, which have further evolved into theconcept of pharmaco-metabonomics,5,6 demonstrate that me-tabonomics is well suited to addressing these challenges.Genomics by comparison does not take into account the effects

of diet or environmental conditions, while proteomics is stillcomparatively expensive and time-consuming. Metabolic pro-filing in order to perform metabonomics was initially per-formed via proton NMR spectroscopy by Nicholson et al.7

However, more recently mass spectrometry, in the form ofliquid chromatography-mass spectrometry (LC-MS), has be-come a useful alternative to NMR and, by providing comple-mentary data, provides greater coverage of the metabolome,8,9

allowing the identification of a wider range of biomarkers. LC-MS is a widely available technology and this will result in itswidespread use for metabonomic research applications. How-ever, the requirement for chromatographic separations to

* Author for correspondence. E-mail: [email protected].† Waters Corporation.‡ AstraZeneca.§ Imperial College, London. Figure 1. Atmospheric solid analysis probe assembly.

3590 Journal of Proteome Research 2010, 9, 3590–3597 10.1021/pr100120g 2010 American Chemical SocietyPublished on Web 05/07/2010

maximize metabolome coverage often results in analysis timesin the region of 15-20 min, although the application of sub 2µm particle LC-MS (UPLC-MS) has allowed run times to bereduced to just 5-10 min or less.10 LC-MS also brings therequirement for an experienced analytical chemist and solvents,calibration, and knowledge to run the system. This is not achallenge when operating in a diagnostic laboratory; however,if MS is to be applied in a field-based diagnostics approachthen a simpler technology is required that can be appliedwithout the need for highly trained analysts, etc.

The advent of direct MS technologies such as desorptionelectrospray ionization (DESI) and direct analysis in real time(DART) has allowed the direct analysis of samples without theneed for an LC separation. While these technologies are notas informative as LC-MS, or especially UPLC-MS, and hencewould not lend themselves to biomarker discovery, they maywell prove to be suitable for the rapid analysis of samples tobe mapped against a validated model. Here we present the useof an atmospheric solids analysis probe (ASAP) as describedby McEwan et al.11 for the high throughput metabonomicscreening of biological fluids such as urine and bile from ratsand dogs.

Experimental Section

Chemicals. Methanol, acetonitrile, formic acid, and am-monium formate were obtained from Sigma-Aldrich (St. Louis,MO, USA). Distilled water 18 MOhm was produced in houseusing a Milli-Q system, Millipore (Billerica, MA, USA).

Animal Samples. Control bile samples (24 h) were obtained,as described previously,12 via bile duct cannulation, from 24male Sprague-Dawley rats, with six samples of control dog bileobtained from two dogs (three samples/dog on three differentoccasions). Quality control (QC) samples, comprised of amixture of each of the bile samples analyzed (e.g., see refs 13and 14) were prepared by careful mixing of an equal volume(20 µL) from each of the bile samples to create a pool samplethat contained a representative mixture of all of the bile

Figure 2. ASAP source.

Figure 3. Rat urine spectrum from ASAP analysis.

High Throughput Screening of Biological Fluids research articles

Journal of Proteome Research • Vol. 9, No. 7, 2010 3591

samples tested. This QC mixture was then treated as per thenormal bile samples.

Rat urine samples (24 h) were also obtained from the bilecannulated rats. All of the samples were stored frozen at -20°C prior to analysis by MS.

Mass Spectrometry Conditions. Mass spectrometry wasperformed on a Waters Xevo QToF hybrid quadrupole orthogo-nal time-of-flight mass spectrometer. The instrument wasoperated in combined ESI/APCI mode (ESCI). This enabled theacquisition of analyte data in APCI mode and reference datain ESI mode. Ionization was performed in ESCI in positive ionmode for the urine samples and negative ion mode for the bileacids using a capillary voltage (ESI) of 3.0 kV with a coronacurrent (APCI) of 5 µA. The source conditions were optimizedto use a cone voltage of 40 V, and aperture 1 setting of 15 V,desolvation temperature of 100-450 °C, desolvation gas 500L/h, and a source temperature of 120 °C. The MS data wereacquired over a range of 100-1000 m/z with a scan durationof 0.5 s, using leucine enkephanin as the lock reference (ESI).

Sample Loading. The samples were analyzed in a randomorder, with biological QC samples intermixed at regular inter-vals between them (every six samples) as well as water blanks.The samples were coated onto the outside of a sealed glasscapillary by simply dipping the capillary into the biofluid andallowing the bulk of the excess to drain off. Any remainingexcess was then removed using a stream of nitrogen and thesample was taken immediately for analysis. A fresh capillarywas used for each analysis.

Data Analysis. The resulting low collision energy MS datafrom the QToF was processed using MarkerLynx XS. The datawas treated, automatically, by the summation of all of thespectra produced in the sample. As there was no separation

technology involved, it was not necessary to time align the dataset for retention time drift. The spectra/intensity from eachsample was normalized for intensity against the sum of thepeak intensities and then multiplied by 10 000 to give a finalresponse for each peak in the spectrum. The data from eachsample was then combined to give an overall data set. Zeroswere inserted in the response table for each sample which didnot detect an ion that was present in the master data set. Afull description of this methodology can be found in U.S. Patent7,418.352.B2.

Results and Discussion

The ASAP process for liquids such as the urine and bileexamined here was both rapid and technically undemanding.The operator simply dips the end of the sealed glass capillaryinto the sample, thereby coating the outside with liquid,followed by removal of the excess. The capillary is then attachedto the MS probe and the whole assembly (Figure 1) is insertedinto the mass spectrometer; a new glass capillary is used foreach sample to minimize the chances of carry over and crosscontamination.

Once the probe has been inserted into the mass spectrom-eter the sample was desorbed from the probe by a heatedstream of gas, which vaporized any volatile compounds. Thesample is ionized in the gas phase by an APCI-like process asdescribed by McEwan et al.,11 Figure 2. This process can beperformed rapidly such that MS data can be acquired at rateof one sample analyzed every 40-60 s (even though currentlysample changing is performed manually). In this study, weobserved no carry over between samples as evidenced by theabsence of signals observed for the water blanks. The acquired

Figure 4. Extracted ion pseudo chromatograms from rat urine analysis by ASAP.

research articles Twohig et al.

3592 Journal of Proteome Research • Vol. 9, No. 7, 2010

MS data takes on the appearance of a broad peak as differentions are vaporized from the probe, as shown in the specificexamples given below.

Urine Analysis. Analysis of the urine samples showed thatit was possible to detect many of the ions previously reportedin LC/MS metabolic profiling studies. Thus, Figure 3 shows thefull scan MS obtained from the positive ion ASAP analysis ofurine, where a total of some 7000 ions were observed including,for example, diagnostic ions from hippurate (m/z )180),pantothenic acid (m/z ) 220), kynurenic acid (m/z ) 190),xanthenic acid (m/z ) 206), phenylpuruvate (m/z ) 165) andindican (indoxyl sulfate) (m/z ) 212). The extracted ion MStraces for these ions are displayed in Figure 4. This result pro-

vides some confidence that, while not providing as compre-hensive a metabolite profile as, for example, UPLC-MS, thetechnique is capable of analyzing the urine samples derivedfrom metabonomic studies and detecting known diagnosticions, and may thus potentially be capable of being used in asample screening/diagnostic role. Multivariate statistical analy-sis of these data for sample classification is discussed below.

Bile Analysis. Bile analysis was initially undertaken onuntreated biofluid samples as for urine. The positive ion MSdata displayed in Figure 5 shows the result obtained for atypical dog bile sample. The two inset MS spectra, A and B,

Figure 5. ASAP pseudo chromatogram and spectra from bile analysis.

Table 1. Table of Masses Identified As Contributing to theObserved Variance in the Data from Bile Analysisa

mass (m/z) p[1]P p(corr)[1]P

229.153 0.0703617 0.77996686.0959 0.113362 0.89805190.064 0.0714874 0.924479

263.241 -0.130702 -0.90099265.256 -0.087938 -0.9417271.27 -0.061647 -0.89616116.066 0.0824036 0.915434118.087 0.151155 0.876469120.072 0.143226 0.801986287.242 -0.073602 -0.9512295.268 -0.064527 -0.97194297.288 -0.061282 -0.9819130.067 0.115261 0.826178299.299 -0.080289 -0.96876132.103 0.286665 0.925362133.107 0.0825892 0.918857132.304 0.0706576 0.758948313.277 -0.170618 -0.98413314.282 -0.076245 -0.98172319.266 -0.073797 -0.96333147.114 0.154719 0.790233331.286 -0.162064 -0.9816332.29 -0.068899 -0.97549337.28 -0.077394 -0.92417339.295 -0.081274 -0.9594341.309 -0.152918 -0.98743342.315 -0.071995 -0.98863

a The values p[1]P relate to the loadings value and p(corr)[1]P to thecorrelation value from the OPLS-DA analysis.

Table 2. Positive Ion ASAP-QToF-MS of Bile Acid Standards

bile acid standard MX- iona second ion third ion

glycholic acid 407 446 464taurocholic acid 406 433 452cholic acid 407Na glycochenodeoxycholate 575 fragment ion 255dehydrocholic acid 401 447lithocholic acid 375 421sodium glycholate 575 fragment ion 255deoxycholic acid 391 437chenodeoxycholic acid 391 437Na taurochenodeoxycholate 390 436Na taurodeoxycholate 436

a The MX- ion is the negative ion nominal mass of detected analyte,with the second and third ions related to fragment or aduct ionsobserved with each bile acid.

Table 3. Marker Ions Derived From The PCA Analysis of theBile Samples Following Addition of Methanol, where p[1] andp[2] Relate to the Coefficient of Variance from the PCALoadings Plot

mass (m/z) p[1] p[2]

101.0212 -0.0892013 -0.034492143.032 -0.0825043 -0.0402257161.0399 -0.149759 0.0368924255.2302 -0.144704 0.0823256279.2279 -0.156761 0.129895283.2599 -0.159877 0.0699177281.2448 -0.145005 0.0777991304.2354 -0.0777542 0.0476693303.226 -0.164246 0.104272390.2989 -0.0756819 0.0440647406.2911 -0.0849576 0.109549436.2953 -0.130356 0.109581



were derived from the beginning and end of the vaporizationprocess, respectively. The data presented clearly show that theMS spectrum obtained from the beginning of data acquisitionwas different from that at the end. The tail end of the peakshows compounds with a much higher m/z value than thoseearly in the sample vaporization, and this is most likely due tothe different boiling points and vapor pressures of the variouscompounds contained in the sample.

These MS data showed a significant number of ions (570 innegative ion mode), but interestingly ions for the bile acids,which might be expected to be significant contributors to theprofile, were absent. As seen with the urine samples, the

number of ions detected for bile was significantly less, by afactor of 10, than that normally observed by UPLC-MS of bile.12

In addition, a simple comparison of the ASAP analyses withthose previously undertaken by UPLC-MS12 confirmed that theions observed were quite different, tending to be more frag-ment-ion related rather than for the expected intact bile acids(the major ions detected for bile are listed in Table 1). Thisdifference may in part have been be due to the mode ofionization, which, as we have indicated, is more APCI-like innature rather than electrospray but may also have been dueto bile acid ionization having been suppressed by theirincorporation into micelles in the bile. When standards of the

Figure 6. Comparison of dog bile and rat bile spectra from ASAP analysis.



Figure 7. PCA (PC1 vs PC2) analysis of entire data set including bioanalysis.

Figure 8. (a) PCA (PC1 vs PC2) analysis of rat urine (blue), rat bile (red), and dog bile (green); (b) PCA (PC1 vs PC2) analysis of rat bile(purple) and dog bile (black) only.



individual bile acids were dissolved in water/methanol 1:1 v/vand analyzed using the ASAP the bile acids were readilydetected providing some support for this notion. The ionsproduced by the ASAP analysis of the bile acid standards werenot the MH- ions but pseudo molecular ions such as, forexample, that observed for taurocholic acid (m/z 452) whichcan be rationalized as resulting from the formation of thesodium adduct (ASAP-MS data for these bile acid standardsare given in Table 2).

The addition of methanol to bile samples provides a meansto disrupt the micelles and should thus facilitate their detectionusing the ASAP probe. The rat and dog bile samples in thepresent study were therefore diluted with an equal volume ofmethanol and then reanalyzed by ASAP QToF MS whereuponsignals relating to bile acids (probably taurocholic, deoxycholicacid, and taurodeoxycholic acid) were detected (see Table 3).Representative MS spectra of the rat and dog bile followingmethanol dilution and analysis using the ASAP are displayedin Figure 6. Following this type of treatment some 7500 ionswere seen in the bile samples. In the dog bile spectrum ionscan clearly be seen corresponding to taurocholic acid at m/z) 452, deoxycholic acid m/z ) 391, and Na taurodeoxycholateat m/z ) 436. In the rat bile data, the taurocholic acid ions atm/z ) 452 and 406, and the Na taurodeoxycholate at m/z )436 were clearly observed.

Metabonomic Analysis. There were approximately 7000 MSfeatures in the combined data set for these (untreated) biofluidswhich, as might be expected from the method of analysis, wassomewhatlowerthanthatpreviouslyreportedusingUPLC-MS.12,14,15

However, this still represents a large number of ions, andcertainly many more than could be examined manually, andtherefore chemometric analysis was performed. Thus, the datafrom the ASAP-QToF MS were subjected to multivariatestatistical analysis, initially using the unsupervised approachof principal components analysis (PCA). The PCA scores plotdisplayed in Figure 7 shows the result (PC1 vs PC2) for all thesamples in the original data set, including rat urine andunprocessed rat and dog bile, together with QC samples,blanks, etc.

From this analysis, the QC samples clustered tightly together,as did the blanks, thus indicating that the intrasample samplerepeatability was acceptable providing some confidence in thequality of the data.13,14 The next step of the data analysis was

to remove the QC samples from the data set for PCA (theresulting scores plot is shown in Figure 8a). Here the dog bilesamples were clearly separated from the rat bile samples, aswas the case when UPLC-MS was performed on the samesamples.12 Perhaps not unexpectedly, given the major differ-ences in metabolite composition between urine and bile, thisanalysis showed that both of the bile data sets were clearlyseparated from those resulting from the rat urine samples.However, while not unexpected the separation of the urinefrom the bile samples, and the rat and dog bile samples fromeach other, demonstrates that sufficient information wasobtained to allow the discrimination of the data set. This givesconfidence that the data generated by the ASAP probe tech-nique, while not suitable for obtaining full metabolite profiles,nevertheless do contain sufficient information to allow the twogroups to be separated suggesting potential, for example, fordiagnostic applications. When the data analysis was restrictedto just the bile samples (Figure 8b), the dog bile samples wereseen to cluster tightly while the rat bile samples showed agreater degree of variability, similar to that seen when thesesamples were analyzed using UPLC/MS.12

The statistical analysis of the bile data obtained aftermethanol treatment also showed excellent clustering of thevarious sample types with excellent grouping of the QCs asbefore. The PCA scores plot shown in Figure 9 shows the resultsfor PC1 and PC2, with the QCs and blanks removed. As can beseen rat and dog bile samples are, as seen without methanoltreatment, clearly separated from each other with the dog bilesamples once again clustering tightly together while those forthe rats showed greater variability. The ions shown as respon-sible for the observed statistical separation of these rat and dogbile samples were then determined by means of an OPLS-DAS-Plot. In this type of statistical treatment, the ions at theextremities of the S-Plot are those that contributed most to theobserved clustering (data not shown) with the ions m/z ) 406,m/z ) 436 (probably representing taurocholic acid ion and theNa taurodeoxycholate ion) identified as key contributing ionsto the statistical analysis.

Discussion

These preliminary studies on urine and bile samples usingthe ASAP probe show that a simple and direct analysis of thebiological samples using ASAP-MS can be performed enabling

Figure 9. PCA analysis of rat and dog bile following addition of methanol prior to ASAP analysis.



the rapid and technically undemanding metabonomic screen-ing of biological samples. Thus, the removal of the hyphenationof the MS to either LC or GC systems means that this approachcan be performed without the need for a highly trainedanalytical chemist and complex separation step, perhapsmaking it a possible platform for in-clinic analysis. However,as the results for bile show, it may still be necessary to performsome sample manipulation in order to detect particularcompounds/metabolites. Thus, while the ASAP technique, aswould indeed be expected for any direct MS analysis method-ology, revealed fewer features for both urine and bile than onebased on, for example, LC-MS successful discrimination be-tween the various sample types was possible via a variety of“diagnostic” ions. The limitations of this device for generatingcomprehensive metabolite profiles clearly indicated that, atleast in its present form ASAP should not be used in studieswhere the aim is investigative sample profiling/biomarkerdiscovery. However, where biomarkers of disease or toxicity,etc. have already been identified using a more comprehensiveapproach such as, for example, UPLC-MS then it may wellprove to be possible to build a rapid and cost-effective methodbased on them that can be deployed in a clinical setting or forpopulation screening in, for example, epidemiological inves-tigations where the aim is diagnosis/screening. However, insuch clinical applications it will clearly be necessary to havefully defined and annotated the sets of marker ions identifiedas discriminating between healthy and diseased subjects forthem to be accepted as biomarkers.

Conclusions

The ASAP-MS process enables rapid analysis of biologicalsamples such as bile and urine. Although the data generatedare less comprehensive than that which can be obtained using

LC or GC-MS, making it unsuited to biomarker discovery, itdoes enable effective sample classification. Simple types ofsample preparation can extend the range of ions detected forcomplex, multicompartment, sample types such as bile.

References(1) Holmes, E.; Wilson, I.; Nicholson, J. Cell 2008, 134, 714–717.(2) Lindon, J. C.; Holmes, J. C.; Bollard, M. E.; Stanley, E. G.; Nicholson,

J. K. Biomarkers 2004, 9, 1–31.(3) Brindle, J. T.; Antti, H.; Holmes, E.; Tranter, G.; Nicholson, J. K.;

Bethell, H. W.; Clarke, S.; Schofield, P. M.; McKilligin, E.; Mosedale,D. E.; Grainger, D. J. Nat Med. 2002, (12), 1439–44.

(4) Mowatt, G.; Zhu, S.; Kilonzo, M.; Boachie, C.; Fraser, C.; Griffiths,T.; N’dow, J.; Nabi, G.; Cook, J.; Vale, L. Health Technol. Assess.2010, 14 (4), 1–356.

(5) Clayton, T. A.; Lindon, J. C.; Cloarec, O.; Antti, H.; Charuel, C.;Hanton, G.; Provost, J. P.; Le Net, J. L.; Baker, D.; Walley, R. J.;Everett, J. R.; Nicholson, J. K. Nature 2006, 440, 1073–1077.

(6) Clayton, T. A.; Baker, D.; Lindon, J. C.; Everett, J. R.; Nicholson,J. K. Proc. Natl. Acad. Sci. U.S.A. 2009, 106, 14728–14733.

(7) Wilson, I. D.; Nicholson, J. K. Prog NMR Spectrosc. 1989, 21, 449–501.

(8) Theodoridis, G.; Gika, H. G.; Wilson, I. D. TrAC - Trends Anal.Chem. 2008, 27, 251–260.

(9) Wu, Z.; Huang, Z.; Lehman, R.; Zhao, C.; Xu, G. Chromatographia2009, 69, S23–S32.

(10) Plumb, R. S.; Granger, J. H.; Stumpf, C. L.; Johnson, K. A.; Smith,B. W.; Gaulitz, S.; Wilson, I. D.; Castro-Perez, J. Analyst 2005, 130,844–849.

(11) McEwan, C. N.; McKay, R. G.; Larsen, B. S. Anal. Chem. 2005, 77,7826–7831.

(12) Plumb, R.S.; Rainville, P. D.; Potts, W. B.; Johnson, K. A.; Gika, E.;Wilson, I. D. J Proteome Res. 2009, 8, 2495–2500.

(13) Gika, H. G.; Theodoridis, G. A.; Wingate, J. E.; Wilson, I. D. J.Proteome Res. 2007, 6, 3291–3303.

(14) Gika, H. G.; Macpherson, E.; Theodoridis, G.; Wilson, I. D.J Chromatogr., B 2008, 871, 299–305.

(15) Wilson, I. D.; Nicholson, J. K.; Castro-Perez, J.; Granger, J. H.;Johnson, K.; Smith, B. W.; Plumb, R. J. Proteome Res. 2005, 4,591–598.

PR100120G



Metabolic Profiling and the Metabolome-Wide Association Study: Significance Level For Biomarker...

Documents

Transcript of Metabolic Profiling and the Metabolome-Wide Association Study: Significance Level For Biomarker...