Biomarker Discovery by Imaging Mass Spectrometry

Adverse drug effects are often associated with pathological changes in tissue. An accurate depiction of the undesired affected area, possibly supported by mechanistic data, is important to classify the effects with regard to relevance for human patients. MALDI imaging MS represents a new analytical tool to directly provide the spatial distribution and the relative abundance of proteins in tissue. Here we evaluate this technique to investigate potential toxicity biomarkers in kidneys of rats that were administered gentamicin, a well known nephrotoxicant. Differential analysis of the mass spectrum profiles revealed a spectral feature at 12,959 Da that strongly correlates with histopathology alterations of the kidney. We unambiguously identified this spectral feature as transthyretin (Ser28–Gln146) using an innovative combination of tissue microextraction and fractionation by reverse-phase liquid chromatography followed by a top-down tandem mass spectrometric approach. Our findings clearly demonstrate the emerging role of imaging MS in the discovery of toxicity biomarkers and in obtaining mechanistic insights concerning toxicity mechanisms.

Proteomic strategies have been exploited to generate differential protein expression that can be correlated with environmental disease or toxicant exposure. In particular, improvements in two-dimensional gel electrophoresis and liquid chromatography coupled to mass spectrometry and the use of other emerging techniques have demonstrated the usefulness of a proteomic approach in toxicology (1)(2)(3). The early detection of disease markers and the rapid screening of experimental compounds either for toxicity or protective efficacy are of particular interest in this field (4).
Imaging mass spectrometry (IMS) 1 is a technique for the direct analysis of peptides and proteins from thin tissue sections using conventional MALDI-TOF mass spectrometers while preserving the abundance and spatial distribution of each analyte (5). The usefulness of this technique has been demonstrated in experimental models and in clinical settings (6,7) as in the accurate and sensitive classification of nonsmall cell lung cancer in tumor biopsies (8).
In this study we further demonstrated the potential applications of an IMS strategy in the study of a well known nephrotoxicant, the aminoglycoside antibiotic gentamicin (9), in the rat kidney. Aminoglycoside antibiotics are widely used, but nephrotoxicity (10) is a clear risk and occurs in 10 -20% (11) of treated patients. Gentamicin-induced nephrotoxicity is seldom fatal and is usually reversible but often results in long hospital stays. Thus, there is a great interest in finding potential markers for the toxicity event and to further elucidate the toxicity mechanism.
We first established our IMS strategy by determining the differential protein expression within the main areas of the rat kidney (cortex, medulla, and papilla) to ensure that we could accurately differentiate the main substructures of this organ. We then investigated whether the kidney lesions secondary to gentamicin treatment could be visualized using IMS. We studied kidney sections from control and drug-treated rats using both a profiling strategy (a few localized measurements per sample, several replicates per animals) and an imaging strategy (systematic, regiospecific measurement on the whole kidney section, one replicate per animal) to evaluate the best approach to depict the histopathology. As we were able to pinpoint several potential markers for gentamicin kidney toxicity, we set up a protein identification strategy combining a protein liquid microextraction step followed by fractionation by reverse-phase (RP) HPLC and sequencing by top-down tandem mass spectrometry. In this study we demonstrated that the fragment (Ser 28 -Gln 146 ) of transthyretin (pre-albumin), a 13-kDa transporter plasma protein, accumulates in large amounts in the cortex of the kidney following gentamicin treatment. We propose therefore that transthyretin might have value as a marker for nephrotoxicity in the monitoring of gentamicin administration.

EXPERIMENTAL PROCEDURES
Animal Treatment-Permission for animal studies was obtained from the Swiss regulatory agencies, and all study protocols were in compliance with animal welfare guidelines. Male Wistar rats of ϳ12 weeks of age (300 g Ϯ 20%) were obtained from Invitrogen. Animals were treated for 7 consecutive days with 100 mg/kg/day gentamicin (Sigma; dissolved in saline) or vehicle by subcutaneous injections and sacrificed by CO 2 inhalation 24 h after the last application.
Sample Preparation-Kidneys were dissected out, snap frozen in liquid nitrogen, and stored at Ϫ80°C prior to further processing. Sections of 12-m thickness were obtained on a cryostat (LEICA CM 3000, Leica Microsystems) at Ϫ18°C and deposited on indium tin oxide-coated conductive glass slides (Delta Technologies). The sample preparation was performed according to a published procedure (12). Tissue sections were fixed by immersion in ethanol baths and allowed to dry for 30 min under vacuum. Matrix deposition was performed either manually with a pipette or automatically with a spotter. Manually 2 ϫ 200 nl of freshly prepared 20 mg/ml sinapinic acid (Fluka) dissolved in 50% acetonitrile, 0.1% trifluoroacetic acid were sequentially deposited onto the tissue section. For automatic spotting, ground sinapinic acid (0.5-3-m particles) was obtained by grinding with a ceramic mortar. Seeding of the tissue was achieved by manual deposition of ground matrix material onto the tissue, and excess crystals were removed with a blow of dry air. A rectangular array of matrix droplets was deposited using a prototype Acoustic Reagent Multispotter (ARM) (13). Matrix spots of 200 -220-m diameter were printed on demand with a 10-Hz drop ejection rate and an array pitch of 250 m. The matrix solution was 25 mg/ml sinapinic acid prepared in 50% acetonitrile, 0.2% trifluoroacetic acid. Two print iterations with 13 droplets each were necessary to obtain dense crystal spots on the tissue.
Mass Spectrometry-MALDI MS spectra of the manually spotted tissue sections were acquired on an UltraFlex MALDI-TOF/TOF instrument (Bruker Daltonics) with a standard 337-nm N 2 laser operating at 50 Hz and a laser spot size of 50 m. The instrument was operated in positive linear mode (2-30 kDa) at constant laser power. A total of 8 ϫ 100 laser shots were accumulated from each deposited matrix droplet. MALDI MS spectra of the robotic spotted tissue sections were obtained on an Applied Biosystems DE STR instrument in linear positive mode. The instrument was operated in delayed extraction mode with a 337-nm N 2 laser operating at 20 Hz. A total of 200 spectra were summed from each matrix spot by rastering the spot with the laser. The data were base line-subtracted and rendered into an ordered array of spectra (image) using a modified Analyze 7.5 (Mayo Clinic) format. The Biomap software package (Novartis) was used for image visualization and to extract selected ion images from the data.
Data Analysis-Base-line subtraction, normalization, peak detection, and spectral alignment were performed using the ProTS-Data software package (Efeckta Technologies, Inc.) A list of 13 common peaks was selected to be internal alignment points, and these common peaks were used to align the spectra according to m/z using a quadratic calibration function. The criteria used to select common peaks were that the peak must occur in greater than 80% of the spectra to be aligned, the peak must not have interfering or overlapping peaks, and the peaks must have a standard deviation of the observed centroid values less than 7 mass units.
The peak lists were binned according to their m/z values using an in-house program developed at Vanderbilt University. The exported MALDI-TOF MS peaks were aligned across samples by use of a genetic algorithm parallel search strategy (8). Briefly peaks were binned together such that the number of peaks in a bin from different samples is maximized while the number of peaks in a bin from the same sample is minimized. A series of mass windows or peak bins were generated that separated similar peaks across multiple spectra. The spectral features were ranked according to the extent of the observed difference to determine relevant biomarkers for gentamicininduced kidney toxicity using a combination of three different criteria: weight value (14,15), average signal-to-noise ratio (S/N), and t test.
Protein Identification-Protein liquid microextraction was achieved by directly pipetting up and down for 5 s 1 l of extraction solvent (50% acetonitrile, 0.1% trifluoroacetic acid) on the region of interest of the tissue section. After collection of ϳ20 l (the pool of 40 microextractions over three sections), the sample was dried and resolubilized in water/acetonitrile (95:5), 0.1% trifluoroacetic acid. The peptide mixture was fractionated onto a 1-mm inner diameter polymeric column (catalog number 219TP5110, Vydac) using a linear acetonitrile gradient delivered at 50 l/min. Eluting fractions were directly collected into a 96-well microtiter plate at a rate of 30 s/fraction. The elution position of the peaks of interest was assessed by spotting 1 l of each fraction with 1 l of 1.5 g/liter sinapinic acid solution in water/acetonitrile/TFA (90:100:0.1) onto a 384 AnchorChip MALDI target (Bruker Daltonics). The mass profiles were recorded by MALDI MS using the same acquisition parameters as for tissue imaging. For sequencing, the HPLC fractions of interest were pooled, dried in a SpeedVac, and resolubilized in water/acetonitrile (1:1) containing 1% formic acid. About 1 ml of the resolubilized fraction was transferred directly into a nanoelectrospray capillary needle. Mass spectra were acquired on a QSTAR Pulsar i quadrupole TOF tandem mass spectrometer (Applied Biosystems/MDS-Sciex) equipped with a nanoelectrospray ion source (Proxeon). The proteins of interest were identified by tandem mass spectrometry using a top-down approach (16,17). One or several multiply charged parent masses were selected to be fragmented for sequence analysis in the QSTAR instrument. Raw spectra analysis and peak list generation were performed using the instrument-provided Analyst/Bioanalyst QS software, Version 1.4 (Applied Biosystems/MDS-Sciex). After manual inspection of the tandem mass spectrum for determination of a sequence tag, protein identification was performed with the Mascot sequence query search program (Matrix Science, Mascot Version 2.1.03) using the Swiss-Prot database (release 46.6) filtered for the taxonomy "Rattus" (5,327 sequences) and a tolerance of 1.0 Da for fragment and precursor masses. MS/MS score evaluation was performed as follows. We routinely performed a first database search using the singly charged fragment ions manually determined to form a sequence tag, and we considered all protein sequences for which a MOWSE score higher than 33 (indicating significant homology as displayed by the Mascot search engine) could be obtained. The hit had to be further confirmed by taking into account all observed sequence ions in the spectrum, in particular the multiply charged fragment ions typically present around the parent mass. Only protein sequences with a MOWSE score higher than 46 (indicating extensive homology or identity, expectancy p Ͻ 0.05) obtained through a second database search including all pieces of evidence were considered in this work. Methionine oxidation (as variable modification) and N-terminal acetylation (as fixed modification) were only considered if there was no unmodified protein candidate satisfying the conditions described above.
Western Blot-Cytosolic fractions from whole kidney tissue extract were obtained by homogenization of a piece of control and treated kidney cortex followed by successive centrifugation steps at 680 ϫ g, 10,000 ϫ g, and 100,000 ϫ g. For Western blot analysis, 8 g of a cortex homogenate or an HPLC fraction of interest (dried in a Speed-Vac and resuspended in SDS buffer) were applied onto a 4 -20% Tris-glycine SDS-polyacrylamide gel electrophoresis system (Invitrogen). Gels were subsequently transferred onto a polyvinylidene fluoride membrane (Polyscreen, PerkinElmer Life Sciences). Blots were blocked in 5% milk in PBS containing 0.1% Tween 20 for 1 h and incubated overnight at room temperature with anti-sheep transthyretin (Abcam, catalog number ab9015) diluted 1:5,000. Detection was performed by enhanced chemiluminescence (Amersham Biosciences) after a 1-h incubation with horseradish peroxidase-conjugated anti-sheep IgG (Silenus).
Immunohistochemistry-Immunohistochemistry was performed using a standard peroxidase-based staining method. Paraffin-embedded tissue sections (6 m) were deparaffinized with xylene, and a) FIG. 1. IMS analytical approach. a, frozen rat kidney cut longitudinally and immobilized on a cryostat target. b, 12-m section deposited on an indium tin oxide-coated conductive glass slide. c, profiling strategy by manual deposition of saturated sinapinic solution on the tissue section. Matrix spots were ϳ1.5 mm in diameter. d, imaging strategy using a matrix-dispensing robot. An ordered array of matrix spots with 200 -220-m diameter was printed on the whole tissue section. e, acquisition of MALDI MS spectra. f and g, typical MALDI MS spectra from manually deposited spots in the cortex and in the papilla, respectively. h-j, examples of MALDI imaging data. Tissue distribution of masses representative for the cortex (red), the medulla (blue), and the papilla (green). endogenous peroxidase activity was quenched with 3% H 2 O 2 in methanol for 1 h. Blocking was performed with an avidin-biotin blocking kit (Vector Laboratories, catalog number SP-2001) and with 2.5% horse serum albumin (from Vectastain kit, see below) for 1 h in each solution. Successive incubations with the primary transthyretin antibody (Abcam, catalog number ab9015, at a dilution of 1:1,000), the secondary antibody, streptavidin, and the chromogen (3,3Ј-diaminobenzidine-nickel, Peroxidase Substrate DAB Kit SK-4100, Vector Laboratories) were performed according to the kit manufacturer's instructions (Vector Laboratories, RTU Vectastainா Universal Quick Kit PK-7800). Contrasts were increased by counterstaining the sections with cresyl violet.

RESULTS
Differential Protein Expression in the Kidney Cortex, Medulla, and Papilla-The potential of an IMS strategy to detect toxicity-related differences in protein expression was explored by exploiting the structural differences of the kidney and by analyzing peptides and proteins specific to each of those regions. The cortex, medulla, and papilla of kidneys from control rats were analyzed by two different analytical strategies ( Fig. 1): a profiling strategy in which spots of matrix were manually deposited on selected areas of the tissue sections and an imaging strategy in which whole sections were arrayed with a robotic matrix spotter.
Spectra generated from manual spotting clearly differentiated the cortex from the papilla (Fig. 1, f and g). In each spectrum, around 250 peaks could be detected from m/z 3,500 to m/z 25,000 with the majority of the peaks between m/z 5,000 and m/z 13,000. The peak at m/z 5,486 was clearly one of the major spectral features of the cortex, whereas the peak at m/z 9,940 was especially abundant in the papilla. In contrast, the analysis of a whole kidney section by an imaging strategy clearly differentiated the cortex and the papilla from the medulla, an intermediate region of the kidney that cannot be easily spotted by hand due to its relatively small size. Fig.  1, h-j, shows a representative example of ion images from regionally discriminating peaks. The cortex was clearly delineated by the peak centered around m/z 17,514, the papilla was demarcated by the peak at m/z 9,940 (also highlighted in FIG. 2. Protein identification strategy. The protein extraction was performed by directly pipetting up and down solvent on the surface of the section. This process was repeated several times to obtain enough material for sequencing purpose. The pool was then dried, resolubilized, and separated by RP HPLC. Fractions were collected and first spotted on a MALDI target to select the fraction of interest containing the marker highlighted by IMS. The fraction was finally dried and resolubilized for sequence analysis by nano-ESI tandem mass spectrometry. the manual spotting), and the medulla was marked by the peak at m/z 6,542.
Protein Identification Strategy-A systematic identification of the spectral features (peaks) analyzed by IMS provides valuable information, further validating the accuracy of the technique and the biological significance of the candidate biomarkers.
The strategy presented here is based on an approach mimicking the imaging process (Fig. 2). Proteins and peptides were directly eluted from tissue by microextraction, fraction-ated by RP HPLC, and surveyed by MALDI MS. After manual spectral correlation, fractions of interest were collected and analyzed by top-down sequencing. Using this strategy we attempted to identify the major components of each kidney substructure as observed in IMS. The major peak in the cortex at m/z 5,486 was identified as cytochrome c oxidase polypeptide VIIc (Ser 17 -Lys 63 ). Several additional cortex polypeptides were also characterized, and these were mostly ubiquitous proteins such as cytochrome c oxidase polypeptide VIc-2

TABLE I Potential biomarkers for gentamicin-induced kidney toxicity using the profiling strategy
Ranking is by weight value, maximum average, and p value. N/A denotes that the average signal-to-noise ratio cannot be accurately computed due to the frequent absence of that ion in the corresponding group. C, cortex; P, papilla. . In contrast, the major peak in the papilla at m/z 9,940 could not be identified. Although our analytical procedure was rather straightforward, ϳ50% of the polypeptides analyzed could not be identified. For some, we could not generate an informative tandem mass spectrum due to low signal intensity or because we could not find a suitable parent mass for fragmentation. In other cases, however, we were able to obtain high quality tandem mass spectra that could not be correlated to a polypeptide sequence in the database, probably due to an unknown amino acid modification or other post-translational modification. Application to Gentamicin Nephrotoxicity-Gentamicin, a well known nephrotoxicant, has been associated with proximal tubular cell damage in the cortex due to the reduction of protein synthesis and altered phospholipid metabolism (18). Histopathological examination revealed marked degeneration and regeneration of proximal tubules in rats treated with 100 mg/kg/day gentamicin for 7 days (Fig. 3). We attempted to correlate this finding by IMS as follows. We ran a pilot study comprising three kidneys from rats treated with 100 mg/kg/ day gentamicin for 7 days and three kidneys from control rats. Three sections were generated from each kidney, and each section was manually profiled using five spots in the cortex and two spots in the papilla (Fig. 1c). Table I shows a representative excerpt of the masses that were shown to be differentially regulated in the control compared with the treated samples based on the weight value, the maximum average S/N, and p value. Most of the discriminating peaks were found in the cortex, consistent with the site of action of gentamicin and with the fact that the renal cortex, which represents the major portion of the kidney and receives 90% of the kidney blood flow (compared with 6 -10% for the medulla and 1-2% for the papilla), would be most sensitive to a toxicant (19). The first ranked peak at m/z 12,959 was found in the cortex as being massively up-regulated upon gentamicin treatment (maximum average S/N, 5) as was the fourth peak (maximum average S/N, 2) at m/z 6,479, its doubly charged species. The third peak at m/z 4,497 was of particular interest as this spectral feature was ubiquitously found in the cortex and in the medulla/papilla area in controls and was significantly down-regulated in the whole treated kidney, offering a first line of evidence that the site of action of gentamicin could also include the more distal tubular structures.
The differential features highlighted in the profiling mode were confirmed through an imaging approach using one kidney section of each group (Fig. 1d). Data analysis from the imaging experiment was in most cases consistent with results obtained from the profiling approach. For example, the peaks at m/z 12,959 (Fig. 4a) and its doubly charged species at m/z 6,479 (data not shown) were significantly detected in the cortex of the treated kidney section but not in the control section. Furthermore the imaging approach highlighted a number of additional discriminating spectral features that were not detected in the profiling dataset. For example, a peak at m/z 8,242 (Fig. 4b) was detected both in the medulla and in the cortex of the control kidney section, but it was significantly down-regulated only in the cortex of the gentamicin-treated section. A similar pattern was also observed for several other peaks, such as m/z 8,127, m/z 8,457, and m/z 8,369 (data not shown).
We focused our efforts in identifying the peak of m/z 12,959 that were clearly highlighted by both the imaging and the profiling approaches to be discriminating for gentamicin nephrotoxicity. Pools of ϳ120 protein liquid microextractions from the renal cortex of treated and control sections were fractionated in three HPLC runs to gain enough material for sequencing purposes. The peak of m/z 12,959 was only found in fraction 39 of the treated sample by survey MALDI MS, and this fraction was then analyzed by top-down tandem mass spectrometry. We identified the peak of m/z 12,959 as transthyretin (Swiss-Prot:TTHY_RAT; Ser 28 -Gln 146 ; protonated calculated mass, 12,959.5 Da) (Fig. 5), a blood transporter protein characterized as a biomarker of nutritional status (20 -22). This finding was first confirmed by Western blot analysis using an antibody highly specific for this protein. A band was clearly detected around 13 kDa both in situ (whole tissue extract) and in the relevant HPLC fraction from treated kidneys, whereas it was absent from control kidneys (Fig. 6). The distribution of transthyretin was further investigated by immunohistochemistry. Elevated levels of transthyretin were observed in the cortex of kidney sections from treated animals, whereas there was no apparent accumulation in the medulla or in the kidney section from control animals (Fig. 7), corroborating the results obtained by IMS. DISCUSSION The cortex and medulla/papilla represent two distinct structural and functional areas of the kidney that are clearly delineated by IMS, substantiating that regiospecific patterns of protein expression can directly be observed at the proteome level using this strategy. In the present study, we attempted to identify directly at the tissue level structural and functional changes in an organ that has been affected by a toxicant. It was important to use a well established biological model whose outcome is confirmed by an independent methodology. Gentamicin-induced nephrotoxicity is extensively described in the literature (9,11,18), and gentamicin-induced nephrotoxicity in rat was confirmed in this study by histopathology.
We applied a mixed profiling and imaging strategy followed by targeted protein identification by protein microextraction and top-down tandem mass spectrometry to study the lesions secondary to gentamicin administration in the rat kidney. The uptake of gentamicin (cationic at a physiological pH) is thought to be carried out either passively by diffusion with acidic phospholipids of the brush-border membrane from the proximal tubular cells or through receptor-mediated endocytosis, leading to lysosomal accumulation and ultimately resulting in renal toxicity (10). We demonstrated that the nephropathy secondary to treatment with aminoglycoside antibiotics might be characterized by the accumulation in the cortex of affected kidneys of transthyretin Ser 28 -Gln 146 . Transthyretin is a plasma carrier of both retinol-binding protein and thyroxine (22) and has been described as a biomarker for nutritional status in various diseases (20,21) such as pancre- atitis (23) and end stage renal disease (24). A truncated form of transthyretin lacking the first NH 2 -terminal 10 amino acids has also been proposed for the detection of early stage ovarian cancer (25). More relevant to our study, transthyretin has also been described as a ligand of megalin (26), an endocytic receptor expressed at the apical membrane of renal proximal tubules that has been reported to also mediate gentamicin uptake (27,28). Among a number of hypotheses that could explain the accumulation of transthyretin in proximal tubules, several lines of evidence indicate that megalin could preferentially bind gentamicin and prevent the efficient reabsorption of transthyretin into the bloodstream (26 -29). Other examples of competition with aminoglycosides for megalin have been reported. Calcium binding to megalin is inhibited in vitro by aminoglycosides (29), and Cui et al. (30) reported that gentamicin almost completely inhibits the uptake of bovine serum albumin by purified megalin in proximal tubular cells.
Further investigations to refine a model for transthyretin accumulation upon gentamicin treatment could include the quantitative evaluation of transthyretin by IMS and immunohistochemistry from animals treated at different doses and durations to confirm its dose-dependent accumulation at the target tissue. Our first analysis by immunohistochemistry confirmed the elevated levels of transthyretin in the cortex of kidney sections from gentamicin-treated rats as observed by IMS. However, a direct correlation between the intensity of the mass spectrometric signal and transthyretin abundance in the tissue remains to be demonstrated.
In addition, larger studies including several types of known nephrotoxicants could demonstrate whether transthyretin holds as a more general biomarker for nephrotoxicity and whether the detected changes of protein patterns precede FIG. 7. Immunohistochemical examination of kidney from control and gentamicin-treated rat. Scanned digitized images of whole kidney sections from a treated (A) and a control (B) rat show high levels of transthyretin immunoreactivity in the treated kidney cortex. High magnification photomicrographs from the cortex (C and E) versus the medulla (D and F) reveal specific localization of transthyretin immunoreactivity in the cortex tubules of the treated kidney. Transthyretin immunoreactivity is visualized by the gray precipitate of diaminobenzidine-nickel counterstained with cresyl violet. renal dysfunction and histopathological alteration. If so, IMS could represent a powerful complementary approach to histopathology in discovering early protein targets of the toxicant and would facilitate the classification of new compounds with known toxicants according to common pharmacological and toxicological features. Such predictive models exist based on gene expression data (31) but not on proteomic platforms so far.
The present study clearly highlights the advantages and limitations of the present IMS approaches. A profiling strategy by manually spotting matrix on predefined areas of interest has the advantages of speed, robustness, and good reproducibility. However, this approach is biased by design as it requires manual intervention, and only a fragmentary analysis of the tissue at low spatial resolution can be obtained. Alternatively an imaging strategy by automatic spotting of matrix on the tissue in an array format results in a comprehensive structural analysis at a higher spatial resolution. Although we were able to uncover some treatment-related differences using a profiling strategy, a clear advantage of the imaging strategy is to go further into structural or morphological detail. In this study, a differential comparison of the much richer imaging dataset was essential to ascertain the regiospecificity of the peaks selected for protein identification in the profiling dataset. Further reducing the raster width or optimizing a reproducible homogeneous matrix coating deposition would be an absolute requirement if the functional anatomy of the organ, such as the proximal tubule or glomerulus of the kidney, is to be assessed in a differential study. However, an imaging strategy will heavily depend on the development of software packages for the rapid and accurate differential analysis of vastly more complex datasets than those generated in this study.
A critical step of an IMS-based biomarker discovery approach remains the unambiguous identification of a potential protein target of interest. In this study, we applied an innovative protein identification strategy that closely preserves the biological content observed by IMS. In particular, a local and specific liquid microextraction allows the conservation of the protein mixture composition observed by IMS, thus avoiding the dilution effect of a classical lysis protocol. Proteins of interest can be further purified using standard fractionation methods, and the biologically relevant isoforms can then be unambiguously determined using top-down sequencing by tandem mass spectrometry. Protein identification strategies, such as described in this study, are essential to enable the study and the validation of candidate biomarkers using orthogonal methods in a large number of samples using, for example, Western or ELISA-based assays.
* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
¶ To whom correspondence should be addressed: F. Hoffmann-La