Anal. Chem. 2004, 76, 1560-1570
Serum Peptide Profiling by Magnetic Particle-Assisted, Automated Sample Processing and MALDI-TOF Mass Spectrometry Josep Villanueva,†,‡ John Philip,† David Entenberg,§ Carlos A. Chaparro,† Meena K. Tanwar,|,⊥,# Eric C. Holland,|,⊥,# and Paul Tempst*,†,‡
Protein Center, Research Engineering, Department of Surgery, Department of Neurology, Molecular Biology Program, and Cancer Biology and Genetics Program, Memorial Sloan-Kettering Cancer Center, New York, New York 10021
Human serum contains a complex array of proteolytically derived peptides (serum peptidome) that may provide a correlate of biological events occurring in the entire organism; for instance, as a diagnostic for solid tumors (Petricoin, E. F.; Ardekani, A. M.; Hitt, B. A.; Levine, P. J.; Fusaro, V. A.; Steinberg, S. M.; Mills, G. B.; Simone, C.; Fishman, D. A.; Kohn, E. C.; Liotta, L. Lancet 2002, 359, 572-577). Here, we describe a novel, automated technology platform for the simultaneous measurement of serum peptides that is simple, scalable, and generates highly reproducible patterns. Peptides are captured and concentrated using reversed-phase (RP) batch processing in a magnetic particle-based format, automated on a liquid handling robot, and followed by a MALDI TOF mass spectrometric readout. The protocol is based on a detailed investigation of serum handling, RP ligand and eluant selection, small-volume robotics design, an optimized spectral acquisition program, and consistent peak extraction plus binning across a study set. The improved sensitivity and resolution allowed detection of 400 polypeptides (0.8-15-kDa range) in a single droplet (∼50 µL) of serum, and almost 2000 unique peptides in larger sample sets, which can then be analyzed using common microarray data analysis software. A pilot study indicated that sera from brain tumor patients can be distinguished from controls based on a pattern of 274 peptide masses. This, in turn, served to create a learning algorithm that correctly predicted 96.4% of the samples as either normal or diseased. The complex nature of the molecular, cellular, and clinical information needed for better cancer diagnosis and therapy suggests that a single parameter may not answer all of the critical questions.1 Instead, a panel of individual tests and measurements * To whom correspondence should be addressed: (e-mail) p-tempst@ mskcc.org; (phone) (212) 639-8923. † Protein Center. ‡ Molecular Biology Program. § Research Engineering. | Department of Surgery. ⊥ Department of Neurology. # Cancer Biology and Genetics Program. (1) Etzioni, R.; Urban, N.; Ramsey, S.; McIntosh, M.; Schwartz, S.; Reid, B.; Radich, J.; Anderson, G.; Hartwell, L. Nat. Rev. Cancer 2003, 3, 243-252.
1560 Analytical Chemistry, Vol. 76, No. 6, March 15, 2004
that could be used in sum could provide the needed information. Cancer diagnostics has therefore begun to shift from traditional single markers to molecular “signatures”, also described as “bar codes”, derived from the simultaneous detection of multiple bioanalytes.2,3 Diagnostic mRNA profiling by microarray analysis is a prominent example of this new trend.4-6 The more discrete data points used in sum, the greater the ability for the collection of data to describe the biology of a tumor. Ideally, these measurements would be done in a single readout and on material that is easily accessible. In the case of brain tumors, for example, where multiple measurements of the tumor tissue over time are difficult, the analysis of biologic samples such as urine or serum has great advantages. Human serum contains thousands of peptides, most of which are thought to be fragments of larger proteins that have been partially degraded by endogenous, proteolytic enzymes, but the precise identities remain undetermined.7-9 The complex array that they create, however, may provide a novel and robust correlate of the biological events occurring in the entire organism. The presence and molecular mass of polypeptides in unfractionated mixtures can be directly determined by matrix-assisted laser desorption/ionization (MALDI) time-of-flight (TOF) mass spectrometry (MALDI-TOF MS) at the sensitivities and resolution that would make it a great technique for serum peptide profiling.10,11 (2) Wulfkuhle, J. D.; Liotta, L. A.; Petricoin, E. F. Nat. Rev. Cancer 2003, 3, 267-275. (3) Petricoin, E. F.; Zoon, K. C.; Kohn, E. C.; Barrett, J. C.; Liotta, L. A. Nat. Rev. Drug Discovery 2002, 1, 683-695. (4) Perou, C. M.; Sorlie, T.; Eisen, M. B.; van de Rijn, M.; Jeffrey, S. S.; Rees, C. A.; Pollack, J. R.; Ross, D. T.; Johnsen, H.; Akslen, L. A.; Fluge, O.; Pergamenschikov, A.; Williams, C.; Zhu, S. X.; Lonning, P. E.; BorresenDale, A. L.; Brown, P. O.; Botstein, D. Nature 2000, 406, 747-752. (5) Alizadeh, A. A.; Eisen, M. B.; Davis, R. E.; Ma, C.; Lossos, I. S.; Rosenwald, A.; Boldrick, J. C.; Sabet, H.; Tran, T.; Yu, X.; Powell, J. I.; Yang, L.; Marti, G. E.; Moore, T.; Hudson, J., Jr.; Lu, L.; Lewis, D. B.; Tibshirani, R.; Sherlock, G.; Chan, W. C.; Greiner, T. C.; Weisenburger, D. D.; Armitage, J. O.; Warnke, R.; Levy, R.; Wilson, W.; Grever, M. R.; Byrd, J. C.; Botstein, D.; Brown, P. O.; Staudt, L. M. Nature 2000, 403, 503-511. (6) Golub, T. R.; Slonim, D. K.; Tamayo, P.; Huard, C.; Gaasenbeek, M.; Mesirov, J. P.; Coller, H.; Loh, M. L.; Downing, J. R.; Caligiuri, M. A.; Bloomfield, C. D.; Lander, E. S. Science 1999, 286, 531-537. (7) Adkins, J. N.; Varnum, S. M.; Auberry, K. J.; Moore, R. J.; Angell, N. H.; Smith, R. D.; Springer, D. L.; Pounds, J. G. Mol. Cell. Proteomics 2002, 1, 947-955. (8) Tirumalai, R. S.; Chan, K. C.; Prieto, D. A.; Issaq, H. J.; Conrads, T. P.; Veenstra, T. D. Mol. Cell. Proteomics 2003. (9) Anderson, N. L.; Anderson, N. G. Mol. Cell. Proteomics 2002, 1, 845-867. 10.1021/ac0352171 CCC: $27.50
© 2004 American Chemical Society Published on Web 02/14/2004
Samples are typically introduced as solid crystals on metal “targets” after prior mixing with a UV-absorbing chemical (the “matrix”). Vaporization and ionization are accomplished by a pulsed laser beam, and gas-phase peptide ions are accelerated in an electric field and flight times through a field-free zone recorded and converted to mass-to-charge values. Using a related technology known as surface-enhanced laser desorption/ionization (SELDI) TOF MS, a differential display method for peptides in cancer patient sera has been described by Petricoin, Liotta, and coworkers, and others.12-14 The patterns that they generated appear to hold important information that may have direct clinical utility as a surrogate marker for detection and classification of cancer.2,3 The term SELDI, often used to make a distinction from MALDI, is an ambiguous denotation in that “matrix” is in fact also used. The real difference is that peptides are captured and concentrated in situ on the LDI targets, which are surfacederivatized with certain ligands (e.g., reversed phase (RP), ion exchange, etc.) that bind polypeptides. SELDI-TOF spectra of serum peptides shown in most published reports exhibit low complexity, i.e., less than 100 peaks, whereas previous studies have put the peptide population (“peptidome”) in human blood plasma at ∼5000 different species.15 The lower peptide count could be the result of overly selective capture, insufficient concentration of the analytes before analysis, or both, and also because the commonly used commercial instrument has lower mass resolution. SELDI solves one major problem, however, the removal of nonpeptidic serum components and salts that interfere with proper crystal formation, ionization, and detection. Groups of peptides can also be analyzed using various statistical algorithms to discover hidden patterns as cancer markers and for more subtle aspects of disease management such as tumor recurrence, response to therapy, and a prediction of survival.16,17 As the significance levels of the statistical tests that select discriminatory peptides are increased accordingly, and ever fewer peptides will make the cut, larger numbers will be needed to start a selection process. To this end, true high-resolution mass spectrometric analysis is required. We describe a novel technology platform for the simultaneous measurement of large numbers of serum polypeptides using reversed-phase, magnetic particle-based sample processing with a MALDI-TOF MS readout. High sensitivity and resolution allow detection, within a molecular mass range of 800-15 000 Da, of over 400 polypeptides in a single droplet of serum. The technology (10) Mann, M.; Hendrickson, R. C.; Pandey, A. Annu. Rev. Biochem. 2001, 70, 437-473. (11) Hillenkamp, F.; Karas, M.; Beavis, R. C.; Chait, B. T. Anal. Chem. 1991, 63, 1193A-1203A. (12) Petricoin, E. F.; Ardekani, A. M.; Hitt, B. A.; Levine, P. J.; Fusaro, V. A.; Steinberg, S. M.; Mills, G. B.; Simone, C.; Fishman, D. A.; Kohn, E. C.; Liotta, L. A. Lancet 2002, 359, 572-577. (13) Li, J.; Zhang, Z.; Rosenzweig, J.; Wang, Y. Y.; Chan, D. W. Clin. Chem. 2002, 48, 1296-1304. (14) Adam, B. L.; Qu, Y.; Davis, J. W.; Ward, M. D.; Clements, M. A.; Cazares, L. H.; Semmes, O. J.; Schellhammer, P. F.; Yasui, Y.; Feng, Z.; Wright, G. L., Jr. Cancer Res. 2002, 62, 3609-3614. (15) Richter, R.; Schulz-Knappe, P.; Schrader, M.; Standker, L.; Jurgens, M.; Tammen, H.; Forssmann, W. G. J. Chromatogr., B: Biomed. Sci. Appl. 1999, 726, 25-35. (16) Qu, Y.; Adam, B. L.; Yasui, Y.; Ward, M. D.; Cazares, L. H.; Schellhammer, P. F.; Feng, Z.; Semmes, O. J.; Wright, G. L., Jr. Clin. Chem. 2002, 48, 1835-1843. (17) Wu, B.; Abbott, T.; Fishman, D.; McMurray, W.; Mor, G.; Stone, K.; Ward, D.; Williams, K.; Zhao, H. Bioinformatics 2003, 19, 1636-1643.
is automated on a liquid handling robot for throughput and reproducibility. Using this new analytical platform and GeneSpring statistical software, we found that sera from brain tumor patients can be readily distinguished from controls based on a complex pattern of 274 discriminatory peptides. When used with a class prediction algorithm, that pattern allowed us to correctly classify over 96% of the samples as either normal or diseased. In sum, the system that we describe here satisfies all criteria of MALDI compatibility, high resolution, reproducibility, and throughput, and the limited application provides further proof of the concept that sera from solid tumor cancer patients contain peptides detectable by MALDI-TOF MS that reflect the activity of the cancer. MATERIALS AND METHODS Materials. Acetonitrile was obtained from Burdick and Jackson (Muskegon, WI), trifluoroacetic acid (TFA) and urea were from Pierce (Rockford, IL), n-octylglucoside was from BoehringerMannheim (Indianapolis, IN), dithiothreitol (DTT) was from BioRad (Hercules, CA), and R-cyano-4-hydroxycinnamic acid (HCCA), dihydroxybenzoic acid (DHB), and sinapic acid were from Bruker Daltonics (Billerica, MA). Serum peptide preparations were done in 0.2-mL polypropylene tubes (8-tube strips; or 8 × 12-tube Temp Plate II) from USA Scientific (Ocala, FL). Serum Samples. Blood samples from volunteer subjects with no known malignancies and from consenting patients diagnosed with glioblastoma were collected in 8.5-mL, BD Vacutainer, glass “red-top” tubes (366430; BD, Franklin Lakes, NJ), allowed to clot at room temperature for up to 1 h, and centrifuged at 4 °C for 5 min at 1000 rpm (1000g). Sera (upper phase) were aliquoted and stored frozen at -80 °C until further use. Patient and control sera were collected following a clinical protocol approved by the Memorial Sloan Kettering Institutional Review Board; YKL-40 protein levels were determined by ELISA prior to mass spectrometric analysis described in this study.18 Serum Pretreatments. In selected cases, sera were additionally subjected to incubation with additives, proteins were removed by precipitation or filtration, or serum albumin was removed by affinity chromatography. An aliquot corresponding to 50 µL of untreated serum was then used for RP-magnetic particle batch separation. Additives. Sera were incubated for 30 min at room temperature in either 8 M urea, 8 M urea/20 mM DTT, or 0.2% n-octylglucoside. Ethanol Precipitation. Sera were mixed with an equal volume of 60 or 100% ethanol by vortexing for 1 min at room temperature and centrifuged at 15000g for 10 min, and the supernatant was taken for further analysis. Filtration. Sera were passed through a Centricon 30 ultrafiltration device (Amicon; Beverly, MA), operated as per the manufacturer instructions. In separate experiments, sera were incubated in 8 M urea for 30 min at room temperature before ultrafiltration. Cibachron Blue Chromatography. Sera were incubated with magnetic, Cibachron Blue-derivatized beads (Cortex Biochem; San Leandro, CA), using a ratio of 20 µL of serum/10 mg of swollen beads, by 10 cycles of pipeting up and down. Particles were pulled (18) Tanwar, M. K.; Gilbert, M. R.; Holland, E. C. Cancer Res. 2002, 62, 43644368.
Analytical Chemistry, Vol. 76, No. 6, March 15, 2004
1561
to the side of the tube with a magnet (see below), and the supernatant was collected for further analysis. Reversed-Phase Batch Separations. Superparamagnetic, silica-based particles (e1 µm diameter; 80% iron oxide), surfacederivatized with common reversed-phase ligands, were obtained from Chemicell (Berlin, Germany). Particles bearing ligands of increasing carbon chain length (SiMAG-C1, -C2, -C3, -C8, -C18), and with nonporous or porous (C1/K, etc..) surfaces were comparatively tested to optimize RP batch separation of serum peptides using a MALDI-TOF MS readout (see below). Following selection of “C8/K” beads, ligand density and magnetic properties were empirically optimized in a series of pilot-scale syntheses and serum peptide binding tests. Three different batches of particles with relatively similar properties were used throughout this study. Particles for use in robotic systems are preferably of somewhat lesser hydrophobicity to prevent aggregation and clogging of pipeting tips. The standard protocol, except where modified for investigative and optimization purposes, was as follows. A suspension of C8/K magnetic particles (500 000 particles/µg; 50 µg/µL DD water) was thoroughly mixed for 2 min by vortexing to obtain homogeneous dispersion. Then, 50 µL of bead suspension was added to 50 µL of untreated human serum and mixed by slowly pipeting up/down five times. After pulling the magnetic particles to the side of the tube with a magnet (specifications under Automated Sample Preparation), the supernatant was removed and discarded. Beads were washed twice with 200 µL of 0.1% TFA in water by moving the tubes, in either an 8-tube strip or 8 × 12-tube plate layout, five times back and forth over one or more magnets, resulting in a laterally jarring motion. Peptides were then eluted with 5 µL of 50% acetonitrile by pipeting the beads up/down 10 times. After collecting the magnetic particles with a magnet to the side, 3 µL of the eluate was transferred to another tube and mixed with 6 µL of premade matrix solution, and 1 µL was deposited for MALDI-TOF MS (see Mass Spectrometry). In some experiments, peptides were stepwise eluted from the RP particles with 5-µL volumes of 20% and then 70% acetonitrile, and the two eluates were analyzed by MS. Two-Dimensional Batch Separations. 2D-RP. Fifty microliters of a C1- or C2-derivatized magnetic particle suspension (in water) was mixed with 50 µL of serum by pipeting up/down five times. After pulling the magnetic particles to the side, the supernatant was aspirated and transferred to a different tube and used to carry out standard RP C8 batch separation. Polypeptides bound to the C1/C2 beads were also eluted with 50% acetonitrile and separately analyzed by MALDI-TOF MS. Cation Exchange-RP C8. Serum (500 µL) was diluted with two volumes of 75 mM sodium acetate, pH 4.0, mixed with a 300-µL bed volume of Toyopearl SP-550C (TosoHaas, Montgomeryville, PA) strong cation exchange (SCAX) resin by tube inversion in a rotating wheel for 2-5 min, and centrifuged at 2000g for 1 min, and the supernatant (designated “flow-through”) was kept for further analysis. The SCAX resin was then washed with 50 mM sodium acetate, pH 4.0, by pipetting up/down 5-10 times, and the supernatant was discarded. Polypeptides were stepwise eluted from the resin with 300-µL volumes of 0.2, 0.5, and 1.5 M KCl in 50 mM acetate buffer, pH 4.0, by pipetting up/down 5-10 times and centrifugation (2000g for 1 min). All three eluates (0.2, 0.5, 1562 Analytical Chemistry, Vol. 76, No. 6, March 15, 2004
Figure 1. Magnetic particle-assisted processing of serum peptides for MALDI-TOF MS analysis. Magnetic beads are surface-derivatized with reversed-phase ligands. The automated batch protocol consists of six discrete segments; the manual procedure is identical, except where noted. More details can be found in the Materials and Methods section. Bead suspension: Magnetic bead pellets are resuspended by pipeting up/down. Note that, in the manual mode, this is done by vortexing. Binding: A measured volume of bead suspension is transferred to a tube containing an aliquot (typically 50 µL) of serum, magnetic beads and serum are mixed by pipeting up/down, beads are pulled to the side by magnetic force, and “supernatant” is removed and discarded. Washing: Washing solution is added, beads are pulled five times from left to right and back by alternately positioning the tube(s) left/right of the magnet(s), beads are pulled to the side, washing solution is removed, and the entire procedure is repeated. Beads pull-down: At the conclusion of the washing step, beads are further resuspended by pipeting up/down, beads are pulled to the tip of the tube by magnets positioned underneath, and supernatant is carefully removed. Elution: A minimal volume of elution solvent is added to the bead pellet and mixed by pipeting up/down, and beads are pulled to the side and a fraction of the eluate transferred to another tube. MALDI sample preparation: A measured volume of premade matrix solution is added to the eluate and mixed, and 1 µL is transferred to the MALDI target.
1.5 M KCl) and the flow-through were used to carry out standard RP C8 batch separations in different tubes and were separately analyzed. Automated Sample Preparation. The manual RP batch protocol was incorporated with minimal changes into two alternative automated platforms, the MAP II (Bruker Daltonics) and Genesis Freedom 100 (Tecan, Research Triangle Park, NC) work stations. A 96-well plate cooler (Eppendorf, Westbury, NY) was added to the systems to reduce solvent evaporation during some steps. Magnetic beads were generally pulled to the side of the tubes (see Figure 1; Binding, Washing, and Elution). To this end, magnetic strips were embedded in the 96-tube holding plates, between the rows of eight. This device comes standard with the Map II system or can be constructed using epoxy-coated, neodymium-iron-boron (NdFeB) magnets (2 3/4 in./1/4 in./1/8 in.; L/W/H) from K&D Magnetics (Boca Raton, FL). To resuspend the beads in a minimal volume of elution solvent, they must first be collected at the bottom of the tubes, requiring the magnets to be positioned right underneath in a 96-well microtiter plate layout. A plate holder, containing 96 NdFeB magnetic disks, 1/4 in. in diameter and 1/4 in. thick (obtained from Forcefield; Fort Collins, CO), was therefore constructed (see Figure 1; Beads pull-down). To facilitate programming of unique protocols, a graphical user interface, based upon Bruker’s device drivers for the MAP II, was
developed in Visual Basic. The Genesis Freedom 100 system was programmed either directly via its standard software or, when individual wells needed to be accessed independently, indirectly through its work-lister capability. Once in the 96-well format, this system automates all of the liquid handling steps, including the magnetic separation via a robotic manipulating arm and spotting onto Bruker 384-spot MALDI target plates. Both systems thus modified were used to develop the automated procedure, which also included mixing of eluates with MALDI matrix and deposition on the target plates. Mass Spectrometry. Peptide profiles were analyzed with an Ultraflex MALDI-TOF mass spectrometer (Bruker, Bremen, Germany) equipped with a 337-nm nitrogen laser, a gridless ion source, delayed-extraction (DE) electronics, a high-resolution timed ion selector (TIS), and a 2-GHz digitizer. The standard protocol, except where modified for investigative and optimization purposes, was as follows. Separate spectra were obtained for two restricted mass-to-charge (m/z) ranges, corresponding to polypeptides with molecular mass of 0.8-4 kDa (“ e4kD”) and 4-15 kDa (“g4kD”) (assuming z ) 1), under specifically optimized instrument settings. Each spectrum was the result of 300 laser shots, per m/z segment per sample, delivered in three sets of 100 shots (at 20-Hz frequency) to each of three different locations on the surface of the matrix spot. This was preceded, in each location, by 20 shots at a higher laser power (“blast” step; data not acquired), determined empirically to be about twice the arbitrary setting of the “acquisition” step. The entire irradiation program was automated using the instrument’s AutoXecute function. Spectra were acquired in linear mode geometry under 20 kV (18.6 kV during DE) of ion accelerating and -1.3 kV multiplier potentials, and with TIS deflection of mass ions of g500 m/z (g4kD segment) or g2000 m/z (g4kD segment). DE was maintained for 80 (g4kD) or 330 ns (g4kD) to give appropriate time lag focusing after each laser shot. Peptide samples were typically mixed with two volumes of premade R-cyano-4-hydroxycinnamic acid (ACCA) matrix solution (Agilent, Palo Alto, CA), deposited onto the stainless steel target surfaces and allowed to dry at room temperature. Thirty femtomoles (per peptide) and 500 fmol (per protein) of commercially available calibration standards (Bruker Daltonics products no. 206195 (4kD)) were also mixed with ACCA matrix and separately deposited onto the target plates, centrally located to eight neighboring serum samples, together arrayed in a 3 × 3 pattern on a standard 384-spot plate. Spectra acquired via AutoXecute were processed using several macros that have been custom written for this task in AURA, which is the macro language built into the XMASS software (Bruker Daltonics). These macros provide automated smoothing, baseline correction, and peak designation of spectra during acquisition. After manually implemented external calibration, a macro extracts peak (i.e., m/z) lists and saves them to a file server. A custom FileMaker Pro database (Filemaker Inc., Santa Clara, CA) was used to transform the numerical data into a text file format required for subsequent statistical analysis (see below). Peak lists were imported into the database for a series of data transformations. To first create a simple binary system for initial pattern analysis, peak intensities were reduced to indicate mere
presence or absence in any of the resulting bins of the peptides observed in any particular sample. Next, the peaks were aligned across all samples within a particular set by binning within a window expanding proportionally with peptide mass (e.g., 1500 ppm). Binning is done by merging all m/z values from all samples into one long list, sorted by increasing value. The first mass is then marked as “real” and compared to the adjacent sorted masses. Any adjacent masses within a user-defined window are called “duplicate”. The process is repeated with the next larger m/z value that has yet to be marked until all the masses in the sorted list are tagged as either “real” or “duplicate”. “Duplicate” masses are then discarded. In the current application, the tolerance was either 2 Da or 1500 ppm (0.15%), depending on the experiment. Note that the assignment of the first m/z value in each bin of masses as the “real” mass is arbitrary. It is merely used as a designation for the binsany value in the bin could have been used as the “real” mass. Once the m/z values are binned, a spreadsheet is created with the results. The first column is a list of all the “real” masses surviving the binning process. The remaining columns represent the samples and whether each sample has a peak binned with the corresponding “real” mass. Statistical Data Analysis. After binning of m/z peaks across all samples of a study set, the resulting spreadsheet represents a digital readout of ∼2000 unique peptides (see Results). Commercial GeneSpring software 6.0 (Silicon Genetics, Redwood City, CA) was used to evaluate proteomic data. A virtual “experiment” was created in software to represent the masses. No normalizations were applied to the experiment since the intensities associated with each mass were reduced to only indicate presence or absence. In the parameter section of the experiment, the samples were labeled as either Glioma or Normal. In experiment’s Interpretation section, the Analysis mode was set to “log of ratio” and all measurements were used. No Cross-Gene Error model was used. Sample Names were displayed as noncontinuous parameter. Once the experiment was created, the masses were filtered by using a one-way ANOVA nonparametric test (MannWhitney U test) and no multiple test correction at p
