Differential Plasma Glycoproteome of p19ARF Skin Cancer Mouse Model Using the Corra Label-Free LC-MS Proteomics Platform
© Humana Press 2008
Received: 14 March 2008
Accepted: 11 September 2008
Published: 2 October 2008
A proof-of-concept demonstration of the use of label-free quantitative glycoproteomics for biomarker discovery workflow is presented in this paper, using a mouse model for skin cancer as an example.
Materials and Methods
Blood plasma was collected from ten control mice and ten mice having a mutation in the p19ARF gene, conferring them high propensity to develop skin cancer after carcinogen exposure. We enriched for N-glycosylated plasma proteins, ultimately generating deglycosylated forms of the tryptic peptides for liquid chromatography mass spectrometry (LC-MS) analyses. LC-MS runs for each sample were then performed with a view to identifying proteins that were differentially abundant between the two mouse populations. We then used a recently developed computational framework, Corra, to perform peak picking and alignment, and to compute the statistical significance of any observed changes in individual peptide abundances. Once determined, the most discriminating peptide features were then fragmented and identified by tandem mass spectrometry with the use of inclusion lists.
Results and Discussions
We assessed the identified proteins to see if there were sets of proteins indicative of specific biological processes that correlate with the presence of disease, and specifically cancer, according to their functional annotations. As expected for such sick animals, many of the proteins identified were related to host immune response. However, a significant number of proteins are also directly associated with processes linked to cancer development, including proteins related to the cell cycle, localization, transport, and cell death. Additional analysis of the same samples in profiling mode, and in triplicate, confirmed that replicate MS analysis of the same plasma sample generated less variation than that observed between plasma samples from different individuals, demonstrating that the reproducibility of the LC-MS platform was sufficient for this application.
These results thus show that an LC-MS-based workflow can be a useful tool for the generation of candidate proteins of interest as part of a disease biomarker discovery effort.
KeywordsSkin cancer LC-MS Label-free protein quantification Biomarker discovery Systems biology Targeted peptide sequencing Glycoproteomics Plasma
Cancer is a leading cause of mortality in the USA  and other developed countries. Years of research have revealed cancer to be a complex disease typically involving both genetic and environmental factors, which results in the molecularly heterogeneous disease that we know . Collectively, the various genetic and environmental factors combine to cause activation and inhibition of multiple cellular pathways, resulting in a range of pathophysiologies, such as angiogenesis, immune system evasion, metastasis, altered cell growth, death, and metabolism . In turn, the heterogeneous nature of cancer has presented a significant challenge for finding new biomarkers for the disease, a topic which thus continues to generate significant research interest [4–6].
There are many types of biomarkers, each fulfilling a particular function. These would include diagnostic, prognostic, predictive, and pharmacodynamic markers . However, regardless of the purpose for which a biomarker is used, it relies on our ability to measure a change in abundance for one or more molecules of interest. Thus, in any new biomarker discovery effort, analytical methods need to be able to robustly and reproducibly measure changes in biomolecules, and be able to identify and quantify with confidence the molecular species that are changing as a result of disease.
In this study, we demonstrate the feasibility of using a liquid chromatography mass spectrometry (LC-MS)-based approach to identifying protein changes in the blood, as a result of the disease, using plasma obtained a mouse model of skin cancer and controls. We also show how the use of protein interaction networks and functional analyses can further inform a biomarker discovery effort. This approach is not just valid for skin cancer, or even cancer in general, but should be equally applicable to most other diseases if suitable biospecimens are available for analysis.
Blood, perfusing the whole body, moving molecules between organs, is a rich body fluid for biomarker discovery. It is assumed that blood and plasma, where the blood cells have been removed, contains clues about the health status of most organs and tissues of the body. Because it is readily accessible, blood plasma is very attractive for biomarker discovery efforts. The main hurdle with plasma is its high degree of complexity, containing probably millions of distinct molecular species, spanning more than ten orders of magnitude in concentration . Since it is assumed that potential protein biomarkers are most likely present in the lower abundance range and whole plasma analyses overwhelm even the most powerful LC-MS system, the adoption of one of many available strategies for reduction of sample complexity is additionally required.
Sample fractionation seeks to divide one complex sample into many samples of lesser complexity and takes advantage of separation techniques that are orthogonal or complimentary to the C18 chromatography used in LC-MS. Protein separation , strong cation exchange [9, 10], and 1D  and 2D gel electrophoresis [12, 13] are among the most common fractionation methods used to date for complex samples in proteomics. Although fractionation is an effective way to perform an extensive inventory of a complex sample, such as plasma, the multiplication of samples it creates becomes a limitation for analyzing large sample populations, as one would do for biomarker discovery.
Another way to reduce complexity is to remove the proteins that are likely not relevant to the disease, such as highly abundant and ubiquitous proteins, which are generally dismissed as biomarker candidates. This is a particularly important consideration for plasma protein analyses, since albumin alone accounts for over half of the total protein content , and just the top 22 most abundant make up about 99%. Immunoaffinity subtraction [14, 15] columns have thus been used to selectively deplete these most abundant proteins from plasma samples, opening up the dynamic range for potentially detecting more interesting proteins. One problem with this approach is that some of the potential biomarkers might bind to abundant proteins and be removed either by non-specific binding or by binding to proteins that are removed.
An alternative strategy to complexity reduction involves specific enrichment for a subfraction of analytes, rather than removing some of them as above. Such approaches typically focus on taking advantage of unique chemical properties of a subfraction of the proteome to enrich for it. Examples of this approach include selective enrichment of cysteinyl-containing peptides , phosphorylated peptides , and N-glycosylated peptides . Glycosylated proteins are particularly attractive as potential candidate biomarkers that might be detectable in plasma since, by definition, they almost exclusively exist as secreted proteins or are embedded in cell surface membranes. They may thus represent a sub-proteome more likely to be detectable in the blood, either through direct secretion or via deposit from cell membranes via a process of cell shedding, leaching, or cell death. It is also known that the oncogenic process itself causes altered cell surface protein expression patterns in tumor cells [19, 20], making glycoproteins also interesting from the disease perspective, as most cell surface proteins are glycosylated. In addition, their hydrophilic sugar moieties tend to make them quite soluble, further increasing our chance of detecting them in blood. Finally, protein N-glycan structures provide a convenient chemical ‘handle’ that can readily be used for their selective enrichment.
Since this approach typically generates only one or a few peptides per protein (more than enough to identify most proteins), the significant reduction in complexity it affords results in a sample that can be directly analyzed by LC-MS without the need for additional fractionation, though this can still be an option in some workflows. This allows us to analyze only one sample per individual, making the approach ideal for population-based studies, as typically required for biomarker discovery. Another advantage of selective enrichment of N-glycopeptides comes at the data analysis stage, where the known consensus sequence for N-glycosylation, N-X-S/T (where X is any amino acid but proline and S/T is serine or threonine), serves as a useful confirmation of peptide identification and can thus be used as a filter to reduce the false discovery rate (FDR). In addition to this, the enzymatic deamidation step that is used to remove the N-glycan structure and make the peptides amenable to LC-MS results in conversion of the formerly glycosylated asparagine to aspartic acid. This generates a small mass shift for the resultant peptide, easily measurable in higher accuracy mass spectrometers. This serves as yet another confirmation of peptide identity and further provides positive identification of the N-glycosylation site on the protein, information that may turn out to also be biologically relevant down the road. Finally, with the availability of sequence databases and the existence of an N-glycosylation motif, the finite number of possible N-glycosylated peptides, along with their expected sequences, is known. This has allowed for the building of N-glycosylation databases, such as UniPep , that can be a useful resource in a candidate-based biomarker discovery workflow.
For relevant biomarker discovery research, one also needs to perform quantitative measurements on populations of samples and identify discriminating peptides for which the abundance segregates the populations into disease states. Traditional quantification methods for proteomics involve stable isotope tags of distinct masses, which enable multiplexing of samples in one LC-MS measurement [22–27]. However, there is a strict limit to the number of samples that can be multiplexed, restricting their use in population studies. Multiplexing also reduces the available dynamic range, as the amount of sample loaded into the LC-MS system is a finite quantity. Multiplexing five samples is equivalent to loading only one fifth of the original material on a per sample basis.
Traditional shotgun proteomic workflows have significant limitations for biomarker discovery. Automatic precursor selection introduces a bias toward high intensity precursor ions, leaving the lower intensity ones that frequently correspond to lower abundance peptides unidentified. Typically, the majority of the reliably detected peptides at the MS1 level do not get identified by automatic precursor selection . To concentrate our efforts on the specific and discriminating features, our approach compares LC-MS profiles first, to find changes in peptide abundance between different populations. Once those peptides are found, their m/z and retention times are placed in an inclusion list to be sequenced and identified, on reanalysis of the same samples by tandem mass spectrometry (MS/MS) .
In label-free LC-MS quantification, only one sample is analyzed at a time, which requires a highly reproducible analytical platform. To infer relative peptide abundance, peak areas from different LC-MS experiments analyzed under similar conditions are compared. The first step in label-free LC-MS quantification is to find all the peaks detected in a sample, a process called peak picking. The next step is peak alignment, where the peaks found in some samples are tentatively mapped to other samples, compensating for small changes in m/z and retention times. This type of analysis assumes that the samples have a majority of peaks in common and that only a small fraction of those will change between subjects. This approach has been formulated in various software packages [30–33]. However, it was found that most tools were designed with a particular kind of data in mind and did not perform well with data acquired with another type of mass spectrometer. Factors such as mass accuracy, noise level, and resolving power, which are specific not only to a particular instrument but to how that instrument is operated, can limit the scope of such tools. The Corra framework (M. Brusniak et al., manuscript submitted for publication) was designed to facilitate the use of those tools and combine them with statistical methods for data analysis within a common interface where we can mix and match different software tools and instruments.
The goal of this study was thus to demonstrate the applicability of a highly integrated label-free LC-MS proteomics workflow to a disease biomarker discovery effort. We used a p19ARF mouse model for skin cancer  to perform a proof-of-concept experiment to show that we were able to identify peptides and proteins that were differentially abundant between plasma samples from skin-cancer-bearing mice and normal controls. We utilized N-glycopeptide enrichment for plasma protein complexity reduction, followed by LC-MS and Corra data analysis to identify differentially abundant peptides. We then used inclusion list MS/MS to identify the peptides and the proteins they originated from. Finally, we used protein interaction and functional data to identify protein networks and molecular functions that were enriched in the regulated proteins. This confirmed that cancer-related proteins and processes were indeed detectable and quantifiable in plasma by this method, thus demonstrating its utility as part of any larger biomarker discovery project.
Materials and Methods
The p19ARF mouse model is a well-characterized in vivo model of epithelial neoplasia, as deletion of the p19ARF tumor suppressor makes them highly susceptible to tumor formation upon topical application of carcinogenic agents. The two-stage chemical protocol involves treatment of mice with a carcinogen, 7,12-dimethylbenz[a]anthracene (DMBA) followed by multiple applications of the promoting agent, 12-O-tetradecanoylphorbol-13-acetate (TPA). This treatment induces benign squamous cell papillomas, with sustained activating mutations in the H-ras oncogene. Early papillomas consist of folded epidermal or follicular hyperplasias that begin to protrude from the skin surface. In wild-type mice, a small fraction of these benign papillomas will progress over time to malignant carcinomas as further genetic mutations occur. Mice heterozygous for the tumor suppressor p19ARF were used for the skin tumor model in this study due to their reduced latency in developing papillomas and carcinomas compared to wild-type mice .
Breeder pairs of genotype p19ARF+/− X p19ARF +/+ on a NIH01a background strain were set up to generate a cohort of ten experimental and ten control mice of genotype p19ARF +/−. Each experimental mouse was paired with a control mouse from the same litter and sex and housed in the same cage during the course of the experiment. Care was taken to ensure that no systematic biases were introduced between cases and controls. All mice were maintained on a 12-h light–dark cycle and had access to autoclaved food and water ad libitum. For the experimental mice, the backs of 8-week-old male and female mice were shaved and treated with a single application of DMBA (Sigma, St. Louis, MO) 25 μg in 200 μl acetone, followed a week later by twice weekly applications of TPA (Sigma) (200 μl of 10−4 M solution in acetone) for 15 weeks. Control mice were treated with TPA alone for 15 weeks. Benign squamous cell papillomas appeared by 8 weeks post-DMBA treatment that progressed to malignant squamous cell carcinomas beginning at 24 weeks. The experimental mice had an average of five papillomas and one carcinoma. No tumors were detected on the backs of control mice. Mice were monitored daily and any abnormal health or behavior noted. Benign to malignant conversion was accompanied by a dramatic change in the appearance of the tumor and its invasion and growth into the underlying dermis. Conversion was easily quantified by visual inspection and confirmed using histological analysis of tumor sections. Carcinoma-bearing experimental mice and the matched control were killed within 1–2 weeks after first visual detection of carcinoma. Mice were euthanized by CO2 inhalation, and whole blood was collected by cardiac puncture. Plasma was purified from the whole blood by K3EDTA addition and centrifugation. Plasma was stored in conical cryovials in 100 μl aliquots and stored in a liquid nitrogen tank.
The isolation of N-linked glycopeptides from total mouse plasma samples (40 μl total from each mouse) was performed essentially as described elsewhere [35, 36]. Unless otherwise noted, all chemicals and reagents were from Sigma. Individual plasma samples were diluted with 40 μl trypsin buffer (0.1% Rapigest (Waters, Milford, MA) in 25 mM KHPO4, pH 8) and desalted using spin columns (SNS P060, The Nest Group, Southborough, MA) pre-equilibrated in trypsin buffer, eluting in a final volume of 80 μl. Proteins were denatured with the addition of 75 μl trifluoroethanol for 1 h, 60°C. Proteins were then reduced with the addition of 10 μl of 80 mM Tris (2-carboxyethyl) phosphine (Pierce, Rockford, IL) for 30 min, 60°C, followed by alkylation with 20 μl of 100 mM iodoacetamide for 30 min, room temperature, in the dark. Proteins were next proteolyzed with the addition of 550 μl trypsin buffer and 40-μl 0.5 mg/ml trypsin (Promega, Madison, WI) for 2 h, 37°C. Glycopeptides were next oxidized with the addition of 100 μl coupling buffer (1 M sodium acetate, 2.5 M NaCl, pH 5.5) and 100 μl of 100 mM NaIO4 (Sigma) for 1 h, 4°C, in the dark and then captured to a solid-phase with the addition 100 μl of a 50% slurry of hydrazide resin (Bio-Rad, Hercules, CA) pre-washed with coupling buffer, 3 h at room temperature with gentle mixing. Beads were washed 3× each with 1.5 M NaCl, H2O, and freshly made bicarbonate buffer (100 mM NH4HCO3, pH 8.3) and the N-linked glycopeptides released by the addition to the beads of 25 μl bicarbonate buffer and 3 μl protein N-glycosidase F (PNGaseF, New England Biolabs, Ipswich, MA) and incubation overnight at 37°C with gentle mixing. Released peptides were then recovered with two additional washes with 100 μl bicarbonate buffer and desalted and cleaned up on Sep-Pak C18 cartridges (Waters) eluting with 50% acetonitrile, 0.1% trifluoroacetic acid, and finally evaporated to dryness under vacuum in clean glass vials.
Peptides were separated on an 1100 Series HPLC system (Agilent, Santa Clara, CA) equipped with a nanoflow pump, operating at a flow rate of 1 μl/min. Mobile phase A was a 0.1% formic acid in water and mobile phase B was 0.1% formic acid, 5% water, and 95% acetonitrile. A binary gradient from 5% to 35% B was used to separate the peptides on a monolithic C18 10 cm long × 100 μm inner diameter column (Merck KGaA, Germany). A self-packed integraFrit column (New Objective, Woburn, MA) with a bed of Magic C18 5 μm particles (Michrom Bioressources, Auburn, CA) 2 cm × 100 μm was used as a pre-column. Sample volumes of 5 μl were injected by the autosampler, and every batch of replicates was randomized before injection.
Mass analysis was performed on a MicrOTOF electrospray time-of-flight mass spectrometer (Bruker Daltonics, Billerica, MA) with a mass accuracy of 5 ppm and a resolving power of 9,000 or better. The mass scale was calibrated using glu1-fibrinopeptide B (Sigma), and mass spectra were acquired at one spectra/s over the range of 300–1,600 m/z. High mass accuracy for the differential mapping on the MicrOTOF was maintained by automatic instrument recalibration between every sample. This was achieved by injecting 320 fmol of glu1-fibrinopeptide B (Sigma) with a 15-min gradient and increasing the cone voltage to 220 V, inducing in-source CID. The fragments were used by the visual basic script to recalibrate on-the-fly the mass spectrometer, insuring the same mass accuracy from the first to the last sample. This measure also had the benefit of preventing carryover between samples and provided a way to monitor the sensitivity and reproducibility of the system during large batch analyses.
Corra Statistical Analysis
Targeted Sequencing of Discriminatory Peptides by LC-MS/MS
The top 300 discriminatory peptides of charge 2 and 3 were made into an inclusion list for sequencing on an LTQ-FT mass spectrometer (ThermoFisher, San Jose, CA), as described elsewhere . A mass tolerance of 25 ppm was specified, and ion accumulation time was set at 500 ms for both IT-MS and FT-MS scans. The scan rate for the FT-MS scans was set to 0.89 s over the range of 300–1,600 at a resolving power of 100,000. MS/MS scan rate was set at 0.2 s. The LTQ-FT connected to an 1100 Series HPLC system, but used a self-packed, 15-cm capillary column with a 150 μm inner diameter packed with a bed of Magic C18 5 μm particles (Michrom Bioresource), without a pre-column. The flow rate was 1.2 μl/min, and the gradient was the same as described for the differential mapping. The identified peptides were mapped back to the aligned features, keeping the same mass accuracy and retention time tolerances as were used for the inclusion list.
Protein Identification and Function
The proteins from the inclusion lists were identified using the Trans Proteomic Pipeline  using Sequest version 27 and the mouse IPI database version 3.5. The searches were performed using a mass tolerance of 0.1 Da on the precursor ion and of 3 Da on the fragment ions. Variable modifications were included for oxidized methionine (16.0 Da) and conversion of asparagine to aspartic acid due to the deglycosylation (0.98 Da) were added to the search, as well as a static modification for cysteine carbamidomethylation (57.02 Da). Trypsin was selected as the proteolytic enzyme. Peptide and Protein FDRs are automatically calculated in a dataset-dependant manner by the PeptideProphet and ProteinProphet components of the Trans Proteomic Pipeline .
Because biological function is more commonly annotated to genes than proteins, the first step of the protein data analysis was to find the Entrez GeneIDs that coded for the identified proteins. These GeneIDs allowed us to identify enriched biological processes and molecular functions, protein–protein interactions, and protein associations. The data mining was performed using Cytoscape  and Gaggle , through the Protein Function Exploration WorkBench  using the following data sources: Entrez GeneID identifications were obtained primarily from the mouse database version 3.5 (ftp://ftp.ebi.ac.uk) and then supplemented (for IPIs that had no Entrez GeneID) by searching the IPI protein sequence against the NCBI ‘nt’ database, using tblastn (http://blast.wustl.edu/) dynamically translating in all six reading frames, selecting only high scoring, complete matches with Entrez GeneIDs. Enriched Gene Ontology biological processes, and molecular functions were calculated using the Bioconductor GOstats package (http://bioconductor.org) on annotations provided by the Affymetrix Mouse Genome 430 2.0 Array annotation data. Protein–protein interactions were inferred from HPRD, the Human Protein Reference Database (http://www.hprd.org) via homology, using NCBI’s Homologene (http://www.ncbi.nlm.nih.gov). Other gene associations were provided by EMBL String (http://string.embl.de).
Results and Discussion
LC-MS Analysis of N-glycosylated Tryptic Peptides by LC-MS
For this experiment, we enriched formerly N-glycosylated tryptic peptides from the blood plasma of ten cancer-bearing mice and ten control mice. All 20 samples were initially analyzed by LC-MS on an ESI-TOF system. Sample run order was randomized for all 20 samples. This randomization was to both average out any cross contamination that might occur due to sample carryover and to ensure that no statistical bias for run order was introduced for subsequent data analysis. To further minimize sample carryover, a standard injection of 320 fmol of the calibration standard peptide glu1-fibrinopeptide B was inserted between each of the 20 LC-MS injections. An advantage of doing this between each LC-MS run was that this calibration standard could be used to recalibrate the machine between each run in an automated fashion. This gave us great confidence in, and knowledge of, the mass resolution and accuracy from run to run. This, in turn, gave us increased confidence in the LC-MS data alignments, which is done on the basis of accurate mass and LC retention time.
Feature Selection and Clustering
All 20 LC-MS maps were processed with Corra. The number of detected features per LC-MS injection varied from 1,096 to 1,407, with a median of 1,354 features on the ESI-TOF platform. To increase our chances of identifying cancer-specific regulated proteins, we only made use of features that aligned across, minimally, half of the LC-MS runs. This approach would, in theory, find any differentially abundant feature observed in all of the cancer mice and not the controls and reduce the processing time to a manageable amount, as it grows exponentially with the number of features to compare. In using this filter, we would expect to loose some genuine cancer marker candidates. However, this was done to help control the FDR. While we would expect that some N-glycoproteins that might be upregulated and detectable in cancer and below the level of detection in controls, they may not be detected in all cancer mouse LC-MS runs. This likelihood increases as their relative abundance decreases. However, for the purposes of this study, it was important to have a low false-discovery rate in terms of the alignments, so that we could have high confidence in the downstream protein function analyses. The next step in the data analysis was to superimpose observed signal intensities for the aligned peptide features to determine which showed differential abundance between case and control mice. This information was then used to see whether the aligned LC-MS data alone could be used to distinguish between the two disease populations in a blind analysis.
Targeted Sequencing by LC-MS/MS for Peptide and Protein Identification
With the aligned peptide feature signal intensities for the cancer versus normal mice, the next step was to identify the peptides and proteins of most interest, according to apparent differential abundance, via inclusion list-based targeted LC-MS/MS. To do this, we selected 300 aligned features (based on raking according to Corra-generated p values) to form the inclusion list. In turn, this was used for LC-MS/MS sequencing using and LTQ-FT MS system and was performed on six randomly chosen plasma N-glycopeptide isolates, three each from the cancer and normal sample sets.
The mass spectrometer was set to monitor selected ion traces of all the targeted features throughout the chromatographic gradient. Thus there were more identified peptides than targeted features as the mass (i.e., m/z) of a feature can correspond to multiple peptides, especially for a complex sample such as isolates from plasma, within the resolving power of the LTQ portion of the mass spectrometer. Corra then determined the correct assignment of identified peptides to a particular feature, based not just on observed mass but also on expected retention times. The MS/MS spectra obtained from the inclusion lists were then searched against a sequence database so that the corresponding peptides sequences could be mapped back to their respective LC-MS features. In this way, we identified 80 peptides out of the 300 differentially abundant features with a PeptideProphet score of ≥0.9. In addition, we identified another eight glycopeptides that had a score lower than 0.9 but were manually validated to be correct (shown in Supplemental Figure S1). The PeptideProphet-calculated FDR for a cut-off of ≥0.9 for this data set was ∼2%.
Variability and Reproducibility of the Analytical Platform
For an LC-MS platform such as this to be useful for studies such as biomarker discovery, it is important that the reproducibility of the experimental platform as a whole is not so great as to confound subsequent data analyses. We thus performed some additional analyses to look at the reproducibility of the platform used in this study.
As was discussed in the “Materials and Methods” above, the use of an internal MS calibration standard in the wash cycles between analytical runs effectively removed variation in the accuracy of the MS measurements themselves. Much more variation, however, could be expected to come from the LC-MS system. We had previously determined that the normalization process and alignment tools of Corra adequately control for inherent variation in global signal intensity and LC retentions times for the LC systems used in this study (Brusniak et al., manuscript submitted for publication). However, since we were also mapping back subsequent MS/MS identifications onto the original, aligned LC-MS data, we looked to see how well these data correlated.
It is generally assumed that the largest form of variation in a LC-MS based disease biomarker study is likely to be between the individuals (patients, mice, etc.) being used for the study. While this variation is expected to be high in human studies, transgenic mice, as used in this study, might generate less variation due to their common genetic background and the use of littermates (i.e., siblings) as controls. To be able to accurately identify differences in the LC-MS profiles of the individuals in a given study, it is important that the observed variation between the individuals is less than the variation observed for the analytical platform as a whole. To investigate whether this was the case for the experimental platform being used for this study, we thus performed additional LC-MS profiling of the 20 samples, however, in triplicate, this time (60 runs in all) to better examine the variation introduced into the data via the experimental platform.
Biological Inference Via Protein Network and Functional Analyses
Gene name, description, IPI number, log2 ratio, PeptideProphet, and ProteinProphet probabilities for the differentially abundant glycoproteins that were identified by targeted MS/MS analysis
Complement component 9
Complement component factor h
Complement component factor i
Immunoglobulin heavy chain 1a (serum IgG2a)
Immunoglobulin heavy constant gamma 1 (G1m marker)
Inter-alpha trypsin inhibitor, heavy chain 3
Inter alpha-trypsin inhibitor, heavy chain 4
Kallikrein B, plasma 1
Lecithin cholesterol acyltransferase
Leukemia inhibitory factor receptor
Similar to Ig heavy chain V region 102 precursor
Leucine-rich alpha-2-glycoprotein 1
Pregnancy zone protein
Serine (or cysteine) peptidase inhibitor, clade A (alpha-1 antiproteinase, antitrypsin), member 10
Serine (or cysteine) peptidase inhibitor, clade A, member 6
Serine (or cysteine) peptidase inhibitor, clade D, member 1
Serine protease inhibitor 1–6
Another way to add information to the network is to look for enrichment of biological process from the Gene Ontology (GO). The GO terms are a controlled vocabulary that describe gene and gene product attributes according to their molecular functions, cellular components, or biological process. The p values, probabilities that a certain term is over-represented in a gene list, are used as a measure of likelihood of these processes having been activated in the mouse cancer model studied. Filtering for low p values will highlight the processes that are unlikely to appear within a group of randomly selected genes or proteins. The yellow rectangles represent those GO biological processes that were prevalent in this data set. As it is shown in Fig. 5, the major differences between the two populations are the result of the cancer-bearing mice being seriously sick due to the advanced state of the cancer. Hence, we observed proteins whose function is related to the host response to disease, such as inflammatory response, regulation of multicellular organism processes, and complement activation. For example, we observed proteins such as C9, Cfi, Cfh, and C1qa, which are complement factors related to innate immunity, a general defense mechanism against pathogens and infections. Likewise, the increased abundance of proteins involved in blood coagulation, such as Klkb1, Fga, Apoh, Serpind1, and Serpinc1, is likely an indication of wounding if we view the morphology of the carcinoma as akin to an open wound. The functional enrichment for these pathways confirmed our ability to detect significant changes in protein abundances caused by disease. Considering the advanced state of the cancer in this study, it was not unexpected that it would generate a very strong host response and that Corra would detect those proteins in our data.
Another protein cluster can be seen that includes Lcat, Hpx, Serpina6, Apob, which are clustered together around the transport and localization GO categories. A post-translational protein modification enrichment region has been identified featuring where Anapc7 has been identified. Anapc7 is regulated by Trp53 , which is a tumor repressor protein. Proteins related to cell death and cell cycle include complement component 9 (C9) and Ighg1. C9 protein is a membrane attack protein and can form large pores that will lyse a cell. Ighg1 appears to be more abundant in the cancer samples, and it has been shown to be produced by cancer cells . Hyaluronan metabolic process was also represented by the differentially abundant Itih3. Hyaluronan contributes significantly to cell proliferation and migration and may also be involved in the progression of some malignant tumors [42, 43]. Using a different mouse skin cancer model but the same glycopeptide enrichment technique, Zhang et al.  previously found carboxipeptidase N, hemopexin, and complement component factor h to be differentially regulated.
Taken together, while the proteins identified in this study and discussed in this paper are insufficient to formulate a strong cancer hypothesis, they can be a starting point for a more targeted type of analysis such as high sensitivity multiple reaction monitoring experiments [45, 46]. Additionally, the inferred proteins from the protein–protein interaction network would also make prime candidates for follow-up analysis. It is important to note that proteins that were not identified by MS/MS but present in this network were either not differentially abundant or were present below the detection limit of the technique.
This study demonstrated that it is possible to start from a population of complex plasma samples and perform label-free LC-MS quantitative proteomics to identify a list of differentially abundant features that segregate two or more sample populations and link those features to proteins present in serum. We also showed that the changes detected do reflect actual biological processes that are related to the presence of disease, such as host response, cell cycle, cell death, cell adhesion, and DNA damage.
We demonstrated that a modern LC-MS platform has sufficient reproducibility and can perform label-free protein quantification of small sample populations, which is challenging for most other techniques used for biomarker discovery. We also confirmed that run-to-run variations were smaller than individual variation, an important prerequisite for this mode of analysis. It was possible to map the retention times from one instrument to another to perform targeted peptide sequencing. Simpler and affordable LC-ESI-TOF platforms can thus be used to analyze the bulk of the data and keep the smaller number of sequencing experiments on the more costly tandem mass spectrometer.
Measuring the difference in protein abundance is the first step of biomarker discovery. The proteins identified were clearly not all specific markers for skin cancer. Many were related to the host immune responses that are expected during tumor growth. However, some proteins seem to be involved in biological processes that are critical to cancer. To detect early disease biomarkers, it would be interesting to repeat the experiment with earlier stages of cancer. This data may still be of value to inform follow-up targeted MRM experiments for the cancer-related proteins that were identified, as well as those that were not identified, but inferred from the building of the protein interaction network.
Finally, this methodology has no definite limit in the number of samples that can be compared, making it suitable for large-scale studies. The principle was demonstrated with a small number of cancer mouse model samples, but can also be directly applied to other organisms and other diseases.
This work was supported with federal funds from the National Cancer Institute, National Institutes of Health, under contract No. N01-CO-12400 (to J.W.), the National Heart, Lung, and Blood Institute, National Institutes of Health, under contract N01-HV-28179 (to R.A.), the National Cancer Institute by Grants R21-CA-114852 (to H.Z.), and by Grant No. 31000-107-67 by the Swiss National Science Foundation (to R.A.).
- Jemal A, Siegel R, Ward E, Murray T, et al. Cancer statistics, 2007. CA Cancer J Clin 2007;57:43–66.PubMedView ArticleGoogle Scholar
- vant Veer LJ, Bernards R. Enabling personalized cancer medicine through analysis of gene-expression patterns. Nature 2008;452:564–70.View ArticleGoogle Scholar
- Hanahan D, Weinberg RA. The hallmarks of cancer. Cell 2000;100:57–70.PubMedView ArticleGoogle Scholar
- Srivastava S, Srivastava RG. Proteomics in the forefront of cancer biomarker discovery. J Proteome Res 2005;4:1098–103.PubMedView ArticleGoogle Scholar
- Anderson KS, LaBaer J. The sentinel within: exploiting the immune system for cancer biomarkers. J Proteome Res 2005;4:1123–33.PubMedPubMed CentralView ArticleGoogle Scholar
- Sawyers CL. The cancer biomarker problem. Nature 2008;452:548–52.PubMedView ArticleGoogle Scholar
- Anderson NL, Anderson NG. The human plasma proteome: history, character, and diagnostic prospects. Mol Cell Proteomics 2002;1:845–67.PubMedView ArticleGoogle Scholar
- Wang H, Hanash S. Intact-protein based sample preparation strategies for proteome analysis in combination with mass spectrometry. Mass Spectrom Rev 2005;24:413–26.PubMedView ArticleGoogle Scholar
- Fortier M-H, Bonneil E, Goodley P, Thibault P. Integrated microfluidic device for mass spectrometry-based proteomics and its application to biomarker discovery programs. Anal Chem 2005;77:1631–40.PubMedView ArticleGoogle Scholar
- Ranish JA, Yi EC, Leslie DM, Purvine SO, et al. The study of macromolecular complexes by quantitative proteomics. Nat Genet 2003;33:349–55.PubMedView ArticleGoogle Scholar
- Chelius D, Zhang T, Wang G, Shen R-F. Global protein identification and quantification technology using two-dimensional liquid chromatography nanospray mass spectrometry. Anal Chem 2003;75:6648–57.View ArticleGoogle Scholar
- Stuart LM, Boulais J, Charriere GM, Hennessy EJ, et al. A systems biology analysis of the Drosophila phagosome. Nature 2007;445:95–101.PubMedView ArticleGoogle Scholar
- Gilchrist A, Au CE, Hiding J, Bell AW, et al. Quantitative proteomics analysis of the secretory pathway. Cell 2006;127:1265–81.PubMedView ArticleGoogle Scholar
- Sitnikov D, Chan D, Thibaudeau E, Pinard M, Hunter JM. Protein depletion from blood plasma using a volatile buffer. J Chromatogr B Analyt Technol Biomed Life Sci 2006;832:41–6.PubMedView ArticleGoogle Scholar
- Liu T, Qian WJ, Mottaz HM, Gritsenko MA, et al. Evaluation of multi-protein immunoaffinity subtraction for plasma proteomics and candidate biomarker discovery using mass spectrometry. Mol Cell Proteomics 2006;5:2167–74.PubMedPubMed CentralView ArticleGoogle Scholar
- Liu T, Qian WJ, Strittmatter EF, Camp DG, et al. High-throughput comparative proteome analysis using a quantitative cysteinyl-peptide enrichment technology. Anal Chem 2004;76:5345–53.PubMedView ArticleGoogle Scholar
- Trinidad JC, Specht CG, Thalhammer A, Schoepfer R, Burlingame AL. Comprehensive identification of phosphorylation sites in postsynaptic density preparations. Mol Cell Proteomics 2006;5:914–22.PubMedView ArticleGoogle Scholar
- Zhang H, Li XJ, Martin DB, Aebersold R. Identification and quantification of N-linked glycoproteins using hydrazide chemistry, stable isotope labeling and mass spectrometry. Nat Biotechnol 2003;21:660–6.PubMedView ArticleGoogle Scholar
- Kakugawa Y, Wada T, Yamaguchi K, Yamanami H, et al. Up-regulation of plasma membrane-associated ganglioside sialidase (Neu3) in human colon cancer and its involvement in apoptosis suppression. Proc Natl Acad Sci U S A 2002;99:10718–23.PubMedPubMed CentralView ArticleGoogle Scholar
- Orntoft TF, Vestergaard EM. Clinical aspects of altered glycosylation of glycoproteins in cancer. Electrophoresis 1999;20:362–71.PubMedView ArticleGoogle Scholar
- Zhang H, Loriaux P, Eng J, Campbell D, et al. UniPep—a database for human N-linked glycosites: a resource for biomarker discovery. Genome Biol 2006;7:R73.PubMedPubMed CentralView ArticleGoogle Scholar
- Gygi SP, Rist B, Gerber SA, Turecek F, et al. Quantitative analysis of complex protein mixtures using isotope-coded affinity tags. Nat Biotechnol 1999;17:994–9.PubMedView ArticleGoogle Scholar
- Ong S-E, Blagoev B, Kratchmarova I, Kristensen DB, et al. Stable isotope labeling by amino acids in cell culture, SILAC, as a simple and accurate approach to expression proteomics. Mol Cell Proteomics 2002;1:376–86.PubMedView ArticleGoogle Scholar
- Ross PL, Huang YN, Marchese JN, Williamson B, et al. Multiplexed protein quantitation in Saccharomyces cerevisiae using amine-reactive isobaric tagging reagents. Mol Cell Proteomics 2004;2:1154–69.View ArticleGoogle Scholar
- Conrads TP, Alving K, Veenstra TD, Belov ME, et al. Quantitative analysis of bacterial and mammalian proteomes using a combination of cysteine affinity tags and 15N metabolic labeling. Anal Chem 2001;73:2132–9.PubMedView ArticleGoogle Scholar
- Veenstra TD, Martinovic S, Anderson GA, Pasa-Tolic L, Smith RD. Proteome analysis using selective incorporation of isotopically labeled amino acids. J Am Soc Mass Spectrom 2000;11:78–82.PubMedView ArticleGoogle Scholar
- Zhou H, Ranish JA, Watts JD. Quantitative proteome analysis by solid-phase isotope tagging and mass spectrometry. Nat Biotechnol 2002;19:512–5.View ArticleGoogle Scholar
- Finney GL, Blackler AR, Hoopmann MR, Canterbury JD, et al. Label-free comparative analysis of proteomics mixtures using chromatographic alignment of high-resolution muLC-MS data. Anal Chem 2008;80:961–71.PubMedView ArticleGoogle Scholar
- Schmidt A, Gehlenborg N, Bodenmiller B, Mueller L, et al. An integrated, directed mass spectrometric approach for in-depth characterization of complex peptide mixtures. Mol Cell Proteomics. 2008. doi:10.1074.Google Scholar
- Li X-J, Yi EC, Kemp CJ, Zhang H, Aebersold R. A software suite for the generation and comparison of peptide arrays from sets of data collected by liquid chromatography-mass spectrometry. Mol Cell Proteomics 2005;4:1328–40.PubMedView ArticleGoogle Scholar
- Bellew M, Coram M, Fitzgibbon M, Igra M, et al. A suite of algorithms for the comprehensive analysis of complex protein mixtures using high-resolution LC-MS. Bioinformatics 2006;22:1902–9.PubMedView ArticleGoogle Scholar
- Jaffe JD, Mani DR, Leptos KC, Church GM, et al. PEPPeR: a platform for experimental proteomic pattern recognition. Mol Cell Proteomics 2006;5:1927–41.PubMedView ArticleGoogle Scholar
- Mueller L, Rinner O, Schmidt A, Letarte S, et al. Superhirn—a novel tool for high resolution LC-MS based peptide/protein profiling. Proteomics 2007;19:3470–80.View ArticleGoogle Scholar
- Kelly-Spratt KS, Gurley KE, Yasui Y, Kemp CJ. p19Arf suppresses growth, progression, and metastasis of Hras-driven carcinomas through p53-dependent and -independent pathways. PLoS Biol 2004;2:E242.PubMedPubMed CentralView ArticleGoogle Scholar
- Tian Y, Zhou Y, Elliott S, Aebersold R, Zhang H. Solid-phase extraction of N-linked glycopeptides. Nat Protoc 2007;2:334–9.PubMedPubMed CentralView ArticleGoogle Scholar
- Zhou Y, Aebersold R, Zhang H. Isolation of N-linked glycopeptides from plasma. Anal Chem 2007;79:5826–37.PubMedView ArticleGoogle Scholar
- Keller A, Eng J, Zhang N, Li XJ, Aebersold R. A uniform proteomics MS/MS analysis platform utilizing open XML file formats. Mol Syst Biol 2005;1:2005 0017.PubMedPubMed CentralView ArticleGoogle Scholar
- Shannon P, Markiel A, Ozier O, Baliga NS, et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res 2003;13:2498–504.PubMedPubMed CentralView ArticleGoogle Scholar
- Shannon PT, Reiss DJ, Bonneau R, Baliga NS. The gaggle: an open-source software system for integrating bioinformatics software and data sources. BMC Bioinformatics 2006;7:176.PubMedPubMed CentralView ArticleGoogle Scholar
- Ramos H, Shannon P, Aebersold R. The protein information and property explorer: an easy-to-use, rich-client web application for the management and functional analysis of proteomic data. Bioinformatics 2008;24:2110–1.PubMedPubMed CentralView ArticleGoogle Scholar
- Hearnes JM, Mays DJ, Schavolt KL, Tang L, et al. Chromatin immunoprecipitation-based screen to identify functional genomic binding sites for sequence-specific transactivators. Mol Cell Biol 2005;25:10148–58.PubMedPubMed CentralView ArticleGoogle Scholar
- Chen Z, Gu J. Immunoglobulin G expression in carcinomas and cancer cell lines. FASEB J 2007;21:2931–8.PubMedView ArticleGoogle Scholar
- Heldin P, Karousou E, Bernert B, Porsch H, et al. Importance of hyaluronan-CD44 interactions in inflammation and tumorigenesis. Connect Tissue Res 2008;49:215–8.PubMedView ArticleGoogle Scholar
- Zhang H, Yi EC, Li X-J, Mallick P, et al. High throughput quantitative analysis of serum proteins using glycopeptide capture and liquid chromatography mass spectrometry. Mol Cell Proteomics 2005;4:144–55.PubMedView ArticleGoogle Scholar
- Anderson L, Hunter CL. Quantitative mass spectrometric multiple reaction monitoring assays for major plasma proteins. Mol Cell Proteomics 2006;5:573–88.PubMedView ArticleGoogle Scholar
- Stahl-Zeng J, Lange V, Ossola R, Eckhardt K, et al. High sensitivity detection of plasma proteins by multiple reaction monitoring of N-glycosites. Mol Cell Proteomics 2007;6:1809–17.PubMedView ArticleGoogle Scholar