Protein separation

Although the final dimension of separation for MS-based proteomics is generally a reversed-phase separation at the peptide level, upstream separations can use a number of different properties for fractionation of complex samples at the peptide or protein level. Many options exist for protein-level separation, each of which can utilise a different physical property of proteins to obtain varying but complementary first-dimension separations. Cation exchange (Ref. Reference Shen11) and anion exchange (Ref. Reference Zhou12), for example, offer good separation orthogonality in fractionating proteomic samples before liquid chromatography (LC)-MS/MS. Two different technologies, however, have emerged recently and joined the mainstream: chromatofocusing (CF) (Refs Reference Sluyterman and Elgersma13, Reference Sluyterman and Wijdenes14, Reference Schlautman15), a variant of ion-exchange chromatography that separates proteins based on pH; and isoelectric focusing (IEF) (Ref. Reference Michel16), which separates proteins based on their isoelectric point (pI), as is done in the first dimension of 2D gel electrophoresis. These methods are easily automatable and provide powerful protein-level separation and useful information of the physical properties of the proteins, while keeping the analytes soluble and compatible with proteomic experiments, and are therefore widely used (Ref. Reference Fang, Robinson and Foster17).

Another commonly used method for protein-level sample fractionation prior to MS is the ‘GeLC’ approach. It harnesses the well-established ability and available equipment for running gels, by separating a complex sample by molecular weight at the protein level in a single 1D SDS-PAGE gel lane, and using that lane as the first dimension in a multidimensional separation. After staining, the entire lane is excised, cut into bands and each band is treated as a fraction of the same sample. After the proteins in these bands are enzymatically digested, each band's peptide mixture can be analysed on an LC-MS/MS instrument and the results combined. The benefits of gel-based protein-level first-dimension separation are threefold: gels are often a good way of making biological samples compatible with MS analysis (e.g. by removal of detergents), method development is not needed as SDS-PAGE is a well-established technique, and the number of identifications obtained per experiment is currently second to none. In fact, of the aforementioned protein-level separation techniques, the GeLC approach has been found to provide the highest number of confident protein or peptide identifications, although the alternative approach of immobilised pH gradient (IPG)-based IEF has the benefit of slightly higher sample recovery over the GeLC separation (Ref. Reference Fang, Robinson and Foster17). A recently developed fractionation method termed GELFrEE (gel-eluted liquid fraction entrapment electrophoresis) also separates proteins based on their size using a gel column; however, in contrast to the GeLC method, the proteins are eluted and collected in the liquid phase (Ref. Reference Tran and Doucette18). The researcher needs to be aware of these performance differences in first-dimension separations when deciding on the priorities for a given experiment.

Peptide separation

The development of powerful techniques and chemistries for separation at the peptide level has led many MS-based workflows to forgo the aforementioned protein-level separations entirely. Many options exist for the first-dimension separation, which have varying degrees of orthogonality: for example, reversed-phase chromatography, which resolves issues with solvent compatibility; or strong ion exchange, different forms of which can provide good sample complementarity. One of the most common first-dimension separations for large-scale proteomic experiments is peptide-level strong cation exchange (SCX) chromatography. There is no ‘best’ answer for separation; despite each of the techniques being optimal for particular sample types, they will all provide complementary results.

First-dimension separations can be performed off-line with a fraction collector, although when they are performed in-line with a reversed-phase column as the second dimension prior to MS, in a workflow called multidimensional protein information technology (MudPIT) (Refs Reference Washburn, Wolters and Yates19, Reference Lin, Alpert and Yates20), the experiment is capable of significant proteome coverage (approximately 60%) in one – albeit very long – experiment (Ref. Reference Bendall21). This type of workflow is described thoroughly in a published protocol (Ref. Reference Delahunty and Yates22). Reproducing an experiment of this type also provides 60% of the proteome, with a large number of the peptides sequenced being species that were not sequenced in the first experiment. It is estimated that it would take five MudPIT experiments performed in this way to achieve near-complete sequence coverage, a phenomenon attributed to MS/MS peptide-sampling rates. As with any MS-based analysis of complex samples, the limitations of this method are time and instrument duty cycle – issues that should be considered when designing experiments and choosing which proteomic approach to use. These limitations can be attenuated by conducting biological and technical repeats and maximising separation and fractionation prior to MS analysis. Also, the development of dynamic exclusion lists to avoid run-to-run resequencing of peptides has recently increased the number of extracellular proteins identified in repeat analyses of the human embryonic stem cell secretome by an order of magnitude (Ref. Reference Bendall21). Expanding the MudPIT workflow to include a third dimension of separation has also been shown to work well (Ref. Reference Wei23), and a study of the proteome of the serum of patients with sepsis utilised immunodepletion of abundant serum proteins followed by a 3D peptide-level separation, allowing the identification of low-abundance serum proteins while identifying ten potential serum biomarkers for sepsis (Ref. Reference Shen24). Although this type of technique shifts the limitation of the method towards separation time and away from the duty cycle of the instrument, the deployment of fast, ultrahigh-pressure liquid chromatography (UPLC) (Ref. Reference Shen25; http://www.waters.com/waters/nav.htm?locale=en_US&cid=10136122) in many laboratories is now proving that these methods are more powerful than ever.

Biochemical fractionation methods

The protein- or peptide-separation techniques described above allow in-depth analysis of a complex sample. However, biochemical fractionation procedures, which add an additional dimension of separation, can lead to even deeper resolution as the separation methods described previously can be performed on a less complex sample. This can be especially important in highly complex samples, such as human plasma or serum, which have a high dynamic range spanning at least ten orders of magnitude. These samples contain a small number of highly abundant proteins, whose signals can dominate MS-based analysis. Depletion of these proteins can be highly advantageous in allowing access to lower abundance species, including potential disease biomarkers, and there are many tried and tested depletion strategies available (Refs Reference Pernemalm, Lewensohn and Lehtio26, Reference Dwivedi27). For cellular studies, spatial information (e.g. which proteins are found in which organelles or which proteins interact with each other) can be extremely important for understanding a complex system, and can be obtained by applying either subcellular fractionation or affinity enrichment techniques.

Isotopic labelling techniques

Mass spectrometers are not inherently quantitative. Differences in ionisation, transmission and detection efficiency dictate that the intensity of a signal from a particular molecule is a relative measure of its abundance, but not an absolute measure. For this reason, all quantitative proteomics, even ‘absolute’ quantitation is relative – relative to an internal standard’ (Ref. Reference Elliott45). MS-based proteomic quantitation was therefore not thrust into the mainstream until 1999, when isotope-coded affinity tags (ICATs) were introduced (Ref. Reference Gygi46). These tags were the first widely available method to quantify the relative concentrations of peptides or proteins in a sample, by way of an isotope-coded chemical modifier. Briefly, each of two samples is treated with either one of a ‘light’ or ‘heavy’ chemical reagent that binds specifically to cysteine residues. The light and heavy tags are chemically identical, except for isotopic differences. The two samples are then mixed and digested, and the tagged peptides are enriched using avidin or streptavidin chromatography against the biotin moiety embedded in the tag. On performing MS analysis on these enriched samples, the chemically identical species from the two samples will coelute from a column and ionise with identical efficiency; however, the peptide that is modified with the ‘light’ form of the reagent will appear at a known lower mass in the spectrum than the ‘heavy’ tagged equivalent peptide from the other sample. One can then directly compare the peak areas of the two chemically identical coeluting peptides and thereby obtain a relative measure of their abundance. Relative quantitation, performed through isotope-coding methods similar to this, is the best way to obtain information about quantitative differences in protein expression, especially from the complex samples usual in proteomics (Fig. 2).

Figure 2.

Workflows for implementation of stable isotope labelling via metabolic (SILAC) and chemical (iTRAQ or TMT) methods. (a) SILAC incorporates isotopes early in the sample preparation procedure, maximising accuracy and reproducibility, and is generally used only for quantitation of samples from cells that can be cultured. Isotope incorporation into the amino acids themselves means peptides to be compared have different masses; therefore quantification occurs from the MS scan. (b) Chemical isotope-coded tags are applied later in the workflow, but can be applied to any biologically derived sample. Isobaric chemical tags (iTRAQ, TMT) add equivalent masses to the peptides in the sample, but produce specific marker ions upon fragmentation, allowing quantification from the MS/MS scan. Abbreviations: iTRAQ, isobaric tags for relative and absolute quantitation; MS, mass spectrometry; MS/MS, tandem mass spectrometry; SILAC, stable isotope labelling of amino acids in cell culture; TMT, tandem mass tag.

One issue with the ICAT method described above, however, is its dependence on the modification of cysteine residues, which account for only 1.42% of the amino acids in a sample (Ref. Reference Gevaert47). Many peptides, and even whole proteins, do not contain a cysteine, and are therefore unquantifiable by means of ICAT. This problem was resolved in 2004 with the introduction of the isobaric tags for relative and absolute quantitation (iTRAQ) label (Ref. Reference Ross48) (Applied Biosystems; http://www.appliedbiosystems.com). With this tag, initially four, and now up to eight, samples can be compared together, using labels with identical mass shifts. This is achieved through the differential placement of the stable isotopes onto ‘balance’ and ‘reporter’ pieces of the tag, which are separated by a labile bond. Each of the different tag ‘flavours’ adds the same overall mass to a peptide, bound through the balance group to the primary amines on lysine side chains and the N-termini of peptides. On mixing of the samples, unlike ICAT-labelled samples, tagged peptides will appear as one signal in a normal MS scan. Only upon fragmentation does the labile bond holding the iTRAQ modification together dissociate, forming an intense marker ion from the reporter group, each of which will be specific for one of the samples to be compared. This behaviour allows for high specificity (marker ion intensity can only be from the peptide currently being fragmented) and high sensitivity because the isobaric nature of the tag dictates that the intensities of the signals from all samples are additive for initial detection in the MS scan, and subsequent sequencing. This technique has also been successfully used in the tandem mass tag (TMT) strategy from Thermo Scientific (http://www.piercenet.com).

These types of chemical labelling strategies are a good choice when probing the proteome of human cells or tissues that cannot or should not be cultured. Unfortunately, as the combination of the samples to be compared occurs late in this workflow, there is a chance of systematic errors during sample handling (Fig. 2). Metabolic labelling techniques, in which cells are grown in isotopically labelled media and compared with those grown in normal media, have been shown to be the most accurate proteomic quantitation method mainly due to the ability to combine samples very early in the procedure (e.g. immediately after lysis), thus minimising errors involved with differential sample handling in the subsequent isolation and purification steps (Ref. Reference Blagoev38). Unfortunately, SILAC (Ref. Reference Ong49) can only be used on cells cultured in vitro, as controlling the isotopes available to a biological sample can be problematic. Although SILAC quantitation on a whole mouse has been successfully performed (Ref. Reference Kruger50), this type of experiment is prohibitively expensive and time consuming for most research projects and species types. Recently, however, a technique that uses a combination of five SILAC-labelled cell lines, pooled together, has been introduced as a ‘physical proteome database’, which was then compared with a nonlabelled carcinoma tissue sample, allowing the SILAC quantitation of uncultured human tissue cells (Ref. Reference Geiger51). This new technique allowed quantitative comparison of lobular and ductal tumours, revealing significant differences, with very low coefficients of variance, in the expression of focal adhesion and glycolytic proteins, in a clinically relevant human tissue sample.

Label-free quantitation

Given that the intensity of the signal in a mass spectrometer is innately a proxy of the abundance of the species in the sample, label-free quantitation approaches have recently entered the mainstream because of their apparent ease, simplicity and cost savings. One of the numerous label-free quantitation approaches is ‘spectral counting’, in which the MS/MS spectra collected for a given species are counted and compared with those collected for the same species in a different sample. This technique uses the assumption that unbiased, intensity-based precursor ion selection leads to intense ions being selected for sequencing more frequently. The number of MS/MS spectra collected for a given analyte would therefore be a proxy of its intensity, and therefore its abundance. Like all label-free quantitation methods, systematic errors in the analysis, such as signal suppression, detector saturation and differential sample loading, occur. These errors need to be minimised and accounted for by performing many replicates, normalising the data, and statistical validation (Refs Reference Elliott45, Reference Li and Roxas52). Although this approach has been shown to be adequate for quantitation of high-abundance components of a mixture (Ref. Reference Old53), isotopic labelling techniques, which correct for these systematic errors and also allow for sample multiplexing, are still the method of choice for quantitative proteomics. The commonly used quantitative proteomic methods, along with their strengths and weaknesses, are shown in Table 1.

Table 1.

MS quantitation techniques

Abbreviations: AQUA, absolute quantitation; ICAT, isotope-coded affinity tags; ICPL, isotope-coded protein labelling; iTRAQ, isobaric tags for relative and absolute quantitation; MS/MS, tandem mass spectrometry; N15, ¹⁵N isotope; SILAC, stable isotope labelling of amino acids in cell culture; SRM, selected reaction monitoring; TMT, tandem mass tag.

Experimental design

Targeting the most appropriate peptides and fragment ions for the protein of interest is key to a successful SRM experiment; therefore some prior experimental knowledge is required. This translates into knowing the mass to charge ratio (m/z) of an abundant, consistently produced (or ‘proteotypic’) peptide (Ref. Reference Mallick58) as well as the m/z of one of its fragment ions that is generated with high intensity. These ‘transitions’ (specific precursor–fragment ion pairs) allow targeted analysis of a particular peptide in a complex mixture. There are many guidelines that can aid in the selection of appropriate transitions, based on prior experimentation, physicochemical parameters and in silico predictions (Ref. Reference Walsh59). These take into account factors that are too numerous to describe in detail here, but are outlined in several recent reviews (Refs Reference Lange54, Reference Han and Higgs60). There are multiple software options available to aid the design and optimisation of these transitions (Ref. Reference Cham Mead, Bianco and Bessant61). Many are reviewed elsewhere (Ref. Reference Yocum and Chinnaiyan62), with a selection listed in Table 2. Although this design stage can take considerable time, once transitions are established, they can be used indefinitely for experiments studying the protein of interest.

Table 2.

Examples of selected reaction monitoring (SRM) resources

Abbreviations: MIDAS, MRM-initiated detection and sequencing workflow; MRM, multiple reaction monitoring.

SRM and quantitation

The SRM approach can be used to quantitate proteins. Relative quantitation can be conducted simply by comparing the absolute peak area of the individual samples (label-free quantitation), although it is difficult to obtain precise measurements because of differences in ionisation efficiency, analyte composition and chromatography. SRM experiments can also be combined with many of the standard isotope labels used in quantitative proteomic experiments, including ICAT, SILAC, ICPL and iTRAQ. Additionally, several methods that aid greatly in speeding up the assay development aspect of SRM have emerged, including databases such as MRMAtlas (Ref. Reference Picotti63) and a method of crude synthetic peptide library production, which allow the rapid generation of validated SRM assays for whole proteomes (Ref. Reference Picotti56). These approaches have been pioneered using the yeast proteome, but the development of databases and resources such as this for clinically relevant tissues could help thrust SRM-based quantitation firmly into the clinical arena.

Applications of SRM

The advantages of SRM experiments have led to an almost exponential increase in the number of studies using this approach in recent years (Ref. Reference Yocum and Chinnaiyan62), and SRM has now been applied to many diverse biological questions, from the quantitation of the biomarker C-reactive protein (CRP) in the serum of patients with rheumatoid arthritis (Ref. Reference Kuhn64) to the absolute quantitation of the human liver alcohol dehydrogenase ADH1C1 isoenzyme (Ref. Reference Janecki65) and pyruvate kinase M2 (PKM2), a potential endometrial cancer marker (Ref. Reference DeSouza66). The approach has been shown to be extremely powerful, both in confirmation of potential biomarkers and in discovery of novel biomarkers, as seen in a recent study that integrated high-throughput and SRM-based approaches to explore breast cancer in a mouse model (Ref. Reference Whiteaker67). SRM is also a promising tool to study splice variants that are putative biomarkers for cancer cells. For example, using the breast cancer mouse model mentioned above, 216 of 608 splice variants were found only in tumour cells and SRM is ideally suited to study these (Ref. Reference Menon and Omenn68). Using SRM coupled with stable isotope dilution MS (SIDMRM-MS), quantitative, multiplexed assays were developed for the analysis of six proteins clinically relevant to cardiac injury (Ref. Reference Keshishian69). These widely applicable assays were conducted using plasma samples, with proteins of interest spanning four orders of magnitude. These were quantitated using a few signature peptides from each target protein, with limits of quantitation ranging between 2 and 15 ng/ml. Similarly, low ng/ml sensitivity quantitation of the prostate-specific antigen biomarker was achieved using LC-MS/MS SRM, from 100 µl of serum, demonstrating good correlation with ELISA measures (Ref. Reference Fortin70). Even greater sensitivity was achieved in a recent study that captured and enriched peptides with antipeptide antibodies and then used SRM-based analysis to quantitate aberrant GlcNAcylated tissue inhibitor of metalloproteinase 1 (TIMP1), a protein implicated in colorectal cancer (Ref. Reference Ahn71). Following enrichment and digestion of glycoproteins from patients' serum, SISCAPA (stable isotope standards and capture by antipeptide antibodies) (Ref. Reference Anderson72) and SRM-MS permitted highly sensitive quantitation of TIMP1 at attomolar concentrations. Automation and multiplexing of this approach shows great potential for analysing large numbers of biomarkers with sufficient sensitivity, reproducibility and precision for clinical applications (Ref. Reference Whiteaker73).

Still, proven reproducibility is essential for SRM assays to move beyond basic research and become a force in clinical or diagnostic assay development and application. A recent, multisite review demonstrated high reproducibility across different laboratories using different instrument platforms (Ref. Reference Addona74). Another recent review questioned whether SRM-MS will replace antibody-based testing in the validation of biomarkers (Ref. Reference Issaq and Veenstra75). SRM-MS has several advantages over antibody assays for biomarker validation: SRM has exquisite sensitivity, with no crossreactivity and less specificity issues than are often associated with antibody assays; SRM is ‘reagent independent’; SRM can be used for any MS-observable ion, making it generally cheaper than antibody assays (Ref. Reference Latterich, Abramovitz and Leyland-Jones76); and these assays are quantitative and easily multiplexed. This is a key point, as reality dictates that having a single biomarker for a disease is unlikely; panels of biomarkers are the more likely future of disease diagnostics, and SRM technologies are very well placed to study these. Issues still remain however, particularly regarding assay throughput and precision, which have not been thoroughly tested and currently do not meet the US Food and Drug Administration (FDA) requirements for routine clinical tests (Ref. Reference Issaq and Veenstra75). Another issue is operator familiarity, as these assays are only beginning to enter the mainstream, and it will take time for users to become comfortable with applying these new techniques. However, the hurdles facing the use of SRM in biomarker validation are slowly being overcome, and although antibody assays will still be used for biomarker validation, increasingly we can expect to see the application of SRM assays.

Protein identification

Identification of peptides and their corresponding proteins is generally conducted using search algorithms that correlate experimental MS/MS spectra to theoretically derived spectra created from known peptide sequences. There are several different search engines available, which differ in their approaches to identifying peptide sequences. The most common search algorithms include Sequest (Ref. Reference Eng, Mccormack and Yates78), Mascot (Ref. Reference Perkins79) and X!Tandem (Ref. Reference Craig and Beavis80). It is worth noting that these search engines are complementary to some extent, so it is often useful to use at least two different algorithms to analyse MS/MS data to increase confidence and sensitivity, and there are tools available to aid this (Ref. Reference Price81). Validation of the peptide and protein identifications is necessary and is often conducted by determining false discovery rates using decoy databases and other statistical methods (Ref. Reference Li, Pizarro and Grosser82). For de novo sequencing of proteins, an approach that is useful in the analysis of PTMs or of organisms whose genome has not been sequenced, there are several software options available, including PEAKS (Ref. Reference Ma83).

Analysis of quantitative proteomic data

Tools for the analysis of quantitative data from proteomic experiments are continuously emerging and being refined. The Trans-Proteomic Pipeline (Ref. Reference Keller84) is a collection of integrated MS/MS analysis tools, including XRPESS and ASAPRatio that are used for the relative quantitation of isotopically labelled peptides and proteins. MaxQuant is a recently developed software suite for the analysis and quantitation of SILAC experiments (Ref. Reference Cox and Mann85). Similarly, Mascot Distiller, from Matrix Science, determines quantitation based on the relative intensities of extracted ion chromatograms for precursors (http://www.matrixscience.com). This approach can be used for ‘label-free’ approaches, or with any chemistry that creates a precursor mass shift, for example ¹⁸O, AQUA, ICAT, ICPL, metabolic labelling and SILAC. ProteinPilot, from Applied Biosystems (http://www.appliedbiosystems.com), provides protein identification and quantitation of SILAC- and iTRAQ-based labels. For label-free approaches, there are many open-source and commercial software packages available, which are discussed in a recent review (Ref. Reference Wang86). It is worth noting that for the analysis of quantitative proteomic data, no standard procedure has been developed that is broadly applicable to all experiment types. As is evident, many software tools exist, and the user still needs to understand what the software is doing in order to be able to critically analyse the results.

Data-mining approaches

With the rapid growth in large-scale proteomic experiments comes the generation of longer and longer lists of proteins. However, the sound biological interpretation of these data lags behind (Ref. Reference Malik77). There are now several analytical strategies and tools available to extract biologically relevant information (e.g. regarding protein–protein interactions, signalling pathways and biological networks) from these large proteomic datasets. These ‘data-mining’ approaches have the potential to contribute to a deeper understanding of biological systems, but need to be applied and interpreted correctly. One of the most powerful tools available, and often the first tool used to conduct analysis on a large dataset, is Gene Ontology (GO) (Ref. Reference Ashburner87). This is a controlled vocabulary that is used to standardise the way in which proteins are described across different species and databases. The consistency in terminology that this ontology provides makes it an invaluable resource for both experimentalists and bioinformaticians. GO annotation of a large MS dataset can be used to determine whether there is any enrichment or depletion for a particular GO category, or can be used to compare two different datasets.

Pathway and network analysis

Another useful approach is pathway analysis, which explores proteomic data in terms of biological pathways, based on known physical and functional interactions between proteins that are present. It is estimated that there are around 300 biochemical pathway analysis tools currently available (Ref. Reference Malik77), with the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome representing the largest databases. Many of the pathway analysis tools are freely available, but there are also some commercially available tools – for example, Ingenuity Pathways Analysis from Ingenuity Systems, and GeneGo from GeneGo Inc. With so many pathway analysis options to choose from, Pathguide (http://www.pathguide.org), which contains information on about 317 biological pathway tools, is an invaluable resource to help guide users in selecting the most appropriate resource to use (Ref. Reference Bader, Cary and Sander88). Pathguide also covers tools that model network and functional interaction information, which takes the data beyond pathway analysis and groups proteins based on participation in larger, multiprotein assemblies. For visualisation of molecular networks, Cytoscape is a useful open-source platform, which also allows integration of genetic and other information (Refs Reference Shannon89, Reference Cline90). There are now several meta-databases for interaction information, including STRING (Refs Reference Snel91, Reference Jensen92), which generates interaction networks by incorporating data from many curated databases, as well as predicted interactions and pathway information. Data can be input to STRING as protein lists, and it has a user-friendly interface. MiMI, from the National Institute for Integrative Biomedical Informatics, merges data from numerous interaction databases as well as other sources and also has a Cytoscape plug-in to allow easy visualisation of networks (Ref. Reference Tarcea93). For all interaction databases, which can have high error rates, care needs to be taken when interpreting information and the source of the interaction information should be checked manually if possible.

Meta-data analysis and data integration

One of the key challenges currently facing researchers is the integration of all these available data. There are several tools available for meta-data analysis of proteomic data, including the database for annotation, visualisation and integrated discovery (DAVID), from the National Institute of Allergy and Infectious Diseases (NIAID), which provides a comprehensive set of functional annotation tools for investigators to understand the biological meaning behind large lists of genes. Other meta-tools include PANTHER (protein analysis through evolutionary relationships, http://www.pantherdb.org), which was designed to classify proteins (and their genes) in order to facilitate high-throughput analysis, and Babelomics, which is a suite of interconnected tools used to functionally annotate genome-scale experiments. Conceptgen, from the National Institute for Integrative Biomedical Informatics, is a web-based tool designed to explore networks of relationships between biological concepts (Ref. Reference Sartor94). The Global Proteome Machine (http://www.thegpm.org), a search engine and database for MS/MS data, links the protein identifications directly to annotation resources such as GO and KEGG within the same platform, allowing efficient examination of the GOs and pathways under- or over-represented in a particular dataset.