|Home | About | Journals | Submit | Contact Us | Français|
The ability to conduct validated analyses of biomarkers is critically important in order to establish the sensitivity and selectivity of the biomarker in identifying a particular disease. The use of stable-isotope dilution (SID) methodology in combination with LC–MS/MS provides the highest possible analytical specificity for quantitative determinations. This methodology is now widely used in the discovery and validation of putative exposure and disease biomarkers. This review will describe the application of SID LC–MS methodology for the analysis of small-molecule and protein biomarkers. It will also discuss potential future directions for the use of this methodology for rigorous biomarker analysis.
The development of clinically useful biomarkers is a two-stage process. The initial bioanalytical validation stage uses control samples in appropriate biofluids to establish analytical limits of quantification and determine errors introduced by sample handling that are typically used in conventional bioanalytical validation studies . This stage is followed by clinical validation using patient samples to determine the assay’s sensitivity for detecting the disease and specificity in discriminating the particular disease. Without rigorous bioanalytical validation, time-consuming and expensive clinical validation studies cannot succeed. Conventional validation strategies require quality control samples to be conducted in the biofluid being examined. Therefore, bioanalytical validation of endogenous bio markers is often very challenging, due to the target analyte’s presence in the biofluid of interest. Related to this hurdle, though more difficult to resolve, is the fundamental uncertainty as to whether the endogenous signal is actually the analyte of interest and not an interfering substance with similar physicochemical properties. To best distinguish the target analyte from chemical-background peaks, assays have to be conducted with the maximal specificity and sensitivity that is possible using stable isotope analogs as internal standards. GC– and LC–MS/MS are the two most widely used instrument platforms to employ stable-isotope dilution (SID) methodology. LC–MS/MS is more applicable to the analysis of a wider range of biomarkers than GC–MS/MS and is also inherently easier to use for rigorous validation. Hence, the present review will focus on the use of SID LC–MS/MS for biomarker analysis. However, the underlying principles relating to the specificity of SID methodology are relevant to both techniques. It should also be noted that SID has been used for the elemental analysis of biomarkers, using such techniques as inductively coupled plasma–MS, though this is beyond the scope of this review and will not be discussed.
Quantitative studies often require the most sensitive means of detection possible. The MS platform and analysis mode best suited for a particular analysis needs to be determined empirically and will depend on the molecule and matrix involved. However, in general, a triple quadrupole (TQ) operated in the multiple reaction monitoring (MRM)-MS (also known as selective reactive monitoring [SRM], but referred to herein as MRM) mode will show exceptional levels of sensitivity and selectivity when coupled to LC. In this mode of operation, a precursor ion is preselected and resolved in Q1 of the TQ, fragmented by collision-induced dissociation (CID) in Q2 and the resultant product ion is analyzed in Q3. Under optimal operating conditions, the precursor to product ion ‘reaction’ is monitored many times per second, resulting in extremely reproducible chromatographic peak shapes and intensity. In this way, a stable (heavy) isotope-labeled standard is used in SID LC–MRM-MS to establish the presence of an endogenous analyte using both the LC retention time and MS/MS mass selection of the TQ platform. This level of specificity cannot be attained with any other bioanalytical technique employed for biomarker analysis.
An authentic stable isotope-labeled analog of a compound is identical to the endogenous molecule except for mass. The term SID most often refers to the use of a stable isotope-labeled internal standard spiked into a sample at a known concentration. The response ratio between the analyte and labeled compound can then be interpolated onto a standard curve to calculate the absolute amount of analyte in the unknown sample. Variations of this technique are also used, most extensively in proteomics, where chemically and metabolically labeled proteins and peptides are often used at unknown concentrations for relative quantification purposes. In either case, the internal standard offers a means to verify the presence of the analyte and normalize experimental variables, such as sample storage and matrix suppression. In our article, the SID technique will be reviewed under this broader understanding of its application in LC–MS experiments. The use of structural analogs as internal standards, rather than authentic isotope-labeled analogs, is undesirable because they will have different retention times and ionization properties compared with the analyte of interest. Therefore, differential ionization can occur between an analyte and a structural analog in the source of the mass spectrometer. This difference arises in part from suppression of ionization by constituents present in the biofluid that is being analyzed and can lead to significant imprecision during quantitative analyses . Unfortunately, suppression effects vary with chromatographic retention time and with biofluid samples from different individuals. It is therefore impossible to standardize the amount of suppression occurring within any particular sample. The ideal control offered by an authentic isotope-labeled internal standard is not always possible because, for many biomarkers, only deuterated and structural analogs are available. Deuterated forms of a compound are not perfect internal standards, since there is a small but significant separation of the deuterium analog internal standards and their corresponding endogenous protium forms during chromatography. This slight difference in chromatography can result in differential suppression or enhancement of ionization and affect the quality of the analytical data. Structural analogs are even less representative of the endogenous compound since, in addition to differences in LC retention time, the structural analog can show different absorptive loss. Selective binding to active sites on glassware or other surfaces can occur during extraction and chromatography, leading to significant analyte loss. Whereas a structural analog might not account for this loss, an isotope-labeled internal standard has identical physicochemical properties and is therefore lost at exactly the same rate as the endogenous analyte. Due to this feature of stable-isotope analogs, they may act as a carrier, preventing the loss of trace amounts of analyte during extraction and analysis . This is an often- overlooked benefit offered by the isotope-labeled internal standard. Finally, variability introduced during compound isolation can be fully controlled by an authentic isotope-labeled standard. In many cases, compound enrichment is required to improve the analytical performance, as we found with amyloid β-peptides  and DNA adducts , which required immunoaffinity purification.
The search for small-molecule biomarkers that are predictive of cardiovascular disease, neurodegenerative diseases and cancers has been particularly intense [5–8]. By linking biomarker discovery and validation with known pathways of endogenous metabolism it is possible to rationally approach the development of a biomarker for a specific disease. This is exemplified in the case of breast cancer research. Based on figures from 2002 to 2006, 87.3% of breast cancer diagnoses were made in women over the age of 45. Hence, aging is an important risk factor and postmenopausal women are at much higher breast cancer risk than premenopausal women. Paradoxically, postmenopausal women lack the capacity to synthesize ovarian estrogens, yet are at increased risk through exposure to elevated levels of circulating estrogens . Rationally, this increased risk could result from a genotoxic effect of estrogen metabolites , from the direct proliferative effect of estrogens  or from a combination of both mechanisms. A substantial effort over the last decade to develop methods for analyzing serum estrogen biomarkers that are predictive of breast cancer risk has therefore been underway . Such a biomarker could designate women at a high risk of breast cancer, and chemoprevention strategies could then be implemented [12,13].
Small-molecule biomarkers are derived from modifications to glutathione (GSH), DNA and cellular metabolites, such as lipids and folates (FIGURE 1). The most common biofluids that are used to monitor small-molecule disease biomarkers include, blood, serum, urine, cerebrospinal fluid, red blood cells (RBCs) and nasal and lung lavages . Modifications to major biomolecules can occur through the direct and indirect actions of reactive intermediates and/or reactive oxygen species (ROS), which in turn can arise from the metabolism of both exogenous and endogenous chemicals during inflammation, radiation and the metabolism of hormones, drugs and environmental toxins . Cellular GSH provides a major defense against ROS and reactive intermediates through GSH peroxidase-mediated reduction of ROS and GSH-S-transferase (GST)-mediated conversion of reactive intermediates to GSH adducts (FIGURE 1) . As GSH is consumed there is a concomitant increase in glutathione disulfide (GSSG) concentrations . In addition, increased activity of oxygenases such as cyclooxygenase (COX) and lipoxygenase occurs, which contributes to the oxidative metabolism of xenobiotics and endogenous compounds [17,18]. These processes can ultimately lead to DNA, protein and lipid damage and disrupt cellular metabolite formation (such as homocysteine and folates). Both the ‘damaged’ molecules as well as the aberrant metabolite concentrations may therefore provide potential biomarkers of diseases (FIGURE 1). For example, the direct action of ROS can lead to the formation of 8-oxo-7,8-dihydro-2′-deoxyguanosine (8-oxodGuo)  and the F2-isoprostanes (IsoPs) . Indirectly, ROS-mediated lipid peroxidation can lead to the formation of DNA adducts such as heptanone-etheno-2′-deoxyguanosine (HεdGuo) (FIGURE 1) . Since DNA adducts are normally excised from DNA by nucleotide excision repair (NER) as the corresponding 2′-deoxynucleotides or by base excision repair (BER) as the corresponding purine or pyrimidine bases, the resulting DNA adducts excreted in the urine can serve as disease biomarkers. In rare cases when the DNA adducts are not excised from DNA, this can lead to a mutation, such as a G to T transversion. Such mutations may ultimately result in aberrant mRNA and protein expression (FIGURE 1), which in turn can lead to increased cellular proliferation and tumorigenesis. Following this rationale, the analysis of urinary DNA adducts has been employed to provide biomarkers of cancer risk .
The analysis of small-molecule biomarkers using LC–MS/MS-based methodology most often involves the use of reversed-phase chromatography coupled to a TQ mass spectrometer or ion trap, utilizing an atmospheric pressure ionization (API) source such as electrospray ionization (ESI), nanospray or atmospheric pressure chemical ionization (APCI). In some instances, such as our targeted lipidomic profiling of various eicosanoids and other metabolites of lipid peroxidation, chiral LC separations were accomplished using normal phase, rather than reversed-phase chromatography . Normal-phase solvents do not allow efficient ionization so this is performed by electron capture atmospheric pressure chemical ionization (ECAPCI)–MS . Ion traps provide high sensitivity in full scan mode and can be used to acquire data-dependant MS/MS information on multiple ions at extremely fast cycle speeds. Additionally, when coupled to a high-resolving and high-mass-accuracy trap, as in a hybrid instrument, this data can provide outstanding compound identifications during small-molecule biomarker discovery. However, TQs are used most often during biomarker validation due to their improved sensitivity for quantification of targeted analytes. Although modern TQ instruments can conduct MRM scans very quickly, operation under optimal scan conditions for biomarker sensitivity and reproducibility does limit the number of compounds monitored.
For brevity, we will focus here on estrogens as a case-in-point of the usefulness and analytical challenges associated with measuring endogenous steroids by SID MS/MS. Estrogen biosynthesis in the breast tissue of post-menopausal women arises from circulating C-19 endogens formed in the adrenal cortex. Estradiol is formed directly from androstenedione or from reduction of estrone by 17β-hydroxysteroid dehydrogenases (HSDs)-1, -7 and -12 and/or aldo-keto reductase 1C3 (FIGURE 2). Conversely, estradiol can be reduced back to estrone by 17-HSDs 2 and 14. Both estrone and estradiol are converted to catechol estrogens by cytochromes P450 (CYPs) and then to 2- and 4-methoxy-estrogens by cate-chol-O-methyltransferases (COMTs). The ratio of 4-methoxy-estrogens to 2-methoxy-estrogens provides an indirect measurement of the catechol estrogens 4-hydroxy-estradriol/2-hydroxy-estradiol and 2-hydroxy-estrone/4-hydroxy-estrone (FIGURE 2). This is potentially important since the catechols are considered to be genotoxic estrogens while 2-methoxy-estrogens are considered to be antiproliferative and protective against mammary carcinogenesis [22,23]. 16α-hydroxy-estrone, which is formed by CYP-mediated metabolism of estrone, is also considered to be procarcinogenic (FIGURE 2) . The ability to routinely conduct very high sensitivity analyses of 4-methoxy-estrogens and 2-methoxy-estrogens together with estrone, 16α-hydroxy-estrone and estradiol in serum will make it possible to develop and evaluate new and improved models of breast cancer risk [13,25]. The development of such models would impact significantly on the implementation of chemoprevention strategies for women newly identified to be in a high breast cancer risk category. Previous studies have shown that this would significantly improve breast cancer prevention [12,13]. Therefore, the ability to routinely analyze plasma and serum estrogens with very high sensitivity could potentially save a large number of women from this devastating disease .
For reliable measurements of multiple estrogens in plasma or serum, it is necessary to employ SID methodology in combination with LC–MRM-MS. Unfortunately, endogenous estrogens are not effectively ionized using conventional ESI or APCI methodology. This has restricted the use of these ionization techniques to samples with higher concentrations of plasma and serum estrogens [26–30]. Fortunately, the ionization characteristics can be readily enhanced by derivatization of the 3-hydroxy-group on the A-ring of the estrogen. Three derivatization approaches for plasma and serum estrogens have been reported. The first approach, which we have pioneered, involves the preparation of an electron capturing pentafluorobenzyl derivative of the estrogen coupled with the use of electron capture atmospheric pressure chemical ionization (ECAPCI)-MS for analysis (FIGURE 3) . Higashi et al. have also explored the utility of ECAPCI-MS for estrogen analysis by using different electron capturing derivatives . We showed that it was possible to quantify estrogens in the low picogram per milliliter range in plasma using LC–ECAPCI-MS (TABLE 1) . This second approach uses conventional derivatization coupled with LC–ESI-MS. This approach is exemplified by studies that have employed the dansyl (D) derivative [26,29,33] and others that made use of picolinoyl (P) or pyridyl-3-sulfonyl (PS) derivatives [34,35] (FIGURE 3). The third approach involves the preparation of preionized (quaternized) derivatives, so that ionization is not required in the ESI source of the mass spectrometer. This has involved preparation of N-methyl-2-pyridyl (NMP), 1–2,4-dinitro-5-fluorphenyl-4,4,-dimethylpiperazinyl (MPPZ) or N-methylnicotinyl (NMN) groups attached to the 3-hydroxy moiety of the estrogen moiety (FIGURE 3) [36–38]. The three derivatization strategies described above make it possible to quantify plasma and serum estrogens by SID LC–MS/ MS with limits of quantification (LOQs) in the high femtogram per milliliter to low picogram per milliliter range (TABLE 1) [32–35,38]. Additionally, until recently, only deuterated analogs were available for use as internal estrogen standards. Fortunately, the recent availability of [13C6]-estrogen analogs from Cambridge Isotope Laboratories (Andover, MA, USA) makes it possible to use internal standards that have identical chromatographic retention times as their corresponding analytes but a mass difference of 6 Da. The availability of methodology based on these novel ionization techniques and [13C]-labeled internal standards will greatly facilitate future studies to rigorously establish the precise levels of individual estrogens that are present in the plasma and serum of postmenopausal women.
The ESI process requires ionization to occur in solution followed by desolvation of the resulting protonated molecules in the source of the mass spectrometer. It is therefore difficult to achieve complete ionization of all analyte molecules. This contrasts with preionized derivatives, which are already completely ionized (FIGURE 3). The NMN preionized derivative for the highly sensitive analysis of plasma and serum estrogens has been employed by Adamec et al. (TABLE 1) . Our laboratory has also explored this approach using the Girard T (GT) derivative, which has been previously used to analyze keto steroids [39,40]. Extremely high sensitivity can be obtained with the preionized estrone GT derivative (FIGURE 3) when analyzed using a modern TQ mass spectrometer and nanoflow chromatography. For example, using [13C6]-estrone as the internal standard, we have demonstrated a limit of detection of 0.6 fg (2.2 amol) of estrone on column as its GT derivative . This suggests that it will be possible to conduct LC–MS/MS assays on multiple estrogen metabolites in serum and plasma an order of magnitude lower than is currently achievable using LC–ECAPCI-MS or LC–ESI-MS methodology (TABLE 1). The ease with which the preionized derivatization strategy can be implemented will make it possible to readily introduce highly sensitive methodology in laboratories that are currently employing LC–MS/MS methodology. We anticipate that the use of preionized estrogen derivatives will also help conserve important plasma and serum samples as it will be possible to conduct high sensitivity analyses using limited sample volumes. This will allow the use of existing banked plasma and serum samples without significantly depleting the amounts that are available. In this way, multiple studies may share the same samples, which may be critical to our understanding of the multivariate factors that cause an increase in breast cancer risk.
There are eight classes of lipids: fatty acyls, glycerolipids, glycerophospholipids, sphingolipids, sterol lipids, prenol lipids, saccharolipids and polyketides . Lipid structures are extremely diverse, which results in their wide range of polarity, although most lipids are sparingly soluble in water at neutral pH. Classes 1–4 have similar fatty acid and polyunsatu-rated fatty acid substituents at their sn-1 and sn-2 positions. Polyunsaturated fatty acids are susceptible to oxidation through both nonenzymatic actions of ROS and by enzymatic processes such as the action of COXs, lipoxygenases and CYPs (FIGURE 4). Lipid oxidation, the mechanisms by which this process can take place and the numerous potential biomarkers that are formed have generated a substantial amount of biomarker research [19,42,43]. This burgeoning interest is due, at least in part, to the importance of lipids in biological systems and to the negative effects of the lipid oxidation on human health . SID LC–MS/MS-based methodology has had a significant impact on the analysis of five of the lipid classes – fatty acyls, glycerolipids, glycerophospholipids, sphingolipids and sterol lipids .
Fatty acyls represent one of the most important classes of bioactive lipids and they have been extensively investigated using MS methodology. In early studies, GC–MS was widely used, but its applicability was limited due to the thermal stability and volatility of the molecules. Therefore, the availability of API-based methodology [44,45] has had a dramatic impact on the field . Fatty acyls form protonated molecules [M + H]+ or negatively charged molecules arising from the loss of a proton [M − H]− under positive or negative ESI conditions, respectively. Fatty acyls are ionized more efficiently under negative ESI conditions so this mode is usually employed [47,48]. Using a combination of ESI and LC–MS/MS, it is now possible to analyze extremely complex mixtures of fatty acyl biomarkers derived from diverse biological samples [49–51]. LC–ECAPCI-MS analysis of fatty acyls, which have been derivatized as pentafluorobenzyl esters, is two orders of magnitude more sensitive when compared with negative ESI of underivatized fatty acyls . The LC–ECAPCI-MS methodology also allows the chiral lipid peroxidation products such as hydroxyeicosatetraenoic acids to be resolved (FIGURE 5) , which makes it possible to determine whether the fatty acyls are derived from nonenzymatic or enzymatic pathways (FIGURE 4) [20,53].
Prostaglandins (PGs) are COX-mediated bioactive metabolites derived from arachidonic acid (FIGURE 4). As with other arachidonic acid metabolites, they form carboxylate anions under ESI and ECAPCI conditions. The MS fragmentation pathways resulting from CID of PGs is complicated, due to the formation of ‘second-generation’ metabolites . However, these fragmentation pathways make it possible to employ specific product ions for selective LC–MRM-MS analyses . PG metabolites have been widely used as biomarkers to study the effects of COX inhibitors. Many of these studies were performed using older SID GC–MS-based methodology . isoPs are a class of arachidonic acid metabolites that are produced by the action of ROS on intact lipids [56,57]. As they are generated nonenzymatically, the isoPs are produced as a mixture of isomers, derived from their multiple chiral centers. There are four classes of isoPs that undergo class-specific fragmentation when subjected to CID, making them good candidates for SID LC–MRM-MS analysis . Using SID GC–MS methodology, they were rigorously validated as biomarkers of oxidative stress and have been widely used as disease biomarkers . Unfortunately, currently many of the isoPs cannot be quantified in biological samples because appropriate heavy isotope-labeled standards are not available. In the near future, however, we anticipate that these standards will be prepared through the autoxidation of [13C20]-arachidonic acid. This will then open up the possibility of conducting the analysis of multiple isoPs using high-sensitivity SID LC–MRM-MS.
Folate/homocysteine metabolism is required for the methylation of many important intracellular substrates including DNA, proteins and lipids, as well for the generation of thymidylate and purines required for RNA and DNA synthesis (FIGURE 6) [60–63]. These, important cellular processes require different intracellular folate derivatives (FIGURE 6) . Biomarker studies have revealed that low folate status is associated with elevated levels of circulating homocysteine (hyperhomocysteinemia) and a phenotype characterized by low RBC and serum folate, together with high homocysteine . This phenotype has been implicated in diverse diseases ranging from neural-tube defects, such as spina bifida [65,66], to aging-related conditions, such as cardiovascular disease  and colorectal cancer . The potentially deleterious effects of hyperhomocysteinemia are a consequence of inadequate levels of the methyl donor 5-methyltetrahydrofolate (5-MTHF) (FIGURE 6) [64,68]. Folate/homocysteine metabolism also modulates GSH biosynthesis through the cystathionine/cysteine pathway, which is in turn is crucial for controlling intracellular redox status (FIGURE 6) .
Over the past 10 years, several functional polymorphisms in enzymes involved in folate/homocysteine metabolism have been described . The functional polymorphism with the most readily observed impact on phenotype is the C to T transition at nucleotide 677 (677C→T) of the MTHFR gene, which results in a change in amino acid residue from Ala→Val at position 222, located at the bottom of the (βα)8 barrel of the catalytic domain of the enzyme . The 677T allele encodes an enzyme that is ‘thermolabile’ and less efficient at generating the 5-MTHF that is needed for both homocysteine remethylation and the generation of S-adenosyl methionine for methylation reactions (FIGURE 6). It is well established that MTHFR 677TT homozygotes with low folate status are at a greatly increased risk of being hyperhomocysteinemic . In RBCs of individuals with this genotype, 5-MTHF comprises only 60% of total RBC folate, whereas this form predominates in the RBCs of their 677CC peers . Furthermore, the MTHFR C677T genotype is the primary determinant of nonmethylfolate accumulation in RBCs . The homozygous MTHFR 677TT genotype confers a significantly increased risk of many of the conditions with which a low folate, high homocysteine phenotype has been associated. However, the excess individual risk of developing such conditions in relation to their prevalence is insufficient to warrant genetic testing and counseling. Therefore, we established SID LC–MS/MS methodology to determine the degree of variation in the RBC folate phenotypes between and within the three MTHFR 677C→T genotypes.
Individual folates are present as folylpoly-glutamates, which prevents their release from the RBCs. Therefore, it was first necessary to convert them to the corresponding monoglutamates prior to analysis. This was accomplished using an individual’s own plasma pteroylpolyglutamate hydrolase by simply lysing the RBCs in a whole blood sample. A SID LC–MRM-MS assay was then developed and used to assess the relationship between different MTHFR 677C>T genotypes and RBC folates in thirty genotyped subjects. This showed that there were four different folate biomarker phenotypes that were differentially distributed between the MTHFR 677C>T genotype classes . The RBC folate biomarker assay is now being employed for extensive phenotyping studies in human populations with defined genotypes [75,76]. Availability of this new biomarker assay will facilitate the identification subgroups of individuals with genotype/phenotype profiles that confer excess risk of defects known to be associated with dysfunction in folate/homocysteine metabolism. Such genotype/phenotype-based risk estimation may in the future be used in the conduct of clinical studies and to develop predictive and diagnostic screening protocols.
Intracellular GSH provides one of the major defenses against oxidative stress in mammalian cells. During oxidative stress, reduced GSH (γ-glutamylcysteinylglycine) is converted to GSSG. GSH is the most abundant small molecule thiol in cells with concentrations in the millimolar range . By contrast, concentrations of GSSG in mammalian cells are usually two orders of magnitude lower . The high abundance of GSH and low abundance of GSSG helps to maintain cells under a reducing environment and prevents oxidative damage to cellular macromolecules. Hydrogen peroxide and lipid hydroperoxides undergo GSH peroxidase-mediated reduction to water and lipid hydroxides, respectively, with GSH providing the reducing equivalents . GSH also readily forms adducts with a great variety of both exogenously and endogenously derived electrophilic reactive intermediates (FIGURE 1) [6,79]. Formation of the GSH adducts is generally facilitated by GSH S-transferases (GSTs) and is normally considered to represent a detoxification of the relevant reactive intermediate. The resulting GSH adducts are then exported from cells by transporters [80,81]. GSH/GSSG homeostasis plays an important role in maintaining cellular redox status . Changes in the half-cell reduction potential of the 2GSH–GSSG couple correlate with the biological status of the cell. Therefore, determinations of the reduction potential can be used to better understand the redox biochemistry that results from oxidative stress. The redox potential of the 2GSH–GSSG couple can be readily obtained from the Nernst Equation:
with lower (more negative) redox potential representing more reducing conditions. According to this equation, the cellular redox potential is a second-order function of GSH concentration, which means that a change in concentration of GSH even without a change in GSH/GSSG ratio could alter the cellular redox status.
Determinations of the cellular redox state require a methodology for accurately measuring the intracellular concentrations of both GSSG and GSH. Many different methods have been published for the quantitative analysis of GSH and GSSG. Despite the large variety of methods that are available for GSH and GSSG analysis, many of them still have limitations. It is well known that GSH oxidation is a serious problem, which can lead to overestimating the amount of GSSG, particularly in cell lysates where redox cycling with abundant protein thiols can readily occur (FIGURE 7). Hence, we developed a SID LC–MRM-MS method for the simultaneous quantitation of cytosolic GSH and GSSG. The method utilizes 4-fluoro-7-sulfamoylbenzofuran as a thiol-derivatizing reagent that can rapidly and completely derivatize GSH. This fast and efficient derivatization is essential in preventing oxidation and enabling reliable measurements of GSSG to be made in the presence of a large excess of GSH. The method was validated and shown to give parallel GSH standard curves (FIGURE 8) when GSSG was added in increasing amounts to cell lysates (solid triangles) and buffer samples (solid circles). The intercept on the y-axis reflected the endogenous concentration of GSH in the cell lysate and the modest rise reflected the small amount of GSH present in the GSSG that was added. These data contrast with that obtained using a conventional SID LC–MRM-MS procedure without thiol derivatization. The slope of the GSH standard curve (FIGURE 8) when GSSG was added to the cell lysate (solid squares) deviated considerably from that obtained in buffer alone (solid circles). This parallelism experiment is illustrative of a general method to determine whether a SID LC–MRM-MS method is quantifying the correct endogenous analyte or some interfering substance. The assay was able to detect subtle changes in the redox potentials of two macrophage cell lines with different phenotypes when they were treated with the endogenous reactive electrophile, 4-oxo-2(E)-nonenal (ONE) . It is currently being employed to examine the redox status of epithelial cells from different animal models and circulating lymphocytes from different human disease states. Elaboration of the SID LC–MRM-MS biomarker methodology for GSH and GSSG to other important cellular thiols such as homocysteine, cysteine and coenzyme A, together with their corresponding disulfides, is currently under development. This will then provide a comprehensive method to monitor thiol biomarkers of oxidative stress to complement the isoP methodology.
Intracellular GSH is present in the cytosol (85–90%), with the remainder being found in the mitochondria, nuclear matrix and peroxisomes . This means that intracellular reactions with endogenous electrophiles occur with a large molar excess of GSH. With the exception of bile, which contains up to 10-mM GSH , extracellular concentrations of GSH are relatively low. For example, GSH concentrations in plasma are in the range of 2–20 µM . GSH is involved in the formation of endogenous bioactive eicosanoids [6,85] and is a source of reducing equivalents in a number of biosynthetic reactions . GSH also plays an important role in the detoxification of endogenous reactive intermediates such as ONE, 4-hydroxy-2(E)-nonenal, leukotrienes, oxo-eicosatetraenoic acids (ETEs) and estrogen quinones, as well as reactive metabolites from chemicals and drugs such as acetaminophen, arylamines, clozapine, carbamazepine, 3-methylindole and valproic acid [6,85]. Furthermore, a trend was observed in the association between increased GSH adduct formation in vitro and increased drug-induced toxicity in vivo . Although some reactive intermediates can form adducts with GSH directly, GST-mediated reactions generally predominate . Therefore, the actual adduct structures are dependent upon the regioselectivity conferred by the particular GST that is involved. When extensive reactive intermediate formation occurs (as in the metabolism of acetaminophen), the resulting GSH adduct formation causes depletion of intracellular GSH, allowing the free sulfhydryl groups on cellular proteins to form adducts through nonenzymatic processes [88,89]. The resulting modified proteins can be immunogenic and stimulate toxic immune reactions . In this manner, GSH adducts are useful biomarkers of potential drug-induced toxicity.
Positive ion LC–MRM-MS has been used extensively to detect GSH adduct biomarkers in cell culture and perfused organs. Analyses are performed either in the full scanning mode or by constant neutral-loss scanning for 129 Da (γ-glutamyl moiety) to detect GSH-derived metabolites, followed by CID and MS/MS analysis of MH+ . Negative-ion ESI methodology in combination with precursor ion scanning of m/z 272 (M-H-H2S) can provide a complementary method for detecting GSH adducts in cellular incubations . More recently, SID methodology was used in combination with an isotope pattern-dependent scanning method applied to the data acquisition of GSH adducts from drug-derived reactive metabolites . In these experiments, recorded full-scan MS and MS/MS data sets were further processed with neutral-loss filtering, product-ion filtering and extracted ion chromatographic analysis to search for protonated molecules and MS/MS spectra of GSH adducts. This approach proved very effective in detecting low levels of GSH adducts, regardless of their fragmentation patterns. The characterization of GSH adduct biomarkers from drug candidates during metabolism studies is critical, since it allows potential sites and mechanisms of bioactivation to be determined. Such studies can be coupled with structural modification of the drug candidate to block sites susceptible to metabolic activation and formation of a reactive intermediate. This knowledge can lead to the development of drug candidates that are less able to bind to cellular macromolecules and reducing their potential for causing adverse events if they are ultimately used in clinical studies [79,89,94].
There is increasing interest in the use of SID LC–MRM-MS-based methodology for the analysis of endogenous GSH adducts as biomarkers of oxidative stress . The 4-hydroxy-2(E)-nonenal-GSH adduct can exist as a mixture of eight diastereomers , four of which can be separated during LC–MS analysis . Analysis of GSH adduct biomarkers has been found to provide insights into the pathological role of 4-hydroxy-2(E)-nonenal . GSH-adducts derived from 4-hydroxy-2(E)-nonenal have also been used as biomarkers for monitoring oxidative stress in animal models  and Alzheimer’s disease . Characterization of the lipid hydroperoxide-derived thiadiazabicyclo-ONE-GSH adduct (TOG) has provided a biomarker that can also be used to monitor intracellular oxidative stress (FIGURE 9) . Using SID LC–MRM-MS analysis of TOG, we recently showed that the lipid hydroperoxide 15(S)-hydroperoxyeicosatetraenoic acid (HPETE) is detoxified more rapidly when it is formed in the cytosol, compared with that formed on intact lipids (FIGURE 9) . Other ONE-like bifunctional electrophiles can be formed from the homolytic decomposition of lipid hydroperoxides. Dioxododecenoic acid  and dioxooctenoic acid  contain the carboxylate terminus from linoleic acid- and arachidonic-derived lipid hydroperoxides, respectively. Both dioxododecenoic acid and dioxooctenoic acid also form TOG-like GSH adducts . In the future it might be possible to use these GSH adduct biomarkers to identify the particular polyunsaturated fatty acid-derived lipid hydroperoxides that are involved in the induction of intracellular oxidative stress. Leukotriene (LT)A4- and oxo-ETE-derived GSH adducts (OEGs), cysteinylglycine adducts (OECs) and FOG-7 (FIGURE 4) can also be analyzed using SID LC–MRM-MS methodology [6,17,85,104]. However, these studies have been restricted primarily to cell culture models owing to the extensive metabolism of the GSH adducts that occurs in vivo as described below. Recently, 15-oxo-ETE was identified as a 15-lipoxygenase-derived metabolite formed by the action of 15-PGDH on 15(S)-HETE . The 15-oxo-ETE was found to inhibit endothelial cell proliferation in vitro and also to rapidly form 15-OEG (FIGURE 10). It is noteworthy that 15-PGDH is downregulated in colon cancer tissues . This implies that it will be possible to use SID LC–MRM-MS analysis of 15-OEG and/ or its metabolites such as 15-OEC (FIGURE 4) to determine whether they are useful biomarkers of 15-PGDH deficiency and/or tumor progression.
Glutathionone adducts are secreted from cells by various transporters [80,81]. These adducts are then metabolized by GGT, an enzyme that exists both bound to the extracellular surface of the plasma membrane and circulating freely in plasma . The result of GGT metabolism is the formation of a cysteinylglycine adduct, as exemplified in the case of acetaminophen (FIGURE 11). The cysteinylglycine adduct can undergo further dipeptidase-mediated metabolism to a cysteine adduct. Finally, the cysteine adduct can be metabolized by N-acetyl transferase (NAT) to an N-acetyl cysteine adduct, also known as a mercapturic acid derivative. This pathway of metabolism also occurs for endogenous GSH adducts . For example, the GSH adduct LTC4 undergoes GGT-mediated metabolism to LTD4, dipeptidase-mediated metabolism to LTE4, which is excreted in the urine and so can be analyzed by SID LC–MRM-MS . LTE4 then undergoes NAT-mediated metabolism to N-acetylLTE4 . Mercapturic acid derivatives (N-acetylcysteine adducts) have been used as biomarkers of exposure to many exogenously derived chemicals [108–112] and endogenously derived bioactive substances [6,113–115].
Stable isotope dilution LC–MS has played a critical role in the establishing DNA adducts as biomarkers for genotoxins derived from environmental chemicals, drugs and endogenous reactive metabolites [116–118]. Availability of these biomarkers has provided insight into both toxic processes and to the multiple steps involved in carcinogenesis. The action of NER and BER enzymes results in the excretion of DNA adducts in the urine (FIGURE 1) [119,120]. Urinary DNA adducts can also be obtained by hMTH1-mediated hydrolysis of oxidized bases that are present in the trinucleotide pool . This makes it possible to use noninvasive LC–MS-based techniques to monitor urinary DNA adducts as both exposure and biological response biomarkers. The SID LC–MRM-MS approach has proven to be particularly useful for assessing the presence of DNA adducts urine and in tissue DNA samples [122–128].
Over the last decade, we have conducted a series of studies using SID LC–MRM-MS to develop biomarkers of lipid hydroperoxide-induced DNA damage . During these studies we discovered that COX-2 can convert arachidonic acid into 15(S)-HPETE, which then undergoes intracellular reduction to 15(S)-HETE . We reasoned that, in settings of oxidative stress, where reducing pathways are compromised, 15(S)-HPETE might survive long enough to undergo homolytic decomposition to DNA-reactive bifunctional electrophiles that would then generate biomarkers of lipid hydroperoxide-derived DNA damage. In previous studies we had shown that 15(S)-HPETE (and other HPETEs) decomposes to 4-hydroperoxy-2(E)-nonenal (HPNE) and ONE and that they form ε- and Hε-DNA adducts, respectively (FIGURE 12). Importantly, COX-2 is localized to the nuclear membrane, which maximizes the potential translocation of any bifunctional electrophiles into the nucleus. Using RIES cells that stably express COX-2, it was possible to show that basal endogenous production of 15(S)-HPETE resulted in the formation of HεdGuo adducts . As predicted from in vitro studies [130,131], there was a dose-dependent increase in endogenous HεdGuo adduct formation when vitamin C was added to the RIES cells . HεdGuo-adduct formation and 15(S)-HPETE biosynthesis were both inhibited by a selective COX-2 inhibitor and by aspirin, a nonselective inhibitor.
The role of COX-2-induced DNA damage in vivo was studied using the Min mouse model of colon carcinogenesis . Min mice provide a useful model to explore the consequences of increased in vivo COX-2 expression. The expressed dominant negative 95 kDa-truncated APC protein in Min mice is thought to prevent processing of β-catenin. This results in increased nuclear β-catenin, which, together with T-cell factor, activates expression of COX-2. In our experiments, DNA was isolated from the entire small intestine of C57BL/6J and C57BL/6JAPCmin mice and then hydrolyzed in the presence of [15N5 or 15N3]-internal standards. In separate experiments, it was shown that the Hε-DNA adducts were not generated as an artifact during the isolation and hydrolysis by adding [13C15N]-DNA. The use of labeled DNA provides a general method to confirm whether an endogenous DNA adduct can arise artifactually during isolation and hydrolysis of DNA or during LC–MS analysis. SID LC–MRM-MS employed in these studies showed that HεdGuo was increased in the Min mice compared with the wild-type mice, while Hε-2′-deoxycytidine (dCyd) was much less abundant. This suggested that DNA repair pathways favored HεdAdo and HεdCyd over HεdGuo in the DNA and that the excised adducts might be excreted in the urine.
DNA is very efficiently repaired and only a limited amount of the total DNA damage results in permanent mutations (7). This means that DNA adducts, which are repaired and appear in the urine, can be used as surrogate biomarkers of DNA damage [119,120]. SID LC–MRM-MS in combination with immunoaffinity purification can provide the specificity and sensitivity for analysis of endogenous and exogenous DNA adducts in mammalian tissue and biofluid samples [133–137]. We anticipate that immunoaffinity purification in combination with highly sensitive SID LC–MRM-MS, will be required for the specific quantification of urinary Hε-DNA adducts. Since the finding that HεdCyd is highly mutagenic in human cells [138,139], the quantification of urinary Hε-DNA adducts could potentially allow the identification of human populations at risk of cancer through lipid peroxidation-mediated DNA-damage.
Elevated levels of oxidatively damaged DNA bases have been measured as biomarkers to monitor numerous diseases including Alzheimer’s disease , Parkinson’s disease , autoimmune diseases , cardiovascular diseases [143–145] and cancer [146–148]. It has been suggested that such damage plays an important role in the etiology of these diseases [149,150]. Oxidative DNA damage is also increased through exposure to environmental chemicals [117,151]. In addition, there is some evidence of a link between ingestion of antioxidants and reduced oxidative damage to DNA . 8-oxo-dGuo is by far the most extensively studied lesion that arises as a consequence of oxidative DNA damage [144,150,153]. Unfortunately, it is extremely easy to induce artifactual formation of 8-oxo-dGuo in cellular DNA during the isolation and hydrolysis procedures that are commonly used [144,154]. Therefore, numerous studies have been conducted to address the problem of quantifying 8-oxo-dGuo in DNA, including the important contributions of the Cadet group [155,156] and the European Standards Committee on Oxidative DNA Damage [157–163]. The comet assay has predicted that basal endogenous concentrations of 8-oxo-dGuo in cellular DNA should be at the level of three to five molecules of 8-oxo-dGuo/107 dGuo . Physicochemical methods developed to analyze 8-oxo-dGuo have consistently obtained values that are almost an order of magnitude higher at 34–37 molecules of 8-oxo-dGuo/107 dGuo . These elevated values were obtained even when highly specific and sensitive methodology based on immunoaffinity purification and stable isotope dilution LC–MS was used , which suggests that artifactual formation of 8-oxo-dGuo might occur during DNA isolation and hydrolysis [144,154]. Significant advances have been made to the comet assay by introducing the use of hOGG1 to specifically remove 8-oxo-dGuo from the DNA so that it is now possible to reliably measure basal cellular oxidative DNA damage . hOGG1 is known to excise both 8-oxo-dGuo and 8-oxo-Gua, producing a single -trand break. It will also remove formamidopyrimidine-dGuo, derived to yield a single-strand break. Unfortunately, a linear dose response to KBrO3-induced oxidative DNA damage was not obtained, even with this significantly improved method . This implies that lesions other than 8-oxo-dGuo can be detected by the comet assay, even when the assay is used in combination with the selective hOGG1 enzyme.
We developed an immunoaffinity purification SID LC–MRM-MS assay for the analysis of 8-oxo-dGuo in cellular DNA. This technique provides the ultimate in analytical sensitivity and specificity that can be attained using physicochemical methodology for analysis of endogenously generated lesions in DNA . Yet, even when this assay was used with conventional DNA isolation procedures, basal levels of 8-oxo-dGuo were similar to those described previously . These experimentally generated adducts were only reduced to acceptable levels when extreme care was taken during sample preparation. First, artifactual generation of 8-oxo-dGuo was reduced when cellular DNA was isolated using commercially available DNAzol (which contains high concentrations of guanidine thiocyanate)  instead of other chaotropic- [167,168] or phenol-based  extraction methodology. Second, Chelex-treated water was used to reduce transition metal-ion contamination in the buffers that were used . Last, deferoxamine was added to complex any residual transition metal ions that were still present in buffers following the Chelex treatment. These steps prevented Fenton chemistry-mediated generation of ROS and artifactual oxidation of DNA bases [144,154], resulting in the levels of 8-oxo-dGuo obtained using these precautions to agree with levels previously determined using the comet assay (FIGURE 13). The LC–MS assay was shown to sensitively and specifically measure 8-oxo-dGuo formed in the DNA of human lung bronchoalveolar H358 cells treated with the redox-active benzo[a]pyrene (B[a]P)-7,8-dione, a reaction product of aldoketo reductase-mediated polycyclic aromatic hydrocarbon-trans-dihydrodiol activation . The immunoaffinity SID LC–MRM-MS assay has now been elaborated to the analysis of urinary 8-oxo-dGuo where it was able to show a significant increase in the urine of smokers [Mangal D, Mesaros A, Blair IA, Unpublished Data]. Therefore, it holds promise as a biomarker to identify subjects at increased risk for oxidative stress.
Exogenously derived DNA adducts, such as those arising from environmental chemicals, can also be employed as biomarkers of exposure . Polycyclic aromatic hydrocarbons, such as B[a]P, are ubiquitous environmental chemicals found in vehicle exhaust emissions and tobacco smoke . A substantial research effort has been conducted on B[a]P-mediated DNA adduct formation to provide biomarkers of exposure to B[a]P . We developed SID LC–MRM-MS methodology in order to quantify eight anti-7,8-dihydroxy-9,10-epoxy-7,8,9,10-tetrahydro-B[a]P(B[a] PDE)-derived DNA adduct biomarkers in four H358 human bronchoalveolar cell lines with different phenotypes . In CYP1A1/1B1-induced H358 cells exposed to (±)-B[a]P-7,8-dihydrodiol, (+)-anti-trans-B[a]PDE-N2-dGuo was the major DNA adduct . Surprisingly, the greatest amount of (+)-anti-trans-B[a]PDE-N2-dGuo was formed in the control H358 cells. These data raise the intriguing possibility that CYP1A1/1B1 might be protective against (+)-B[a]PDE-mediated DNA damage. Going forward we will assess the relationship between cellular DNA and urinary DNA adducts. Such studies will require the sensitivity offered by new generation LC–MS instrumentation, which will make it possible to conduct analyses in circulating lymphocytes.
The need for SID MS quantification of protein and peptide biomarkers arises from both basic and clinical researchers. Driving these assays is a push for increased selectivity and sensitivity during both the discovery and validation phases of biomarker development. Currently, the great majority of validated procedures used to quantify biological macromolecules are based on immunoassay methodologies. Although these antibody-based techniques are exceptionally sensitive (typically picogram per milliliter range), long development time, variability and crossreactivity have stimulated the interest in an equally sensitive, though more selective and controlled, assay platforms. The proven capability of SID MS, considered the ‘gold standard’ for small-molecule quantification, offers an attractive solution to the limitations associated with antibody-based detection. Peptide and protein measurement by SID MS does, however, pose unique challenges and the field is striving to match the lower limits of quantification typical of those seen using ligand-binding assays, such as ELISAs [3,175–180].
There are several key advantages characteristic of SID MS when compared with antibody-based methods alone. As noted by Carr and Anderson, analyte detection by MS can be structurally unambiguous, essentially offering absolute selectivity . Even in instances where selectivity may be compromised in a LC–MS/MS experiment (i.e., a co-eluting analyte has isobaric precursor and product m/z), these interferences can often be identified and the LC–MS method altered to relatively quickly eliminate the problem. The ability to assay multiple peptides in a single LC–MS/MS experiment is another strength not offered by standard immunoassays. Several SID MS experiments have been conducted, showing the ability to quantify peptides correlating to multiple proteins in a single run [176,179,181]. This multiplexing significantly increases assay throughput, which may prove critical to realizing SID MS of macromolecules in the clinical setting, where biomarker panels will most likely be the reality . Last, unlike immunobased detection, which is reliant on the binding interaction of two macromolecules, LC–MS detection is direct and only depends on the chemical nature of the analyte. This characteristic of SID MS often results in excellent assay linearity over several orders of magnitude, whereas ligand-binding assays are often limited in dynamic range and often show a nonlinear responses . The utility of SID MS in all stages of biomarker development, from identification, through validation and ultimately into clinical practice has proven this technology holds a central role in biomarker research [184–187]. The extension of SID MS to protein quantification, however, is an evolving field and its application is not trivial. In this review, we will cover the historical development, critical challenges and current state of the field, highlighting select publications during the discussion.
A typical workflow for the analysis of peptides and proteins by a ‘bottom-up’ approach involves the enzymatic cleavage of the large molecule followed by the analysis of the resultant peptides by LC–MS/MS. Traditionally, the endopeptidase trypsin is used to digest a protein into its constituent peptides. Trypsin cleaves proteins C-terminal of arginine and lysine, except when blocked by an adjacent C-terminal proline residue . The average tryptic peptide is ideal for LC–MS/MS sequencing, due to its relatively small length and single C-terminally located basic residue. Positive-mode ESI of tryptic peptides will normally produce M+2H+ and M+3H+ species with m/z between 400–1000 and, upon CID, will show a predominant y- and b-fragment ion series . A protein is qualitatively identified in such an experiment when one or several peptides are sequenced that are unique to the protein . The bottom-up approach is similarly used for protein quantification, where the amount of one or several unique peptides corresponds stoichiometrically to the amount of the whole protein. Such peptides, when they also ionize well by positive ESI, are sometimes referred to as ‘prototypic’ or ‘signature’ peptides [191,192]. Protein quantification using a bottom-up workflow involving protease digestion is shown in FIGURE 14. In FIGURE 14, the tryptic peptide AVFPSIVGR, corresponding to human β-actin, was quantified following trypsin digestion of the whole protein. The isotopically labeled peptide AVFPSIVG[13C615N4]R was used as an internal standard. The peptide pair in this example was analyzed using a TQ instrument operated in the MRM mode, where the transition masses are the peptides’ monoisotopic doubly charged M+2H+ ions to singly charged y6 and y7 ions, respectively. The resultant chromatographic trace for each peptide is then represented by the sum of both transitions (FIGURE 14).
In 1996, Barr et al. reported the first attempt to quantify a protein by ‘bottom-up’ isotope dilution-MS methodology . In this seminal study, apolipoprotein (apo)-A-1 was quantified using isotope-labeled reference peptides synthesized using standard N-(9-fluorenyl)methoxycarbonyl (Fmoc) chemistry. The reference peptides incorporated [D3]-leucine and [13C3]-alanine and corresponded to three apo A-1 tryptic peptides. quantification of the standard reference peptides was accomplished by nitrogen analysis and amino-acid analysis. The experimental design entailed the addition of two reference peptides prior to tryptic digestion. One peptide provided a primary measurement, while the other was used to confirm this calculated amount. A third standard peptide was added postdigestion and was used to determine an absolute recovery value. The results showed the assay to be exceptionally precise, with approximately 4% coefficient of variation (% CV). The goal of this assay was to develop an alternative approach to immuno-based detection for the quantification of reference protein. A noted caveat of the experiment was the requirement that the sample material must be abundant, well characterized and pure. In this analysis, material quantified was purified apo-A1 (~ 1 mg/ml of a certified reference material) that had been extensively characterized and shown to be free from degradation and modification. Since the peptides quantified in the study contained modifiable amino acids and did not control for the rate of trypsin hydrolysis, the ability of the assay to quantifiy trace amounts of apo-A1 in a complex biological matrix would be questionable. In fact, although the assay was extremely reproducible, 44% of the experimental imprecision was attributable to differences between sample digestions, rather than purely analytical error.
Following the work of Barr et al., in independent studies, the laboratories of Aebersold and Chait used isotope-labeling strategies to quantify proteins expressed in yeast . Using an isotope-coded affinity tag (ICAT), working in the Aeborsold group, Gygi et al. were able to selectively label the sulfhydryl groups of cysteine-containing proteins with either deuterium-labeled (D8) or unlabeled reagent. Tagging two experimentally conditioned groups of yeast with the ICAT it was possible to perform relative quantification and determine differential protein expression between the two groups. A limitation to the technique, however, was that only cysteine residues were labeled and, hence, only cysteine-containing peptides served in the analysis . In an alternative approach, Oda et al., in the Chait laboratory, uniformly labeled yeast with 15N. In these experiments, two batches of yeast, one in natural abundance 14N (99.6%) and the other in 15N (>96%) enriched cell growth media were prepared. The yeast strains used were mutants that expressed or were deficient in, a G1 cyclin protein that regulated G1→S transition in budding yeast. The comparison of the yeast strains, using this labeling strategy, not only revealed alterations in protein expression levels, but showed differences in phosphorylation on important signaling proteins. As the authors’ note, the gel fractionation performed would not alter the observed unlabeled to labeled peptide ratios, since the 15N-labeled proteome was mixed with the unlabeled proteome prior to sample workup . The added control offered by the addition of the labeled proteome during early stages of sample preparation is a great strength of the technique.
Today, most LC–MS-based biomarker-discovery experiments using stable isotopes are accomplished by either tagging with chemical reagents or metabolic labeling of cells in culture. The majority of current chemical tagging techniques label peptides on the N-terminus and epsilonamino of lysine using N-hydroxysuccinimide chemistry. Isobaric mass tags are particularly popular for biomarker-discovery experiments since the mass difference between labeled forms occur on fragment ions rather than on the parent analyte [196,197]. In this manner, MS spectra are not complicated and the peptide MS signal is increased by multiplexing. These reagents and their application have been reviewed [198,199]. Notably, demonstrating the multiplexing capabilities and utility of the iTRAQ reagent, Zhang et al. monitored the time-resolved tyrosine phosphorylation events downstream of the epidermal growth factor receptor . Following on the promise of this earlier work, the group more recently was able to probe the EGF receptor signaling network in glioblastoma cell lines, which revealed a cross-activation between the epidermal growth factor receptor and c-MET receptor pathways. This result led to the exciting and clinically relevant finding that the dual inhibition of both receptors had greater cytotoxicity than either treatment alone on cultured glioblastoma cells . This and similarly successful experiments showed the utility of the chemical labeling approach in biomarker discovery type experiments [202–206]. A limitation to this application, however, is that the isotope label is introduced postprotein extraction, meaning extreme care must be taken to ensure error is not introduced during sample preparation. Additionally, other factors, such as mixed MS/MS spectra, resulting from the co-isolation of a contaminating ion during MS/MS acquisition, can lead to aberrant quantification .
Another powerful technique employed in biomarker discovery experiments is stable isotope labeling by amino acids in cell culture (SILAC). Following the work performed in the Chait laboratory, in which uniformly 15N-labeled yeast were prepared , Ong et al. extended the concept to mammalian cells and termed the method SILAC . Rather than uniform 15N labeling, SILAC metabolic labeling introduces specific 13C-and 15N-labeled amino acids. In a typical SILAC workflow, an immortalized cell line is grown in media containing unlabeled essential amino acids or media containing labeled counterparts (i.e., [13C615N2]-lysine and [13C615N]-leucine). Unlabeled and labeled cells are then mixed following a control or experimental treatment and a bottom-up LC–MS/MS analysis is performed . Unlike isobaric labeling, SILAC results in a mass difference at the MS level. Due to this feature, MS spectra do become more complicated upon multiplexing. To improve quantification at the MS level in such cases, high resolution MS instruments are often employed [210,211]. There are numerous examples illustrating the successful identification of candidate biomarkers using SILAC methodology [212–216]. In a recent study, Pan et al. in the laboratory of Matthias Mann, described the global evaluation of phosphorylation alterations in response to kinase inhibition . A multiplexed SILAC method using ‘light’ ([12C6]-arginine, [12C6]-lysine), ‘medium’ ([13C6]-arginine, [2H4]-lysine) and ‘heavy’ ([13C615N4]-arginine, [13C615N2]-lysine) cell growth media was developed. The group was able to relatively measure thousands of phosphorylated peptides from the three SILAC cell populations, which were exposed to control treatment, EGF stimulus or growth factors in the presence of kinase inhibitors. Following a peptide fractionation and phospho-peptide enrichment step, the fractions were analyzed by LC–MS and the response pattern of the three labeled proteomes was determined. In this manner, phosphorylation events were delineated into nonresponders, growth factor responders and those activated by growth factors but suppressed by inhibitors . As SILAC methodologies become more streamlined and data analysis software is more effectual , a greater variety of candidate biomarkers will be uncovered.
A critique often encountered when comparing chemical labeling to metabolic labeling strategies is that only the former is applicable to investigating primary source materials such as biofluids, tissues and biopsies. Following the experimental workflow outlined above, this is indeed true. However, the Blair laboratory has shown the utility of using the SILAC-labeled proteome as a reference standard that can be applied to any source material. Proof of principle for this approach was established in the work of Yu et al. . The CAPAN-2 pancreatic cancer cell line was labeled by SILAC and secreted proteins from the conditioned media were collected. This stable isotope-labeled proteome (SILAP) standard was added to pooled pancreatic cancer or control serum (FIGURE 15). Over 100 differentially expressed biomarker candidates were identified, with the proteins intercellular adhesion molecule 1 and basal cell-adhesion molecule independently validated. This approach was extended by Rangiah et al. by developing a MRM method combined with a SILAP standard to study biomarkers in the Apcmin mouse, a colon cancer model . Candidate biomarkers were identified in the secreted proteome of the CT26 colon cancer cell line and an MRM method was developed to measure 12 biomarkers of interest. The CT26-derived SILAP standard was added to pooled Apcmin mouse or normal serum samples. Levels of all 12 biomarkers could be quantified with differential expression validated by Western blot analysis for five biomarkers. Demonstrating the versatility of this approach, Shah et al. pooled SILAC-labeled human epithelial endocervical-1 and vaginal cell secretomes. This SILAP standard was then used in the characterization and relative quantification of proteins present in patient cervicovaginal fluid in a preterm birth study. The result of the investigation showed that three proteins were significantly elevated in preterm birth cases (desmoplakin isoform 1, stratifin and thrombospondin 1 precursor), providing a foundation for further validation experiments . Limiting the analysis to proteins secreted and overexpressed in the cell line model makes it possible to exclude acute phase proteins and other abundant proteins found in serum or cervicovaginal fluid, while simultaneously focusing on proteins with biological relevance to the disease of interest.
Gygi and colleagues have extended the methodology of protein quantification by stable isotope-labeled peptide standards to cellular proteins resolved on an SDS-PAGE gel and termed the technique AQUA . In their studies, an isotope-labeled phosphopeptide standard was used to quantify endogenous amounts of a phosphorylated protein. This innovation offered accessibility to many important protein biomarkers related to kinase signaling . Additionally, the application of the technique to proteins separated by SDS-PAGE gel provided an attractively simple, established and widely applicable means of reducing sample complexity. AQUA has become a critical technique in the validation of candidate protein biomarkers .
An important development of the AQUA technique was reported by Mayya et al. in a 2006 study that monitored the cell cycle-dependant phosphorylation of CDKs on multiple residues. These experiments used synthetic peptides representing the CDK regulatory sites T14 and Y15 contained in the tryptic peptide IGEGT14Y15GVVYK in three phosphorylated and one unphosphorylated form. In this manner, it was possible to quantify not only total CDK, but also calculate the amount of the phosphorylation forms present . Although the experiments were not able to control for CDK protein immunoprecipitation, results were reproducible and in agreement with quantitative immunoblotting. Many regulatory phosphorylation sites, such as those that occur within the catalytic loop of a kinase, exist on multiple residues within a short stretch of amino acid residues. Therefore, quantification of all phosphorylated and unphosphorylated forms of these peptides can provide valuable information indicating activation state [224,225].
In their initial study, Barr et al. highlighted many of the confounding issues associated with the AQUA technique, chief among them being reference material purity/stability and variability of protein hydrolysis. For a peptide standard quantified by SID MS to match its value calculated by amino acid analysis, any resolubilization of the peptide post-amino acid analysis must be complete. Additionally, no modifications to the standard can occur (e.g., oxidation or carbamylation) during peptide storage or early sample preparation. Lastly, degradation and absorptive losses incurred during peptide storage must be minimized [226,227]. Recent reports have investigated the accuracy and robustness of SID MS of proteins using standard peptides [227,228]. Mirzaei et al. compared peptide standards generated synthetically with those expressed from a synthetic gene in vitro. In this second approach, termed quantification concatamers (QconCAT), a series of targeted tryptic peptides are concatenated into a single plasmid and expressed using a bacterial transcription/translation system in vitro. The expression of the QconCAT is performed in a medium containing isotopically labeled amino acids (e.g., [13C615N2]-lysine and [13C615N4]-arginine, used in place of the unlabeled species). A series of 25 tryptic peptides were assembled into QconCATs or synthetically expressed individually. The ability of the two standard methods to report equimolarity between the 25 peptides was then investigated. Multiple resolubilization buffers and digestion parameters were compared, with the result that neither method was proven optimal and both suffered from a combination of the above issues. Interestingly, synthetic peptides appeared to suffer from solubility and modification, while the QconCAT peptide showed variability in digestion and required extensive digestion optimization . This study emphasizes that neither a single nor a concatenated peptide is a true authentic standard for a whole protein and, unless great care is taken, accuracy can suffer.
Quantifying trace amounts of protein in complex biofluids, such as tissue, serum and plasma, using SID MS requires sample preparation strategies that enrich the analyte and/or reduce interferences [230,231]. It is estimated that 99% of serum’s total protein mass is due to the top 20 most abundant protein species . The dynamic range of protein abundance in serum may cover eleven orders of magnitude with albumin, the most abundant protein, being present at the extraordinarily high concentration of 60 mg/ml . A recent review of the literature identified that direct analysis of plasma is possible for proteins in the microgram per milliliter range . Bonder et al. demonstrated the quantification of the candidate prostate cancer biomarker Zn-α2 glycoprotein directly in serum without fractionation or abundant protein depletion. The results showed a linear dynamic range of 0.32–10 µg/ml and a inter- and intra-assay imprecision of less than 6% C V. The assay was able to quantitatively differentiate between pilot groups of normal men (3.65 µg/ml) and those suffering from nonmalignant prostate disease and PCa (6.21 and 7.59 µg/ml respectively). This assay is unique among the many LC–MS assays reported, in that a robust LC configuration utilizing analytical flow rates (250 µl/min) was used . Yet, although it is possible to measure a selected protein directly in serum, the level of sensitivity required to quantify many clinically relevant protein biomarkers, such as prostate-specific antigen (PSA) and cardiac troponins, which exist in the low nanogram per milliliter range, require additional sample preparation.
Similar to the benefits seen when quantifying small molecules following solid-phase extraction, peptide fractionated prior to LC–MS can dramatically improve the analytical performance. Coupled to reversed-phase separation, strong cation exchange (SCX) chromatography can serve as an effective and orthogonal form of peptide chromatography. SCX is often utilized in qualitative proteomics experiments to increase proteome coverage [235,236]. Recently, the technique has been used to quantify low level serum biomarkers. In their work, Keshishian et al. systematically explored the impact of abundant serum protein depletion followed by SCX fractionation on the ability to quantify a panel of six proteins spiked into human serum. The report indicated a ten- to 40-fold increase in signal-to-noise ratio, comparing fractionated to complex samples. As stated by the authors, the benefits of SCX fractionation was apparently achieved through both a reduction in background noise and an increase in analyte signal attributable to more target analyte being loaded on column per injection .
Since the report of Keshishian et al. illustrating the ability to quantify serum proteins in the 1–10 ng/ml range using limited fractionation , other studies have used the same or similar workflows with success. For example, PSA was quantified over 5–40 ng/ml in serum with excellent accuracy and precision using a combination of abundant serum protein depletion and mixed cation exchange fractionation. A benefit noted by the authors of this assay over existing methods using immunoenrichment of the PSA is that results are not complicated by free and bound forms of PSA. In the later case, some PSA species may have their antibody recognition region buried within a complex resulting in high assay variability. The issue of bound PSA can also cause disagreement between laboratories using antibodies that target different epitopes . In addition to analytical benefits, an antibody-free method may offer an economic advantage due to a reduction in reagent usage and time per assay.
Another means to reduce sample complexity at the peptide level is the technique of stable isotope standards and capture by antipeptide antibodies (SISCAPA). Anderson et al. introduced the SISCAPA technique in 2004 , and several excellent studies have recently established the approach [177–179,239]. Using the SISCAPA method, Hoofnagle et al., working in the Heinecke laboratory, successfully targeted three thyroglobulin peptides for quantification in patient serum. Thyroglobulin is a clinical tumor marker of thyroid carcinoma; however, commercial ligand-binding assays suffer from endogenous antibody interferences leading to a high false positive rate. One of the three peptides targeted was capable of quantifying thyroglobulin with the necessary level of sensitivity (2.6 ng/ml) and was in good correlation with a commercial immunoassay . Further extending the usefulness of SISCAPA, Kuhn et al., working in the Carr laboratory, verified the ability to multiplex assays. In their work, protein quantification was performed on peptides correlating to troponin I (cTnI) and IL-33, an established and emerging marker of cardiac injury and cardiovascular health. This assay showed satisfactory precision and a limit of LOQs of approximately 3 ng/ml (cTnI) and 5 g/ ml (IL-33). Good correlations were found when patient cTnI levels determined by a commercially available immunoassay were compared with the amount calculated using the SISCAPA assay .
Although fractionation steps at the peptide level are beginning to show great promise, protein immunoprecipitation prior to SID MS is exceptionally sensitive . N-terminal probrain natri-uretic peptide (NTproBNP), a clinically validated marker of heart failure, was measured following protein immunoprecipitation by Berna et al. In these experiments, NTproBNP was immunoprecipitated from serum, digested with trypsin and then quantified using a single isotope-labeled NTproBNP tryptic peptide. Following assay optimization, a lower LOQ was determined to be 0.1 ng/ml with a less than 16% CV. The assay was used to monitor drug-induced cardiac hypertrophy in rats, showing a quantitative correlation between heart weights and circulating NTproBNP concentration. Results indicated that following a 14-day drug treatment, the heart weight increased by 34%, while the NTproBNP amount increased from 0.107 (± 0.009) to 0.332 (± 0.03) ng/ml . These data are not only remarkably accurate but are an order of magnitude more sensitive than those seen using other LC–MS/MS methodologies.
Highlighting the limited degree of control offered by a standard peptide alone, in their report, Berna et al. determined that only approximately 63% of NTproBNP protein was recovered. The authors indicate this recovery value was most likely affected by immunoprecipitation, digestion and ion suppression . Similar levels of protein recovery following immunoprecipitation from serum have been reported . To ultimately control these variables and provide traceability across differing LC–MS platforms and enrichment anti bodies, a whole protein standard is required. The production of isotope-labeled whole protein standards was first developed and utilized in NMR analysis [242,243]. The cell-free synthesis of isotopically labeled proteins described by these NMR investigations has been used by our laboratory and others. The production of the cell-free recombinant protein is accomplished using extracted bacterial translation and transcription machinery and, therefore, requires that the target protein be inserted into a suitable expression plasmid. Isotopically labeled amino acids of the experimenter’s choice can then be added to the in vitro reaction to produce a variety of labeled species. Brun et al. investigated the accuracy of using isotope-labeled protein standards (PSAQ), as compared with AQUA or QconCAT, in the quantification of staphylococcal enterotoxins. The group investigated the ability to quantify the toxins in water and urine following a protein capture and fractionation method. It was shown, as expected, that both AQUA and QconCAT quantification severely underestimated the amount of toxin present, while the measurement using the PSAQ calibrant was acceptable . Other groups have since implemented the PSAQ approach in the quantification of therapeutic antibodies and in discerning closely related protein isoforms [245,246].
The Blair laboratory has significantly extended the PSAQ application with the creation of a post-translationally modified PSAQ standard. To facilitate our studies, full-length isotope-labeled and phosphorylated human focal adhesion kinase (FAK) protein was produced in vitro. The focus of our investigation was the Src kinase specific phosphorylation within FAK’s catalytic loop region. The amounts of total FAK protein, as well as phosphorylated forms, were quantified using standard peptides. Using this characterized reference material, it was then possible to calculate the amount of endogenous FAK and determine the absolute amount of phosphorylation present within the catalytic loop region relative to the amount of total FAK protein following immunoprecipitation . Applying this technique to a cell culture model, we were able to correlate the amount of FAK catalytic loop phosphorylation with Src kinase inhibition, confirming FAK phosphorylation as a biomarker of drug exposure in culture. These experiments are the first use of a post-translationally modified isotope-labeled whole protein standard and its usage provided novel data unattainable via other established methods.
Improved sensitivity using lower flow rates and newer generation mass spectrometers should make it possible to use readily accessible circulating lymphocytes as sources of potential in vivo biomarkers. This will prove to be particularly useful for DNA adduct analyses as it will be possible to compare DNA adducts in cellular DNA with urinary DNA adduct metabolites (FIGURE 1). These data will ultimately link the activity of DNA-repair enzymes and DNA-adduct formation in the prevention of tumorigenesis . Increasing instrument sensitivity will also revolutionize the approach to folate analysis (FIGURE 6), allowing analyses to be conducted in metabolically competent cells rather than the current use of RBCs . Lastly, the simultaneous quantification of free and oxidized thiols in the lymphocytes will make it possible to assess interindividual differences in oxidative stress and how they relate to the folate phenotype .
In addition to instrumentation, methodologies to produce preionized derivatives will lead to increases in assay sensitivity . This will hopefully prove particularly useful in the analysis of circulating steroids, making it possible to rigorously evaluate the role of CYP-mediated estrogen metabolism  and facilitate the development of new models of breast cancer risk . The use of preionized derivatives as described by Hong and Wang could also have applicability for the analysis of urinary DNA adducts . Although LC–ECAPCI-MS still provides the best sensitivity for analysis of lipids and permits chiral analyses to be conducted , it is not clear that reduced flow rates and modern TQ instruments will improve ECAPCI sensitivity still further. The requirement for normal phase solvents in order to accomplish chiral separations is currently very restrictive and will most likely stimulate the development of alternative approaches to improving ESI sensitivity, such as using preionized carboxylate derivatives.
The use of appropriate protein standards in SID LC–MRM-MS assays is critically important and is an active area of research within the feld of protein biomarkers. This is because the standard will define the scope and accuracy of an assay. While global relative quantification can be accomplished using an isotopically labeled proteome, no measurement in the analysis is absolute. Conversely, the measurement of a specific protein using labeled synthetic peptides can be absolute, but a specific protein or group of proteins is targeted. The choice of standard should be determined based on the fitness for the experiment at hand and no single methodology will likely serve every paradigm. During discovery and the early qualification stages of biomarker development, many candidates need to be screened for their usefulness. Hence, many measurements need to be made with acceptable analytical precision, but no specific degree of accuracy is required. Later stages of biomarker validation will be targeted by nature. As a biomarker assay moves into the clinical setting, the use of a traceable reference standard will become more critical and serve to benchmark interlaboratory comparisons.
For reprint orders, please contact moc.ecneics-erutuf@stnirper
Financial & competing interests disclosure
Support of biomarker research by NIH grants UO1ES16004, RO1CA091016 and RO1130038, held by Ian A Blair and P30ES0130508 held by Trevor Penning at the University of Pennsylvania is gratefully acknowledged. Ian A. Blair is the AN Richards Professor and Vice-Chair of the Department of Pharmacology, Director of the Center for Cancer Pharmacology, Director of the Proteomics and Systems Biology Facility and an employee of the University of Pennsylvania. Eugene Ciccimaro is an LC–MS applications chemist employed by the Scientific Instruments Division of Thermo Fischer Scientific. Description of the studies described in the article was not influenced by the authors’ current appointments, although research in many of the areas is currently ongoing. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.
No writing assistance was utilized in the production of this manuscript.
Papers of special note have been highlighted as:
of considerable interest