Abstract
BACKGROUND: Many avenues have been proposed for a seamless transition between biomarker discovery data and selected reaction monitoring (SRM) assays for biomarker validation. Unfortunately, studies with the abundant urinary protein uromodulin have shown that these methods do not converge on a consistent set of surrogate peptides for targeted mass spectrometry. As an alternative, we present an empirical peptide selection work flow for robust protein quantification.
METHODS: We compared the relative SRM signal intensity of 12 uromodulin-derived peptides between tryptic digests of 9 urine samples. Pairwise CVs between the 12 peptides were 0.19–0.99. We used a correlation matrix to identify peptides that reproducibly tracked the amount of uromodulin protein and selected 4 peptides with robust and highly correlated SRM signals. Absolute quantification was performed with stable isotope–labeled versions of these peptides as internal standards and a standard curve prepared from a tryptic digest of purified uromodulin.
RESULTS: Absolute quantification of uromodulin in 40 clinical urine samples yielded interpeptide correlations of ≥0.984 and correlations of ≥0.912 with ELISA data. The SRM assays were linear over >3 orders of magnitude and had typical interdigest CVs of <10%, interinjection CVs of <7%, and intertransition CVs of <7%.
CONCLUSIONS: Comparing the apparent abundance of a plurality of peptides derived from the same target protein makes it possible to select signature peptides that are unaffected by the unpredictable confounding factors inevitably present in biological samples.
Selected reaction monitoring (SRM),5 also known as multiple reaction monitoring, is a quantitative mass spectrometry (MS) technique that targets predefined precursor and product ions specific to a particular analyte of interest (1). Proteins are typically quantified by cleaving them into peptides with a specific protease, such as trypsin, measuring the concentration of at least one signature peptides, and inferring the concentration of the parent protein (2–5).
Uromodulin was selected as an exemplary target to test SRM peptide selection work flows because of its physiological importance, biological complexity, and association with disease phenotypes. Uromodulin, also known as UMOD or Tamm-Horsfall glycoprotein, is the most abundant protein in healthy human urine, but its functions remain incompletely understood (6). Data from genetically modified mice suggest that uromodulin protects against urinary tract infections and calcium oxalate crystals (6) and participates in the regulation of sodium reuptake to control blood pressure (7) (8) and glomerulocystic kidney disease (6). In these diseases, impaired uromodulin processing leads to its accumulation in the endoplasmic reticulum (9). Additionally, common uromodulin variants are associated with chronic kidney disease (10) (11) and hypertension (12), possibly via effects on salt reabsorption in the kidney (13). Some disease-associated variants are present at lower concentrations in urine (10). Exact quantification of urinary uromodulin as a novel biomarker of susceptibility to chronic kidney disease and hypertension is therefore of clinical interest and may represent a future diagnostic biomarker to monitor blood pressure–lowering treatment.
Uromodulin is well represented in proteomic MS databases. For example, aside from a 99-aa N-terminal region with only 1 tryptic cleavage site, PeptideAtlas has MS data representing 97% of the mature protein. Nevertheless, MS analysis is complicated by the existence of 4 major isoforms; a variety of silent, protective, and disease-associated single nucleotide polymorphisms and mutations; and multiple glycosylation sites and disulfide bonds. In addition, urine is challenging to analyze because its pH is inconsistent between samples and it has widely varying concentrations of uromodulin, albumin, total protein, urea, salts, creatinine, and other metabolites (14).
A variety of methods have been previously used to identify signature peptides for protein quantification (2, 15, 16). One common approach is to target peptides identified by a data-dependent MS screen on related samples, as these peptides are guaranteed to be detectable by MS (2, 17). A limitation of this approach is that discovery MS and quantitative MS are traditionally performed on different types of MS instruments with different liquid chromatography (LC) systems, ionization, collision cells, and fragmentation patterns. Consequently, the dominant peptides that provide for highly confident protein identification on one instrument do not always yield sufficient MS signals for quantification on a different instrument. In addition, long peptides (e.g., >10 aa) generally yield more MS/MS fragment ions for confident identification, whereas shorter peptides are more likely to yield a limited number of dominant fragment ions for sensitive SRM quantification. A related approach is to target peptides found in spectral peptide libraries (18). Available libraries contain spectra representing many thousands of peptides collected from hundreds of MS runs, thereby facilitating the selection of target peptides and transitions that have been reproducibly observed (e.g., http://chemdata.nist.gov/dokuwiki/doku.php?id=peptidew:start). However, current MS spectral databases are primarily populated with data from discovery MS instruments and are therefore not directly applicable to SRM assays. SRMAtlas, an online resource designed to overcome this limitation, has MS spectra from natural and synthetic peptides collected on triple-quadrupole mass spectrometers, the most common instrument for SRM. A prepublication SRMAtlas preview covers 99.9% of the human proteome (19). A third approach, in silico prediction of proteotypic peptides based solely on a protein's amino acid sequence, provides an alternative to relying on previously acquired spectra that is especially useful for pioneering work on biological samples that have not been subjected to extensive proteomic analysis (20).
Peptide selection for an SRM assay requires more than the mere identification of detectible peptides (21). If the goal of the experiment is to quantify the total protein concentration, the selected peptides should not contain genetically encoded variations and should not be susceptible to in vivo or in vitro posttranslational modifications. On the other hand, if the goal is to monitor a specific isoform, single nucleotide polymorphism, or posttranslational modification, peptide selection is constrained by the need to target specific peptides that may have relatively weak SRM signals and therefore require extensive optimization (22).
Here we demonstrate that unpredictable confounding factors can interfere with MS quantification. Thus, selection of peptides for a robust assay requires experimental data. We present an empirical peptide selection work flow to identify surrogate peptides suitable for determining the concentration of targeted proteins in a complex biological milieu by identifying peptides with highly correlated SRM signals.
Materials and Methods
URINE SAMPLES
We purchased pooled healthy human urine and 10 urine samples from healthy males from Bioreclamation and obtained clinical urine samples from 42 participants of the Atherosclerosis Risk in Communities (ARIC) study, for which a detailed description of sample selection and characteristics has been published (23, 24).
URINE SAMPLE PREPARATION
The sample preparation process is illustrated in Supplemental Fig. 1, which accompanies the online version of this article at http://www.clinchem.org/content/vol62/issue1. To prepare urine for MS analysis, samples stored at −80 °C were thawed, gently mixed, and centrifuged for 10 min at 10 000g at room temperature. Urine (5 μL) was supplemented with 3 μL of 1 mol/L NH4HCO3, 5 μL water, 2 μL of 1% Rapigest, 2 μL of 1000 fmol/μL stable isotope–labeled (SIL) peptides, and 0.16 μL of 0.5 μg/μL β-galactosidase, which was used as a QC probe to monitor the consistency of sample processing and analysis. Proteins were reduced with 1 μL of 100 mmol/L Tris(2-carboxyethyl)phosphine for 30 min at 60 °C, alkylated in the dark with 1 μL of 50 mmol/L iodoacetamide for 30 min at 37 °C, and incubated with 0.8 μL of 0.125 μg/μL trypsin (Promega Gold) in a 37 °C shaker for 6 h. Digested peptides were purified on an HLB microplate and resuspended in MS loading buffer.
MASS SPECTROMETRY
We performed SRM assays on an LC-MS system comprising a high-flow HPLC (Shimadzu Prominence) with an XBridge BEH 30 C18 reversed-phase column (Waters) linked to a triple-quadrupole mass spectrometer (Q-Trap 6500 or Q-Trap 5500, Sciex) with a TurboV ion source (Sciex). A detailed description of SRM LC-MS/MS methods and parameters is provided in the online Data Supplement. SRM data was processed with Multiquant (Sciex).
Data-dependent MS experiments for discovery were performed on an Orbitrap Elite MS (Thermo Scientific) coupled to an Easy-nLC 1000 chromatography system (Thermo Scientific) and a TripleTOF® 5600 MS (Sciex) coupled to an Ekspert nanoLC 415 chromatography system as described in the online Data Supplement. Data was processed through Sorcerer™ (Sage-N-Research), ProteinPilot™ (Sciex), or PASS (Integrated Analysis) software.
QUANTIFICATION OF UROMODULIN
We measured the absolute concentration of uromodulin with SIL peptides as internal standards and purified uromodulin (EMD Millipore) as an external standard, as described in online Supplemental Methods.
PEPTIDE SELECTION METHODS
For data-dependent LC-MS/MS, we analyzed a tryptic digest of purified uromodulin on an Orbitrap MS, in both higher-energy collisional dissociation and collision-induced dissociation fragmentation modes, and by triple-TOF MS. Proteome Discoverer was used to search MS spectra files and rank peptides. Peptides are commonly ranked by intensity and spectral counting. These methods can give different results, so they were compared. The database methods involved searching human proteome databases from NIST, PeptideAtlas, and SRMAtlas for uromodulin peptides. Predictions were obtained through the PeptideAtlas interface.
Results
COMMONLY USED SIGNATURE PEPTIDE SELECTION METHODS YIELD DIVERGENT RESULTS
The first major step in developing an SRM assay is to choose signature peptides for the quantitative analysis. In uromodulin 1, there are 27 predicted tryptic peptides with lengths in the useful range of 6–21 aa (Fig. 1A). From these, potential signature peptides were identified by data-dependent acquisition, database, and predictive methods. Remarkably, these methods yielded almost completely different results. No clear patterns emerge when comparing the top 10 uromodulin peptides selected with 12 different, but not entirely independent, peptide selection methods (Table 1). Urine matrix and the choice of algorithm for searching discovery data each had a profound influence on peptide ranking (see online Supplemental Table 1). There was, however, modest overlap in the ranking of transitions on the basis of fragment ion intensity (see online Supplemental Table 2). These results demonstrate that current peptide selection methods do not converge on a consistent set of recommended peptides and transitions for quantitative analysis.
(A), Amino acid sequence features of uromodulin 1. Candidate tryptic peptides of 6–21 aa include 2 signature peptides reporting the concentration of total uromodulin (yellow), 2 signature peptides that discriminate between uromodulin isoforms (dark blue), 3 peptides identified by data-dependent acquisition that were found to have nonlinear responses (orange), and 6 other peptides included in the correlation matrix (purple). Potential posttranslational modifications include N-linked glycosylation (red), disulfide bonds (light green), and methionine oxidation (pink). (B), r2 matrix. The schema at top presents structural features of the 4 uromodulin isoforms and identifies the location of 12 candidate peptides, which are identified by their first 5 amino acids. To empirically identify signature peptides that can accurately report the concentration of uromodulin protein, each peptide was individually compared with every other peptide for a total of 72 (12 × 12 ÷ 2) comparisons. For each peptide pair, a plot was constructed with SRM measurements from 9 urine samples. Values for the area under the curve for 1 peptide were plotted on the x axis, and values for the area under the curve for the other peptide were plotted on the y axis. A line was fitted to the 9 data points, and r2 was calculated and entered into the matrix. GPI, glycosylphosphatidylinisotol.
Comparison of SRM peptide selection methods.a
EMPIRICAL WORK FLOW FOR SRM PEPTIDE SELECTION
To identify the best signature peptides for quantifying uromodulin in urine, the first step was to eliminate peptides that were never detected by MS on any instrument, were not unique to uromodulin, or were located within a C-terminal region thought to be absent from the mature protein. Several peptides with methionine or cysteine residues were also eliminated, as these are susceptible to in vivo and in vitro modifications affecting their m/z ratios, were also eliminated. This process narrowed the original set of 27 theoretical peptides to 12 candidates for further testing (see online Supplemental Table 3).
We used a tryptic digest of purified uromodulin to identify a set of transitions for each peptide that had high and reproducible peak intensities on a triple-quadrupole mass spectrometer. The digest was then repeatedly injected to optimize the collision energy for each transition. The resulting parameters were used to investigate the performance of the 12 candidates in urine matrices. After establishing robust procedures for trypsin digestion and peptide cleanup, we evaluated each peptide in a set of tryptic digests of urine samples obtained from healthy individuals. For this initial analysis, raw area-under-the-peak measurements were compared without normalization.
Measured amounts of the uromodulin signature peptides used for quantifying uromodulin protein should be linearly related to the amount of input protein and the amount of other well-behaved signature peptides. To identify peptides with this property, we calculated the coefficient of determination (r2) for pairwise comparisons between each of the 12 candidate peptides across 9 urine samples (Fig. 1B). As expected, r2 values for pairs of transitions from the same peptide were always >0.998, indicating that any variations from true linearity were due to the effects of differences between individual urine samples on the overall detectability of specific peptides. In contrast, low correlations were observed between several pairs of peptides, indicating that ≥1 peptide in each of these pairs was not accurately reporting the protein concentration. The identity of the peptides with poor correlations could not have been predicted from SRM chromatograms, as all of the peptides had symmetrical and unambiguously quantifiable peaks, with no indication of interference, in all urine samples.
Notably, the peptides with the lowest correlations were highly accessible to trypsin digestion, suggesting that these peptides may be derived from regions of the protein that are sensitive to endogenous proteases that vary between individuals (see online Supplemental Fig. 2). Also, the poorly correlated SGSVIDQSR peptide, although routinely detected in urine and purified uromodulin, is thought to be located within a C-terminal propeptide associated with the glycosylphosphatidylinisotol anchor and may be absent from the mature protein.
From the r2 data, we selected a set of 4 signature peptides that were all highly correlated with each other. Two of these peptides, DWVSVVTPAR (DWVSV) and YFIIQDR (YFIIQ), were present in all uromodulin isoforms. The other 2, TLDEYWR (TLDEY) and FVGQGGAR (FVGQG), can discriminate between isoforms (Fig. 1 and online Supplemental Fig. 3). In making our selections, we also considered the total SRM signal intensity of each peptide, background noise, LC retention time, and peak shape (see online Supplemental Table 4). Four Met-containing peptides included in the empirical test had acceptable raw pairwise correlations but were excluded because the extent of Met oxidation was highly variable (see online Supplemental Fig. 4).
BUILDING A QUANTITATIVE SRM ASSAY
For absolute quantification, SIL peptide versions of the empirically selected uromodulin signature peptides were spiked into each trypsin digest and used to normalize the data. Our expectation was that the SIL peptides would behave similarly to natural peptides with the same sequence, such that any loss of natural peptides during sample processing would be accompanied by an equivalent loss of SIL peptides. Normalization was found to be remarkably effective in a test where peptide cleanup procedures were deliberately manipulated to alter peptide recovery (see online Supplemental Fig. 5). The SIL peptides were also used to further optimize the MS parameters (see online Supplemental Table 5).
On a standard curve constructed from a serial dilution of purified uromodulin, the SRM response for 12 abundant transitions representing the 4 signature uromodulin peptides was linear over ≥3 orders of magnitude (r2 ≥0.998) (see online Supplemental Fig. 6). The lower limits of quantification (LLOQs) were 0.4–14.1 μg/mL (Table 2). The upper limits of quantification for all transitions were >446.4 μg/mL, the highest concentration tested. At 446.4 μg/mL uromodulin, recoveries were nearly 100%, and CVs were <5%.
Lower and upper limits of quantification of selected uromodulin peptides.
REPRODUCIBILITY AND RECOVERY
To establish the reproducibility and recovery of the final method, we quantified uromodulin in pools of healthy and diseased urine. For reproducibility, 5 aliquots of each pooled sample were processed on 5 different days (25). The interassay, intraassay, and total CVs were 1%–13%, 1%–11%, and 5%–13%, respectively (see online Supplemental Table 6). For recovery, healthy and diseased urine pools were mixed at ratios of 5-μL healthy pool:15-μL diseased pool, 10-μL healthy pool:10-μL diseased pool, 15-μL healthy pool:5-μL diseased pool. Recoveries ranged from 83% to 118%, with a mean (SD) of 104% (6%) (see online Supplemental Table 7).
QUANTITATIVE SRM ASSAY YIELDS REPRODUCIBLE RESULTS COMPARABLE TO ELISA
We evaluated the quantitative SRM assay by measuring the uromodulin concentration in 42 urine samples that had been previously analyzed with ELISA (24). The absolute concentration for each peptide was calculated with reference to a standard curve prepared from data collected in the same sequence of MS runs. Three independent digests were prepared for each urine sample, and the SRM assay was run 3 times on each digest. Two urine samples were eliminated from further analysis: 1 had a uromodulin concentration below the LLOQ, and the other was enriched for uromodulin isoforms 1 and 4 over isoforms 2 and 3, as shown by relatively high amounts of the TLDEY and FVGQG peptides.
The results for the remaining 40 samples, acquired from a total of 360 MS runs, were internally consistent (Fig. 2). CVs comparing the 3 digests for each sample were typically <10%, and CVs comparing the 3 injections for each digest were typically <7%. CVs comparing peptide concentrations measured with different transitions were typically <10%, with a trend toward higher CVs for low-concentration peptides.
Four uromodulin peptides were quantified by SRM in 40 urine samples with SIL internal standards for normalization to a standard curve. For presentation, the samples are arranged according to the concentration of the DWVSV peptide. Absolute concentration (micrograms per milliliter) and reproducibility (CV) are compared between LC-MS injections (n = 3) for quantitation of the DWVSV-y7 transition in each digest (A), trypsin digests (n = 3) (B), and different SRM transitions (n = 2, 3, or 4) for the same peptide (C). See Table 2 for a list of transitions for each peptide.
Notably, the uromodulin concentration determined by SRM was greater than that determined by ELISA. This discrepancy could be due to (a) inconsistency in the documented concentration of the standards used for SRM and ELISA or (b) reduced antibody binding to endogenous uromodulin caused by interference from unknown matrix components or structural modifications (e.g., posttranslational modifications, proteolysis) within 1 of the uromodulin epitopes. In addition, the calculated concentration of the isoform-discriminatory FVGQG peptide was consistently higher than that of the other peptides, suggesting that the purified uromodulin calibrator had a different ratio of isoforms than the clinical samples or lacked an interfering contaminant common to all urine samples. Alternatively, the FVGQG peptide could have a different decay rate than the other peptides (26).
There were strong correlations (≥0.98) between the calculated concentrations of the 4 uromodulin signature peptides (Fig. 3). These results represent a substantial improvement over the >0.90 correlations for these peptides observed during the peptide selection phase. This improvement was achieved by normalizing to the SIL internal standards, thereby controlling for variations in peptide recovery. In contrast to the superior results for the empirically selected signature peptides, normalized data for 3 peptides that had been previously selected from data-dependent acquisition (Fig. 1) correlated poorly with each other (r2 = 0.28–0.70) and with the 4 empirically selected peptides (r2 = 0.38–0.74). Importantly, there was also a high correlation between the SRM data for the 4 empirically selected peptides and results from an ELISA assay that had been performed 2 years earlier on the same samples (Fig. 3). These results demonstrate that choosing signature peptides on the basis of experimental results generates more reliable SRM data.
Normalized SRM and ELISA data from 40 urine samples are presented as a correlation matrix.
Discussion
The accuracy of protein quantification by SRM, SWATH, and other MS techniques is completely dependent on the selection of appropriate surrogate peptides to represent the protein of interest. Empirically testing a plurality of candidate peptides to identify those with correlated MS signals makes it possible to select peptides that will generate robust data in the real world. Reliance on other popular methods can lead to confounding results, because unpredictable factors can interfere with accurate quantification.
USE OF A CORRELATION MATRIX TO IDENTIFY PROTEOTYPIC PEPTIDES
In principle, when a protein is completely digested into peptides, the derivative peptides should be present in equimolar amounts. Thus, if one complex biological sample has twice as much of a protein of interest as another, it should, after proteolysis, have twice as much of every derivative peptide. Consequently, in a set of unknown biological samples, the measured amounts of 2 peptides derived from the same protein should have a linear relationship regardless of the amount of protein in each sample. If the relationship deviates from linearity for any reason, at least one, and possibly both, of the peptides is not suitable for determining the concentration of the parent protein.
We propose an efficient work flow to select representative peptides for absolute MS quantification of a target protein (Fig. 4). The process begins by identifying the set of all potential peptides from an amino acid sequence that are within a detectible m/z range. If the goal of the experiment is to monitor a specific posttranslational modification, proteolytic cleavage, isoform, or mutation, peptides representing the desired feature must be retained. Otherwise, the initial set can be narrowed down by eliminating peptides that are not present in all forms of the protein to be quantified. Peptides subject to oxidation and other in vitro artifacts should also be eliminated, if possible.
Preliminary SRM assays are designed to target as many peptides as practically possible and then test them in biological samples representative of the milieu that will be used for quantitative assays. If the peptide is readily detected, these preliminary assays do not have to be fully optimized for MS performance or absolute quantification, and they can be developed with purified protein, enriched protein, or native biological samples. The goal is to quickly measure the relative amounts of each peptide in the full range of appropriate biological samples. An r2 value is calculated for each pair of peptides and then arranged in a matrix, making it possible to identify a subset of well-behaved peptides whose scores are all relatively correlated with each other. The final signature peptides can then be selected on the basis of practical criteria including signal strength and LC elution time.
There are many potential reasons for the measured amount of a peptide to vary from expectation. Differences in chemical composition, pH, or ionic strength of the biological matrix can influence proteolysis, peptide stability, aggregation, or ionization in an MS instrument (22). Oxidation and other artifactual chemical modifications can change the mass of a peptide and thereby interfere with MS detection. Peptide mass can also be affected by unknown posttranslational modifications or polymorphisms. In addition, background noise could arise from unknown components in the biological matrix. By following the proposed work flow, peptides with poor correlations can be readily identified with a correlation matrix and then expeditiously eliminated without actually determining precisely why they are unsuitable for quantification.
LIMITATIONS OF PREVIOUS PEPTIDE SELECTION METHODS
The most important concept arising from this work is that one cannot take shortcuts in peptide selection and expect to be rewarded with a robust assay. A variety of common peptide selection methods were tested and gave wildly inconsistent results. Notably, 14 different uromodulin peptides were ranked among the top 3 by at least one method (Table 1; online Supplemental Table 1), but none of these top-3 peptides were included in the empirically derived SRM assay (see online Supplemental Table 5). The most commonly recommended peptide, DSTIQVVENGESSQGR, with 6 different endorsements, had a low SRM signal and a relatively low correlation with other uromodulin peptides. Five other top-3 peptides, including 2 recommended by SRMAtlas, contained methionine residues, which can have a high degree of variability in the percentage of oxidation. Additionally, 2 top-3 peptides predicted by purely computational methods were not detected on any MS instruments.
COMPARING SRM AND ELISA ASSAYS
All 4 uromodulin peptides in our final assay yielded quantitative SRM results comparable to those obtained with the ELISA (Fig. 3). The correlation between different peptides measured by SRM was somewhat higher than the correlation with the ELISA data. This difference may arise because the same tryptic digests were used for all peptides in the SRM assay, whereas the ELISA was performed 2 years earlier (24).
SRM assays have several advantages over ELISA. Most importantly, ELISA is completely dependent on antibodies. It takes a long time to produce antibodies with sufficient affinity and specificity (27), and their corresponding epitopes may be suboptimal for quantification because of incomplete accessibility, interferences, or variation between protein forms (28). These concerns are magnified by the fact that epitopes are not even disclosed for the commercially available ELISA assays targeting uromodulin. Furthermore, SRM assays are more flexible than ELISA, as they can target multiple peptides, including ones that discriminate between isoforms and posttranslational modifications (21).
In conclusion, the empirical peptide selection work flow described here is useful to identify signature peptides for quantitative MS assays that are demonstrably free from unpredictable artifacts that could interfere with accurate and reproducible quantification.
Acknowledgments
The authors thank Zongming Fu and Ronald Holewinski for running samples on the Orbitrap ELITE and Triple-TOF 5600 LC MS/MS instruments.
Footnotes
↵† Q. Fu and E. Grote contributed equally.
↵5 Nonstandard abbreviations:
- SRM,
- selected reaction monitoring;
- MS,
- mass spectrometry;
- ARIC,
- Atherosclerosis Risk in Communities;
- SIL,
- stable isotope–labeled;
- LLOQ,
- lower limit of quantification.
Author Contributions: All authors confirmed they have contributed to the intellectual content of this paper and have met the following 3 requirements: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; and (c) final approval of the published article.
Authors' Disclosures or Potential Conflicts of Interest: Upon manuscript submission, all authors completed the author disclosure form. Disclosures and/or potential conflicts of interest:
Employment or Leadership: None declared.
Consultant or Advisory Role: None declared.
Stock Ownership: None declared.
Honoraria: None declared.
Research Funding: J.E. Van Eyk, National Heart, Lung, and Blood Institute, Johns Hopkins Proteomic Innovation Center in Heart Failure (HHSN268201000032C), Chronic Kidney Disease Biomarker Consortium funded by the National Insitute for Diabetes and Digestive and Kidney Diseases (U01-U01DK085689).
Expert Testimony: None declared.
Patents: Q. Fu, E. Grote, and J.E. Van Eyk (US Provisional Application).
Role of Sponsor: The funding organizations played a direct role in the design of study, review and interpretation of data, and preparation and approval of manuscript.
- Received for publication April 23, 2015.
- Accepted for publication November 3, 2015.
- © 2015 American Association for Clinical Chemistry
References
In this issue
