Measurement by a Novel LC-MS/MS Methodology Reveals Similar Serum Concentrations of Vitamin D Binding Protein in Blacks and Whites

Materials

Stable isotope-labeled internal standard peptides, labeled at lysine [(¹³C)₆H₁₂(¹⁵N)₂O, Δm = +8 amu] residues and their unlabeled counterparts were synthesized by New England Peptides (sequences are listed in Table 1). Other materials are listed in Supplemental Methods.

Human samples

Samples from the community-based ARIC Study were collected at multiple sites in the United States under protocols approved by institutional review boards at each site. All study participants provided written informed consent. The serum samples utilized here were obtained from 200 participants who attended the ARIC visit 5 clinical exam (2011–2013) and who returned for a repeat quality control visit, approximately 4–8 weeks later. Race was self-reported. ARIC phlebotomists were centrally trained and certified, and specimens were processed and stored using standardized procedures. Of the 200 participants, a total of 10 people were excluded from the analysis of correlations between immunoassay and LC-MS/MS as well as the calculation of kappa based on having a race with low sample numbers (Asians; N=2), missing genetic data from the parent study (N=6), or having non-matching genotype data from the parent study for rs4588, which was measured using two different platforms (N=2). There were 3 participants excluded from the comparison between immunoassay and LC-MS/MS results due to highly discrepant measurements (>100 µg/mL) between the first blood draw and the second blood draw as determined by immunoassay (N=187 for comparison of LC-MS/MS with immunoassay). For the evaluation of isoform identification by LC-MS/MS vs. genotyping, there were 3 samples that did not have a sample available from the second blood draw for LC-MS/MS (N=187 for comparison of LC-MS/MS and genotyping). Leftover clinical samples were used in assay validation in accordance with guidelines from the Human Subjects Division at the University of Washington.

Genotype and data from ARIC

Single nucleotide polymorphisms rs7041 and rs4588 are located on chromosome 4 in the GC gene. Genotypes for rs7041 and rs4588 were obtained from the ITMAT-Broad-CARe Chip, a custom 50K SNP genotyping array. Quality control procedures have been previously published (27). The genotype at the rs4588 SNP was also determined using the Affymetrix Genome-wide Human SNP array 6.0 chip, which interrogates 906,600 SNPs. Quality control measures (28, 29), and methods for imputation to ~2.5 million SNPs in whites (29) and blacks (30) have been published. All genotyping was performed at the Broad Institute of the Massachusetts Institute of Technology and Harvard. Haplotypes are as follows: Gc1f = rs7041-T (Asp), rs4588-C (Thr); Gc1s = rs7041-G (Glu), rs4588-C (Thr), Gc2 = rs7041-T (Asp), rs4588-A (Lys).

Peptide Selection

The peptides used for quantification and identification of the three most predominant haplotypes of VDBG in human plasma (Table 1) were based on sequences obtained from the NCBI Protein Database [(Gc2 (GI:139641); Gc1f (GI:324021745); Gc1s (GI:119626050)] (22). Each of the sequences were imported into Skyline (version 2.5.1.5962) to generate in silico tryptic peptide precursor ion isolation lists and collision energies for Xcalibur (Q Exactive Series, version 2.3) to detect peptides using tandem mass spectrometry (Q Exactive Plus, Thermo) (31). Criteria used for the selection of peptides are described in Results.

Sample preparation

The details for the preparation of samples are provided in Supplemental Methods. Briefly, samples were denatured with trifluoroethanol, reduced with dithiothreitol, and alkylated with iodoacetamide. After dilution of the trifluoethanol, proteins were proteolyzed with trypsin for 30 min with agitation. Samples were further diluted prior to being injected onto an Acquity UPLC connected to a Xevo TQMS triple quadrupole mass spectrometer (Waters, MA).

Liquid Chromatography and Tandem Mass Spectrometry

The details of the chromatographic and mass spectrometric conditions for nanoflow and normal flow chromatography are available in the Supplemental Methods. The transitions monitored and example chromatograms for targeted methods at normal flow rates are provided in Supplemental Tables 2 and 3 and Supplemental Figure 1, respectively.

Calibration, Quality Control, and Quality Assurance

Working calibrators consisted of pooled human serum containing 246 µg/mL VDBG, which was used undiluted or diluted with chicken serum (Equitech-Bio, Inc., TX) to 185 µg/mL, 123 µg/mL and 62 µg/mL. Working calibrator concentrations were assigned by assaying the undiluted serum pool against 4 reference calibrators containing recombinant VDBG (genotype Gc1f, Novoprotein, NJ) gravimetrically spiked into chicken serum at the following concentrations: 475, 238, 119, and 59 µg/mL. Quantitative amino acid analysis (Molecular Structure Facility; University of California, Davis) was utilized to determine the concentration of each amino acid present in a hydrolyzed sample of the stock calibrator solution. This information was used to establish the concentration of VDBG in the stock solution that was then spiked to make the reference calibrators. For calibrators, patient samples, and quality control materials, the measured concentration of peptides VLEPTLK and ELPEHTVK were averaged to determine the final concentration of the protein. Two single-person human serum samples were used as quality control materials, which were aliquoted and kept frozen at −20°C. Batches were repeated when one quality control sample was more than 3SD away from the mean or when both quality control samples were more than 2SD away from the mean. Clinically relevant potential interferences were evaluated using the Assurance Interference test kit (Sun Diagnostics, ME) as previously described (32). Quality control parameters were evaluated for each sample: (1) the 3 ratios of peptides VLEPTLK, ELPEHTVK and THLPEVFLSK to one another, (2) internal standard peak areas, and (3) transition ion ratios for each of the peptides. Due to the low abundance of the secondary (qualifying) ions for several of the peptides, the final method uses the three peptide ratios and internal standard peak areas to monitor quality. Samples failed quality control if two internal standard peak areas were more than 2 SD from the mean or if two peptide ratios fell outside 2 SD from the mean. In this study, 8 samples failed quality control. On repeat, two samples still failed quality control and were not included in the analysis.

Isoform identification

Three peptides (sequence) were used to distinguish between Gc1s (LPEATPTELAK), Gc1f (LPDATPTELAK), and Gc2 (LPDATPK) isoforms. To normalize for VDBG concentration and digestion variability, the peak area ratio of each isoform-specific peptide was divided by the peak area ratio of one of the quantifying peptides (VLEPTLK), where the peak area ratio for each peptide is defined as the chromatographic peak area of the endogenous peptide divided by the chromatographic peak area of its corresponding stable isotope-labeled internal standard peptide. Comparison of the genotyping data from the ARIC cohort with the data from the LC-MS/MS method was used to empirically establish cut-offs for each peptide using the first blood draw. The cut-offs were optimized for specificity and therefore set at the lowest peptide ratio above which the isoform was always correctly identified as present in samples from the first blood draw (0.014, 0.082, and 0.10 for Gc1s, G1f, and Gc2, respectively; Supplemental Figure 2, raw data presented in Supplemental Material Appendix). The method was then evaluated for accuracy using the data acquired for the samples from the second blood draw. We did not attempt to assign the relative amount of each isoform in each sample, which will not be possible without the analysis of pure protein isoform calibrators. The threonines in the sequence ATPTEL are glycosylated in vivo (25). Deglycosylation at the protein and peptide level with Oglycosidase (New England Biolabs, MA) did not increase the peak areas observed for the Gc1f and Gc2 peptides but did double the peak area for the Gc1s peptide when deglycosylated after proteolysis (data not shown).

Statistical Analysis

Descriptive statistics are provided using means and proportions. T-tests and single-factor ANOVA were used to evaluate differences between means. Percent variation in VDBG explained by the haplotypes was calculated as R² from the following model [VDBG] = β₀ + β₁*haplotype. We tested whether the association between haplotype and VDBG levels varied significantly by VDBG measurement method using a generalized estimating equation (GEE) implemented in SAS PROC GENMOD, with repeated ID. Specifically, each person in the dataset had 2 observations (1 for ELISA and 1 for LCMS); ‘method’ was a dichotomous variable. The following model was specified: [VDBG] = β₀ + β₁*Method + β₂*Haplotype + β₃*Method*Haplotype.

Assay development

To identify the peptides used for the quantification of VDBG, recombinant protein was digested with trypsin and analyzed using nanoflow chromatography and targeted tandem mass spectrometry on a high resolution mass spectrometer. All tryptic peptides between 6 and 25 amino acids in length were targeted (N=26), representing 68% coverage of the protein. Of these peptides, many were excluded: 12 contained non-ideal amino acids, 1 had poor precision (>20%), 3 were not stable to freeze-thaw of the digest, 2 had relatively low signal intensity in human plasma, and 1 had multiple potentially interfering peaks in a digest of human plasma (Figure 1). From the remaining peptides, stable isotope-labeled internal standard peptides were synthesized for the top 4 most intense peptides (by peak area of the most abundant fragment), of which 2 were precise enough for use in protein quantification. A third peptide was monitored for quality control/assurance. The peptides used for isoform identification were the tryptic peptides surrounding the polymorphic amino acids. One peptide with a missed cleavage was tested as well, but was never detectable in human plasma. Mass spectrometric parameters for the normal flow method were determined empirically and compared with those predicted from Skyline (Supplemental Table 4). Improvement in signal intensity (peak area of the longest fragment for each peptide) was 1- to 13-fold with empirical optimization, which on average was a higher improvement than that seen in previous studies (33, 34). Time course analysis of digestion suggested that 30 min was optimal and that internal standard peptides should be added prior to digestion (Supplemental Figure 3) (35).

Validation of the quantitative assay

For the quantification of VDBG, assay validation fit within the recommended guidelines that we recently proposed (32). Raw data from the validation experiments are available as a Skyline file in Supplemental Material. The total imprecision of the assay was determined by 5×5 analysis to be 7.3–9.0% CV (at 234 and 266 µg/mL, respectively), which includes contributions due to within-day (4.9–6.7% CV) and between-day variability (5.4–6.0% CV). Peptide degradation during digestion was determined to introduce a 37% negative bias, which was overcome by adding internal standard peptide prior to digestion (35) and by including an external calibration scheme of normal human sample diluted into chicken serum. The limit of detection was determined as the mean plus 10 times the signal-to-noise observed for the measurement of fetal bovine serum (N=23) and was 5.4 µg/mL. Linearity and the lower limit of quantification were determined by mixing studies with two normal human samples and by dilution of a normal human sample with chicken serum. The assay was linear from 62 to 434 µg/mL (Supplemental Figure 4). The lower limit of quantification was 71 µg/mL using interpolation (<20% CV), which was validated at 109 µg/mL (14.2 %CV). Hemoglobin (≥500 mg/dL) and triglycerides (≥3,083 mg/dL) were demonstrated to interfere with protein quantification, but airfuging eliminated the interference from triglycerides (Supplemental Table 5). Spike-recovery was performed by spiking pooled purified VDBG (Sigma-Aldrich, MO) at 108 µg/mL (N=28) or 318 µg/mL (N=14) into random serum samples. Recovery was 104±17% (range 76–141%, expected range based on the variability of the assay: 39–161%) and 103±12% (range 78–123%, expected range: 67–133%), respectively. Protein was stable to 2 freeze-thaw cycles (+1.9% bias). Incubation at room temperature for 48 h (+6.9% bias, 3.4% CV) and refrigerated for 48 h (+8.9% bias, 4.2% CV) showed a slightly positive bias. Evaluating chromatographic peak area, pure internal standard peptides were stable in the autosampler for 24 h in a 96-well plate (≤16% loss), except peptide THLPEVFLSK (26% loss).

Evaluation of the assay’s ability to identify isoforms

The identification of isoforms was based on the presence of isoform-specific peptides. To control for the variability of VDBG concentration and the extent of digestion across samples, the isoform-specific peptide signal was normalized to the signal of a quantitative peptide (common across all three isoforms). To evaluate the ability of the assay to reliably identify the haplotype of participants, we took advantage of the genotypic data available for the ARIC cohort. The accuracy of the new assay to identify the presence of the three haplotypes was 124/126 (98.4%), 85/86 (98.8%), and 58/58 (100%) and the absence of the three haplotypes was 59/61 (96.7%), 101/101 (100%), and 128/129 (99.2%) for Gc1s, Gc1f, and Gc2, respectively (raw data presented in Supplemental Material Appendix). The kappa coefficient (95% confidence interval) for agreement of the genotype of each individual (i.e., both haplotypes) between the two methods was 0.97 (0.95–1.00).

Method comparison with R&D Systems immunoassay

Given recent questions about whether the R&D Systems immunoassay binds to VDBG differentially by genotype (26), we performed a method comparison in a subset of ARIC participants who at clinic visit 5 (2011–2013) had undergone a repeat blood-draw. In this group, the mean age was 76.2 y, 62% were female, and 28% were black. Mean (SD) VDBG concentrations were higher when measured with LC-MS/MS compared with the immunoassay [264.5 (32.4) vs. 183.9 (69.7) µg/mL, p<0.0001]. Figure 2, Supplemental Figure 5, and Table 2 demonstrate that the influence of the haplotype on VDBG levels varied by VDBG measurement method. The association between haplotype and VDBG levels varied significantly based on whether VDBG was measured by LC-MS/MS or immunoassay (p_interaction = <0.0001). Haplotypes explained 81% of the variability in immunoassay-measured VDBG concentrations, but only 12% of the variability in LC-MS/MS-measured concentrations.

Racial distributions

In both blacks and whites, VDBG concentrations were normally distributed when measured by LC-MS/MS but were skewed or bimodal when measured by the R&D immunoassay (Figure 3). Mean (SD) levels of VDBG differed by race when measured by the immunoassay [133 (57) vs. 203 (64) µg/mL for blacks and whites, respectively; p<0.0001]. In contrast to immunoassay, when measured by LC-MS/MS, VDBG levels did not differ by race [262 (25) vs. 266 (35) µg/mL, respectively; p=0.43].

Potential isoform-specific bias in the mass spectrometric assay

It is possible that the Gc1s, Gc1f, and Gc2 polymorphisms cause variability in the digestion of VDBG and variable liberation of the quantification peptides. Ruling this possibility out completely would require purified protein of each isoform. Lacking the availability of these important reagents, we investigated the crystal structure of VDBG and found that the quantifying peptides were >25Å away from the polymorphic amino acid residues (Supplemental Figure 6). We also examined the ratio of the two quantification peptides across the ARIC cohort and found no difference between the genotypes (p=0.75, single-factor ANOVA, Supplemental Table 6), suggesting that the peptides were liberated similarly in all six genotypes.

Validation per new guidelines

Protein biomarkers are an essential component of patient care. The identification of putative novel biomarkers happens almost daily, but the translation of initial observations from discovery experiments into the clinical care of patients happens only rarely. The fact that most preliminary observations from federally-funded investigations are not repeatable can, at least in part, be traceable back to poorly validated assays, which may be preventable. In the validation of the assay presented herein, we have followed the recommendations that we recently published (32). The fundamental characteristics of a reliable assay, namely precision, linearity, specificity, and stability can be assessed with a relatively limited number of injections. It is hopeful that this proof-of-principle application of the suggested guidelines will be helpful for the validation of other biomarkers as they make their way through the biomarker discovery pipeline (36).

Need for purified isoforms

Our novel assay reports the concentration of VDBG and the haplotypes detectable in plasma. For the calculation of free and bioavailable vitamin D in heterozygotes, it would be ideal to use the actual concentrations of each different isoform present in plasma. While we present data that the quantifying peptides from VDBG are at a distance from the polymorphic amino acids in the protein and that they appear to be liberated at the same rate during trypsin proteolysis, it is not yet clear that the polymorphic peptides are released at the same rate. To evaluate the rate of peptide liberation in each haplotype and thus validate the ability to measure the concentration of each haplotype in plasma, purified VDBG protein from homozygotes will be required (37).

Estimates of free and bioavailable vitamin D metabolites

Vitamin D deficiency is associated with poor bone health and increased incidence of many chronic diseases. Using a standard cut-off of 20 ng/mL for serum 25(OH)D, a huge proportion of the black population (81%) is vitamin D deficient (19). However, studies recurrently demonstrate that low 25(OH)D is a stronger risk factor for adverse bone and cardiovascular outcomes in whites than blacks (6, 19–21). These and other studies suggest that alternative markers of vitamin D metabolism may be more important in predicting pathophysiology compared with 25(OH)D. For example, it is now widely recognized that localized, tissue-specific conversion of 25(OH)D to active 1,25-dihyroxyvitamin D is likely taking place and that extra-renal tissues likely acquire 25(OH)D that is not bound to VDBG (38). In other words, free or bioavailable vitamin D metabolites (<15% of the total) may be more important in predicting systemic vitamin D sufficiency.

Epidemiological investigations of free and bioavailable vitamin D are just beginning. For example, using a polyclonal ELISA for VDBG, free and bioavailable 25(OH)D were more strongly correlated with bone mineral density than total 25(OH)D (39), but VDBG was not associated with heart failure (40). Importantly, a study using the R&D Systems monoclonal immunoassay for VDBG in a community setting revealed that blacks and whites have comparable concentrations of bioavailable 25(OH)D (22). The results presented here suggest that a monoclonal immunoassay that is sensitive to protein polymorphisms may generate misleading concentrations of bioavailable vitamin D metabolites. However, as mentioned above, it will be important to continue to study the impact of polymorphisms on protein quantification by LC-MS/MS. Of note, the results of LC-MS/MS are in contrast to those from certain immunoassays and suggest that blacks and whites have similar plasma concentrations of VDBG.

Targeted mass spectrometric assays in the biomarker discovery pipeline

High-throughput automated immunoassays provide timely protein biomarker results for patient care. Their central role in the clinical laboratory has led to the proposal that the biomarker discovery pipeline should end in a validated, FDA-approved immunoassay for distribution to clinical laboratories (36). Unfortunately, the initial steps of the biomarker pipeline generally utilize platforms other than those present in the clinical laboratory, and the results of pre-clinical studies are therefore not always relevant to clinical practice. The sensitivity and specificity of protein assays by mass spectrometry, their use in earlier stages of biomarker discovery, and the availability of the necessary instrumentation in a growing number of clinical laboratories, makes protein assays by mass spectrometry promising tools in the validation of novel biomarkers, including VDBG and bioavailable vitamin D metabolites.