Choice of Peptide Standards.

The peptides that constitute the QconCAT were previously selected on the basis of experimental design followed by theoretical assessment. This approach was limited by options in a data-dependent experiment yielding one to two peptides per UGT (Russell et al., 2013). Extensive sequence homology between UGTs also contributed to this limitation. Initially, peptides that provided higher abundance were used in line with widely accepted literature (Brownridge et al., 2011; Lawless et al., 2016). Instead, we propose adoption of a better appraisal of quantification that uses as a basis several peptides for each UGT. In this work, assessment of suitability of peptides was carried out on the basis of theoretical appraisal by two independent analysts and in silico evaluation using CONSeQuence algorithm (Eyers et al., 2011). Quantification was subsequently considered on the basis of the more favorable peptide choice (Table 1). Theoretical criteria for peptide assessment are included in Supplemental Methods. Briefly, selected peptides should have unique sequences, and mass-to-charge ratios (of parent and fragment), should not be mapped to membrane-associated domains or subject to polymorphisms or biologic (post-translational) modifications, should be readily cleavable, of suitable length (6–20 amino acids), and have favorable stability (to chemical modification owing to handling/storage) and moderate hydrophobicity (Kamiie et al., 2008; Carroll et al., 2011). Choice of monitored peptide charge state (z = +2 or +3) was also considered.

Choice of Peptide Fragments.

Fragment selection was initially conducted in silico using Skyline 1.4, with fragment ratios expected to return consistent quantification. Initially, two to three transitions per peptide were monitored with mean ratios being used for quantification. In this work, the uniqueness of fragment sequences and consistency between estimates on the basis of monitored transitions were assessed, especially for low abundance proteins; less-specific fragments returning inconsistent ratios were excluded from analysis. In addition, the chromatographic trace of different fragment ions was assessed and poor quality signals were excluded. Since retention time and m/z values monitored had relatively large filters in the LC-MS/MS assay, the m/z values for the selected fragments were assessed, with isobaric and isomeric fragments being excluded (Table 1).

Correction for Efficiency of Label Incorporation.

Assessment of efficiency of ¹³C-label incorporation into QconCAT protein synthesis was previously reported as an in-house quality control step in QconCAT expression; constructs of ≥95% purity are accepted as quantitative standards (Achour et al., 2015). The level of incorporation can vary batch-to-batch, and impurity is expected to affect quantification, especially of low abundance proteins (Carr et al., 2014). Uncorrected and corrected quantitative ratios were generated using eqs. 1 and 2, respectively (Fig. 2, A and B):(1)(2)where and are uncorrected and corrected ratios, respectively, used to quantify a peptide representing a UGT enzyme or the QconCAT (using a QconCAT-based internal standard); is the intensity of the light peptide signal measured in the quantitative experiment; is the peak intensity of the light peptide originating from the QconCAT measured in quality control experiments; and is the signal intensity of the heavy QconCAT peptide measured in the quantitative experiment.

Normalization of Abundance Measurements.

Normalization is commonly applied relative to protein mass, leading to abundance levels expressed in units of picomoles per milligram of fractional protein. Protein mass measurement is normally done by a colorimetric assay, generating data that may not be reproducible. Commonly used assays include: Bradford assay (Bradford, 1976), bicinchoninic acid (BCA) assay (Smith et al., 1985), and tryptophan fluorescence assay (Wiśniewski and Gaugaz, 2015). Abundance and activity were measured by independent laboratories and different protein assays were used (Bradford and BCA assays), demonstrating differences in reported contents for matched samples (Fig. 2C). In this study, a proposed approach to resolving this issue was to normalize abundance and activity data using protein levels measured by the same assay (BCA assay), using eq. 3.(3)where is target enzyme abundance (expressed in picomoles per milligram of HLM protein); represents the concentration of the unlabeled internal standard used to quantify the QconCAT [a Glu¹-fibrinopeptide B analog, modified to reduce the incidence of missed cleavage owing to the glutamate at the N-terminus (Lawless and Hubbard, 2012)]; is a volume correction factor relating the analyzed volume to the volume of HLM sample; and is the protein content determined for each sample (BCA assay). The terms assessed in this report were: target peptide and QconCAT-based–internal-standard ratios (parent-fragment selection and correction for label incorporation) and protein mass used for normalization (Fig. 2D).

Statistical Assessment of the Optimization Process.

Correlations were assessed at each stage using Spearman correlation test (Rs) and scatter of data was assessed with linear regression (R²). The following criteria were used: α-value of 0.05 (Bonferroni-corrected for correlation matrices), strong correlation (Rs > 0.50) and limited scatter (R² > 0.30), taking into account the effect of abundance/activity units, as previously advocated (Achour et al., 2017a).

Effects of the Choice of Monitored Peptides/Fragments.

Mass spectrometry-based proteomic strategies rely on using peptides as surrogates for proteins, and the limitations of this approach are still being uncovered. In eukaryotes, protein truncation and splice variants can result in misleading measurements, and technical issues include signal overlap and variable peptide responses between runs. QconCAT design normally follows a pragmatic approach, with two or more peptides included for each protein (Pratt et al., 2006); however, only one peptide is ultimately used for quantification. Two peptides representing each UGT were therefore included in the QconCAT whenever possible. Table 1 shows sequences used in data acquisition, scored on the basis of theoretical and in silico criteria; Supplemental Fig. 1 shows peptide elution profiles. Theoretical assessment of peptide suitability involves consideration of several parameters, which tend to be prioritized somewhat subjectively, and was therefore conducted by two independent analysts; the analysts’ scores were in agreement and compared well with in silico assessment. Three pairs of peptides (representing UGTs 1A1, 1A4, and 2B7) returned consistent quantification (Supplemental Fig. 2), whereas three pairs (UGTs 1A6, 2B4, and 2B15) showed significant differences.

When two or more peptides are used to quantify a protein, preference has conventionally been given to peptides that return higher concentrations (Brownridge et al., 2011; Lawless et al., 2016); the assumption is that underestimation can occur owing to differences in efficiency of release of peptides. It has recently become clear that several peptide-related factors can affect measurements and should therefore be considered when peptide choice is made, ideally in the design stage, although this is not always possible, especially in global proteomics. The strategy of monitoring the “highest reporter” will probably work well when soluble proteins are quantified; however, for membrane-bound proteins, especially those with unknown structures (e.g., UGTs), a more secure strategy is to monitor “best performer” peptides. Discerning best performers will almost always involve visual examination of MS and MS/MS spectra. For example, isobaric sequences representing UGT1A6 (WIYGVSK) and UGT2B7 (ADVWLIR) overlapped on the chromatogram, leading to overestimation of abundance, especially for UGT1A6. Discrepancies in abundance in relation to peptide choice have previously been reported to reach up to 30-fold in the case of P-glycoprotein in hepatocytes (Chen et al., 2017), and therefore applying suitable criteria is essential to ensure quality of surrogate peptides (Bhatt and Prasad, 2018). Such criteria were previously used in a label-free experiment on matched liver samples, which resulted in good agreement with QconCAT measurements of several UGT enzymes (Achour et al., 2017a).

Because targeted quantification uses MS/MS data as a basis, the properties of fragments are as important as those of parent peptides. Carr et al. (2014) recommended monitoring three to five fragments, allowing inconsistencies in measurements to be reconciled. Fragments are typically selected in silico on the basis of predicted peak intensities rather than sequence properties (Carr et al., 2014). There are several potential pitfalls with this approach. First, MS/MS spectra may have several low intensity peaks (especially with proline-containing peptides), so that a lower number of consistent measurements can be made. Two peptides, YLSIPTVFFLR (UGT1A3) and ESSFDAVFLDPFDNCGLIVAK (UGT1A9), were subject to this error, when peaks of lower quality returned inconsistent ratios. Although optimizing transition selection resulted in improved correlation with activity, quantitative methods for these two enzymes still require improvement. Additionally, erroneous quantification can occur when uniqueness of peptide-to-fragment m/z values cannot be ensured, most often in complex biologic mixtures, with short sequences being most affected (Carr et al., 2014). The MRM filters (retention time, parent ion m/z, and fragment ion m/z) are normally sufficient to ensure exclusive selection of fragments, but not always, especially in nonscheduled experiments, where retention times are not specified. Peptide sequences WIYGVSK (UGT1A6), ADVWLIR (UGT2B7), and WIYGVSK (UGT2B15) were eventually excluded from analysis for isobaric interference despite returning levels higher than their alternatives. The combined effect of peptide and fragment selection was shown to be substantial (Supplemental Table 1), with abundance values changing 0.5- to 3-fold upon reassessment. This led to improved abundance-activity correlation for most UGT enzymes. The initially weak correlation for UGTs 1A4, 1A6, and 2B15 (Rs = 0.19–0.35, P > 0.05; R² = 0.01–0.14) became moderate (Rs = 0.52–0.56, P < 0.01; R² = 0.24–0.31). For other UGTs (1A1, 1A3, 2B7), this improved mainly in terms of data scatter (from R² = 0.30 to R² = 0.47–0.57), although correlation for UGT1A9 remained weak (Rs = 0.20, P = 0.33, R² = 0.07).

Effects of Correction for the Quality of Isotopically Labeled Standards.

QconCATs are artificial proteins expressed in-house, and the extent of labeling varies, depending on the construct and culture conditions, from ∼95–99% (Achour et al., 2015). This means that QconCATs can contribute unlabeled peptide that may affect quantification when the analyte is expressed at low levels (Carr et al., 2014). Table 1 shows the extent of label incorporation into each peptide, reflecting more efficient labeling of arginines than lysines (Russell et al., 2013). The outcome of correction for purity was that levels of UGTs had a variable artifactual component of up to 10%, with little effect on correlation with activity. A similar trend was observed with efflux transporters BCRP and MRP2, quantified in human jejunum using QconCAT methodology (Harwood et al., 2015), which were overestimated on average by 10% and 7%, respectively, when corrections for purity were not considered (Harwood et al., 2016a). However, these errors do not always lead to meaningful differences in pharmacokinetic outcomes (Harwood et al., 2016b).

Effects of Normalization of Abundances.

Normalization relies on protein content determination using colorimetric assays, which are prone to interference from reagents commonly used in routine sample processing, including detergents, chaotropes, and reducing agents (Wiśniewski and Gaugaz, 2015). In particular, BCA assay is incompatible with commonly used concentrations of urea and dithiothreitol (Smith et al., 1985), whereas Bradford reagent tends to interact with sodium dodecyl sulfate (Bradford, 1976), a detergent used for gel-based sample preparation. A limitation of the cross-laboratory study (Achour et al., 2017b) was the use of Bradford assay with QconCAT measurements, whereas both activity and SIL-based measurements were normalized using BCA assay. Comparing the two protein measurements (Fig. 2C) indicates that, although average protein content in the samples was similar, there was no correlation between individual values. Normalization against the same protein content resulted in changes in enzyme levels reaching up to 50%, with improved correlation with activity, mainly in terms of scatter for six UGTs (from R² = 0.24–0.57 to R² = 0.48–0.73), and UGT1A9 showed substantial improvement to moderate correlation (Rs = 0.47, P = 0.02; R² = 0.34). It is not clear, however, whether BCA-normalization leads to better results; it may just lead to more consistent error. We have recently illustrated that normalization relative to tissue mass instead of fractional protein introduces less artifactual variability to end-point measurements (Achour et al., 2017a).

Overall Effects of Systematic Reassessment on UGT Measurements.

Specific effects of the assessed factors on abundance of each enzyme are described in Supplemental Results. Collectively, the proposed strategy led to 0.5- to 3.3-fold change in UGT levels, with substantial improvement in correlation with activity (Fig. 3) and tighter levels of interindividual variability in abundance (26%–86%), matching variability in activity (27%–67%), in line with recent literature (Margaillan et al., 2015). In addition, cross-laboratory comparison of UGT abundances seemed to indicate overall agreement, returning generally interchangeable abundance values (Supplemental Figs. 3 and 4). On the basis of these considerations, a list of established UGT-specific methods was generated (Supplemental Table 4).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Correlation between individual UGT enzyme abundances and activity rates (n = 24) using the original dataset (A) and the reassessed data on the basis of the proposed strategy (B). Moderate to strong, statistically significant correlations are shown in blue and weak correlations in gray. Units of abundance-measurement are picomoles per milligram of HLM protein, and units of catalytic activity are nanomoles (glucuronide) per minute per milligram of HLM protein. Substrates used for activity measurement are: β-estradiol (UGT1A1), chenodeoxycholic acid (UGT1A3), trifluoperazine (UGT1A4), 5-hydroxytryptophol (UGT1A6), propofol (UGT1A9), zidovudine (UGT2B7), S-oxazepam (UGT2B15). Rs, Spearman correlation coefficient. Dashed lines represent lines of regression.

The correlation matrix (Fig. 4) confirms specificity of protein and activity data, with no evidence of spurious abundance-activity relationships. A similar complementary approach was used previously to discern tissue-specific glycolytic and gluconeogenic pathways (Wiśniewski et al., 2015). Expression intercorrelations uncovered in the UGT dataset (Supplemental Table 3) were also in line with literature (Achour et al., 2014a; Margaillan et al., 2015). Correlations of enzyme expression have recently been adopted for more realistic model-based predictions of drug clearance and drug-drug interactions (Barter et al., 2010; Doki et al., 2018), with additional established correlations making their way into commonly used platforms, such as Simcyp.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Correlation matrix of QconCAT-derived individual UGT enzyme abundances (n = 24) and activity rates (abundance vs. activity). Strong, statistically significant correlations are shown in blue. Units of abundance-measurement are picomoles per milligram of HLM protein, and units of catalytic activity are nanomoles (glucuronide) per minute per milligram of HLM protein. AZT, zidovudine; CDCA, chenodeoxycholic acid; EST, β-estradiol; 5HTOL, 5-hydroxytryptophol; OXAZ, S-oxazepam; PRO, propofol; TFP, trifluoperazine. Supplemental Table 2 shows the statistical analysis used to generate the abundance-activity correlation matrix.

Conclusions.

The QconCAT approach offers several advantages (Al Feteisi et al., 2015a), and therefore it has recently been adopted for various clinical and biologic applications (Dzieciatkowska et al., 2015; Wang et al., 2015; Kito et al., 2016). Although initial applications were primarily aimed to quantify soluble proteins, often in simple organisms, it is clear that QconCAT can be applied to membrane-bound mammalian proteins. However, monitored peptides and fragments need to be chosen carefully, preferably using a priori selection, and corrections are required for relatively low purity standard peptides targeted at low abundance proteins. We continue to advocate using tissue mass for abundance normalization and activity data for quality control. We propose optimized QconCAT methods for the quantification of several UGTs (1A1, 1A6, 2B4, 2B7, and 2B15) and a robust data analysis strategy for targeted proteomic quantification, particularly applicable for QconCAT-based measurements.