Optimized Approaches for Quantification of Drug Transporters in Tissues and Cells by MRM Proteomics

Surrogate Peptide Selection

Selection of surrogate peptides for a protein is the first and most critical step in MRM proteomics. Until few years back, the surrogate peptide designing was based on experimental approach involving tedious protein fractionation by SDS-PAGE gel electrophoresis followed by capillary LC nano-electrospray ionization quadruple time-of-flight (nano-ESI-Q-TOF) or Orbitrap MS instruments [23, 24, 35]. While this approach is accurate, it is a laborious, cost-intensive procedure, and requires a protein standard with reasonable purity and amount. To overcome these shortcomings, in silico approach, first adopted by Kamiie et al. [27] for drug transporters, is preferred these days. The uniqueness, LC retention, MS response, peptide stability, quantitative reliability, and signal quality are the important aspects that are considered for surrogate peptide selection. While integrated software tools, e.g., SRMAtlas, MRMaid, or GPMDB (see below), are available for surrogate peptide design, these tools do not cover all the proteins across species. Therefore, we present below a basic generalized approach starting from finding a protein sequence to selecting a quantitatively reliable surrogate peptide for any transporter protein of interest.

1. Protein sequence: For targeted quantification of a transporter, first a full protein sequence is obtained from the databases such as NCBI, Ensembl, UniprotKB, or Protein knowledgebase (Table II). For example, the sequence for human P-gp (MDR1) was obtained from UniprotKB (Fig. 3) by putting protein name and species in the search box. The protein accession number for the transporter (e.g., P08183 (Uniprot) or ENSP00000265724 (Ensembl) for P-gp) can be obtained in this step for subsequent use.

2. In silico peptide digestion: For any protein, a list of potential peptides can be generated using in silico digestion tools listed in Table II. The parameters like protein accession number, corresponding database (e.g., Uniprot), protease type (e.g., trypsin), mass range (800–3,000), and minimum peptide length [7] are entered, and a list of potential peptides is obtained with information on the end and start amino acids and peptide molecular weights. We used Protein Prospector and Skyline (see below) to perform this task for P-gp. While these tools allow protein digestion by various proteases (e.g., trypsin, chymotrypsin, Lys-C, etc.), trypsin is commonly used in proteomics because of the following advantages. Trypsin specifically hydrolyzes peptide chains primarily at the carboxyl side of the amino acids lysine (K) or arginine (R), except when either is followed by proline (P). Trypsin forms peptides that are favorable for detection in triple quadrupole LC-MS/MS instrument (e.g., peptides with mass range of 800–3,000 and charge (typically +2)). Further, most of the publically available MS proteomic data, i.e., spectral libraries, are derived from trypsin-digested peptides.

3. Exclusion of regions of uncertainty: When peptides are generated by in silico prediction, it is important that only those candidates that are suitable for quantitative work are selected. Only peptides with 7–22 amino acids should be considered as this mass range is detectable in triple quadrupole MS. Also, potential (or known) transmembrane regions are excluded as they can affect protein digestion of transporter proteins. Where experimental information on the location of the transmembrane regions is not available, non-experimental methods can be used to predict these regions. For this, Uniprot marks transmembrane regions as (i) “potential regions” indicating some conclusive in silico evidence, (ii) “probable regions” with at least some experimental data, and (iii) “regions by similarity” indicating that some experimental information is available on a similar protein (or part of it). Uniprot predicted 12 transmembrane regions for MDR1, i.e., positions 45–67, 117–137, 187–208, 216–236, 293–316, 331–352, 712–732, 757–777, 833–853, 855–874, 935–957, and 974–995 (Fig. 3). Posttranslational modification (PTM) can change the mass of a peptide and hence should be avoided. When literature data on PTMs are not available, predictive tools, e.g., NetNGlyc, NetOGlyc, and NetPhos, can be used to predict such PTMs (Table II). Using such tools, amino acid residues, 91, 94, 99, and 666 in MDR1 (Fig. 3), were excluded as they are expected to be either phosphorylated or glycosylated. It is also important to avoid any peptide with possible changes in amino acid sequence due to non-synonymous single nucleotide polymorphism (SNP) or other genetic variations. For example, a common variant rs2032582 in ABCB1 [36] causes changes in amino acid S → A (position 893) which restricts the use of ELEGSGK as a quantitative surrogate peptide. While this information can be obtained from literature, useful databases exist that compile such data (Table II). Consideration of genetic polymorphism is important, but this criterion can be relaxed if the allele frequency is low (e.g., <1%) particularly in the absence of suitable alternative peptides. In general, all the peptides containing cysteine (C) and methionine (M) residues (underlined in Fig. 3) can be excluded because they are susceptible to rapid oxidation [37]. Accuracy of reported protein sequence is also important to be considered. For example, conflicting sequences can be submitted to the databases from different research groups, different types of experiments, or different biological samples, e.g., positions 23 [38] and 336 [39] in P-gp. Conflicting information on protein sequence can be sometime obtained from databases (e.g., Uniprot). Continuous sequence of arginine (R) or lysine (K) (e.g., RR, KK, RK, or KR) (ragged ends) should be excluded as it may cause incomplete trypsin digestion. While trypsin does not cleave K and R that are followed by P, this phenomenon is not always true [40]. Nevertheless, peptides containing such sequences should be excluded if other peptides are available for quantification.

4. Uniqueness: While specificity can be tested right after step 2 above, it can be exercised on the limited number of potential proteotypic peptides after excluding the regions of uncertainty. Specificity can be verified by using online tools like BLAST search analysis or homology analysis in Protein Prospectus (Table II). These tools can also confirm whether the peptide is conserved across species as shown for MDR1 in Table III. Use of the conserved peptides is advantageous when transporter expression is compared across species. The use of the same peptide reduces variability introduced by trypsin digestion.

5. Additional criteria: Amino acids that are prone to deamination (asparagine and glutamine [41, 42]) can be avoided. N-terminal glutamine and glutamic acid can form cyclic pyroglutamate under acidic conditions [43]. Aspartic acid, when paired with glycine, proline, or serine, can undergo hydrolysis and cause peptide cleavage in acidic environment [44]. Tryptophan and histidine residues are also prone to chemical degradation [45–47]. Despite these stringent selection criteria, it is sometime difficult to control modification of peptide. In that case, a combination of both unaltered and altered peptides can be selected for reliable quantification. Some peptides are difficult to synthesize. For example, it is challenging to remove asparagine N-terminal-protecting agent during peptide synthesis [48]. A peptide containing multiple serine or proline residues is also difficult to synthesize. Optimum solubility of peptide is important for LC retention as well as for solubilization of peptides during various steps (synthesis to LC-MS/MS analysis) of quantitative proteomics. The incomplete solubility of peptide might be due to hydrophobicity of peptides or due to β sheet formation [49]. It is generally difficult to dissolve a peptide if more than 50% of the amino acid residues are hydrophobic (e.g., valine, isoleucin, leucin, phenylalanine, tryptophan, or cysteine). A series of glutamine, isoleucine, leucine, phenylalanine, threonine, tyrosine, or valine can form β sheets that result in incomplete solvation [37]. These criteria are important, but if applied stringently, potential MS-sensitive peptides might be excluded. Therefore, if peptides with such red flags are selected, it is important to test their stability and solubility before using them for protein quantification.

6. Prediction of chromatographic (LC) parameters: Only those peptides that are retainable in the LC column should be selected for the next phase. Generally, a reverse phase column is used for peptide quantification. Software tools, e.g., SSRCalc, could be used for predicting peptide LC retention. In general, peptides with hydrophobicity index below (HI) 7 are not retained on the column, and those with HI above 50 strongly bind with reverse phase material. Such peptides should be excluded. Skyline can also predict retention time (RT) by regression equation generated from experimental data. Typical LC mobile phase conditions involve initial elution of polar interferences by high aqueous mobile phase (e.g., 98% aqueous) followed by gradually increasing organic phase (e.g., up to 60% acetonitrile). For P-gp, we predicted RT for its surrogate peptides based on peptide sequences and Skyline. Only 11 peptides that have optimum RT in conventional reversed-phase columns were considered for the next stage (Table III). Alternatively, Thermo Scientific Pierce peptide RT calibration mixture contains 15 standard peptides that can be used for RT prediction based on peptide hydrophobicity.

7. Prediction of relative MS response: The MRM signal intensity depends on the MS response of parent as well as product ions. Not all peptides are MS responsive and not all MS/MS fragments are equally abundant. Therefore, real spectral libraries or experimental MRM data from databases (Table II) can be used to select the best peptide and fragment ions. However, as these data are not always available for drug transporters, bioinformatic tools such as ESPPredictor, PeptideSieve and CONSeQuence can be used for this purpose. These tools typically work based on the machine learning algorithms that are taught to identify highly abundant peptides from experimental data. This is an active area of research and it is anticipated that improved bioinformatic tools, developed based on larger datasets, will be available in the future. Fragmentation pattern with relative intensities of product ions can be obtained or predicted by tools, such as PeptideART, MRMPilot, PinPoint, Skyline, and SRMAtlas (Table II). The potential high abundant fragment ions can also be generated for heavy labeled peptide at this stage using these tools. We used SRMAtlas and PeptideART to predict the most intense fragments of P-gp peptides (Table III). We found that >90% of the top five fragments predicted by both the software were similar. The collision energy and fragmentor (cone) voltage for each peptide were predicted using SRMAtlas. Typical collision energy for MRM peptides is from 10 to 30 eV, and fragmentor (cone) voltage is from 100 to 160 V in Agilent triple quadrupole instrument. These parameters generally increase with peptide length. Dwell time can be optimized to allow at least 15 data points for each peak. If multiple peptides or transitions are used, dynamic or scheduled MRM can be adopted which allows targeted data acquisition for a given transition only around the predicted RT.

While the above discussion is about theoretical criteria-based manual method of surrogate peptide selection, more advanced bioinformatic tools are emerging for this purpose. Integrated bioinformatic tools Skyline, GPMDB, PeptideAtlas, SRMAtlas, and MRMaid for surrogate peptide selection are discussed below:

Qualification of Surrogate Peptide

Once a surrogate peptide is predicted by the in silico tools, it is important to test whether this peptide works in the real sample. One can procure the synthetic surrogate peptides at this stage and tune LC-MS/MS parameters using standards. However, peptide synthesis generally takes 4–6 weeks, and sometime the selected surrogate peptide turns out to be less sensitive in MS. Therefore, an alternative approach that is independent of peptide standard can be used to facilitate method development. The latter involves screening potential quantitative surrogate peptides in the real samples (membrane fractions from tissues or cells) using generic LC and MS conditions. For transporter protein, vesicles or cells expressing transporters are preferred for qualification as they generally have higher transporter expression. For example, 11 surrogate peptides of P-gp were detected in cell expression P-gp in a single dynamic MRM method (Fig. 4). The peptide signal can be qualified by (i) superimposability of multiple product ion chromatograms and (ii) elution of peptides at predicted RT (Fig. 5). Reproducibility of (i) and (ii) with another column generally adds confidence to the results. MS sensitivity can be optimized by playing with critical MS parameters such as collision energy, resolution, and dwell time for any given set of parent to product ion. An alternative approach is to purchase crude peptide libraries for tuning LC-MS/MS parameters. Such peptide libraries are currently available from Thermo Scientific Pierce Biotechnology (PEPotec SRM Peptide libraries) as aliquots of ~0.1 mg/peptide with purity between 75 and 125%. The cost of these peptides is much lower, $15–25/peptide, than the analytical grade peptide standard (~$600/peptide). Once two to three best peptides are qualified, e.g., IATEAIENFR, and NTTGALTTR (Table III) for P-gp, these can be used for protein quantification.

Peptide Synthesis and Characterization

Peptide can be procured from custom peptide synthesis companies (e.g., New England Peptides, Thermo Scientific Pierce Biotechnology, or Sigma Aldrich). The corresponding internal standard peptide that is labeled by incorporation of [¹³C₆¹⁵N₂]lysine or [¹³C₆¹⁵N₄]arginine can also be purchased. For absolute quantification of transporters, the purity and concentration of standard solutions are important. Amino acid analysis (AAA) is a critical test to ensure the absolute quantity of peptide standard used as calibrator. Chromatographic purity should also be established at this stage. Peptide aliquots should be stored at temperature <−20°C. New England Peptides, Thermo Scientific, and Sigma Aldrich also offer AAA for an additional cost of ~$250/peptide.

Sample Procurement and Membrane Protein Isolation

Endogenous factors such as genetic polymorphism, posttranslational modification, disease condition, and medication can affect drug transporter expression. Transporter quantification in tissues can also be affected by the quality of samples that depends on variables associated with sample procurement and storage (temperature, microbial contamination, etc.). Regarding transporter abundance in cell lines, factors such as passage number, confluency, or culture conditions can affect transporter expression. Cells grown in the presence of fetal bovine serum should be thoroughly washed if the total protein is used to normalize transporter expression.

Total transporter protein expression in the tissue homogenate or total membrane fraction (i.e., plasma membrane plus membranes from intracellular organelles) may not all be functionally active because some of it might be in the trafficking process to the membrane. Therefore, isolation of pure plasma membrane is ideal for predicting transporter function. To do so, there are a few protocols and commercially available kits that purport to isolate the plasma membrane [50, 51]. However, in our experience, these kits/methods do not yield pure plasma membrane fraction reproducibly for quantitative applications. Therefore, we and others have used total membrane transporter expression for this purpose [9, 11, 12, 19, 22, 52–54]. Although it is possible that the transporter trafficking is altered by disease condition or drug treatment [55–57], we believe that obtaining pure plasma membrane is not necessary if the ratio of expression of transporter protein in the plasma membrane relative to that in the entire cell is assessed to be similar in tissues (e.g., human livers) and in transporter-expressing cell lines. This assumption can be tested by crudely isolating various subcellular fractions from tissues and cells and comparing the expression of the transporters relative to the marker of plasma membrane (e.g., Na⁺K⁺ATPase) in these fractions. Such studies are ongoing in our laboratory and have yielded promising results.

Protein Digestion

The surrogate peptide-based absolute quantification of protein completely relies on the efficiency of protein digestion. Typical trypsin digestion process involves (i) denaturation of transporter proteins by denaturant like urea or sodium deoxycholate (DOC) and heating, (ii) reduction of cysteine residues by dithiothreitol (DTT), (iii) alkylation of thiols by iodoacetamide (IAA), and (iv) protein digestion by addition of sequencing grade trypsin. Optimization of protein/trypsin ratio, incubation time, temperature, pH, surfactant concentration, and stability of trypsin should be considered for the efficient trypsin digestion [58, 59]. Maximum digestion of transporter protein is achieved by incubating protein/trypsin in a ratio of 1:10 to 1:50 for 4–24 h at pH 7.5–8.5 and temperature 37°C. The use of surfactant is important to ensure solubilization of transporter protein. However, high concentration of surfactant can adversely affect trypsin digestion efficiency [58]. Trypsin digestion efficiency can be improved by using specialized equipment such as microwave [60], ultrasonic probe [61], and hydraulic pressure pulse activator [62]. Moreover, addition of serine protease (Lys-C) with trypsin improves digestion efficiency [63, 64]. The use of DOC is preferred over urea as a denaturant [58]. When urea is used, the temperature should not exceed 60°C as the latter results in carbamylation of proteins or peptides. Repigest is another denaturant that is becoming common in shotgun proteomics due to its compatibility with LC-MS/MS [65, 66].

Using the purified transporter protein as a calibration standard addresses the issue of less than 100% protein digestion (Table IV) [67]. For instance, based on pure P-gp protein as a quality control, we found that trypsin digestion of P-gp was close to 100% [11]. Digestion efficiency is not a problem in relative quantification if the variability in trypsin digestion is addressed. Variability can be addressed theoretically by using labeled protein (SILAC protein) as an internal standard. However, we observed that if trypsin digestion is optimized and sample-to-sample variability in the digestion is low, labeled peptide behaves equally well or better than the labeled proteins as an internal standard [13]. Alternatively, QconCAT relies on concatamers of tryptic peptides in an artificial protein as standards and therefore addresses digestion efficiency with the assumption that the rate of trypsin digestion remains similar in the sample and the artificial protein (Table IV) [68–70].

To increase confidence in protein quantification, one can use multiple peptides (whenever available). Here, we should note that despite superior precision of the MRM technique, it might be misleading to use the term “absolute” quantification because the latter depends on the degree of protein digestion. Irrespective of the latter, if protein quantification is performed using the same peptide(s) and similar conditions to achieve maximum protein digestion, such data can be used to derive scaling factors for IVIVE from cell lines to human tissues. However, this is only possible if the ratio of intracellular to plasma membrane expression of transporter is similar in tissues and cells.

Sample Preparation for Analysis

Sample cleanup is important for reliable MS signal. Protein or peptides can be enriched by using ultrafiltration, solid phase extraction, or chloroform/methanol extraction described elsewhere [71–73]. If enrichment methods are used, recovery should be evaluated. To do so, an endogenous marker (Na⁺K⁺ATPase) or a spiked labeled protein can be used as an enrichment marker. Labeled peptide as an internal standard can be used to address variability in post-digestion steps (e.g., matrix effect). Nonspecific binding to sample container and LC-MS/MS tubing should also be taken into consideration. It is generally preferred to use glass vials or LoBind plastic tubes to store samples and standards.

LC-MS/MS Parameter Optimization

LC separation is critical in MRM proteomics because of the complexity of tissue/cell protein sample. While nano-LC-MS/MS can be used for transporter quantification [14], our discussion is focused on conventional LC-MS/MS. A typical LC run is 15–60 min long in a reverse phase UHPLC column (e.g., 100 × 2.1 mm, particle size <2 μ). The initial mobile phase conditions are low organic solvents (e.g., 0–5% acetonitrile and 100–95% of water with 0.1% formic acid) for the first 3–5 min. This initial step washes polar interferences from the sample. During this step, the flow can be diverted to the waste to avoid MS source contamination and matrix effect. Mobile phase compatibility with the sample should be taken into consideration particularly if the analyte peptides are very polar. The organic phase can be gradually increased, and the slope of the gradient can be adjusted based on the hydrophobicity of the surrogate peptides. The predicted MRM parameters can be systematically tuned using the peptide standard similar to small molecules. Parent and product ions, collision energy, voltage, resolution, and dwell time are the key parameters that should be tuned using the standard.

Method Validation

US FDA bioanalytical method validation guidance can be adopted for LC-MS/MS protein quantification method (http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM368107.pdf). The validation should include demonstration of parameters such as selectivity, accuracy, precision, recovery, calibration curve, sensitivity, reproducibility, and stability. While MRM methods are generally selective, it is not surprising to observe multiple peaks due to similar parent-product pairs from tryptic digest. Therefore, it is important that a longer LC method and multiple transitions are used to confirm the target peptide signal. Nonavailability of a blank matrix is an important concern in the bioanalysis of endogenous proteins, including transporters. The potential interferences in a trypsin digest not only include other peptides but also include endogenous small molecules and xenobiotics. Therefore, fragments from each parent ion in the actual processed tissue samples (typically from multiple donors, e.g., n = 6) should be used to demonstrate the selectivity. Accuracy can be demonstrated by spiking standard peptide into the processed tissue samples. Within-run and between-run precision should be performed using processed tissue sample with and without adding synthetic standard. Considering the higher variability in protein quantification as compared to small molecules, acceptance criteria of ±25% for accuracy and precision can be used. Sensitivity can be demonstrated by measuring lowest analyte concentration (in sample buffer) that can be measured with acceptable accuracy and precision. It should be noted that the latter is not the actual lower limit of quantification (LLOQ) as the sample matrix does not represent the tissue sample. Therefore, if possible, a sample can be identified that lacks analyte protein (e.g., due to genetic polymorphism or mutation), and LLOQ standard can be spiked into such blank sample to measure the real sensitivity of the method. Effect of freeze and thaw cycles, longer storage, and bench-top exposure on peptide stability should be established. The effect of the above conditions on protein quality can be demonstrated because the quality of protein can alter trypsin digestion. Processed sample stability in auto-sampler should be studied.

Sample Analysis

The ratio of analyte and internal standard response is calculated and plotted against concentration for a given peptide. The calibration equation is then used to calculate concentration in the actual sample. Assessment of peak quality is important as multiple interfering peaks can be observed in the biological sample. One way to assure quality of peaks is by calculating correlation between peptides (or fragments). Vendor-specific software are available for high-throughput post-acquisition data analysis. Tools such as Skyline, AuDIT, and SRMCollider are also helpful in the analysis of large sample size.

While the above discussion mainly focuses on the peptide MRM proteomics, the proposed approach can be extended to other MRM protein assays summarized in Table IV [67–70, 74–79].

P-gp Protein Quantification in Human Liver Tissue, Hepatocytes, and Cell Lines

We selected two surrogate peptides for P-gp quantification, NTTGALTTR and IATEAIENFR, based on the above-discussed approaches (Table III). P-gp protein expression was quantified in the School of Pharmacy, University of Washington human liver bank, hepatocytes, and LLCPKII cells (Fig. 6) using a validated method [11]. The cell line expressed ~120-fold higher P-gp protein than that expressed in the hepatocytes or human liver. NTTGALTTR was used for this quantification because it resulted in ~1.4-fold higher yield than IATEAIENFR perhaps due to the better trypsin digestion efficiency of the former.