Quantitative Proteomics in Translational Absorption, Distribution, Metabolism, and Excretion and Precision Medicine

Abstract

A reliable translation of in vitro and preclinical data on drug absorption, distribution, metabolism, and excretion (ADME) to humans is important for safe and effective drug development. Precision medicine that is expected to provide the right clinical dose for the right patient at the right time requires a comprehensive understanding of population factors affecting drug disposition and response. Characterization of drug-metabolizing enzymes and transporters for the protein abundance and their interindividual as well as differential tissue and cross-species variabilities is important for translational ADME and precision medicine. This review first provides a brief overview of quantitative proteomics principles including liquid chromatography–tandem mass spectrometry tools, data acquisition approaches, proteomics sample preparation techniques, and quality controls for ensuring rigor and reproducibility in protein quantification data. Then, potential applications of quantitative proteomics in the translation of in vitro and preclinical data as well as prediction of interindividual variability are discussed in detail with tabulated examples. The applications of quantitative proteomics data in physiologically based pharmacokinetic modeling for ADME prediction are discussed with representative case examples. Finally, various considerations for reliable quantitative proteomics analysis for translational ADME and precision medicine and the future directions are discussed.

Significance Statement

Quantitative proteomics analysis of drug-metabolizing enzymes and transporters in humans and preclinical species provides key physiological information that assists in the translation of in vitro and preclinical data to humans. This review provides the principles and applications of quantitative proteomics in characterizing in vitro, ex vivo, and preclinical models for translational research and interindividual variability prediction. Integration of these data into physiologically based pharmacokinetic modeling is proving to be critical for safe, effective, timely, and cost-effective drug development.

I. Introduction

Drug development is a resource-intensive process that relies on a wide range of in vitro, preclinical, and clinical experiments, costing approximately 1.5 billion dollars over a period of 10–12 years. A poor translation of in vitro and preclinical data to humans and high interindividual variability in drug disposition and response pose significant challenges in drug development. Translational absorption, distribution, metabolism, and excretion (ADME) encompasses quantitative understanding and integration of in vitro, preclinical, and exploratory human drug disposition data to predict the right dose for the right patient at the right time toward precision medicine. Because drug disposition and response can be affected by genetic and nongenetic factors, precision medicine is considered a promising approach for safe and effective drug development and pharmacotherapy.

Although physiologically based pharmacokinetic (PBPK) modeling is emerging as a promising approach for predicting drug disposition using in vitro data, it requires a myriad of physiological data including protein abundance of drug-metabolizing enzymes and transporter (DMET) proteins in the in vitro models and tissues as well as their interindividual variability and differential tissue abundance. In vitro data are often generated using recombinant systems (e.g., Supersomes), cell lines or vesicles overexpressing transporters, and human-derived models (e.g., microsomes or cultured hepatocytes). However, these models differ in the protein abundance of DMET proteins from the corresponding human tissues (Xu et al., 2018). Similarly, a quantitative understanding of interspecies differences in DMET abundance is required for a reliable allometry scaling of drug disposition and tissue distribution data from preclinical models to humans, especially for interpreting drug toxicity data. For example, SGX-523, an investigational mesenchymal-epithelial transition factor inhibitor, was discontinued after the phase 1 clinical trials due to nephrotoxicity, which was not captured during the animal toxicity studies. SGX-523 is primarily metabolized by aldehyde oxidase (AO) in the human liver; however, human-specific AO isoform is poorly expressed in preclinical species, particularly in rats and dogs (Diamond et al., 2010; Basit et al., 2021). Similarly, the renal abundance of organic anion transporter 2 (OAT2) is >5-fold higher in mice as compared with humans and monkeys (Basit et al., 2019). Such species differences in DMET abundance often lead to discordance between preclinical and clinical data.

Quantitative proteomics has emerged as a crucial technique in the translational ADME and precision medicine. Typically, quantitative proteomics involves selective quantification of the proteotypic peptide(s) unique to the protein(s) of interest by liquid chromatography–mass spectrometry (LC-MS/MS), without the need for an antibody, thus making it a superior technique over conventional methods such as western blotting. The technique has been applied for characterizing DMET abundance for the assessment of (1) interindividual variability, i.e., the effect of age, genotype and sex, (2) differential tissue variability, (3) interspecies differences, and (4) the viability of in vitro models as illustrated with examples in Tables 1–5. Together, these data are important for in vitro to in vivo extrapolation (IVIVE) and PBPK modeling. For example, morphine and zidovudine pharmacokinetics (PK) and clearance were predicted in the neonatal population using ontogeny data of UDP-glucuronosyltransferase 2B7 (UGT2B7) and organic cation transporter 1 (OCT1) in pediatric and adult human liver samples (Emoto et al., 2017; Emoto et al., 2018; Bhatt et al., 2019). Similarly, the effect of disease states and genetic variations on drug PK and clearance have been successfully predicted by integrating quantitative proteomics data in PBPK modeling (Emoto et al., 2017; Bhatt et al., 2019; Vildhede et al., 2020). Moreover, quantitative proteomics data have also been shown to predict the fractional contribution of individual enzymes (f_m) and transporters (f_t) in drug disposition and drug-drug interaction (DDI) (Jamei et al., 2014; Li et al., 2019a; Anoshchenko et al., 2020; Kimoto et al., 2020). Similarly, the technique has been used for the selective quantification of biomarkers for monitoring drug response and disease diagnosis (Heo et al., 2007; Warth et al., 2012; Basit et al., 2020) and for identifying nonsynonymous single nucleotide polymorphism (SNPs) and post-translational modification (PTMs) (Shi et al., 2018b). Further, the selectivity of quantitative proteomics results in high signal-to-noise ratio for a sensitive determination of DMET abundance in a small sample volume (e.g., 24-hepatocytes on-column, and 10 ng microsomal protein on-column) (Ahire et al., 2021).

TABLE 1.

Successful examples of applications of quantitative proteomics for characterization of in vitro ADME models

Organ	Study Objective	Target DMET Protein Quantified	Reference
Liver	Effect of pregnancy related hormones on protein abundance of UGTs in primary human hepatocytes	UGTA1 and UGT2B7	Khatri et al., 2021
	Hepatocyte model selection to predict the uptake and biliary efflux of drugs in suspended, plated, and sandwich-cultured human hepatocytes	OATP1B1, NTCP, MRP3, and BSEP	Kumar et al., 2019
	Characterization of HepG2 3D spheroid model for hepatotoxicity	CYPs, UGTs, GSTs, MRP1, and MRP4	Hurrell et al., 2019
	Comparison of protein abundance between human liver microsomes and hepatic cell lines (e.g., HepG2, Hep3B, and Huh7)	CYPs, UGTs, SULTs, AKRs, GSTs, ALDHs, ADHs, AO, CESs, PONs, and EPXs	Shi et al., 2018a
	Comparative proteomics analysis of HLM and HLS9 fractions for in vitro drug metabolism studies	CYPs, UGTs, SULTs, AKRs, GSTs, ALDHs, ADHs, AO, CESs, PONs, and EPXs	Wang et al., 2020
	Effect of pregnancy-related hormones and induction of CYP3A activity over the course of pregnancy in HepaRG and SCHHs cells	CYP3A	Zhang et al., 2015b
	Test primary hepatocytes as a suitable model for hepatic drug metabolism study	CYPs, UGTs, MDR1, MRPs, BCRP, NCP, OCTs, OATPs, CNTa, ENT1, and ATE1	Schaefer et al., 2012
	Interindividual variability in DMEs in HLMs	CYPs and UGTs	Ohtsuki et al., 2012
	Compare protein abundance between hepatic cell lines (HepG2, Hep3B, and Huh7)	CYPs, UGTs, and ALDHs	Shi et al., 2018a
Intestine	Drug metabolizing capabilities of human primary intestinal monolayers	CYP3A4, UGT1A1, UGT1A10, UGT2B17, and CES2	Speer et al., 2019
	Effect of culture duration on transporter abundance in Caco-2 cells	MDR1, BCRP, PEPT1, OSTα and OSTβ, MRP2, MRP4, OATP2B1 and MCT1	Uchida et al., 2015
	Regional DMEs protein changes along the line of the small intestine using cryopreserved human intestinal mucosa (CHIM) model	UGTs, CESs, and SULTs	Zhang et al., 2020a
Brain	Determine the protein abundance of transporters and receptors in human brain microvessels	BCRP and P-gp	Uchida et al., 2011
	Protein abundance of key membrane proteins in hCMEC/D3 cells and human brain microvessels	ABCA2, MDR1, MRP4, BCRP, GLUT1, 4F2hc, MCT1, and ENT1	Ohtsuki et al., 2013
	Identify DMET protein expression in brain microvascular endothelial cells	GLUT1, EAAT1, EAAT2, BCRP, MDR1, CYP1B1, CYP2U1, GSTP1, COMT, GSTM3, GSTO1, and GSTM2	Shawahna et al., 2011
	Quantification of brain transporter abundance in brain microvascular endothelial cells	BCRP, P-gp, OATP2B1, ENT1, and GLUT1	Billington et al., 2019
Skin	Characterization of in vitro models of human skin	CYPs, ADHs, ALDHs, AKRs, AO, CESs, EPHx, and PONs	Couto et al., 2021
Eye (retina)	Transporter protein abundance in the plasma membranes of the human retinal pigment epithelium cells and immortalized arpe19 cell line)	MRP1, MRP5, GLUT1, 4F2hc, TAUT, CAT1, LAT1, MPR7, OAT2, MCT1, MCT4, MRP4, RFC1, and MATE1	Pelkonen et al., 2017

TABLE 5.

Interspecies differences in DMET abundance (pmol/mg protein)

Tissue	DMET Proteins	Human	Monkey	Dog	Rat	Mouse	References
Liver	MDR1/Mdr1a	0.40 ± 0.20	0.46 ± 0.10	0.59 ± 0.11	0.26 ± 0.03	—	Wang et al., 2015
		0.12 ± 0.01			0.02 ± 0.04	0.01 ± 0	Hammer et al., 2021
	MATE1/Mate1	0.48 ± 0.15	—	—	0.72 ± 0.15	—	Wang et al., 2015
	MRP2/Mrp2	1.54 ± 0.64	5.15 ± 1.53	1.93 ± 0.19	7.33 ± 1.16	—	Yin et al., 2015
					2.84 ± 0.44		Hammer et al., 2021
	MRP3/Mrp3	0.51 ± 0.15	0.76 ± 0.21	0.38 ± 0.05	BLQ	—	Wang et al., 2015
					0.08 ± 0.01	0.13 ± 0.06	Hammer et al., 2021
	MRP4/Mrp4	BLQ	BLQ	BLQ	BLQ	—	Wang et al., 2015
	OCT1/Oct1	4.45 ± 1.89	18.8 ± 6.8	0.50 ± 0.30	2.14 ± 0.47	—
	BSEP/Bsep	2.11 ± 0.50	4.45 ± 1.48	1.87 ± 0.37	2.31 ± 0.45	—
		0.23 ± 0.01			1.02 ± 0.16	0.94 ± 0.26	Hammer et al., 2021
	NTCP/Ntcp	2.17 ± 0.93	1.90 ± 0.24	—	7.91 ± 1.36	—	Wang et al., 2015
		0.21 ± 0.03			1.74 ± 0.23	19.21 ± 3.52	Hammer et al., 2021
	Oatp1a2			0.70 ± 0.17			Wang et al., 2015
	Oatp1b4			10.4 ± 2.3
	OATP2B1/Oatp2b1	1.7 ± 0.6	0.30 ± 0.10	4.55 ± 2.00
	OATP1B1/Oatp1b1	2.0 ± 0.9	12.7 ± 1.4
	OATP1B3/Oatp1b3	1.1 ± 0.5	14.4 ± 2.1
	Oatp1a1				3.91 ± 0.84
	Oatp1a4				5.56 ± 1.40
	Oatp1b2				6.43 ± 1.40
	CYP1A1/Cyp1a1	0.09 ± 0.01			0.05 ± 0		Hammer et al., 2021
	CYP1A2/Cyp1a2	5.93 ± 0.45			1.3 ± 0.49
	CYP2B6	0.59 ± 0.04
	CYP2C8	7.01 ± 0.55
	CYP2C9	14.87 ± 0.67
	CYP2C19	2.29 ± 0.07
	CYP2D6	4.35 ± 0.25
	CYP2E1	3.81 ± 0.2
	CYP3A4	4.8 ± 0.2
	CYP3A5	0.77 ± 0.04
	Cyp2b10					0.21 ± 0
	Cyp2c29					8.07 ± 0.93
	Cyp2c38					0.03 ± 0
	Cyp2c39					0 ± 0
	Cyp2c55				0.01 ± 0	0.18 ± 0.03
	Cyp2d9					18.4 ± 1.8
	Cyp2d10					16.7 ± 2.0
	Cyp2e1				5.86 ± 0.97	32.8 ± 2.2
	Cyp3a25					0.24 ± 0.09
	Cyp2b1				0.1 ± 0.03
	Cyp2c11				57.8 ± 15.7
	Cyp2c12				0.43 ± 0.41
	Cyp2c13				2.7 ± 2.7
	Cyp2d3				11.1 ± 2.2
	Cyp3a9				0.48 ± 0.15
	Cyp3a18				1.0 ± 0.4
	AO	11.96	17.87				Basit et al., 2021
	CES1	569.06	153.55
	CES2	64.85	491.69
	SULT1A1	5.97	23.01
	SULT1B1	0.23	0.31
	SULT1E1	1.07	1.60
	SULT2A1	18.59	17.63
	SULT1A3	1.98
	UGT1A1	0.87	0.29
	UGT1A3	0.3	0.28
	UGT1A6	1.02	5.70
	UGT2B4	5.83	32.25
	UGT2B7	6.01	20.73
	UGT2B15	3.76	7.92
	UGT2B17	0.21
Kidney	MDR1/Mdr1a	2.11 ± 0.68	1.86 ± 0.34	1.08 ± 0.35	1.13 ± 0.28	0.56 ± 0.21	Basit et al., 2019
	MATE1/Mate1	5.21 ± 2.11	5.76 ± 1.75	—	—	—
	MRP4/Mrp4	0.91 ± 0.62	2.58 ± 0.74	—	1.09 ± 0.21	0.32 ± 0.14
	OAT1/Oat1	5.21 ± 1.82	8.51 ± 1.59	2.37 ± 0.94	8.96 ± 1.97	5.19 ± 3.25
	OAT2/Oat2	0.97 ± 0.32	0.96 ± 0.3	—	BLQ	4.12 ± 1.37
	OAT3/Oat3	3.84 ± 1.3	4.44 ± 1.03	—	—	—
	OAT4/Oat4	0.51 ± 0.23	0.62 ± 0.2	—	—	—
	OCT2/Oct2	8.16 ± 2.34	16.72 ± 4.84		7.32 ± 1.76	14.7 ± 4.88
	OCTN1/Octn1	1.32 ± 0.51	0.63 ± 0.3	—	—	—
	OCTN2/Octn2	0.64 ± 0.2	1.55 ± 0.33	0.53 ± 0.18	2.8 ± 0.59	1.48 ± 0.31
	SGLT2/sglt2	3.68 ± 1.51	6.39 ± 1.34	11.89 ± 2.87	11.6 ± 2.25	13.66 ± 2.91
	AO	1.53	3.10				Basit et al., 2021
	CES1	3.65	44.14
	CES2	35.1	72.48
	SULT1A1	BLQ	2.97
	SULT1B1	BLQ	0.15
	SULT1E1	BLQ	1.05
	SULT2A1	BLQ	0.40
	SULT1A3	BLQ
	UGT1A1	BLQ	0.06
	UGT1A3	BLQ	0.04
	UGT1A6	0.22	0.41
	UGT2B4	BLQ	0.23
	UGT2B7	1.74	2.46
	UGT2B15	BLQ	1.17
Intestine	AO	BLQ	1.25				Basit et al., 2021
	CES1	3.5	62.37
	CES2	224.34	388.08
	SULT1A1	3.44	15.18
	SULT1B1	1.74	1.25
	SULT1E1	1.08	5.29
	SULT2A1	4.88	7.56
	SULT1A3	6.8
	UGT1A1	0.97	4.32
	UGT1A3	0.03	4.58
	UGT1A6	BLQ	0.88
	UGT2B4	BLQ	0.12
	UGT2B7	1.88	0.83
	UGT2B15		2.23
	UGT2B17	1.31

Here, we provide a summary of various MS techniques, acquisition methods, and sample preparation approaches used in quantitative proteomics, with a detailed compilation of the reported applications of the technique in the in vitro model characterization, IVIVE, interindividual variability, intertissue differences, interspecies variability, PBPK modeling, and drug response prediction. Various considerations and recommendations for a robust quantitative DMET proteomics analysis, including the future directions, are also discussed.

II. Quantitative Proteomics: Basic Principles

Quantitative proteomics has emerged as a major approach for protein quantification due to the significant developments in both LC-MS hardware and software. The technique is primarily based on the bottom-up or shotgun proteomics that relies on the analysis of digested proteins by a protease enzyme(s) (e.g., trypsin, trypsin plus Lys-C, chymotrypsin, or proteinase C) in biologic samples. Due to the limitations of the spectral counting method of protein quantification (Lundgren et al., 2010), chemical and metabolic labeling approaches were developed in the late 1990s for relative and absolute protein quantification (Fig. 1A). The labeling techniques include isotope-coded affinity tag (Gygi et al., 1999), stable isotope labeling by amino acids in cell culture (SILAC) (Ong et al., 2002), tandem mass tags (TMT) (Thompson et al., 2003), and isobaric tags for relative and absolute quantification (iTRAQ) (Wiese et al., 2007). The labeling methods provided advantages of greater accuracy and precision but suffer from limitations of multistep sample preparation, higher cost, and higher sample requirement. Recent approaches like absolute quantification (AQUA) or multiple reaction monitoring (MRM) (Gerber et al., 2003), quantitative concatemers (QconCAT) (Beynon et al., 2005; Carroll et al., 2011), protein standards for absolute quantification (PSAQ) (Kaiser et al., 2011), and stable isotope labeling and capture by anti-peptide antibodies (SISCAPA) address some of these limitations (Beynon et al., 2005).

Fig. 1.

Absolute and relative LC-MS/MS–based protein quantification (A) and proteomics data acquisition (B) approaches. (C) Schematic depiction of various mass-spectrometry data-acquisition approaches for targeted and untargeted proteomics. MRM can be performed using low-resolution MS analyzers such as triple quadrupole and Q-trap, but PRM, DDA, and DIA require high-resolution MS analyzers such as orbitrap or TOF. AQUA, absolute quantification.

A. Liquid Chromatography–Mass Spectrometry Techniques and Data Acquisition Approaches

The LC separates analytes based on lipophilicity, and the MS measures the mass-to-charge (m/z) ratio of ions in the gas phase. In the MS, an ion source first ionizes analytes in the gas phase, ions are separated in a mass analyzer(s) based on the m/z ratio, and the product ion intensity is recorded by a detector. The breakthrough of electrospray ionization in the 1990s solved the problem of ion generation from large and nonvolatile analytes (e.g., proteins and peptides), making it suitable for liquid chromatography. Atmospheric pressure chemical ionization has also been used in the proteomics study, especially for analysis of smaller peptides, albeit to a lesser extent than electrospray ionization. Atmospheric pressure chemical ionization is a softer ionization technique in which the ionization of analytes is performed by proton transfer at atmospheric pressure outside of the vacuum chamber in the MS (Doerge et al., 1996; Toribio et al., 2000).

Typical mass analyzers used in proteomics are, (1) quadrupole (Q), (2) ion trap (IT), (3) time-of-flight (TOF), (4) orbitrap, and (5) Fourier-transform ion cyclotron resonance. Although the analyzers can be used standalone, hybrid instruments are designed to combine the unique capabilities of mass analyzers for greater analytical performance (Supplemental Table 1). The hybrid MS analyzers include triple quadrupole, hybrid quadrupole ITs (Q-q–linear IT or Q-trap), Q-TOF, TOF-TOF, LTQ orbitrap, and LTQ Fourier-transform ion cyclotron resonance.

MS-based data acquisition falls under two major categories: targeted (hypothesis-driven) and untargeted (hypothesis-generating) approaches (Fig. 1B). The selection of the MS acquisition approach depends on the experimental goals, such as the need for proteome coverage versus sensitivity and precision (Elias et al., 2005; Pino et al., 2020). For example, a selected list of DMET proteins can be quantified by the targeted approaches such as MRM or parallel reaction monitoring (PRM) for sensitive and precise quantification. On the other hand, the untargeted acquisition approaches [e.g., data-dependent acquisition (DDA) and data-independent acquisition (DIA)] are employed for achieving a dynamic peptide coverage. DDA method acquires MS ions based on their intensities in each sample (e.g., top 10), whereas DIA acquires all theoretical ions irrespective of their intensities for deeper proteome coverage that is possible with higher data acquisition speed. In general, targeted proteomics methods are more sensitive and precise for a small number of proteins, whereas untargeted proteomics is superior for obtaining proteome-wide coverage and simultaneous quantification of a large number of proteins, thus providing information at biologic pathway levels. A detailed comparison of different protein acquisition approaches and conventional immunoquantification is provided in Supplemental Table 2 and summarized below.

1. Targeted Proteomics

MRM is the most popular technique for DMET proteomics analysis, in which proteotypic or surrogate peptide(s) unique to the target protein is quantified (Fig. 1C) (Wolf-Yadlin et al., 2007; Kamiie et al., 2008). MRM is performed on a triple quadrupole or Q-trap mass spectrometers, where the first quadrupole is set to allow only a selected m/z ratio of a precursor ion into the second quadrupole. The MS fragmentation of the selected ion by collision-induced dissociation in the second quadruple yields characteristic product ions. The product ions are filtered in the third quadrupole and detected by a MS detector (e.g., photomultiplier tube). Generally, the top 3 high intensity product ions are monitored for a reliable peptide quantification (Lange et al., 2008; Vidova and Spacil, 2017). Although MRM-based proteomics is selective and precise, it can only quantify a limited number of product ions (10–100) concurrently. Scheduled or dynamic MRM partially addresses this problem by analyzing target ions in a prefixed elution time window, thus allowing analysis of >300 transitions per 30 minutes (Carr et al., 2014; Ronsein et al., 2015). PRM or MRM high resolution provides a more advanced alternative due to its ability to quantify all fragment ions from a single or multiple precursor ions simultaneously (Fig. 2) (Peterson et al., 2012). PRM offers advantages such as the ability to quantify 100s–1000s of proteins, higher specificity, and high-throughput quantification (Peterson et al., 2012; Kim et al., 2016) without the need for additional method development and optimization efforts (Rauniyar, 2015). Skyline (Genome Sciences, University of Washington, Seattle, WA) is currently the most used open-access software for targeted proteomics method development and analysis.

Fig. 2.

Steps involved in proteotypic peptide selection for the targeted quantitative proteomics. First, protein sequence is acquired from databases such as Uniprot (https://www.uniprot.org/). MRM transitions are predicted using Skyline (https://skyline.ms/project/home/begin.view?). Finally, proteotypic peptides can be selected based on criteria listed on the top-right corner using a series of software tools such as SRM Atlas (https://db.systemsbiology.net/sbeams/cgi/PeptideAtlas/GetTransitions), Protein Prospector (https://prospector.ucsf.edu/prospector/cgi-bin/msform.cgi?form=mshomology), sequence-specific retention calculator (SSRC) (http://hs2.proteome.ca/SSRCalc/SSRCalcx.html), or Protter tool within Skyline. SRM, Selected Reaction Monitoring.

Proteotypic or surrogate peptide selection is the first step in both MRM and PRM (Kamiie et al., 2008). Briefly, the proteotypic peptides should be unique, have an optimum length of 7–22 amino acids, should be devoid of transmembrane region, and should not be prone to PTM or change by a SNP site. Ideally, unstable amino acid residues (e.g., C, M, W), ragged ends (RR, KK, RK, and KR), or potential missed cleavage sites are avoided. The peptide should also have an optimum hydrophobicity (∼30% hydrophobic residues) to be retained on a chromatographic column (Fig. 2). In MRM, a synthetic stable isotope-labeled (SIL) or heavy peptide with ¹³C and/or ¹⁵N-labeling at the carboxyl-terminal of lysine or arginine residues are used as internal standards to normalize the matrix effect (MS ion suppression) and other postdigestion variables (Lange et al., 2008). Alternatively, an external multipoint calibration curve of a standard light peptide can be used for absolute peptide quantification, especially when the linearity is warranted over a large dynamic range.

2. Untargeted Proteomics

Untargeted proteomics is a hypothesis-independent approach that utilizes high-resolution MS data in conjunction with bioinformatics tools such as MaxQuant (Max Planck Institute of Biochemistry, Germany) for unbiased system-wide proteome identification and quantification (100s–1000s of proteins). Untargeted proteomics data are acquired using one of the two techniques, i.e., DDA and DIA. In DDA, full-scan mass spectrum determines the number of peptides present in a sample, and the most abundant precursor ions (e.g., Top 10) are selected and fragmented in the second stage. Although DDA is selective for high-intensity peptides, it is not efficient in the quantification of low-abundant proteins (Meyer, 2019; Smith et al., 2019; Tiwary et al., 2019). Further, each peptide species in DDA is purposefully detected only a few times, resulting in limited precision. On the other hand, DIA is a more advanced approach that allows for the detection of all theoretical masses in the preselected m/z windows, which are fragmented and analyzed to record all theoretical second stage scans (Reubsaet et al., 2019). DIA generally involves one of the two approaches: sequential windowed acquisition of all theoretical fragment ions (SWATH) and collision energy alteration (LC-MS^E). In SWATH, the full mass range is divided into several smaller m/z windows, and all precursor ions in a preselected m/z isolation window are fragmented (Ludwig et al., 2018; Smith et al., 2019). During each cycle time (2–4 seconds), the precursor ions are selected and fragmented multiple times, providing time-resolved product ions of all eluted precursor ions. In MS^E, the precursor and product ions are analyzed at low- and high-collision energy scans, and the precursor ions are fragmented within a selected m/z range. DDA and DIA data are generally analyzed using open-access software, e.g., MaxQuant and OpenSWATH.

B. Protein Quantification Approaches

Protein quantification can be achieved utilizing either labeled-based or label-free approaches. The first approach is based on chemical or metabolic labeling for a sensitive and reproducible relative or absolute quantification. For example, in stable isotope labeling by amino acids in cell culture, the heavy amino acids (lysine, arginine, or both) are introduced into the proteins in the cell culture for one condition (e.g., test), whereas the cells are grown in regular media for another condition (control). The harvested cells are mixed and homogenized before digestion (Ong et al., 2002). The light to heavy peptide signal ratio provides an estimate of quantitative changes in test versus control. Isotope-coded affinity tag utilizes chemical labeling of cysteine residues (Gygi et al., 1999), whereas in iTRAQ and TMT, the digested peptides are labeled by chemical tags at the free amines (Gygi et al., 1999; Thompson et al., 2003). The labeled samples are mixed for multiplexed analysis by targeted or untargeted proteomics approaches. Because the tags are isobaric in iTRAQ and TMT, no mass shift is detected in the labeled peptides in the full parent ion scan, and the signals from a single peptide from all samples are summed, whereas the distribution of isotopes in the different tags represents the tag-specific “reporter” ion intensity. The ratio of MS signal intensities from the tags provides the relative amounts of the peptide in different samples (Wiese et al., 2007).

Isotope dilution is another approach that utilizes spiked-in SIL peptide standards (Stöcklin et al., 1997; Brun et al., 2007). A known and consistent amount of SIL peptides is spiked into all samples, and the ratio is monitored to estimate the absolute levels. However, isotope dilution assay can only be used for a limited number of proteins because of the cost (Li et al., 2016; Vildhede et al., 2018; Wiśniewski et al., 2019). AQUA, QconCAT, and PSAQ all use the isotope dilution or spiked-in standards, where AQUA standards consist of synthetic peptides that are spiked into the sample after proteolysis (Gerber et al., 2003), QconCAT concatemers are chimeric proteins that are composed of different proteotypic peptides from multiple target proteins (Beynon et al., 2005), and PSAQ standards are full-length protein(s) that match the biochemical properties of the target protein(s) and spiked in at the beginning of the analytical process (Kaiser et al., 2011).

The total protein approach (TPA) is an emerging global quantitative proteomics strategy that utilizes a computational method to convert spectral intensities to protein concentrations (Vildhede et al., 2018). First used in 2012 (Wiśniewski et al., 2012), TPA has been applied in protein quantification for a variety of applications in ADME research (Wiśniewski and Rakus, 2014; Bryk and Wiśniewski, 2017; Ölander et al., 2020). TPA is centered on the quantification of individual proteins and groups of functionally related proteins in any protein mixture without the use of spike-in standards. Thus, TPA protein quantification approach can be applied to diverse large-scale proteomic data with reasonable precision. However, TPA is reliant on the depth of the proteomic analysis, which, in theory, requires complete digestion of all proteins in a sample using more than one protease to achieve precise protein quantification (Wiśniewski et al., 2019).

C. Proteomics Sample Preparation Approaches

Efficient and reproducible sample preparation is the key to quantitative proteomic analysis. Although many sample preparation approaches are used for the reproducible quantification of proteins (Supplemental Table 3), DMET proteins pose unique challenges in the sample preparation because of the transmembrane nature of DMET proteins. Quantitative proteomics analysis of transmembrane proteins is often challenging because these proteins are coagulated during the protein precipitation step, making them non-accessible to the proteolytic enzymes. To achieve optimum digestion, membrane proteins require additional treatments such as membrane solubilization or enrichment using immunoprecipitation, centrifugation, or cell surface biotinylation. Membrane solubilization utilizes MS-compatible detergents, organic solvents, and organic acids (Han et al., 2001; Blonder et al., 2002) to make them more accessible for proteolysis. Plasma membrane enrichment by surface biotinylation and enrichment via streptavidin beads is another method that reduces the contamination from other subcellular organelles and enhances the quantification efficiency (Shin et al., 2003; Zhang et al., 2003; Zhao et al., 2004; Kumar et al., 2017b). For example, the biotinylation approach has been used to enrich the apical and basolateral membrane transporters of rat kidney collecting ducts before LC-MS/MS analysis (Yu et al., 2006) as well as for the quantification of plasma membrane transporters expressed in CHO and MDCK cells (Kumar et al., 2017b).

D. Technical Challenges and Quality Controls

Rigor and reproducibility are important attributes of a quantitative method. The selection of optimum quality controls provides confidence in the results by minimizing technical variability in data. Interlaboratory variability in protein quantification is common in the absence of harmonized guidelines for quantitative proteomics. The factors contributing to the inter- or within-laboratory variability in protein quantification typically include the use of different matrices (e.g., homogenate versus cell fractions), different digestion protocols [e.g., filter-assisted sample preparation (FASP) versus immunoenrichment], different measurement techniques (targeted versus untargeted), variability in sample extraction efficiency, and inadequately optimized proteotypic peptides. Therefore, the use of internal and external quality controls (discussed below) is required for minimizing the technical variability in protein quantification results.

1. Multiple Proteotypic Peptides and Multiple Product Ions

The use of multiple proteotypic peptides of a protein provides confidence in protein quantification by eliminating bias caused by changes in amino acid due to SNP, PTM, and peptide stability. A good correlation between the responses of multiple peptides across samples can be used to rule out peptide-specific variability. Similarly, the selection of multiple product ions of a peptide should be measured for reproducible protein quantification. A strong correlation (generally r² > 0.99) between the responses of the selected product ions confirms the absence of any background interferences.

2. Reanalysis of Incurred Samples

Reanalysis of a sample with known concentrations of proteins of interest [also referred to as positive quality control (PQC) sample] assists in evaluating the batch-to-batch variability. Data normalization by the PQC sample minimizes intra- and interday variations in the sample preparation and analysis steps. Further, the use of a universal PQC sample for data normalization can be implemented to reduce interlaboratory variability.

3. Peptide and Protein Internal Standards

Externally added SIL peptides as an internal standard address the variations due to postdigestion factors such as the matrix effect, MS instrument sensitivity, retention time shift, and sample evaporation. Similarly, the spiking of an exogenously added protein (e.g., bovine serum albumin) during the digestion can serve as an internal control for the predigestion, sample processing, and instrumentation variability.

4. Data Normalization by Marker Proteins

Subcellular fractions (e.g., tissue homogenate, S9 fractions, cytosol, and microsomes) used for in vitro drug metabolism and transporter studies are susceptible to interbatch or interlaboratory variability, which often leads to a poor IVIVE. For example, to extrapolate enzyme-mediated drug clearance from subcellular fraction to a tissue or organ clearance, universal scaling factors [e.g., a milligram of protein per gram of liver tissue (MPPGL)] are required. Identification of marker (housekeeping) proteins (Supplemental Table 4) can be used to calculate recovery, enrichment, and purity in the in vitro preparations (Xu et al., 2018). The marker proteins can also be utilized for characterizing the in vitro subcellular fraction that is critical for IVIVE. Similarly, the marker proteins can identify variability in tissue sampling in heterogeneous tissue, e.g., kidney (Li et al., 2019b).

5. Identifying Sample Integrity Issues

Tissue samples are particularly susceptible to quality issues. Principal component analysis (PCA) is a dimensionless method that reduces the complexity of a large number of variables by transforming them into smaller ones (principal components, e.g., PC1, PC2, etc.). Since smaller data sets are easy to visualize and analyze, the overall variability can be interpreted by machine learning algorithms in PCA (Rao and Li, 2009). Thus, PCA can be used not only to interpret data for changes caused by biological factors but also in identifying poor quality samples in a big sample cohort (Yan et al., 2008; Rao and Li, 2009; Bhatt and Prasad, 2018).

III. Applications of Quantitative Proteomics in Translational Absorption, Distribution, Metabolism, and Excretion and Precision Medicine

A. Characterization of In Vitro Models

In vitro models are crucial for ADME screening because of their ability to recapitulate human mechanisms, generate data in a cost-effective and high-throughput manner, avoid cross-species differences, and minimize animal use. Human liver microsomes, cytosol, hepatocytes, recombinant proteins, and transporter-expressing cell lines or vesicles are commonly used for in vitro screening of drug metabolism and transport as well as for predicting in vivo clearance and DDI potential. However, the in vitro models suffer from several limitations e.g., (1) pooled human microsomes or cytosol do not represent the population variability, (2) DMET abundance or activity changes with cell culture conditions or between lots, and (3) DMET abundance in recombinant enzyme systems or transporter-overexpressing cell lines and vesicles is not similar to corresponding human tissue. To address these limitations, quantitative proteomics has been used to assess the effect of cell culture conditions on transporter abundance in human hepatocytes in relation to the freshly isolated or cryopreserved human hepatocytes (Kumar et al., 2019). Similarly, quantitative proteomics allows the determination of the relative expression factor (REF) values between the recombinant models and human tissue for tissue-specific extrapolation of the intrinsic clearance data (Parvez et al., 2021).

Microsomal enrichment and recovery are influenced by the processing variables such as the choice of homogenizer, buffer composition, and centrifugation speed resulting in technical variability in the enzyme abundance or activity (Xu et al., 2018). Interestingly, the impact of microsomal or cytosolic processing variables is not consistent for different enzymes as it depends on the subcellular localization of proteins. For example, carboxylesterases (CESs) are soluble endoplasmic reticulum luminal proteins, which are susceptible to release into cytosolic fraction during the microsome preparation. Although microsomal fraction is used for CES activity, it does not account for the loss into the cytosolic fraction during the microsomal preparation. Plasma membrane isolation for transporter quantification suffers from similar challenges of contamination from other subcellular organelles (Kumar et al., 2015). Multiplexed quantification of subcellular markers (Supplemental Table 4) by quantitative proteomics is capable of assessing the purity of microsomal, cytosolic, and membrane preparations (Xu et al., 2018).

Overexpressed cell lines (MDCK or LLC-PK1 cells) are commonly used in vitro models to characterize transporter substrate or inhibitor potential of new drugs. However, large interlaboratory differences in transporter expression across laboratories reduce confidence in translating these data to humans. Quantitative proteomics has been used to delineate differences between the overexpressed in vitro models and tissue preparations (Table 1). Breast cancer resistance protein (BCRP/ABCG2) and bile salt export pump (BSEP/ABCG11) were quantified in overexpressing insect membrane vesicles and rat, dog, monkey, and human liver samples for the extrapolation of PK parameters of biliary excretion from in vitro models to humans (Li et al., 2009). Similarly, systemic characterization of four different P-glycoprotein (P-gp)-overexpressing cell lines for the abundance resolved IVIVE disconnect in another study (Li et al., 2021).

Because of the high cost and logistical challenges of obtaining good quality human tissues and primary cells, cell lines such as Caco-2, HepG2, HepaRG, HUH7, and LS180 cells are also used in ADME assays. Characterization of the ability of these models to metabolize or transport drugs, including the effect of culture conditions on the DMET abundance, is important for better translation of the in vitro data (Table 1). Recently, advances in cell culture and microfluidics techniques have led to the development of microphysiologic systems (MPS) that are proving to be promising in drug screening. However, accurate determination of DMET proteins in MPS models is essential for the validation of these tools. Proteomic characterization in MPS models is often challenging due to low initial cell counts (5000–20 000 cells), especially using the modest sensitivity of conventional LC-MS instruments. Advances in LC-MS/MS sensitivity and miniaturized sample preparation are expected to address this limitation.

B. In Vitro to In Vivo Extrapolation

Although in vitro models play crucial roles in the early screening of undesirable characteristics of an NCE, accurate scaling of in vitro data to humans is critical for safer and cost-effective drug development. In vitro metabolic or transporter-mediated clearance data are used to predict in vivo clearance, DDI, as well as in vivo human PK to estimate the first-in-human dose. However, the lack of knowledge concerning physiologic differences, including DMET abundance between in vitro models and human tissues, often leads to the poor scaling of in vitro data. The relative activity factor (RAF) approach (eq. 1) that relies on differential activity measured using a probe substrate between recombinant system and human tissue is used for IVIVE of drug intrinsic clearance. However, specific probe substrates are not available for the majority of DMET proteins. Moreover, this approach is not viable for transporter assays as the uptake and efflux activity cannot be measured in tissue samples.

Proteomics-informed IVIVE utilizing the REF approach (eq. 2) is an emerging alternative to address the limitations of the RAF approach. Typically, the enzyme or transport kinetics data in the recombinant system provide an estimation of in vitro velocity maximum (V_max) and substrate affinity (K_m) values to derive intrinsic clearance (CL_int). Since V_max is proportional to the protein expression, and K_m is considered constant between systems (tissue versus in vitro system), the REF approach can be used to normalize the kinetics data by protein abundance in tissues to estimate tissue-intrinsic clearance by individual enzymes (eq. 2). The DMET abundance–normalized CL_int data can be extrapolated to organ clearance by multiplying it with two systems scaling factors, e.g., MPPGL or cytosolic protein per gram of liver (CPPGL) and organ weight (eqs. 1 and 2). Proteomics-informed scaling of in vitro transport data from OCT1-transporter–expressing HEK293 cells and plated human hepatocytes resulted in a reasonable prediction of in vivo clearance (Sachar et al., 2020). Likewise, IVIVE of transporter-mediated renal clearance of 26 OAT substrates by proteomics-based REF approach was able to predict the renal clearance within 2- to 3-fold of the observed values (Kumar et al., 2021). Quantitative proteomics data can also be used to predict tissue-specific toxicity by integrating DMET abundance and in vitro metabolism and transporter data. For example, selective intestinal toxicity of SN-38 can be explained by the higher abundance of CES2 relative to UGT1A1 in the intestine in comparison with the liver (Parvez et al., 2021). Further, quantitative proteomics analysis of microbial gut β-glucuronidases showed a good correlation with higher intestinal toxicity of SN38 as the former facilitates the conversion of SN38-glucuronide to the toxic SN38 (Parvez et al., 2021). DMET proteins are expressed in various tissues besides the liver. Quantification of DMET proteins in different tissues can be used to determine tissue-specific f_m or f_t to tease out the effect of DDIs and pharmacogenetic variability on local tissue concentration and toxicity. For example, MK-7246, a selective CRTH2 prostaglandin 2 antagonist, was discontinued from clinical development due to high variability in PK (Wang et al., 2012). Although MK-7246 is metabolized by multiple UGTs in the liver, retrospective pharmacogenetic data suggested 96% metabolism by UGT2B17 (f_m = 0.96), which could only be explained by the high intestinal abundance of UGT2B17 resulting in the high first-pass metabolism of the investigational drug (Wang et al., 2012). Similarly, proteomics data were important for explaining preferential MRP2-mediated biliary secretion of androgen glucuronides of testosterone and other androgens. Although vesicle data indicated the predominant role of sinusoidal MRP3 in the excretion of androgen glucuronides into the blood, the higher abundance of canalicular MRP2 suggested a preferential elimination of androgen glucuronides into the intestinal lumen and explained the enterohepatic recycling (Li et al., 2019a). Quantitative proteomics has also been used to assess the impact of genetic variability on clinical outcomes. For instance, the interindividual variability in the in vitro metabolism of (S)-mephenytoin and clopidogrel catalyzed by CYP2C19 in the human liver was successfully predicted by the changes in CYP2C19 protein abundance (Shirasaka et al., 2016).

Primary human hepatocyte culture is considered the gold standard for the evaluation of enzyme induction potential of new chemical entities, which relies on specific substrates for the activity assays. However, the selective probe substrates are often not available, especially for the highly homologous proteins, such as CYP3A4, CYP3A5, and CYP3A7. Although mRNA expression could predict changes in transcription, it often does not provide quantitative changes in protein abundance. Quantitative proteomics serves as a better alternative to determine the enzyme or transporter induction potential because it offers multiplex protein quantification with greater throughput, sensitivity, and robustness as compared with the currently recommended methods (mRNA and activity assay) by the regulatory agencies. Quantitative proteomics was used for the accurate determination of the induction potential of CYP1A2, CYP2B6, CYP3A4, and CYP3A5 in human hepatocyte incubations (Williamson et al., 2011). Similarly, immunoaffinity-based proteomic approach was used in another study for the assessment of enzyme induction potential in a small sample amount (MacLean et al., 2017). Recent data further validate the use of quantitative proteomics for assessing the induction of the enzyme protein abundance in human hepatocytes, which reasonably precited the in vitro enzyme activity data using probe substrates (Stresser et al., 2021; Savaryn et al., 2022). Because protein quantification better reflects the functional activity as compared with mRNA, multiplexed DMET quantification by quantitative proteomics has the potential to screen the protein induction and suppression potential during routine early drug discovery stages.

C. Prediction of Interindividual Variability

Drug ADME and response is influenced by several population factors such as age, sex, genotype, disease conditions, and other health conditions like pregnancy. However, due to ethical, logistical, and financial reasons, clinical trials are typically performed on a limited number of adult subjects. With the emergence of PBPK modeling, potential interindividual variability in drug metabolism and PK can be predicted using virtual populations as far as the physiologic data like DMET protein abundance in different populations are known. Table 2 provides a compilation of the quantitative proteomics studies used for predicting interindividual variability in drug disposition. The principle of drug clearance translation from healthy adults to a special population, such as pediatric, depends on the fact that most of the extrinsic and intrinsic factors (Fig. 3) affect protein abundance or V_max without influencing K_m. Therefore, characterization of the influence of population factors on DMET abundance is critical information for a successful prediction of the interindividual variability using PBPK modeling.

TABLE 2.

Successful examples of the application of quantitative proteomics for the characterization of interindividual variability in DMET abundance

Enzyme/Transporter	Fold Change	Reference
Effect of age (liver): fold change (relative to infants)
CYP1A2	>2-fold (adults); >1- to 2-fold (juveniles)	Sadler et al., 2016
CYP2A6	>2-fold (adults)
CYP2B6	>2-fold (juveniles)
CYP3A5	>2-fold (juveniles, adults)
UGT1A1	>1- to 2-fold (adults)	Bhatt et al., 2019
UGT1A4	>2-fold (children, adolescents, adults)
UGT1A6	>2-fold (children, adolescents, adults)
UGT1A9	>2-fold (adults)
UGT2B7	>2-fold (adolescents, adults)
UGT2B15	>2-fold (adults)
UGT2B17	>2-fold (children, adolescents, adults);	Bhatt et al., 2018
ALDH1A1	>1- to 2-fold (children, adolescents)	Bhatt et al., 2017
ADH1B	>2-fold (children); >1- to 2-fold (adolescents, adults)
ADH1C	>2-fold (children, adolescents, adults)
CES1	>2-fold (adults); >1- to 2-fold (children, adolescents)	Boberg et al., 2017
FMO3	>1- to 2-fold (children, adult)	Zane et al., 2018
SULT1A1	>1- to 2-fold (children)	Ladumor et al., 2019a
SULT1A2	>1-- to 2-fold (adolescents, adults)	Dubaisi et al., 2019
SULT1B1	>2-fold (infants, children, adolescents, adults)	Ladumor et al., 2019a)
OCT1	>1- to 2-fold (adolescents, adults)	Prasad et al., 2016; van Groen et al., 2018
OATP1B3	>1- to 2-fold (children, adolescents, adults)	Prasad et al., 2016; van Groen et al., 2018
P-gp	>1- to 2-fold (adolescents, adults)	Prasad et al., 2016; van Groen et al., 2018
MRP3	>1- to 2-fold (adult)	Prasad et al., 2016; van Groen et al., 2018
Effect of age (kidney): fold change (relative to infants)
P-gp	>1- to 2-fold (children, adults)	Cheung et al., 2019
URAT1	>2-fold (children)
OAT1	>1- to 2-fold (children, adults)
OAT3	>1- to 2-fold (children, adolescents, adults)
OCT2	>1- to 2-fold (children)
Effect of genotype: fold change (relative to the corresponding reference allele)
BCRP	<0.5-fold (rs2231142 (C421A))	Prasad et al., 2013
MRP2	>1- to 2-fold (SNP 21214G>A (V417I)	Deo et al., 2012
OATP1B1	>2-fold (SNP c.463C>A); >1- to 2-fold (388 A>G, c.597C>T)	Prasad et al., 2014
CYP2A6	>1- to 2-fold (*1/*4, ↑ *1/*9); <1–0.5-fold (*1/*17, *1/*2)	Tanner et al., 2017; Tanner et al., 2018
FMO3	>1- to 2-fold (E158K:G308G)	Xu et al., 2017
UGT2B17	>1- to 2-fold (*1/*2); < 0.5-fold (*2/*2)	Gaedigk et al., 2018
Effect of sex: fold change (male versus female adults)
BCRP	>1- to 2-fold	Prasad et al., 2013
UGT2B17	>2-fold	Tanner et al., 2017
CYP2A6	<1- to 2-fold
FMO3	>1- to 2-fold	Xu et al., 2017
Effect of disease condition: fold change (relative to healthy adults)
Alcoholic liver disease (CPS A)	>2-fold (P-gp); < 0.5-fold (BSEP, MRP2, OAT2, OATP1B1, OATP2B1)	Wang et al., 2016; Drozdzik et al., 2020
Alcoholic liver disease (CPS B)	>2-fold (P-gp); <0.5-fold (BSEP, MRP2, NTCP, OCT1, OCT3, OAT2, OATP1B1, OATP1B3, OATP2B1)	Drozdzik et al., 2020
Alcoholic liver disease (CPS C)	>2-fold (P-gp, MRP3, BCRP, OCT3); <0.5-fold (MRP2, NTCP, OAT2, OATP1B1, OATP2B1)	Drozdzik et al., 2020
Nonalcoholic fatty liver disease	<0.5-fold (CYP3A4)	Ali et al., 2017
Nonalcoholic steatohepatitis	>2-fold (OATP1B1, MRP4); <0.5-fold (NTCP, OATP1B3)
HCV	>2-fold (MRP3, MRP4, P-gp); <0.5-fold (CYP3A4, CYP2C9, CYP2E1, CYP1A2, CYP2A6, CYP2C8, UGT1A4, UGT1A6, UGT2B7, UGT2B15, CES1, ADH1B, ADH1C)	Wang et al., 2016
HCV (CPS A)	>2-fold (P-gp); <0.5-fold (BSEP, MRP1, MRP2, BCRP, NTCP, OCT1, OCT3, OATP1B1, OATP2B1)	Drozdzik et al., 2020
HCV (CPS B)	>2-fold (P-gp, MRP1); <0.5-fold (BSEP, MRP2, OCT1, OCT3, OATP1B3, OATP2B1)
HCV (CPS C)	>2-fold (P-gp, MRP3); <0.5-fold (BSEP, MRP1, NTCP, OCT1, OCT3, OATP1B3, OATP2B1)
Primary biliary cirrhosis; PSC (CPS A)	>2-fold (P-gp, BSEP, MRP3, OCT3, OATP1B3); <0.5-fold (MRP2)
PSC (CPS B)	>2-fold (P-gp, MRP3, OCT3); <0.5-fold (MRP2, OATP1B3)
Autoimmune hepatitis (AIH) (CPS A)	>2-fold (P-gp, MRP3, BCRP, OCT3); <0.5-fold (MRP1, MRP2)
AIH (CPS B)	>2-fold (P-gp, MRP3, BCRP, OCT3); <0.5-fold (MRP2)
AIH (CPS C)	>2-fold (P-gp, MRP3, OCT3); <0.5-fold (MRP2, NTCP, OCT1, OATP1B1)
Ulcerative colitis	<0.5-fold (MCT1)	Erdmann et al., 2019

Fig. 3.

Effect of internal and extremal factors on CL_int of drugs via changes in DMET protein abundance or substrate affinity. Most of these factors affect the velocity maximum (V_max) by altering protein abundance (E), whereas substrate affinity (K_m) can be affected by changes in amino acid sequences at active site by factors such as PTMs or nonsynonymous SNPs affecting active site. K_cat, turnover number; miRNA, microRNA.

1. Effect of Age

It is well recognized that children are not small adults. Age is associated with multifactorial physiological changes, especially during early life, and the abundance of different DMET proteins changes with age in a nonmonotonic fashion, resulting in variable drug clearance and unpredictable drug toxicity or efficacy. For example, the underdevelopment of UGT enzymes has been shown to correlate with the immature glucuronidation activity in neonates and infants, which could lead to toxicity of UGT substrates when prescribed based on body weight or body surface area–normalized doses. Chloramphenicol caused cardiovascular collapse characterized by Gray baby syndrome in neonates due to the poor glucuronidation capacity of younger children (Mulhall et al., 1983). Similarly, lower levels of UGTs are associated with higher exposure of zidovudine in neonates and infants, leading to hematological toxicity (Capparelli et al., 2003). Further, age-dependent maturation of DMET abundance could confound the effect of genetic variability in the case of highly polymorphic enzymes or transporters (e.g., CYP2D6, UGT2B17, OATP1B1, and OCT1). The well-documented adverse effects of codeine in breastfed infants can be explained by extensive maternal CYP2D6 activity as well as the low abundance of the detoxifying enzyme UGT2B7 in infants (Bhatt et al., 2019).

Age-dependent changes in DMET quantitative proteomics data in the human liver are compiled in Table 2 and Fig. 4. The developmental patterns of DMET proteins are generally characterized into three groups. Some proteins show the highest abundance in neonates which decreases with age. For example, CYP3A7 sharply decreases during the first year of life and is rarely detected in adults. Similarly, BCRP is decreased with age in the human liver, but the change in its abundance remains within ∼2-fold between neonates and adults. The second group of DMET proteins is relatively constant (<1.5-fold change) throughout the development. Examples of these proteins include CYP2C9, CYP2E1, CYP3A5, SULT1A1, SULT2A1, OATP1B1, OATP2B1, NTCP, BSEP, and MATE1. However, the majority of DMET proteins belong to the third group, which mature with age (group 3). The group 3 proteins include CYP1A2, CYP2A6, CYP2B6, CYP2C8, CYP3A4, UGT1A1, UGT1A4, UGT1A6, UGT1A9, UGT2B7, UGT2B15, UGT2B17, ADH1A, ADH1B, and ADH1C and show a >5-fold difference in the abundance from neonatal stage to adulthood. These ontogeny data, which are now routinely derived using quantitative proteomics, have been integrated into PBPK software such as Gastroplus and Simcyp for the extrapolation of adult PK data to children. Since the regulatory agencies are encouraging clinical trials in children (https://www.nichd.nih.gov/research/supported/bpca), protein abundance–informed pediatric PBPK modeling is important for designing safer clinical studies in children.

Fig. 4.

Age-dependent changes in the abundance of CYPs (A), UGTs (B), transporters (C), and other non-CYPs (D) in the liver from human donors representing neonatal, infants, early-childhood, middle-childhood, adolescents, and adults.

2. Effect of Genetic and Epigenetic Factors

Genetic polymorphisms in DMET proteins contributes significantly to the PK variability of drugs. In general, there are approximately 14–16 million SNPs that are observed every 300–1000 nucleotides across the human genome (Roden and George, 2002). The V_max of metabolism/transporter is affected by a direct effect on the protein abundance as a result of changes such as a mutation in transcription factor–binding site or promotor region, insertion of a new stop codon in an exon, gene deletion, loss of start codon, mRNA splicing, or protein degradation (Fig. 3), although the nonsynonyms SNPs can also affect K_m. The influence of genetic polymorphism on drug PK may require dosing adjustment for certain individuals, especially for the narrow therapeutic index drugs (Collins, 1991). For example, genetic variations in patients have been shown to be associated with unpredictable toxicity or the lack of efficacy (Shuldiner et al., 2009; Avivi et al., 2014). Table 2 presents representative examples of clinically relevant SNPs, copy number variation (CNV), and epigenetic changes that are commonly associated with population variability in drug PK and/or response.

Genetic polymorphisms that are caused by variations at a SNP or haplotype in a gene of a DMET protein are one of the most common causes of interindividual variability in PK. For example, CYP2D6 exhibits more than 100 genetic variants due to point mutations, duplications, and insertion or deletion of single or multiple nucleotides. Individuals with different CYP2D6 allelic variants are classified as poor metabolizers (PMs), intermediate metabolizers (IMs), extensive metabolizers (EMs), and ultrarapid metabolizers (UMs), corresponding to the extent of involvement of these variants in drug metabolism (Zhou et al., 2008). For example, tamoxifen, a selective estrogen receptor (ER) modulator used in ER⁺ breast cancer patients, is extensively metabolized by CYP2D6 to endoxifen (Filipski et al., 2014). Due to the genetic variability in CYP2D6, ultrarapid metabolizers experience a higher plasma level of endoxifen than poor metabolizer and intermediate metabolizer patients and encounter severe hot flashes and a higher risk of disease relapse (Goetz et al., 2005). In addition, SNPs in the coding region of an enzyme or transporter could change the K_m without changing the protein abundance. For example, CYP2D6*17 allele, which carries T107I, R296C, and S486T mutations, leads to a reduced function of CYP2D6 by increasing the K_m (decreased affinity) for bufuralol and dextromethorphan (Marcucci et al., 2002).

CNV is a structural variation in which a section of the gene is repeated or deleted (∼1 kb to 3 Mb). CNVs are reported for several DMEs such as CYP2A6, CYP2C19, CYP2D6, CYP3A5, UGT2B17, GSTM1, GSTT1, and SULT1A1 (Ménard et al., 2009; Zhou et al., 2010; Santos et al., 2018). Interestingly, UGT2B17 shows a very high ethnic variability in the frequency of individuals with the gene deletion (*2/*2) (Xue et al., 2008), i.e., ∼25% in the African and European populations but >70% in the East Asian population (e.g., >90% in the Japanese population). Not only CNV, but several SNPs (rs9996186, rs4860305, and rs28374627) are also associated with UGT2B17 abundance (Bhatt et al., 2017). Similarly, the changes in CYP2C19 abundance due to genetic polymorphisms strongly correlated (r² = 0.984) with (S)-mephenytoin hydroxylase activity (Fig. 5A) (Shirasaka et al., 2016). The activity score of CYP2D6 also showed a reasonable correlation with the protein abundance (Fig. 5B) (Gaedigk et al., 2018). Several other proteins such as OATP1B1, BCRP, MRP2, and FMO3 show a strong association of genetic polymorphism with protein abundance that correlates with the in vivo activity (Table 3). Like the expression quantitative trait loci (eQTLs) for the association of genotype and transcript levels, the protein abundance quantitative trait loci (pQTLs) can be used as a noninvasive approach to assess the effect of pharmacogenomic variability. The epigenetic modifications such as DNA methylation, histone modification, and microRNA (miRNA) also contribute to interindividual variability in drug PK due to their effects on DMET protein abundance (Kacevska et al., 2012). For example, miR-491-3p regulates UGT1A3, UGT1A6, and UGT1A1 expression and alters enzyme activity (Dluzen et al., 2014). Thus, integration of quantitative proteomics data into PBPK modeling is an important noninvasive approach to predict the effect of genetic and epigenetic variability on drug PK and response, hence providing a promising approach for precision medicine.

Fig. 5.

Activity and protein abundance correlation of CYP2C19 (A) and CYP2D6 (B) in human liver microsomes (HLMs) (Shirasaka et al., 2016; Gaedigk et al., 2018).

TABLE 3.

Common single nucleotide polymorphism in DMET proteins that affect their activity or abundance

Protein	Nucleotide Change; Allele; Amino Acid Change (rs Number)	MAF (Caucasians, %)	Activity (Abundance)	Clinical Application	References
CYP1A1	3801 T>C; *2A; I462V (rs4646903)	19	Increased (unknown)	Predictor for clinical outcome in liver cancer patients treated with EGFR-TKI therapy	Zhou et al., 2009
CYP1A2	−163C>A; *1F; NA (rs762551)	33.3	Increased (increased)	Associated with the risk for colorectal adenomas in humans	Sachse et al., 2003; Moonen et al., 2005; Pilgrim et al., 2012
	−246T>delT; *1D; NA	4.82	Decreased (decreased)	Higher olanzapine plasma concentration	Sachse et al., 2003; Uslu et al., 2010; Czerwensky et al., 2015
CYP2C9	430 C>T; *2; R144C (rs1799853)	19.0% *1/*2 1.6% *2/*2 1.8% *2/*3	Decreased (unknown)	Poor warfarin metabolism	Aithal et al., 1999; Hillman et al., 2004; Lindh et al., 2009
	1075A>C; *3; I359L (rs1057910)	9	Decreased (unknown)	Decrease metabolism of phenytoin compared with reference genotype	Scordo et al., 2001; van der Weide et al., 2001; Rosemary et al., 2006
CYP2C19	681G>A; *2; splicing variant I331V (rs4244285)	16	Decreased (unknown)	In Asians with ulcer treatment of Helicobacter pylori infection with omeprazole varying treatment effectiveness	Aynacioglu et al., 1999
	−806C>T and -3402C>T; *17; I331V (rs12248560)	18	Increased (increased)	Increase clinical response to clopidogrel treatment by better antiplatelet activity and higher risk of bleeding	Sibbing et al., 2010; Santos et al., 2011
CYP2D6	1749A>G, 2549delA; *3; frameshift mutation, N166D; 259 (rs1135824)	2.04	Decreased (decreased)	Decreased metabolism of risperidone	Sachse et al., 1997
	1846G>A; *4; P34S; L91M; H94R; Splicing variant; S486T (rs3892097)	20.7	Decreased (decreased)	Poor metabolism of tricyclic antidepressants may lead to increased toxicity in Caucasians	Sachse et al., 1997; Bijl et al., 2008
	100C>T; 1039C>T; 1661G>C; 1846G>A; 4180G>C; *4D; P34S; Splicing variant; S486T (NA)	3.4	Decreased (decreased)		Shimada et al., 2001
	100C>T; 997C>G; 1661G>C; 1846G>A; 4180G>C; *4L; P34S; Splicing variant; S486T (NA)	4.5	Decreased (decreased)		Shimada et al., 2001
	whole gene deletion; *5; CYP2D6 deleted (NA)	4.1	Deletion (deleted)	Require dose adjustment of CYP2D6 substrates	Griese et al., 1998
	1707delT; *6; frameshift mutation, 118F (rs5030655)	1.3	Nonfunctional (decreased)	Person may experience haloperidol-induced extrapyramidal side effects	Griese et al., 1998; Šimić et al., 2016)
	2935A>C; *7; H324 (rs5030867)	1	Decreased (unknown)		Evert et al., 1997
	2615-2617delAAG; *9; K281del (rs5030656)	2	Decreased (unknown)		Griese et al., 1998
	100C>T; *10; P34S; S486T (rs1065852)	8	Decreased (unknown)		Griese et al., 1998; Raimundo et al., 2004
	2988G>A; *41; R296C; Splicing variant; S486T (rs28371725)	8	Decreased (decreased)		Raimundo et al., 2004
CYP2A6	NA; *12; substitutions 10aa (esv2663194)	2.9	Decreased (unknown)		Haberl et al., 2005
	NA; *1B; NA;NA	32.6	Increased (unknown)	Faster nicotine metabolism	Mwenifumbo et al., 2008; Djordjevic et al., 2013
	1799 T > A; *2; L160 (rs1801272)	2.3	Decreased (unknown)	Shifting of coumarin metabolism from 7-OH coumarin to 2-hydroxyphenylacetic acid	Hadidi et al., 1997; Ingelman-Sundberg, 2001
	NA; *4; deletion		Deletion (deleted)	Impairment of nicotine metabolism	Ingelman-Sundberg, 2001
	− 48 T > G; *9; TATA box (rs28399433)	7.1	Decreased (decreased)		Schoedel et al., 2004
CYP3A4	566T>C; *17; F189S (rs4987161)		Decreased (unknown)	Decreased metabolism of nifedipine	Dai et al., 2001
	NA; *1B; NA (rs2740574)	17	Increased (unknown)	Higher docetaxel metabolism	Kadlubar et al., 2003; Tran et al., 2006
	664T>C; *2; S222P (rs5585340)	2.7	Decreased (unknown)		Sata et al., 2000
CYP3A5	6986A>G; *3C; Splicing defect (rs776746)	81.3	Decreased (decreased)	Prognosis of non–small cell lung cancer patients undergoing chemotherapy and surgery	Quaranta et al., 2006; Jiang et al., 2016
	6981A>G, 29748T>C; *3k / *10; Splicing defect; F446S (rs41279854)	2.0	Decreased (decreased)		Lee et al., 2003
CYP2B6	64C>T; *2; R22C; rs8192709	5.3	Increased (unknown)		Hiratsuka et al., 2002; Honda et al., 2011
	1459C>T; *5; R487C (rs3211371)	14.0	Decreased (unknown)	Decreased S-mephenytoin N-demethylase activity	Lang et al., 2001
	785A>G; *4; K262R (rs4802101)	5.0	Increased (unknown)		Lang et al., 2001; Kirchheiner et al., 2003
	516G>T, 785A>G; *6; Q172H; K262R (rs3745274, rs2279343)	25.2	Decreased (unknown)		Lang et al., 2001; Kirchheiner et al., 2003
	−1456T>C, -750T>C, .785A>G, 1459C>T; *7; Q172H; K262R; R487C (rs3745274, rs2279343, rs3211371)		Decreased (unknown)	Decreased S-mephenytoin N-demethylase activity	Hiratsuka et al., 2002; Lamba et al., 2003
	Gene deletion, *2; NA (rs10025771)		Decreased (decreased)		Bhatt et al., 2018
UGT2B17	−155G>A,840A>T, 788G>T, 2603T>G; H2/H2; NA (rs7436962, rs9996186, rs4860305, and rs28374627)	22	Increased		Bhatt et al., 2018
UGT1A1	NA; *28; TATA box (rs8175347)	26–31	Decreased (decreased)	Decreased glucuronidation of bilirubin to about 30% of wild-type levels, adverse reactions to irinotecan treatment	Bosma et al., 1995; Marques and Ikediobi, 2010
SULT2A1	187G>C, 679A>G, 781G>A; NA; A63P, K227D and A261T (rs6639786, rs2270112, and rs17268988)		Decreased (NA)	Decreased metabolism of tibolone	Thomae et al., 2002; Miller et al., 2018; Wong et al., 2018
GSTP1	313A>G; NA; I105V (rs1695)	0.08–0.33	Decreased (unknown)	Alters the pharmacokinetics of cyclophosphamide	Sharma et al., 2014
UGT2B7	802C>T; *2; H268Y (rs7439366)	50	Decreased (NA)	Higher exposure of epirubicin	Bhasker et al., 2000; Parmar et al., 2011; Uchaipichat et al., 2013
UGT2B15	253G>T; *2; D85Y (NA)	50	Decreased (NA)	Interindividual variability in the clearance of oxazepam and lorazepam
BCRP	421C>A; NA; Q41K (rs2231342)	4.5–12	Decreased (decreased)	AUC of sulfasalazine, simvastatin, and rosuvastatin >2-fold higher in comparison with reference allele	Prasad et al., 2013; Tanaka et al., 2015
OATP2B1	388A>G, 521T>C and 388A>G; c.521T>C, Asn130Asp; p.Val174Ala; *1b, *5, and *15; N130D, V174A, (rs2306283, rs2306283)	30, 16 and 14	Decreased (NA)	Interindividual variability in statin therapy	Gao et al., 2017; Mori et al., 2019
OATP1B1	463C>A; *14; R155C (rs149535236)	2	Increased (increased)	Cerivastatin-induced rhabdomyolysis	Couvert et al., 2008; Romaine et al., 2010; Tamraz et al., 2013
OCT1	181C>T, 1201G>A, 1393G>A, 1257delATG; *2, *3, *4, *5; R61C, G401S, G465R and deletion of M420 (rs12208357, rs34130495, rs34059508, rs35167514)	9, 3, 16	Decreased (NA)	Impaired metformin uptake and lower morphine clearance in pediatrics	Saadatmand et al., 2012; Fukuda et al., 2013; Arimany-Nardi et al., 2015
OCT2	808G>T; NA; A270S (rs316019)		NA (NA)		Meyer et al., 2017)

3. Effect of Sex

Sex-dependent changes in DMET protein abundance can also contribute to variability in human PK study, drug efficacy, and toxicity. Erythromycin is primarily metabolized by CYP3A4 in the human liver to N-demethylated metabolite, which showed 40% more metabolism in women than men. Other CYP3A4 substrates such as verapamil, nifedipine, mirtazapine, cyclosporine, and diazepam also show higher metabolic clearance in women (Nicolas et al., 2009). Regarding phase 2 metabolism, acetaminophen clearance is 22% higher in men than women due to the higher glucuronidation rate (Miners et al., 1983), whereas aspirin exhibits higher bioavailability in women due to the reduced conjugation by glucuronic acid and glycin (González-Correa et al., 2007). The levels of UGT2B17 in the human liver quantified by quantitative proteomics showed ∼3-fold higher abundance in men (Bhatt et al., 2018). Likewise, BCRP, CYP2A6, and FMO3 showed ∼1- to 2-fold higher abundance in men (Table 2). Although the sex differences in DMET proteins are small in humans, more dramatic sex-dependent changes have been reported in rodents. In particular, quantitative proteomics data revealed the highest sex-dependent changes in kidney transporters in mice, followed by rats and dogs. Kidney OAT1 is ∼3.2- and 1.3-fold higher in male mice and rats, respectively, whereas OAT2 is 1.6-fold higher in male mice than in female mice. Mdr1 is 2- and 1.4-fold higher in the kidneys of female mice and dogs, respectively (Basit et al., 2019). These proteomics data in preclinical species correlate with the in vivo PK and toxicity data (Diamond et al., 2010; Lolkema et al., 2015). Overall, quantitative tissue proteomics data on sex-dependent differences in DMET proteins in both human and preclinical species are emerging to fill this critical knowledge gap.

4. Effect of Disease States

Disease states such as cirrhosis, chronic kidney disease (CKD), cancers, gastrointestinal disorders, autoimmune diseases, and diabetes can alter PK by changing blood flow to organs, plasma protein binding, and apparent volume of distribution, as well as the protein abundance of DMET (Sharma et al., 2020). In particular, the changes in DMET protein abundance can lead to a subtherapeutic response, drug toxicity, DDI, and variable PK profile of drugs. However, it is both ethically and logistically challenging to measure the impact of a disease condition on PK/PD and measure such changes in clinical trials in the disease population. For example, mild, moderate, and severe hepatic or renal dysfunction differentially impacts ADME processes. Since biobanks of postmortem tissues from deidentified donors are becoming increasingly available for research purposes, quantitative proteomics is expected to find broader applications in characterizing disease effects on DMET proteins.

Hepatic dysfunction is one of the major diseases in the United States affecting ∼2% adult population (https://www.cdc.gov/nchs/fastats/liver-disease.htm). The causes of hepatic dysfunction are diverse, including chronic alcohol abuse, viral/bacterial infection, fatty liver disease, and drug-induced injury. Liver diseases significantly affect PK of drugs cleared via hepatic metabolism, albeit the change in the PK depends upon disease severity, mechanisms of drug elimination, and alteration in f_m and f_t. Understanding the impact of various liver diseases on DMET abundance is important to predict drug disposition in a patient with liver disease. The quantitative changes in DMET proteins in hepatic dysfunction are summarized in Table 2. In general, the abundance of DMEs is decreased in the disease state, whereas transporter protein abundance is variable (increased or decreased). The Child-Pugh Class C liver damage is associated with an increase in the expression of P-gp and MRP4 by >200% and a decrease in MRP2, NTCP, OCT1, OATP1B1, and OATP2B1 by <25%–46%. Whereas in the alcoholic liver disease, MRP2, NTCP, OAT2, OATP1B1, OATP1B3, and OATP2B1 are downregulated by 26%–76%. Other hepatic conditions such as Hepatitis C, primary biliary cholangitis, primary sclerosing cholangitis, and autoimmune hepatitis are associated with a reduction in BSEP and OATP2B1 and increase in P-gp and MRP4 (Drozdzik et al., 2020). Wang et al. (2016) used a targeted quantitative proteomics approach and observed a reduced transporter protein expression (except MRP3) in alcoholic cirrhosis patients. Similarly, the effect of nonalcoholic fatty liver disease (NAFLD) of DMET protein expression in steatosis and nonalcoholic steatohepatitis (NASH) liver condition revealed the reduced abundance of OATP1B1/1B3/2B1, OAT2, and NTCP in liver diseases, whereas MRP3 expression was elevated (Vildhede et al., 2020).

CKD is associated with changes in drug PK due to its effect on glomerular filtration rate, protein binding, water loading capacity, and changes in transporter abundance and activity. For example, mirabegron area under the plasma concentration-time curve (AUC) is increased by 1.7-fold in renal impairment patients due to the alteration in P-gp abundance (Dickinson et al., 2013). Similarly, pitavastatin (an OATP substrate) showed 1.4-fold elevated AUC in CKD patients (Morgan et al., 2012). Thus, CKD not only impacts renal elimination of drugs but also modulates the nonrenal clearance of drugs. The impairment of drug metabolism and transport in CKD is considered to be an effect of alterations in protein translations and enzyme degradation, reduced hemoprotein biosynthesis, cofactor depletion, and competitive inhibition of DMEs by circulating uremic constituents (Elston et al., 1993; Nolin et al., 2008). Tan et al. reported a 60% reduction in OPTP1B activity, which can modulate the clearance of nonrenally eliminated drugs such as pitavastatin, repaglinide, etc. (Tan et al., 2019). However, limited quantitative proteomics data are available on the effect of CKD on DMET proteins due to the difficulty in obtaining disease tissue samples for the study. Relying on clinical studies to monitor these multifactorial effects is a resource and cost-demanding effort. Therefore, quantitative proteomics of tissue samples coupled with PBPK modeling is being considered as a promising alternate to clinical studies for predicting disease effects (Prasad et al., 2018; Ladumor et al., 2019b; Rowland Yeo et al., 2020; Effinger et al., 2021).

D. Physiologically Based Pharmacokinetic Modeling

The use of PBPK modeling in drug development and regulatory filing is continuously increasing as evident by more than a twofold increase since the year 2013 (Perry et al., 2020). Although the concept of PBPK modeling is not new (Teorell, 1937), there has been tremendous progress in the application of PBPK models in the recent past. A reliable prediction of drug PK using PBPK modeling requires knowledge of quantitative physiologic data, including DMET abundance. Once validated, these models are considered promising in drug development, such as for dose optimization and clinical study design (Wang et al., 2019). Knowledge of the abundance and activity of DMET proteins in tissues such as the liver, kidney, and intestine is critical for accurate prediction of drug PK in healthy and special populations, such as patients with renal or hepatic impairment, children, and pregnant women (Al Feteisi et al., 2015; Prasad et al., 2017). The proteomics information has been integrated into PBPK modeling (Fig. 6) for predicting drug disposition in special populations (Table 4).

Fig. 6.

Schematic workflow of proteomics-informed PBPK model development and validation for extrapolation of interindividual variability in DMET abundance to predict the effect of population variability on drug pharmacokinetics from healthy adult population data. The model structure is based on a middle-out approach, and the reported clinical PK study data or in-house experimental data can be used as input parameters (i.e., system- and drug-specific parameters). CL, intrinsic clearance; HA, healthy adults; SP, special populations such as pediatric, pregnant women, geriatric, and patients with organ dysfunction/disease state.

TABLE 4.

Successful examples of proteomics informed PBPK modeling and IVIVE for predicting drug disposition

Drugs	Enzymes and/or Transporters Involved	Study Objective	Study Population	Predictive Accuracy (P/O fold ratio)	Software Used	References
Zidovudine	UGT2B7	Liver cirrhosis	Adult Child Pugh C subjects	0.97	Simcyp	Prasad et al., 2018
		Age-dependent PK	Neonates, infants and children	0.5–2*		Bhatt et al., 2019
Morphine	UGT2B7 and OCT1	Liver cirrhosis	Adult Child Pugh C subjects	0.97	Simcyp	Prasad et al., 2018
		Age-dependent PK	Neonates, infants and children	0.5–2*		Prasad et al., 2018
Acetaminophen	SULT1A1/1A3/1B1/2A1	Age-dependent PK and its effect on fraction of drug metabolized	Neonates and infants	0.98	GastroPlus	Ladumor et al., 2019a
Rosuvastatin	OATP1B1/1B3/2B1, NTCP	Effect of body weight on rosuvastatin PK	Adult healthy subjects with varying body weights	0.95	Semimechanistic model using R	Wegler et al., 2021
Lamotrigine	UGT1A3/1A4/2B7	Alcoholic cirrhosis	Adult hepatic impaired subjects	0.83	GastroPlus	Ladumor et al., 2019b
Oseltamivir	CES1/2	Age-dependent PK	Infants	0.5–2.1*	Simcyp	Boberg et al., 2017
Ethionamide	FMO3	Drug-drug interaction with methimazole and impact of genetic polymorphism	Healthy subjects	0.6–2.6*	Simcyp	Nguyen et al., 2019
Dexamethasone, betamethasone, darunavir, and lopinavir	P-gp	Maternal-fetal PK	Nonpregnant (control) and pregnant population	0.8–1.25	Simcyp	Anoshchenko et al., 2021
Digoxin	OATP4C1	Moderate and severe renal impairment	Renal impaired population	0.5–2	Simcyp	Scotcher et al., 2017
Cidofovir, oseltamivir, carboxylate and cefuroxime	OAT	Severe renal impairment	Renal impaired population	0.5–2	Simcyp	Hsu et al., 2014
Adefovir, avibactam, entecavir, famotidine, ganciclovir, oseltamivir, carboxylate and sitagliptin	OAT	Moderate and severe renal impairment	Renal impaired population	0.67–1.5	Simcyp	Hsueh et al., 2018

^*Range (multiple doses and route of administration were tested).

Pediatric PBPK modeling accounts for 20% of all Food and Drug Administration submissions (Zhang et al., 2020b). There are multiple examples of drugs (Table 4) with successful applications of DMET ontogeny data in predicting neonatal and infant PK. For example, a proteomics-based PBPK model successfully predicted the metabolic profile of acetaminophen (a substrate of UGTs, SULTs, and CYPs) in various age groups from neonates to adults (Ladumor et al., 2019a). The abundance data of UGT, SULT, and CYP enzymes were incorporated into the PBPK model to predict the age-dependent change in the fraction of acetaminophen metabolized by individual enzymes (f_m). Because f_m is directly proportional to the relative enzyme abundance, the nonmonotonic patterns of enzyme maturation with age is associated with the metabolic switching, i.e., a PBPK model predicted acetaminophen f_{m, ratio} values (i.e., f_m,UGT/f_m,SULT) of 0.46, 0.56, and 1.71 in neonates, children, and adults, respectively, which was consistent with the observed clinical data. A PBPK model was used to predict morphine PK in newborns (age 0 days) by using age-dependent changes in the organic cation transporters OCT1 and UGT2B7 (Balyan et al., 2017; Bhatt et al., 2019). Similarly, proteomics-based PBPK modeling has been applied to predict the impact of disease conditions on hepatic drug clearance of drugs such as bosentan, olmesartan, repaglinide, zidovudine, morphine, and telmisartan (Li et al., 2015; Prasad et al., 2018).

E. Characterization of Interspecies Differences in Drug Metabolism and Transport

Interspecies differences in physiologic parameters including DMET protein abundance/activity result in the poor translation of animal data. For example, biliary excretion of drugs or metabolites varies significantly in animals versus humans due to species differences in hepatic blood flow and bile flow along with transporter abundance, and it is difficult to predict drug or metabolites hepatic excretion of drugs and metabolites (Fallon et al., 2016). For example, an eightfold higher in vitro bile canalicular clearance of 2,4-dinitrophenyl-S-glutathione was observed in rats than dogs in hepatocytes (Ishizuka et al., 1999). Therefore, the characterization of interspecies differences in drug metabolism and transport is useful in (1) better translation of preclinical to clinical data, (2) distinguishing toxicity mechanisms, and (3) avoiding clinically relevant DDIs. Quantitative proteomics has been used to determine species differences in DMET proteins for better interpretation and scaling of animal data (Table 5). For example, the abundant expression of AO in human and monkey explain higher metabolite formation and toxicity in these species as compared with dogs and rats (Diamond et al., 2010; Basit et al., 2021). In particular, the ability to quantify multiple proteins and the high precision of the quantitative proteomics approach allows for an accurate determination of interspecies differences in DMET protein abundance, which can be integrated into the PBPK model to extrapolate PK parameters in humans. However, it is important to consider that not only the abundance of DMET proteins is different between animals and humans, but the K_m can also vary due to differences in the amino acid sequence at the active sites. Therefore, when scaling the intrinsic clearance data from animal to human, both K_m and abundance differences between the species must be considered as illustrated in eqs. 3–5. Where, K_cat is the turnover number, i.e. the number of times each enzyme site converts substrate to product per unit time and [S] is the substrate concentration.

F. Protein Quantification in Liquid Biopsies and Extracellular Vesicles

Biofluids such as blood and urine serve as liquid biopsies to characterize interindividual differences and offer a multitude of safety and logistical advantages due to the ease of access and the involvement of less or noninvasive procedures. Biomarker quantification in biofluids is a fundamental approach in drug discovery and development for monitoring drug effectiveness, safety, and toxicity. Since most biomarkers are proteins, quantitative proteomics has emerged as one of the most powerful techniques for biomarker discovery and quantification. For example, CD14 and α-fetoprotein were identified as potential biomarkers of hepatocellular carcinoma in high-risk cohort serum samples using iTRAQ-based quantitative proteomics (Guo et al., 2017). Similarly, the utility of quantitative proteomics in the characterization of protein induced by vitamin K antagonist-II (PIVKA-II) biomarker for blood clotting factor deficiency, vitamin K–related disorders, and diagnosis and monitoring of hepatocellular carcinoma (Basit et al., 2020). Further, the multiplex quantification of protein biomarkers in biofluids can provide information on changes in both individual proteins as well as biologic pathways perturbed by a drug or disease, hence allowing for systems-level information.

Protein quantification in extracellular vesicles (EVs) secreted in blood or urine from individual tissues is emerging as a promising approach in precision medicine. EVs are small extracellular, nonreplicating nanovesicles that are considered a rich source of biomarkers. EVs contain a collection of proteins, DNA, mRNA, and lipids derived from the cell of origin. EVs are broadly classified into three classes based on their sizes, such as exosomes (50–150 nm), microvesicles (100–1000 nm), and apoptotic bodies (50–5000 nm) (Yáñez-Mó et al., 2015), and they can be distinguished from other EVs from their surface-expressed set of tetraspanins (CD9, CD63, and CD81). The EV isolation technique typically involves (1) ultracentrifugation, (2) size-exclusion chromatography, (3) resin precipitation, and (4) membrane affinity chromatography (Fig. 7) (Rodrigues et al., 2021). The immunocapture or immunoprecipitation methods are particularly useful because of their ability to selectively enrich exosomes from a specific tissue using antibodies against a marker protein that is exclusively expressed in the tissue of interest.

Fig. 7.

Utilizing circulating small extracellular vesicles (sEVs) for predicting tissue DMET abundance for noninvasive prediction of drug disposition, DDI, and effect of genetic polymorphism. To isolate sEV as liquid biopsy from plasma, samples are centrifuge or enriched using size exclusion chromatography or immunocapture to obtain concentrated sEV collection from individual (Rodrigues et al., 2021).

Besides monitoring disease severity and drug response, EVs have been recently used to quantify DMET proteins. For example, the quantification of DMET proteins in EVs has been successfully used to quantify the effect of genetic polymorphism (Rodrigues and Rowland, 2019), DDI (Rodrigues and Rowland, 2019), and overall interindividual variability (Achour et al., 2021). The total plasma exosomes from rifampicin-dosed individuals have shown a strong correlation between CYP3A4 exosomal expression (mRNA and proteins) versus midazolam oral clearance (Rowland et al., 2019). Similarly, rifampicin-mediated induction of CYP3A4 in HepaRG cells was precited by analyzing exosomes. The utility of quantitative proteomics in liquid biopsy analysis has also been shown for the determination of the effect of pregnancy on DMET abundance (Rodrigues et al., 2021), which revealed that CYP3A4 is induced; however, CYP2D6 and OATP1B1 showed no apparent induction during pregnancy. Other studies investigated alcohol-induced oxidative stress (Cho et al., 2017) and drug-induced toxicity (Kumar et al., 2017a) using CYP2E1 in EVs or plasma exosomes as a marker. Similarly urinary exosomes were used to quantify makers of IgA nephropathy (Moon et al., 2011) and thin basement membrane nephropathy (Rahman et al., 2019). Thus, liquid biopsy is a promising technique, which, in conjunction with quantitative proteomics, has potential for broader applications in biomarker research. In particular, the use of liquid biopsies can be applied for patient stratification based on their individual ability to metabolize or excrete drugs for precise drug dosing. Because of the noninvasive nature of the liquid biopsy approach, the technique can also be potentially applied to characterize the effect of chronic diseases and the long-term use of medication on DMET proteins.

IV. Recommendations for the Reproducible Use of Quantitative Proteomics

Although quantitative proteomics is a promising technique to predict interindividual variability, these data can be confounded by the quality of tissue samples. Banked tissues can often be affected by confounding variables associated with harvesting, processing, and storage. Similarly, the presence of scarred tissue, medication use, and zonal (spatial) variability in DMET abundance can confound the determination of interindividual variability using banked tissues. Harmonized protocol of tissue collection and storage should be employed along with the collection of critical demographic and medical records to allow the comparison of data from different laboratories. In the case of a heterogenous tissue (e.g., placenta or kidney), the localization of DMET proteins is an important factor. For instance, the expression pattern of DMET proteins is different in the kidney cortex versus the medulla. Hence, to account for the contamination from the medulla in the cortex during fraction preparation, anatomic markers (e.g., aquaporins) can be used to account for the contamination.

The purity of subcellular fractions such as microsomes is often not characterized for contamination from other organelles, e.g., cytosolic contamination in microsomes. Quantitative proteomics can help in detecting contamination by measuring marker proteins, e.g., calnexin and calreticulin as markers of the ER membrane and ER lumen, respectively. CYPs are present in the ER membrane, and non-CYP enzymes are present in the cytosol or lumen of cellular organelles (Xu et al., 2018). Therefore, it is recommended that DMET proteins should be ideally quantified directly in tissue homogenate (Prasad et al., 2016). Optimization of tissue-specific homogenization methods, buffers, the volume of extraction buffer per gram of tissue, and optimized temperature conditions are important for generating reliable models. The use of EVs as a source of DMET protein quantification can be confounded by cell debris, which could be addressed by considering the shedding factor for normalization of DMET protein abundance (Achour et al., 2021).

Interday variability due to subtle changes in the sample preparation and LC-MS platforms may result in trypsin digestion variability. The use of an external protein standard such as bovine serum albumin as an internal control can resolve this issue (Bhatt et al., 2018). Solubility and stability of SIL peptides should also be checked and validated experimentally for the accurate estimation of peptide concentration in the biologic matrix. Also, contamination of light peptides in the heavy peptide standard should be checked to avoid false-positive results. Absolute protein abundance is affected by several factors such as trypsin digestion efficiency, quality of the calibrator peptides, and differential enrichment across laboratories. These challenges require that a common calibrator, for example, a well-characterized pooled human tissue sample, can be used to monitor such crosslaboratory variability. Digestion efficiency can also be assessed using codigestion of exogenously added proteins such as bovine serum albumin or QconCAT.

V. Conclusions and Future Directions

Applications of quantitative proteomics in translational ADME and precision medicine have been increasingly reported in the recent past. With its ability to support IVIVE and PBPK modeling, as well as characterization of in vitro models, intertissue, and interspecies DMET protein variability and disease progression, quantitative proteomics serves as a key tool from early stages of drug discovery to clinical development and pharmacotherapy. Furthermore, an evolving trend of the use of quantitative proteomics in liquid biopsy (exosomes) allows the noninvasive quantification of DMET proteins to monitor real-time disease progression and therapeutic response. Increased sensitivity of LC-MS and miniaturized sample preparation are expected to be useful in the analysis of DMET proteins in the samples harvested using microsampling techniques. High interlaboratory variability in DMET proteomics data are a concern, which requires collective efforts to develop robust protocols and quality controls. Further, there is a need for harmonized protocols for biospecimen collection, archiving, sample preparation, and MS data analysis to avoid crosslaboratory variability. A universal control sample can be developed to be shared across sites to ensure interlaboratory reproducibility. Global proteomics data are emerging in the literature that can be archived in well-open access libraries for future use.

Abbreviations

ADME

absorption, distribution, metabolism, and excretion

AO

aldehyde oxidase

AQUA

absolute quantification

AUC

area under the plasma concentration-time curve

BCRP

breast cancer resistance protein

BSEP

bile salt export pump

CES

carboxylesterases

CKD

chronic kidney disease

CL_int

intrinsic clearance

CNV

copy number variation

DDA

data-dependent acquisition

DDI

drug-drug interaction

DIA

data-independent acquisition

DMET

drug-metabolizing enzymes and transporter

ER

estrogen receptor

EV

extracellular vesicle

f_m

fractional contribution of an enzyme in drug metabolism

f_t

fractional contribution of a transporter in drug transport

IT

ion trap

iTRAQ

isobaric tags for relative and absolute quantification

IVIVE

in vitro to in vivo extrapolation

K_m

substrate affinity

LC-MS/MS

liquid chromatography–mass spectrometry

m/z

mass-to-charge

MPPGL

milligram of protein per gram of liver tissue

MPS

microphysiological systems

MRM

multiple reaction monitoring

OAT2

organic anion transporter 2

OCT1

organic cation transporter 1

PBPK

physiologically based pharmacokinetic

PCA

principal component analysis

P-gp

P-glycoprotein

PK

pharmacokinetics

PQC

positive quality control

PRM

parallel reaction monitoring

PSAQ

protein standards for absolute quantification

PTM

posttranslational modification

Q

quadruple

QconCAT

quantitative concatemers

RAF

relative activity factor

REF

relative expression factor

SIL

stable isotope-labeled

SNP

single nucleotide polymorphism

SWATH

sequential windowed acquisition of all theoretical fragment ions

TMT

tandem mass tags

TOF

time-of-flight

TPA

total protein approach

UGT2B7

UDP-glucuronosyltransferase 2B7

V_max

velocity maximum.

Authorship Contributions

Participated in research design: Ahire, Prasad.

Performed data analysis: Ahire, Kruger, Sharma, Mettu, Basit, Prasad.

Wrote or contributed to the writing of the manuscript: Ahire, Kruger, Sharma, Mettu, Basit, Prasad.