HIV Pharmacology Data Repository (PDR): Setting the New Information Sharing Standard for Clinical and Preclinical Pharmacokinetic Studies

Abstract

Drug development approaches increasingly harness computational modeling to predict drug behavior. These in silico approaches, collectively termed “pharmacometrics”, have significant value in deriving biological meaning from analysis of pooled drug concentration vs. time (CvT) datasets. However, the field lacks standardization for pharmacokinetic data description, requiring expert annotation to enable aggregate mining and sharing. These limitations impede data sharing and preservation as mandated by current NIH policies. To this end, we propose a minimum information standard for pharmacokinetic studies composed of three categories (Intervention, System, and Concentration). We implement this standard in the development of a web-based database: the HIV Pharmacology Data Repository (HIV PDR). We describe our technical approach for creating the HIV PDR, the protocols we established for standardized data deposition, and the current content of the database. We also demonstrate the utility of the HIV PDR for pharmacometrics research through computational modeling of CvT data extracted from this new database. Based on these efforts, we propose the HIV PDR as a standard to preserve and share pharmacokinetic data generated through preclinical and clinical HIV research.

Introduction

The implementation of GenBank in 1982 established the feasibility of sharing large-scale datasets generated through the Human Genome Project and beyond.^1,2 Around the same time, HIV was recognized as a unique pathogen with high genetic variability, thus prompting the development of a specialized resource, the HIV Sequence Database, to address what was anticipated to be a complex research need.³ Thus, the field of HIV was at the vanguard of a data-sharing movement that has expanded beyond genomics to include multiple databases created across biomedical disciplines. The rise of these databases reflects a widely adopted philosophy that data-sharing accelerates biomedical discovery. In 2023 the NIH solidified the future of data-sharing across biomedical fields, through the revised Policy for Data Management and Sharing (NOT-OD-21–013). In this policy the agency mandates that investigators “promote the management and sharing of scientific data” generated from all NIH-supported research.⁴ The policy also strongly suggests preserving data through established repositories. To support compliance with the new policy, the NIH has created a list of public data-sharing resources.⁵ The proposed list is extensive but not exhaustive, and there is an expressed need to establish novel repositories for disciplines where existing platforms are inadequate or lacking altogether. Pharmacokinetics is an example of such a discipline as searching the aforementioned NIH resources⁵ with the keywords “pharmacokinetics”, “pharmacology”, and “drug” produced no relevant hits.

Although data sharing resources for “big data” do exist for genomics and proteomics, as well as small molecule characterization,⁶ the field of clinical pharmacology lacks a comprehensive, real-time resource for sharing raw drug concentration vs. time (CvT) data obtained from preclinical and clinical pharmacokinetic (PK) studies, including mechanistically relevant tissue and cellular values. The only three existing repositories we know of from searching FAIRSharing⁷ and PubMed⁸ are either limited to clinical plasma data⁹, have an environmental toxicokinetic focus¹⁰, or have a specialized focus (i.e. HIV) but with limited breadth¹¹, and none have demonstrated a clear ability for real-time data sharing.

The small number and limitations of existing drug CvT databases may reflect the slowness of a transition away from traditional field practices where PK data are shared predominantly through journal publications in nonmachine-readable format as brief data summaries or graphs rather than complete datasets. When PK data are published, they may contain nomenclature inconsistencies or omission of critical elements. These practices hinder innovations in PK data modeling. Over the last decade, model-informed drug development methods have challenged the earlier drug development paradigm where clinical pharmacology knowledge is derived from siloed studies with limited applicability outside the context of individual study design. In silico research approaches such as pharmacometrics integrate data collected from a wide range of study designs and populations into robust datasets enabling population-level and cross-species extrapolation, as well as systems biology modeling approaches.^12–14 To support these efforts, the International Society of Pharmacometrics (ISoP) established a Standards and Best Practices Committee in 2012 to provide recommendations for standard pharmacometric analyses, and the first guidance document was published in 2023 for population PK modeling.^15,16 Yet, the rate-limiting step of such therapeutic innovations is the ability to aggregate and harmonize high quality pharmacokinetic data collected over a range of species or populations and anatomical compartments.

Here we introduce a minimum information standard for CvT data in humans and animals to facilitate PK data sharing and help define the framework for database design. The concept of reporting minimal information standards has been gaining popularity in different research disciplines, with examples including Minimum Information About a Bioactive entity (MIABE),¹⁷ Minimum Information About a Microarray Experiment (MIAME),¹⁸ and Minimal Information About Nanomaterials (MIAN).¹⁹ The new standard we propose here addresses gaps in the International Society of Pharmacometricians’ (ISOP) Basic Data Structure for Population PK Analysis¹⁶ to guide data sharing for PK studies comprehensively, including CvT data generated in animals and in therapeutically relevant physiological spaces. Our proposed standard consists of key reportable variables divided into 3 categories following the operative structure of PK studies: Intervention (drugs administered, including how and when they are administered at a given quantity), System (the species the intervention is administered to, including anatomical compartments being sampled), and Concentration (small-molecule concentrations measured upon intervention, including prodrugs and drug metabolites, using the gold-standard methodology LC-MS/MS). We apply this standard in curating CvT data collected from previously siloed studies into a user-friendly database named the HIV Pharmacology Data Repository (HIV PDR). We position this database as a critical research resource to support data sharing, management, mining, and modeling for the community of translational scientists working to optimize HIV therapeutics. Finally, we demonstrate the scientific utility of our approach in a proof-of-concept pharmacometrics modeling exercise using CvT data extracted from the HIV PDR.

The Minimum Information Standard

Intervention

The Intervention category is composed of two portions of information belonging to the operative terms of “pharmaco” and “kinetics”. Thus, the Intervention category sets the stage for a PK study, capturing the drug being administered and its dose, route, and frequency of administration (Table 1), as well as basic formulation attributes to delineate between immediate- or extended-absorption profiles. These key reportable variables compose the “pharmaco” portion of a PK study. Variables related to time, the “kinetic” portion of a PK study, include time elapsed since administration of the last (most recent) dose (for generating, or interpreting data via, a PK concentration-time curve) and time elapsed since the first dose (for identifying the probability that a steady-state condition has been achieved).

Table 1:

Minimum Information Standard for PK Studies and Its Expansion of the ISOP PopPK Basic Data Structure^a

Category and Variables	Units	Requirements or Preferences	Corresponding ISOP PopPK Variable
Intervention Category
“Pharmaco”-Related Variables
Administered Drug	n/a	Generic name preferred, with any coformulations disclosed	Not defined
Administered Dose	mg, mg/kg, μg, μg/kg, μg/day	May be prescribed/intended dose (DOSEP) in lieu of Actual Treatment Dose or Actual Amount of Dose Received if actual values are unknown	DOSEA, Actual Treatment Dose or AMT, Actual Amount of Dose Received
Weight (human/animal)	kg	Required for total dose calculations when Administered Dose is mg/kg or μg/kg	WT, Body Weight
Route of Administration	n/a	Includes formulation attributes such as long- vs. short-acting and implants	ROUTE, Route
Frequency of Administration (multiple dosing)	n/a	qXXhour or qXXweek nomenclature preferred (over BID, etc.) to better represent structured preclinical dose schedules, “once” or “baseline” defined	IIb, Dosing Interval
“Kinetic”-Related Variables
Time Elapsed Since Last Dose	Days, Hours, Minutes	Actual time preferred and required for capturing missed doses relative to frequency prescribed, but nominal/planned time is acceptable	APRELTM, Actual Rel Time From Previous Dose
Time Elapsed Since First Dose	Days, Hours, Minutes	Inclusion to infer the probability of steady state conditions	AFRELTM, Actual Rel Time From First Dose
System Category
Species	n/a	Controlled vocabulary and taxonomy in Figure 1	Not defined
Matrix	n/a	Controlled vocabulary and taxonomy in Figure 1	Not defined
Concentration Category
Analyte	n/a	Matched to nomenclature used in Administered Drug	DVID, Dependent Variable Name
Analyte Concentration	ng/mL, ng/g, g, fmol/million cells, BLQ, ALQ, LLOQ	Biologically meaningful units reported and overview of BLQ/ALQ imputation	DV, Analysis Value; BLQFN, BLQ Flag; ALQFN, ALQ Flag; or PCLLOQ, Lower Limit of Quantitation, respectively

^aInternational Society of Pharmacometrics (ISOP) Basic Data Structure for ADaM PopPK Implementation Guide, available at https://www.cdisc.org/

^bThe ISOP Structure matches Frequency of Administration with units of Time Elapsed as hours, which differs from the proposed standard herein that incorporates preclinical studies and long-acting therapies. Time unit conversation may be needed to translate data between these formats.

For PK studies in plasma, the Intervention category reports essential data required to generate a drug CvT profile (or PK curve) to calculate the noncompartmental PK parameters of a drug: Cmax (maximum observed concentration), Tmax (time it takes to reach Cmax), AUC (area under the curve reflecting total exposure to the drug), and elimination rate constant (used to calculate drug half-life). Individual or population pharmacokinetic analysis can also be performed to characterize absorption, distribution, metabolism, and elimination of the drug, depending on the goals of the study.

Notably, the ISOP standard does not distinguish between the administered drug and the quantified analyte, as both fall under “Dependent Variable Name (DVID)” with the assumption that they are interchangeable.¹⁶ From a population PK perspective, the observed CvT data are the most critical element for identifying sources of inter- and intra-individual variability among large pooled datasets. However, from a data sharing/archiving perspective, either in publications or databases, the administered drug may differ from the quantified analyte and is therefore required as intrinsic context for PK studies. This is particularly well illustrated when interpreting CvT data for different prodrugs that are converted to the same drug metabolite. For example, the antiretroviral tenofovir (TFV) is administered as one of two prodrugs, tenofovir disoproxil fumarate (TDF) and tenofovir alafenamide (TAF), which are administered at different dose amounts and exhibit distinct plasma TFV PK profiles, along with unique safety profiles.²⁰

System

Another way that our standard expands the capacity for PK data archiving is by the System category, which contextualizes a PK study within the physiological properties of the species employed, including distinguishment of healthy volunteers vs. end-user populations in clinical studies and the anatomical compartments sampled by collection of distinct biological matrices (tissues, cells, and fluids) within the species assessed (Table 1 and Figure 1). Preclinical studies in animals permit sample collection of a variety of matrices obtained at discrete time points following administration of a drug in a controlled environment. Likewise, clinical or exploratory studies in humans define the expected concentrations from a given dose and can be coupled with the effects of the drug to predict efficacy in end-user populations. This may involve measurement of drug concentrations in the plasma or different regions of the body, including effector sites where the drug enacts an intended therapeutic objective (indication).²¹ Data obtained from PK studies in animals can inform studies in humans, and vice versa, through in silico modeling such as for scaling up studies from small animals to nonhuman primates to humans in drug development programs²² or repurposing existing FDA-approved drugs for alternative therapeutic uses investigated in animal models. Examples of the latter include adapting oncology agents to reverse latent infection in HIV cure approaches^23,24 or using the HIV protease inhibitor ritonavir to boost the PK of nirmatrelvir for SARS-CoV-2 treatment.²⁵

Figure 1. Animal and anatomical taxonomies proposed for the minimum information standard for PK studies.

The species (A) and anatomical source (B) associated with CvT data are delineated into field-relevant hierarchies, where the latter may be expanded as needed to fit the context of a given therapeutic application, such as distinguishing CD4+ target cells in the setting of human immunodeficiency virus. In general, drug concentrations derived from cells may be differentiated by anatomical origin (B, left to right): homogenized bulk tissues, isolated cells, and suspended in fluid. Additional distinctions may be made for protein-unbound fractions where needed to support modeling activities.

Direct quantification of drugs and drug metabolites in distinct biological matrices illustrates drug distribution to sites of effect and metabolism, including sites where prodrugs are converted to pharmacologically active moieties. Thus, individual PK studies may be designed to characterize drug distribution in specific anatomical compartments (e.g., CNS, GI, genital) to answer specific questions about predicted local therapeutic effect or toxicity. Physiologically based PK modeling draws inferences about how these compartments relate as drugs flow through the body.²⁶ By categorizing biological matrices within appropriate animal and anatomical taxonomies (Figure 1), the System category supports these types of modeling activities. The organizational structure in Figure 1 provides a framework which may be expanded as needed to fit the context of a given therapeutic application, such as distinguishing target cells or protein unbound fractions.

Clinical covariates, defined as covariates that may affect PK, are not routinely provided to analytical laboratories generating CvT data and are outside the scope of the proposed minimum information standard, which poses some limitations for population PK modeling. However, these variables may be provided as metadata according to the Basic Data Structure for Population PK Analysis published by ISOP.¹⁶ Given the overlap between these standards, data linkage should be achievable based on unique specimen identifiers.

Concentration

The Concentration category coalesces concentrations of prodrugs and drug metabolites into two standardized variables: Analyte and Analyte Concentration (Table 1), where the latter distinguishes results reported as below or above the limits of quantification for the assay (BLQ or ALQ, respectively), as well as the assay’s lower limit of quantification (LLOQ) to inform how to impute these results. Assay performance characteristics are outside the scope of our PK-focused standard but may be summarized in publications of CvT data.

Linking concentrations with variables under the Intervention and System categories generates a minimum information standard with the capacity to support biologically meaningful data queries in both the clinical and preclinical spaces. The standard incorporates data for mechanistically related molecules to allow comparisons of a drug’s metabolic pathway with potential determinants of efficacy, such as competitive substrates. For example, concentrations of the antiretroviral tenofovir in plasma and its active metabolite (tenofovir diphosphate) in cells may be compared with concentrations of dideoxyadenosine triphosphate, the competitive endogenous substrate governing efficacy.^27,28 The flow of drugs can be followed over the relative time course from absorption to urinary or gastrointestinal excretion in different species and following distinct routes of administration.

Piloting the New Standard with the HIV Pharmacology Data Repository

HIV Pharmacology as a Model System for PK Database Generation

One approach to ensure a minimum information standard balances minimalism with broad applicability is to trial the standard with a complex system. Antiretroviral therapy (ART) involves complex pharmacological issues identified and addressed by over three decades of HIV treatment evolution.²⁹ Studies in HIV pharmacology have pioneered or advanced concepts such as U=U (Undetectable=Untransmittable by sexual intercourse), also known as treatment as prevention (TasP),^30,31 pre-exposure prophylaxis (PrEP),^27,32 and pharmacokinetic enhancement with pharmacokinetic boosting agents.³³ Yet, despite broad implementation of effective therapy capable of sustaining virologic suppression in people living with HIV and preventing infection in high-risk populations, annual incidence continues to drive the global burden of disease.³⁴ The success of pharmacological strategies to reduce HIV incidence relies on high adherence to daily oral dosing in people living with, and at risk for, HIV who may not be well-positioned for adherence success due to lack of access to care, low health literacy, and/or discrimination and stigma.^35,36 Efforts to reduce pill burden through the advent of long-acting formulations are only beginning to come to fruition with FDA approval of the first such complete ART regimen (i.e., injectable cabotegravir and rilpivirine).^37,38 Ultimately, an HIV cure would negate the need for lifelong treatment, yet cure strategies are complicated in part by pharmacological challenges (e.g., limited drug distribution to viral reservoirs in tissues, inadequate efficacy at those sites, and unfavorable systemic drug toxicity).^39,40

Quality of Bioanalytical Data

The clinical pharmacology field has recognized the need for standardization in bioanalytical quality control to ensure scientific rigor and reproducibility across laboratories generating drug concentration data for clinical studies. To this end, the AIDS Clinical Trials Group (ACTG) established a comprehensive quality assurance/quality control program, including routine proficiency testing consistent with Clinical Laboratory Improvement Amendments of 1988 (CLIA) requirements for clinical tests.⁴¹ The Clinical Pharmacology Quality Assurance (CPQA) program was established in 2008 to expand on these efforts by enforcing bioanalytical standardization, including passing criteria for instrument runs, across participating laboratories (currently 12) to ensure LC-MS/MS methods consistently yield accurate data values over time.^42,43 Yet, a minimum information standard for PK studies is not documented, representing a missed opportunity for the CPQA network. Implementation of the standard proposed herein would harmonize large drug concentration datasets with meaningful biological context required for aggregate PK modeling exercises.

The HIV Pharmacology Data Repository: A Resource for Real-Time PK Data Sharing

As a CPQA-participating laboratory, the UNC Center for AIDS Research (CFAR) Clinical Pharmacology and Analytical Chemistry (CPAC) Core is uniquely positioned to trial the minimum information standard proposed herein using diverse preclinical and clinical CvT datasets generated over 20 years of service. To this end, we have consolidated our electronic data archives into a central, searchable HIV Pharmacology Data Repository (HIV PDR; Figure 2). This data portal uses a relational (SQL) database structure to map study variables according to the minimum information standard for PK studies, including drug concentration data generated according to field-established quality standards. Importantly, the HIV PDR was designed according to FAIR principles for data sharing⁴⁴ and curated according to HIV pharmacology nomenclature to bypass the need for literature deep dives and ultimately support user-driven in silico research.

Figure 2. CvT data standardization efforts in the initial development of the HIV Pharmacology Data Repository (HIV PDR).

Archived CvT data generated by the Clinical Pharmacology and Analytical Chemistry (CPAC) Core through Center for AIDS Research (CFAR) service requests were consolidated from 688 Excel files (example file in top panel) into a single SQL database composed of 109,167 unique CvT data points. In doing so, synonym data dictionaries were created to parse 1,140 CvT data descriptors into 174 defined variables, thereby reducing heterogeneity by 85%.

Currently, ~62% of electronically archived data generated from CFAR service requests are maintained in a database with the front-end developed using an ASP.NET core with Angular and the back-end on an SQL Server. Our initial developmental activities collapsed 1,140 unique column headers describing bioanalytical results within 688 Excel files into 174 unique bioanalytical variables, thereby populating the Concentration category of information described above. Synonym data dictionaries were created to harmonize terminology for sample attributes (i.e., Intervention and System categories) transmitted by CFAR service requesters through the manifest Excel file (Figure 2). In creating these dictionaries, we collapsed 246 descriptors of sample attributes into 81 variables of unique anatomical sources obtained from 15 species, applying the taxonomies shown in Figure 1 and further distinguishing CD4+ cell subsets vs. others where possible. The result of these development efforts is a searchable database with controlled vocabularies housing 109,167 unique drug CvT values resulting from 47,003 samples collected across investigations of 76 drug molecules. Subsequent “back-end” harmonization efforts of existing CPAC data will liberate an additional >40,000 CvT datapoints archived in approximately 350 files generated prior to 2016 and requiring minor reformatting for upload.

To minimize unnecessary expansion of synonym dictionaries, we created a web-based service intake questionnaire through the HIV PDR to standardize how we receive relevant variables on the “front-end” of CPAC data generation (Figure 3). Responses to a brief series of questions about study design instruct the HIV PDR to select and tailor a manifest template (Excel file) for service requesters to complete with sample attributes under the Intervention and System categories. Excel data validations constrain selections in the template according to the HIV PDR’s SQL database variables, and drug concentrations resulting from LC-MS/MS sample analysis are populated in downstream columns. Additionally, the template follows one of four possible overarching structures based on HIV pharmacology study type: Clinical PK, Preclinical PK, Adherence, and In Vitro/Ex Vivo. Thus, a controlled vocabulary is enforced thereby reducing the need for data curation, minimizing term ambiguity, and improving searchability for data mining and artificial intelligence (AI)-driven exploration of the data.

Figure 3. HIV PDR tools enforcing PK data standardization.

The HIV PDR utilizes front- and back-end tools to standardize receipt of relevant variables before and after CPAC generation of drug concentration data. The synonym data dictionary can also be applied and expanded for submission of PK data generated by participating external laboratories.

Extractable Data Queries Promote Scientific Exploration

Datasets are extractable from the HIV PDR as machine-readable CSV files that can be tailored for a particular study question through employing a series of filters including species, biological matrix, and drug. Taxonomical classifications distinguish bodily compartments and the subcompartments within to aggregate data obtained from an entire organ (e.g., GI tract) or individual areas (e.g., Peyer’s patches, upper and lower GI tract), with delineation at the cellular level (e.g., vaginal CD4+ cells vs. vaginal epithelial cells) and fluid level (e.g., blood plasma vs. cervicovaginal fluid).

As a proof-of-concept demonstration of scientific utility, we performed a secondary data analysis using the HIV PDR’s data archives to develop a population PK model for tenofovir (TFV) in plasma, derived from the prodrug tenofovir disoproxil fumarate (TDF), and its active metabolite, tenofovir diphosphate (TFVdp), in peripheral blood mononuclear cells (PBMCs). We extracted 922 TFV and 785 TFVdp concentrations from 88 human study participants across 5 clinical studies including 3 dosing levels of TDF (150, 300, and 600 mg) under both first-dose and steady-state conditions (Figure 4).^27,45–49 In contrast to the traditional approach where such a dataset is compiled through laborious manual retrieval and annotation of data from discrete publications or laboratory archives, this dataset was extracted from the HIV PDR through a simple query of the following keywords: “Human”, “blood plasma”, “TDF”, and “PBMC”. We successfully fit a two-compartment population PK model to this dataset, incorporating one gut transit compartment to address the observed delay in oral absorption and a PBMC compartment to describe the metabolite disposition, using the PK modeling software NONMEM. The PK model featured two gut transit rate constants for the 150/300 (K_tr1) and 600 (K_tr2) mg doses and unequal rate constants for TFV disposition in and out of PBMCs (K_cpb, K_pbc). The condition number of the final model was 29.48, suggesting that the model has an appropriate number of estimated parameters relative to the amount of observed data. We curated the dataset using R (v. 4.4.1) to make the structure compatible with NONMEM. We did not encounter any errors related to the original data, demonstrating that HIV PDR datasets are “analysis-ready” according to the Basic Data Structure for Population PK Analysis (Table 1).¹⁶ Interindividual variability was estimated on the clearance and volume of central and peripheral compartments, and the model parameters were estimated with reasonable precision (CV% < 11.5). The model structure and parameter values were consistent with existing literature (Table 2).^45,49–51

Figure 4. Tenofovir and Tenofovir Diphosphate Pharmacokinetic Model Schematic.

Tenofovir in plasma was described as a two-compartment distribution model, with first-order absorption and first-order elimination. For tenofovir disoproxil fumarate doses above 300mg, a series of gut transit compartments were used to further describe oral absorption. Tenofovir diphosphate (TFVdp) concentrations in peripheral blood mononuclear cells (PBMC) were linked to the central tenofovir plasma compartment, with first-order rate constants describing the transfer and metabolism of TFVdp. CL: tenofovir plasma clearance; V_c: volume of distribution of the central plasma compartment; Q: intercompartmental clearance between the central and peripheral plasma compartments; V_p: volume of distribution of the peripheral plasma compartments; K_a: first-order absorption rate constant; (K_tr,n): gut transit compartments; K_cpb, K_pbc: rate constants describing the transfer and metabolism of tenofovir in plasma to and from TFVdp in peripheral blood mononuclear cells

Table 2:

Comparison of PK Parameter Estimates for Tenofovir in the Published Literature Versus Model Estimates from Pooled Data in the HIV PDR

		PK Parameter Estimate with Units (%RSE)a
Source	Populationb	CLc, L/hr	Vcd, L	Qe, L/hr	Vpf, L	Kag, 1/hr	KTR1, KTR2h, 1/hr	Kcpb, Kpbci, L/hr
Jullien 2005	M/W with HIV	90.9	534	144	1530	NR	NR	NR
Baheti 2011	M/W with HIV	42	273	181	440	1.03	NR	NR
Greene 2019	M with/without HIV	40	161	73.6	618	3	NR	0.0344, 0.0309
Leung 2023	W without HIV	58.7	331	142	843	0.863	NR	NR
HIV PDR Pooled Data	M/W with/without HIV from 5 studies	51.1 (3.2%)	223 (n/a)	173 (4.4%)	687 (4.7%)	1 (n/a)	1.36, 6.10 (n/a)	0.0255, 0.0269 (11.1%, 11.5%)

^aPercent RSE is provided for the HIV PDR model parameter estimates only.

NR=not reported, n/a=not applicable.

^bCisgender men (M) and cisgender women (W) with or without HIV.

^cTenofovir clearance (CL) from the central plasma compartment.

^dVolume of distribution of the central plasma compartment (V_c)

^eIntercompartmental clearance between the central and peripheral plasma compartments (Q)

^fVolume of distribution of the peripheral plasma compartments (V_p)

^gFirst-order absorption rate constant (K_a)

^hGut transit compartments used to describe absorption of tenofovir disoproxil fumarate for the 150/300 mg (K_TR1) and 600 mg (K_TR2) doses

ⁱRate constants describing the transfer of tenofovir in plasma to (K_cpb) and from (K_pbc) tenofovir diphosphate in peripheral blood mononuclear cells. The volume of the PBMCs is fixed at 1 L, giving these terms units of drug clearance.

Discussion

Here we propose a minimum information standard for PK studies, using HIV pharmacology as a model system, and apply this standard in the design of a web-based data portal to support in silico research driving next-generation therapeutics and data management needs for our community (the HIV PDR). The proposed standard was intentionally designed to be truly minimal following the basic tenets of pharmacokinetics. The basic structure of the standard, divided into three categories (Intervention, System, and Concentration), allows flexibility to include additional variables if supported by demonstrated potential for broad applicability. Applying the minimum information standard through the HIV PDR provides the opportunity to establish suitability and determine areas of enhancement through end-user engagement.

We demonstrated the scientific utility of the HIV PDR by simultaneously modeling TFV and TFVdp using plasma and PBMC data from 5 clinical studies extracted from the repository with different designs and dosing schemes (Figure 5).^27,45–49 This dataset provided a unique opportunity to discover PK characteristics of TFV that had not been previously reported (Table 2). Notably, we observed that the 600 mg dose exhibited a longer oral absorption delay compared to the 150 or 300 mg doses, requiring two transit rate constants to achieve the best fit. Interestingly, this finding was not evident when fitting separate models for individual datasets, nor was it reported in the original study on the 600 mg dose.²⁷ In addition, only Study 4 and Study 5 included dense sampling around Cmax,^27,48,49 allowing robust estimation of parameters dependent on these observations (Vc, Ka, Ktr1, Ktr2). Thus, we estimated these parameters for the two studies, and then fixed them to the estimated values when fitting the five data sets. This highlights the advantage of aggregating data from PK data repositories to enable more comprehensive PK analyses.

Figure 5. Tenofovir (A) and Tenofovir Diphosphate (B) Model-Predicted vs. Observed Concentrations, by Clinical Study and Dosing Status.

For each study that contributed data to the tenofovir/tenofovir diphosphate model, the median predicted concentrations are depicted by the colored lines (Study 1: Greene 2019 Clinical Pharmacology & Therapeutics; Study 2: Devanathan 2024 JAIDS; Study 3: Cottrell 2019 Clinical Infectious Diseases; Study 4: Thurman 2021 EClinicalMedicine; Study 5: Cottrell 2016 Journal of Infectious Diseases and Leung 2023 CPT Pharmacometrics & Systems Pharmacology). Observed data from first dose (closed) and steady state (open) dosing are shown as the diamond symbol. TFV: tenofovir; TFVdp: tenofovir diphosphate; nM: nanomolar; h: hour

The HIV PDR innovates in the field of clinical pharmacology, as existing platforms (Drugs@FDA⁵² and HIVDrugInteractions⁵³) are helpful for clinical decision-making but limited in pharmacometric utility. Drugs@FDA provides historical product labels and FDA review documents for most drug products approved since 1939, whereas HIVDrugInteractions offers prescribing guidance and tools for treatment selection, including curated PK fact sheets referencing key clinical studies. These platforms archive summary estimates of PK/PD parameters housed within the drug label rather than raw extractable CvT data. Furthermore, these repositories lack standardization in parameter reporting and provide data exports in nonmachine-readable format, creating bottlenecks in the model development process. Large clinical cohort networks such as MACS/WHIS⁵⁴ and ACTG⁵⁵ may aggregate CvT data archived within the network, but they lack key data descriptors and require lengthy application processes hindering data accessibility.

Importantly, existing clinical pharmacology databases have little opportunity to share data generated in real-time for investigational agents. Most of the data provided is sourced directly from the new drug application (NDA) or the product label. The HIV PDR overcomes this limitation through synergy with existing mechanisms for PK data stewardship within the CFAR network. As a generator of bioanalytical data that participates at every phase of the drug development process, our CFAR Core facility has established processes to streamline information transfer for investigational compounds assessed in federally funded studies. For example, the HIV PDR houses >3,500 concentrations of islatravir (a first-in-class, investigational antiretroviral) and its active metabolite in plasma, cells, and tissues from mice, rats, rabbits, dogs, sheep, and nonhuman primates following intravenous, subcutaneous, and oral administration. Collating and sharing data from these federally funded projects in the HIV PDR offer an unprecedented opportunity to conduct modeling studies that support dose translation and obviate the risk of research duplication.

Based on a comprehensive search of the indexing sites FAIRSharing and PubMed, we are only aware of three existing PK data repositories. One archives 16,267 CvT datapoints for 187 chemical entities and their metabolites.¹⁰ However, the repository’s rigid upload criteria (requiring data reformatting into a template form), lack of a user-friendly interface, and environmental toxicokinetic focus make it an unlikely site of data sharing for HIV researchers. Additionally, this database’s website indicates that few if any additions to the initially published archived have been achieved since 2020. In contrast, another other database has expanded its reported clinical or experimental outputs from 73,017 to 121,607 as of October 6, 2023. Of these 121,607 outputs, 36,839 are associated with raw concentration values for drugs and their metabolites (excluding 28,922 calculated means or missing values), and the vast majority were derived from studies published more than 10 years ago (113 of 118 studies).⁹ Although the database contains certain PK parameters and other clinical test results extracted from these publications, it is constrained to mostly clinical plasma CvT data (>95%) and is extremely limited in the breadth of archived drugs. Of the 36,839 CvT values, >50% are caffeine, morphine, paracetamol, dextromethorphan, and their derivatives, and none are relevant to HIV or infectious disease. The third database is focused on HIV and has a user-friendly data display, but is limited to small clinical studies of three antiretrovirals (efavirenz, nevirapine, and dolutegravir), two of which have limited clinical relevance today.¹¹

Future Directions for the HIV PDR as a Data Sharing Resource

The HIV PDR is designed for growth in keeping with research strategies executed through CFAR activities and beyond. Future activities include establishment of a collaborative governing body to devise best-practices for data curation and validation that will guide the design of a web-based portal for submission of PK data generated by external laboratories. The synonym data dictionary developed in our initial database build can be easily expanded to translate descriptors of PK data utilized by participating external sources, such as LDMS⁵⁶ and SCHARP⁵⁷, into our minimum information standard. Additionally, we will develop data validation tools to flag bioanalytical data laying outside of the normal distribution for data archived within the HIV PDR. These tools can be designed to match newly generated datapoints against historical controls according to the specified variables (analyte, concentration units, biological matrix, species, drug, dose, and route of administration) to add another level of quality assessment prior to publication in the database. Finally, we will develop an accession system that records data query attributes, harnesses system-versioned temporal tables for each extracted PK data pool, and provides references such as associated publications or National Clinical Trial (NCT) numbers to identify study populations or other metadata. This design feature will ensure reproducibility despite the dynamic nature of the data archives and provide a persistent unique identifier in keeping with FAIR principles and NIH data management policies.^4,44 The HIV PDR will become publicly available when data permissions are established.

The HIV PDR was initially envisioned as a specialized platform to meet an immediate need for HIV PK data management and sharing within our CFAR Core’s userbase, yet we recognize its suitability for broader application outside the setting of HIV. In addition to antiretrovirals, the HIV PDR contains PK data for therapeutics used for HIV comorbidities and co-occurring illnesses, such as tuberculosis, other STIs and infections, and cancer, among others. This diversity of PK data is achievable because the minimum information standard the database employs was borne of a pharmacologically complex disease state. In other words, PK data standardization in the HIV PDR is truly “minimal,” providing the basic framework for adaptability beyond HIV.

Another potential future direction for the HIV PDR is to incorporate mass spectrometry imaging (MSI) data, which is an alternative strategy to LC-MS/MS for quantifying drug distribution in tissues and cells that preserves the spatial distribution of analytes while maintaining high sensitivity and specificity. We are exploring how to incorporate MSI data generated in our laboratory into the HIV PDR, while leveraging existing MSI data repositories like MassIVE.⁵⁸ Drug concentrations by MSI, provided as nanograms of drug per gram of tissue in a thin section, could easily be deposited into the HIV PDR along with an accession number associated with the original annotated imaging datasets residing in an MSI database. This approach could leverage the relative strengths of respective databases without duplicating effort.

As expectations for data sharing become convention across research disciplines, investigators will increasingly rely on data repositories. Specialized platforms for PK data, such as the HIV PDR, capture field-relevant data descriptors for targeted data mining to support secondary analyses. In setting a minimum information standard for PK studies, a key element of rational database design, we provide framework for sharing CvT data with enough context to infer meaning and drive advancements in clinical therapeutics.