The human hepatocyte TXG-MAPr: gene co-expression network modules to support mechanism-based risk assessment.

Mechanism-based risk assessment is urged to advance and fully permeate into current safety assessment practices, possibly at early phases of drug safety testing. Toxicogenomics is a promising source of mechanisms-revealing data, but interpretative analysis tools specific for the testing systems (e.g. hepatocytes) are lacking. In this study, we present the TXG-MAPr webtool (available at https://txg-mapr.eu/WGCNA_PHH/TGGATEs_PHH/ ), an R-Shiny-based implementation of weighted gene co-expression network analysis (WGCNA) obtained from the Primary Human Hepatocytes (PHH) TG-GATEs dataset. The 398 gene co-expression networks (modules) were annotated with functional information (pathway enrichment, transcription factor) to reveal their mechanistic interpretation. Several well-known stress response pathways were captured in the modules, were perturbed by specific stressors and showed preservation in rat systems (rat primary hepatocytes and rat in vivo liver), with the exception of DNA damage and oxidative stress responses. A subset of 87 well-annotated and preserved modules was used to evaluate mechanisms of toxicity of endoplasmic reticulum (ER) stress and oxidative stress inducers, including cyclosporine A, tunicamycin and acetaminophen. In addition, module responses can be calculated from external datasets obtained with different hepatocyte cells and platforms, including targeted RNA-seq data, therefore, imputing biological responses from a limited gene set. As another application, donors' sensitivity towards tunicamycin was investigated with the TXG-MAPr, identifying higher basal level of intrinsic immune response in donors with pre-existing liver pathology. In conclusion, we demonstrated that gene co-expression analysis coupled to an interactive visualization environment, the TXG-MAPr, is a promising approach to achieve mechanistic relevant, cross-species and cross-platform evaluation of toxicogenomic data.

Characterisation of the NRF2 transcriptional network and its response to chemical insult in primary human hepatocytes: implications for prediction of drug-induced liver injury.

The transcription factor NRF2, governed by its repressor KEAP1, protects cells against oxidative stress. There is interest in modelling the NRF2 response to improve the prediction of clinical toxicities such as drug-induced liver injury (DILI). However, very little is known about the makeup of the NRF2 transcriptional network and its response to chemical perturbation in primary human hepatocytes (PHH), which are often used as a translational model for investigating DILI. Here, microarray analysis identified 108 transcripts (including several putative novel NRF2-regulated genes) that were both downregulated by siRNA targeting NRF2 and upregulated by siRNA targeting KEAP1 in PHH. Applying weighted gene co-expression network analysis (WGCNA) to transcriptomic data from the Open TG-GATES toxicogenomics repository (representing PHH exposed to 158 compounds) revealed four co-expressed gene sets or 'modules' enriched for these and other NRF2-associated genes. By classifying the 158 TG-GATES compounds based on published evidence, and employing the four modules as network perturbation metrics, we found that the activation of NRF2 is a very good indicator of the intrinsic biochemical reactivity of a compound (i.e. its propensity to cause direct chemical stress), with relatively high sensitivity, specificity, accuracy and positive/negative predictive values. We also found that NRF2 activation has lower sensitivity for the prediction of clinical DILI risk, although relatively high specificity and positive predictive values indicate that false positive detection rates are likely to be low in this setting. Underpinned by our comprehensive analysis, activation of the NRF2 network is one of several mechanism-based components that can be incorporated into holistic systems toxicology models to improve mechanistic understanding and preclinical prediction of DILI in man.

Assessing Concordance of Drug-Induced Transcriptional Response in Rodent Liver and Cultured Hepatocytes.

The effect of drugs, disease and other perturbations on mRNA levels are studied using gene expression microarrays or RNA-seq, with the goal of understanding molecular effects arising from the perturbation. Previous comparisons of reproducibility across laboratories have been limited in scale and focused on a single model. The use of model systems, such as cultured primary cells or cancer cell lines, assumes that mechanistic insights derived from the models would have been observed via in vivo studies. We examined the concordance of compound-induced transcriptional changes using data from several sources: rat liver and rat primary hepatocytes (RPH) from Drug Matrix (DM) and open TG-GATEs (TG), human primary hepatocytes (HPH) from TG, and mouse liver/HepG2 results from the Gene Expression Omnibus (GEO) repository. Gene expression changes for treatments were normalized to controls and analyzed with three methods: 1) gene level for 9071 high expression genes in rat liver, 2) gene set analysis (GSA) using canonical pathways and gene ontology sets, 3) weighted gene co-expression network analysis (WGCNA). Co-expression networks performed better than genes or GSA when comparing treatment effects within rat liver and rat vs. mouse liver. Genes and modules performed similarly at Connectivity Map-style analyses, where success at identifying similar treatments among a collection of reference profiles is the goal. Comparisons between rat liver and RPH, and those between RPH, HPH and HepG2 cells reveal lower concordance for all methods. We observe that the baseline state of untreated cultured cells relative to untreated rat liver shows striking similarity with toxicant-exposed cells in vivo, indicating that gross systems level perturbation in the underlying networks in culture may contribute to the low concordance.

What general conclusions can we draw from kinase profiling data sets?

Understanding general selectivity trends across the kinome has implications ranging from target selection, compound prioritization, toxicity and patient tailoring. Several recent publications have described the characterization of kinase inhibitors via large assay panels, offering a range of generalizations that influenced kinase inhibitor research trends. Since a subset of profiled inhibitors overlap across reports, we evaluated the concordance of activity results for the same compound-kinase pairs across four data sources generated from different kinase biochemical assay technologies. Overall, 77% of all results are within 3 fold or qualitatively in agreement across sources. However, the agreement for active compounds is only 37%, indicating that different profiling panels are in better agreement to determine a compound's lack of activity rather than degree of activity. Low concordance is also found when comparing the promiscuity of kinase targets evaluated from different sources, and the pharmacological similarity of kinases. In contrast, the overall promiscuity of kinase inhibitors was consistent across sources. We highlight the difficulty of drawing general conclusions from such data by showing that no significant selectivity difference distinguishes type I vs. type II inhibitors, and limited kinase space similarity that is consistent across different sources. This article is part of a Special Issue entitled: Inhibitors of Protein Kinases (2012).

A preclinical secondary pharmacology resource illuminates target-adverse drug reaction associations of marketed drugs.

In vitro secondary pharmacology assays are an important tool for predicting clinical adverse drug reactions (ADRs) of investigational drugs. We created the Secondary Pharmacology Database (SPD) by testing 1958 drugs using 200 assays to validate target-ADR associations. Compared to public and subscription resources, 95% of all and 36% of active (AC50 < 1 µM) results are unique to SPD, with bias towards higher activity in public resources. Annotating drugs with free maximal plasma concentrations, we find 684 physiologically relevant unpublished off-target activities. Furthermore, 64% of putative ADRs linked to target activity in key literature reviews are not statistically significant in SPD. Systematic analysis of all target-ADR pairs identifies several putative associations supported by publications. Finally, candidate mechanisms for known ADRs are proposed based on SPD off-target activities. Here we present a freely-available resource for benchmarking ADR predictions, explaining phenotypic activity and investigating clinical properties of marketed drugs.

Neurofilament light chain and dorsal root ganglia injury after adeno-associated virus 9 gene therapy in nonhuman primates.

In nonhuman primates (NHPs), adeno-associated virus serotype 9 (AAV9) vectorized gene therapy can cause asymptomatic microscopic injury to dorsal root ganglia (DRG) and trigeminal ganglia (TG) somatosensory neurons, causing neurofilament light chain (NfL) to diffuse into cerebrospinal fluid (CSF) and blood. Data from 260 cynomolgus macaques administered vehicle or AAV9 vectors (intrathecally or intravenously) were analyzed to investigate NfL as a soluble biomarker for monitoring DRG/TG microscopic findings. The incidence of key DRG/TG findings with AAV9 vectors was 78% (maximum histopathology severity, moderate) at 2-12 weeks after the dose. When examined up to 52 weeks after the dose, the incidence was 42% (maximum histopathology severity, minimal). Terminal NfL concentrations in plasma, serum, and CSF correlated with microscopic severity. After 52 weeks, NfL returned to pre-dose baseline concentrations, correlating with microscopic findings of lesser incidence and/or severity compared with interim time points. Blood and CSF NfL concentrations correlated with asymptomatic DRG/TG injury, suggesting that monitoring serum and plasma concentrations is as useful for assessment as more invasive CSF sampling. Longitudinal assessment of NfL concentrations related to microscopic findings associated with AAV9 administration in NHPs indicates NfL could be a useful biomarker in nonclinical toxicity testing. Caution should be applied for any translation to humans.

Identifying multiscale translational safety biomarkers using a network-based systems approach.

Animal testing is the current standard for drug and chemicals safety assessment, but hazards translation to human is uncertain. Human in vitro models can address the species translation but might not replicate in vivo complexity. Herein, we propose a network-based method addressing these translational multiscale problems that derives in vivo liver injury biomarkers applicable to in vitro human early safety screening. We applied weighted correlation network analysis (WGCNA) to a large rat liver transcriptomic dataset to obtain co-regulated gene clusters (modules). We identified modules statistically associated with liver pathologies, including a module enriched for ATF4-regulated genes as associated with the occurrence of hepatocellular single-cell necrosis, and as preserved in human liver in vitro models. Within the module, we identified TRIB3 and MTHFD2 as a novel candidate stress biomarkers, and developed and used BAC-eGFPHepG2 reporters in a compound screening, identifying compounds showing ATF4-dependent stress response and potential early safety signals.

"3D, human renal proximal tubule (RPTEC-TERT1) organoids 'tubuloids' for translatable evaluation of nephrotoxins in high-throughput".

The importance of human cell-based in vitro tools to drug development that are robust, accurate, and predictive cannot be understated. There has been significant effort in recent years to develop such platforms, with increased interest in 3D models that can recapitulate key aspects of biology that 2D models might not be able to deliver. We describe the development of a 3D human cell-based in vitro assay for the investigation of nephrotoxicity, using RPTEC-TERT1 cells. These RPTEC-TERT1 proximal tubule organoids 'tubuloids' demonstrate marked differences in physiologically relevant morphology compared to 2D monolayer cells, increased sensitivity to nephrotoxins observable via secreted protein, and with a higher degree of similarity to native human kidney tissue. Finally, tubuloids incubated with nephrotoxins demonstrate altered Na+/K+-ATPase signal intensity, a potential avenue for a high-throughput, translatable nephrotoxicity assay.

A Set of Six Gene Expression Biomarkers Identify Rat Liver Tumorigens in Short-term Assays.

Chemical-induced liver cancer occurs in rodents through well-characterized adverse outcome pathways. We hypothesized that measurement of the 6 most common molecular initiating events (MIEs) in liver cancer adverse outcome pathways in short-term assays using only gene expression will allow early identification of chemicals and their associated doses that are likely to be tumorigenic in the liver in 2-year bioassays. We tested this hypothesis using transcript data from a rat liver microarray compendium consisting of 2013 comparisons of 146 chemicals administered at doses with previously established effects on rat liver tumor induction. Five MIEs were measured using previously characterized gene expression biomarkers composed of gene sets predictive for genotoxicity and activation of 1 or more xenobiotic receptors (aryl hydrocarbon receptor, constitutive activated receptor, estrogen receptor, and peroxisome proliferator-activated receptor α). Because chronic injury can be important in tumorigenesis, we also developed a biomarker for cytotoxicity that had a 96% balanced accuracy. Characterization of the genes in each biomarker set using the unsupervised TXG-MAP network model demonstrated that the genes were associated with distinct functional coexpression modules. Using the Toxicological Priority Index to rank chemicals based on their ability to activate the MIEs showed that chemicals administered at tumorigenic doses clearly gave the highest ranked scores. Balanced accuracies using thresholds derived from either TG-GATES or DrugMatrix data sets to predict tumorigenicity in independent sets of chemicals were up to 93%. These results show that a MIE-directed approach using only gene expression biomarkers could be used in short-term assays to identify chemicals and their doses that cause tumors.

Machine learning guided association of adverse drug reactions with in vitro target-based pharmacology.

BACKGROUND: Adverse drug reactions (ADRs) are one of the leading causes of morbidity and mortality in health care. Understanding which drug targets are linked to ADRs can lead to the development of safer medicines. METHODS: Here, we analyse in vitro secondary pharmacology of common (off) targets for 2134 marketed drugs. To associate these drugs with human ADRs, we utilized FDA Adverse Event Reports and developed random forest models that predict ADR occurrences from in vitro pharmacological profiles. FINDINGS: By evaluating Gini importance scores of model features, we identify 221 target-ADR associations, which co-occur in PubMed abstracts to a greater extent than expected by chance. Amongst these are established relations, such as the association of in vitro hERG binding with cardiac arrhythmias, which further validate our machine learning approach. Evidence on bile acid metabolism supports our identification of associations between the Bile Salt Export Pump and renal, thyroid, lipid metabolism, respiratory tract and central nervous system disorders. Unexpectedly, our model suggests PDE3 is associated with 40 ADRs. INTERPRETATION: These associations provide a comprehensive resource to support drug development and human biology studies. FUNDING: This study was not supported by any formal funding bodies.

A Novel Open Access Web Portal for Integrating Mechanistic and Toxicogenomic Study Results.

Applying toxicogenomics to improving the safety profile of drug candidates and crop protection molecules is most useful when it identifies relevant biological and mechanistic information that highlights risks and informs risk mitigation strategies. Pathway-based approaches, such as gene set enrichment analysis, integrate toxicogenomic data with known biological process and pathways. Network methods help define unknown biological processes and offer data reduction advantages. Integrating the 2 approaches would improve interpretation of toxicogenomic information. Barriers to the routine application of these methods in genome-wide transcriptomic studies include a need for "hands-on" computer programming experience, the selection of 1 or more analysis methods (eg pathway analysis methods), the sensitivity of results to algorithm parameters, and challenges in linking differential gene expression to variation in safety outcomes. To facilitate adoption and reproducibility of gene expression analysis in safety studies, we have developed Collaborative Toxicogeomics, an open-access integrated web portal using the Django web framework. The software, developed with the Python programming language, is modular, extensible and implements "best-practice" methods in computational biology. New study results are compared with over 4000 rodent liver experiments from Drug Matrix and open TG-GATEs. A unique feature of the software is the ability to integrate clinical chemistry and histopathology-derived outcomes with results from gene expression studies, leading to relevant mechanistic conclusions. We describe its application by analyzing the effects of several toxicants on liver gene expression and exemplify application to predicting toxicity study outcomes upon chronic treatment from expression changes in acute-duration studies.

Prioritizing hits from phenotypic high-content screens.

In the past decade, advances in the field of high-content screening (HCS) have provided researchers with a powerful new screening tool to observe treatment effects on multiple experimental parameters. While extremely useful, HCS has resulted in the collection of large datasets of increased complexity that require intensive analysis. Recently, approaches have been developed to analyze multi-parametric HCS data more completely and, when used in conjunction with RNA interference, target-based biochemistry and structural analysis, these approaches have begun to unlock the potential of this screening format in aiding drug discovery. This review illustrates how the combination of these technologies has been used to successfully drive the drug discovery process.

Chemical fragments as foundations for understanding target space and activity prediction.

The use of small inhibitors' fragment frequencies for understanding kinase potency and selectivity is described. By quantification of differences in the frequency of occurrence of fragments, similarities between small molecules and their targets can be determined. Naive Bayes models employing fragments provide highly interpretable and reliable means for predicting potency in individual kinases, as demonstrated in retrospective tests and prospective selections that were subsequently screened. Statistical corrections for prospective validation allowed us to accurately estimate success rates in the prospective experiment. Selectivity relationships between kinase targets are substantially explained by differences in the fragment composition of actives. By application of fragment similarities to the broader proteome, it is shown that targets related by sequence exhibit similar fragment preferences in small molecules. Of greater interest, certain targets unrelated by sequence are shown to have similar fragment preferences, even when the chemical similarity of ligands active at each target is low.

Medication Exposure in Highly Adherent Psychiatry Patients.

Medication exposure is dependent upon many factors, the single most important being if the patient took the prescribed medication as indicated. To assess medication exposure for psychotropic and other medication classes, we enrolled 115 highly adherent psychiatry patients prescribed five or more medications. In these patients, we measured 21 psychotropic and 38 nonpsychotropic medications comprising a 59 medication multiplex assay panel. Strict enrollment criteria and reconciliation of the electronic health record medication list prior to study initiation produced a patient cohort that was adherent with 91% of their prescribed medications as determined by comparing medications detected empirically in blood to the electronic health record medication list. In addition, 13% of detected medications were not in the electronic health record medication list. We found that only 53% of detected medications were within the literature-derived reference range with 41% below and 6% above the reference range specific to each medication. When psychotropic medications were analyzed near trough-level, only sertraline was found to be within the literature-derived reference range for all patients tested. Concentrations of the remaining medications indicated extensive exposure below the reference range. This is the first study to empirically and comprehensively assess medication exposure obtained in comorbid polypharmacy patients, minimizing the important behavioral factor of adherence in the study of medication exposure. These data indicate that low medication exposure is extensive and must be considered when therapeutic issues arise, including the lack of response to medication therapy.

Assessment of Patient Medication Adherence, Medical Record Accuracy, and Medication Blood Concentrations for Prescription and Over-the-Counter Medications.

Importance: Inaccurate medication records and poor medication adherence result in incomplete knowledge of therapy for patients. Objective: To study accuracy of medical records and patient adherence by measuring blood concentrations of medications. Design, Setting, and Participants: This cross-sectional study validated a serum-based liquid chromatography-tandem mass spectrometry assay to simultaneously quantify 263 medications used for acute and chronic conditions. The assay panel was applied to 3 clinical patient cohorts: residual serum from 1000 randomly selected samples sent for routine clinical chemistry testing between April 8 and October 6, 2015 (residuals cohort), 50 prospectively enrolled patients in a gastroenterology clinic between March 1 and March 15, 2016, who were prescribed more than 5 medications (gastroenterology care cohort), and a convenience cohort of 296 patients with hypertension who sought care in an emergency department (ED care cohort) between July 1, 2012, and April 25, 2013. Integrated data analysis of the cohorts was performed from August 22 to November 29, 2017. Main Outcomes and Measures: Medication serum concentrations, electronic health record medication lists, and predicted drug interactions. Results: Of the 1346 total samples, 1000 came from the residuals cohort (640 women and 360 men; median age, 60 years [interquartile range (IQR), 44-71 years]), 50 from the gastroenterology care cohort (30 women and 20 men; median age, 66 years [IQR, 62-70 years]), and 296 from the ED care cohort (160 women and 136 men; median age, 59 years [IQR, 52-66 years]). Median medication adherence, defined as the subset of detected medications from the prescription record, was 83% (IQR, 50%-100%) in the residuals cohort, 100% (IQR, 84%-100%) in the gastroenterology care cohort, and 78% (IQR, 57%-100%) in the ED care cohort. Patients adherent to 1 medication were more often adherent to other medications. Among patients prescribed 3 medications or more, there were no significant associations between medication adherence and sex or number of prescribed medications, and there was a modest association between adherence and age. By comparing detected vs prescribed medications, we detected a median of 0 (IQR, 0-2) medications per patient that were not listed in the electronic health record in the residuals cohort, 1 (IQR, 0-2) medication per patient that was not listed in the electronic health record in the gastroenterology care cohort, and 1 (IQR, 0-2) medication per patient that was not listed in the electronic health record in the ED care cohort. A total of 435 patients (43.5%) in the residuals cohort had no discrepancy between the electronic health record and detected medication lists, 22 patients (44.0%) in the gastroenterology care cohort had no discrepancy between the electronic health record and detected medication lists, and 41 patients (13.9%) in the ED care cohort had no discrepancy between the electronic health record and detected medication lists. Half of adverse drug reaction alerts occurred among medications detected without prescription. Conclusions and Relevance: Comprehensive medication monitoring offers promise to improve adherence, the accuracy of medical records, and the safety for patients with polypharmacy.

Managing Psychotropic Medications in Complex, Real-World Patients Using Comprehensive Therapeutic Drug Monitoring.

There are multiple treatment options for depression, anxiety, psychosis, and other psychiatric disorders, and psychiatry patients are often comorbid with complex, polypharmacy treatment regimens. Unlike cardiovascular disease and diabetes, there are no readily available biomarkers to gauge treatment success with psychotropic medications, often resulting in subjective determination of medication therapy effectiveness. The physiochemical properties of psychiatric medications in general lend themselves to quantitative measurement in blood, offering an avenue to optimize treatment for each patient. Herein, we describe a novel application that employs comprehensive therapeutic drug monitoring of both psychiatric and nonpsychiatric medications to holistically personalize therapy for complex psychiatry patients.

Medication adherence, medical record accuracy, and medication exposure in real-world patients using comprehensive medication monitoring.

BACKGROUND: Poor adherence to medication regimens and medical record inconsistencies result in incomplete knowledge of medication therapy in polypharmacy patients. By quantitatively identifying medications in the blood of patients and reconciling detected medications with the medical record, we have defined the severity of this knowledge gap and created a path toward optimizing medication therapy. METHODS AND FINDINGS: We validated a liquid chromatography-tandem mass spectrometry assay to detect and/or quantify 38 medications across a broad range of chronic diseases to obtain a comprehensive survey of patient adherence, medical record accuracy, and exposure variability in two patient populations. In a retrospectively tested 821-patient cohort representing U.S. adults, we found that 46% of medications assessed were detected in patients as prescribed in the medical record. Of the remaining medications, 23% were detected, but not listed in the medical record while 30% were prescribed to patients, but not detected in blood. To determine how often each detected medication fell within literature-derived reference ranges when taken as prescribed, we prospectively enrolled a cohort of 151 treatment-regimen adherent patients. In this cohort, we found that 53% of medications that were taken as prescribed, as determined using patient self-reporting, were not within the blood reference range. Of the medications not in range, 83% were below and 17% above the lower and upper range limits, respectively. Only 32% of out-of-range medications could be attributed to short oral half-lives, leaving extensive exposure variability to result from patient behavior, undefined drug interactions, genetics, and other characteristics that can affect medication exposure. CONCLUSIONS: This is the first study to assess compliance, medical record accuracy, and exposure as determinants of real-world treatment and response. Variation in medication detection and exposure is greater than previously demonstrated, illustrating the scope of current therapy issues and opening avenues that warrant further investigation to optimize medication therapy.

Dependence of molecular properties on proteomic family for marketed oral drugs.

An association of drugs with their proteomic family reveals that molecular properties of drugs targeting proteases, lipid and peptide G-protein-coupled receptors (GPCRs), and nuclear hormone receptors significantly exceed limits for some properties in the "rule of five", while drugs targeting cytochrome P450s, biogenic amine GPCRs, and transporters have significantly lower values for certain properties. Also, the variation in drug targets appears to be a factor explaining increasing molecular weight over time.

Kinomics: characterizing the therapeutically validated kinase space.

The annotation and visualization of medicinally relevant kinase space revealed that kinase inhibitors in the clinic are, on average, of higher molecular weight and more lipophilic than all other clinically investigated drugs. Tyrosine kinases from the vascular endothelial growth factor and epidermal growth factor receptor families are the most pursued targets. Furthermore, oncological indications account for 75% of all kinase-related clinical interest. In addition, analysis of the similarity between kinase targets with respect to sequence, selectivity and structure has revealed that kinases with > or =60% sequence identity are most likely to be inhibited by the same classes of compounds and have similar ATP-binding sites. The identification of this threshold, together with the widely accepted representation of the sequence-based kinase space, is expanding our understanding of the clinical and structural space of the kinome.

Co-prescription trends in a large cohort of subjects predict substantial drug-drug interactions.

Pharmaceutical prescribing and drug-drug interaction data underlie recommendations on drug combinations that should be avoided or closely monitored by prescribers. Because the number of patients taking multiple medications is increasing, a comprehensive view of prescribing patterns in patients is important to better assess real world pharmaceutical response and evaluate the potential for multi-drug interactions. We obtained self-reported prescription data from NHANES surveys between 1999 and 2010, and confirm the previously reported finding of increasing drug use in the elderly. We studied co-prescription drug trends by focusing on the 2009-2010 survey, which contains prescription data on 690 drugs used by 10,537 subjects. We found that medication profiles were unique for individuals aged 65 years or more, with ≥98 unique drug regimens encountered per 100 subjects taking 3 or more medications. When drugs were viewed by therapeutic class, it was found that the most commonly prescribed drugs were not the most commonly co-prescribed drugs for any of the 16 drug classes investigated. We cross-referenced these medication lists with drug interaction data from Drugs.com to evaluate the potential for drug interactions. The number of drug alerts rose proportionally with the number of co-prescribed medications, rising from 3.3 alerts for individuals prescribed 5 medications to 11.7 alerts for individuals prescribed 10 medications. We found 22% of elderly subjects taking both a substrate and inhibitor of a given cytochrome P450 enzyme, and 4% taking multiple inhibitors of the same enzyme simultaneously. By examining drug pairs prescribed in 0.1% of the population or more, we found low agreement between co-prescription rate and co-discussion in the literature. These data show that prescribing trends in treatment could drive a large extent of individual variability in drug response, and that current pairwise approaches to assessing drug-drug interactions may be inadequate for predicting real world outcomes.