Drug-induced liver injury (DILI) can result in severe clinical outcomes such as acute liver failure, and, although rare, DILI is encountered frequently during the drug development process and therefore presents a significant challenge to drug developers and regulators. Importantly, the complexity of the disease, the lack of predictive biomarkers as well as the difficulty to diagnose and determine causality are obstacles that need to be overcome [1
]. Our current understanding of DILI pathogenesis is limited, and therefore identifying risk factors is critical for a better understanding and avoidance of DILI risks.
An increased risk of developing DILI is thought to be caused by interactions among the host, drug, and environmental factors [3
]. Genetic studies have found single nucleotide polymorphisms in a number of genes, including human leukocyte antigen (HLA) regions that were associated with increased risk for DILI as shown for flucloxacillin, amoxicillin-clavulanate, and lapatinib [5
]. Non-genetic host factors, such as age, gender, and underlying liver disease, were also found to be associated with increased DILI risk; however, many of these factors may be associated only with specific drugs [10
]. Underlying liver diseases affect the activities of certain CYP enzymes such that drugs that are substrates of these enzymes should be given at lower doses [11
]. Furthermore, there is evidence for serum acute phase reactants to hallmark healthy individuals at risk to develop DILI prior to drug treatment, therefore carrying the potential to identify individuals likely to develop DILI [12
When considering drug properties that may have influence on DILI, lipophilicity is an important criterion that strongly contributes to the distribution of drugs throughout the body and is generally associated with a large volume of distribution. There is also increasing interest in investigating drug characteristics and their relationship to DILI risk. Lammert et al. found that exposure to oral medications given at doses greater than 50 mg/day was associated with a higher DILI risk [13
]. This finding was later confirmed in the Spanish DILI registry, where 77% of the drugs that caused DILI were given at doses of 50 mg/day or more [14
]. In addition to dose, extensive metabolism (>50%) is also associated with increased risk of hepatic adverse events from orally administered drugs [15
]. Recently, we identified drugs with daily doses (DD) greater than 100 mg/day and high lipophilicity (measured by log value for octanol-water partition coefficient p
, i.e., logP ≥ 3) to be associated with increased DILI risk [16
]. These two drug characteristics (i.e., extent of metabolism and lipophilicity) look promising and have been widely cited by seminal reviews [3
], but further validation to test the reliability of the findings in larger datasets is warranted.
In this study, we aim to comprehensively revisit the relationship between DILI risk and lipophilicity and the extent of metabolism using five independent annotations for drugs within the DILIrank dataset, a large published drug reference dataset with 1036 Food and Drug Administration FDA-approved drugs [17
]. We found that lipophilicity and the extent of metabolism alone were associated with increased DILI risk, but that the strength of association varied considerably across different annotations. However, the combination of high daily dose with high lipophilicity or significant hepatic metabolism was consistently associated with increased DILI risk across all annotated datasets.
The Venn diagram (Figure 1
) shows the commonality and the specificities of the DILI datasets retrieved from the five independent annotations for n
= 1036 DILIrank drugs. A large portion of the drugs (n
= 398) are annotated by at least two datasets, but only n
= 38 drugs are in common among all five datasets. Chen et al. provides the largest annotation dataset of n
= 504, with n
= 192 v
Most-DILI-concern and n
= 312 v
No-DILI-concern drugs, followed by Xu et al. of n
= 343, with n
= 195 DILI positives and n
= 148 negatives, and Greene et al. of n
= 275, with n
= 189 drugs having human evidence of toxicity and n
= 86 drugs with no evidence of human toxicity. Additionally, the Sakatis dataset annotated n
= 178 drugs, including n
= 92 hepatoxicants and n
= 86 as non-hepatotoxicants, and the Zhu et al. dataset labeled n
= 217 drugs, of which n
= 161 were hepatotoxic and n
= 56 were non-hepatotoxic. Notably, a total of n
= 275 DILIrank drugs were not evaluated in the analysis due to their ambiguous annotations.
2.1. Lipophilicity and DILI Risk
We analyzed 763 medications retrieved from the DILIrank dataset with both daily dose and logP data available. To comprehensively assess the relationship between lipophilicity and hepatotoxicity risk, five independent DILI annotations were used, as detailed above. We used Fisher’s exact test to determine the statistical significance of the association between high daily dose, high lipophilicity (measured by logP), and DILI risk. The extent of the association was measured using odds ratio (OR) calculated by logistic regression.
As illustrated in Table 1
, the combination of high lipophilicity (logP
≥ 3) and DD ≥ 100 mg, as defined by the Rule of Two (RO2) [16
], is statistically significantly associated with the risk of DILI across all datasets (p
< 0.05); however, the estimated OR ranges from 2.32 to 11.50, depending on the annotations used.
The association between DILI risk and high lipophilicity is statistically significant only when considering the annotations generated by Chen, Greene, and Zhu; however, no significant relationship was seen when the Sakatis or Xu annotations were considered. Note, DD ≥ 100 mg was significantly associated with DILI risk in all datasets. Furthermore, when the datasets of Chen, Greene and Zhu were evaluated, the OR for logP alone is significantly lower than those calculated for the combination of DD ≥ 100 mg and logP ≥ 3, suggesting a weaker association between logP alone with DILI risk as compared to the combination of DD ≥ 100 mg and logP ≥ 3.
We also analyzed the RO2 results across the five annotated datasets. As shown in Figure 2
, there were 13% of RO2 positives but <1% RO2 negatives which were shared by the five datasets. Similar trends were observed for the percentage of RO2 positives and negatives shared by three or four datasets. The higher consensus in RO2 positives than in RO2 negatives agrees with our previous report regarding the unique characteristics of the RO2 model, which has a high true positive rate but a high false negative rate [16
2.2. Metabolism and DILI Risk
Similarly, a total of 559 DILIrank drugs with daily dose and available metabolism data were used to assess the relationship between DILI risk and the extent of metabolism. As given in Table 2
, the combination of hepatic metabolism ≥50% and DD ≥ 100 mg was significantly associated with an increased risk of DILI across all datasets (p
< 0.05), and the extent of association, as measured by ORs, ranged from 3.79 to 11.09 depending on the annotations used. Significant hepatic metabolism (≥50%) alone was significantly associated with DILI risk in most annotations except for the Zhu et al. and Sakatis et al. datasets. Similar to logP
, the calculated ORs for the combination of metabolism and daily dose were higher than those for metabolism alone. This was seen in any of the five annotated datasets, which demonstrates that the combination with daily dose enhances the association between the extent of metabolism and DILI risk. As in the preceding analysis, DD ≥ 100 mg alone was significantly associated with DILI risk across all annotated datasets.
The identification of DILI risk factors is of critical importance to protect individuals from adverse drug reactions. Recent seminal reviews include lipophilicity and the extent of metabolism as potential risk factors in the pathogenesis of DILI [3
]. Given the inconsistencies among DILI annotations, it is important to further validate contributing factors to DILI onset and progression. In the present study, we utilized a large published DILIrank dataset with 1036 FDA-approved drugs and five independent DILI annotations to comprehensively assess the association between DILI risk and two drug properties, namely lipophilicity and the extent of metabolism. Based on a consensus approach, our results suggest that lipophilicity and the extent of metabolism alone had weak, but statistically significant, associations with an increase in DILI risk, and that factoring daily dose into the algorithm significantly strengthened these associations.
In our analysis, we did observe significant variation across different annotations regarding the strength of the association between DILI risk and lipophilicity as well as the extent of metabolism. Obviously, the accuracy of annotating DILI risks varied among the datasets and this could affect the outcome when assessing risk factors. No golden standard for DILI annotation has been established so far, and each annotation utilizes different sources and criteria to define the risk of liver injury. For example, in the Xu et al. dataset, drugs with more than 10 clinical reports of Hy’s law cases are positive, while in the Sakatis et al. dataset, having 50 case reports of clinically significant hepatotoxicity was set as a threshold for drugs to be considered as DILI positive. Furthermore, to establish causality, the data needs to be evaluated by accepted methods for causality assessment such as RUCAM [24
], though expert opinion is of equal importance [25
]. Unfortunately, different outcomes among various causality assessment scales have been observed [26
]; nonetheless, for some drugs the reported evidence of hepatotoxicity is vague [27
]. Although not all of the datasets compared in the present study considered the RUCAM score, determining dose and lipophilicity in combination (or RO2) greatly improved the coherence among the different datasets, as shown in Figure 2
DILI negative annotation are even more problematic, as demonstrated by a recent report [17
] which suggested that as many as 40% of drugs defined as negatives by some means were also reported as DILI positives in other registries with established causality. Such contradictions, together with the significant variation in our assessment of risk factors, highlight the importance of selecting appropriately annotated datasets for the identification of DILI risk factors.
We therefore established a consensus annotation by cataloging the majority vote of the five selected annotations. Based on this consensus annotation, both lipophilicity and the extent of metabolism were significantly associated with increased DILI risk, and the strength of the association increased when combined with daily dose. The association between DILI risk and lipophilicity or metabolism was also consistently significant when assessed by the Chen and Greene annotations. In addition to annotation accuracy, an appropriate data analysis is crucial to ascertain correct conclusions. In a recent article, Weng et al. [29
] reported that the combination of daily dose and metabolism or logP
did not improve the prediction of drug-induced liver injury (DILI). Unfortunately, the data analysis is flawed, particularly when considering the methods for defining DILI negatives. Here, the authors made the assumption that DILI negatives could be defined by simply subtracting the positives for a given condition from the total number of drugs. In our reanalysis, however, we removed inappropriately assigned DILI negative drugs (e.g., “all human Adverse Drug Reactions (hADRs)” or “severe hADRs” in the study by Weng et al.), and found that the odd ratios for the combined daily dose and metabolism or logP
were statistically significant, a finding distorted by the inappropriate assignment of DILI negatives in the study by Weng et al. [29
]. This highlights the importance of a careful assessment of DILI annotations and the necessity of the consensus approach as employed in the present study.
The mechanisms by which DILI occurs are complex and are frequently idiosyncratic in nature; nonetheless, lipophilicity and metabolism are contributing factors. Lipophilic drugs often display high volumes of distribution and are therefore distributed amongst many different tissues and organs and need to be converted into hydrophilic metabolites to be eliminated [23
]. Furthermore, extensively metabolized drugs have a greater potential to form toxic reactive metabolites [15
]. Toxic reactive metabolites may irreversibly form covalent bonds, inhibiting transport proteins or triggering an immune response. Reactive metabolites may also interfere with mitochondrial function or cause oxidative stress. Whether or not these interactions lead to liver injury appears to be dependent on the accumulation of the reactive metabolites beyond a critical level [2
]. Hepatic exposure and the amount of parent drug or metabolite that accumulates depend on dose, which might explain why the combination of daily dose significantly improves the association between DILI risk and lipophilicity and metabolism.
Several limitations must be considered in this study. First, the lipophilicity and metabolism data used in this study were collected from different literature sources, yielding data that were likely not consistently measured or that may have been derived from different study protocols. Second, defining DILI risk associated with a drug is challenging and, without a “golden standard” annotation, relies on a variety of different sources and methods to collect data and define DILI risk, resulting in DILI annotations that differ for certain drugs. Third, our study is based on setting thresholds for daily dose and logP
which could lead to some bias, especially for the drugs with values close to the thresholds. Fourth, not all of the datasets in the present study considered the causality assessment (e.g., RUCAM score), which is essential in future studies to characterize DILI, as reports suggest that some hepatotoxicity recorded in the literature is vague [27
]. Finally, this study is a retrospective analysis and requires further validation based on a prospective study design.
Despite these limitations, our evaluation permitted the evaluation of five independently annotated datasets to overwhelmingly suggest that both lipophilicity and the extent of metabolism can be considered as risk factors of DILI. However, neither lipophilicity nor the extent of metabolism alone is a strong predictor for DILI risk, but together with daily dose, the association is clear.