Article pubs.acs.org/crt
Cite This: Chem. Res. Toxicol. 2018, 31, 494−505
Impact of Physicochemical Properties on Dose and Hepatotoxicity of Oral Drugs Paul D. Leeson* Paul Leeson Consulting Ltd, The Malt House, Main Street, Congerstone, Nuneaton, Warks CV13 6LZ, U.K.
Downloaded via UNIV OF CALIFORNIA SANTA BARBARA on June 22, 2018 at 14:21:36 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
S Supporting Information *
ABSTRACT: A database containing maximum daily doses of 1841 marketed oral drugs was used to examine the influence of physicochemical properties on dose and hepatotoxicity (drug induced liver injury, DILI). Drugs in the highest ∼20% dose range had significantly reduced mean lipophilicity and molecular weight, increased fractional surface area, increased % of acids, and decreased % of bases versus drugs in the lower ∼60% dose range. Drugs in the ∼20−40% dose range had intermediate mean properties, similar to the mean values for the full drug set. Drugs that are both large and highly lipophilic almost invariably do not have doses in the upper ∼20% range. The results show that oral druglike physicochemical properties are different according to these dose ranges, and this is consistent with maintenance of acceptable safety profiles as efficacious exposure increases. Verified DILI annotations from a compilation of >1000 approved drugs (Chen, M.; et al. Drug Discov. Today, 2016, 21, 648) were used. The drugs classified as “No DILI” (n = 163) had significantly lower dose and lipophilicity, and higher Fsp3 (fraction of carbon atoms that are sp3 hybridized) versus the “Most DILI” (n = 163) drugs. The percentages of acids were reduced and bases increased in the “No DILI” versus the “Most DILI” groups. Drugs classified as “Less DILI” or “Ambiguous DILI” had intermediate mean values of dose, lipophilicity, Fsp3, and % acids and bases. The impact of lipophilicity and Fsp3 on DILI increases in the upper 20% versus the lower 80% dose range, and a simple decision tree model predicted “No DILI” versus “Most DILI” outcomes with 82% accuracy. The model correctly classified 19 of 22 drugs (86%) that failed in development due to human hepatotoxicity. Because many oral drugs lacking DILI annotations are predicted to be “Most DILI”, the model is best used preclinically in conjunction with experimental DILI mitigation.
■
INTRODUCTION The concept that increasing drug dose and exposure increases risk of toxicity, especially idiosyncratic toxicity, is well documented.1,2 Daily dose can be viewed as the ultimate composite or multiparameter property of an oral drug, since it relies on the effective therapeutic concentration, the required target occupancy over time, absorption, clearance, volume of distribution, and dosing frequency.3 Absorption in turn is dependent on permeability and solubility. A combination of dose, solubility, and lipophilicity criteria has been recommended to guide selection of high quality oral drug candidates.4 Physicochemical properties are known to influence each of the parameters affecting dose, and free drug concentrations are inversely related to lipophilicity.5,6 This is important to consider if both dose and physical properties are to be used together in structure−toxicity studies. An example is hepatotoxicity, or drug-induced liver injury (DILI), the most common form of idiosyncratic toxicity in humans. DILI is a complex, rare, multifactorial event, occurring after weeks or months of treatment; drug exposure and formation of reactive metabolites are generally accepted as key factors.7−10 The “rule of two”, derived from 164 oral drugs, proposed that high daily dose (>100 mg) combined with high lipophilicity (cLogP > 3) increases DILI risk.11 This could be useful guidance but might be misleading because the data set is small (100 mg, but not lipophilicity, was associated with various human hepatic adverse effects among a group of 975 oral drugs.13 In contrast, hepatotoxicity models have been developed using the rule of two in combination with reactive metabolite formation14 and hepatic metabolism.15 High DILI risk is associated with Biopharmaceutics Drug Disposition Classification System (BDDCS) class 2 drugs,16 which are poorly soluble but highly permeable and generally highly lipophilic. Topological molecular properties17,18 and substructural chemical features19 could also be linked to hepatotoxicity. Quantitative structure−hepatotoxicity relationships,17,19−22 developed using differing data sets, provide better performance than the rule of two, but are not easy to understand on a physicochemical basis. Clearly, a major problem hampering analysis of DILI has been the use of data sets based on differing DILI assessments;23,24 the need for a large, well-curated, and consistently accurate database was Received: February 18, 2018 Published: May 3, 2018 494
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology
versus −p[MDD] (x-axis) were therefore assembled and trends in the complex data visualized and simplified by using moving average plots and 5% or larger dose bins, as shown in Figure 1 for lipophilicity (cLogP). There is no impact of dose on cLogP when the −p[MDD] is ∼ −2.5 (the DDD set behaves similarly). Mean cLogP in the 5% dose bins follows the moving average (Figure 1b; the binned cLogP distributions are shown in Figure 1c). The downward trajectory of lipophilicity with increasing dose affects significant numbers of drugs: −p[MDD] values above −3 and −2.5 comprise 40% and 20% of the total, respectively (Figure 1d). The general pattern seen with cLogP versus −p[MDD] is recapitulated with all other measures of lipophilicity examined (Table S1), including cLogDpH7.4 and the property forecast index34 (PFI = chromatographic cLogDph7.4 plus number of aromatic rings). In Figure 2a, the 5% dose bin mean properties for cLogP (Figure 1c) are reproduced, and Figure 2b−i shows the same 5% dose bin analysis for several other physicochemical properties, each referenced to the mean property value for the whole set. Molecular weight (Figure 2b) is highest in the lowest 5 percentile dose group and decreases below the average in the highest 15% dose range. Total polar surface area shows a complex pattern (Figure 2c), but when normalized for size using heavy atom count (Figure 2f), the pattern is consistent with the lipophilicity trend, showing increasing polar fraction in the highest 20% dose range. Aromatic ring count is reduced at both the lowest 5% and highest 20% of doses (Figure 1d), while Fsp335 shows little change except for the lowest 5% dose range, where it is increased (Figure 1e). It is notable that in the upper 20% dose range the proportion of drugs that are acids increases above the average, while bases are reduced (Figure 2g,h); there is a complex pattern in the distribution of neutral drugs by dose (Figure 2i). In Figure 2, each of the property distributions appears near to their average at the upper 20 percentile dose value. Using relative property values in each 5% bin, as % changes from the average value, shows that oral drugs in the ∼20−40 percentile dose range have mean property values close to the overall averages, unlike the upper 20% and lower 60% ranges where there is also greater variability (Figure 3). The data in Figure 3 show that the ∼20−40% dose range represents a crossover region for those properties that change as dose increases. These percentile dose ranges are not hard cutoffs but approximate values based on visual inspection of the dose−property trends in Figures 1−3; it should be noted there is considerable property variability in each of these dose ranges (Figure 1c,d). The mean physical properties of the upper 20% versus the 20− 40% and lower 60% dose ranges are listed in Table S1. Because molecular weight is reduced in the upper 20% dose range, some physical properties, including chiral atom count and sp2 atom count, show no dose-dependent differences when normalized for size (Table S1). The compression of lipophilicity and molecular weight in the upper 20% dose range has practical consequences relevant to selection of drug candidates. Of the 368 oral drugs with upper 20% dose values, only 9 (2.4%) have both cLogP > 4 and molecular weight > 400, where ADMET risk is increased (Figure S2).36 Of the nine drugs, three are iodinated
recently addressed by combining annotations from FDA drug labeling with human causality information.25 The resulting “verified” data set for severity of DILI annotated >1000 approved drugs as No DILI, Most DILI, Less DILI, and Ambiguous DILI.22,25 In this study, we have assembled an oral drug dose database containing >1800 marketed oral drugs. We show that (1) there are meaningful trends between dose and oral drug properties including lipophilicity, size, and ionization state; and (2) daily dose, in combination with the fraction of carbon atoms that are sp3-hybridized (Fsp3) and lipophilicity (cLogP), can distinguish the verified Most DILI from No DILI categories with >80% accuracy.
■
ORAL DRUG DOSE DATABASE An oral drugs database26 was updated to 2016 approvals and annotated with maximum daily dose (MDD, mg), the highest reported total dose per day, obtained from literature1,9,20,27 and online28 sources. A total of 1841 oral drugs were assigned an MDD value. The defined daily dose (DDD), “the assumed average maintenance dose per day for a drug used for its main indication in adults”,29 is a useful single-source comparative data set,13 and a total of 1261 oral drugs were assigned a DDD value. Although drug doses by weight, in milligrams, are almost universally used, the application of molar doses, instead of doses by weight, is best practice for quantitative structure− activity studies.30 Daily dose values were therefore converted to the logarithm of the molar doses, −p[MDD or DDD] = Log10 ((dose (mg)/mol wt)/1000), adjusting for salts where necessary. Increasing values of −p[MDD or DDD] reflect increasing dose size. Standard physicochemical properties of the oral drugs were calculated.31
■
RELATIONSHIPS BETWEEN PHYSICAL PROPERTIES AND ORAL DOSE Mean and quantile DDD values are overall less than 2-fold lower than MDD values (Table 1), and both sets showed Table 1. Mean and Quantile Dose Data for Oral Drug Data Setsa MDD, n = 1841 median mean 10%, 90%
DDD, n = 1261
−p[MDD]
dose, mg
−p[DDD]
dose, mg
−3.23 −3.68 −4.75, -2.11
200 853 7.5, 2000
−3.48 −3.59 −4.97, −2.30
100 542 4.1, 1500
a
MDD is maximum daily dose; DDD is defined daily dose from the World Health Organisation. −p[MDD or DDD] = Log10 ((dose (mg)/mol wt)/1000).
similar dose versus property trends. The MDD values are used in all analyses. It is interesting to note that mean −p[MDD] values have reduced over time but have been constant for the past 4 decades; in contrast, cLogP and molecular weight have increased significantly over the same period26,32 (Figure S1). Hence oral drug molecular inflation33 is not on average associated with changes to daily dose. The intention here is not to establish structure−activity relationships for oral dose, a challenging proposition,27 but to examine how druglike physicochemical properties vary, if at all, by dose. In this sense, the properties, not the dose, are the dependent variables. Scatter plots of physical properties (y-axis) 495
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology
Figure 1. Analysis of the effect of increasing dose (MDD) on oral drug lipophilicity. (a) Scatter plot of cLogP versus −p[MDD]; in the highest 20% of doses, lipophilicity is reduced. (b) Moving average plot in black, using mean cLogP values of 2.5% of drugs higher and lower in dose for each data point, and in red circles, mean cLogP values of 5 percentile dose groups. (c) Box and whisker plots generated in Microsoft Excel, showing cLogP distributions of 5 percentile dose bins. The box covers the 2nd and 3rd quartiles of data in each bin, the cross indicates the mean value, the horizontal bar marks the median value, and the whiskers show the upper and lower extents of the distribution, excluding outliers >3 quartiles from the median. (d) Box and whisker plots of the distributions of cLogP in the lower 60% range of doses, the 20−40% range, and the upper 20%, these cutoffs coming from inspection of the trends shown in panels a−c. In panels c and d, categories connected by the same letter are not different (p > 0.05, Tukey HSD; http://astatsa.com/OneWay_Anova_with_TukeyHSD/).
■
radiocontrast agents, three are prodrugs, and only two, nelfinavir and vemurafenib, were discovered in the last 40 years. The cLogP > 4 and molecular weight > 400 region is a popular chemical space for current drug discovery, since it is the most populated of the four possible cLogP/molecular weight categories in the recent patent literature37 (45% of compounds, Figure S2). These observations suggest that candidates having these properties together with −p[MDD] values > −2.5 (the upper 20 percentile value) are highly unlikely to succeed. Hence avoiding doses in the upper 20% range is critical for success in oral candidates possessing more extreme physical properties.
HEPATOTOXICITY
Of the 1036 drugs given a “verified” DILI annotation in humans,25 731 were assigned oral MDD values (Table 2). The remainder are biologicals, injectables, topicals, and seven oral clinical candidates that were not marketed due to human hepatotoxicity. These seven compounds were combined with other candidates that failed because of hepatotoxicity to provide a set of 22 compounds useful for testing physiochemical models (see later). The 731 drugs having a DILI annotation are about evenly distributed between the four categories, leaving 1110 drugs as a reference set lacking DILI annotation (Table 2). The most commonly prescribed 200 drugs (Top 200)38 contain 155 oral drugs, of which 17 (11%) are classified as No DILI and 16 496
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology
Figure 2. Property means of 5 percentile dose bins versus mean −p[MDD]P: (a) cLogP (data taken from Figure 1c); (b) molecular weight; (c) total polar surface area; (d) aromatic ring count; (e) Fsp3 (fraction of carbon atoms that are sp3 hybridized); (f) total polar surface area divided by heavy atom count; (g) % acids (negatively charged at physiological pH); (h) % bases (positively charged at physiological pH); and (i) % neutrals (uncharged at physiological pH). Mean −p[MDD] values for acids, bases, neutrals, and zwitterions are respectively −2.91, −3.59, −3.36, and −3.09. Acids have significantly higher doses than neutrals or bases (p > 0.05, Tukey HSD; http://astatsa.com/OneWay_Anova_with_TukeyHSD/). The mean values for each property are shown in red and referenced to the 20 percentile upper dose range in blue.
the lowest mean −p[MDD] and cLogP values, and the highest mean Fsp3 value. Although the mid-50% quantile property ranges of all groups overlap, it is notable that the Less DILI, Ambiguous DILI, and No Data sets showed no differences in these properties, yet their mean values are intermediate between the No DILI and Most DILI groups. There was varying statistical significance in the differences between the Less DILI and Ambiguous DILI groups versus the No DILI and Most DILI groups (Figure 4). The ion class observations are striking: acids comprise 9% and bases 54% of the No DILI group compared with 27% and 25% respectively in the Most DILI group (Figure 4). Other properties that were significantly different between the No DILI and Most DILI groups were
(10%) Most DILI (Table 2). It is interesting to note that 59% of the most prescribed oral drugs are classified as Less DILI, suggesting that this designation has not significantly restricted therapeutic application. The No DILI group contain older drugs than the other groups (Table 2), perhaps because it takes longer to identify unambiguously “clean” molecules. Spreadsheet S1 contains the DILI annotated oral drugs and their doses. The physical property distributions of the five groups in Table 2 were examined in detail, and the key results are shown in Figure 4. Four properties significantly distinguished the No DILI versus Most DILI groups: dose (−p[MDD]), lipophilicity (cLogP), Fsp3, and ionization class. The No DILI group had 497
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology
Figure 3. Percent deviation from the mean for each property in the 5 percentile dose bins, versus −p[MDD], for eight molecular properties shown in Figure 2. By visual inspection, properties are close to the overall means in the ∼20−40 percentile dose range, which represents a crossover region for those properties which are changing as dose increases.
classes where cLogP and Fsp3 are dominant properties and −p[MDD] is less important. A caveat is the relatively small number of No DILI acidic drugs (15); further data are needed to confirm the lack of a dose effect on the hepatotoxicity of acidic drugs. On the 44 Most DILI acidic drugs, 32 are carboxylic acids, which are prone to reactive metabolite formation via glucuronidation,1 where the chemical mechanism is well understood.39 With basic drugs, −p[MDD] and cLogP have the greatest impact, and Fsp3 is not different between the No DILI and Most DILI groups. It should be noted that basic drugs have higher Fsp3 values than other ion classes.26 Among neutral drugs, there is a different pattern, with mean cLogP not different between No DILI and Most DILI groups, in contrast to lower mean −p[MDD] and higher mean Fsp3 in the No DILI group. Models distinguishing No DILI from Most DILI were generated by partition (decision tree) analysis40 using dose, Fsp3, and lipophilicity (Figure 7a−c). This approach identifies property cut-off values that optimally differentiate the DILI groups. Models were tested systematically by altering the property entry sequence. Dose was the most significant single parameter, with 60 of 63 drugs having −p[MDD] < −4.2 being in the No DILI category; these drugs are categorized by dose alone. Subsequent sequential optimal cuts of Fsp3 < 0.28 provided a group dominated by Most DILI drugs, and only the remaining drugs required lipophilicity, cLogP ≥ 2.4, to differentiate No DILI and Most DILI; the resulting model differentiated No DILI from Most DILI with 82% accuracy (Figure 7c). Using milligram doses, the MDD cut off was 35 mg, with an accuracy of 80% (Table S3, entry 2). Individual partition models were generated for each ion class, using the same properties, and are consistent with the data in Figure 6, but these did not improve accuracy further over the full model (Table S3, entries 2−5). The partition DILI model provides a prediction of all the No DILI and Most DILI drugs, significantly improving on the rule of two by adding an additional parameter (Fsp3); the binary nature of the rule of two restricts prediction to about 50% of drugs (Figure 7d).
Table 2. Distribution of Verified DILI Assignments and Oral Drugs in the Top 200 Prescribed Medicines, Together with Median Year of First Drug Publication for Each Groupa oral drug category
n, MDD
No DILI Most DILI Less DILI Ambiguous DILI No Data
163 163 232 173 1110
a
n, % of Top 200 (155 orals) 17, 16, 91, 22, 9,
11% 10% 59% 14% 6%
median year of publication 1955 1969 1971 1969 1966
Note the No DILI drugs are appreciably older than the other groups.
alternative measures of lipophilicity (LogD7.5, LogD6.5, LogP (all Chemaxon), and PFI), aromatic ring count, sp2 atom count and sp2−sp3 atom count (Most DILI greater in all cases), and sp3 and chiral atom counts (Most DILI lower). Properties that were not different across all DILI groups were molecular weight, heavy atom count, rotatable bond count, hydrogen bond donors and acceptors, and total polar surface area (Table S2). Since the upper 20% daily dose range of drugs show changed properties versus lower doses (Figure 3), the impact of cLogP and Fsp3 in the upper 20% versus lower 80% doses was examined for the No DILI and Most DILI classes (Figure 5). As expected from Figure 2, among the combined set of No DILI and Most DILI drugs, cLogP was lower in the upper 20% dose range versus the lower 80%, but the cLogP difference between No DILI and Most DILI was increased by 2 units in the higher dose group (Figure 5a). Again, in agreement with Figure 2, the combined set of No DILI and Most DILI drugs did not differ in mean Fsp3, but the difference between No DILI and Most DILI is significantly increased by 0.23 in the higher dose group (Figure 5b). The results show that cLogP and Fsp3 become increasingly important in influencing hepatotoxicity as dose is increased. The influence of −p[MDD], cLogP, and Fsp3 on No DILI and Most DILI is different according to ion class, as illustrated in Figure 6; mean properties are provided in Table 3. With acidic drugs, there are clear differentiations between DILI 498
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology
Figure 4. Distributions of (a) −pMDD, (b) Fsp3, (c) cLogP, and (d) ion class in the annotated DILI groups and No Data oral drug groups (Table 2). For box and whisker plot definitions, see Figure 1. Categories connected by the same letter are not different (p > 0.05, Tukey HSD; http:// astatsa.com/OneWay_Anova_with_TukeyHSD/).
Figure 5. Distributions of (a) cLogP and (b) Fsp3 in No DILI and Most DILI groups, divided by lower 80% and upper 20% daily doses. For box and whisker plot definitions, see Figure 1. Categories connected by the same letter are not different (p > 0.05, Tukey HSD; http://astatsa.com/ OneWay_Anova_with_TukeyHSD/).
499
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology
Figure 6. Properties of No DILI and Most DILI drugs by ion class: (a) cLogP vs −p[MDD]; (b) Fsp3 vs −p[MDD]; and (c) Fsp3 vs cLogP.
Analysis of a group of 22 oral drugs (including seven from ref 25, see above) which were reported to have failed in clinical development due to human hepatotoxicity,8,41,42 plus one withdrawn marketed drug, sitexsentan,43 provide support for the partition model in Figure 7c (Table 4). Of the 22 drugs, 19 (86%) are predicted in the Most DILI category, compared to 13 (59%) predicted by the rule of two (Table 4). Recently, the rule of two was used to examine the hepatotoxicity risk of some
drugs marketed for treatment of hepatitis C virus (HCV), namely paritaprevir, ombitasvir, dasabuvir, simeprevir, asunaprevir, and ritonavir.44 Except for ombitasvir, all these drugs have doses >100 mg and have cLogP > 3, and therefore fail the rule of two. Application of the partition model gave an identical result (Table S4; ritonavir is a Most DILI drug used in the model generation). 500
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology Table 3. Mean Properties and Standard Deviations of “No DILI” and “Most DILI” Drugs According to Ion Class mean (standard deviation) acids −p[MDD] cLogP Fsp3
bases
neutrals
No DILI, n = 15
Most DILI, n = 44
No DILI, n = 88
Most DILI, n = 41
No DILI, n = 48
Most DILI, n = 72
−3.30 (1.87) −0.4 (2.58) 0.54 (0.28)
−2.70a (0.78) 3.7b (2.06) 0.2b (0.20)
−3.82 (0.89) 2.2 (0.28) 0.51 (0.22)
−3.09b (0.53) 3.5b (0.42) 0.44a (0.24)
−3.56 (1.52) 2.2 (3.25) 0.50 (0.27)
−2.94b (0.63) 2.5a (2.47) 0.30b (0.25)
a
No difference between No DILI and Most DILI. bNo DILI and Most DILI are significantly different (p < 0.05, Tukey HSD; http://astatsa.com/ OneWay_Anova_with_TukeyHSD/).
Figure 7. Partition or decision tree analysis of No DILI (n = 163) versus Most DILI (n = 163) oral drugs. (a) Plot of Fsp3 versus −p[MDD]; the first and most significant cut of the data is at −p[MDD] < −4.2, where 60/63 drugs are assigned No DILI. The second cut is at −p[MDD] ≥ −4.2 and Fsp3 < 0.28, where 81/92 drugs are assigned to Most DILI. (b) The remaining drugs are differentiated by cLogP, with an optimal cut of cLogP ≥ 2.4. (c) Counts of No DILI and Most DILI drugs in the property ranges defined by the cuts in panels a and b; the left-hand four bars predict No DILI (128/163, 78.5% correct), and the right-hand four bars predict Most DILI (139/163, 85.3% correct). The model is 81.9% accurate overall, odds ratio = 21.2 (95% confidence interval = 12.0−37.5; p < 0.0001; https://www.medcalc.org/calc/odds_ratio.php). (d). Performance of the model across all DILI categories and No Data sets, in comparison with the rule of two.
an effect of molecular three-dimensionality on DILI. Since by definition, for carbon atoms in a molecule, Fsp3 = 1 − Fsp2 − Fsp1, increasing Fsp3 is identical to reducing Fsp2 and Fsp1. The chemical mechanisms of bioactivation leading to idiosyncratic toxicity often occur at via π-electron-rich carbon
What is the physical interpretation of the role of the Fsp3 parameter in the DILI model? Moments of inertia are alternative molecular shape parameters describing rod-, sphere-, and disc-likeness,45 but these did not distinguish the DILI groups (data not shown), suggesting that Fsp3 may not reflect 501
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology Table 4. DILI Predictions for Drugs That Failed in Clinical Development Due to Hepatotoxicity in Humansa DILI prediction compound Darbufelone Fialuridine Pralnacasan Zamifenacin TAK-875 LY-2409021 MK-0893 Fiduxosin CP-457920 CP-085958 Falnidamol Pafuramidine Sitaxentand ADX-10059 CP-368296 Telcagepant CP-724714 CP-422935 Tasosartan Solithromycin CP-456773 Aplaviroc predicted Most DILI
ref
max daily dose, mg
−p[MDD]
cLogP
Fsp3
partition modelb
8 25 25 8 8 8 8 25 8 8 25 25 43 8 8 41 8 8 25 42 8 25
10 19 1200 40 50 90 120 120 120 200 200 200 300 200 300 560 500 500 600 800 1200 1600
−4.52 −4.29 −2.64 −4.02 −4.02 −3.79 −3.69 −3.67 −3.43 −3.37 −3.29 −3.26 −3.18 −3.08 −3.18 −3.01 −2.97 −2.88 −2.84 −3.02 −2.53 −2.56
3.7 0.0 2.2 6.0 4.7 7.4 7.8 4.9 2.2 4.6 3.8 4.8 3.4 4.1 2.4 4.0 4.6 6.8 2.5 3.7 3.4 3.9
0.44 0.56 0.46 0.33 0.34 0.38 0.16 0.33 0.17 0.20 0.33 0.10 0.22 0.13 0.30 0.46 0.19 0.52 0.22 0.70 0.45 0.55
No DILI No DILI No DILI Most DILI Most DILI Most DILI Most DILI Most DILI Most DILI Most DILI Most DILI Most DILI Most DILI Most DILI Most DILIe Most DILI Most DILI Most DILI Most DILI Most DILI Most DILI Most DILI 19/22 (86%)
rule of twoc No DILI
Most DILI Most DILI Most Most Most Most Most
DILI DILI DILI DILI DILI
Most DILI Most DILI Most DILI Most DILI Most DILI Most DILI 13/22 (59%)
a
The partition model is 86% correct, compared with the rule of two, which cannot assign all drugs and is 59% correct. bFrom Figure 7c. cMost DILI: cLogP >3, dose >100 mg. No DILI: cLogP < 3, dose 200 compounds. Chem. Res. Toxicol. 25, 2067−2082. (10) Schadt, S., Simon, S., Kustermann, S., Boess, F., McGinnis, C., Brink, A., Lieven, R., Fowler, S., Youdim, K., Ullah, M., Marschmann, M., Zihlmann, C., Siegrist, Y. M., Cascais, A. C., Di Lenarda, E., Durr, E., Schaub, N., Ang, X., Starke, V., Singer, T., Alvarez-Sanchez, R., Roth, A. B., Schuler, F., and Funk, C. (2015) Minimizing DILI risk in drug discovery - a screening tool for drug candidates. Toxicol. In Vitro 30, 429−437. (11) Chen, M., Borlak, J., and Tong, W. (2013) High lipophilicity and high daily dose of oral medications are associated with significant risk for drug-induced liver injury. Hepatology 58, 388−396. (12) Yu, K., Geng, X., Chen, M., Zhang, J., Wang, B., Ilic, K., and Tong, W. (2014) High daily dose and being a substrate of cytochrome P450 enzymes are two important predictors of drug-induced liver injury. Drug Metab. Dispos. 42, 744−750. (13) Weng, Z., Wang, K., Li, H., and Shi, Q. (2015) A comprehensive study of the association between drug hepatotoxicity and daily dose, liver metabolism, and lipophilicity using 975 oral medications. Oncotarget 6, 17031−17038. (14) Chen, M., Borlak, J., and Tong, W. (2016) A model to predict severity of drug-induced liver injury in humans. Hepatology 64, 931− 940. (15) McEuen, K., Borlak, J., Tong, W., and Chen, M. (2017) associations of drug lipophilicity and extent of metabolism with druginduced liver injury. Int. J. Mol. Sci. 18, 1335.
ASSOCIATED CONTENT
* Supporting Information S
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.chemrestox.8b00044. Table S1, mean physicochemical properties of oral drugs by lower 60%, 20−40%, and upper 20% dose ranges; Table S2, mean physical properties by DILI class; Table S3, most DILI vs NO DILI partition analyses; Table S4, DILI prediction of HCV drugs; Figure S1, distributions of −p[MDD], molecular weight, and cLogP for 1841 oral drugs by decade of first publication; Figure S2, oral drugs in the lower 80% and upper 20% dose ranges; Figure S3, DILI category of oral drugs according to BDDCS classification; and Spreadsheet S1, DILI annotations and doses of oral drugs (PDF)
■
REFERENCES
AUTHOR INFORMATION
Corresponding Author
*E-mail:
[email protected]. ORCID
Paul D. Leeson: 0000-0003-0212-3437 Notes
The author declares no competing financial interest.
■
ACKNOWLEDGMENTS The author is grateful for many helpful discussions with Martin Bayliss, Scott Boyer, James Butler, Minjun Chen, Darren Green, 503
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology (16) Chan, R., and Benet, L. Z. (2017) Evaluation of DILI predictive hypotheses in early drug development. Chem. Res. Toxicol. 30, 1017− 1028. (17) Xu, Y., Dai, Z., Chen, F., Gao, S., Pei, J., and Lai, L. (2015) Deep learning for drug-induced liver injury. J. Chem. Inf. Model. 55, 2085− 2093. (18) Rincón-Villamizar, E., and Restrepo, G. (2014) Rules relating hepatotoxicity with structural attributes of drugs. Toxicol. Environ. Chem. 96, 594−613. (19) Zhang, H., Ding, L., Zou, Y., Hu, S.-Q., Huang, H.-G., Kong, W.B., and Zhang, J. (2016) Predicting drug-induced liver injury in human with naive Bayes classifier approach. J. Comput.-Aided Mol. Des. 30, 889−898. (20) Chen, M., Hong, H., Fang, H., Kelly, R., Zhou, G., Borlak, J., and Tong, W. (2013) Quantitative structure-activity relationship models for predicting drug-induced liver injury based on FDA-approved drug labeling annotation and using a large collection of drugs. Toxicol. Sci. 136, 242−249. (21) Mulliner, D., Schmidt, F., Stolte, M., Spirkl, H.-P., Czich, A., and Amberg, A. (2016) Computational models for human and animal hepatotoxicity with a global application scope. Chem. Res. Toxicol. 29, 757−767. (22) Hong, H., Thakkar, S., Chen, M., and Tong, W. (2017) Development of decision forest models for prediction of drug-induced liver injury in humans using a large set of FDA-approved drugs. Sci. Rep. 7, 17311. (23) Björnsson, E. S., and Hoofnagle, J. H. (2016) Categorization of drugs implicated in causing liver injury: critical assessment based on published case reports. Hepatology 63, 590−603. (24) Luo, G., Shen, Y., Yang, L., Lu, A., and Xiang, Z. (2017) A review of drug induced liver injury databases. Arch. Toxicol. 91, 3039− 3049. (25) Chen, M., Suzuki, A., Thakkar, S., Yu, K., Hu, C., and Tong, W. (2016) DILIrank: the largest reference drug list ranked by the risk for developing drug-induced liver injury in humans. Drug Discovery Today 21, 648−653. (26) Leeson, P. D., St-Gallay, S. A., and Wenlock, M. C. (2011) Impact of ion class and time on oral drug molecular properties. MedChemComm 2, 91−105. (27) Matthews, E. J., Kruhlak, N. L., Benz, R. D., and Contrera, J. F. (2004) Assessment of the health effects of chemicals in humans: I. QSAR estimation of the maximum recommended therapeutic dose (MRTD) and no effect level (NOEL) of organic chemicals based on clinical trial data. Curr. Drug Discovery Technol. 1, 61−76. (28) Online prescription information, including: (a) DSSTox (FDAMDD) FDA Maximum (Recommended) Daily Dose Database, https://pubchem.ncbi.nlm.nih.gov/bioassay/1195#section=Top; (b) U.S. Food & Drug Administration, Drugs@FDA: FDA Approved Drug Products, http://www.accessdata.fda.gov/scripts/cder/ drugsatfda/index.cfm; (c) Drugs.com, http://www.drugs.com/; (d) RxList, Drugs A-Z, http://www.rxlist.com/drugs/alpha_a.htm; (e) U.S. Food & Drug Administration, Liver Toxicity Knowledge Base (LTKB), http://www.fda.gov/scienceresearch/bioinformaticstools/ livertoxicityknowledgebase/default.htm; (f) Mims, https://www. mims.com. (29) World Health Organisation, Defined daily dose: Definition and general considerations, http://www.whocc.no/ddd/definition_and_ general_considera/. (30) Dearden, J. C., Cronin, M. T. D., and Kaiser, K. L. E. (2009) How not to develop a quantitative structure−activity or structure− property relationship (QSAR/QSPR). SAR QSAR Environ. Res. 20, 241−266. (31) (a) Biobyte, ClogP, http://www.biobyte.com/index.html; (b) Chemaxon, Calculators and Predictors,https://www.chemaxon.com/ products/calculator-plugins/property-predictors/; (c) GSK internal property calculation. (32) Leeson, P. D., and Springthorpe, B. (2007) The influence of drug-like concepts on decision-making in medicinal chemistry. Nat. Rev. Drug Discovery 6, 881−890.
(33) Leeson, P. D. (2016) Molecular inflation, attrition and the rule of five. Adv. Drug Delivery Rev. 101 (Suppl. C), 22−33. (34) Young, R. J., Green, D. V., Luscombe, C. N., and Hill, A. P. (2011) Getting physical in drug discovery II: the impact of chromatographic hydrophobicity measurements and aromaticity. Drug Discovery Today 16, 822−830. (35) Lovering, F., Bikker, J., and Humblet, C. (2009) Escape from flatland: increasing saturation as an approach to improving clinical success. J. Med. Chem. 52, 6752−6756. (36) Gleeson, M. P. (2008) Generation of a set of simple, interpretable ADMET rules of thumb. J. Med. Chem. 51, 817−834. (37) Leeson, P. D., and St-Gallay, S. A. (2011) The influence of the ’organizational factor’ on compound quality in drug discovery. Nat. Rev. Drug Discovery 10, 749−765. (38) Symphony Health Solutions, Top 200 Drugs of 2014, http:// symphonyhealth.com/wp-content/uploads/2015/05/Top-200-Drugsof-2014.pdf (39) Potter, T., Lewis, R., Luker, T., Bonnert, R., Bernstein, M. A., Birkinshaw, T. N., Thom, S., Wenlock, M., and Paine, S. (2011) In silico prediction of acyl glucuronide reactivity. J. Comput.-Aided Mol. Des. 25, 997−1005. (40) Wessa, P. (2016) Recursive partitioning (regression trees) (v1.0.5) in Free Statistics Software (v1.2.1), Office for Research Development and Education, URL: http://www.wessa.net/rwasp_ regression_trees.wasp/. (41) Ho, T. W., Connor, K. M., Zhang, Y., Pearlman, E., Koppenhaver, J., Fan, X., Lines, C., Edvinsson, L., Goadsby, P. J., and Michelson, D. (2014) Randomized controlled trial of the CGRP receptor antagonist telcagepant for migraine prevention. Neurology 83, 958−966. (42) Owens, B. (2017) Solithromycin rejection chills antibiotic sector. Nat. Biotechnol. 35, 187−188. (43) Gov.UK, Sitaxentan (Thelin): worldwide withdrawal from the market, https://www.gov.uk/drug-safety-update/sitaxentan-thelinworldwide-withdrawal-from-the-market. (44) Mishra, P., and Chen, M. (2017) Direct-acting antivirals for chronic hepatitis C: can drug properties signal potential for liver injury? Gastroenterology 152, 1270−1274. (45) Sauer, W. H. B, and Schwarz, M. K. (2003) Molecular shape diversity of combinatorial libraries: a prerequisite for broad bioactivity. J. Chem. Inf. Comput. Sci. 43, 987−1003. (46) Bowes, J., Brown, A. J., Hamon, J., Jarolimek, W., Sridhar, A., Waldron, G., and Whitebread, S. (2012) Reducing safety-related drug attrition: the use of in vitro pharmacological profiling. Nat. Rev. Drug Discovery 11, 909−922. (47) Liu, T., Oprea, T., Ursu, O., Hasselgren, C., and Altman, R. B. (2016) Estimation of maximum recommended therapeutic dose using predicted promiscuity and potency. Clin. Transl. Sci. 9, 311−320. (48) Smith, D. A., Beaumont, K., Maurer, T. S., and Di, L. (2015) Volume of distribution in drug design. J. Med. Chem. 58, 5691−5698. (49) Sutherland, J. J., Raymond, J. W., Stevens, J. L., Baker, T. K., and Watson, D. E. (2012) Relating molecular properties and in vitro assay results to in vivo drug disposition and toxicity outcomes. J. Med. Chem. 55, 6455−6466. (50) Muller, P. Y., and Milton, M. N. (2012) The determination and interpretation of the therapeutic index in drug development. Nat. Rev. Drug Discovery 11, 751−761. (51) Tarcsay, Á ., and Keserű , G. M. (2013) Contributions of molecular properties to drug promiscuity. J. Med. Chem. 56, 1789− 1795. (52) Williams, H. D., Trevaskis, N. L., Charman, S. A., Shanker, R. M., Charman, W. N., Pouton, C. W., and Porter, C. J. H. (2013) Strategies to address low drug solubility in discovery and development. Pharmacol. Rev. 65, 315−499. (53) Blomme, E. A. G., and Will, Y. (2016) Toxicology strategies for drug discovery: present and future. Chem. Res. Toxicol. 29, 473−504. (54) Muthas, D., Boyer, S., and Hasselgren, C. (2013) A critical assessment of modeling safety-related drug attrition. MedChemComm 4, 1058−1065. 504
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
Article
Chemical Research in Toxicology (55) Proctor, W. R., Foster, A. J., Vogt, J., Summers, C., Middleton, B., Pilling, M. A., Shienson, D., Kijanska, M., Ströbel, S., Kelm, J. M., Morgan, P., Messner, and Williams, D. (2017) Utility of spherical human liver microtissues for prediction of clinical drug-induced liver injury. Arch. Toxicol. 91, 2849−2863. (56) Bueters, T., Gibson, C., and Visser, S. A. (2015) Optimization of human dose prediction by using quantitative and translational pharmacology in drug discovery. Future Med. Chem. 7, 2351−2369. (57) Morgan, P., Van Der Graaf, P. H., Arrowsmith, J., Feltner, D. E., Drummond, K. S., Wegner, C. D., and Street, S. D. A. (2012) Can the flow of medicines be improved? Fundamental pharmacokinetic and pharmacological principles toward improving phase II survival. Drug Discovery Today 17, 419−424.
505
DOI: 10.1021/acs.chemrestox.8b00044 Chem. Res. Toxicol. 2018, 31, 494−505
