Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD 50 Values: An Attempt to Make Some Reparation

Abstract

The prediction of human toxicities from animal toxicity tests is often poor, and is now discouraged and in some cases banned, especially those involving the LD₅₀ test. However, there is a vast number of historical LD₅₀ data in both public and in-house repositories that are being put to little use. This study examined the correlations between human lethality (doses and concentrations) of 36 MEIC chemicals and the median values of a large number of mouse and rat LD₅₀ values obtained for four different routes of administration. The best correlations were found with mouse and rat intraperitoneal LD₅₀ values (r² = 0.838 and 0.810 for human lethal dose, and r² = 0.753 and 0.785 for human lethal concentration). The results show that excellent prediction of human lethal dose and concentration can be made, for this series of chemicals at least, by using uncurated rodent LD₅₀ values, thus offering some reparation for the millions of rodent lives sacrificed in LD₅₀ testing.

Keywords

correlation human lethal concentration human lethal dose MEIC chemicals QAAR

Introduction

Rodent acute toxicity is a standard requirement in the assessment of the potential acute toxicity of chemicals to humans.¹ However, the correlation of rodent toxicity with human lethal dose (HLD) is often low.^2,3 Many factors can affect rodent toxicity values, such as species, genetic variability, age, sex, weight, health, diet, administration route, time of assessment after administration, ambient temperature, housing conditions (e.g. isolated or aggregated), time of day/night and time of year.⁴ Many of these factors also apply to assessments of HLD values, and thus contribute to uncertainty concerning their accuracy.

Furthermore, concern has been expressed for many years about the use of animals in toxicity testing as a proxy for toxicity in humans, nowhere more so than in regard to lethality tests such as the LD₅₀ test, which determines the dose of a chemical that will kill 50% of test animals in a given time. Because of this concern, many efforts have been made to find acceptable substitutes for the rodent LD₅₀ test, such as in vitro cytotoxicity assays.^5

–11

Bernson et al.¹² proposed a ‘Multicentre Evaluation Study of In vitro Cytotoxicity’ (MEIC), based on a set of 50 very diverse organic and inorganic chemicals. The MEIC chemicals, as they came to be known, have been widely studied. Indeed, it is fair to say that the 50 chemicals can be regarded as a gold standard data set. Ekwall et al.¹³ examined the human and rodent acute oral toxicities of the first 10 of these chemicals, and found excellent correlations. They did not give correlation statistics, but from their data, the coefficients of determination r² were calculated by the present authors to be 0.917 for human versus mouse oral toxicities and 0.836 for human versus rat oral toxicities. A similar study was carried out¹⁴ on the same 10 chemicals; this study did not give HLD or rodent LD₅₀ data, but reported that the human versus mouse oral correlation had a lower prediction error (0.68 log units) than had the human versus rat oral correlation (1.04 log units). For 39 MEIC chemicals, Calleja et al.¹⁵ reported prediction errors of 0.57 log units for the human versus mouse oral LD₅₀ correlation and 0.61 log units for the human versus rat oral LD₅₀ correlation. Similar prediction errors for a slightly smaller set of MEIC chemicals were reported by Calleja et al.¹⁶

Weiss and Sawyer¹⁷ examined human–rodent acute toxicity correlations for the 27 chemicals for which they could find human lethal dosages (HLDs) from the 50 MEIC chemicals. They used mouse (M) and rat (R), oral (or) and intraperitoneal (ip), LD₅₀ values, and gave the Spearman’s Rank coefficients as: Mor = 0.737, Mip = 0.597, Ror = 0.676 and Rip = 0.439, again indicating that the mouse was a better model for the prediction of HLD.

Ekwall et al.¹⁸ published rodent and human toxicity data for the 50 MEIC chemicals, and Ekwall et al.² used those rodent LD₅₀ values and results from 61 in vitro methods to predict HLD values of the 50 MEIC chemicals. They found that mouse oral LD₅₀ values correlated reasonably well with HLD (coefficient of determination r² = 0.65), while rat oral LD₅₀ values gave a somewhat poorer correlation with HLD (r² = 0.61), again indicating that the mouse is a better model than the rat for HLD prediction. Lessigiarska et al.¹⁹ also found that mouse oral LD₅₀ values correlated better (r² = 0.683) than did rat oral LD₅₀ values (r² = 0.627) with HLD for 25 MEIC chemicals.

Although the primary aim of the MEIC studies was to examine whether cytotoxicity data could replace rodent LD₅₀ values, recent work^11,20 indicates that the apparent inherent lack of reliability in some in vitro data is an additional reason that the use of cytotoxicity alone may not be enough to model acute toxicity. There is also, of course, much variability in rodent LD₅₀ values. Karmaus et al.²¹ have recently pointed out that “in vivo acute systemic toxicity studies can produce variable results, even when conducted according to accepted test guidelines.” They examined over 21,000 rat oral LD₅₀ values, and found that some chemicals had reported LD₅₀ values ranging over at least three orders of magnitude.

It is clear that mouse and rat oral LD₅₀ values correlate reasonably well with HLD values for the MEIC chemicals. However, with the exception of the work of Weiss and Sawyer,¹⁷ there does not appear to have been any investigation of whether rodent LD₅₀ values obtained using other routes of administration correlate better with HLD values of the MEIC chemicals.

Using average rodent LD₅₀ values, calculated from a very large number of published values, covering mouse and rat oral, intraperitoneal and intravenous routes of administration, Dearden²² found that the best correlations of HLD values for up to 18 psychoactive drugs were with mouse subcutaneous (n = 10, r² = 0.842), rat intravenous (n = 14, r² = 0.823) and mouse intravenous (n = 18, r² = 0.756) LD₅₀ values. All other HLD–rodent LD₅₀ correlations had r² < 0.6.

It should be mentioned that Nendza et al.²³ stated that while “in silico predictions must include an assessment of the error range of experimental (individual) data quality,” they also commented that “less reliable data can still be adequate for risk assessment in combination with other evidence. The pooling of several studies, one or more of which may be inadequate by itself, may collectively satisfy the overall requirement for valid data.” That is probably why the average rodent LD₅₀ values used by Dearden²² gave such good HLD predictions.

The HLD–rodent LD₅₀ correlations for the MEIC chemicals discussed above are reasonable but not good, with the exception of those obtained from Ekwall et al.,¹³ which could be high because they were based on only 10 chemicals. Encouraged by the results obtained by Dearden²² using averaged rodent LD₅₀ values to predict the HLD values of psychoactive drugs, an attempt is made in the present study to use median LD₅₀ values for the MEIC chemicals to try to improve HLD predictions. Median rather than average LD₅₀ values are used,⁷ as they place less weight on the contributions of extreme values.²⁴ Although the MEIC data set is relatively small, it is very diverse, representing many closely-studied chemical classes having a wide range of physicochemical and other properties. It is therefore appropriate to use this data set to test the hypothesis that median rodent LD₅₀ values can be used to predict HLD values well. Future work could expand upon the MEIC data set, in order to validate the results of the present study.

It is stressed that this work is an attempt to make some reparation for the millions of rodents sacrificed over many decades, often for little or no purpose. It exemplifies how existing data can be used to derive new insights, and in no way does it imply that more rodent LD₅₀ tests should be performed. Madden et al.²⁵ recently reviewed the use of in silico approaches, such as quantitative structure–activity relationships (QSARs), as alternatives to animal testing, wherein predictions are based on models incorporating molecular structural information. The present study demonstrates the application of an analogous concept — namely, quantitative activity–activity relationships (QAARs). In this approach, new knowledge is derived from existing data from one species to make predictions for another. Maximising the use of data available from in-house or public repositories conforms to the Three Rs principles.²⁶

It is noted that Bailey and Balls²⁷ recently stated that: “We reiterate that we welcome any objective efforts to shed light on the value — or lack of value — of animal tests for drugs intended for human use.” The present work accords with that statement.

Methods

Data collection

Fifty MEIC chemicals were used in the historical study of Ekwall et al.¹⁸ However, since the present study used rodent LD₅₀ values calculated by software (as outlined below), as well as measured values, inorganic chemicals, xylene (because there are three isomeric xylenes) and paraquat (as the molecule carries a positive charge) were excluded. This left 36 MEIC chemicals to be studied. Rodent LD₅₀ values were collected from a number of sources (see online Supplementary Information 1). HLD values were taken from Ekwall et al.¹⁸ with the exception of atropine²⁸ and digoxin.²⁹

In addition to providing HLD values for the MEIC chemicals, Ekwall et al.¹⁸ also reported the human lethal blood concentrations (HLCs) of the chemicals in three formats: the clinical, the forensic and the peak values. The present study sought to examine how well median rodent LD₅₀ values modelled HLC values for the 36-chemical data set.

As has been reported elsewhere,^21,30,31 a wide variation in reported rodent LD₅₀ values is often evident. For example, the mouse oral LD₅₀ values for atropine ranged from 32.1 to 1040 mg/kg, the rat oral values for 2,4-dichlorophenoxyacetic acid ranged from 300.6 to 4700 mg/kg, and for lindane from 57 to 6000 mg/kg. All LD₅₀ values collected, and their sources, are provided in the Supplementary Information 1, and the calculation of median values is presented in the online Supplementary Information 2. The HLD, HLC and median LD₅₀ values are given in Table 1.

Table 1.

Human lethal doses, human forensic lethal concentrations and rodent median LD₅₀ values, for 36 MEIC chemicals.

Chemical	HLD^a (mg/kg)	HFLC^a (mg/l)	Rodent median LD₅₀ values (mg/kg)
Chemical	HLD^a (mg/kg)	HFLC^a (mg/l)	Mor	Mip	Msc	Miv	Ror	Rip	Rsc	Riv
Acetylsalicylic acid	385.2	801.1	742.3	331.0	947.3	—	1305	662.2	—	495.0
Amitriptyline	37.4	4.2	196.1	70.9	134.2	24.2	334.3	65.6	663.0	11.1
Amphetamine	5.0	6.3	29.0	16.3	6.9	16.4	43.0	25.9	80.4	20.0
Atropine	1.47	0.20	187.0	60.2	300.3	40.9	337.0	67.6	183.4	28.2
Caffeine	140.7	119.7	231.6	163.4	242.7	92.7	334.2	157.3	212.1	105.0
Chloramphenicol	2878.0	99.9	2355.0	1206.9	979.5	673.9	2764.8	1356.8	3463.4	431.5
Chloroform	993.0	97.0	246.4	698.9	704.0	—	1070.1	1110.5	—	—
Chloroquine	51.9	14.0	310.5	54.2	136.3	21.6	464.9	102.0	190.0	60.0
Dextropropoxyphene	9.1	8.0	205.4	105.4	154.1	25.1	133.8	53.3	108.6	14.2
Diazepam	71.5	31.2	486.3	111.5	489.8	25.0	638.0	118.0	6.4	32.0
Dichloromethane	1377.4	362.3	1316.7	1172.9	6460.0	—	1829.9	916.0	—	—
2,4-Dichlorophenoxyacetic acid	384.1	568.2	440.5	182.0	—	—	708.9	666.0	—	—
Digoxin	0.368	0.016	17.9	3.6	10.2	12.4	26.6	4.0	16.3	25.0
Ethanol	4714.3	4824.1	6960.4	4405.4	8871.6	2057.5	10042	4612.1	—	1540.5
Ethylene glycol	1559.1	2587.5	8892.0	3769.9	5194.0	948.4	7146.0	5010.0	2897.3	3260.0
Hexachlorophene	213.6	34.6	115.9	32.3	46.0	—	55.1	21.9	10.6	7.5
Isoniazid	172.7	131.0	194.7	120.9	156.7	170.0	1203.0	335.0	329.0	365.0
Isopropanol	2563.8	1509.6	4013.3	4200.8	6000.0	1714.0	5246.1	2047.4	—	1094.0
Lindane	241.9	1.0	95.0	80.0	—	—	185.4	35.0	750.0	—
Malathion	690.2	281.2	1109.4	637.3	221.0	184.0	1065.0	645.7	632.0	50.0
Methanol	1569.3	1886.7	9686.0	6370.4	9800.0	4710.0	6943.2	7559.6	—	2131.0
Nicotine	0.71	22.9	8.7	9.1	26.4	0.72	56.5	18.5	27.0	1.67
Orphenadrine	41.7	12.0	116.1	58.7	107.8	27.0	289.7	81.9	203.0	25.1
Paracetamol	268.8	228.8	706.1	507.1	697.7	—	2023.0	1324.3	1952.1	—
Pentachlorophenol	28.5	99.0	106.2	45.4	63.0	—	95.4	57.9	81.5	—
Phenobarbital	111.2	121.9	223.2	189.9	193.6	213.9	371.9	142.0	178.1	126.0
Phenol	156.2	76.5	327.3	180.0	347.5	112.0	678.7	127.0	371.5	—
Phenytoin	303.3	79.8	202.2	141.8	216.6	95.2	1675.7	207.7	212.0	91.4
Propranolol	62.2	11.1	374.6	80.0	165.4	30.5	518.0	66.6	100.8	16.0
Quinidine	61.8	43.8	415.0	129.9	400.0	42.8	455.8	106.0	264.9	26.9
Tetrachloromethane	1309.0	227.5	10651.0	1941.0	30690.0	—	4092.6	3202.6	—	—
Theophylline	156.9	149.9	315.1	155.1	165.1	164.6	272.6	179.7	115.4	197.2
Thioridazine	54.8	6.4	445.3	104.1	313.7	41.9	983.3	150.0	640.0	67.7
1,1,1-Trichloroethane	7165.0	171.9	2521.8	2227.9	16000.0	—	3670.7	3328.5	—	—
Verapamil	114.2	7.9	147.3	60.3	50.9	4.8	118.3	61.0	102.9	12.0
Warfarin	106.9	99.8	220.6	361.4	3576.8	165.9	48.2	232.1	—	89.5

^a Converted from data in Ekwall et al.¹⁸

HLD = human lethal dose; HFLC = human forensic lethal concentration; Mor = mouse, oral administration; Mip = mouse, intraperitoneal administration; Msc = mouse, subcutaneous administration; Miv = mouse, intravenous administration; Ror = rat, oral administration; Rip = rat, intraperitoneal administration; Rsc = rat, subcutaneous administration; Riv = rat, intravenous administration.

Rat and mouse LD₅₀ values can also be calculated by a few software programs. Gonella Diaza et al.³² found the best to be the ACD/Labs ToxSuite³³ (now called I-Lab 2) and the US Environmental Protection Agency’s Toxicity Estimation Software Tool (TEST) program.³⁴ The former calculates mouse oral (or), intraperitoneal (ip), subcutaneous (sc) and intravenous (iv) LD₅₀ values, as well as rat oral and intraperitoneal LD₅₀ values. The latter calculates rat oral LD₅₀ values only. The ACD/Labs software is a commercial program, but at present it is available free of charge to academics at British universities via the Royal Society of Chemistry website. The present study used these two software programs, and the rodent LD₅₀ values predicted by them are given in Table 2.

Table 2.

ACD I-Lab 2-predicted and TEST-predicted rodent LD₅₀ values for 36 MEIC chemicals.

Chemical	Predicted LD₅₀ values (mg/kg)
Chemical	ACDMor	ACDMip	ACDMsc	ACDMiv	ACDRor	ACDRip	TESTRor
Acetylsalicylic acid	1100	460	760	500	1400	710	757
Amitriptyline	180	67	120	26	330	120	309
Amphetamine	130	130	13	28	120	81	770
Atropine	440	110	430	39	760	160	464
Caffeine	370	260	370	100	150	100	278
Chloramphenicol	2900	710	1000	260	1700	1700	2059
Chloroform	1300	760	470	110	1600	950	1794
Chloroquine	310	140	250	19	550	95	866
Dextropropoxyphene	200	110	110	40	330	110	346
Diazepam	610	230	790	51	870	260	1772
Dichloromethane	1200	230	1300	83	1200	570	400
2,4-Dichlorophenoxyacetic acid	340	510	830	180	480	740	586
Digoxin	8.6	4.8	19	7.6	27	24	12.7
Ethanol	4400	1700	5500	1300	3500	25	672
Ethylene glycol	12000	4900	2700	2200	10000	4600	754
Hexachlorophene	130	41	44	37	170	40	308
Isoniazid	240	200	310	290	1200	180	528
Isopropanol	4100	2000	4900	890	3800	2000	2217
Lindane	220	220	12	43	710	130	513
Malathion	1100	920	120	210	1600	100	459
Methanol	4300	2500	5900	1500	3400	2600	171
Nicotine	68	57	45	7.9	220	59	1019
Orphenadrine	290	94	350	38	530	100	720
Paracetamol	830	520	410	220	2300	950	1807
Pentachlorophenol	140	57	100	56	110	100	227
Phenobarbital	450	240	290	390	180	200	539
Phenol	450	100	370	87	260	160	416
Phenytoin	570	280	310	87	1600	1100	1228
Propranolol	280	100	240	34	750	84	1793
Quinidine	730	390	170	75	380	110	553
Tetrachloromethane	2800	460	1900	99	3500	280	1803
Theophylline	160	96	270	130	61	120	171
Thioridazine	460	140	320	54	790	110	1360
1,1,1-Trichloroethane	2100	840	2000	77	3500	280	3469
Verapamil	220	69	64	2.9	380	57	365
Warfarin	1000	390	200	85	170	480	47

ACD = ACD/Labs; TEST = US Environmental Protection Agency’s Toxicity Estimation Software Tool; Mor = mouse, oral administration; Mip = mouse, intraperitoneal administration; Msc = mouse, subcutaneous administration; Miv = mouse, intravenous administration; Ror = rat, oral administration; Rip = rat, intraperitoneal administration.

All toxicity values were converted to mmol/kg and negative log-transformed for purposes of correlation. Values given for salts were corrected to the corresponding free base or acid. Carai et al.³⁵ have commented that this may not give the correct value for the free acid or base; for example, for a sodium salt, the sodium ion could have affected mortality by changing the electrolytic homeostasis. However, such salt effects could not be corrected in literature data.

Statistical analysis

Statistical analysis was carried out by using regression analysis in Minitab 19 software.³⁶

It is now accepted practice to carry out an external validation of a predictive correlation, which is achieved by using it to predict endpoint values for chemicals not included in the data set that was used to develop the correlation (i.e. the training set).³⁷ Frequently, this is done by removing, say, 20% of the chemicals from the training data set, re-developing the correlation with the remaining 80% of chemicals, and using that correlation to predict the endpoint values of the removed chemicals (i.e. the test set). Selection of the test set chemicals was done by the well-used method³⁸ of removal of every fifth chemical of the alphabetical list, namely: caffeine, diazepam, ethylene glycol, malathion, pentachlorophenol, quinidine and verapamil.

Results and Discussion

Correlation of the HLD and rodent LD₅₀ values

Correlations of the HLD values with the median measured rodent LD₅₀ values, as well as the calculated rodent LD₅₀ values for the whole data set, are given in Table 3 (see Figure 1 for the best correlation between the values). Table 4 shows the best training set QAARs for HLD values correlated with both measured and calculated rodent LD₅₀ values. The best training set QAARs were used to predict the HLD values of the seven test set chemicals, and these values are given in Table 5.

Table 3.

Correlations of the HLD values of the MEIC chemicals with experimental and calculated rodent LD₅₀ values.

Eqn	Correlation	n	r²	q²	s	F
1	log 1/HLD = 0.516 (0.101) + 1.119 (0.104) log 1/Mor	36	0.773	0.743	0.572	116.1
2	log 1/HLD = 0.189 (0.081) + 1.084 (0.082) log 1/Mip	36	0.838	0.815	0.484	175.8
3	log 1/HLD = 0.490 (0.113) + 0.940 (0.099) log 1/Msc	34	0.737	0.702	0.634	89.8
4	log 1/HLD = −0.003 (0.126) + 1.001 (0.110) log 1/Miv	26	0.776	0.729	0.601	83.1
5	log 1/HLD = 0.695 (0.115) + 1.122 (0.114) log 1/Ror	36	0.739	0.701	0.614	96.4
6	log 1/HLD = 0.252 (0.088) + 1.046 (0.087) log 1/Rip	36	0.810	0.781	0.524	145.0
7	log 1/HLD = 0.542 (0.151) + 0.790 (0.193) log 1/Rsc	26	0.411	0.237	0.765	16.7
8	log 1/HLD = −0.040 (0.143) + 0.961 (0.126) log 1/Riv	27	0.699	0.650	0.683	58.1
9	log 1/HLD = 0.702 (0.111) + 1.244 (0.120) log 1/ACDMor	36	0.759	0.733	0.590	107.3
10	log 1/HLD = 0.282 (0.103) + 1.254 (0.128) log 1/ACDMip	36	0.739	0.708	0.614	96.3
11	log 1/HLD = 0.392 (0.122) + 1.050 (0.133) log 1/ACDMsc	36	0.648	0.594	0.714	62.5
12	log 1/HLD = −0.196 (0.130) + 1.088 (0.141) log 1/ACDMiv	36	0.635	0.577	0.726	59.2
13	log 1/HLD = 0.808 (0.138) + 1.240 (0.149) log 1/ACDRor	36	0.671	0.633	0.690	69.4
14	log 1/HLD = 0.221 (0.141) + 1.110 (0.189) log 1/ACDRip	36	0.504	0.448	0.847	34.5
15	log 1/HLD = 0.756 (0.188) + 1.255 (0.243) log 1/TESTRor	36	0.439	0.378	0.901	26.6

n = number of chemicals in the model; r² = coefficient of determination (a measure of goodness of fit); q² = cross-validated coefficient of determination (a measure of robustness of the correlation); s = standard error of the prediction; F = Fisher statistic, or variance ratio (a measure of predictivity). The numbers in brackets are the standard errors on each coefficient.

HLD = human lethal dose; ACD = ACD/Labs; TEST = US Environmental Protection Agency’s Toxicity Estimation Software Tool; Mor = mouse, oral administration; Mip = mouse, intraperitoneal administration; Msc = mouse, subcutaneous administration; Miv = mouse, intravenous administration; Ror = rat, oral administration; Rip = rat, intraperitoneal administration; Rsc = rat, subcutaneous administration; Riv = rat, intravenous administration.

Figure 1.

Correlation of human lethal doses and mouse intraperitoneal LD₅₀ values, for 36 MEIC chemicals. HLD = human lethal dose; Mip = mouse, intraperitoneal administration.

Table 4.

Best correlations of the HLD values of the MEIC training set chemicals with experimental and ACD-calculated rodent LD₅₀ values.

Eqn	Correlation	n	r²	q²	s	F
16	log 1/HLD = 0.525 (0.120) + 1.167 (0.122) log 1/Mor	29	0.773	0.734	0.616	91.9
17	log 1/HLD = 0.208 (0.098) + 1.120 (0.096) log 1/Mip	29	0.835	0.806	0.525	136.9
18	log 1/HLD = 0.757 (0.139) + 1.177 (0.133) log 1/Ror	29	0.742	0.693	0.657	77.7
19	log 1/HLD = 0.283 (0.106) + 1.085 (0.102) log 1/Rip	29	0.807	0.770	0.568	112.9
20	log 1/HLD = 0.721 (0.124) + 1.329 (0.136) log 1/ACDMor	29	0.773	0.747	0.609	95.2
21	log 1/HLD = 0.264 (0.116) + 1.353 (0.146) log 1/ACDMip	29	0.753	0.720	0.635	85.4
22	log 1/HLD = 0.868 (0.159) + 1.326 (0.173) log 1/ACDRor	29	0.677	0.635	0.726	58.8

HLD = human lethal dose; ACD = ACD/Labs; Mor = mouse, oral administration; Mip = mouse, intraperitoneal administration; Ror = rat, oral administration; Rip = rat, intraperitoneal administration.

Table 5.

Observed HLD values of the MEIC test set chemicals, and those predicted from equations 16–22.

Chemical	Observed HLD (mg/kg)	HLD predicted from equation number:
Chemical	Observed HLD (mg/kg)	16	17	18	19	20	21	22
Caffeine	140.7	71.4	81.0	64.4	80.5	87.0	156.9	18.7*
Diazepam	71.5	159.0	61.5	71.9	57.1	149.0	116.1	169.7
Ethylene glycol	1559.1	6066	3829	2897	3793	12894**	2473	7104
Malathion	690.2	405.9	426.4	229.3	356.3	310.6	719.1	362.6
Pentachlorophenol	28.5	27.3	22.7	13.9	26.5	21.5	18.0	13.6
Quinidine	61.8	129.4	71.9	84.7	50.2	181.1	226.6	54.2
Verapamil	114.2	36.6	29.2*	16.3*	26.8*	32.9	19.3*	48.6

HLD = human lethal dose.

* Prediction error > 1 standard deviation; **Prediction error > 2 standard deviations.

Equations 16–19 gave predicted HLD values within 1 standard deviation (sd), with the exception of verapamil, for which three predictions were between 1 and 2 sd (see Table 5) This is still acceptable, bearing in mind the wide range of some rodent LD₅₀ values, and Lipnick’s comment³⁹ that only predicted values with errors over 3 sd should be regarded as unacceptable. Hence, equations 16–19 have been successfully externally validated.

HLD predictions from equations 20–22, based on ACD/Labs-predicted rodent LD₅₀ values, were also mostly within 1 sd, with the exception of two values with an error greater than 1 sd and two values with an error greater than 2 sd. Therefore, on the whole, the ACD/Labs I-Lab 2 software yielded good HLD predictions, with 20 out of 21 predictions being acceptable.

It can be seen from Table 3 that mouse intraperitoneal (Mip) and rat intraperitoneal (Rip) LD₅₀ values correlate excellently with HLD values; it is generally acknowledged that r² values of about 0.8 are as good as can be expected from in vivo data. Mouse oral (Mor) and rat oral (Ror) LD₅₀ values also correlate well with HLD values. These correlations are much better than that reported for HLD–Ror LD₅₀ correlation (r² = 0.571) for a 30-chemical data set.³⁰ Figure 1 illustrates the HLD–Mip LD₅₀ correlation obtained in the present work. It is interesting to note that Fry et al.⁴⁰ reported that rodent intraperitoneal LD₅₀ values were more widely available than were those for other routes of administration. However, in the Supplementary Information 2 of the present work, it can be seen that more LD₅₀ values were found for oral administration than for intraperitoneal administration.

The training set QAARs in Table 4 are very similar to those developed for the full data set, given in Table 3. The predicted HLD values in Table 5 are almost all within 1 sd of the observed HLD values, and only one (from the ACDMor prediction) has an error of more than 2 sd. Thus, any of the equations 16–22, and particularly equations 16–19 based on median measured rodent LD₅₀ values, can be used to predict human lethal doses of MEIC chemicals.

HLD values also correlated well with ACD/Labs-predicted Mor and Mip LD₅₀ values, but less well with ACD/Labs-predicted LD₅₀ values via other routes of administration. As can be seen from Table 6, the ACD/Labs I-Lab 2-predicted LD₅₀ values all correlated very well (r² ≥ 0.88) with the respective median measured LD₅₀ values, with the exception of the Rip LD₅₀ values (r² = 0.70). One reason for these good correlations may be that, with one exception, all the MEIC chemicals are in the ACD/Labs I-Lab 2 database.³³ However, it should be noted that all but one of the ACD/Labs I-Lab 2-predicted LD₅₀ values were different from the measured values in their database.

Table 6.

Correlations of the measured and calculated rodent LD₅₀ values.

Eqn	Correlation	n	r²	q²	s	F
38	log 1/Mor = 0.144 (0.058) + 1.059 (0.063) log 1/ACDMor	36	0.892	0.872	0.311	280.3
39	log 1/Mip = 0.087 (0.058) + 1.166 (0.068) log 1/ACDMip	36	0.897	0.885	0.326	296.0
40	log 1/Msc = −0.035 (0.069) + 1.174 (0.076) log 1/ACDMsc	34	0.882	0.870	0.388	240.1
41	log 1/Miv = 0.009 (0.064) + 1.066 (0.062) log 1/ACDMiv	26	0.925	0.907	0.305	297.1
42	log 1/Ror = 0.096 (0.061) + 1.094 (0.066) log 1/ACDRor	36	0.890	0.876	0.305	275.9
43	log 1/Rip = −0.027 (0.094) + 1.128) (0.126) log 1/ACDRip	36	0.703	0.677	0.564	80.4
44	log 1/Ror = 0.063 (0.120) + 1.135 (0.155) log 1/TESTRor	36	0.611	0.554	0.575	53.5

The TEST-predicted Ror LD₅₀ values correlated very poorly with HLD values (r² = 0.439), which is disappointing. The TEST software-predicted Ror LD₅₀ values also correlated only poorly (r² = 0.611) with the median measured Ror LD₅₀ values. Thus, it can be concluded that the ACD/Labs-predicted mouse and rat oral LD₅₀ values are the better choice of the two to predict the HLD values of the MEIC chemicals.

Ekwall et al.² found that rat and mouse oral LD₅₀ values correlated well with each other (r² = 0.852) for the 50 MEIC chemicals, a finding also reported by Hoffmann et al.³⁰ in their study based on 40 MEIC chemicals (r² = 0.800). Table 7 shows rat–mouse LD₅₀ correlations for the log-transformed values of the data given in Table 1, with a better correlation value (r² = 0.883) than those calculated by Ekwall et al.² and by Hoffmann et al.³⁰ It is not surprising that rat–mouse LD₅₀ values correlate better than do HLD–rodent values, since the two rodent species share more similar physiology. Indeed, some researchers have used rat and mouse LD₅₀ values interchangeably in studies.⁸

Table 7.

QAARs of rat–mouse LD₅₀ values.

Eqn	Correlation	n	r²	q²	s	F
45	log 1/Ror = −0.184 (0.056) + 0.917 (0.057) log 1/Mor	36	0.883	0.870	0.316	255.3
46	log 1/Rip = −0.061 (0.035) + 0.997 (0.036) log 1/Mip	36	0.958	0.954	0.212	772.3
47	log 1/Rsc = −0.017 (0.112) + 0.819(0.151) log 1/Msc	25	0.562	0.488	0.542	29.5
48	log 1/Riv = 0.046 (0.061) + 0.919 (0.052) log 1/Miv	25	0.932	0.919	0.282	315.1
49	log 1/ACDRor = −0.141 (0.057) + 0.874 (0.061) log 1/ACDMor	36	0.857	0.847	0.300	204.2
50	log 1/ACDRip = 0.024 (0.068) + 0.792 (0.085) log 1/ACDMip	36	0.721	0.666	0.406	87.8

ACD = ACD/Labs; QAAR = quantitative activity–activity relationship; Mor = mouse, oral administration; Mip = mouse, intraperitoneal administration; Msc = mouse, subcutaneous administration; Miv = mouse, intravenous administration; Ror = rat, oral administration; Rip = rat, intraperitoneal administration; Rsc = rat, subcutaneous administration; Riv = rat, intravenous administration.

Table 7 shows that the correlation between log 1/Rsc and log 1/Msc is very poor. Table 3 shows that the correlation between log 1/HLD and log 1/Rsc (r² = 0.411) is much poorer than the correlation between log 1/HLD and log 1/Msc (r² = 0.737), suggesting that it is the Rsc data that are largely at fault. This could be because only relatively few Rsc LD₅₀ values were found (see Supplementary Information 2), with only a single Rsc LD₅₀ value having been found for many of the studied MEIC chemicals.

The use of rodent LD₅₀ values to model human lethal blood concentrations

Of the three HLC formats (i.e. forensic, clinical and peak), the forensic values correlated best with rodent LD₅₀ values. For example, r² values for the correlation of forensic, clinical and peak HLC values with Mip values were 0.753, 0.683 and 0.651, respectively. Therefore, only human forensic lethal concentration (HFLC) values were subsequently used, and their correlations with median measured and calculated rodent LD₅₀ values for the whole data set are given in Table 8 (see Figure 2 for the best correlation between the values). Table 9 gives the best training set QAARs for HFLC values correlated with both measured and calculated rodent LD₅₀ values. The best training set QAARs were used to predict the HFLC values of the seven test set chemicals, and these values are given in Table 10.

Table 8.

Correlations of the HFLC values of the MEIC chemicals with experimental and calculated rodent LD₅₀ values.

Eqn	Correlation	n	r²	q²	s	F
23	log 1/HFLC = 3.930 (0.138) + 1.188 (0.142) log 1/Mor	36	0.672	0.619	0.784	69.7
24	log 1/HFLC = 3.583 (0.113) + 1.170 (0.115) log 1/Mip	36	0.753	0.716	0.680	103.7
25	log 1/HFLC = 3.846 (0.148) + 0.953 (0.130) log 1/Msc	34	0.628	0.563	0.829	54.1
26	log 1/HFLC = 3.259 (0.175) + 1.112 (0.152) log 1/Miv	26	0.689	0.610	0.834	53.2
27	log 1/HFLC = 4.126 (0.151) + 1.204 (0.149) log 1/Ror	36	0.657	0.605	0.802	65.1
28	log 1/HFLC = 3.655 (0.106) + 1.172 (0.105) log 1/Rip	36	0.785	0.749	0.635	123.8
29	log 1/HFLC = 3.939 (0.205) + 0.850 (0.261) log 1/Rsc	26	0.307	0.111	1.033	10.6
30	log 1/HFLC = 3.188 (0.179) + 1.123 (0.158) log 1/Riv	27	0.670	0.615	0.854	50.8
31	log 1/HFLC = 4.158 (0.133) + 1.393 (0.144) log 1/ACDMor	36	0.734	0.704	0.706	94.0
32	log 1/HFLC = 3.693 (0.110) + 1.457 (0.136) log 1/ACDMip	36	0.771	0.742	0.655	114.4
33	log 1/HFLC = 3.823 (0.131) + 1.232 (0.142) log 1/ACDMsc	36	0.689	0.648	0.763	75.5
34	log 1/HFLC = 3.108 (0.120) + 1.353 (0.131) log 1/ACDMiv	36	0.759	0.714	0.672	106.9
35	log 1/HFLC = 4.231 (0.180) + 1.298 (0.194) log 1/ACDRor	36	0.568	0.507	0.899	44.8
36	log 1/HFLC = 3.624 (0.153) + 1.324 (0.204) log 1/ACDRip	36	0.554	0.498	0.914	42.2
37	log 1/HFLC = 4.222 (0.216) + 1.411 (0.279) log 1/TESTRor	36	0.429	0.341	1.035	25.5

HFLC = human forensic lethal concentration; ACD = ACD/Labs; TEST = US Environmental Protection Agency’s Toxicity Estimation Software Tool; Mor = mouse, oral administration; Mip = mouse, intraperitoneal administration; Msc = mouse, subcutaneous administration; Miv = mouse, intravenous administration; Ror = rat, oral administration; Rip = rat, intraperitoneal administration; Rsc = rat, subcutaneous administration; Riv = rat, intravenous administration.

Figure 2.

Correlation of human forensic lethal concentrations and rat intraperitoneal LD₅₀ values, for 36 MEIC chemicals. HFLC = human forensic lethal concentration; Rip = rat, intraperitoneal administration.

Table 9.

Best correlations of the HFLC values of the MEIC training set chemicals with experimental and ACD-calculated rodent LD₅₀ values.

Eqn	Correlation	n	r²	q²	s	F
38	log 1/HFLC = 3.979 (0.165) + 1.210 (0.167) log 1/Mor	29	0.661	0.591	0.845	52.6
39	log 1/HFLC = 3.651 (0.135) + 1.191 (0.132) log 1/Mip	29	0.751	0.705	0.724	81.3
40	log 1/HFLC = 4.231 (0.179) + 1.244 (0.172) log 1/Ror	29	0.660	0.647	0.846	52.3
41	log 1/HFLC = 3.736 (0.124) + 1.203 (0.120) log 1/Rip	29	0.789	0.746	0.667	100.8
42	log 1/HFLC = 4.224 (0.154) + 1.466 (0.167) log 1/ACDMor	29	0.740	0.707	0.740	76.7
43	log 1/HFLC = 3.727 (0.127) + 1.558 (0.159) log 1/ACDMip	29	0.781	0.749	0.679	96.2

HFLC = human forensic lethal concentration; ACD = ACD/Labs; Mor = mouse, oral administration; Mip = mouse, intraperitoneal administration; Ror = rat, oral administration; Rip = rat, intraperitoneal administration.

Table 10.

Observed HFLC values of the MEIC test set chemicals, and those predicted from equations 38–43.

Chemical	Observed HFLC (mg/kg)	HFLC predicted from equation number:
Chemical	Observed HFLC (mg/kg)	38	39	40	41	42	43
Caffeine	119.7	25.2	35.3	22.4	27.7	29.8	57.3
Diazepam	31.2	57.1	20.8	45.7	18.1	51.9	38.3
Ethylene glycol	2587	2648	1845	1336	2243	8335	10517
Malathion	281.2	150.3	161.4	83.2	135.8	115.1	305.5
Pentachlorophenol	99.0	9.2*	7.2**	4.4**	7.8**	6.2**	4.5**
Quinidine	43.8	45.8	24.4	29.1	15.5	63.5	81.1
Verapamil	7.9	12.2	9.2	5.0	7.5	9.4	4.5

HFLC = human forensic lethal concentration.

* Prediction error > 1 standard deviation; **Prediction error > 2 standard deviations.

In Table 8, it can be seen that most of the median rodent log 1/LD₅₀ values correlate well with log 1/HFLC values, with only two correlations having r² values of < 0.65. Two correlations based on ACD/Labs-predicted values also had r² values of < 0.65. Table 9 shows the best of those correlations developed from the training set, and Table 10 gives the predicted HFLC values derived from equations 38–43 for the test set chemicals. With the exception of those for pentachlorophenol, the predicted HFLC values were all within 1 sd of the observed values, and even the pentachlorophenol values were less than 3 sd in error, which is acceptable under Lipnick’s guidance.³⁸ Hence, any of the equations 38–43 can be used to predict human forensic lethal concentrations of MEIC chemicals with acceptable accuracy.

Ekwall et al.² also examined correlations between the HLC values of the MEIC chemicals and the IC₅₀ values from 61 in vitro cytotoxicity tests, and found that on the whole they were poor, with only 5 out of 150 correlations having r² values of ≥ 0.7. It should be noted that in these five instances, IC₅₀ values from fewer than the full set of 50 MEIC chemicals were used in the correlations, implying perhaps that some data were omitted to improve the extent of correlation.

The use of ‘lowest’ rodent LD₅₀ values to model human acute toxicity values

Several studies^8,17,41 have used the lowest available rodent LD₅₀ values, in order to obtain worst case estimates of toxicity. From the very large number of rodent LD₅₀ values collected for the present study (see Supplementary Information 2), lowest values were used to determine whether better predictions of HLD could be obtained from these data than from the use of median LD₅₀ values. The QAARs developed from these data are given in Table 11.

Table 11.

Correlations of the HLD values of the MEIC chemicals with minimum (worst case) experimental rodent LD₅₀ values.

Eqn	Correlation	n	r²	q²	s	F
51	log 1/HLD = 0.113 (0.120) + 0.914 (0.117) log 1/Mor_min	36	0.642	0.591	0.719	61.0
52	log 1/HLD = −0.150 (0.101) + 1.093 (0.103) log 1/Mip_min	36	0.769	0.732	0.578	113.0
53	log 1/HLD = 0.370 (0.106) + 0.927 (0.092) log 1/Msc_min	34	0.760	0.725	0.607	101.1
54	log 1/HLD = −0.158 (0.133) + 0.945 (0.104) log 1/Miv_min	26	0.775	0.719	0.601	82.8
55	log 1/HLD = 0.257 (0.138) + 0.876 (0.141) log 1/Ror_min	36	0.534	0.440	0.821	38.9
56	log 1/HLD = 0.063 (0.091) + 1.040 (0.091) log 1/Rip_min	36	0.794	0.760	0.546	131.1
57	log 1/HLD = 0.457 (0.147) + 0.814 (0.177) log 1/Rsc_min	26	0.468	0.310	0.727	21.1
58	log 1/HLD = −0.107 (0.153) + 0.895 (0.124) log 1/Riv_min	27	0.674	0.623	0.711	51.8

HLD = human lethal dose; Mor = mouse, oral administration; Mip = mouse, intraperitoneal administration; Msc = mouse, subcutaneous administration; Miv = mouse, intravenous administration; Ror = rat, oral administration; Rip = rat, intraperitoneal administration; Rsc = rat, subcutaneous administration; Riv = rat, intravenous administration.

A comparison of the QAARs given in Table 11 with equations 1–8 in Table 3 shows that almost all those in Table 3 have better statistics than do those in Table 11, indicating that the use of median rodent LD₅₀ values is preferable to the use of the worst case (i.e. ‘lowest’) rodent LD₅₀ values for the prediction of HLD and probably other toxicity endpoints. The main exceptions involve intravenous LD₅₀ values, as equations 4 and 54 are almost identical. It can be seen from Supplementary Information 2 that not only are there generally fewer intravenous LD₅₀ values available, but their range is usually much narrower than those obtained from other routes of administration. It would thus appear that the intravenous route gives more accurate LD₅₀ values than does administration by other routes.

Weiss and Sawyer¹⁷ used the 50 MEIC chemicals in their study of the correlation of HLD values with worst case Mor, Mip, Ror and Rip LD₅₀ values, so their results can be compared directly with those found in the present study. The r² values from their data were found to be: 0.636 (Mor); 0.784 (Mip); 0.578 (Ror); and 0.769 (Rip). In each case, these r² values are appreciably lower than those of the corresponding equations 1, 2, 5 and 6 in Table 3, again indicating that the use of median rodent LD₅₀ values leads to better correlations with HLD values than does the use of ‘lowest’ rodent LD₅₀ values. It should be noted that both approaches require extensive literature searches.

Do more rodent LD₅₀ values yield more accurate median LD₅₀ values?

It might be expected that the more rodent LD₅₀ values that can be found for a given endpoint, the more accurate should be their median value. Equations 1–8 (Table 3) were developed using median rodent LD₅₀ values (given in Supplementary Information 2), while equations 51–58 (Table 11) were developed using single worst-case LD₅₀ values (given in Supplementary Information 2). The difference in the statistics between corresponding equations (e.g. 1 and 51) in Tables 3 and 11 should be a function of the number of rodent LD₅₀ values employed to calculate the median LD₅₀ values used in equations 1–8. The following correlations were obtained:

Diff . {in r}^{2} = - 0.0 766 (0.0194) + 0.0323 (0.0043) (mean no . {of LD}_{50} values)

{n = 8  r}^{2} = 0.903 q^{2} = 0.820s = 0.0289 F = 55.6

Diff . in s = 0.0 744 (0.0 292) + 0.0 352 (0.00 50) (mean no . {of LD}_{50} values)

{n = 8  r}^{2} = 0 {.893   q}^{2} = 0.816s = 0.0330 F = 50.3

Equations 59 and 60 show clearly that the more rodent LD₅₀ values are available for calculation of median values, the greater the improvement in prediction of HLD values. It follows that studies carried out using only a single set of LD₅₀ values, rather than using average or median values, may be less accurate, although this would be dependent on the quality of the individual data set used. The advantage of the approach demonstrated here is that use of median values improves predictivity without the need for extensive data quality assessment.

Conclusions

Human lethal dosage (HLD) and human forensic lethal concentration (HFLC) values for 36 organic MEIC chemicals have been shown to correlate very well with mouse and rat median intraperitoneal LD₅₀ values, and slightly less well with mouse and rat median oral LD₅₀ values. The correlations of HLD values with subcutaneous and intravenous LD₅₀ values were not as good. The use of median LD₅₀ values gave better results than the use of lowest available LD₅₀ values, with the advantage that no curation of data is required. Therefore, it appears feasible that publicly available median rodent LD₅₀ values could be effectively used to obtain very good predictions of human toxicity, and thus make some reparation for the millions of rodent lives lost in LD₅₀ testing. To this end, it would be admirable if other researchers could also undertake such investigations embracing a wider range of endpoints, in order to maximise — and indeed extend — the utility of the existing data.

Supplemental material

Supplemental Material, sj-pdf-1-atl-10.1177_0261192921994754 - Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD₅₀ Values: An Attempt to Make Some Reparation

Supplemental Material, sj-pdf-1-atl-10.1177_0261192921994754 for Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD₅₀ Values: An Attempt to Make Some Reparation by John C. Dearden and Mark Hewitt in Alternatives to Laboratory Animals

Supplemental material

Supplemental Material, sj-pdf-2-atl-10.1177_0261192921994754 - Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD₅₀ Values: An Attempt to Make Some Reparation

Supplemental Material, sj-pdf-2-atl-10.1177_0261192921994754 for Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD₅₀ Values: An Attempt to Make Some Reparation by John C. Dearden and Mark Hewitt in Alternatives to Laboratory Animals

Footnotes

Acknowledgements

The authors thank Dr Grace Patlewicz and Dr David Ebbrell for help and advice given.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Ethical approval

Ethics approval was not required for this research paper.

Informed consent

Informed consent was not required for this research paper.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Supplemental material

Supplemental material for this article is available online.

References

Gadaleta

Vuković

Toma

, et al. SAR and QSAR modeling of a large collection of LD₅₀ rat acute oral toxicity data. J Cheminform 2019; 11: 58.

Ekwall

Barile

Castano

, et al. MEIC evaluation of acute systemic toxicity. Part VI. The prediction of human toxicity by rodent LD₅₀ values and results from 61 in vitro methods. Altern Lab Anim 1998; 26 (Suppl 2): 617–658.

Clark

Steger-Hartmann

. A big data approach to the concordance of the toxicity of pharmaceuticals in animals and humans. Regul Toxicol Pharmacol 2018; 96: 94–105.

Zbinden

Flury-Roversi

. Significance of the LD₅₀ test for toxicological evaluation of chemical substances. Arch Toxicol 1981; 47: 77–99.

Balls

Horner

. The FRAME interlaboratory programme on in vitro cytotoxicity. Food Chem Toxicol 1985; 23: 209–213.

Clothier

Hulme

Smith

, et al. Comparison of the in vitro cytotoxicities and acute in vivo toxicities of 59 chemicals. Mol Toxicol 1987; 1: 571–577.

Walum

. Acute oral toxicity. Environ Health Perspect 1998; 106 (Suppl 2): 497–503.

Kitagaki

Wakuri

Hirota

, et al. SIRC-CVS cytotoxicity test: an alternative for predicting rodent acute systemic toxicity. J Toxicol Sci 2006; 31: 371–379.

Kinsner-Ovaskainen

Bulgheroni

Hartung

, et al. ECVAM’s ongoing activities in the area of acute oral toxicity. Toxicol in Vitro 2009; 23: 1535–1540.

10.

Prieto

Kinsner-Ovaskainen

Stanzel

, et al. The value of selected in vitro and in silico methods to predict acute oral toxicity in a regulatory context: results from the European Project ACuteTox. Toxicol in Vitro 2013; 27: 1357–1376.

11.

Prieto

Graepel

Gerloff

, et al. Investigating cell type specific mechanisms contributing to acute oral toxicity. ALTEX 2018; 36: 39–64.

12.

Bernson

Bondesson

Ekwall

, et al. A multicentre evaluation study of in vitro cytotoxicity. Altern Lab Anim 1987; 14: 144–146.

13.

Ekwall

Bondesson

Castell

, et al. Cytotoxicity evaluation of the first ten MEIC chemicals: acute lethal toxicity in man predicted by cytotoxicity in five cellular assays and by oral LD₅₀ tests in rodents. Altern Lab Anim 1989; 17: 83–100.

14.

Jover

Ponsoda

Castell

, et al. Evaluation of the cytotoxicity of ten chemicals on human cultured hepatocytes: predictability of human toxicity and comparison with rodent cell culture systems. Toxicology in Vitro 1992; 6: 47–52.

15.

Calleja

Persoone

Geladi

. Human acute toxicity prediction of the first 50 MEIC chemicals by a battery of ecotoxicological tests and physicochemical properties. Food Chem Toxicol 1994; 32: 173–187.

16.

Calleja

Persoone

Geladi

. The predictive potential of a battery of ecotoxicological tests for human acute toxicity, as evaluated with the first 50 MEIC chemicals. Altern Lab Anim 1993; 21: 330–349.

17.

Weiss

Sawyer

. Cytotoxicity of the MEIC test chemicals in primary neurone cultures. Toxicol in Vitro 1993; 7: 653–667.

18.

Ekwall

Clemedson

Crafoord

, et al. MEIC evaluation of acute systemic toxicity. Part V. Rodent and human toxicity data for the 50 reference chemicals. Altern Lab Anim 1998; 26 (Suppl 2): 571–616.

19.

Lessigiarska

Worth

Netzeva

, et al. Quantitative structure–activity–activity and quantitative structure–activity investigations of human and rodent toxicity. Chemosphere 2006; 65: 1878–1887.

20.

Armstrong

Frieauff

, et al. Predictive toxicology: latest scientific developments and their application in safety assessment. In: Chackalamannil

Rotella

Ward

. (eds) Comprehensive medicinal chemistry III. Amsterdam: Elsevier, 2017, vol. 2, pp. 94–115.

21.

Karmaus

Fitzpatrick

Allen

, et al. Variability of LD50 values from rat oral acute toxicity studies: implications for alternative model development. In: NICEATM Abstract, Society of Toxicology 2018 Annual Meeting, San Antonio, TX. 11–15 March 2018.

22.

Dearden

. Prediction of human lethality of psychoactive drugs from rodent LD₅₀ values. Int J Quant Struct-Prop Relat 2019; 4(2): 1–27.

23.

Nendza

Aldenberg

Benfenati

, et al. Data quality assessment for in silico methods: a survey of approaches and needs. In: Cronin

MTD

Madden

(eds) In silico toxicology: principles and applications. Cambridge: RSC Publishing, 2010, pp. 59–117.

24.

Rowe

. Essential statistics for the pharmaceutical sciences, 2nd edn. Chichester: John Wiley, 2016, 409 pp.

25.

Madden

Enoch

Paini

, et al. A review of in silico tools as alternatives to animal testing: principles, resources and applications. Altern Lab Anim 2020; 48: 146–172.

26.

Russell

WMS

Burch

. The principles of humane experimental technique. London: Methuen, 1959, 238 pp.

27.

Bailey

Balls

. Recent efforts to elucidate the scientific validity of animal-based drug tests by the pharmaceutical industry, pro-testing lobby groups, and animal welfare organisations. BMC Med Ethics 2019; 20: 16.

28.

Arena

. Atropine poisoning: a report of two cases from Jimson Weed. Clin Pediatr 1963; 2: 182–184.

29.

National Institutes of Health. Toxnet, https://www.nlm.nih.gov/toxnet/index.html (2021, accessed 22 January 2021).

30.

Hoffmann

Kinsner-Ovaskainen

Prieto

, et al. Acute oral toxicity: variability, reliability, relevance and interspecies comparison of rodent LD₅₀ data from literature surveyed for the ACuteTox project. Regul Toxicol Pharmacol 2010; 58: 395–407.

31.

Zhu

Martin

, et al. Quantitative structure–activity relationship modeling of rat acute toxicity by oral exposure. Chem Res Toxicol 2009; 22: 1913–1921.

32.

Gonella Diaza

Manganelli

Esposito

, et al. Comparison of in silico tools for evaluating rat oral acute toxicity. SAR QSAR Environ Res 2015; 26: 1–27.

33.

Anon. ACD/Labs, https://www.acdlabs.com (undated, accessed 22 January 2021).

34.

United States Environmental Protection Agency. Toxicity estimation software tool (TEST), https://www.epa.gov/chemical-research/toxicity-estimation-software-tool-test (undated, accessed 22 January 2021).

35.

Carai

MAM

Colombo

Gessa

. Protection by the GABAB receptor antagonist SCH50911, of γ-hydroxybutyric acid-induced mortality in mice. Eur J Pharmacol 2004; 503: 77–80.

36.

Anon. Minitab, https://www.minitab.com/en-us/ (undated, accessed 22 January 2021).

37.

Dearden

Cronin

MTD

Kaiser

KLE

. How not to develop a quantitative structure–activity or structure–property relationship (QSAR/QSPR). SAR QSAR Environ Res 2009; 20: 241–266.

38.

Raevsky

Grigor’ev

VYu

Weber

, et al. Classification and quantification of the toxicity of chemicals to Guppy, Fathead Minnow and Rainbow Trout. Part 1. Nonpolar narcosis mode of action. QSAR Comb Sci 2008; 27: 1274–1281.

39.

Lipnick

. Outliers: their origin and use in the classification of molecular mechanisms of toxicity. Sci Total Environ 1991; 109–110: 131–153.

40.

Fry

Garle

Hammond

. Choice of acute toxicity measures for comparison of in vivo/in vitro toxicity. Altern Lab Anim 1988; 18: 175–179.

41.

Freidig

Dekkers

Verwei

, et al. Development of a QSAR for worst case estimates of acute toxicity of chemically reactive compounds. Toxicol Lett 2007; 170: 214–222.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.32 MB

0.04 MB

Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD 50 Values: An Attempt to Make Some Reparation

Abstract

Keywords

Introduction

Methods

Data collection

Statistical analysis

Results and Discussion

Correlation of the HLD and rodent LD50 values

The use of rodent LD50 values to model human lethal blood concentrations

The use of ‘lowest’ rodent LD50 values to model human acute toxicity values

Do more rodent LD50 values yield more accurate median LD50 values?

Conclusions

Supplemental material

Supplemental Material, sj-pdf-1-atl-10.1177_0261192921994754 - Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD50 Values: An Attempt to Make Some Reparation

Supplemental material

Supplemental Material, sj-pdf-2-atl-10.1177_0261192921994754 - Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD50 Values: An Attempt to Make Some Reparation

Footnotes

Acknowledgements

Declaration of conflicting interests

Ethical approval

Informed consent

Funding

Supplemental material

References

Supplementary Material

Correlation of the HLD and rodent LD₅₀ values

The use of rodent LD₅₀ values to model human lethal blood concentrations

The use of ‘lowest’ rodent LD₅₀ values to model human acute toxicity values

Do more rodent LD₅₀ values yield more accurate median LD₅₀ values?

Supplemental Material, sj-pdf-1-atl-10.1177_0261192921994754 - Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD₅₀ Values: An Attempt to Make Some Reparation

Supplemental Material, sj-pdf-2-atl-10.1177_0261192921994754 - Prediction of Human Lethal Doses and Concentrations of MEIC Chemicals from Rodent LD₅₀ Values: An Attempt to Make Some Reparation