Acute Toxicity and Eye Irritancy 1123

If P is expressed by a value of y on the y axis (standard The following steps should be taken for graphic estima-
deviations), then tion of LD50 by probit analysis.
y
y2 1. Convert response probabilities to probit units
exp
1
P= dy by a probit transformation table (see Ref. [45,
2 2
pp.5455]).
2. Convert all doses into log dose units (e.g., log10
x
( x u )2 dose = x). (Steps 1 and 2 may be eliminated if

1
= exp dx
2 2 2 probit-log graphic paper is available.)
3. Using the probit as the abscissa and log10 dose as the
ordinate, plot the response probit units against the
The solution of this equation is x = u + y or log10 dose.
4. Draw a straight line such that the vertical deviations
y=
( x u) = 1 x u (22.10)
of points (the probits) at each x value are as small as
possible. Extreme probits, for example, those out-
side the range of probits 7 and 1, carry little weight
Therefore, the probability when expressed in terms of y (the in the fitting of the probitlog doseresponse line
NED scale) is related linearly to x, the log dose. If x is plotted and thus should be excluded.
against the corresponding y, a straight line with slope=1/ 5. From the regression of the probitlog dose line,
will be obtained. To further facilitate calculation, Bliss [11] extrapolate the log dose corresponding to probit
suggested a slightly different NED unit called the probit, units of 5, which also correspond to the P = 0.5.
such that the new y value is equal to [(x u)/] + 5. This Thus, this extrapolated dose should be the median
procedure eliminates the negative values of NED when P has lethal log dose, and the LD50 value would be the
a value of less than 50%. Therefore, the probit is equal to the antilog of this log dose value.
NED plus 5. The linear relationship between probits and log 6. Calculate the slope of the probitlog dose line. This
dose is similar to the relationship between NED and log dose. slope, = 1/, is defined as the number of increases
Thus, when y = 5, from Equation 22.10 in probit units for a unit increase in log dose. The
slope defined by Litchfield and Wilcoxon [116] is
( x u) + 5 equal to
5=
1 LD84 LD50
+ =
2 LD50 LD16
and x = (i.e., the median log dose that has a probability of
response of 50%).
This slope is different but related to the slope
described here, thus the larger the slope value, the
ESTIMATION OF LD50 BY PROBIT ANALYSIS steeper the probitlog dose response. The opposite
The basic linear equation for the probit analysis as described is true in the Litchfield and Wilcoxon definition.
in the previous section is 7. An 2 test should be conducted to determine if the
fitted line is adequate. A small value of 2 statistic
1 (within the limits of random variation) may indicate
y = 5+

( x u) satisfactory agreement between the theoretically
expected line and the fitted line. A significantly large
where y is the probit ( is the standard deviation of a log- 2 statistic may indicate either that the animals do
normal distribution with mean u, and x is the log dose). not respond independently or that the fitted line (pro-
This equation is linear with respect to y and x often can be bitlog dose) does not adequately describe the dose
expressed as a linear equation, for example, y = + x, where response relationship of the test substance. If the
= 1/ = slope, and = 5 (u/). When y = 5, (x u)/ = 0; latter is true, forms of the doseresponse curve other
thus x = (the median log dose). Furthermore, y is related to than the probitlog dose linearity may exist, and fur-
P (the probability of response that has a value of 01) by the ther transformation may be needed [75]. If the former
following equation: is the case, then precision of the line is reduced.
8. Determination of precision is by weighting the coef-
y ficient. The standard deviation of a binomial dis-
1 y
P=
2 exp 2 dy
x
tribution is PQ / n , where P and Q are the mean
probabilities, P equals (1 Q), and n is the num-
ber of test subjects. Thus the variance is PQ/n, the
The reader should bear in mind that both the u and x are in square of the standard deviation. It is obvious that
log dose scale. the variance (i.e., the spread of a distribution) is
1124
inversely related to n. This relationship means that
the larger the number of test subjects, the smaller
the variance and the better the precision. The recip-
rocal of the variance is invariance, which measures
the weight, nW. Here W (weighting coefficient) = Z2/
PQ, where Z = (1 / 2 ) exp( y 2 / 2) and is related to
the normal frequency function corresponding to the
NED. A table of weighting coefficients (see Ref. [45,
p. 55]) corresponding to probits (y) is available [76].
The standard error for the log LD50 is given by
nW
if the estimated log LD50 does not greatly differ
from the true mean log LD50, because this estima-
tion does not take into consideration the error in the
estimation of for the probitlog doseresponse
line. Abetter equation for the estimation of the vari-
ance of the estimated log LD50 is given by
V ( m ) = 2
1
+
( m x )2
nW
nW ( x x )
2
where
V(m) is the variance of LD50
x is the weighted mean log dose
m is the median log dose
x is the log dose
1 / = 1 / nW ( x x )
If the 2 is large, indicating that the test subjects
do not respond independently to the dose, the esti-
mation of variance of log LD50 may not apply, and
adjustment due to the sampling variation of the
slope (1/) of the probitlog dose line may have to
be made [75]. For a quick estimation of the LD50,
this adjustment may be dropped, and the SE would
be the square root of the variance, that is, V (m).
One must remember that the dose is expressed in
log dose; therefore, the estimation of the SE for the
LD50 in the original dose unit (e.g., mg/kg) is impos-
sible. However, an approximation is given:
SE (LD50) = (10m) ([loge(10) (Sm)]
where Sm (which equals / nW or V ( m ) ) is the
estimated SE for the median log dose m (i.e., m =
log LD50 or 10m = LD50).
A more rapid approximation of the SE of log
LD50 was given by Litchfield and Wilcoxon [116] as
S beyond the scope of this chapter, and interested readers are
Sm =
N /2
Hayes Principles and Methods of Toxicology
where
S is the difference between two log doses of
expected effects (as indicated by the probiting
dose line) that differ by one unit of probit
N is the total number of animals between the log
dose limits, corresponding to the expected pro-
bit 4.06.0 (i.e., the 16% and 80% responses)
9. Fiducial limits. The concept of fiducial limit is sim-
ilar to the confidence limit. The value of the two
may be the same, but they are not always identical.
The fiducial probability F (e.g., 95%) can be defined
as the situation when the true value of a parameter
lies between the calculated upper and lower limits,
which would not be contradicted by a significance
test at the 1/2 (1 F) probability level. These higher
and lower limits are called the fiducial limits. For
rapid analysis, the fiducial limits at the F = 95%
level can be estimated by log LD50 1.96 (Sm).
Amore detailed estimation can be obtained by the
maximum likelihood estimation [75].
Another simple approximation of the fiducial
limits is given by Litchfield and Wilcoxon [116]
as LD50/f LD50 and LD50 f LD50 for the lower and
upper limits, respectively, where LD50 is defined as
(
the LD50 factor equal to ( s ) 2.77 N .)
Here s is the slope, which is defined as
1 LD54 LD50 1
+ = ( 3.55 + 3.55 ) = 3.55
2 LD50 LD16 2
In this example, and N is the total number of ani-
mals used between response probabilities 16% and
84% (i.e., probit 4 and 6, equal to 30 in this example).
Then f LD50 equals 1.896. Therefore, the lower fidu-
cial limit is equal to 8.91/1.896 = 4.70, and the upper
fiducial limit is equal to 8.96 1.896 = 16.90.
Logistic Transformation
Waud [202] suggested a logistic approach to calculate the
LD50. Thus
DE
P=
( D + KE
E
)
where
P is the probability of response
D is the dose
E and K are scale and location parameters, respectively
K corresponds to the LD50
With the procedure of iteration, K and E can be estimated
with a range of confidence. The derivation of this equation is
referred to the original article by Waud [202].
Acute Toxicity and Eye Irritancy
NONLETHAL PARAMETERS
Although the LD50 and the slope of the doseresponse curve
can provide valuable information on the toxicity of a com-
pound, the LD50 is not equivalent to toxicity. Chemicals can
induce damage to the physiological, biochemical, immuno-
logical, neurological, or anatomical systems not character-
ized by the LD50. Depending on the severity and the extent of
the disturbance of the normal biological functions, the animal
may survive the toxic response even though some irreversible
tissue damage may have occurred. Nonlethal, adverse effects
are as undesirable as lethality and certainly should be taken
into consideration during the risk assessment of a chemical.
A major problem in analyzing nonlethal responses is that
in many cases the data are not quantal. For example, der-
mal toxicity ranges from slight to severe. These polychoto-
mous data may be handled by RIDIT analysis, which was
designed to analyze quantal responses with more than two
outcomes [1,18,92].
While toxic effects may contribute to lethality, any attempt
to correlate a particular nonlethal response to mortality may
be irrational [189] unless that response is the only one respon-
sible for the eventual death of the animal. Identification of
the response or responses related to mortality is not often a
straightforward matter. Nonlethal responses that affect the
general well-being of an animal should be considered in the
risk assessment of a compound.
If nonlethal responses can be viewed as true quantal data,
the median effective dose (ED50) and the corresponding
doseresponse curve may apply. The ED50, which often is
used in the standardization of biologically active compounds
such as a drug, has a meaning similar to the LD50 except
that it is designated to examine nonlethal parameters such
as pharmacological responses and other nonlethal adverse
effects. The ED50 is defined as a statistically derived single
dose of a substance that can be expected to cause a partic-
ular effect to occur in 50% of the animal population. The
therapeutic index (TI), defined by the ratio of LD50/ED50
or LD1/ED99, has been applied to establish the safety margin
of some biologically active drugs. The higher the index, the
greater the margin of safety with the drug, that is, a large
difference exists between the amount of compound predicted
to kill 50% of the animals and the amount of compound pre-
dicted to elicit a particular response in 50% of the animals.
The TI gives an even greater estimate of safety when the LD1
is compared to the ED99.
REVERSIBILITY OF NONLETHAL PARAMETERS
In general, reversible responses are those that diminish with
elimination of the chemical from the body. A true revers-
ible response will cause no residual effects when the chemi-
cal is completely eliminated from the body. Such responses
are commonly seen in drugs used at therapeutic dose levels.
Asthe amount of drug in the body increases, the magnitude
of the effect also increases. If it is truly reversible, the effect
will wear off when the drug is completely eliminated.
1125
The reversibility of a particular response is dependent on
the organ or system involved, intrinsic toxicity of the chemi-
cal, length of exposure, total amount of the chemical in the
body at a specific time, and the age and general health of
the animal. If the amount of chemical in the body is high
enough, the intensity of the response may overwhelm a par-
ticular organ. Effects indicated through hormonal imbal-
ance such as thyroid effects generally are reversible unless
the threshold is surpassed. Damage in rapidly regenerating
organs, such as liver, is usually more likely to be reversible
than damage in nonregenerating tissues, such as nerves.
Agood example is the delayed onset of neuropathy caused
by many organophosphate insecticides. The chemical may be
completely eliminated from the body before the effect man-
ifests itself. Animals with renal or liver diseases are often
more susceptible to damage (reversible or irreversible) by a
chemical insult because of decreased ability to eliminate the
chemical. Exposure to a chemical at an early age may induce
irreversible damage more easily than at an older age because
of the limited development of the kidneys and/or functional
capacity of other organs, such as the liver.
In risk assessment, it is important to know whether a toxic
effect is reversible. Irreversible effects seen in animals obvi-
ously are weighted more heavily in reaching a conclusion on
the toxicity and hazard a chemical may pose for humans.
ACUTE TOXICITY TESTING
The objectives of acute toxicity testing are to define the
intrinsic toxicity of the chemical, predict hazard to nontarget
species or toxicity to target species, determine the most sus-
ceptible species, identify target organs, provide information
for risk assessment of acute exposure to the chemical, pro-
vide information for the design and selection of dose levels
for prolonged studies, and, the most important and practical
of all, provide valuable information for clinicians in the pre-
diction, diagnosis, and treatment for acute overexposure (poi-
soning) to chemicals. Acute studies often are called the first
line of defense in the absence of data from long-term stud-
ies. These data help industrial, governmental, and academic
institutions formulate safety measures for their researchers
and for limited segments of their worker population dur-
ing the early stage of the development of a chemical. From
a regulatory standpoint, acute toxicity data are essential in
the classification, labeling, and transportation of a chemical.
From an academic standpoint, a carefully designed acute tox-
icity study can often provide important clues on the mecha-
nism of toxicity and the structureactivity relationship for a
particular class of chemicals.
Many acute toxicity studies have been conducted solely for
the purpose of determining the LD50 of a chemical. However,
the reader is reminded that acute toxicity is not equivalent
to the LD50, and that the LD50 is not an absolute biological
constant to be equated, as some investigators have, with such
chemical constants as pH, pKa, melting point, and solubility.
The LD50 is only one of many indices used in defining acute
toxicity. A well-designed acute toxicity study should include