P 0 ¼130 s andtlag¼ 0 h. The computer fitting
gave 0.2629.46 for the IC 50 , 0.03317.9 for
kd,2.6839.6 fornand 12158 forP 0 (limits
are CV%) with no lag time. Precision increased
when a finite lag time was included in the fitting.
As stated earlier, these are two of the many exam-
ples that can be chosen to illustrate principles. These
two cases, however, are especially relevant to the
relationship between animal work and phase I studies
in which only the simplest effects, such as counter-
action of a painful stimulus or raising/lowering of a
physiological parameter such as PCA, are likely to be
commonly measured. The reader is again referred to
standard texts for more thorough treatment of models
of this kind (Sharma and Jusko, 1997).
8.4 Commentary
We have not sought in this chapter to describe phase
I studies as such. This is a postgraduate textbook,
and we wish to convey howin vitroandin vivodata
of various kinds may be used to help extrapolate
observed drug effects from simple experimental
systems to the more complex clinical situation.
The ultimate need is to obtain useful predictions
of response in healthy human subjects (phase I
studies) from observed drug effects in animals or
in the test tube.
What are the strengths and weaknesses of these
approaches? The use of intrinsic clearancein vitro
permits predictions between species for the parti-
cular enzyme/route of metabolism concerned. If
humans have qualitatively different routes of
metabolism for any particular compound, then
this will weaken the predictive value of thein
vitro observation. Similarly, allometric scaling
works best for compounds with a high component
of nonenzymatic elimination, such as our model
compound with approximately 90% excretion as
unchanged drug. This prediction weakens as var-
iations in rates of enzymatic reactions become
more important. The PK–PD modeling appro-
aches use the existingin vivodata to calculate
constants which can be applied to otherin vivo
data but does not, in its present form, linkin vitro
andin vivodata.
Significantly, none of these approaches uses
drug-receptor binding data. AlthoughKdvalues
are generated during initial screening of the scores
of compounds emerging from medicinal chemistry
laboratories, it has been a traditional problem that
relative efficacy remains unknown (this does not
detract from their value in chemical, structure–
activity analyses). Neither does any of these
approaches uses results ofin vitrofunctional assays
which emerge from screening of the compounds in
biochemistry laboratories. It should be added that
there are exceptions, however: drug–receptor
binding constants and EC 50 values fromin vivo
studies in animals were used by Danhof and
Mandema (1995) to model drugs effects at benzo-
diazepine receptors and effects on EEG (Figure
8.7). Rowleyet al. (1997) have taken a similar
approach with NMDA antagonists.
Prospectus
In the future, models will exist which will link
constants forin vitrobinding to cloned human
receptors (Kd), data fromin vitrofunctional assays
(IC 50 ) and animal and humanin vivoEC 50 values.
A composite prediction matrix will be applied
rapidly and accurately to the process of synthesis
of new compounds for phase I testing.
In the shorter term, what can we now do to
expedite the drug selection process? Figure 8.8
represents a flow chart illustrating one form of
metabolism/pharmacokinetics input into the drug
discovery process. Arrows (indicating the flow of
work and communication) pointing to the right
represent perceived progress, whereas arrows point-
ingtotheleftrepresent‘disappointments’(andother
feedback) leading to corrections and revisions. The
numbered asterisks indicate continuations. The
‘flow of time’ is from left to right and from the top
panel to the bottom panel. The rectangles indicate
tasks that are to be completed, and rectangles in a
column within a panel represent work done by
different departments which may be simultaneous
or not simultaneous but does not require much
interaction between the investigators involved.
Unlike the flow chart of a computer program, after
which the diagramis modeled, most of the decisions
8.4 COMMENTARY 95