An Approachto Immunoassay
William H. C. Walker
Foreword
These notes were prepared for a series of lectures in
our continuing-education program. They are intended
to enable an analyst given labeled ligand, standard, and
antibody to set up an optimized assay, and to define
those steps in the assay that require meticulous control
of reaction conditions in order to achieve reproducible
and valid results.
Much of the text concerns practical details and these
have all been described in the extensive scientific lit-
erature, some so often that original attribution would
be difficult. Moreover, the notes are didactic, painted
in black and white where in fact innumerable shades of
gray exist. Much of the emphasis derives from my own
practical experience, rather than published work, which
tends to emphasize the end rather than the means. For
these reasons, I have omitted references and instead
provided a further reading list that will help to round
out the broader perspectives.
Immunoassay has more experimental variables than
most other analytical techniques; it is clear that there
is no one best way to set up an assay. These notes are
intended for those analysts who have limited experience
in immunoassay, in the hope that they will increase
understanding of what is happening in the assay tubes
and which variables have most effect on the outcome.
Despite the great daily contributions of immunoassay
to diagnosis and management, we still have a long way
to go when an international quality-control survey can
report an erroneous prolactin result of 0.25 to be within
two standard deviations of the group mean when the
result should have been 250.
Department of Pathology, McMaster University Medical Centre,
1200 Main Street West, Hamilton, Ontario, L8S 4J9, Canada.
384 CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977
Radioactivity-counting
A radionuclide is an atom that spontaneously decays,
with release from the nucleus of radioactivity, which
may be a, j3, or radiations or X-rays derived from en-
ergy transfers in the orbital electrons. Alpha rays are not
encountered in immunoassay work. Beta rays are elec-
trons of different energies characteristic of each isotope;
they have poor penetrating power. Gamma rays are
electromagnetic radiations similar to X-rays. Like X-
rays, they have variable penetrating power, depending
on their energy but will travel several centimetres
through low-density material.
Tritium and carbon-14 are pure beta emitters. The
beta rays from a solution will not penetrate the few
millimetres of the wall of a containing glass tube and
they have to be counted by their ability to release pho-
tons from organic phosphors occurring in the same so-
lution. This is the principle of liquid scintillation
counting. The efficiency of the phosphor is impaired by
the presence of water, and scintillation fluids are always
nonaqueous based, with a wetting agent added to enable
limited amounts of water to be dissolved. Quenching
caused by water and pigments (as in jaundiced serum,
for example) may require correction to each tube.
Gamma ray or X-ray emitters, such as iodine-125, are
counted after the radiation has passed through the wall
of the containing tube. A solid sodium iodide crystal
apposed to the tube contains fluor, which releases one
photon for one quantum of gamma radiation. Iodine-
125 has gamma radiation of relatively weak energy.
There will be a different counting rate if a solution is
counted in a glass tube or a plastic tube.
The unit or radioactivity is the curie (Ci). 1 curie =
3.7 x iO’#{176}
disintegrations per second. Note that disin-
tegration rate is inversely related to half life. One mole
of ligand containing 1 atom of 1251 per molecule will
disintegrate faster than 1 r#{241}ol
of ligand containing 1
atom of 3H per molecule. The maximum theoretical
specific activity for mono-labeled compound is:
= 2.92 X 10 Ci/mol
= 6.24 X 10 Ci/mol
125J = 2.17 X 10 Ci/mol
Highest counts will be obtained if 125! is used as label,
but of course the material will retain a useful degree of
radioactivity for only a few months rather than for
hundreds of years.
Statistical Inference
Every step in an immunoassay is imprecise, and the
sum of all random errors leads to an overall uncertainty
in the derived assay value, which is seldom better than
±5% within-batch and ±10-15% between-batch for 1
coefficient of variation.
Errors cost money and the objective of good assay
design is to minimize error, both systematic (bias, ac-
curacy) and random (imprecision). It is possible to
consider the individual identifiable components of
random error separately and thereby identify the
dominant sources of error so that they can be given
special remedial attention.
The total random error is definable in two ways. Ei-
ther from basic logical precepts (called a priori) or from
experiment (a posteriori). The probability of an unbi-
ased coin yielding “heads” tends to 0.5 as n tends to
infinity. This is a priori reasoning from the laws of
chance. We may come to the same conclusion by spin-
ning a coin many times-this is an a posteriori defini-
tion. We shall see examples of both approaches when
we consider overall assay error. A priori reasoning fol-
lows established laws of probability, but is only as good
as the premises it starts from. Is the coin unbiased; is the
counter free of power-line artifact? A posteriori infer-
ence tests the validity of the premises. Be alert to these
types of statistical inference and be wary of accepting
formidable mathematics in place of real data. There is
no better way to find out how reliable an assay is than
to run the same samples over again.
Counting Imprecision
Counting error contributes to overall random error,
usually forming only a minor part, but it can be the
dominant part.
A priori, because radionuclide disintegrations are
random and uninfluenced by past events, a Poisson
distribution is expected. Poisson distributions are very
common: the daily snowfall, waiting time for a bus, in-
teraction in an air-liquid segmented stream-these are
all Poisson distributions. Poisson distributions have the
characteristic that their standard deviation is the square
root of the mean. An increase in counts (x) of 100-fold
results in an increase in standard deviation (SD) of
10-fold. The coefficient of variation in percent is (SD/x)
X 100, which decreases because the standard deviation
has increased less than the mean.
Counts (x) SD CV,%
100 10 10
10 000 100 1
Beware of confusing standard deviation and coeffi-
cient of variation; with increasing counts, the first in-
creases, the second decreases. Confusion is easy and may
lead to such statements as “the more counts, the better
the precision,” which is the opposite of the truth in some
common situations in immunoassay.
Consider a tube of ligand 1%bound to antibody with
total counts 10 100 cpm. After separation, the bound
fraction and the free fraction are counted for 1 mm
each.
Cpm SD
Total 10000 ±100
Bound 100 ± 10
Free 10000 ±100
The result may be shown diagrammatically:
0#{149}none
freeT all bound#{149}I0
100
free 10000 ±100 - - bound 100 ± 10
all free
10 100 #{149} ‘- none bound #{149}
0
The free count, although much larger than the bound,
leads to such uncertainty that the amount that is free
cannot be distinguished from 100%, although the tiny
bound count clearly indicates a binding different from
zero. If the counting time were increased to 100 mm, the
free count would now have the same precision as the
bound count for 1 mm. Note that the uncertainty of
location on the no-free-all-free continuum shown by the
free count will not be altered by changing the scale to
free/bound (F/B), B/F, or any other transform.
Of course, the amount bound could be fixed still more
precisely by counting the bound fraction for 100 mm,
although the improvement would be illusory because
the minute error would then be swamped by other ex-
perimental errors. The general rule is established: al-
ways count the smaller of the two fractions. Because
the bound fraction is seldom greater than 0.5, this
means count the bound fraction only. Counting time is
money; counting both fractions is less profitable than
counting bound only for twice as long. Counting a total
count for each tube is unhelpful unless it is known that
each tube differs as, for instance, when total counts are
used as a measure of extraction recovery. Otherwise, one
can always pipette total labeled ligand with a precision
of ± 1%, and counting precision will seldom improve on
this. Counting total counts in each tube is not justifiable
as a means of picking up pipetting blunders. if blunders
in pipetting labeled ligand are likely, the same blunders
will occur in pipetting standard, antibody, and unknown
samples. The solution here is to change pipettes or op-
CLINICAL CHEMISTRY, Vol. 23. No. 2, 1977 385
erator. A good quality-control scheme is the proper way
to monitor quality; redundant counting is no help.
Counting Statistics
With total counts/mm = L
bound counts/mm = B L mol/litre
free counts/mm = F Fig. 2.
counting time per tube = t mm
y = BC/LC
the standard deviation of y [S(y)] caused by counting Ab
error is given below:
Fractions ]Counts
0.5Ab
counted S(y)
I B,L,, Fy(y#{247}
i)]’/ - (Lt)-”2
II [(1 - y)(2 - y)]’/2 . (Lt)”2 0]
III B,F,, [y(1 - y)]112. (Lt)’2
IV B only [y(O.0001 Lty + 0.5)11/2. (Lt)-”2 Fig. 3.
In IV, L is counted only once, but is added to all tubes
with pipetting precision 1% (1 CV). B only is counted,
but for twice the time allowed in the other protocols.

Interpretation of Binding Data

0.54
In a system with a binding protein (Ab) having a
single species of binding sites with equilibrium constant
K, the law of mass action gives:
Fig. 4.
Unoccupied binding sites or “free antibody” (Ab B) -

+ free ligand (F) bound ligand or “bound antibody”

(B)
binding capacity. Regardless of how much more L is
added, B can never exceed this total binding capaci-
F(Ab - B)/B = K-’ (1) ty.
if F, Ab, andB are all in mol/litre, thenKwill have units Can we use this curve to derive Ab and K? We must
of litres/mol, and K-’ will be in mol/litre. Strictly, K is remember that B is seldom available in mol/litre units.
dimensionless, because the quantity logK is used in It is usually available in the form of a decreasing
thermodynamics and the logarithm of a dimensional counting rate that represents the amount of isotope that
number, such as 10 apples, does not exist. LogiolO = 1; has been bound, when studies are done with increasing
logiolO apples = nothing. We can convert to dimen- amounts of L that contain a small fixed proportion of
sionless K values later when we consider thermody- isotopically labeled L. Run an assay with increasing
namics by the simple device of expressing F, Ab, and B labeled ligand only. The experimental curve will then
as dimensionless moles of solute per mole of water. look like Figure 2 and even Ab is no longer easily avail-
Obviously, a solution of binding protein, with binding able. We will use two relationships derived from equa-
capacity Ab moles of ligand per litre, cannot bind more tion 1
than Ab moles of ligand. Also the bound component B (a) when B is small, the slope
cannot exceed the total ligand concentration, L. The dB Ab
relationship between B and L will therefore look like
dL K-’+Ab
Figure 1. The curve starts off almost linear with a slope
somewhat less than the 45#{176} angle that represents B = (if you want to derive this, differentiate with respect
L. It flattens at a B value that represents Ab, the total to L,
K-1B
L = Ab B + B
-

Total saturation of binding sites which comes directly from equation 1, putting L = F
+B)
(b) when B = 0.5 Ab, F =
Now consider Figure 3. We can add an ordinate
scale-0, 0.5 Ab, Ab-where Ab is set at the counts
plateau, even though as yet we don’t know what Ab
is.
L
Next draw a tangent (OT) to the curve at B 0-

FIg.1. (Figure 4) and read off L1 where the tangent crosses the

386 CLINICAL CHEMISTRY, Vol. 23, No. 2. 1977

plateau. Knowing the slope and the height of the
triangle, it follows that
L, = K-1 + Ab
Y1!b
Now read off L2 at 0.5 Ab.
L + La

L2=F+B=K’+0.5Ab La

So Fig. 5.

2L2 = 2K’ + Ab
and
K = 2L2-L,andAb = 2L, -2L2
This simple approach yields an estimate for K’ and Ab
by use of only limited parts of the curve: the initial slope, y
the mid-binding point, and the plateau.
You will see later that no assay can be optimized
without knowing Ab and K’, and the above procedure Fig.6.
is an absolute minimum. This type of curve in which
increasing bound counts occur with increasing ligand
concentration is also seen in one form of the immu- immunoassay, plotting y’ or 1/counts bound (i.e., re-
nometric radioassay system. ciprocal counts bound) vs. ligand concentration. A
straight line should result, and many assays depart little
Radloimmunoassay from linearity when such a plot is used. This approach,
In radioimmunoassays, increasing concentrations of now obsolete, is no longer used by serious workers,
the analyte ligand, L0, are reacted with antibody in the mainly because of the enormous uncertainty of points
presence of a fixed concentration of radiolabeled ligand, far from the origin. Figure 6 shows such a plot with ±2
L Some assays operate for most of their range at sat- SD included. Notice that although the higher points
uration of their antibody sites and it follows that the might suggest a lower slope to the line, their uncertainty
more La, the greater is the dilution of L When a fixed ‘.
isso great that they would divert proper attention from
amount (Ab) of this mixture is bound, there will be the lower points, which are located with great preci-
progressively fewer counts in the bound moiety as La sion.
increases. This phenomenon of large changes of precision along
In such a system with total ligand La + L* and a fixed either axis is called a non-uniformity of variance or
amount bound B = Ab, the proportion of the total li- heteroscedasticity. We shall return to this frequently
gand that is bound in considering plotting procedures. Notice that this
defect of 1/y plots applies equally to 1/counts bound,
B B - Ab to total counts/counts bound, or for that matter to
L - La + L* - La + L* free/bound counts.
If labeled and unlabeled ligand react on equal terms Never use a reciprocal scale as a standard curve. The
with the antibody, changes in variance mislead the eye (or a least-squares
regression, which usually assumes uniform variance).
counts bound = B* = Ab The fact that K is never infinitely large results in cur-
total counts L* La + L* vature at the lower end of the line, which is the only
Setting counts bound/total counts = y region of good precision, and this is another good reason
for avoiding a linearizing procedure that should not be
Ab expected to give a truly linear result.
y - La + L*
Scatchard Plot
La = (Ably) - L*
With this insight into why increasing dose of L in a
This is the relationship between added ligand and radioimmunoassay results in decreasing counts of
counts bound for antibody with infinitely high K if L * bound ligand and a decrease in the response y = bound
> Ab. It explains why counts bound decline as La in- counts/total counts, we can consider a famous formal
creases and why the curve has a hyperbolic shape, be- relationship that enables us to derive Ab and K.
cause the equation has the general form y = (aix) + From equation 1,
b.
Graphically the relationship is shown in Figure 5.
Note that the La scale can be obtained by offsetting the
origin a distance equal to L*. Ab-B
An equation of this sort was used in the early days of

CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977 387

 B=Ab- (.K-1) (2)

This is the Scatchard equation, and is of fundamental y

1-y Ab
importance. Assuming no discrimination of binding
between L * and La,
B B* Fig. 7.
= = y; and B = Ly
but
BFLF
B+F=L; =1-y

So,
y
y B I-y

1-yF
The Scatehard equation may be rewritten as Ly

LyAb- y Fig. 8.
1 -y
Here, La is known, L* may be calculated and y is
known. This is a linear equation with unknowns Ab and Ab
K’. Plottingy/(1 y) againstLy (Figure 7), a straight
-

line results, with abscissa intercept Ab and slope -K.

notice that by comparison with the three-point ap- y
proach we started with, we can now use all points on any
radioimmunoassay curve to define Ab and K. It is true
that the points have slight variation so that their sta- L
tistical weights are unequal, but this is not a serious Fig. 9.
factor except at the highest La values. The area of un-
certainty is shown in Figure 8. 0
There are pitfalls in applying data to a Scatchard plot,
and these will be considered in detail later. $

Standard Curve I-y

From

LyAb- ‘
1 -y K’
Fig. 10.
it follows that

(3)
y 1-y
This is a general relationship between L and y, and
when Ab, K, and L* are known, Ab

La K-’
y 1-y
Notice how this compares with the reciprocal relation-
ship obtained earlier by considering K infinitely large L
Fig. lOa.
La
y
The K’ term disappears, because with K = , K’ =
0. We have now introduced a K term that gives a formal
relationship over the whole range of La and applies even
when L * is less than Ab. Graphically, the reciprocal
relationship was as in Figure 9. The relationship L = y
K’/l - y has a similar reciprocal form, with 1 y re- -

placing y and K1 replacing Ab as in Figure 10. Adding i0

Figures 9 and 10 (Figure lOa) yields a sigmoid curve L
(Figure 11), and this curve can be generalized as Fig. 11.

388 CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977

 L Ab/K 1

K-’ y 1-y
to give a family of standard curves of differing Ab/K
ratios (Figure 12). All immunoassay curves for which
the antibody has a single species of binding sites and
which do not discriminate between L* and La will fall
Fig. 12.
on one of these curves. We shall consider the practical
consequences of this relationship for assay sensitivity
and optimization later.

Roots of a Quadratic
Let us return to
K1
L=- K
y 1-y
Sb +
This is a disguised quadratic: y
1-y
...
Ly-Ly2 Ab-Aby-K’y -I-

Ly2-(Ab+K’+L)y+AbO Ab
NN\
In a quadratic ax2 + bx + c = 0; the roots x andx2 are
related such that x x 2 = c/a
Fig. 13.
Hence yly2 = Ab/L; but y, = BIL
so
AbL Ab
Y2

The Scatchard plot

w
Ly=Ab y K-1 1-w
1 -y
is applicable to both roots
B*
t
= + I
I-w

I-i
Ab* W2
8-Sb
y2 =

Fig. 14.
The plot is shown in Figure 13. Note that for y,, the
ordinate has limits 0 and AbIK’, the abscissa has limits
0 and +Ab. All data points exist between these finite
limits. Fory2 the limits start at (Ab + K’) and -1, but
extend to infinity. These open-ended limits lead to in-
appropriate emphasis of high L points that have poor
Again, this quadratic has another root, put
precision. Compare this with y,, where all high L values
are confined to an area near the abscissa intercept and B-Ab Ab
W2 B
their poor precision is not exaggerated.
The only other linear relationship for
and:
(Ab_B)=K1
Lw2 = w2 Ab -

1- W2
is given by putting
Graphically (Figure 14) the limits for w, are 0 and
F B K’/Ab for low L and for high L.
w1 = 1 - w, =
For w2,the limits are -(Ab + K’) and -1 for lowL
and -K’ and 0 for high L. This is the opposite of the
Then:
Scatchard plot. The root w2 is now well behaved for
W1 (Ab-L+Lw,)=K-’ plotting data in terms of B and Ab. The ordinate limits
1 - Wi can be converted from -1 to 0 to 0 to 1 by using
w2 B
Lw, = Wi Ab-K’ (1 + 1 - w2) = =
1 - Wi

CLINICALCHEMISTRY,Vol. 23, No. 2, 1977 389

 We can now return to the original relationship be-
tween bound counts and ligand (Figure 15) in which the
plateau is represented on the ordinate by Ab, expressed
in counts This was originally solved for Ab and K by
-(K +Abi
the three-point procedure. Now let us use the inverse
Scatchard, which enables us to use all 10 points. w2 is
obtained as (B Ab)/B
- = (bound counts plateau -

counts)/bound counts. Now plot the data points on the

Fig.15.
coordinates of Figure 14, using Lw2 on the abscissa. The
ordinate has limits of 0 and 1 if Y21 = Ab/B is used;
alternatively, bound counts may be used unchanged for The two-site immunoradiometric assay first immo-
the ordinate scale with limits 0 and Ab. bilizes the target ligand and then reacts this with labeled
immunoradiometric Assay antibody. The second reaction comes to equilibrium as
if the ligand were in free solution, and decanting the free
In immunoassay, there are three reactive components antibody does not disturb that equilibrium. The initial
that affect the equilibrium position: immobilization of target ligand is achieved by allowing
F + (Ab-B) it to bind to solid-phase unlabeled antibody in great
free ligand free antibody excess.
B
Step I S:Ab + La -* S:Ab:L,
bound ligand-antibody
Step 2 ISAb:I.a
The standard curve defines antibody binding capacity
(Ab) and equilibrium constant (K). An unknown li-
gand concentration (La) may be calculated from
_Ab K’ S:Ab:F + (AbB)* S:Ab:B*
L0
y l-y
where the two roots of the equation are given by Because each ligand molecule is required to react with
antibody twice in this system, first with solid-phase
B Ab
unlabeled antibody and then with labeled antibody in
Yi = Y2 = solution, it is required that the ligand shall have at least
B/L is defmed in radioimmunoassay by introducing a two antigenic sites available. These antigenic sites need
small fixed amount of labeled ligand, L*, and measuring not be identical; it is therefore possible to selectively
B*/L*. assay only antigens that have both sites X and Y on the
In immunoradiometric assay, Ab/B is defined by la- same molecule if the first-stage antibody is directed
beling partly purified antibody with I and measuring against X and the second-stage antibody against Y.
Ab*/B*.
Immunoradiometric assays have sensitivities similar
to those of radioimmunoassays. There are some differ-
In either case, separation of B from F or from (Ab-B)
is required in order to defme they value. ences that may influence the choice of technique.
In immunoradiometric assay, separation of labeled In favor of immunoradiometric assay:
bound antibody from free is always by solid phase. The (a) Simple.
one-site assay takes (AbB)* onto solid phase; the (b) Suited to ligands that cannot easily be labeled or
two-site assay takes B* onto solid phase. that have poor stability in labeled form.
The one-site immunoradiometric method uses a solid (c) Less prone to matrix effects because of washing
phase-ligand to remove free antibody from the equi- after uptake of target ligand in the two-site assay.
librium mixture: (d) The two-site assay can be directed to two dif-
ferent antigenic determinants, thereby measuring only
those ligand molecules that possess both determinants.
Step 1 F + (Ab-B) B
Ligands with one but not both determinants will not be
free
measured.
antibody
Against immunoradiometric assay:
Step 2 S:L + (Ab-B) - S:B (a) Consumes 50-fold or more antibody per assay
excess solid- than would be required for the corresponding ra-
phase ilgand dioimmunoassay.
(b) Washing of immobilized phase disrupts equi-
Removal of free antibody from the initial equilibrium librium and so has to be reproducible.
mixture will disriapt the equilibrium position unless the (c) Multiple steps are tedious for large numbers of
antibody is effectively irreversible. The one-site assay samples as compared with radioimmunoassay; some
is therefore potentially sensitive to timing and washing assays call for six washes, which must be timed and re-
routines. producible.

390 clJ’IICAL ClEMSTRY, Vol. 23, No.2, 1977

 (d) Potential for its own matrix problems, e.g.,
rheumatoid factor.
y
Measurement of Labeled Ligand
Mass and Binding Energy
A known mass of ligand will be taken for iodination,
La
but the subsequent isolation of labeled ligand from the
Fig. 16.
reaction mixture will result in variable loss of ligand.
Further, some labeled ligand may no longer be reactive
with antibody because of damage to its antigenic site
during the relatively harsh iodination process. It is
therefore not possible to calculate from manufacturers’ L*

“specific activity,” the mass of labeled ligand added to

each tube. Fortunately, it is very easy to measure the (La)n
mass of labeled ligand in terms of the unlabeled stan- n

dard used for the calibration curve. Consider Figure 16.

A relationship exists between y and total ligand, La +
L*. Since L* is needed to measurey, the curve starts at
the beginning of the solid line where La = 0. Data cor- 0 y
responding to the broken line cannot be experimentally Fig. 17.
obtained.
Suppose twice the labeled ligand dose were used. The
curve would now start at a point displaced to the right converted to L * equivalents when calculating bound
to the extent of one extra dose of L * But that point is values for a Scatchard plot, because the bound/free
already defined in terms of La. Hence, L* can be read values are expressed as y/(1 y), which relates exclu-
-

from the L0 standard curve. sively to labeled bound and labeled free. With y/(1 y) -

A good practice is to set up tubes with zero La and
increasing amounts of L*: two L*, four L*, six L*, eight
L These represent increments of one, three, five, and
seven L * on the regular standard curve set up with one
dose of L*. The increments serve to allow measure-
ments on the most precise part of the curve (coefficient
of variation increases at very low and very high La) and
indicates whether antibody is discriminating between
labeled and unlabeled ligand. The ratio ra = K*/Ka can
be defined from these data. If a dose of labeled ligand
(L* + nL*) (L* is the regular amount used in preparing
the standard curve; n = 1, 3, 5, 7) produces a response
(L0)0, it can be shown that:
nL* = (La)n/(yn + ra - rayn)
or
(La)n/n = L*r0 ±yn[L*(1 - r0)]
With (La)n, n, and y known, r0 and L* are derived if
two values of n are available. It is wise to have four,
spaced as widely as possible over they scale, to improve
precision.
Graphically, with (La )0/n as ordinate and y0 as ab-
scissa, the linear relationship has (La )/n = L * when y
= 1; and (La)n/n = L*ra wheny = 0 (Figure 17). Defi-
nitions of L * and r0 are essential preliminaries to deri-
vation of Ab and K, by use of a Scatchard plot or any
other graphical or computational approach. The mea-
surement of L * and r0 is simple; it is an integral part of
assay development. This is the method of choice for
measuring L * in the assay system. It is better than using
stated specific activity because losses may have occurred
in the iodination and purification steps. If r0 is different
from unity, the La values of the standard curve must be
as ordinate, the appropriate abscissa is y(L* + La/1a);
[ya = y + ra ray]. In this way, nonlinearity
- of the
Scatchard plot from differing K* and K0 is avoided and
values for Ab and K* are obtained free of artifact.
Derivation of Binding Constants
Estimates of binding capacity, Ab, and &juilibrium
constant, K, are obtained from a Scatchard plot. This
plot is meaningless unless certain conditions are met:
(a) All concentration units are expressed in terms of
concentration in the final equilibrium mixture before
the separation reagent is added.
(b) Bound/total (y) is expressed as corrected bound
fraction, using y = (y’ f3)/(a
- 3). a and
- are upper
and lower limits of y’; y’ = apparent bound counts/total
counts.
(c) Bound concentration is y(L0 + L*). L* must be
included. Otherwise the zero standard has zero “bound,”
which is nonsense.
(d) The reaction must proceed to equilibrium with
labeled and unlabeled ligand mixed together before
antibody is added.
If a linear Scatchard plot results, the values Ab and
K must be validated by demonstrating that a standard
curve with twice the antibody concentration yields a
binding capacity of two Ab with K unchanged.
A nonlinear plot indicates either severe disruption of
the equilibrium by the separation system or a combi-
nation of several species of binding sites. If more than
one species of binding sites is present, the assay is more
likely to have cross-reactivity problems.
Statistical Weighting
Consider the following data:

CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977 391

 ligand pg (X) counts/mm (Y) the plot of your choice, and you can get some feel for the
I 100 50 effective width of your standard curve. When you have
II 400 420
observed duplicates spaced widely, you may find an
III 900 870
Iv 1600 1680 overwhelming need to purchase an essential piece of
immunoassay equipment: a broad felt-tip pen to draw
It is clear that the counts are roughly the same as the calibration curves that begin to represent the degree of
mass of ligand, all being within 5% except the lowest. uncertainty inherent in the data. The fine line is imag-
If the least-squares fit of Y on X is calculated and the inary and the second decimal place answer that comes
exercise is repeated for 1/Y on 1/X, the two regression from it is equally fictional.
equations can be used to calculate the counts to be ex- No immunoassay will perform better than ±5%, one
pected from 1500 pg of ligand. CV within batch, and some parts of every curve will be
Why are the answers different? Why is one obviously much worse. Be aware and be critical of your data.
misleading? Because of varying statistical weighting on
the reciprocal plot, which attaches great weight to the
deviation of 50 counts from the value expected for I but Plotting Routines
is little affected by a difference of 80 in the vicinity of
Some popular nearly-linear plots are shown below:
point IV.
From
The linear plot of Y on X, on the other hand is equally
affected by a difference of 50, whether it occurs at the L =Ab/y-K’/(l-y)
lowest or the highest level.
assume the binding is very avid, K is very large and K’
For counts directly proportional to mass, the graph
can be ignored, then
should have a slope of 1 regardless of whether Y and X
are both linear, or both reciprocal. You can compare the La+L* Ab/y
slope of the least-squares fit of the present data to linear
and
axes: almost exactly 1; to the slope with use of reciprocal
axes: more than 2. This again shows the reciprocal plot L* = Ab/yo
to be dominated by point I, owing to the great weight
where yo = B*/L* when La = 0
placed on its deviation.
A plot in which the dependent variable has variance
that changes with the size of the variable is called het-
La = Ab (-
eroscedastic. Unweighted least-squares fits to such data
may seriously mislead, as we have just seen. La = i:’ -
Unfortunately, reciprocal plots for radioimmunoassay yo yy
are popular because they tend to convert a curve into This is a linear equation in La and y/y or Bo*/B*. A
almost a straight line. On such a plot, dominance by the similar plot results with ordinate 1/B* or 1/y by
points furtherest from the origin is universal, with re- changing the scale and with F*/B* (as used by Ekins)
sulting serious errors in derived results near the origin. since F*/B* is equal to (L*/B* - 1) = (1/y 1). -

Least-squares fits can be weighted, but this is generally The three reciprocal plots Bo*/B*, 1/B*, and F*/B*
a job for a computer. Without such a facility, never use are similar except for scalar change or a shift of the or-
reciprocal plots for radioimmunoassay. Remember: igin. All suffer from a severe increase in statistical un-
reciprocal plots come in many forms: “bound counts at certainty of data points as y decreases. Since K1 can
zero ligand/bound counts”; “1/bound counts”; “free/ rarely be ignored, the line deviates from linearity at low
bound” (= “1 - total/bound”). They are all equally La values and a least-squares fit is inappropriate on
unsuited to least-squares fit, and of course a “best fit by grounds both of nonlinearity and markedly inconstant
eye” is equally bad. variance of data points.
The logit plot (log [Y/(1 -Y)J where Y = y/yo) is The expression
similar to a reciprocal plot in that great stretching of
parts of the scale result in a linear response. Here, both (4)
ends of the scale are stretched, leaving a compressed yo y
area at mid-range when y = 0.5 yo. Rodbard, who in- is the basis of another common plotting routine.
vented this plot, stresses that it must be used only with
y - y/yo
computer routines that include weighting. Nevertheless,
it has great popularity as a by-eye or simple least-
yo - y - 1 - y/y0
squares means of getting a straight-line response curve. and loge y/(yo - y) = logit y/y
Follow the advice of the inventor and never use the logit Coverting equation 4 to logarithmic form:
plot unless you are going to use computer weighting for
loge L = constant - logit (y/yo)
the least-squares fit.
Always plot duplicate data points separately rather Any immunoassay that will yield a linear reciprocal plot
than first calculating a mean. You do, at least, stand a must also have a linear logit plot. With this plot the
chance of observing how far apart the duplicates are in change in variance of the data points as y changes is

392 CLINICAL CHEMISTRY, Vol. 23, No. 2. 1977

even more marked, and both extremes of the line have
B. //
great uncertainty. Rodbard has emphasized that //
least-squares fits to the logit plot must use a weighting ,,-

algorithm, although this is rarely done. / - -

There are only two plots to consider: ,,%

2
(a) The Scatchard plot is a truly linear relationship
for a homogeneous binding species. Over the range of
Fig. 18.
y values usually found in immunoassay, the variance of
data points does not vary greatly withy if low y values deviations from the experimental variables serve to
( 2 X blank values) are avoided.
(b) The simple linear plot of y (or counts bound) vs. indicate departures from mass-action expectations.
La has a gently sigmoid shape, little change of variance
withy, and is the curve of choice for “best-fit-by-eye” Sensitivity
curve fitting. It can be subjected to a simple nonlinear A lot of controversy has been generated over sensi-
least-squares fit, but because this involves considerable tivity, because of the Alice in Wonderland attitude,
computation it is simpler to use a least-squares fit to a “when I use a word it means what I want it to mean.”
Scatchard plot, which yields more data. There are many definitions of sensitivity; the two most
Two plots that should never be used are: popular are:
(a) B*/Bo vs. La. B*/Bo transforms every curve to (a) The maximum slope of the standard curve.
a common starting point at 1 (or 100%) when La = 0. (b) The least amount of ligand distinguishable from
This destroys the useful data contained in the B0 value, zero.
which provides helpful comparative data on assay Consider the two standard curves, A (left)and B
performance. y is always better than B*/Bo. (right), in Figure 18. The steeper curve looks the more
(b) Any ordinate with log L as abscissa. The loga- sensitive until we examine the ±2 SD limits of the re-
rithmic scale prevents plotting the response at La = 0, sponse shown by the dotted lines. The slope has to be
thereby reducing the standard points by one and de- viewed in the context of the precision with which the
stroying the data relating to a usually critical part of the response can be defined. The precision of the assay is
standard curve required to define the least amount of determined by a combination of these two factors,
ligand distinguishable from zero. The compression of
dL
the high end of the logarithmic scale increases the in- S(L) = S()-
dy
fluence of high La points, of poor precision, on the shape
of the standard curve. where 5(L) and S(.) are standard deviations of L andy,
Never use B*/Bo; never use log La. There are no ex- respectively. Given a nonlinear standard curve (which
ceptions to this rule. is usual in immunoassay) and a changing precision
Remember that the least-squares fit can only produce (counting and experimental) of y, the precision of es-
the best fit to a previously defined function: linear, timation of L will vary widely. Slope of the standard
cubic, etc. If the function is not known, or the wrong curve alone is no measure of sensitivity; curve B was
function is prescribed (e.g., a linear fit to what is ob- obtained from A by doubling the scale of y. The data
viously a curve), then the resulting fit will be nonsense. were identical for both curves. Sensitivity will be used
The fitting routine that is the closest equivalent to a here exclusively as a shorthand form for “the precision
“best-fit smooth curve by eye” is the spline function. It of La as L0 tends to zero.” This is an important measure
allows no previous definition of the form of the curve. of assay performance because it affects the volume of
The spline function requires more than desk-top cal- plasma needed in an assay, and in general the more di-
culator computing power and is relevant only when lute the plasma, the less the matrix interferences. It is
computerized calculation has to be used and least- relevant in those assays where subnormal concentra-
squares fits cannot reasonably represent the data tions are of clinical importance, e.g., the pituitary hor-
points. mones: somatotropin, lutropin, follitropin, cortico-
The four-parameter plot of Rodbard is equivalent to tropin, thyrotropin, and prolactin. In practice, we can-
the logit plot with upper and lower limits not fixed but not define a lower limit of normal for any of these, and
open to computerized trial and error modification. Al- subnormal concentrations have to be defined indirectly
most any deviation from a straight line can be removed by failure of the pituitary to respond to provocative
by stretching or contracting one or other end of the or- stimuli. In some assays (parathormone, corticotropin)
dinate in this way. The plot requires computerized we have difficulty in measuring normal concentrations
weighting of data points and considerable computing and can only clearly define increased concentrations.
power. The process of assay optimization is designed to in-
The Scatchard plot with optimization of upper and crease sensitivity to its practical limit and requires a
lower y limits is within the compass of several desk-top meticulous dissection of the assay procedure to define
calculators and will produce a linear response curve with the best values for many operational variables.
nearly all data. Often the optimized limits correspond When all variables are optimized, the ultimate de-
closely to experimentally measurable variables, and terminant of sensitivity is the antibody equilibrium

CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977 393

constant, K. The higher K is, the greater the sensitivity. ligand do not compete for the same species of binding
An optimized assay should be able to detect ligand with site, no displacement occurs and an assay is not possible.
a precision, as La tends to zero, of But given displacement, there may well be a difference
in the avidity of binding of labeled and unlabeled ligand.
S(L) = 0.1 K’
This is almost certain in the more recent convenience
K has units of litres/mol and K’ is in mol/litre. The labeling with 125J of small ligands such as steroids and
value of S(La) refers to the final incubation mixture of digoxin, which need bulky side-chain addition before
L “, La, and Ab before the separation phase is added. It iodination. If the side-chain substitution corresponds
follows that the greater the proportion of plasma in the to the site of protein linkage to hapten used to raise the
incubation mixture, the better the precision of the es- antibody, the labeled product may have preferential
timate of La in the plasma. This requirement conflicts binding. In other situations the label may impair
with the need to dilute plasma to avoid matrix effects. binding because of steric hindrance by the bulky iodine
Organic extracts of plasma steroids that are then taken atom.
to dryness may overcome this problem; prior adsorption There may be cross reactivity at a site other than that
of corticotropin onto glass beads and elution into a small involved with the labeled ligand, but this generally will
volume similarly concentrates plasma contents before have no effect on label displacement. A different type
analysis. In assays involving use of unprocessed plasma, of interference occurs if the labeled material contains
20% by volume represents a practical limit for plasma impurities X’ along with the target ligand L*. If im-
in the incubation mixture. It follows that nothing is purity reacts with one species of sites and target ligand
gained by using an incubation volume larger than re- with another, a standard curve containing only target
quired for good pipetting precision and adequate mix- ligand La, will represent competition between La and
ing. A tube containing ligand, labeled ligand, and anti- L*; there will be no competition with X*. Now intro-
body in volumes of 1000, 500, and 1000 iil, respectively, duce unknown samples containing not only target ligand
is no better than one containing 100,50, and 100 tl. The but another species either identical to X or capable of
former tube merely represents the contents of 10 of the competing with it. The assay will measure some function
latter tubes pooled. Pipetting precision of 1000-gil vol- of X + L. Purify the labeled ligand until only L* is
umes is no better (about 1%) than precision of SO-tl present; the assay will now measure La alone. Observa-
volumes. Using the larger volumes therefore wastes la- tions such as impure labeled somatotropin detected only
beled ligand and antibody (and money). by an increase in quality-control results and corrected
At one time it was thought that specific activity of by chromatographic purification of the labeled material
ligand limited sensitivity, but Ekins showed theoreti- are probably due to this cause. Similar effects have been
cally that this is rarely the limiting factor. He demon- seen with labeled human 3-choriogonadotropin. It is
strated that the antibody concentration in the final important to recognize that specificity can be influenced
incubation mixtures should be between 1 K’ and 3K-1 by purity of the labeled ligand. Adequate binding and
and that labeled ligand should be 2 K’ to 4 K1. If displacement by standards is not sufficient.
specific activity is sufficiently high to give about 7000 Simple relationships exist for the effects of a cross-
counts in reasonable counting time (1-5 mm) with a reacting ligand, L, with binding energy K in an un-
labeled ligand concentration in the region of 3K-’, then known mixture of target ligand, La, together with L.
sensitivity will be optimal. Increasing specific activity With r = K*/K, a mass L reacts as though it were
10-fold and working at an L* concentration of 0.3 K’ L/’y equivalents of L*, where = r + y - ray. In
will cause no further increase in sensitivity and may general, the quality of the assay data does not justify
impair it, especially because ligand antigenicity often attempts to quantify La, L, and K separately. It is
is damaged if more than one iodine is attached to each sufficient to observe that any dependence of assayed
molecule. Achievement of adequate specific activity is value of a sample on changes in y value renders that
usually no problem, but the ligand must be properly value suspect. The result may not even represent a
purified after iodination and the mass of ligand must reasonable approximation to the true target ligand
be measured in the assay environment in which it is to concentration.
be used. Because iodination may alter the binding The two standard approaches to measuring an un-
characteristics of the ligand (especially with iodination known sample at different y values are:
of steroid derivatives), it is necessary to define K* and (a) Add unknown to a range of standards.
Ka, the equilibrium constants for antibody-labeled li- (b) Assay serial dilutions of sample.
gand interaction and antibody-unlabeled ligand inter- There are the standard recovery and inhibitor tests
action, and to avoid situations in which these differ of colorimetric and enzymatic assays. Both are man-
greatly. This is a required part of the assay optimization datory in assay development. Further, because matrix
routine. effects may well be patient- or sample-dependent (e.g.,
drugs, lipemia, prohormones, complement, immuno-
Cross Reaction logic disorders), it is sound assay practice to assay every
In radioimmunoassay, we measure the behavior of the sample at two dilutions. As a general guide, failure of the
labeled ligand and hope to infer from this what is hap- assay values to agree within 20% after correction for
pening to the unlabeled ligand. If labeled and unlabeled dilution factor is cause for holding the result suspect.

394 CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977

 Unfortunately, the converse is not true; agreement
of dilutions is not always sufficient to validate an assay
result.
Parallelism
If one plots the results of a dilution study against y,
using a logarithmic scale for La, arbitrarily setting the
result for the undiluted sample at an La concentration
of unity, and compares the resulting curve with a stan-
dard curve by using the same axes, parallelism of the
two lines indicates a fixed ratio between unknown and
standard independent of y. This approach has been
used extensively in bioassays, and statistical tests of
parallelism have been devised. It is entirely equivalent
to correcting assay results for dilution and testing for
identity, with use of the known within-assay preci-
sion.
Specificity
What practical steps can be taken to ensure speci-
ficity:
(a) Assay development:
i. Define L * and ra. With this information use the
standard curve to generate a Scatchard plot. Be aware
that multiplicity of binding sites can lead to nonspeci-
ficity owing to labeled ligand impurity and to changes
in separation stage.
ii. Avoid delayed addition of labeled ligand, which
can result in the “first-come, first-served effect.”
iii.Explore plasma matrix effects by adding a plasma
with low ligand concentration to each standard. A y-
dependent recovery indicates matrix effects.
iv. If solvent extraction is used, examine the effects
of solvent residues remaining after evaporation. Re-
member that this artifact can recur with aging of solvent
or introduction of a new batch of solvent.
v. Examine the y dependence of serial dilutions of
samples of high ligand concentration.
vi. Test the ability of structurally related analogs to
compete with the target ligand.
(b) Routine assay:
i. Always run samples at two different dilutions.
ii. Always include low, medium, and high quality-
control samples at beginning and end of the assay
run.
Some specificity problems inherent in immunoassays
can only be resolved by extensive sample purification
or parallel bioassay measurements.
Polypeptide hormones exist in heterogeneous
Along with the biologically active hormone there are
often precursors (prehormones such as proinsulin) and
fragmentation products. Many of these biologically
inactive forms retain specific antigenic determinants.
Divergence between immunological and biological po-
tency has been described for insulin, parathormone,
glucagon, corticotropin, gastrin, thyrotropin, somato-
tropin, and lutropin. Chemical similarities between
somatotropin, prolactin, and between choriogonado-
tropin, follitropin, and lutropin result in immunologic
cross reactivity.
A similar divergence between immunological and
biological activity is seen with thyroxine and cortisol,
owing to the presence in vivo of variable amounts of
thyroxine-binding protein and transcortin.
The only rational approach to these problems is
awareness. Those discrepancies that are associated with
pregnancy or “the pill” can be anticipated by awareness
of patient status. There can be no substitute for keeping
up with literature reports of discrepant or misleading
situations. This is a rapidly moving field, and increasing
questioning of assay values may be expected as the
honeymoon phase of immunoassay gives way to rea-
soned scepticism.
Separation of Bound and Free
In a equilibrium system of free ligand, antibody-
bound ligand, and unoccupied antibody, a separation
of free and bound components may disrupt the equi-
librium. Such disruption will result in reduced sensi-
tivity and, by its introduction of another variable, loss
of precision, especially since reversibility of binding will
be dependent on the time in contact with the separation
reagent.
Some antibodies are effectively irreversible; they
should be identified because the separation stage, once
completed, will go no further and is therefore not time
dependent.
Many antibodies are rapidly reversible and require
precise timing of a disruptive separation stage. It may
be possible to use a nondisruptive separation system
with such antibodies. If antibody is linked to a solid
phase or is precipitated by a second antibody, the bound
and unoccupied antibody are removed in like proportion
and the free moiety remains unchanged:
F(Ab-B) K F(n[Ab-BJ)
B nB
Other systems that precipitate bound antibody (e.g.,
ammonium sulfate) may change the binding charac-
teristics of the antibody as it is denatured, and lead to
disturbance of the first equilibrium. Even second
antibody may disturb the initial equilibrium if the
second antibody has different reaction energies in its
binding of free first antibody and antibody-ligand
complex.
A disruptive separation system should not be con-
sidered always undesirable. Although it imposes a time
dependence on the separation phase, this may not be
forms. unmanageable. As compensation, the separation system
may aid specificity by reversing the weak binding of
cross-reacting ligands while leaving the less reversible
avid binding to target ligand relatively unchanged.
Abraham has noted changes in assay specificity as be-
tween solid-phase antibody (nondisruptive) and char-
coal (disruptive), and a similar loss of specificity is seen
with an estradiol assay when second antibody replaces
charcoal. With second antibody, values for estradiol in
plasma are 1.5-2.5 times the estradiol value when the
assay is done on a crude diethyl ether extract of plasma.
In contrast, the estradiol assay gives identical values
CLINICAL CHEMISTRY. Vol. 23, No. 2, 1977 395
with second antibody and charcoal when chromato- pre-precipitated floccules. This is a version of the
graphically pure estradiol fractions from the extract are solid-phase approach and suffers from the problems of
run. A similar selectivity results from separation of free that technique.
cortisol by partition into toluene-Liquifluor scintillant.
More-polar cross-reacting ligands fail to enter the or- Precipitation by Protein Denaturation
ganic phase and do not interfere. Ammonium sulfate, alcohol, and polyethylene glycol
We will now consider the various classes of separation are used to precipitate antibody along with other
systems. gamma globulins present.
Advantages: precise pipetting, easily automated, fast
Solid-Phase Antibody reaction. Potentially disruptive, which may improve
Antibody is either adsorbed onto the walls of the tube specificity.
or covalently bonded to inert particles. Linkage to Disadvantages: Disruption of initial equilibrium
magnetic iron oxide is an attractive option, because it often leads to poor sensitivity. Nonspecific adsorption
allows a magnetic flux to replace both mixing (which of free ligand to precipitate appears relatively common.
must be continued throughout the reaction) and cen- Polyethylene glycol has to be pipetted at 30% concen-
trifugation. tration, which is inconveniently viscid.
Disadvantages: More antibody is required because
some sites are inaccessible to ligand, owing to steric Adsorption of Free
hindrance. Sensitivity may be impaired if equilibrium Charcoal, usually coated with dextran, or protein,
constant is reduced. The incubation time may need to talc, or Celite are used to adsorb free ligand, leaving the
be increased. Precision may be impaired, owing to ad- bound ligand in solution.
sorption variation or problems in pipetting slurry. Advantages: Convenient, inexpensive. Disruptive
Advantages: Separation is nondisruptive of equi- and may aid selectivity, especially in steroid assays.
librium and requires only decantation or aspiration Disadvantages: Severely disruptive; requires rapid
(with magnetic flux for coated iron oxide). addition of adsorbent to all tubes of batch with cen-
Comments: A variant consists of bonding to solid trifugation of all tubes simultaneously; speed and timing
phase [e.g., agarose (Sepharose)] mounted in a column. essential unless antibody happens to be effectively ir-
This is very convenient for small batches but tends to reversible (rare with steroid antibodies). Affected by
be costly and tedious for large batches. plasma matrix, especially protein concentration. Some
Second antibody itself may be bonded to solid phase, of the bound fraction (10%) is adsorbed onto the solid
which aids the aspiration step and is not subject to the phase.
disadvantages listed above. This procedure is more Comments: Specific selectivity of adsorption is
costly than other acceptable second-antibody tech- sometimes exploited, e.g.; Sephadex-thyroxine, cat-
niques. ion-exchange resin-gastrin or -thyroxine. Pipetting with
these coarser particles introduces imprecision.
Second Antibody (Double Antibody)

Antibody from a different species of animal is raised Summary

against gamma globulin of the species used for raising 1. For small numbers of samples, consider the con-
first antibody. This procedure is generally nondisrup- venience of solid phase, either in columns or on paper
tive and its main disadvantage of taking 16 h to produce disc.
a precipitate has been overcome by incubating in the 2. For larger batches use charcoal for steroids; second
presence of 20 g/liter polyethylene glycol, which brings antibody for digoxin, polypeptides.
the flocculation reaction to completion within 5 mm. 3. For very large batches remember that automated
This approach is generally applicable except with those solid-phase magnetic iron oxide procedures will soon be
rare ligands (e.g., factor VIII) that are themselves pre- the method of choice, e.g., for thyroxine.
cipitated by polyethylene glycol.
Advantages: Nondisruptive separation improves Optimization of Separation Stage
sensitivity; specificity of reaction minimizes misclassi- Given a series of tubes in which ligand is at equilib-
fication of bound and free components; less influenced rium with antibody, the object of the separation stage
by matrix (e.g., plasma protein) effects. is to separate bound from free without disturbing the
Disadvantages: Complement must be inactivated by equilibrium. Several things can go wrong at this
including ethylenediaminetetraacetate. Assay speci- stage.
ficity may be impaired if cross-reacting species with (a) Some of the antibody, and with it the bound
weak binding are left undisturbed. fraction, will be adsorbed with the free.
Comments: Cost is often cited as a drawback. At $2 (b) Some of the free will not be adsorbed.
per milliliter, second antibody may be diluted 10-fold (c) Because the tubes have to be filled one at a time
for use; 100 zl per tube thus costs only 2u per tube, and are centrifuged all at the same time, results will vary
which is a small proportion of total cost. if the separation stage is time dependent.
A variation consists of pre-precipitation of first and Similar problems will arise if the bound is precipi-
second antibody and reaction of the ligand with the tated.

396 CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977

 Three questions must be answered:
1. What separation system is best for a given
assay?
y
zero std.

2. At what concentration?
high std.
3. For how long? I I I I I

0.1 0.5 I 5
The experimental protocol is simple. Separation
systems are tested one at a time in two extreme situa-
tions: with all the ligand bound and with all the ligand
free. A zero standard tube with 100-fold excess of anti-

(TT
Ay
body suffices for the first situation, a tube with label and
no antibody for the second. Because the separation
system may be influenced by buffer, temperature, and
the presence of serum or plasma in the equilibrium Fig.19.
mixture, these variables should be held constant at the
conditions actually existing in the assay.
Great excess of antibody and no antibody are not
realistic assay conditions, because antibody is routinely
always present at constant concentration. It is therefore
y
preferable to substitute for total binding and zero
binding, two real-life situations: the zero standard and
the highest standard to give, respectively, maximal and
minimal binding. This also economizes on expensive
Fig. 20.
antibody, because the use of many tubes with 100-fold
excess of antibody is avoided.
Replicate pairs of zero and high standard are treated
with separation system at varying concentration for
fixed time. Charcoal, for instance, is used at 0.1%, 0.5%,
1%, 5% concentration (1,5, 10, 50 g/liter) for 30 mm. At
low charcoal concentration, some of the free may not be log y

adsorbed; at high concentration, some of the bound

antibody will be adsorbed (Figure 19). The difference
in the bound fractions at these different concentrations
Fig. 21.
will often show a broad maximum. Use the lowest
charcoal concentration that corresponds to this maxi-
mum. With this concentration of charcoal, proceed to
time studies. Here only zero standards are required. operating range of the apparent bound fraction, y’. Set
Incubate for 30, 20, 15, 10, 5, 3, and 1 mm (Figure 20). up duplicate zero standards with 100-fold the regular
With use of a logarithmic scale for y, the data often show antibody concentration and duplicate standards with
a two-component decline in bound (Figure 21). The first 100-fold the regular highest-concentration standard.
component corresponds to removal of bound from weak These tubes should have, respectively, all labeled ligand
binding sites the second, a slower removal from the most bound and all labeled ligand free. They will never give
avid sites, which usually correspond, respectively, to y’ values of 1 and 0. The y’ values will be a and fi, and
broad specificity sites and highly specific sites. The an approximation to true bound fraction, y, is given
charcoal treatment should continue until all weak sites by
have been stripped of ligand; significant improvement
- f3
in specificity of the assay will result.
There is of course, a rapid initial component that is
seldom seen in which the free is adsorbed. This is often
complete in less than 1 mm. The rapid removal of the Setting Up the Assay
free is responsible for disrupting the equilibrium posi- 1. Use clean, uncluttered space. Isotopes are a health
tion and stripping weakly bound ligand from antibody. hazard; contaminated work areas ruin assays. Be un-
With irreversible binding, the “bound” counts rapidly hurried and methodical.
become constant. 2. Decide exactly what you are going to do; write it
Next, determine the time dependence. It is realistic down in a protocol sheet/work list; do it without modi-
to expect that for 40 assay tubes, 1 mm will be required fication. It is surprising how often a wild result is as-
for charcoal addition. What will be the variation in the cribed to transposition of tubes, confusion of numbers
“bound” fraction if the incubation period varies by 1 on tubes-all trivial, all avoidable. Meticulous tech-
mm? If this variation exceeds 0.5% of total counts, the nique saves time spent on repeats.
assay is likely to be unusable. 3. The work list. Use a common protocol that differs
Finally, obtain an estimate of the upper and lower as little as possible from one assay to the next.

CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977 397

 A. Start with seven standards plus a zero tube. The
lowest standard should be just above the limit of de-
tection of the assay. The highest, at twice the upper
limit of the assay, serves as a guide for dilution of high
samples. The other five may be spaced equally over the
range of the assay or may be concentrated toward the
lower end, depending on the assay ligand. It is wise to
have standards close to the upper and lower limits of the
normal range, because these concentrations are likely
to involve clinical decisions and maximal precision is
needed here.
B. Follow with three quality-control samples: low,
normal, and high.
C. Run samples with a mid-quality control in every
tenth position.
D. End with the same three quality-control sam-
ples.
4. Run every test in duplicate-standards, quality
control, and unknowns. There is no exception to this
rule. Diagnostic and therapeutic decisions may rest on
one result; blunders are unacceptable.
5. With assays that cover an extended concentration
range or those that are known to be subject to matrix
effects, it is sound practice to run every unknown sam-
ple at two dilutions, usually with only a twofold differ-
ence in concentration so that both results fall within the
range of the assay. With assays of intrinsically good
precision, it is acceptable to run each dilution in a single
tube rather than in duplicate. Failure of results for the
two dilutions to agree then indicates either a matrix
effect or a blunder. In either case, repeat testing is in-
dicated. Agreement of two dilution-corrected values,
run singly or in duplicate, from twofold differing vol-
umes of the same plasma is a strong indicator of a valid
assay result, free of matrix inaccuracy. Lack of agree-
ment must cast doubt on the validity of the result. It
may not be possible to remove that doubt, but it is
critically important for analyst and consumer to know
that reasonable doubt exists.
6. Have standards ready diluted as though they were
plasmas containing the specified amount of ligand. Use
the same protocol and the same pipette for standards
and unknowns. Adding a standard in varying volume
to assay tubes to give increasing concentrations is in-
excusable. In immunoassay, everything depends on
repeatability of assay conditions from tube to tube.
There is never a good reason for using a different pro-
tocol for standards.
7. Use a single batch of tubes for an assay. 1251 is a
weak gamma emitter. There is about a 20% loss of
counts if glass tubes rather than plastic tubes are used.
There will be unpredictable losses if tube batch is
changed within assay.
8. Use a protocol that specifies precisely how long
and at what temperature for each tep in the assay, and
stick rigidly to the protocol.“Add charcoal, 1 ml to each
tube, and centrifuge after 30 mm at 5 #{176}C”is a mean-
ingless direction. Is this stage of the assay tempera-
ture-dependent? If so, the charcoal should be at 5 #{176}Cmance is a necessary preliminary
before pipetting, as should the assay tubes, and the
398 CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977
centrifuge must be refrigerated. If not, there is no need
to mess about, do the whole procedure at room tem-
perature. Is the charcoal addition time-dependent? If
so, the 30 mm must have limits: “add charcoal to each
tube, completing all additions within 1 mm, let stand
for 30 mm.. . If not, say so: “centrifuge after a mini-
.“
mum of 15 mm and a maximum of 4 h.”
Statements in the protocol cannot be precise unless
their limits have been defined. To play safe by hedging
every instruction with meticulous restrictions is to en-
courage sloppy technique since staff in a busy laboratory
cannot waste time on irrelevant and unhelpful restric-
tions. Know what matters and what does not and con-
centrate special attention only on steps that are critical.
This is part of good assay development, which we shall
consider later. To any visitor who comes to your labo-
ratory and asks “why are you doing that assay step in
that way,” you should have a clear and logical answer.
The place for working by rote is the kitchen, not the
laboratory-and never the immunoassay laboratory.
9. Do not take counting equipment for granted. Have
calibration for 1251 set up by an expert and monitored
regularly. Remember that counters may be upset by
line-voltage fluctuations. Monitor counter precision by
counting a control sample when the counter is not oth-
erwise occupied. Test the effect of switching on any
other large equipment. Get advice from a physicist and
stick to it.
10. Be consistent in the use of concentration units.
Avoid a plethora of conversion constants for volume,
mass, concentration, and units. The standards can be
expressed in terms of ligand mass per tube, concentra-
tion per final incubation volume (before or after addi-
tion of charcoal or equivalent), or concentration of
standard solution as it is added to each tube. The choice
is arbitrary; none is theoretically best. We have already
imposed an absolute rule that standards and plasmas
(diluted or not) are added with the same pipette. It
follows that the concentrations of standard solutions to
be added to standard tubes are directly comparable with
the concentrations of plasmas added to test tubes.
Identical concentrations of each should give identical
reduction of bound counts. Unknown plasma concen-
trations can thus be read directly off a calibration curve
set up in terms of concentrations of standards. There
are no factors or unit changes other than the trivial
dilution factor of diluted unknowns if these are run both
undiluted and diluted. This approach has the virtue of
simplicity; there is no better way of reducing errors. Use
it always. It is equivalent to saying that arbitrary con-
stants applied many times over to samples should be
applied once only to the standards.
Method Development
Protocol
Given labeled ligand, unlabeled standard ligand of
known concentration, and antibody, the development
of an optimized assay and the evaluation of its perfor-
to routine use. It is
reasonable to expect that in the final assay there should
be a good reason for every step, time, temperature,
etc.
The following protocol provides answers to the more
important questions.
1. Equilibration time at 5 #{176}C
Measured: forward reaction rate
Purpose: define ligand-antibody incubation time for
evaluation of assay
2. Reversibility at 5 #{176}C
Measured: reverse reaction rate
Purpose: (a) is the separation stage potentially time
dependent?
(b) is delayed addition of labeled ligand an
option?
3. Separation system
Measured: effects of concentration, time, and tem-
perature on the chosen separation sys-
tem
Purpose: define conditions required for maximum
discrimination between bound and free
4. Antibody titration at 5 #{176}C
Measured: range and magnitude of “bound” fraction
in absence of unlabeled ligand
Purpose: define upper (a) and lower (f.) bounds for
“bound” fraction and select antibody
concentration to give 50% binding
5. Standard curve at 5 #{176}C with increments of tracer
Measured: binding capacity (Ab); equilibrium con-
stant of antibody-unlabeled ligand (Ka)
and of antibody-labeled ligand (K*);
mass of labeled ligand (L *)
Purpose: selection of optimum antibody and tracer
concentration
6. Temperature dependence of binding
Measured: equilibrium constant at 25 and 37 #{176}C
Purpose: (a) defining appropriate incubation tem-
perature
(b) ascertaining optimum attainable sen-
sitivity
(c) ascertaining thermodynamic constants
of antibody-ligand interaction
7. Equilibrium time at operating temperature
Measured: forward reaction rate at routine assay
incubation temperature
Purpose: ascertaining routine assay incubation
time
8. Working standard curve
Measured: binding under operating conditions at
different standard concentrations
Purpose: definition of range and operational sensi-
tivity of assay
9. Matrix-accuracy-recovery
Measured: (a) low unknown with added standard
increments
(b) high unknown serially diluted
Purpose: ascertaining “how much of the target ligand
is being missed by the assay” = accuracy
or recovery. Is recovery dependent on level
of bound/total counts? Is unextracted
plasma analytically acceptable?
10. Cross reactivity-specificity
Measured: reactivity of high concentrations of po-
tentially cross reacting ligands
Purpose: what are the risks of nontarget ligand being
erroneously included in the result?
11. Influence of temperature, separation system, la-
beled fractions, solvents on matrix effects
Measured: apparent ligand concentration in 40 un-
known samples when assay conditions
are altered
Purpose: is the assay robust? Will deterioration of
label modify results? Is specificity depen-
dent on temperature or type of separation
system?
12. Within-batch precision
Measured: results on replicate samples at selected
ligand concentrations
Purpose: precision of assay at zero, low, medium, and
high concentrations of ligand
13. Between-batch precision
Measured: concentrations of three control sera in
every batch
Purpose: definition of operational coefficient of
variations at low, medium, and high con-
centrations
PracticalAspects
Given antibody, standard ligand, and label, where do
we start?
1. Find a condition of incubation and separation that
gives increasing binding of label with increasing anti-
body concentration. Use 5 #{176}C,
5-day incubation, and
second antibody for peptides, charcoal for steroids.
Duplicate zero-standard tubes with buffer, label, and
antibody diluted 1/10, 1/100, 1/1000, 1/10 000, and 1/
100 000 are sufficient.
2. Select an antibody dilution that gives about 50%
binding. Set up duplicate zero-standard tubes in which
tracer is added at time t = 0 (4 duplicates) 24,48, 72, 96,
108, 114, 117, 119, 119.5, and 119.75 h. Terminate by
addition of separation phase at 120 h (only use one of
the t = 0 duplicates). This gives incubation times of 120,
96, 72, 48, 24, 12, 6, 3, 1, 0.50, and 0.25 h. The extra t =
0 duplicates are available for more lengthy incubation
if the 96- and 120-h incubations are not at plateau. To
one pair of the t = 0 duplicate tubes add 100-fold the
upper projected operating standard. Terminate after
1 h of incubation at 5 #{176}C. This demonstrates revers-
ibility.
3. Using the same antibody concentration and pairs
of zero and top operating standard tubes, test the effect
of different concentrations of separating reagent. With
the optimal concentration, measure the time depen-
dence of separation as already described.
4. Define a and f by using duplicate zero-standard
tubes with antibody in 100-fold excess and duplicate
100-fold excess of standard with regular antibody con-
centration under equilibrium incubation conditions
with optimized separation system.
5. Set up three standard curves, using duplicate tubes
CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977 399
at each standard concentration. Incubate at 5 #{176}C, 25 CC,
and 37 #{176}C.
In the 5 #{176}C
curve, set up four extra zero
standards that contain two-, four-, six-, and eightfold
the regular amount of label.
Calculate L*, Ab, and K* at 5, 25, and 37 #{176}C.
Define ra = K*/Ka
6. If K is temperature dependent, useS #{176}C for incu-
bation. If not, use 25 #{176}C
unless there is peptidase incu-
bation hazard (e.g., corticotropin and parathyrmn). If 25
#{176}C
is used, redefine equilibrium time, and repeat stan-
dard curve using time sufficient for 95% of equilibri-
um.
7. Assay low unknown, added to every standard tube,
against buffer added to every standard tube. Set up
three high-concentration sera, serially diluted to the
limits of detection of the assay.
8. Assay potential cross-reacting ligands. Assay 40
random unknowns and compare with results on using
a different antibody and/or a different separation sys-
tem or assayed in another centre. Check the effect of
using old label, or adjacent fractions from label purifi-
cation on the 40 samples. Check for solvent residue ef-
fects in the zero-standard tube.
9. Assay 40 samples at low, medium, and high con-
centrations within same batch.
10. Set up the quality-control system for between-
batch monitoring.
Quality Control
The Program and the Problems It Reveals
A quality-control program is an integral part of assay
design. No data may be reported without acceptable
quality-control performance. Surveys between labora-
tories reveal vast discrepancies, 10-fold differences may
occur. These differences are due to:
-standard-calibration errors
-differences in assay protocol
-variation with assays
Standard calibration errors can only be avoided by
using the ourest standard material available, storing it
under conditions that give long-term stability, and
calibrating it at regular intervals against international
reference material. An effective quality-control system
will monitor short-term trends in standard concentra-
tion caused by decomposition or evaporation of solvents.
This is one function of the control system that must be
clearly identified.
Given reference standards, variations in results for
plasma samples owing to assay protocol differences
between laboratories imply a failure of the assays to
measure ligand concentrations without discrimination
between standards and plasmas. If such plasma-related
matrix effects are present, there is no validity in ex-
pecting all plasmas to show the same matrix effect. It
is sometimes possible to change separation systems,
antibody, or technique. For instance, some ligands are
measurable by gas chromatography or mass spectrom-
etry.
Other approaches to revealing standard-plasma
discrimination are standard additions to plasmas and
400 CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977
running many different plasma dilutions. These have
been discussed under Method Development. The only
technique in this category that is relevant in routine
monitoring is assay of samples at two dilutions, e.g., 100
and 50 gil. Failure of answers to correspond to plasma
sample volume indicates sample-specific matrix effects,
which may always be present-because, for instance,
of the presence of therapeutic agents or of pathologically
occurring abnormal proteins such as myeloma protein
or rheumatoid factor.
A third area of relevance to quality control is the
monitoring of within-batch precision, i.e., knowing the
confidence limit of a given assay value. This relates to
the least amount distinguishable from zero, our defi-
nition of sensitivity; it also indicates the appropriate
rounding off of answers to give only significant fig-
ures.
The final quality-control function is to reveal
batch-to-batch precision, which reflects the relevance
of small changes in results from individual patients
monitored over a prolonged period.
Thus a quality-control system must indicate:
(a) Changes in accuracy of standards
(b) Changes in assay-dependent matrix effects that
produce general plasma-standard discrepancy
(c) Isolated sample-dependent matrix effects
(d) Within-batch precision, random and drift
(e) Between-batch precision
Note that in (a) and (b), only changes can be moni-
tored; the presence of any inaccuracy is not definable
by quality-control monitoring.
How do we isolate the five messages we may get from
a practical assay?
Set up the followingsamples:
1. Three control samples-low, medium, and high-
from aliquoted pooled sera are run at the beginning and
end of every assay.
2. Every unknown sample is run in duplicate at two
dilutions. With insight into assay performance this is
varied to running each sample either singly at two
dilutions or in duplicate at one dilution.
3. Three samples-low, medium, and high-from the
previous batch are rerun in the next batch as repli-
cates.
Plot control values at two dilutions, start and end of
batch, on a logarithmic ordinate, using different sym-
bols for each dilution.
On the same chart, plot one or more assay parameters
for each assay: yo, La at 0.5 y, Ab, etc. Availability de-
pends on computing routine. We find Ab most useful.
Plot difference in replicates on a linear chart with scope
for positive and negative change.
Calculate within-batch precision by using duplicate
values of grouped patient data in low, medium, and high
range
2n
where d is difference between pairs, n is number of pairs,
and a- is standard deviation of individual result. SD of
a reported mean of two results will be a-// n must be i. Antibody
greater than 30. If necessary, pool several batches. ii. Labeled ligand
How do we use these data to define categories a-e? iii. Buffer
(a) Change in standards: A simultaneous propor- iv. Separation reagent
tionate change in three controls and in Ab (all giving the v. Solvent
same slope on log scale) has to be due to a change in the vi. Tubes: adsorption, counting efficiency
units used to define all four parameters, i.e., standard
units. The same change will occur in the three replicates, Future Developments
but these data will be less obvious if the change in What is coming next in immunoassay? The answer,
standards is a trend over several assays. in the next three years, is
Remedy: prepare fresh standards and run in the next (a) Automation
assay together with old standards. (b) Non-isotope immunoassays
(b) Change in matrix effects caused by assay. Change (c) Receptor assays
in controls and replicates, but no change in Ab-and
often no clear proportionality-indicates assay problem. Automation
This may be a solvent residue effect, deterioration of There are three systems now being evaluated. The
charcoal or buffer, or decomposition of label. CENTRIA, a column-based centrifugal approach; the
(c) Sample matrix: Failure of agreement at two MICROMEDIC, an automated pipetting system; and the
dilutions is often seen. A tendency for the smaller Technicon Automated RIA System, a magnetic solid-
sample volume to give a slightly larger result when phase continuous-flow system. The continuous-flow
corrected for dilution indicates inhibition of binding by technique uses antibodies bonded to magnetic iron
plasma matrix. Isolated samples may show this effect oxide particles, which are passed along a segmented
greatly enhanced, even to the extent of the direct assay stream until they reach the scintillation counter. Here
value on SO il being greater than the result on 100 il a timed magnet holds up the solid-phase antibody for
without any dilution correction. If this effect does not one sample at a time, the bound counts remain sta-
occur in the first 1000 samples in a given assay, drop the tionary, and the free pass on. The bound counts are then
two dilutions. switched into a counter and counted for 1 mm. This
(d) Within-batch precision: system will allow analysis of large numbers of samples,
i. Drift: a consistent difference between controls at such as thyroxine.
start and finish of every batch indicates a time depen-
Non-isotope Immunoassays
dence in the assay that must be identified and elimi-
nated. The charcoal step is often time dependent and The label is essential in immunoassay. It need not be
requires rapid addition of reagent, and batches of no a radioisotope and several alternatives are available.
more than 50 tubes. Some have the attraction of permitting measurement
ii. Random: within-batch SD from grouped dupli- of bound without separating it from free. Fluorescence
cates indicates the best operating precision attainable. quenching, fluorescence polarization, the enzyme
Round answers to ±0.5 SD. Expect 5-10% CV within multiplied immunoassay technique (EMIP5, Syva), and
batch. electron spin resonance all work without prior separa-
(e) Between-batch precision: Calculate the SD for tion. A wide range of enzyme-linked assays have been
30 control samples from the last 30 batches. Draw in ±3 developed with enzyme linked to ligand (enzymo-
SD limits on the chart. Control data outside these limits immunoassay) or to antibody (immuno-enzymometric
are unacceptable. Repeat the batch. Expect between- assay) by using both one- and two-site techniques. The
batch precision of 10-15% (1 CV). enzyme luciferase (Photobacterium fischeri, EC 1.2.-.-)
presents considerable attraction, used in a system with
What to Do Next NAD/antibody linkage, because it allows counting of
Expect assays to go out of control; even experienced photons with standard scintillation-counting equip-
operators in large assay laboratories have frequent ment.
unexplainable problems. All these techniques, excepting luciferase, have
1. Ask what has changed since the last good assay. sensitivities about 100-fold less than with isotopes. They
2. Check which reagents in the unacceptable assay have major usefulness in the drug assay area.
have been used recently in some other assay that gave
no problems. Solvents, buffers, second antibody, char- Receptor Assay
coal are often shared by several assays. In assay devel- Cells have specific receptors for hormones, either on
opment, select conditions that keep reagents in shared the cell surface or in cell sap. Assays now exist for these
use. This saves reagent-preparation labor, eases storage receptor sites, e.g., insulin or monocytes, estradiol in
problems at 5 #{176}C,
and provides valuable cross checks on breast acinar cells. Assay of these receptor levels will
performance. become increasingly important. Circulating antibodies
3. A common problem is the zero tube that binds less to some receptors have been demonstrated (insulin re-
ligand than usual. Remember there are up to six rea- ceptors, long-acting thyroid stimulator, or thyroid-
sons. Check them systematically: stimulating immunoglobulin for thyrotropin receptors).

CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977 401

Assay of estradiol receptors is important in determining Zettner, A., Principles of competitive binding assays. Clin. Chem. 19,
699 (1963).
relevant surgical therapy for breast cancer.
Principles of Competitive Protein-binding Assays, W. D. Odell and
In addition, the receptors may themselves be used for W. H. Daughaday, Eds. Lippincott, Philadelphia, Pa., 1971.
assay of ligand. They are unlikely to replace antibody Radioimmunoa.ssay Methods, K. E. Kirkham and W. M. Hunter, Eds.
because of their lability, but they will complement im- Williams and Wilkins, Baltimore, Md., 1971.
munoassay in the cytochemical assay that allows hor- Handbook of Radioimmunoassay, G. Abraham, Ed. Dekker, New
York, N. Y., In press.
mone to react with slices of target tissue and measures
intracellular response quantitatively by scanning in- Theoretical
tegrating micro densitometry. Measurement of bio-
Rodbard, D., and Feldman, H. A., Theory of protein ligand interac-
logical activity rather than of antigenic reactivity will tion. In Methods in Enzymology, 36, Part A, B. W. O’Malley and J.
be important in difficult diagnoses. G. Hardman, Eds. Academic Press, New York, N. Y., 1975, pp.
3-16.
Applications Ekins, R. P., Newman, G. B., and O’Riordan,
J. L. H., Theoretical
aspects of “saturation” and radioimmunoassay. In Radioisotopes
#{149}
Peptide hormones will be fractionated into pre- in Medicine: In Vitro Studies, R. L. Hayes, F. A. Goswitz, and B. B.
cursor and breakdown fragments before assay. P. Murphy, Eds. U.S. Atomic Energy Commission, Oak Ridge, Tenn.,
1968, pp 59-100.
#{149}
Gastrointestinal hormones, gastrin, cholecystoki-
nm, gastric inhibitory peptide, vaso-intestinal peptide, Optimization
chymomedin, motilin, and pancreatic peptide will be
Ekins, R. P., Newman, G. B., Piyasena, R., Banks, P., and Slater, J.
increasingly important. D. H., The radioimmunoassay of aldosterone in serum and urine:
#{149}
Prostaglandins, bile acids, drugs. Theoretical and practical aspects. J. Steroid Biochem. 3, 289
#{149}
Hepatitis B antigen, other viral and bacterial an- (1972).

tigens and antibodies. Quality Control

#{149}
Enzyme assay by mass rather than activity will
Rodbard, D., Statistical quality control and routine data processing
undergo exponential growth.
for radioimmunoassays. Clin. Chem. 20, 1255 (1974).

Suggested further reading Steroids

General Abraham, G.E., Radioimmunoassay of steroids in biological materials.
Practical Radioimmunoassay, A. J. Moss, G. V. Dairymple, and C. Acta Endocrinol. 75, Suppl. 183 (1974).
M. Boyd, Eds. Mosby, St. Louis, Mo., 1976.
Delayed Addition
Radioimmunoassay, L. M. Freeman and M. D. Blanfox, Eds. Grune
and Stratton, New York, N. Y., 1975. Rodbard, D., Ruder, H. J., Vaitukaitis, J., and Jacobs, H. S., Mathe-
Radioimmunoassay and Saturation Analysis, P. H. S#{246}nksen,
Ed. Br. matical analysis of kinetics of radioligand assays: Improved sensitivity
Med. Bull. 30 (1974). obtained by delayed addition of labelled ligand. J. Clin. Endocrinol.
Yalow, R. S., Radioimmunoassay: Its past, present, and potential. In Metab. 33, 343 (1971).
Progress in Analytical Chemistry, 7,1. L. Simmons and G. W. Ewing, Pratt, J. J., and Woidring, M. G., Radioimmunoassay specificity and
Eds. Plenum, New York, N. Y., 1974, pp 1-32. the “first-come first-served effect.” Clin. Chim. Acta 68, 87 (1976).

402 CLINICAL CHEMISTRY, Vol. 23, No. 2, 1977