103
would remain in molar units (e.g., nM). The regression model
(equation 4-6) and the actual regression equation would be exactly the
same as described above, except that and Bq now would be expressed in
the units fmole/mg protein. The regression would provide estimates of
the parameters Bq, K^, and C-j; and the results of the merged regression
would be plotted for display in the same coordinate system as the
"merged Scatchard" plot described above. This analysis would also
require the assumption that the parameters to be fit do not vary with
the cytosol protein concentration.
Although the absence of a correlation between Bq (fmole/mg) and
cytosol concentration made the above assumption seem reasonable, the
variability of the individual estimates of receptor concentration (Bq)
again suggested that a better merged regression estimate of would be
obtained by performing the nonlinear regression with Bq as an input
variable, instead of using it as a unitary parameter to be estimated.
Therefore a 2-parameter, 3-variable regression was performed using
exactly the same regression model and equation (again weighted as 1/By )
as described above, except that for each value of By, the corresponding
estimate of Bq from a Scatchard analysis of the individual isotherm con
taining the input data point was used, in addition to the total ligand
concentration (S^), as a second (unweighted) independent variable. The
regression program thus generated estimated values of d and C-| and
their "asymptotic" standard errors described above. The resulting com
plete 2-parameter regression model was plotted for display (e.g., figure
4-16) in the "Scatchard-1ike" coordinate system in which By was replaced
by Bt/Bq, the normalized individual value of total bound steroid. The
measured nonspecific binding corrections (B^) were sometimes