You need to conjugate either the protein or the ligand with a fluorophore (if it is not fluorescent). Labeling is usually carried out on thiol groups (Cysteine) or amino groups N-terminal amine or Lysine).
Then, you have to make sure that the conjugation procedure did not significantly affect the interaction. To do that, it is important to compare the labeled vs. the unlabelled protein/ligand.
Note that best results are obtained by tagging the ligand, which is in most cases the smaller (macro)molecule.
To determine the equilibrium dissociation constant, you need to perform a dose-response experiments. I recommend both saturation and competition experiments, the latter using the unlabelled ligand.
Hope this will prompt you to further study the technique and to exploit its full potential.
Article Analysis of protein-ligand interactions by fluorescence polarization
Hope these help. Once you have FP set-up it is pretty straightforward. You can do equilibrium competition binding experiments in FP to then determine Kd.
1. Plot fluorescence anisotropy versus substrate concentration [A] for the protein. You will have a maximal value (y1) to which the data extrapolate and a minimal value (y2). Maximal and minimal values do not necessary correlate to bound or unbound forms. Whichever one happens at higher ligand concentration corresponds to the bound form and whichever one happens at lower concentration corresponds to the unbound form.
a. If the data can be fit to a fraction bound (Xa) equation, do so. The fraction will be calculated as follows:
Xa = {e^(-nH*(Kd-[A]))}/{1+e^(-nH*(Kd-[A]))}
nH is the Hill constant, a measure of cooperativity. It will be “1” if the data is not cooperative, between 0 and 1 if there is negative cooperativity, and no larger than the number of binding sites if it is infinitely cooperative.
b.Then the intensity can be mapped to this fraction bound through the maximal and minimal values of anisotropy:
anisotropy = y2 + (Xa*(y1-y2))
Use the plot of intensity versus concentration and non-linear regression fit of the data to determine the value of Kd for the ligand. When your fit matches the observed data with a minimal chi-squared value, you have the correct value of Kd.
One more thing that may help: you have to put in initial parameter guesses for nH, Kd, y1 and y2 when you do it this way. You may find that setting the Hill constant =1 for your first guess and NOT floating it during the non-linear regression speeds your mathematics up and lets you determine close guesses for the other parameters faster and with greater accuracy. The same would be true if you fixed y1 or y2. The fewer things you have to vary in this calculation, the more reliable your results will be for the rest. It's kind of like having to reckon the position of the sun into a determination of direction you are travelling, and rather than having to calculate the direction of the sun, you just point to it because you know where it is. See if you can get the fit to converge onto your data without having to float all four parameters. If you can, then you should go with that.
Oh, and sometimes the fluorophore will not respond to the binding of ligand, as observed by no change from the maximal and minimal values of anisotropy. In such cases, you cannot use fluorescence to determine ligand binding.