If you look at your regression coefficients you preferably take those coefficients or interactions that are significant. Two factors at a time can be visualized in a contour plot, the others have to be set to normally at their central values. I find often contour plots more easy to interpret than response surface plot - less complexity - same theory. It is only a matter of visualizing the response as 2D or 3D.
Yes, the contourplots are 2D representations of the response surface, the prrinciple is the same as in geography were the height of the landscape is signed with different colours and lines.
According to the files you supported there is interaction between the C and A and C and B parameters becase the increment in the value of the enzyme activity is different according to the C factor is on 1- or +1 level.
According to the CD graph I assume the presence of nonlinear relationships. Maybe a higher level design would also be advantegous.
Risking to be repetitive to what my colleagues already explained, when you have an unknown physical law to explain an effect or response to certain variables, you may adjust the real (physical) phenomenon by a mathematical approximation: the "response surface".
Usually a polynomial of a certain order (not more than 2 or 3) with adjustable parameters is employed.
The parameters are find out empirically from an experimental design. From the graphs you added to your question it seems that a Simplex Centroid model was applied in your case.
But, as the response in your case depends on more than two independent variables, the result cannot be pictured into a three dimensional space as a surface. It is in fact an "hyperspace".
Thus the only visualization possible is through the "projections" of the hyperspace into several two-dimensional spaces, which are in fact the several two-dimensional contour plots that you attached.
It would be as to see the shadow of a mountain at several times (positions) while the Sun is shitfing about the horizon. But in more than three dimensions.