I;m not sure that I correctly understand your experiment but assuming that only membrane or nucleus expresses (but not both simultaneously or neither), then you could just compare the proportion of M expression in group A with the proportion of M expression in group B (Pm A vs Pm B). The proportions of nucleus expression Pn would be (1- Pm) A and (1-Pm) B

Then do a difference in proportions test by normal approx to binomial, chi sq or exact test

Thank you for the answer! James Leigh

In fact, we have the possibility of both (membrane and nucleus) expresses simultaneously or neither.

Does it have any influence on the choice of test?

You could still use chi sq test but consider chi sq components in comparisons .

eg Group A MN M N O

30 40 20 10

Group B 20 30 40 10

do chi sq on 2 x 4 table for overall difference in distributions , then explore components.

You can use chi square for trend.

A more elegant alternative would include the transformation of the expression into numbers (continuous) and use GEE (Generalized Estimating equations).

You may use the Digital HScore to quantify the expression