Test negative factors among correlated determinants

Dear all,

I am working on the obesity research. I found a molecule A that is negatively control the generation of adipocytes. I have a cohort of patients, and collected their adipose tissue samples (n=29). I do found that A is negatively correlated with adipogenesis in lean patients (n=22). However, A is strongly correlated with obesity, and in obesity people (n=7), both adipogenesis and A is increased. I can not tell the effect of A on adipogenesis in those patients.

I tried multiple lineage regression on total samples, but it worked badly. I am going to try partially least square, do you think it is a appropriate way? Whether there is a better statistic model can deal with this situation?

Thanks a lot for your time. I appreciate any suggestions and comments.