suppressor variables

I have built a multiple linear regression model with two independent variables (say X1 and X2). The primary aim is *not* to use the model for prediction, but to understand the total variance of the dependent variable (say Y) explained by the Xs.

The X1 and X2 were derived from a dynamical/physical system in such a way that they are expected to be independent of each other, but reasonably correlated with Y. However, in practice I found that X1 and X2 are correlated with each other (though the correlation is small -0.08).

Now I went on to do the variance partitioning. I found that the joint contribution of X1 and X2 is NEGATIVE. The reason could be that Cor (x1,Y) >0; and Cor (x2,Y)> 0; but Cor (x1,X2)<0.

Could anyone confirm if this is the reason? Could anyone also suggest how to deal with this situation?