The dependent variable of linear regression does not have to be normally distributed. The only normality assumption of linear regression is on error terms. But if the dependent variable is too non-nomral, it affects normality of error terms.