DV Transformation Changes Sign of Correlation in Regression

Hi Everybody,

I've run into an issue with my data and was hoping this community might have some ideas.
I am running several multiple regressions, and I've done data transformations on dependent variables that have excessive skew, either using a log (k-x) function or a sqrt (k-x) function where k is 1 greater than the largest possible value of x.
But I am finding that after these transformations, the sign of the correlation coefficients, as well as the beta weight in regression, has been flipped (from negative to positive or positive to negative). Attached is an example - you will see that the "foodshopqual" IV is positively correlated to the "SEFFIC" variable, but negatively correlated to the log transformed SEFFIC variable.
Any thoughts? Ever seen this before?

Thanks for any and all input!