Linearly increasing trend in PLS residuals

#1
Hi everyone,
I am currently using partial least square regressions to predict a variable from a set of 6 others. In general, residuals are randomly distributed, with sometimes a certain level of heteroscedasticity. For one dataset, however, the residuals are positively correlated to the variable to be predicted (not to the predicted values), see below. The predictions for that dataset are not great. What can cause this? Should I use a different model? residuals.PNG
Thank you