They always increase the $R^2$ of the model.

They have high potential to pull the regression line towards them, exerting significant influence on $\hat{\beta}_1$.

They increase the variance of the error term $\sigma^2$.

They have no effect on the regression coefficients.