A key difference between the correlation coefficients and the regression is that the regression incorporates the effect of all your predictor variables simultaneously.
It is certainly possible that when measuring the individual relationship between leadership and digitalitaet individually, there is a significant correlation, but in a model that also includes partizipation, the relationship is not significant.
For example, it could be that partizipation is somewhat multicollinear with digitalitaet, and the predictive information in digitalitaet partially "exists" in partizipation. And with both variables in the regression, the existance of partizipation makes digitalitaet partially redundant. This is just one possible situation.
Some of the work you can do is to measure the correlation coefficients between all your variables, and compute the VIF for all your predictor variables (with the car or usdm packages) to diagnose the multicollinearity. Multicollinearity is one of the most common things that creates problems for linear regression, and VIF is a common diagnostic.