Yes, it is possible. PLS-DA maximizes the **covariance** between X (data) and Y (group). The variance displayed in the plot above is the **explained variance for X** . Covariance and x-variance may not agree with each other in some cases. For instance, the 1st component may not explain more X-variance than the 2nd component. In summary:

- The first component still explain the most co-variance between X and Y (most predictive)
- There are other factors (in addition to Y) contributing to the variance in the data X