What I mean is, if the inputs of PLS (X predictor variables and Y predicted variables) are discrete variables, can PLS regression be used?
Yes, you can do that by using the Y variable as you normally would in continuous PLS, and then subsequently using either LDA (linear discriminant analysis) or KNN (K nearest neighbors) performed on the scores (T) to classify the samples. I did this here: https://www.ncbi.nlm.nih.gov/pubmed/17059674
Edit: I think I misread your question the first time. There is a method called PLS Discriminant Analysis (PLS-DA). PLS-DA uses a discrete Y and continuous X. I'm not sure if you could modify PLS-DA to use a discrete X as well, but I don't see why it wouldn't work to just try using a discrete X with PLS-DA.