How to implement Backpropagation through Time in tensorflow

Question

I know how to implement Vanilla NN\'s BP using tenosrflow\'s GradientDesentOptimizer but I have no idea in case of RNN. Should I first construct unrolled graph and then calc