tensorflow gradient with repect to a subset of trainable variables

Question

I would like to take the gradient of the loss function just with respect to a single weight in a layer. For taking the derivative with respect to the entire first layer, the fol