algebraic properties affecting a protocol

This is from a past exam paper. The question is as follows:

A protocol designer uses signature to simplify, and hopefully correct, NSPK:

1. $A → B$ : $sign(sk_A, encrypt(pk_B, N_A))$
2. $B → A$ : $sign(sk_B, encrypt(pk_A,(N_A, k)))$

where $N_A$ is a fresh random nonce created by $A$, $k$ is a fresh session key created by $B$, and the sign and encrypt operations use the public key cryptosystem to respectively sign and encrypt under the respective keys. You should assume that $A$ and $B$ know each others’ public keys $pk_A$ and $pk_B$.

Does this work (i) in the case where signature and encryption satisfy no algebraic properties other than those implied by their definitions and (ii) if signature and encryption commute (i.e. $sign(k_1, encrypt(k_2, x))$ = $encrypt(k_2,(sign(k_1, x)))$?

In each case either argue convincingly that the revised protocol works or analyse exactly what your attack on it achieves.

I can't come up with a man in the middle attack or a reflection attack for this protocol - and don't really see how the algebraic properties of these functions are relevant either way? How do I solve this question?