It's likely due to Stockfish's transposition table (essentially a fast lookup table that Stockfish can use to refer back to positions it evaluated before). During your first run, information about visited positions was stored in the table. Then on the second run, Stockfish may have looked at some of Black's moves after depth 1 (since it's already evaluated all states at depth 1).
SmallChess' answer to this question may be useful if it's still confusing:
How to read score from stockfish - inconsistent response every call?
Edit: to test if it is the transposition table behind this, try to reset it after the first run (close the whole program if you can't figure out how). Then on the second run, the number of nodes should be the same (as well as the centipawn evaluation).