Leko won game three with white and drew the second game to take a 2.5-1.5 lead in the match at the halfway point. Leko varied from his eternal 1.e4 with d4, something he's tried on a few other occasions with mixed results. He may have been eager to test Karpov in the Queen's Indian and the long theoretical line they ended up in. 13..c5 has been a hard worker of late, although 13..b5 is still alive and kicking. They line leading to 20.Qg4 was played in Topalov-Anand in San Luis last year, but it goes back to 1997. Everyone has played 20..Bg5 but Karpov went for 20..Bf6, which seems reasonable since Topalov got a lot of pressure from 20..Bg5 21.Qxc4 Nd3 22.Ba3. (Kasparov's evaluation of this entire line with the protected pawn on c6 is "it's just a matter of time" before White wins.)
Karpov has been on both sides of this QID variation and it was deeply analyzed around his 1996 FIDE WCh match with Kamsky. His "new" continuation here was analyzed by Avrukh in the ChessBase MegaBase 2006 and given a predictable +/= after 22.Qxc4. White has a protected passer on c6 but no obvious way to make progress. The good news is he can torture Black all day, which is what Leko did. This is particularly effective in rapid and eventually Karpov blundered and lost. Not pretty, but a legal takedown and the sort of thing Karpov himself was always brilliant at. Small advantage, no counterplay for opponent, grind them on the board and the clock. Leko had made substantial progress before the blunder and 43.Qc2 would have ended things much earlier. (43..Nc5 44.Bd5)
Karpov tried to return the favor in game four in a Nimzo line with 5.a3 to vary from game two. It's another old and well-known line (the stem game I have is Euwe-Rossolimo, Hastings 1949!). White gets a small advantage in Q+R endgame (or Q+R+R as here) and tries to grind for a while. Even Karpov can't beat Leko from such dry toast (as opposed to stale nachos) and it ended on move 37.