In an optimal strategy, you don't need to make any presumptions of your opponent's strategy. You watch what they do, and react to that. If you react to what you predict your opponents will do, then they can just not do the thing.
However, people don't play optimally, therefore there is meta. The fact that there is meta is proof that people aren't playing optimally.
The most obvious kind of meta is when newbies think Thief is really devastating, so the best strategy is to prepare for some Thieves. However, I suspect that the metagame online is more subtle, since the players are better? I dunno, I play offline only.
If there is no dominant strategy, the optimal strategy is dependent on the plays of the other player(s). This is the core of game theory. Meta is trying to optimize if there is no dominant strategy: either beating a large as possible set of strategies (in ccg's, as you can't change your starting deck you just need to build a deck that beats the largest set of predicted strategies. Meta comes in when you have to predict the probability of a certain deck being played: One includes TeCH to try and beat the strategies with largest probabilites), or trying to adapt to your opponents strategy.
Adapting to your opponents strategy is all what dominion is about. If your opponent is playing a rush, you want your engine to go off earlier than if he plays big money. If Tunnel and Militia are on the board (and no other Tunnel enablers), you don't want to buy Tunnel except if your opponent buys 5 Militia, for instance. This is just adapting the strategy set to what the other player buys.
Anyway, I guess there is meta, it predominantly comes from the approach of greening. There is some optimal way to green, but it is not easy to understand it. The meta evolution made greening an ever-changing process. I don't know if there are other large meta-changes. Maybe the acceptance of JOAT and Rebuild after time (and simulations) as good cards can considered to be changing the metagame. But also the inclusion of new sets changes the approach of some cards. I take Upgrade as an obvious example. At first, everybody accepted Upgrade as a bona-fide copper-trasher if it wasn't used for anything better and the copper wasn't completely necessary. But then came Poor House, a card costing 1. It suddenly changes the approach of the card when that was on the board: the copper-trashing ability then is posed with the obvious problem of needing to gain a Poor House.
The small problem is that because of the 'we only use 10 cards'-approach, these additions can change the game, but don't necessary do it. Still, having the cards around in the selection procedure does change the value of some cards. This is most obvious with Alchemy. Most Alchemy cardss are better when fewer sets are included in the selection procedure.
So, there is meta in both ways: The decision making of the players given a certain set is meta, as well as the change in used cards in the selection procedure. It is not as huge as in a CCG like Hearthstone or Magic, or something like Pokemon battling. Also contributing to that is that there is no banlist or something: In Pokemon we consider Mewtwo to be way overpowered battling in Ubers, whilst Tauros is declined to the depth of Never Used. Both metagames with complete different banlists. Dominion doesn't consider these banlists, or rather, has a randomized banlist consisting of all of the kingdom cards except 10 of them. If one considers that, there are an enormous amount of possible metagames. Therefore, we can't analyse them (well, we can, but we don't consider all individual kingdoms obviously), and subsequently don't feel like considering a metagame analyzing kingdoms at all (or at least I do). But I guess they each include their own meta, strictly speaking.