The best contact to use is the one that represents the real life conditions the best. You mentioned bonded contact; you would only use this if your wheel were welded/glued/adhered to the ground this contact would be accurate. As you are analyzing a wheel just being pressed down on the ground the more accurate contact set to use is going to be No Penetration as this will not cause the model to be artificially stiff in the region of contact the way a bonded contact will be.
non-penetration contact will give you more accurate results, but you need to be sure your model is properly constrained under gravity and then add forces in my experience. The bonded method is like having every single part of contacting surface areas welded together. Where as in my experiences is better to bond the edges, simulating welds, using a non-penetration contact and solid bodies throughout. Of course finer mesh means better results, and you can also simulate the contact point using split lines to isolate the force in just the contact point. I have used this trick in the past to make a "point load". The best part is you can draw shapes like a shoe and make that your split line to get a simulation of someone stepping on your model. Hope this clears things up for you.
Just to clarify. If you set two pieces of metal on edge and weld them they are only "bonded" on the edge of the weld, not the entire contact surface area. Therefor whenever possible avoid bonded simulation.
BONDED ACTS LIKE ONE PART
NO-PENETRATION IS TWO PARTS (MORE REALISTIC)