I did not know that about MacDonald. Thanks for pointing it out. Do you agree, though, that such examples are rare? The original premise is that in math it appears self taught is not really a viable route for all but a very small few.

If I'm not mistaken, Paul Lockhart was also self-taught. If I recall correctly, he met Ernst Strauss at some point, who introduced him to Erdos who then somehow managed to get him admitted to the graduate program at UCLA (he dropped out of undergrad). He had published before being accepted to graduate school.