Don't start your reading here, first study the 1993 paper and the home page here.
It is mainly about concepts. Sound science cannot be founded on concepts alone, but science gets in a muddle if the concepts are not properly clarified. So this can be important.
The nature of the autistic syndrome.
An expression "the autistic spectrum" has become popularised. This is regrettable because it promotes a fallacy that autism variation is all or mostly on one dimension from mild to severe. A more reasonable concept is that autism, Aspergers, and related things such as dyslexia, are all part of the autistic syndrome, that is tendency to clustering together of certain characteristics. This clustering is many-dimensional, with the number of dimensions being equal(ish?) to the number of characteristics whose expression can be affected by antiinnatia. In practice, as generally found with factor analysis, there would be a smaller number of main/most important dimensions (such as predisposition to allocating attention to the behaviors of others), some mediumly important ones, and a lot of less important ones.
The inclusion of such things as dyslexia (or what might otherwise be called language disabilities) is warranted by the finding of these being found associated in twin studies, and their symptoms falling within the same theoretical framework of suppression of innatons.
The nature of antiinnatia factors
The 1993 paper indicated that both environmental and genetic factors would be antiinnatia factors. We might add that perhaps epigenetic, genomic imprinting or mitochondrial factors could also have antiinnatia effect.
In any case, there is no reason to suppose a yes/no distinction between variables that are antiinnatia factors and those that are not. Instead, like autism itself it is a matter of degree with no clear cutoff.
And one would expect some antiinnatia factors to be "purer" (more non-specific) than others. This can arguably be seen in the social class differential graphs of my 1993 paper (graphs of table 1 added in author's reprint). The "pure" would be due to the purer antiinnatia, while the "complicated"/"organic" would be due to the less pure causing what we might call side-effects. (I discussed this variable purity of antiinnatia factors already in the 1993 paper itself.)
The numerous antiinnatia genes in combination would act as a relatively pure, general antiinatia factor, even though any one of those genes taken in isolation might have some idiosyncratic side-effects.
One would expect some grossly injurious process such as an untimely bash on the head, or a severe infective brain inflammation of viral encephalitis would have substantial side effects.
I would reckon that mercury, while a relatively pure antiinnatia factor, would be less pure than the 'portfolio' of antiinnatia genes, due to its interference with thiol-dependend enzymes, and its tendency to generate oxidative stress.
One can then go on to reasonably expect that there would be some other factors which have more or less of antiinnatia factor effect. And obviously any gene specifically related to language function is going to tend to be a factor in suppression of at least language function, and hence favour outcomes with a bit of resemblance to autism even though arguably not properly considered a true antiinnatia factor.