HELP: Q about an ATR Vowel Harmony System

[I apologize if this message got sent twice.   I've been having some 
AOL-related trouble.] 

Hi all,

I'm revamping one of my old languages, Gweydr, and turning it into a, well, 
good language.  Since I had originally intended it to be a vowel harmony 
language (before I actually knew what this was), I decided to do two things I'd 
wanted to do with vowel  harmony: (1) Create a Finno-Ugric-type vowel harmony 
system, *and* (2) create an [ATR] vowel harmony system.

The Finnish Vowel Harmony part is no problem at all.  Basically, there is a 
normal five vowel system, and then all the back vowels are fronted in certain 
contexts.  So, your normal system is:

Front: i e
Back: u o A

Then the fronted versions are:

Front: i e
Fronted: y 2 &

And, in the above, the vowels /i/ and /e/ would be opaque, though would be 
treated as [-back] if they came word initially.  This much is clear.  I'm having 
some problems with the ATR system, though.

Looking at everything but the low vowels, the [-ATR] versions of the 
ordinarily [+ATR] vowels would be:

Front: I E
Fronted: Y 9
Back: U O

That's pretty simple, too.  Then, however, I've run into this problem with 
the low vowels.  Low vowels are pretty much always transcribed as [-ATR].  This 
in itself isn't a problem, as the two could be opaque to [ATR] harmony.  
*However*, when they came word initially, this means they would cause [-ATR] 
harmony to spread.  So, if you had:

k&nelymi

It would surface as:

k&nElYmI

I don't like this, and I don't want it.  I want low vowels to *not* spread 
[ATR]-ness at *all*.  What this would mean is that all the vowels would go to 
their default state of [+ATR] for [-low] vowels, and [-ATR] for [+low] vowels.  
This, however, is not something I've ever heard of, so I didn't know if it was 
realistic.  Consider this option 1 (and, if you could, I'd like you to 
evaluate the various options at the end, or propose an option of your own).

(Option 2) Well, this, of course, would be to leave them as [-ATR] and have 
them spread this  harmony.  This is what I don't want.  Is this option, 
however, so realistic that it can't be avoided?

(Option 3) The option that immediately came to me, as a native English 
speaker, was to specify that /&/ and /A/ are [+ATR], and that the [-ATR] version of 
*both* vowels is /@/.  Then, however, I immediately remembered an [ATR] vowel 
harmony system from phonology last quarter where there was a plus and minus 
ATR version of a low vowel.   However, the [-ATR] version was /A/ and the [+ATR] 
version was /@/--exactly backwards from what I was thinking (and what I'd 
want).  What's more, when I asked the professor (since my intuition was that the 
opposite notion would be common), she said it was very common for /@/ to be 
the [+ATR] verison of /A/.  The reason for this would probably be that /@/ is 
higher than /A/, and thus can be [+ATR] or [-ATR] (under the assumption that no 
low vowel can be be [+ATR], which is widely believed).

(Option 4) This is the crazy option, so be prepared.  It's been said that in 
some Midwestern varieties of English that the /&/ in "can" or "that" is 
actually [+ATR], but is *not* either [E] or [e], phonetically--it really is [&].  If 
this is true, I don't see why /A/ couldn't have a [+ATR] version as well.  
And, if this were the case, why not say that the vowels /&/ and /A/ are 
*default* [+ATR]?  The you could have [-ATR] versions of the vowels, and I came up 
with /a/ for /&/ (this is the IPA symbol that actually looks like "a", and is 
lower than /&/--it's the lowest, frontest vowel you get, and certainly [-ATR]), 
and /6/ for /A/ (/6/ in X-SAMPA is the lowest central vowel there is, which is 
an "a" written like this: "a", but flipped upside-down.  It would also would 
be default [-ATR]).  This system would be weird, no doubt, but workable, I 
think.

All right, those are the four options.  Any thoughts?

-David