Thursday, February 18, 2010

I Wuzzint Doin It Right

So my surgeon's partner is yanking on my toe with a great big frown on his face.

“There's less flexibility now than there was last time. I'm going to write you a prescription for physical therapy.”

I think I managed not to groan. It isn't that I really mind physical terrorism — I'd done well playing through the pain before the surgery, and doing the same through far less pain afterward. But we're talking about the Little Piggie that Went to Market. I am kind of known at this podiatry practice as the Patient Who Does So Well because He Always Complies; but really.

Still, Mister Compliant took the scrip in hand, showed up at physical therapy, and was lectured on how, really, a stiff big toe can completely mess up your gait and your sense of balance.

“But I have no sense of balance,” I said, “Never did.”

The doc who runs the PT practice overheard, and swooped in for a test. It's actually not uncommon for folks with a hinkey middle ear to do things like ski, mountain bike, and climb — they just (over)compensate using vision and by feeling the floor (proprioception). But we soon found out I was not one of these.

While I tried to balance on one foot, the doc observed that I wasn't actually doing badly at all — but I was maintaining my balance too much using my ankles and not enough using my toes.

“You have to dig in with the toes to maintain balance,” he said.

The man was telling me I'd never learned to stand properly. I felt like a Lolcat poster:

Standing still: Yer Not Doin It Right

Well, maybe there's more than that I haven't been doing right: hence today's entry, from Brian Duistermars and Dawnis Chow of Mark Frye's lab at UCLA. Using a wind tunnel, they tested whether fruit flies could turn in the right direction to find a scent source when an antenna on one side or the other was “occluded” with a tiny little glob of glue.

I kid you not.

This kind of research nearly always truns up a couple of surprises: for one thing, the flies are left handed: blocking the left antenna affected their ability to find the scent more than blocking the right. The Johnston's organ — the middle part of the insect antenna, which detects motion, unlike the outer part that detects smell — is necessary for proper turning into the scent, though not its detection.

But maybe the most amazing part of the experiment was that it worked at all: that flies need two antennae to track the direction of a smell in the wind, and that therefore it's possible to detect a difference of smell between antennae that are less than a millimeter apart.

To help me wrap my head around this, I actually emailed Mark Frye to ask him whether that inference was really warranted. He said yes, but threw a further monkey wrench:

“If you calculate the mean molecular concentration gradient across the fly antennae it is on the order of thermal flux (noise) ...”

So somehow the little bastards are detecting a difference against a noise background that's as loud as the signal.

Now, I've been bloviating on the dog lists for some time now about how I'm having trouble seeing how a dog's nostrils, which draw scent from about three inches apart, can detect a usable concentration gradient. I'd say the fly results pretty much put any doubt to rest.

Having said that, it occurs to me that what I was tripping over had less to do with whether they could detect a smell difference across that distance than how it would be useful. To see why that's an issue, let me explain how our understanding of what smell looks like has changed since the 1980s, when the scent theory explanation that most dog handlers have read was published.

Starting from the idea that, as wind blew scent along, it would diffuse outward and mix with the surrounding air, creating what we call a “scent plume:”

If you look at how the scent gradually decreases with downwind distance, there's a very subtle change that would be incredibly difficult to detect.

But wait; folks working with smoke in wind tunnels have found out that that old picture of a scent plume is vastly oversimplified. In reality, as the scent mixes with surrounding air, it does it turbulently. Think of how, if you stir a black coffee and then drip cream in, it doesn't mix evenly. Stronger chunks of cream remain visible for a while. Same thing with cigarette smoke, or a smokestack. The real scent plume looks like this (and note you can see a much better drawing, for which I wasn't able to get reprint rights, in this paper):

So even though I had the correct picture in my head of a filamentous scent plume, with really strong chunks of smell interspersed with increasingly large voids, I wasn't putting it all together: dogs don't have to detect subtle scent gradients — or at least not that subtle — because such gradients essentially don't exist in nature. All the pups need to do is detect the transition from no smell to strong smell and back, as it sweeps past: and for that, a different signal in each nostril actually does provide important information.

'Course, it's not as simple as turning toward the stronger scent, like you would in the “old” scent plume. If a filament has just swept by you, that would be the wrong direction. But by checking which nostril got the smell when, and what direction the wind was blowing from, you can develop a more involved strategy, something like the ones seen in insects and birds and fish — and unless my own, humble and as-yet-anecdotal observations are way off base, search-and-rescue dogs as well.

For the flies, that signal-to-noise problem still stands. But at least, finally, I think I may be on track to “doing it right.”


smartdogs said...

I just read Vincent Dethier's "To Know a Fly" (1963). Dethier wrote that he believed that flies find and follow a scent trail by flying in a random pattern until they pick up the scent, then flying upwind until they lose it. Lather, rinse, repeat until source is found.

This adds an interesting layer to the idea. Being able to triangulate a location with forward/left/right data would certainly make the process more efficient.

Do you think dogs are consciously aware of wind direction and correct for it - or do you think they use a kind of mental vector assessment algorithm?

Ken Chiacchia said...

Well, the word "conscious" is always difficult to apply to flies. But we know that they use the middle segment of their antennae to sense movement -- which would definitely pick up movement caused by the wind. I didn't go into any detail, but one of the tests these guys ran showed that this sense is important for knowing which way to turn -- which isn't merely into the wind, but into the wind and toward the center of the scent plume. It's the interaction of wind and scent sensation that allows the latter.

Moreover, we know exactly how some insects do this. In many moths, Dethier's guess is expanded and clarified -- in fact, if all you did was dash into the wind and then wait to get the scent you'd usually lose it, because if you look at the picture of the scent plume (either one), it's wider farther away, and so a dash into the wind will usually put you out of it.

There are two phases: When you encounter a scent filament, you dash into the wind. You'll almost always lose it; but at that point, you start casting back and forth across the wind, widening this cross-wind search until you encounter scent again. Then you dash into the wind once more. That makes the lather, rinse, repeat cycle keep you always moving with the scent plume, even if it takes a sharp turn away from you (which it will when the wind changes direction).

Dog handlers often see dogs hit hard on scent, dash into the wind, then cast about like they've lost it. Now I think we know why this happens, though I wonder if our interaction with them doesn't make it harder for them to repeat the process.

In conditions of low turbulence, like at night or cloudy days with steady wind, and presumably the plume will look more like the "old" picture, the moths don't do this -- they weave back and forth into the wind in successively tighter swings. This behavior may be a superposition of dashing and casting, which the moth can maintain because it's getting continuous scent. Once again, because the plume gets narrower toward the source and may kink because of wind changes, doing this rather then trying to move directly into the wind (or try to sense the subtle increase in smell intensity and move that way) may help the animals keep track of it (there's also the possibilitiy that they're using moving back and forth across the plume's strong center to keep from desensitizing to the continuous scent).

Once again, dogs show this behavior as well.

Finally, if the moths encounter continuous scent without wind, they land. Don't even try to find the source, like they're waiting for the wind to kick up again and give them the critical directional cue. On still nights, when there's no convection but also no movement of scent, dogs can have a lot of trouble finding the center of what we call a "scent pool."

smartdogs said...

Interesting. So for at least some animals (somewhat analogously to vision) scents are easier to discern when they're moving. It sounds like they could be using flow fields of scent to navigate.

How do you think this ties into Gibson's Visual Field Theory?