Ardi and chimps
As I have said before, I don’t know enough paleontology, so I’m just speculating now and would love to be corrected by someone with more knowledge in the field…
One thing that bothers me with the whole Ardipithecus brouhaha is its age and its relationship with chimps. Simply because I cannot completely reconcile it with the genetic estimates of speciation with chimps.
Estimating speciation times
First some observations on species divergence, genetic distance and fossils… If we consider even the simplest scenario for speciation — an allopatric speciation (and more complex scenarios will just exaggerate the problem — a population is split into two that eventually evolve into separate species.
Due to the population genomics in the population before the split, individuals in the two resulting populations will have different degrees of relatedness within and between the populations. On an individuals level, eventually everyone within a population will be more related than between populations, but because of recombination the relatedness between the populations — and eventually species — will vary along the genome and the average sequence divergence will be greater than the divergence between the populations / species.
If we date the speciation by considering genomic differences, we are thus overestimating the time back to the ancestor.
If we consider morphological changes between the two populations — as we do if we consider fossils — those will necessarily have evolved after the split. At least if we are looking at morphological traits that clearly separates the two populations and not just variation within a single species.
So for fossil evidence we will get an underestimate of the speciation time.
(Please ignore that the skull in the illustration is a gorilla, it was what I had lying around when I made the figure, but my focus is on chimps so that is the speciation I am illustrating)
Now, human-chimp sequence divergence looks like it is about 6 million years ago, but remember that this is sequence divergence and not species divergence. The speciation event must be more recent.
Through statistical models of population genetics we can estimate the speciation time from the sequence divergence time and if we do, we get estimates of the human chimp species divergence closer to 4-5 million years ago (Dutheil et al. 2009; Hobolth et al. 2007; Patterson et al. 2006).
This puts Ardi, at 4.4 million years ago, smack on top of the speciation event!
Humans, chimps and Ardipithecus
Now I’m not saying that I trust the genetics more than the paleontology. I understand it a whole lot better, but that is a different issue.
One thing that is clear, though, is that if the genetic estimates are true, then Ardipithecus cannot have evolved a whole lot away from the common ancestor of man and chimp. It actually could be the last common ancestor, just judging from the time estimates.
If we believe that Ardi shows us that many of the traits we believed were derived traits on the human line are actually ancestral traits that changed on the chimp line instead, then we could put Ardi close to the human-chimp split and reconcile the genetic and paleontological estimates for the time of the split.
If it has evolved significantly towards humans it is back to the drawing board for the genetic models.
Which is it?
I really look forward to see what John Hawks has to say about it when he gets to it in his Ardipethicus FAQ…
–
279-311=-32
October 6th, 2009 at 11:54 am
There definitely is a discrepancy there, but where does your “human-chimp sequence divergence looks like it is about 6 million years” number come from?
October 6th, 2009 at 12:36 pm
Actually, that number is a bit dodgy as well :)
It is really just what falls out of the percentage difference in genome sequence, but of course that depends on where you’re looking and more importantly an assumptions about the molecular clock and the calibration point. The clock assumption seems to be okay when looking at apes, but the calibration point is a bit problematic. The calibration point must necessarily be a sequence divergence point, but fossil evidence is needed to get it, so that means that somewhere in the calculations the fossil estimate (which as I argue is probably an underestimate of the speciation time) is equated to the sequence divergence time (which will be an overestimate of the speciation time), so there is a bit of a problem here…
October 6th, 2009 at 1:36 pm
Exactly.
Either a rate is calculated from parent-offspring pairs, in which case we assume that we can extrapolate that point estimate millions of years into the past (probably not true), or a rate is calculated from a fossil calibration of some point, in which case the error is almost certainly substantially greater than the “precise” estimates of statistical error in the algorithms.
So, that “6 million years” number is really some value that is probably older than 5 million and probably less than 10 million, for some value of probably…
October 6th, 2009 at 1:49 pm
Yup!
Still, assuming that the molecular clock assumption is fine — and it looks that way — the relative dates should be okay. Then the errors in the calibration point decreases as you get closer to the present.
If you have a calibration point 20 mya +/- 1 million years, so [19-21 mya], 25% of that divergence would be 5 mya +/- 0.25 million years [4.25-5.25 mya].
So the problems with calibration is not that serious after all.
The real problems are probably 1) if you can trust the molecular clock, 2) if you have enough data to accurately extrapolate from it, 3) if the calibration point is measuring what you think it is (speciation divergence versus species divergence) and 4) if your models are at all capturing reality.
1 and 2 I don’t see as much of a problem.
3 is what I was hinting at in the post, and there certainly is a problem there if the species used for calibration has a large effective population size.
4 is always a problem and would be my guess as to where the problem is if the genetic estimates are wildly off…
October 7th, 2009 at 11:57 am
But the calibration point is NEVER that accurate!! The human-chimp divergence is probably the best calibration point in the primates, and it could easily fall anywhere in a 5 million year range.
The date of some particular fossil _might_ be well constrained to a 1-2 million year interval, but to equate the age of a fossil to the time of origin of its species is a HUGE error.
October 7th, 2009 at 2:21 pm
I didn’t think that we had that accurate an estimate of human-chimp divergence… what do we have in the fossil record for this? I was under the impression that we had some decent calibration points further back, but to be honest I don’t really know that much about what we know from the fossil record.
Anyway, I completely agree that we cannot equate the age of a fossil with the date of a speciation event, but shouldn’t we be able to get a calibration interval from fossils?
October 7th, 2009 at 3:01 pm
Austrolopithecus, Ardipithecus, Orrorin, Sahelanthropus, etc.. It would be nice to have some chimp fossils too (maybe some of these are!), but theoretically, you only need fossil representatives of one of the descendant lineages from a split for information on the time of the split – fossils from the other lineage would be great, but are technically redundant.
As for intervals, I do not believe it it possible to infer any _useful_ upper bound from the fossil record. Lower bounds are fine. For instance, for the human-chimp calibration, we could probably set a pretty hard lower bound at 5 Ma. But what’s the upper bound? In my mind, 20 Ma is close to impossibly old, 15 Ma is highly implausible, but 10 Ma, while in my thinking unlikely, is within the range of plausibility. But I’m predisposed to a split closer to the present. Those predisposed towards an earlier split around 8-9 Ma might be willing to push that top limit out further. And the farther back in time we go, with spottier fossil records, the greater the range to any kind of supportable upper bound gets.
So getting back to the calibration example you gave, if you have a calibration point of 20 +/- 1 Ma, then that’s not that big of a deal error-wise, but if a more realistic calibration interval is 35-17 Ma, then the error IS a big problem, regardless of how perfectly accurately you can calculate the rest of the clock models, etc. If your estimated divergence is ~ 1/4 of the calibration date, then you have a date that ranges from 8.75-4.25 Ma. Add on the uncertainty in the clock estimates, etc. then you have a estimated divergence date that runs from, say, 9.5-3.5 Ma.
The precise divergence dates given in most molecular divergence date estimation papers are terribly misleading, because they’re probably not accurate as the fossil data (the foundation of the entire result) are generally not understood and misinterpreted.