-------- Original Message --------
Subject: Morphometrics of small, variable specimens--embryos
Date: Tue, 29 Mar 2011 11:15:26 -0400
From: [email protected]
To: [email protected]

Hello all,

I have a second question to ask. Again, I am doing some analyses of
midgestational mouse embryos. The different groups are nearly genetically
identical. I am looking for morphometric differences due to differences in
genotype at a single or at two loci. However, the effect of genotype will
be mixed up among the variation in the data owing to heterogeneity of
developmental progress among the specimens. The level of ontogenetic
variation is large.

I generally take the procrustes coordinates (the symmetric component when
I can) and regress on centroid size. I also regress those residuals on
developmental stage by counting somites. This last set of residual
coordinates I use as the data to generate the covariance matrix.

But I still notice that the specimens tend to non randomly clump by litter
in scatter plots (like PCA). I think of this as a "litter effect," but I
imagine that this effect owes to significant residual ontogenetic
variation in the data. For example, when I regress PC1 against
developmental stage, an orderly curve results. The most and least
developmentally advanced specimens occupy the same end of the PC, with the
intermediates scattered in between. Similarly, when I regress the
Procrustes residuals against somite number, a "growth curve" results. It
has the same form as a growth curve, being horizontal at the low and high
ends and positively sloped in between. The intermediately staged embryos
dominate the linear trend. This makes sense, since the dominate trend of
PC1 is the difference between the intermediate embryos and the rest of the
sample, the least and most developmentally advanced. The shape changes at
the extremes are "not seen clearly" be the analysis. They are collapsed
together.  I think this means that the allometry of embryonic growth is,
while a large component, not the only significantly dominating component
of the ontogenetic variation in the sample. The non-linear component of
shape variation is hard to get rid of in embryos.

These patterns persist even when different developmental time spans are
considered. Breaking up a sequence into smaller sequences does not seem to
be equivalent to isolating those specimens along the linear trend from
those that are not. The same non-linear "growth curve" results after
regressing procrustes residuals against somite number for the sub sample
of intermediates alone, though it may not be as pronounced.

So my question is, given that my different groups are nearly identical
genetically, I can expect that they will be very similar looking in the
analysis. If I cannot see a separation of the genotypes along PCs or along
a discriminant function, I feel like I can argue that no morphometric
difference has been detected, despite the presence of residual ontogenetic
variation, which would serve to actually separate my groups, since my
samples are never exactly the same in developmental stage composition.
Would you agree that the data are biased toward finding a shape difference
between the two groups, biased because of remaining shape variation due
not to genotype but to developmental heterogeneity? So that a negative
result of no shape difference detected should be interpreted as
particularly robust?

Thanks.

Eric

[email protected]
University of Calgary
Faculty of Medicine









Reply via email to