Figure 3 Delta rank comparison of highly significant (p
< 0.001) Cladocopium and Durusdinium GO
categories. Higher delta ranks indicate upregulation and lower delta
ranks indicate downregulation.
We were concerned that the variance in the number of symbiont counts per
sample would skew the results of our functional analyses. This was
primarily due to the fact that read counts for each symbiont covaried
with the symbiont’s proportion in the host. While this effect was
formally accounted for in our analyses by including the
log10(total count) as a covariate in DESeq2 models,
doubts remained as to whether this would not create additional
artifacts. One reassuring argument is that the main gene expression
response (to codominance) was not aligned with the minor-major axis
(i.e. with variation in total counts) but rather happened in the middle
of the total counts’ scale. Still, to further test for possible residual
effects of total counts variation on our results, we resampled the
counts across samples so that each sample would have similar counts. We
reduced the counts to 30,000 per sample, 15,000 per sample, and 8,000
per sample, the latter being the lowest number of counts for a symbiont
type observed in a sample. We repeated the DESeq2 and GO_MWU analyses
for each resampled dataset and compared the GO delta ranks of each.
Reassuringly, we observed similar results at the GO delta-ranks level
across all resampling trials, indicating that our functional results
were largely unaffected by differences in coverage across samples (Fig.
S3).