For some ribosomal RNA sequences used to infer the deepest phylogenies, they actually do use the DNA sequences(as they would be directly complementary to RNA), because even at these ages ribosomal RNA sequences are decently well conserved. For example, extant bacterial 23s ribosomal RNA sequences have roughly 55-60% sequence identity to extant archaeal ribosomal sequences.
In an attempt to try to show some evidence for universal common descent, and inspired by a blogpost by evograd, almost a year ago I inferred a bacterial, and an archaeal tree using 23s ribosomal RNA sequences. What I wanted to try to show is that, without using outgroup rooting(thus forcing the result), internal nodes in the two trees(inferred independently of each other, using midpoint rooting) still exhibit ancestral convergence. The idea is that if bacteria and archaea really did evolve from a common ancestor, as we go deeper back in time towards the approximate root of each tree, the sequences should become more similar, both with respect to sequence identity, and by alingment scores.
That does indeed appear to be the case. I never finished aligning up all my sequences. I don’t do this professionally so did a lot of copy-pasting into browser windows using online tools to infer trees and get alignment scores, and I still have hundreds of alignments to do, so forgive me if I won’t finish this up anytime soon. But here is a screenshot of a preliminary result:
Alignment scores go from lower=more green, to higher=more red.
Here are the bacterial and archaeal trees with labeled nodes.
The root nodes for the trees are N1 for both. As you can see, alignment scores between nodes in each tree become progressively more red in color as we move closer to the roots of each tree. Interestingly, the particular data set I had collected seems to imply that the true root of my bacterial RNA tree lies closer to node N2(it gets ever so slightly better alignment scores to archaeal sequences), than the midpoint root node N1.