Opsin evolution: ancestral sequences
Reconstruction of ancestral genes -- indeed whole genomes -- is useful in a variety of contexts. Widely done in opsins to reconstruct historical spectral senstitivities, our purpose here is primarily to reduce the excessive number of available opsin sequences to representative ones that still carry all the information of the opsin class but without the idiosyncracies that might have developed in particular clades. An ancestral ciliary opsin sequence at the agnathan divergence node takes away the subsequent 500 million years of sequence divergences. Suppose the same is done for rhabdomeric opsins. Comparing these to an uncharacterized extant (contemporary) lophotrochozoan or cnidarian opsin greatly sharpens the alignment which previously involved a billion years of round trip evolutionary divergence. It further facilitates comparison of diagnostic signature residues and patches and rare genetic events such as intron gain or loss.
These considerations are very important in opsins, which are embedded in the largest and most complex of all gene families, the GPCR, because it happens that the critical events in the evolutionary origin of eye are quite old, certainly predating the Cambrian. Opsin sequences are well-conserved, less well than some like histone or ribosomal proteins but far more than the median protein, but over the time scales involved the percent identity has dropped off into the unreliable Blast twilight zone (below 30%) where a faster evolving opsin might be confused with a slower evolving GPCR not involved in photoreception. Ancestral sequences thus greatly improve the placement of opsins within their correct homology class.
However the utility really depends on the accuracy of ancestral sequence reconstruction. We hear this or that maximal likelihood or bayesian methodology "should" work -- but are these assertions really testable or just self-serving bioinformatic blather? Ancient dna has two problems -- the sequenceable component is never that ancient and the fossil that it came from is never exactly from the divergence node. On the protein side, even if collagens and other structural proteins can be sequenced from dinosaur femur, that won't help with soft-tissue membrane-bound opsins. Fossil eyes in trilobites are much studied but equally uninformative at the molecular level. So direct tests of reconstructed opsins are not imminent.
Another dimensionality to testing accuracy of reconstructed sequences involves physical construction of the gene and its expression in a contemporary host. If the gene were an enzyme, we might gain confidence looking at binding constants, catalytic efficiency, and substrate specificiity. For opsins we have covalent binding of retinal, spectral sensitivity, 7-transmembrane topology, and signaling capability. Unfortunately we don't know what the ancestral lambda max should be. These functionalities won't prove stringent enough because even the sloppiest reconstruction will get invariant and near-invariant residues correct. These may be quite adequate to produce a satisfactorily functioning opsin that bears little relationship to true ancestral sequence.
It could be argued that the ancestral residues with the most variation are the least important, so it doesn't really matter if the reconstruction gets them right. It's abundantly clear from a quick alignment that amino and carboxy termini are under very relaxed constraints even within a single orthology class, making reconstruction outside the core opsin essentially hopeless. There's also markedly less conservation in some of the extracellular and cytoplasmic connecting loops.
Thus the focus of the restoration effort lies between the hopeless and the slam-dunk invariant residues. Here we don't want to use reconstruction methods developed and vetted for cytoplasmic proteins. The rules -- such as what consititues a conservative substitution -- are very different for integral membrane proteins such as opsins where alpha helices have exterior exposed to hydrophobic rather than hydrophilic except at their cap residues. We know at the outset from the determined 3D structure of bovine RHO1 that opsins will have significant co-evolution, that is ectopic (non-adjacent) residue pairs that shift in a coordinated manner according to their own reduced alphabet despite the disparity in linear position, the best known case being the retinal-bearing lysine and its negative counterion glutamate (where the reduced alphabet contains aspartate at the counterion but not arginine or histidine at the Schiff base, though opsins are known where the counterion position has shifted). These issues are not considered in residue-by-residue and local patch reconstruction methods.
It's proven extremely difficult to determine the 3D structure of additional GPCR despite an immense research effort by the pharmaceutical industry. There's been recent progress but not specifically say on rhabdomeric opsins. The rule of thumb in crystallography of soluble proteins is that an unknown sequence can be reliably fitted to a known 3D structure if homology exeeds the 30% identity level, with the big picture retained even at much lower levels. These may apply to membrane opsins because the 7-transmembrane topology (deduced from hydrophobic periodicity plots) is a very deeply conserved feature. However subtleties of ectopic interactions may not emerge from structural fitting despite the many constraints provided by invariant residues, though residue covariance can sometimes be inferred from direct statistical study of the sequences themselves.
We'll take a heuristic approach here to ancestral sequence reconstruction because not all possible evolutionary nuances have tangible sequaleae to our central focus of disentangling very ancient gene duplications and divergences for the purpose of photoreceptor functional homologenization. It is very likely that a pragmatic hand-curational approach informed by expert opinion and tailored to the particular circumstances of opsin structure/function produces a better product than blind application of statistical web software whose appealing 'objectivity' only masks massive internal subjectivity in parameter choices and mutational processes. However with opsins the outcomes may scarcely differ in the early rounds of reconstruction, for example the opsin portfolio at lamprey node.
We can expect as ancillary benefits (1) a tenfold reduction in the number of sequences under management, (2) a small set of proxy sequences that retains all of the information (including intron and indel rare genomic events) but none of the idiosyncraticies, (3) a blast query that significantly outperforms any of its consitituent sequences on outside opsins because it has taken off 500 million years of divergence time, (4) and a sequence less likely to be fooled by non-opsin rhodopsin superfamily members or generic GPCR.
It's best procede in stages with the actual work of ancestral sequence reconstruction, as determined by phylogenetically dispersed sampling density. That is, lophotrochozoan ciliary opsins are known in too low numbers in too few species, whereas an excessive number of insect rhabdomeric and teleost cone opsins are available. After a bioinformatic push on new genomes, the resultant data set allows ancestral sequence reconstructions at common ancestor with lamprey for all classes of deuterostome opsins and at the ancestral arthropod for rhabdomeric imaging opsins. For ciliary opsins in lophtrochozoa, cnidaria, and early diverging deuterostomes, the sparse set of individual sequences must initially be retained. This will unavoidably mix filtered ancestral sequences with noisy contemporary species-level opsin sequences at the interpretative stage. That's the usual state of affairs in bioinformatics and hardly a show-stopper.
In actual ancestral opsin reconstruction, we won't use consensus sequence except as heuristic because that doesn't exploit the known gene tree and species tree. There's no real benefit to single-species consistency, even though species such as Xenopus have nearly a full set. Profile sequences (which retain the dispersion over the 20 possible amino acids at each reconstructed position) are powerful but unwieldy. Most of the benefit can there skimmed by use of reduced alphabet at positions where this is necessary. That is, most proteins have residue positions where the ancestral value is undeterminable, being a polymorphic mix in perpetuity of more or less equally acceptable alternatives (eg asparagine/glutamine waffling). Here there is an obscure technical issue of replacing special symbol output used in common reduced alphabets with letters recognizable to Blast and other tools. For example, Multalin might ';' for aromatic residue waffling (Y, W, F) whereas we will use lowercase y, and so forth.
Outgroups have an important role in arbitrating ancestral residue choice in the situation two sister clades might disagree. Here simple parsimony drives the decision. If say threonine is used in one clade of cone opsins and serine in another, while pinopsin and the others use threonine, then the ancestral residue is taken as threonine and not as reduced alphabet threonine/serine, ie the serine is taken as a clade-specific change on that stem. This extends to an 8 row decision table that covers all the combinatorial possibilities.
{decision table}
Before going there, let's take a quick overview of stratified invariance in post-lamprey ciliary opsins. That's quickly done by aligning the opsins and taking the consenus line at incrementally declining percent identity requirements, that is 100% invariant, 95%, 90%, etc. We see immediately that there is no information for significant timescales outside the opsin core, which begins not much sooner than a very deeply conserved asparagine, position 55 in the GFPIN region in human RHO1 terminology. That's not to say there's not good information earlier about evolution within cone opsins, but we're looking at a very much deeper time scale for now.
{stratified invariance in cilary opsins}
(to be continued)
