J. Peter Gogarten, Department of Molecular and Cell Biology, University of Connecticut, USA.
The manuscript by Aaron Vogan and Paul Higgs describes a simulation of genome evolution with the aim to characterize the processes that occurred at the emergence of coherent genomes. The simulations illuminate a valid point concerning the Darwinian threshold: accurate replication removes the necessity for high levels of gene transfer. While I agree with this general conclusion, I have a few questions, comments and suggestions. Some of these might seem unfair to invoke in face of the straightforward and limited simulations; however, the authors themselves take a broad view in discussing their findings.
The performed simulations describe a sympatric process, corresponding to sympatric speciation, if viewed as speciation. I do not know how frequent this process is in extant microorganisms, but allopatric speciation, driven by geographical division of a population, appears to be frequent among extant organisms. Its requirement is that gene flow across the geographical barrier is low compared to within group transfer. Geographical isolation could split a population of communal entities (precells, progenotes, or hypercycles) before crossing the Darwinian threshold. The splitting of a population into separate lineages through an allopatric process depends on barriers to gene flow between the separate lineages, and not on the question whether distinct genome genealogies exist within a single lineage.
Author response: In a more complex model, we might have considered geographical barriers. However, the simple model we studied already shows the Darwinian threshold effect that we are interested in, so additional complexities are not required at this point.
As illustrated by the above comment, some of my uneasiness in reading the article stems from the fact the authors do not always distinguish between within species/population gene transfer, that often results in homologous recombination, and gene transfer between divergent organisms often resulting in non homologous recombination. I readily admit that the distinction between these processes is not always clear - in some instances gene transfer between species can result in homologous recombination and the transferred gene, although very divergent in sequence, can be functionally equivalent to alleles in a population [36]. The simulations by Vogan and Higgs treat gene transfer as equivalent to non-homologous recombination (the transferred gene is added to the recipient genome); however, in the discussion they also include processes that result in homologous recombination, the latter process is equivalent to sex in eukaryotes in that it allows organisms to combine beneficial mutations that occurred in different genealogies [37], and prevents Muller's ratchet [38] from operating. The occurrence of homologous recombination in bacterial and archaeal species has been amply documented (see [39] for review). While the benefits and costs of sex continue to be debated, the widespread occurrence of sex suggests that the benefits are substantial. If within population transfers resulting in homologous recombination are included under HGT, then the generalized statement "if the accuracy of genome replication is high, then HGT is unfavourable" does not appear warranted. It might be true for the performed simulations, but not as a general statement for biology.
Author response: We hope it is clear that we are modelling gene gain and loss, which is non-homologous, and we are not modelling homologous recombination because we are not considering variations between alleles for the same gene. The points above are relevant at the level of evolution of a single gene. Our model is working at the whole-genome level. The statement in the abstract is now changed to "if the gene loss rate is lower, then HGT is, on average, unfavourable".
Evolution constitutes a strange loop where the process by which evolution occurs can be subject to selection pressure. Thus parameters such as mutation and gene transfer rates might be subject to selection and change over evolutionary time. It is tempting to speculate that the observation that gene transfer is biased towards transfer between closely related organisms [36, 40, 41] might reflect the fact that transfer between more divergent organisms is more frequently detrimental to the recipient, possibly because more molecular parasites might be acquired through these transfers. However, this bias might also reflect the simple fact that more mechanisms exist for transfer between closely related organisms, and that the transferred gene has a better chance of being integrated into the recipients regulatory and metabolic networks if it originated from an organisms with similar sequence biases and regulatory signals. The question "if HGT is on average beneficial to the organism" needs to be considered for the different types of gene transfer (see the previous paragraph). Another way the question needs to be differentiated is the meaning of "beneficial to the organism". Sometime ago Jeff Townsend and I proposed the neutral or nearly neutral theory of HGT [42] in analogy to Kimura's neutral theory. According to our proposal, most transferred genes found in a population are neutral or nearly neutral to the recipient. This does not mean that all transfers are neutral, many, or even most transfers may be detrimental to the recipient, but because they lower the fitness of the recipient, these genes are quickly eliminated from the population. Neutral, or nearly neutral transferred genes may be fixed in the population through drift, but as Lawrence and Ochman found, these genes often do not persist in the recipient lineage over long periods of time [43]. In addition, most of the genes identified as acquired through HGT in a genome have not been fixed in the population, and most of them never will be. The hypothesis that most of the transferred genes observed in a population are neutral or nearly neutral does not say that most transferred genes are neutral, most of the transferred genes might be detrimental, but because they are subject to purifying selection they are not frequently encountered in the population.
Author response: The question of whether transfer occurs more frequently between closely related species is very interesting, and it is easy to envisage more complex models where the either the attempt rate or the successful incorporation rate of HGT is larger for more similar genomes. Also, we agree that many newly gained genes are not beneficial and will not spread widely. This point is already discussed in our conclusion, and we already pointed out that it would be useful to extend our model by including a category of non-functional genes. Addition of non-functional genes would allow the pan-genome to become much larger, without significantly increasing the size of individual genomes. However, addition of non-functional genes would also make HGT less favourable than the present case. Therefore, once again, this would not change the central result of our model that there is selection to reduce the rate of HGT.
The benefit or detriment of gene transfer to a population is a different matter. The authors claim that their simulations provide evidence against a statement in [1] that barriers to HGT from divergent organisms might not be of significant selective advantage to the group. Or in other words, according to [1] preventing rare transfers, even if most were detrimental to the recipient, would not create a significant selective advantage for the group. An example might illustrate my reasoning (I don't claim to speak for my co-authors). Assume a population of 109 organisms; 200 of these acquire a gene from a divergent organisms. In half of the cases, the recipient is less fit, and these genes will be rapidly eliminated from the population, in 99 cases the transferred genes might be nearly neutral and most of them will be eliminated from the population due to drift. Even if a nearly neutral gene drifts to fixation in the population, as long as it is not a molecular parasite with mechanisms for its own long term survival, the gene will be subject to the deletion bias prevailing in bacterial genomes [44], and the nearly neutral gene will be eliminated on the long run. The rare advantageous gene might open up new resources, be rapidly fixed in the population, and/or spawn divergence into a new ecological niche. Even though in the example half of the acquired genes are detrimental, the fitness of the population is hardly affected negatively. The difference between this reasoning [1] and the simulations by Vogan and Higgs could be that in the simulation gene transfer between organisms is not rare, rather in the simulations transferred genes accumulate in the genomes of the population decreasing the average fitness of its members. In real life the ballooning of genome content is prevented by the deletion bias, even though the adaptive value of the deletion bias might be the inactivation of molecular parasites and not in streamlining the genome.
Thus I still think that if gene acquisition from divergent species is rare, then effective selection for barriers to transfer would not occur. To the contrary, the rare transfers that provide a selective advantage, and that are driven to fixation in the existing population, or that lead to the founding of a new population in a different niche conceivably could select for mechanism that enable HGT events between divergent organisms. Some of these HGTs have changed the face of this planet, e.g., the two photosystems working in series in oxygen producing photosynthesis [45, 46], or the pathway that allows methane production from acetate in Methanosarcina [47]. It is tempting to speculate that the creation of new metabolic pathways and capabilities was a driving force to select for and maintain HGT mechanism. While the idea of gene sharing mechanism providing an adaptive advantage is philosophically appealing to me, I do not know of any evidence that the transfer mechanisms between divergent organisms are under purifying selection due to the selective advantage provided to the recipient (this is different for within group transfer, where the existence of Gene Transfer Agents whose sequences appear to be under purifying selection suggests selection for the gene transfer machinery [48]). Furthermore, the transfer of genes beneficial to the recipient (antibiotic or heavy metal resistance genes) often can be explained from the selfish gene point of view rather than assuming group selection in favour of HGT [49]. It appears possible that these transfer events, however big their impact on the evolution of our biosphere, occurred as by products of other processes, such as uptake of DNA as a nutrient [27] as detailed by Vogan and Higgs, or as a by-product of the propagation of molecular parasites and viruses.
Author response: The argument above seems to rest on whether HGT is 'rare' or 'common'. If it is as rare as in the example above, there will be almost no selection either way on the HGT rate because most individuals will not experience HGT. But Woese's proposal was that, in the early stages, HGT was common. We have shown that this makes sense if vertical inheritance is inaccurate, because high HGT rates are then favoured. But our model shows that when vertical inheritance becomes more accurate, there will be selection to reduce HGT. This would lead to the state described in this example, where HGT is too rare to be selected against. Thus, if we accept Woese's view that HGT was initially very common, there must already have been selection against it to reduce it to the levels that we see now.
We agree that rare advantageous cases of HGT can be important in evolutionary history, as in the examples that the reviewer gives. We already stated in our conclusions that "the few beneficial cases will have important consequences for adaptation of the organism". We also already said that the remaining low level of HGT could be due to selfish genes (molecular parasites and viruses) or selection for uptake of DNA as a nutrient. So the reviewer seems to be largely in agreement with us.
Vogan and Higgs see the high level of inconsistency at the root of the tree of life as an indication for high levels of gene transfer at the time of the Last Universal Common Ancestor (LUCA). I do not agree. At the time of LUCA the use of the 20 canonical amino acids found in modern proteins, the ribosome, signal recognition particles, and chemisosmotic coupling had already been established [50]. While not all of the molecules present in all extant organisms were present in the same organismal common ancestor [51], the phylogenies of the listed traits are congruent and rooting them using ancient paralogs (see [52] for a review) or the echo from the assembly of the genetic code [53] consistently places the root between the archaea and bacteria, with the eukaryotic nucleocytoplasm being the sister to the archaea (which in the different analyses are recovered either as mono- or paraphyletic). In contrast to the statement in the manuscript by Vogan and Higgs, the deepest split in the tree of life is recovered with a surprising consistency, even analyses based on genome content [54] and midpoint rooting of the rather frequently transferred aminoacyl tRNA synthetases [55] place the root of the genome network in the same place. Together these observations (the inferred characteristics at the root of the tree and the congruence of phylogenetic patterns) suggest that LUCA had already crossed the Darwinian threshold.
Author response: Our paper presents a theoretical model only and therefore we cannot add to the debate on the precise nature of the LUCA. For our purposes, it does not matter whether the Darwinian threshold occurred at the time of the LUCA or prior to this.