Is the microbial tree of life verificationist?

The field of microbial phylogenetics has questioned the feasibility of using a tree-like structure to the describe microbial evolution. This debate centres on two main points. First, because microorganisms are able to transfer genes from one to another in zero generations (horizontal gene transfer, or HGT), the use of molecular characters to perform phylogenetic analyses will yield an erroneous topology and HGT clearly makes the evolution of microorganisms non tree-like. Second, the use of concatenated gene sequences in a total evidence approach to phylogenetic systematics is a verificationist endeavour, the aim of which is to bolster support. However, the goal of the total evidence approach to phylogenetic research is based in the idea of increasing explanatory power over background knowledge through test and corroboration, rather than to bolster support for nodes in a tree. In this context, the testing of phylogenetic data is a falsificationist endeavour that includes the possibility of not rejecting the null hypothesis that there is no tree-like structure in molecular phylogenetic data. We discuss several tests that aim to test rigorously the hypothesis that a tree of life exists for microorganisms. We also discuss the philosophical ramifications of background knowledge and corroboration in microbial studies that need to be considered when suggesting that HGT confounds the tree of life. The Willi Hennig Society 2009. Recent attempts to fell the tree of life Recently, whole genome prokaryotic and tree of life (ToL) phylogenetics has been viewed as a ‘‘useless and inscrutable endeavor’’ because of the prevalence of horizontal gene transfer (HGT) (Doolittle and Bapteste, 2007; Boucher and Bapteste, 2009; Doolittle and Zhaxybayeva, 2009; Lopez and Bapteste, 2009). [The above phrase comes from a review of one of the current authors papers that reconstructs the tree of life using a concatenated amino acid matrix. The reviewer was a microbialist.] In an earlier paper, Bapteste et al. (2008) articulated this idea, claiming: ‘‘it is safer to assume a comb-like topology of life’’. This ‘‘safe’’ topology would be a polytomy, reflecting the apparent lack of resolution at deep nodes in the ToL due to the inability (because of HGT) of the data at hand to resolve a bifurcating relationship. These and other publications (Creevey et al., 2004; Dagan and Martin, 2007) have led to the attitude amongst many microbiologists that the tree of life has been felled, and has been called the ‘‘microbialist’’ viewpoint by (Dagan and Martin, 2007). In promulgating this argument, several authors have made the claim that phylogenetic tree building is verificationist. Specifically, Dagan and Martin (2007) suggest that the assumption of a tree-like structure of the relationships of life is a positivist philosophical construct. Bapteste and Boucher (2009) and Bucknam et al. (2006) also erect arguments that lead to the conclusion that the construction of the ToL is verificationist. Although Dagan and Martin (2007) do not offer a philosophical argument for this viewpoint, Bucknam et al. (2006) do. With specific reference to Carnap (see below), they suggest that ‘‘total evidence methods have a defined epistemological background: they belong to the verificationist toolkit’’ (Bucknam et al., 2006, p. 7; emphasis added). The alternative approach is the falsificationist approach as articulated by Popper in several publications. We agree with Bucknam et al. (2006) that the falsificationist approach is the most appropriate one for phylogenetics. However, we take issue with their interpretation of what is and what is not falsificationist in phylogenetic analysis. *Corresponding author: E-mail address: [email protected] The Willi Hennig Society 2009 Cladistics 10.1111/j.1096-0031.2009.00288.x Cladistics 26 (2010) 195–201 As an example of their notion of falsificationism, consider the following statement from Bucknam et al. (2006, p. 7): ... instead of aiming to confirm clades via a progressive increase in their bootstrap support when more data is added, a falsificationist would more likely be interested first by the rejection of impossible groupings, underlining what can not be a natural group. We agree with their assessment of the search for progressive increases of support as being verificationist. Dagan andMartin (2007, p 84) alsomake this observation by stating that recent studies that use concatenation approaches to construct aToL (e.g. Ciccarelli et al., 2006) work ‘‘by accumulating data that are compatible with the null hypothesis of a common treewithout having a chance to refute it’’, an approach that is indeed verificationist, if the initial null hypothesis is that a ToL exists. Bucknam et al. (2006) developed two interesting new approaches [progressive reconstruction analysis (PRA) and concaterpillar analysis (CA)] to reassess the validity of concatenation in ToL inference. However, some, such as Bapteste et al. (2008), see the null hypothesis of phylogenetic systematics as a lack of phylogenetic structure in the data, or a bush (a safe phylogeny) which, if not rejected, could be explained as all organisms being studied being genetically identical, that all organisms are genetically distinct, or that reticulate dissemination of genetic information is the rule. Notwithstanding, Dagan and Martin (2007) interpretation that the goal of total evidence or concatenation is to obtain higher and higher bootstrap support is mistaken. We suggest that the validity of a tree-like structure to represent evolutionary history cannot be tested by any method that purports to test whether the preponderance of data supports this tree over that tree. This leads us to the observation that the microbialist philosophical argument ignores one of the great principles of the philosophy of science as outlined by Popper (2002), namely corroboration. We also suggest that background knowledge as articulated by Popper is an important aspect of phylogenetic science that has not been discussed in any detail by authors addressing theHGTproblem.Herewe attempt to clarify, based on four approaches, some of the philosophical problems brought on by attempts to fell the ToL. First, we ask, ‘‘can we test for a ToL using tree thinking?’’ Next we attempt to clarify what concatenation does in a philosophical context and third we examine the background knowledge inherent in ToL studies. Finally, we examine the principle of corroboration as a guiding light to a falsificationist, bifurcating ToL. Three phylogenetic tests for the existence of a vertical ToL We agree that to deny the existence of HGT and its potential effects on vertical evolution would be frivolous. However, although some studies have demonstrated large amounts of HGT amongst organisms, to date there have been very few rigorous attempts to test the hypothesis that a vertical ToL (vToL) exists. The jump from the existence of a large amount of HGT to the lack of a vToL is usually made in an anecdotal fashion. We suggest that a proper assessment of the futility of a vToL should include tests of the following hypotheses that will address the reason for the current lack of support at the base of the ToL. First, a rigorous test of whether or not a supermatrix composed of information using as much sequence as possible (regardless of HGT) will give a resolved tree needs to be accomplished. Second, if a resolved tree is produced, a rigorous test of whether the nodes in the tree are robust lends credence to the phylogeny; however, see Kluge (1998) for arguments against the utility of support measures in phylogentic analysis, an especially interesting concept when considering noisy data (like the ToL). Any analysis that passes these first two tests could be said to have uncovered a most corroborated signal (sensu Lienau et al., unpublished data) for the massively concatenated supermatrix of all available sequence data. The revelation behind this signal would be antecedent to the most important test, as passing the first two tests as discussed above does not in and of itself test the existence of a vToL. A final and most rigorous test of whether or not the relationships implied by the resultant tree are representative of organismal branching patterns is needed. These tests are meant to assess whether the polytomy at the base of the ToL is hard (adding data gives no resolution) or soft (adding data gives resolution), and if resolution is achieved (whether it is believably supported). For the first test, if a poorly resolved tree results from such an analysis, then we would have to conclude, as have Bapteste et al. (2006), that large amounts of sequence information add little to a ToL hypothesis and we would consider the hard polytomy hypothesis. In this case, we would be forced to admit to the approach as misleading and not useful until more data could be added. But when is enough, enough? That is, it is interesting and timely, given the amount of molecular data that is just becoming available to microbial phylogeneticists, to speculate just what criteria would suffice to test the hard polytomy hypothesis rigorously enough for it to be accepted as well corroborated. Furthermore, we have to consider what a hard polytomy hypothesis would mean in a biological sense for different group of microbes. First, gathering and testing of data never ends, which is why a scientist can say, ‘‘I ain t often right, but I ve never been wrong’’ (from Scarlet Begonias by Hunter and Goscia, 1974; Dodd, 2007). However, given a static taxonomic sample, for genome sequences one can make the reasonable claim that fully sequenced genomes 196 E.K. Lienau and R. DeSalle / Cladistics 26 (2010) 195–201