Data quality and phylogeny inference
The notable incongruence among trees in the published literature clearly indicates that good resolution of and support for clades in published topologies are no proof of the reliability of the data and methods of analysis. For example, crustaceans are polyphyletic in  but paraphyletic in . Tardigrades are the sister taxon of Nematoda in , but the sister group of Onychophora and Euarthropoda in . Since, in each case, at least one of these pairwise incompatible inferences must be incorrect, it is legitimate to ask which data have a better signal-to-noise ratio. It has also been demonstrated in simulations that false clades with high support may be found in inferred trees that do not correspond to the original topology [e.g. [38, 39]]. Therefore, the fact that the Serialia clade of Giribet et al.  had a jackknife support of 95 in the parsimony analysis is not necessarily evidence for the discovery of a new and distinct phylogenetic signal.
In the case of the mollusk phylogeny discussed here, the original alignment seems to be rich in data at first glance (five genes and 108 species were used), however, the quality of these data is very difficult to estimate if one only reads the publication. Since Giribet et al. obtained only a partial sequence (about 1.2 kb of 28SrRNA) of the monoplacophoran Laevipilina antarctica, and the placement of the monoplacophoran can only be the result of similarities shared in the 28S partition of the data set, the relevant information is limited.
One may argue that gaps and missing data do not bias phylogenies, but this depends on the patterns of missing characters. Hartmann and Vision  showed by simulation studies with incomplete alignments that parsimony algorithms in particular, as applied in POY, had the lowest accuracy in finding the correct tree and the highest sensitivity for patterns of missing data. Topological disagreement with the original tree of the simulation (range from 0 showing complete agreement to 1 showing complete disagreement) approached the median value of 0.4 in alignments with 60% missing data (similar to the alignments in ), but 1.0 only in correctly aligned simulated data (very unlikely in the alignments in question).
Giribet  argued that a primary homology (positional homology in the starting alignment) is irrelevant if it does not comply with the congruence test (fit of characters to a tree), hence Giribet et al.  rely on phylogenies inferred with POY. Using POY means that criteria for primary homology are not separated from optimality criteria used for tree inference (co-optimizing topology and homology), or, in other words, that homology hypotheses are adapted with the method to optimize results, which leads to circular reasoning. Quality criteria based on empirical and topology-independent evidence, such as variability and similarity of sequence regions and fit of positions to a secondary structure model, are not considered. This is a fundamental difference from our approach: In any empirical science, data quality can and should be evaluated prior to analyses that aim at hypotheses testing. It should be stressed that evaluation of data quality and the use of alignments to infer an hypothesis can be usefully treated as two independent steps [e.g. [20, 39, 42]], although a discussion of this point is beyond the scope of the present study. For the relevance of primary homologies see [43–50]. For problems with POY see [51–53].
The key question raised by Giribet et al. – the position of Monoplacophora within Mollusca,- was based only on a fragment of the 28S rRNA gene (1280 bp). Moreover, a rather high proportion of taxa is represented by short sequences, 18% of which are represented by less than 350 bp. This raises the question of whether the available information is really sufficient to support radically new ideas such as the Serialia hypothesis. To answer this we used different tools for an a priori data exploration (= prior to traditional tree inference).
A priori analyses
Phylogenetic networks are derived from the split decomposition method originally described by Bandelt and Dress . Networks show support for groups of sequences even when they are mutually incompatible and visualize edge-lengths for signal-like patterns and contradictions [3, 21, 22, 55–59]. If networks are tree-like, one can assume that phylogenetic signal dominates in the data set . Figs. 1, 2 and 3 show long terminal branches connected by networks with short edges, thus indicating conflict and lack of distinct split-supporting nucleotide patterns for deeper nodes of the phylogeny.
In Fig. 1, some clades with distinct elongate stems are present (certain gastropod groups, cephalopods, Caudofoveata, Solenogastres), but these are not relevant to the discussion of the deeper mollusk phylogeny. Interestingly, the monoplacophoran sequence shares character states with bivalves (Figs. 1, 2, 5) and Polyplacophora (Figs. 3, 6).
Few of the groupings are separated by distinct splits, i.e. by sets of parallel edges that are longer than those of conflicting splits. This observation is congruent with the spectrum of split-supporting patterns (Fig. 4), which is obtained with a different method but nevertheless shows the same signal-to-noise relation: cephalopods conserve shared character states, but the Serialia do not appear among the 50 best supported splits. The third method we used, quartet mapping, is an entirely different tool but the results it gave are similar (Fig. 7): there is no phylogenetic structure that allows an unequivocal placement of Laevipilina.
Although all of these tools can be improved to refine their ability to identify signal-like patterns, it is evident that their use shows congruent results. The large number of conflicting patterns, which are not visible when analyses are restricted to conventional tree inference, call for caution in propagation of new hypotheses.
The Serialia in tree topologies
Our tree-reconstruction analyses of the original dataset supported a Serialia clade only with the complete alignment, as was done in the original publication . However, a priori network analyses as well as tree reconstruction with the purified alignment (after masking problematic regions with ALISCORE, Fig. 5) contradicted this hypothesis because Laevipilina antarctica is nested within the bivalves. Moreover, monophyly of Serialia is only achieved when analysing the 28S rRNA partition after exclusion of ambiguous sites and with very low Bayesian support (Fig. 6), which is unacceptable as clear phylogenetic signal. One also has to keep in mind that Bayesian support may be too optimistic 
Biologists often assume that it is sufficient to use correct substitution models for phylogenetic reconstruction. However, missing signals cannot be compensated for even with the best model. Missing signals due to signal erosion have been documented in older radiations of the metazoan tree [61–63] and data transformation using models may even increase the level of noise when deep phylogenies are studied .
A cladistic analysis of morphological characters is beyond the scope of the present study, but documented homologies can contradict hypotheses even without inferring a tree. For example, the fine structure of feathers (as evidence for homology of the plumage) would obviously not be compatible with polyphyly of feathered organisms (birds). We therefore briefly review morphological characters as additional sources of information. For discussion of homology of morphological characters we refer to the malacological literature (see below).
Giribet et al.  stated that "the disparity of mollusk body plans is so great that it is quite difficult to find a single trait shared by all seven classes of mollusks" and that "preconceived ideas on mollusk relationships ...rely almost entirely on shell morphology". This argument implies that morphology is useless for establishing the monophyly of mollusks, yet many unique traits are shared by all larger mollusk taxa in their basic pattern. Some of these attributes were mentioned in the introduction and additional examples are discussed below. The monophyly of Mollusca is supported by homology of mantle and a mantle cavity containing at least one pair of gills, the ventral "foot", the presence of a dorsal heart with paired auricles, a bilateral nervous system with two pairs of conspicuous longitudinal nerve cords with major commissures only anteriorly, and the presence of rhogocytes. The radula is a unique mollusk character and its secondary absence in particle- or filter-feeding species is easily explained  (see latest review of Haszprunar et al. ).
The idea that Monoplacophora might be derived from Polyplacophora conflicts with the Conchifera hypothesis, which places extant Monoplacophora (Neopilinida) at the base of the Conchifera, whereas Polyplacophora are usually regarded as the sister group of Conchifera [6, 64–68]. In the few morphologically based phylogenetic analyses available to date, this position is confirmed or at least not contradicted [6, 68].
Giribet et al.  contend that serially repeated gills and the eight sets of dorsoventral retractor muscles are a synapomorphy of Monoplacophora + Polyplacophora. They do not discuss the characters shared by Conchifera, although these cannot be ignored: Conchifera (comprising Monoplacophora, Gastropoda, Cephalopoda, Bivalvia and Scaphopoda) have a massive shell that protects the dorsal visceral mass and covers a larger dorsal area than the serial shells of Polyplacophora. In addition, the outer mantle surface protected only by cuticle and single calcareous spiculae in non-conchiferans is covered by the single massive shell in conchiferans. The suprarectal commissure of the nervous system seen in non-conchiferans is replaced by a subrectal commissure in Conchifera. Conchifera are further characterized by typical statocysts near the pedal ganglion. Haszprunar  names another synapomorphy of the Conchifera, namely, the cilia with a single ciliary root, rather than two ciliary rootlets typical of metazoans, including Solenogastres, Caudofoveata and Polyplacophora.
The presence of eight dorsoventral retractor muscles in extant Monoplacophora, regarded by Giribet et al. as a synapomorphy shared with Polyplacophora, can be interpreted as a plesiomorphic homology inherited from the common ancestor of all Conchifera and Polyplacophora (the Testaria or Eumollusca hypothesis [e.g. [65, 69, 70]]). Some fossil bivalves with eight pairs are known [71, 72] and less derived bivalves still show six pedal retractor muscles. The Polyplacophora retain a mobile dorsal exoskeleton composed of eight small shells. According to this view, after fusion of these single shells in the stemline of Conchifera, the eight pairs of rectractor muscles were retained in the most primitive Conchifera, of which the Neopilinida survive. Giribet et al.  did not mention that the dorsoventral muscles of Neopilina already show signs of simplification: Polyplacophora have two pairs of muscles arranged in tandem for each dorsal shell, whereas in the well studied Neopilina these are fused – a probable consequence of the simpler shell configuration. It can be assumed that this simplification continued during conchiferan evolution and led to muscle reductions in higher evolved Conchifera [6, 65, 66, 73].
This hypothesis resolves the conflict between homologies shared by Monoplacophora and Polyplacophora on one hand, and homologies present only in Conchifera on the other hand. The first set of homologies consists of plesiomorphies, whereas the second consists of apomorphies of Conchifera. Some similarities of the pharyngeal area of Monoplacophora to the pharynx of chitons as described by Wingstrand  are probably homologous (glandular epithelium of subradular sac, similarity of salivary glands, radular vesicles), but it is not clear if these are plesiomorphies or apomorphies and which variations occur in other Conchifera. Giribet et al.  mentioned traces of seriality that are seen in different mollusk taxa but did not offer an explanation for this observation. In the traditional understanding of mollusk phylogeny these characters are not enigmatic; some are independent of the shell-adductor system (spicules on caudofoveate larvae) and offer no motive to search for segmentation, whereas others (gills and nephridia in primitive cephalopods) can be interpreted as remnants of the seriality inferred for the last common ancestor of Polyplacophora and Conchifera.
Giribet et al  pointed out that shell formation of Neopilinida differs from other Conchifera and concluded that this indicates non-homology. Haszprunar and Schäfer  indicated that the foliated layer of nacre is not homologous to the nacre of gastropods. However, these authors also referred to Poulicek and Jeuniaux , who considered the microstructure and the composition of the chitinous organic matrix of the neopilinid shell to be more similar to other conchiferans than to polyplacophorans.