On type A dataset a really good discrimination level is achievable, with 44 species out of 46 identifiable via DNA barcoding. For two couples of species, the interspecific divergence is less than the optimal threshold and hence two morpho-species of filarioid nematodes are not resolved by DNA barcoding approach. These two species belong to Onchocerca and Cercopithifilaria genus. Despite O. volvulus and O. ochengi are easily identified based on morphology and host specificity, their nucleotide divergence is quite low (mean interspecific divergence 1.9%). If O. volvulus infects human patients only in Africa (originally) and South America (following the transatlantic slave trade) and O. ochengi infects only cattles, the two species could derive from a recent speciation event . This event could decrease the resolution power of DNA barcoding.
Another putative recent speciation has been proposed for two species of Cercopithifilaria genus (C. longa and C. bulboidea), showing a mean interspecific divergence of 0.2%. These parasites are restricted to two Japanese mammals (Naemorhaedus crispus and Cervus nippon), and a recent speciation event has also been hypothesized using both molecular and morphological data [27, 36]. It should be noted that these evolutionary dynamics are often difficult to identify as reported in .
Dataset B (that encompasses all the coxI sequences of filarioid nematodes available in GenBank) has been used to perform DNA barcoding and DNA taxonomy with a tree-based method. Coherently with the results obtained with dataset A, this phenetic approach shows a clear separation of MOTUs representing separated groups of morpho-species with the exception of O. volvulus-O. ochengi and C. bulboidea-C. longa. Anyway, closely related species could be characterized by a certain level of interspecific hybridization, because the reproductive isolation could not be total since the very beginning of the natural history of a species. These effects are particularly evident in mitochondrial gene trees, and represent a serious problem for DNA barcoding (at least in most metazoans, for which mitochondrial markers are widely used). Problems of this nature are likely to have occurred in the O. volvulus-O. ochengi and C. bulboidea-C. longa cases where traditional taxonomy identified good species [27, 38]. As a consequence, the usage of a tree-based method alone for species identification could be dangerous and deceptive. Moreover, in a gene tree, a 'true' species may be wrongly represented by a paraphyletic group of alleles/haplotypes, due to introgression or incomplete lineage sorting (see ). In such cases, the gene tree could appear misleading or uninformative about the species identification because of retention, and consequent random sorting, of ancestral polymorphisms.
It is important to underline that GenBank entries are not absolutely free from identification errors. The results of DNA barcoding analyses performed on coxI sequences obtained from GenBank (dataset B) do not show such type of problem. However, an example of error is represented by the entry [GenBank:AY462911] identified as Litomosoides carinii. This species parasites sciurids in Brazil  and was described by Travassos in 1916. The congeneric species Litomosoides sigmodontis was described by Chandler in 1931, parasites the murid Sigmodon hispidus, and is spread worldwide in the laboratories as model species for the studies on filarioses. For some reasons there is the tendency to confound these two clearly distinct species, and it is relatively common to observe the erroneous name L. carinii used instead of L. sigmodontis for laboratory strains of these filariods. In this context, it should be noted that basically all the results on L. sigmodontis published till now are relative to these laboratory strains established since 1970s. Here we present a molecular identification of L. sigmodontis directly collected from wild hosts. Laboratory strains and wild specimens show no molecular differences.
The five unidentified MOTUs present in dataset B encompass parasites of three avian hosts, a taxonomic group where biodiversity and distribution of filarial nematodes are underestimated. As described above, these are cases where molecular analysis can help to discover new species (DNA taxonomy).
It must be underlined that DNA taxonomy performed with simple molecular data can only suggest the presence of potential new species, whose real existence must be corroborated by integrated approaches .
Type C datasets reveal that two different markers have similar discrimination power, but if coxI shows high manageability in data handling, the marker 12S rDNA is more susceptible to the data handling (especially in gap treatment). Processing 12S rDNA type C dataset with MUSCLE and MEGA (pairwise deletion), DNA barcoding performs 6.3 time better than using MUSCLE and MEGA (complete deletion). In addition, processing 12S rDNA type C dataset with MUSCLE it is possible to obtain 0.3% of MCE (see Table 2), whilst using ClustalX, it is possible to obtain 0.4% of MCE (see Table 2). This is a quite relevant observation: the generation of a reliable alignment is a major impediment limiting the use of 12S rDNA gene sequences for barcoding purposes. For this reason, Chu et al.  have proposed to use ribosomal DNA sequences for DNA barcoding without performing an alignment, showing congruence between their approach and a tree reconstruction (based on neighbour-joining algorithm). Anyway, 12S rDNA offers practical benefits: it is much shorter compared with coxI, and therefore more likely to be readily amplified from chemically damaged (i.e. formalin fixed) or badly conserved specimens .
It is important to underline that the presence of nuclear mitochondrial pseudogenes (numts ) could introduce serious ambiguity into DNA barcoding and their presence cannot be known a priori . In nematodes, numts seems to be rare , despite their presence has been reported (see for example  were a short fragment of the mitochondrial 16S rDNA of W. bancrofti included into the nuclear LDR region is used for the screening of this parasite). In our study, the results of BLAST search, multiple alignment analyses and the quality of trace files for bidirectional processing of our sequences seems to exclude any interference caused by numts.
Our results indicate that the proposal to use the ST (10 times intraspecific variability) as described in  must be evaluated case by case. Indeed, in the case of coxI, the OT is equivalent to ST (both the thresholds generate the same value of MCE), but for 12S rDNA OT performs extremely better than ST (mean MCE relative to OT is 0.7%, mean MCE relative to ST is 50.5%). The extremely high values of MCE relative to ST are caused by the moderately high intraspecific K2P distances of the marker 12S rDNA that are enhanced of a 10 times magnitude. The data handling has also a relevant effect on the mean intraspecific divergence: MUSCLE, TREECON and considering gaps are all alternatives that enhance K2P distances.
The sampling of filarioid nematodes is clearly not exhaustive and particularly difficult, due to complications associated with their collection (i.e. recovery at necropsy in most of the cases), that requires highly skilled personnel and enduring logistic efforts all over the world. The datasets presented encompass also species for which only one sequence is available. This is a circumstance that avoid to evaluate the intraspecific variability of the marker, and consequently the discrimination power of the method decreases. However, we want to remark the importance of the datasets here reported: filarioid nematodes represent a relevant neglected, vector-borne, tropical diseases.