We found an upper limit for intraspecific sequence divergences in a wide range of species of the diverse butterfly family Lycaenidae, but no lower limit for interspecific divergences and thus no barcoding gap. This result is especially well documented in the comprehensively sampled genus Agrodiaetus (114 of ca 130 recognized species sequenced) while the smaller overlap in Arhopala can be attributed to the lower percentage of species sampled (33 of more than 200 species). The choice of species by  was to maximize coverage of divergent clades while minimizing the total number of species which is a common and sensible approach for phylogenetic studies, but undermines the power of such sequence data as critical tests for the barcoding approach. The general level of sequence divergence is not exceptionally low in Lycaenidae compared to other Lepidoptera. The mean congeneric interspecific sequence divergence of 5.1% in Lycaenidae (5.1% in Agrodiaetus and 5.0% in the other genera) was only slightly lower than the mean value of 6.6% found by  in various families of Lepidoptera.
We thus confirm the results of Meyer & Paulay  and Meier et al. . Our results also agree with those from a recent study in the Neotropical butterfly subfamily Ithomiinae (Nymphalidae)  which records highly variable levels of divergence in mtDNA (COI &COII) between taxa of the same rank. Our results however fail to agree with those of Barrett & Hebert  on arachnids. In that study the mean percent sequence divergence between congeneric species was 16.4% (SE = 0.13) and thus three times higher than in our study while the divergence among conspecific individuals was only slightly higher with 1.4% (SE = 0.16). The contradiction between our study and theirs can be explained by the very incomplete and sparse taxon sampling in their data set amounting to just 1% of the species contained within the families. We conclude that the reported existence of a barcode gap in arachnids appears to be an artifact based on insufficient sampling across taxa.
Despite these difficulties, species identification of unidentified samples with the help of barcodes is entirely possible. The NJ tree profile approach which does not rely on a barcode gap enabled the correct assignment of many sequences, and other methods (e.g. applying population genetic approaches) might further increase the success rate. However, 17% of test sequences could still not be identified correctly, even in some sympatric species pairs which clearly differ in phenotype and chromosome number (e.g. Agrodiaetus ainsae [n = 108–110]/fabressei [n = 90], Agrodiaetus hopfferi [n = 15]/poseidon [n = 19–22]). The main reason for this failure is that a large proportion of species are not reciprocally monophyletic, e.g. due to incomplete lineage sorting, which is in accordance with a previous study . Moreover, the success with this method is again completely dependent on comprehensive sampling. If the correct species is not included in the profile, the assignment must by necessity be incorrect and misleading. Because of the non-existence of a barcoding gap, this error will often be impossible to detect. This limits possible applications of the barcoding approach. For example, cryptic species can only be detected with the help of a barcoding approach at high genetic divergence from all phenotypically similar species. An example is Agrodiaetus paulae which was discovered in this way . In contrast, and on the one hand, the sympatric species pairs Agrodiaetus ainsae-fabressei, A. hopfferi-poseidon and A. morgani-peilei would have gone unnoticed by barcoding approaches even though their strong phenotypical and karyological differentiation (n = 108 vs. n = 90, n = 15 vs. n = 19–22 and n = 27 vs. n = 39, respectively) clearly indicates their specific distinctness. On the other hand, sequence divergence in what is currently believed to represent one species does not per se prove the specific distinctness of the entities in question. In Polyommatus icarus or P. amandus, for example, the high divergences between North African and Eurasiatic samples is a strong hint for the presence of unrecognized cryptic species, but this needs to be rigorously tested with sequence data from samples that cover the geographic range more comprehensively. Also in practical application the problem of misidentified specimens and sequences in GenBank remains a real threat to the accuracy of barcode-based identifications. An example is the GenBank sequence AB192475 of Lampides boeticus which is also used in the CBOL database (see above). This underscores the importance of voucher specimens and documentation of locality data, an issue raised by barcoding supporters but unfortunately still much neglected by GenBank. Another case of misidentification (GenBank sequence AF170864 of Plebejus acmon which was originally submitted as Euphilotes bernardino)  has already been corrected with the help of the voucher specimen.
In conclusion, the barcoding approach can be very helpful, e.g. in identifying early stages of insects or when only fragments of individuals are available for analysis. However, correct identification requires that all eligible species can be included in the profile and that sufficient information is available on the amount of intraspecific genetic variation and genetic distance to closely related species.
The barcoding procedure is not very well suited for identifying species boundaries but it may help to give minimum estimates of species numbers in very diverse and inadequately known taxonomic groups at single localities. Our case study on Agrodiaetus shows that a substantial number of species would have gone unnoticed by the barcoding approach as 'false negatives'. Thus, especially in clades where many species have evolved rapidly as a result of massive radiations with minimum sequence divergence, the barcoding approach holds little promise of meeting the challenge of rapid and reliable identification of large samples. Yet, it is exactly these situations which pose the most problematic tasks in the morphological identification of insects.
Although molecular data can be helpful in discovering new species, a large genetic divergence is not sufficient proof since it must be corroborated by other data. Furthermore, most closely related species which are difficult to identify with traditional means, are also similar genetically and would go unnoticed by an isolated barcoding approach. Mathematical simulations have shown that populations have to be isolated for more than 4 million generations (i.e. 4 million years in the mostly univoltine Agrodiaetus species) for two thresholds proposed by the barcoding initiative (reciprocal monophyly, and a genetic divergence between species which is 10 times greater than within species) to achieve error rates less than 10% . This might help to explain why the barcoding approach appears to be more successful in the Oriental genus Arhopala which is thought to represent a phylogenetically older lineage of Lycaenidae estimated to be about 7–11 Million years old , while the origin of the Palaearctic genus Agrodiaetus is dated at only 2.5–3.8 Million years .
Our data show that the lack of a barcoding gap and reciprocal monophyly in Lycaenidae is not confined to the genus Agrodiaetus with its extraordinary interspecific variation in chromosome numbers, but also to other genera of Lycaenidae with stable chromosome numbers. It should also be noted that in Agrodiaetus there is neither evidence for exceptional rapid radiation as in cichlids of the East African lakes  nor for unusual (i.e. sympatric) speciation patterns caused by karyotype evolution. Rather, karyotype diversification seems to have been a mere by-product of the usual mode of allopatric speciation [29, 30, 44].