Comparative performance of the five molecular methods
The percentage of correctly identified strains obtained using the five identification methods, and the number of misidentified non-targeted species greatly depended upon the method used (Tables 1 and 2). The percentage of misidentified strains ranged from 16.8% to 67.4% (Table 2). The m-PCR method of Kabeya et al.  had the worst performance, and produced unreliable results for all three of its targeted species (Tables 1 and 2). Although all strains of A. cryaerophilus and A. skirrowii were correctly identified, a further eight and six non-targeted species, respectively, were mistakenly identified as one of these two species (Table 1). Furthermore, only 4.8% of the A. butzleri strains were correctly identified, with six non-targeted species being confused with this species (Tables 1 and 2). Globally, the Kabeya et al. m-PCR method correctly identified just 32.6% (31/95) of the studied strains. Although this method was also designed to differentiate subgroups 1A and 1B of A. cryaerophilus, not all strains of these subgroups were correctly identified (Table 2). This correlates with the in silico observations of Douidah et al.  who reported that the primer used  were not specific enough to provide correct identification of A. cryaerophilus at the subgroup level. Further to this, Debruyne et al. have suggested, that based on results of AFLP and hsp60 analyses, the subgroup nomenclatures 1A and 1B should be abandoned.
The second least reliable method analysed was the m-PCR technique described by Houf et al.. This correctly identified 55.8% (53/95) of the strains (Table 2), including all those belonging to its targeted species (A. butzleri, A. cryaerophilus, and A. skirrowii; Table 1). This method was 100% reliable for the identification of A. butzleri, and there was no confusion with other species. However, nine of the fourteen non-targeted species generated the typical amplicon of A. cryaerophilus; two that of A. skirrowii; and two simultaneously generated both amplicons (Tables 1 and 2). Only A. cibarius produced no amplification when using this method (Table 2). These results agree with previous studies that showed the existence of misidentifications when using this method [1, 5–7].
A similar number of correctly identified strains (83.2%) were obtained when using the other three evaluated methods (Pentimalli et al.; the combined method of Douidah et al.  and De Smet et al.; and Figueras et al.). However, the number of misidentified non-targeted species differed depending upon the method used (Tables 1 and 2). Most misidentification occurred when using the method of Pentimalli et al.. In this case, four non-targeted species were confused with A. butzleri, one with A. cryaerophilus, and two with A. skirrowii (Tables 1 and 2). Furthermore, the expected amplicons for A. butzleri and A. skirrowii in individual reactions were also obtained for the eight and three strains of A. cibarius, respectively (Table 2). Nevertheless, no cross-reaction with non-targeted species occurred when using primers designed for A. cibarius that reacted only with the eight strains of this species. The combined method of Douidah et al. and De Smet et al., misidentified four of the non-targeted species (Arcobacter defluvii, Arcobacter ellisii, Arcobacter venerupis, and Arcobacter suis) as A. butzleri, and two of the three strains of A. ellisii as A. cryaerophilus (Table 2). The method performed correctly for the four remaining targeted species. Finally, the 16S rRNA-RFLP designed by Figueras et al.  was found to misidentify three species (A. trophiarum, A. thereius, and some A. cryaerophilus strains) as A. butzleri. Further to this, A. suis, and A. defluvii produced the same pattern, and two species (A. venerupis, and Arcobacter marinus) a very similar one (Table 2). Because of these limitations, this method has recently been updated with new endonucleases; these produced specific results for all strains and species . This updated protocol was the one used to identify all strains used in this study.
Comparative evaluation of the targeted genes and designed primers
When the results were evaluated in relation to genes used to identify the species, it was observed that the 23S rRNA gene regions targeted in the Kabeya et al. method for A. butzleri, A. cryaerophilus, and A. skirrowii were unreliable, as was the region employed in the Houf et al. method  for A. cryaerophilus (Tables 1 and Additional file 1: Table S2). However, the regions of the 23S rRNA gene targeted by the m-PCR method of Douidah et al.  were 100% reliable for the detection of A. skirrowii, A. cibarius, and A. thereius, but not for A. butzleri (Tables 1, 2 and Additional file 1: Table S2). With regard to the gyrA gene, the region used for the identification of A. cryaerophilus in the latter method, and the one in the method of Pentimalli et al.  were unreliable. Despite all strains of A. cryaerophilus being correctly identified, A. ellisii was confused with this species when using the Douidah et al. method, and with A. skirrowii when using the Pentimalli et al.  method (Tables 1 and 2). The main reason for the poor performance of the targeted regions of 23S rRNA or gyrA genes (Additional file 1: Table S2) is the limited amount of sequences used to derive the primers. For instance, the sequences of the 23S rRNA gene are, at the time of writing, only available for eight of the seventeen known Arcobacter species (A. butzleri, A. cryaerophilus, A. skirrowii, A. cibarius, A. nitrofigilis, A. thereius, Arcobacter mytili, and A. trophiarum), and the gyrA gene is only available for seven species (A. butzleri, A. cryaerophilus, A. skirrowii, A. cibarius, A. nitrofigilis, A. marinus, and A. halophilus). In contrast, the sequences of the 16S rRNA gene are available for all species of the genus, and this has enabled the identification of endonucleases that produce specific patterns for all species; as described in the recently published update of the 16S rRNA-RFLP method . The 16S rRNA gene has also been used to design specific primers for A. skirrowii and A. butzleri in the Houf et al. method , and for A. butzleri by Pentimalli et al. . However, only the primers that targeted the 16S rRNA region chosen by Houf et al.  for the identification of A. butzleri (Additional file 1: Table S2) were 100% specific, and showed no cross-reaction with other species (Tables 1 and 2).
Literature review of the studied methods
The results of the literature review, which summarised the total number of strains and species identified using any of the five compared methods (Additional file 1: Table S3), revealed that the m-PCR method of Houf et al.  was the most globally referenced, with 71.9% (123/171) of all citations. This method was used to identify 64.8% (2735/4223) of the strains recorded in the literature since 2000 (Additional file 1: Table S3). The next most frequently used methods were the 16S rDNA-RFLP of Figueras et al.  and the m-PCR of Douidah et al., which were used to identify 14.6% and 13.4% of strains, respectively (Additional file 1: Table S3). The overall most prevalent species were A. butzleri (63.7% of strains), followed by A. cryaerophilus (27.3%), and A. skirrowii (4.9%) (Additional file 1: Table S3). The other 14 species represented only 4.1% of the recovered strains (Additional file 1: Table S3). The species diversity may be influenced by the different origins of the strains and/or the isolation methods used in the analysed studies.
When considering the results obtained in the present study, with those of the literature review, the strains identified as A. butzleri (64.5%) using the m-PCR designed by Houf et al.  could be considered to be correctly identified (Additional file 1: Table S3). However, the use of this method has probably led to a global overestimation of the number of A. cryaerophilus and A. skirrowii as some of the strains identified are likely to belong to other species (Tables 1 and 2). For example, when Atabay et al. used the Houf et al. method  they identified six A. skirrowii strains that were not able to hydrolyze indoxyl acetate, despite this being a typical phenotypic characteristic of the species. Interestingly, A. mytili, one of only two Arcobacter species (along with Arcobacter molluscorum) unable to hydrolyze indoxyl acetate, produces the typical band of A. skirrowii when the m-PCR method of Houf et al.  is used. Therefore, the six strains identified by Atabay et al. may belong to that species. Further evidence for this confusion of results can be found in a study on the prevalence of Arcobacter in meat and shellfish in which strains belonging to A. nitrofigilis and A. thereius were recognized . This is because contradictory results were seen when using two identification methods in parallel [14, 18]. When using the Houf method , A. nitrofigilis produced the expected amplicon for A. skirrowii and A. thereius the amplicon expected for A. cryaerophilus. However, when using the method of Figueras et al.  the expected 16S rRNA-RFLP pattern of A. nitrofigilis and A. butzleri was obtained for the A. nitrofigilis and A. thereius strains, respectively. The correct identity of these strains was confirmed as A. nitrofigilis and A. thereius through sequencing of the 16S rRNA and/or rpoB genes . This sequencing approach resolved the discrepancies observed between the two identification methods [14, 18] and has also led to the discovery of the species A. mytili, A. molluscorum, A. defluvii, A. ellisii, Arcobacter bivalviorum, A. venerupis, A. cloacae, and A. suis[5–7, 24–26].
The use of the m-PCR method of Douidah et al. in combination with the PCR method of De Smet et al. enabled A. thereius (17.6%, 100/567), A. trophiarum (1.8%, 10/567), and A. cibarius (0.2%, 1/567) to be recognized in two independent studies [27, 28] (Additional file 1: Table S3). Nevertheless, there is a weakness in this approach as the strains of four non-targeted species may be misidentified as the more frequently isolated A. butzleri (Tables 1 and 2).
Finally, with regard to studies that used the methodology designed by Kabeya et al. , our results revealed that all of the targeted species may have been overestimated; this is because 12 of the 14 non-targeted species could be misidentified (Tables 1 and 2). No studies were found that used the PCR method of Pentimalli et al. , and our results indicate that this method is not reliable (Tables 1 and 2).