A presence/absence (1/0) matrix was created from the data compiled at the Arctic collembolan catalogue by Babenko and Fjellberg [44], considering 16 geographical areas also as defined by Babenko and Fjellberg [44] (Figure 1), Area C (Urals), although indicated in the main description table (Table 1), was excluded from the statistical analyses as this region has been considerably under-sampled in comparison with the remaining areas and the diversity index is unrealistically low [56]. Additionally, and to avoid bias in the dataset, species with dubious records (marked as a ‘?’ in Babenko and Fjellberg [44]), likely recent introductions into Arctic regions (for example Orchesella), recently described species (species described from 2005 onwards, as they may have been overlooked in other regions) and species with unresolved status were not included in the current study. Thus 358 species were considered in the study from a total of 390 species described from Arctic areas.

Information on contemporary climatic regimes from each mainland area was gathered, including factors previously identified as affecting collembolan species distribution: temperature [39,57] and precipitation [58]. Available meteorological data from the Arctic island archipelagos is in most cases scattered and often biased towards summer measurements. Therefore environmental data from the Arctic islands were not included.

Gaussian mixture clustering was significant for K = 9 (statistics value for original data = 0.26; mean for null data = 0.335; p < 0.05). From the 358 species included in the study, 169 were considered noise by NNclean, while the remaining 189 were allocated to any of the 9 clusters. Cluster 1 was composed mainly by Palaearctic/Atlantic species, clusters 2, 5 and 7 by mainly Nearctic species, clusters 3, 4, 6 and 8 mainly by Palaearctic species and cluster 9 by Holarctic species (Table 2).

Common elements were present within Siberian areas, Atlantic areas and the Canadian Arctic (Figure 2), with some wide-ranging species shared among all them. Just one cluster (cluster 9) was found throughout the Arctic, comprising exclusively cosmopolitan and holarctic, species: Agrenia bidenticulata (Tullberg, 1876), Folsomia bisetosa Gisin, 1953, Folsomia quadrioculata (Tullberg, 1871), Pseudoisotoma sensibilis (Tullberg, 1876), Megalothorax minimus (Willem 1900), Oligaphorura groenlandica (Tullberg, 1876) and Sminthurides malmgreni (Tullberg, 1876). This is in contrast to the Beringian area (F, G) which showed a high number of species characteristic from that area (22 species from cluster 2 exclusive from G, and the 13 species from cluster 4, which are mainly found in F), although some species characteristic of area F also appear in area E: Tetracanthella martynovae Potapov, 1997, Psyllaphorura sensillifera (Martynova, 1981), Protaphorura neriensis (Martynova, 1976), Anurida bondarenkoae Tshelnokov, 1988) (Table 2).

Three clusters appear uniquely in one mainland area and are absent from their respective Arctic island groups: Cluster 2 in the American side of Beringia, comprising species present in Western America (G); Cluster 3 in mid-Siberia including species present in west and mid-Siberia (D); and Cluster 5 in the Canadian Arctic. The American-Beringian cluster (Cluster 2) appears as unique and different to the main Asian-Beringian cluster (Cluster 4) (Figure 2). Most of the species that colonize Atlantic areas have European-Western Palaearctic distributions, while species colonizing Eastern Siberia have mostly Asiatic distributions and species found in the Canadian Arctic are mainly Nearctic (Table 2). Species distributed throughout the Arctic are mostly not exclusively known from Arctic areas, while limited-range or endemic species appear in Beringia, mid-Siberia, the Canadian Arctic and Greenland (Table 3).

High Arctic islands seem to be colonized primarily by wide-ranging species, although that is not the case for the Svalbard archipelago which comprises elements from the common holarctic cluster (9) and the Atlantic and East European clusters. There is just one species currently thought to be specific to the high Arctic islands not present elsewhere in the Arctic, Folsomia altamontana Yosii, 1971, described from Severnaya Zemlya and New Siberian Islands.

Geographical areas were found to be clearly different whether considering species composition (analysis of variance on ranks, H = 84.80, df = 7, p < 0.05) or daily temperature (period 1996–2005) (analysis of variance on ranks, H = 5091.93, df = 7, p < 0.05). The post-hoc Dunn's test for pairwise differences between areas revealed that pairwise differences between areas due to temperature were not the same as the pairwise differences between areas due to species distribution (Table 4).

For 8 out of the 9 clusters found by Gaussian mixture clustering, no clear relation was found between the geographical distribution of the species present in each cluster and the environmental variables accounted. Species belonging to each different cluster shows a geographical distribution (Figure 3) which does not relate to the curved representing average temperature (Figure 4A), temperature of the warmest month (July) (Figure 4B), or precipitation (Figure 4B). The distribution of cluster 1 shows a linear relation with both precipitation (r² = 0.830, F = 29.255, p < 0.005) and mean annual temperature (r² = 0.593, F = 8.725, p < 0.05).The distribution of cluster 6 shows a second grade relation with mean temperature of the warmest month (r² = 0.59, F = 11.40, p < 0.05). The distribution of species considered as noise by Gaussian Mixture Clustering does not show a relationship neither with mean annual temperature (r² = 0.210, F = 1.591, p >0.05), nor with precipitation (r² = 0.172, F = 1.244, p > 0.05)

The geographical distribution of clusters (Figure 2) showed the existence of a relationship within different Atlantic areas (A, a, I, as far as area B in East Europe), western Siberia (B, c, D), eastern Siberia (E, F, D), Asian-Beringia (E,F), Alaskan-Beringia (G) and the Canadian Arctic (H). Greenland appears as more closely related to the European Arctic than to the Canadian Arctic. The western and eastern Palaearctic showed a different cluster structure, sharing just the common Holarctic element. Specific clusters, exclusive to a particular region were found in mid-Siberia, Asian-Beringia, Alaskan-Beringia and the Canadian Arctic, while the high Arctic islands possessed mainly components from the most widespread cluster composed of wide-ranging species. Only a small number of cosmopolitan species were present throughout the Arctic (Table 2).

Species colonizing each defined Arctic area also tended to share specific regional distributions: species colonizing the Canadian Arctic were mainly Nearctic, those colonizing Scandinavia had mainly a western Palaearctic distribution and species colonizing the eastern Palaearctic had mostly Asiatic-Beringian distributions. Endemic species were present in mid- and western Siberia (D), particularly in the Taimyr Peninsula, in Beringia (F and G) and only a single endemic species occurred in eastern North America (H) and Greenland (I) (Table 3).

Factors as precipitation and temperature have been previously indicated as limiting the distribution of collembolan [39,57,58]. Differences found in climate between areas also do not match closely those found in species distribution (Table 4). Clustering do not follow environmental parameters either—for instance areas more distantly related in terms of climatic regimes (e.g., Scandinavia (A) and mid-Siberia (D)) share a cluster which is not share by Greenland (I), which has an intermediate temperature and precipitation regime. Each cluster shows a defined distribution, often sharply skewed towards certain geographic locations (Figure 3), showing no consistency with temperature and precipitation gradients (Figure 4A and 4B). Additionally, species considered as noise do not show a relationship with environmental parameters. Although contemporary climatic regimes may play a role in species distribution, this is not capable of explaining the geographical variation found. Additionally, adjacent areas are more likely to be influenced by similar climatic regimes, for instance the western Palaearctic, Svalbard Archipelago and Greenland are under the influence of the North Atlantic Oscillation [68]. Therefore, there is a possibility that these environmental parameters are in effect autocorrelated with the distribution clusters.

Northbound colonization patterns as the Arctic deglaciated would explain why most species colonizing Atlantic areas have European-western Palaearctic distributions, species colonizing eastern Siberia have mostly Asiatic distributions and species found in the Canadian Arctic are mainly Nearctic (Table 1). The strong Atlantic influence on distribution clusters of the Greenland area disagrees with previous Holarctic studies both in Tardigrada [69] and in Cladocera [70] where a Nearctic origin of the Greenland fauna was suggested. Nevertheless, Cladocera and Tardigrada, aquatic groups presenting drought resistance stages, have been highlighted as good airborne dispersers [70]. On the other hand for the non-anhydrobiotic springtails, long distance aerial dispersal in polar areas is unlikely [71] due to mortality following rapid desiccation [72], In spite of collembolan being considered truly soil dwellers, some surface active species can show higher dispersal potential and resistance to desiccation, and aerial dispersal should not to be underestimated for short distance dispersal in moist conditions. Springtails have, however, been shown to survive long periods on sea- water [73], and ocean currents, together with glacial refugia, may have an important role in delineating Antarctic distributions [74]. Collembola has been described to been passively disperse by water [75,76]. Should species of Collembola be effective sea water dispersers, it raises the question of why there is no obvious connection between the Palaearctic and Nearctic areas even though ocean currents and ice flow patterns show a level of physical connection between the Palaearctic and Nearctic across the Arctic Ocean [77] and given that springtails have also been shown to survive extended periods below zero [78–80]. Plant molecular studies have also failed to support the traditional hypothesis that considers the north Atlantic an important dispersal barrier between Palaearctic and Nearctic areas [7,81,82]. Abbott et al [83], applying RFLP techniques in chloroplast DNA of the Arctic plant species Saxifraga oppositifolia L., obtained a general Arctic structure showing a series of similarities with the pattern obtained in the current study (Figure 2). A Beringian component was shared among eastern Siberia areas as far as Taymyr, a mid-Siberian component was shared as far as Scandinavia and the Svalbard archipelago and an Atlantic component was present from Scandinavia to western Greenland. These patterns were associated with glacial refugia and post-glacial colonization phenomena. In both cases, either gaussian mixture clustering in the distribution of the Arctic springtails or cp DNA haplotype distribution of S. oppositifolia, the eastern boundaries of the Atlantic component occur west of the Taimyr Peninsula, while the Taimyr forms the western boundary of the Beringian cluster. The link between Atlantic and west and mid-Siberia shown by cluster 8 suggest a link between these areas and could be indicative of post-glacial colonization from LGM ice-free areas in mid-Siberia.

Beringia refugia as determining factor on biogeographical structure of terrestrial fauna and flora have been repeatedly emphasized [4,5,84]. Insights into the colonization of Siberia from Beringia, as described for Arctic plants [83], are provided from the extension of cluster 6 from Beringia (F) to mid-Siberia (D) as well as cluster 4 into eastern Siberia (E). The low number of species colonizing the high Arctic islands could suggest recent colonization. Nonetheless, the high Arctic islands, although colonized by a lower number of species, appear to have a higher species diversity than that which would be expected by island biogeography theory [85]. In the case of high Arctic Canada and many Siberian Arctic islands, only wide-ranging species colonize them rather than species commonly found on the closer mainland (Figure 2). This suggests that mostly wide-ranging and competent dispersers have had the ability to reach these areas. Wrangel island and the New Siberian islands could be also colonized from eastern Siberia and Beringia, their closest mainland with which they share cluster 6 as has been previously suggested based on local fauna descriptions [50], Novaya Zemlya could have been colonized from either or both Eastern Europe and Mid Siberia, while the Svalbard archipelago seems to have been colonized both from distant (mid-Siberia) and close (Atlantic), mainland sources.

The disjunct distribution of the high Arctic species Folsomia altamontana Yosii, 1971, which is present not only in Severnaya Zemlya and the New Siberian islands but also the Himalaya and Putorana Plateau, could suggest either a magnificent ability for long distance dispersal, relict populations for what during glaciation was a more widespread species, or a cryptic speciation event, as has recently been suggested for the Antarctic Friesea grisea (Shaffer 1891) [41]. Folsomia altamontana is thus an interesting target for further phylogeographical analysis.

Endemic collembolan species (narrow range endemism) occur in north and mid-Siberia (D), Beringia (F and G), the Canadian Arctic (H) and Greenland (I), all of which have been previously indicated as including possible refugia during LGM by a number of authors and for a series of taxa [81,82,86–88]. The two possible glacial refugia that could be identified in mid-Siberia, east of Ural Mountains (D), and Beringia (F and G), defined by cluster characteristics (Figure 2) and inferred from the raw data through presence of endemic species (Table 3), agree with glacial refugia described for Arctic plants [82,6]. In contrast, the exclusive cluster in the Canadian Arctic seems to be composed mainly by species widespread (5) in North America, suggesting a main dispersal route to the Arctic rather than a possible refugium, as discussed in the case of Arctic Tardigrada [69]. Additionally, no indication of glacial survival in the high Arctic islands can be found in the current analysis of Arctic Collembola, although occurrence of endemism in other invertebrate groups could indicate otherwise [89]. A particularly interesting case is provided by the high Arctic islands in the Canadian Archipelago, including Ellesmere island and Baffin island. This region was suggested by Hultén [81] as a possible glacial refugium as it remained largely unglaciated throughout LGM and was postulated as a source for post-glacial Arctic recolonization for plant species such as Dryas integrifolia [88] or the collared lemming Dicrostonyx groenlandicus [86]. However, this area does not show other patterns which might be expected in an area of recent colonization, for example no endemic or limited range species, rather being colonized by a low number of species, and possessing mostly cosmopolitan or Holarctic species. Species-level biogeographical analysis however overlooks obscure regional endemism due to cryptic speciation, a phenomenon that is not unknown in the Collembola [41] and not unexpected in Arctic areas as high levels of intra-specific genetic variation have been found in widespread groups genera as Folsomia [90]. Furthermore, species-specific and individual responses to climate fluctuations, as well as differences in environmental, climatic conditions, geographical extension and duration of the occupation of the different glacial refugia, would determine post- glacial characteristics of the species and communities in the area [91] making some refugia more readily identifiable than others. Cryptic glacial refugia can nonetheless be detected applying the appropriate method, as repeatedly shown by phylogeographical studies [22,23,41].

Exclusive clusters could be an indicator of either glacial refugia or early post-glacial recolonization, although the high number of species restricted to Beringia would support the glacial refugium hypothesis in line with the accumulating additional evidence of the Beringian glacial refugia [83,86,87]. Abbott et al. [83] used haplotype diversity analysis to define Beringia as the major refugium for Saxifraga oppositifolia during the LGM, although the high haplotype diversity in Taimyr area was also suggested as being the result of dispersal to the area from different sources rather than resulting from differentiation in glacial refugia. Later studies on the colonization patterns of high Arctic plants [7] have pointed to drifting ice and wood as important high Arctic dispersal vectors, creating patterns such as the link between mid-Siberia and the Svalbard Archipelago, the latter mainly colonized from a refugium east of the Ural Mountains [6]. This agrees with the link shown in Figure 2 between Atlantic and mid-Siberia through cluster 8. The exclusive cluster from the west and mid-Siberian area (D) might therefore reflect the invertebrate community of the glacial refugium described for Arctic plants [6], a hypothesis strengthened by the existence of endemic springtail species in mid-Siberia (D) (Table 3).