We obtained between 680 and 726 bp (Otus lettia ussuriensis and Aegolius acadicus, respectively) for the myoglobin intron-2 resulting in a final alignment of 749 bp. Among the 749 bp, 151 were variable (20%) and 56 were parsimony informative (7.5%). Maximum Likelihood (ML) analyses yielded one tree (-ln = 2198.81) that slightly differs topologically from the 50% majority consensus rule tree obtained from the Bayesian analyses (-ln = 2442.98). We obtained between 561 and 593 bp (O. leucotis and O. pembanesis/O. senegalensis Allele 1, respectively) for the TGFB2 intron-5 resulting in a final alignment of 605 bp. Three individuals were found to be length-variant heterozygotes: O. leucotis possesses a CCT duplication in a region with a CCT pattern in all other species, O. rutilus (FMNH 431150) possesses a one base pair deletion (G) in position 171 of the alignment, and O. senegalensis possesses a one base pair insertion (A in position 559 of our alignment; this insertion was also found in the two O. pembaensis individuals sequenced). The two O. senegalensis alleles also differ by two further mutations; these two alleles clustered together as the sister-group to O. pembaensis in a ML analysis (tree not shown). Therefore, we use the consensus sequence (the two single nucleotide polymorphisms were coded using the appropriate IUPAC code) from the two O. senegalensis alleles for further phylogenetic analyses. Only the alleles without the insertion/deletion were included in the phylogenetic analyses for O. leucotis and O. rutilus, as the insertion/deletion events were autapomorphic in both cases. Among the 601 base pairs retained for the analyses, 163 were variable (27%) and 71 were parsimony informative (11.8%). ML analyses yielded one tree (-ln = 2131.71) that slightly differs from the 50% majority consensus rule tree obtained from the Bayesian analyses (-ln = 2179.36).

The topologies obtained from the nuclear loci were very similar to each other, delineating the primary clades without achieving resolution at the tips (see Additional Files 1 and 2). The 50% majority consensus rule tree obtained from the Bayesian analyses (Figure 1, -ln = 4638.29) and Maximum Parsimony strict consensus tree (105675 equally parsimonious trees of 412 steps, CI = 0.82, RI = 0.86) of the two concatenated nuclear loci provided a well-resolved topology for inter-generic relationships as well as some resolution of relationships among the primary Otus lineages but failed to provide resolution among members of the Indian Ocean radiation.

The concatenated mitochondrial sequences retained for analyses were 2983 bp long (1047 bp for ND2, 684 bp for ATP6, 351 bp for ND3 and 901 bp for cytochrome-b) and correspond to the positions 5246 to 6281 (ND2), 9240 to 9923 (ATP6), 10776 to 11120 (ND3), and 15011 to 15911 (cytochrome-b) of the Gallus gallus mitochondrial genome sequence 30; GenBank accession number X52392). The ATP6 and cytochrome-b sequences contained no insertions, deletions and stop-codons in the reading frame. The ND2 sequence of Aegolius acadicus exhibits a two-codon insertion (CAA ACC) just before the stop codon. All the ND3 sequences exhibited the pyrimidine insertion (T for O. capnodes, C for all other species analyzed) previously reported for several clades of birds 31; this extra-nucleotide was removed before phylogenetic analyses. Partitioning the gene by codon positions significantly improved the fit of models to the data for all four mitochondrial loci, as inferred from the Bayes Factor (B_F) values (BF_ND2 = 826.3, BF_ATP6 = 681.7, BF_ND3 = 319.1, BF_cytb = 1048.8). Mitochondrial gene trees were very similar to each other (ND2: ML -ln = 10556.45, BI partitioned by codon position = 10161.23 – Additional File 3; ATP6: ML -ln = 6550.30, BI partitioned by codon position = 6232.87 – Additional File 4; ND3: ML -ln = 3266.29, BI partitioned by codon position -ln = 3146.47 – Additional File 5; cytochrome-b, ML -ln = 8623.79, BI partitioned by codon position -ln = 8128.06 – Additional File 6), albeit levels or resolution varied among genes. As expected, no conflict was detected between the individual evolutionary histories of the mitochondrial gene trees (as inferred from posterior probabilities). The 50% majority-rule tree obtained from the concatenated analyses of the mitochondrial genes (partitioned by gene and codon position: -ln = 27725.63, Figure 1) was very similar in terms of topology and number of supported nodes to the MP strict consensus tree (two equally most parsimonious trees of 5922 steps, CI = 0.42, RI = 0.60).

The individual trees obtained from the two nuclear introns and mitochondrial data sets were very similar to each other and no incongruence was detected between the nuclear and mitochondrial topologies (Figure 1), according to the criteria defined in the Material and Methods section. Further, there was usually strong congruence for nodal support among the different methods. Most of the nodes present in the 50% majority-rule consensus tree resulting from the Bayesian analyses performed on the concatenated data set (partitioned by gene and codon position- 14 partitions-, -ln = 32195.91, Figure 2) were very well supported in the parsimony analyses too (two equally most parsimonious trees, 6340 steps, CI = 0.44, RI = 0.62).

The African species O. ireneae was recovered as the sister-taxon of all remaining Otus sensu stricto scops-owls. All the western Indian Ocean taxa clustered in a clade that also contains the Eurasian O. scops, the São Tomé endemic O. hartlaubi, the Pemba Island endemic O. pembaensis, the African mainland O. senegalensis, the Philippine taxa O. mirus and O. longicornis, as well as the Indo-Malayan O. sunia. The western Indian Ocean taxa were not recovered as a monophyletic lineage since the Seychelles O. insularis was more closely related to the Indo-Malayan O. sunia than to any other taxon occurring in the western Indian Ocean region. Uncorrected-p mitochondrial distances among members of the Indian Ocean taxa/O. sunia clade range between 4.6% (between O. capnodes and O. rutilus) and 7.0% (O. sunia and O. rutilus), with a mean of 5.3% (s.d. = 0.7%). Most of the relationships between the Comorian and Malagasy taxa (capnodes, mayottensis, moheliensis, pauliani, rutilus) did not receive statistical support and short inter-nodes characterized most of the branches among these lineages. We attribute this lack of resolution to rapid speciation events ('hard polytomy') rather than a lack of sufficient character sampling ('soft polytomy') because 1) we sampled several genes with different evolutionary properties resulting in a final alignment of more than 4300 bp and, 2) the nodes above and below the polytomy received bootstrap percentages of 70% or more or posterior probabilities of 0.95 or greater. The remaining Otus species clustered in a second large clade. Within this latter clade, O. lettia ussuriensis (eastern Russia) did not cluster with another O. lettia sample from Laos, but with the Philippine taxon O. megalotis, suggesting that further work with more complete geographic sampling is needed to address the evolutionary history of the O. lettia/O. megalotis species complex.

The biogeographic history inferred from the topology of the concatenated analyses is intriguing as several faunistic exchanges involving the Indian Ocean islands and Indo-Malaya region occurred (split O. mirus-O. longicornis from the Indian Ocean taxa and split between O. insularis and O. sunia). These faunistic exchanges imply either multiple colonization events of the western Indian Ocean islands or one re-colonization of the mainland by O. sunia, about 0.25–0.30 mya. The main radiation of western Indian Ocean island taxa (2.5 mya, 95% HPD = 1.2–4.0 Table 1) occurred soon after the initial colonization event (3.6 mya, 95% HPD = 1.8–6.0). Our analyses revealed that O. hartlaubi, endemic to the African Atlantic island of São Tomé, and O. pembaensis, restricted to Pemba Island off the east African coast, have strong affinities with the African mainland species O. senegalensis. A very close relationship between O. senegalensis and O. pembaensis is further supported by the fact that these two taxa share a one nucleotide insertion in the TGFB2 locus. The position of O. harlaubi in the mitochondrial and concatenated tree renders this insertion paraphyletic. Yet, it is worth noting that the two alleles of O. senegalensis differ in length at this site. Considering that the first colonizers of São Tomé and Pemba were probably in small numbers when compared to the continental and widely distributed O. senegalensis, we regard the discrepancy between the mitochondrial and nuclear trees as being due to the different effective population size of the markers (the mitochondrial genome has an effective population size that is one fourth of the nuclear genome) and random processes (coalescence). The basal split within Otus sensu stricto occurred about 11.7 mya (95% HPD = 6.0–19.0) and the divergence between the two primary clades about 9.3 mya (95% HPD = 4.7–15.1, Figure 2).

Previous studies 1820 highlighted that several species usually assigned to Otus (Ptilopsis leucotis, Megascops) are not directly related to Otus. Our analyses, using independent samples and additional genes, confirmed the lack of direct relationships among these three lineages. Given that these results were previously suggested, we do not discuss them further.

Our analyses based on nuclear and mitochondrial sequence data provide the first phylogenetic hypothesis on the diversification of Otus sensu stricto. We suggest that Otus consists of at least three primary lineages. The first lineage consists of the 'relictual' Sokoke Scops-owl (O. ireneae), endemic to the coastal forests of Kenya and Tanzania. This species is sometimes considered to form a superspecies with the patchily distributed Sandy Scops-owl (O. icterorhynchus) 22, the only African Otus species we were not able to sample.

The second major lineage is restricted to South-East Asia and consists of O. spilocephalus, the O. bakkamoena-O. lettia-O. lempiji superspecies, as well as Otus lettia ussuriensis and O. megalotis. As inferred from our results, the current taxonomy within this group appears to be problematic as O. lettia lettia and O. l. ussuriensis are not recovered as sister-taxa. O. l. ussuriensis, restricted to Sakhalin, Ussuriland and North-East China, is sometimes considered to be related to the Japanese Scops-owl (O. semitorques), based on voice, plumage and iris color data 22. We could not include the latter species in the present work, but highlight here that this relationship needs to be further tested using molecular data.

The third major lineage includes all the remaining Otus species we sampled and is divided in two subclades that consist of 1) O. scops, O. senegalensis, O. hartlaubi and O. pembaensis, and 2) O. longicornis, O. mirus, O. mayottensis, O. rutilus, O. capnodes, O. moheliensis, O. pauliani, O. insularis and O. sunia. Relationships among these taxa are discussed below.

Data accumulated in recent years on the biology and distribution of the western Indian Ocean scops-owls 26272829 together with our analyses based on DNA sequences of six loci, shed new light on the evolutionary history of these birds. The phylogeny we propose here is in full agreement with recent work based on phenotypic characters suggesting that O. pauliani, O. capnodes, O. moheliensis and O. mayottensis represent distinct evolutionary lineages 252829. In addition we find good support for Asian biogeographic affinities of the western Indian Ocean islands Otus spp., and highlight that they may constitute a paraphyletic assemblage.

The Indo-Malayan O. sunia formed a well-supported clade with O. insularis from the granitic Seychelles, which was very closely related to the Indian Ocean lineage, the latter forming a sequentially paraphyletic assemblage. The nested position of O. sunia/O. insularis within the Malagasy-Comorian clade, supported by a posterior probability of 0.98 (but not by MP bootstrap percentage), suggests a recent re-colonization of the mainland from an island-distributed ancestor. Yet, we also acknowledge that posterior probabilities could be misleading when short internodes/polytomies are involved 33, which is the case here, and we await the implementations of reversible-jump Markov chain Monte Carlo algorithms to explore these aspects 33.

It is also worth noting that in a similar biogeographic comparison of a different group of birds, non-monophyly of western Indian Ocean taxa was also retrieved by Warren et al. 13; the Indo-Malayan bulbul Hypsipetes madagascariensis was nested within the western Indian Ocean taxa, suggesting a similar scenario to the one we present for Otus. Colonization of the mainland from an island distributed ancestor is generally regarded as unlikely because: 1) mainland taxa are considered more competitive than island taxa and, 2) insular populations are smaller and produce fewer emigrants compared to those on continents 3435. However, empirical cases of continental re-colonization from islands are accumulating 16363738, including cases from Madagascar 39; this suggests that islands can also act as colonization sources for continental faunas. Data on the phylogeographic structure within the Indo-Malaya O. sunia complex may help to decipher which hypothesis (multiple colonizations of the Indian Ocean islands or re-colonization of the mainland) best explains the current pattern. Indeed, if strong and ancient phylogeographic structure occurs within O. sunia, we would expect two independent colonizations of the western Indian Ocean islands. In contrast, if genetic data indicate weak differentiation among O. sunia populations and patterns of population expansion from western to eastern Indo-Malaya, this would support the hypothesis of re-colonization of the mainland from the Seychelles Islands.

The genus Otus is fairly homogeneous in plumage relative to many bird genera. However, the Indian Ocean taxa show strong differences from each other in structural, plumage and vocal characters 2628. Considering, for example, only the songs of the Comorian taxa (which inhabit islands as close as 50 km from each other), O. pauliani gives a very long series of chaw notes repeated at about 2/sec; O. moheliensis a sequence of hisses; O. capnodes a high-pitched whistled peeooee; and O. mayottensis a series usually of 3–11 deep, single hoots. Intuitively, these differences may be used to argue against close evolutionary relationships; however, our data indicate that the relationship is indeed close in all cases, as could be predicted from the islands' proximity to one another. This study confirms that the vocal and morphological differences are indeed associated with distinct evolutionary lineages, but suggests that they are not related in any simple or obvious way to the evolutionary distance between these lineages, and therefore must be used with caution in identifying affinities between taxa (or lineages). Therefore, we highlight here that the close relationship suggested by Marshall 24 between O. insularis and O. magicus should be further tested using molecular data, especially if we consider the considerable distance between the two areas (over 6000 km).

The geographic and temporal origins of certain western Indian Ocean island bird taxa have received attention in recent years [e.g. 59101316404142434445 this study] (Table 2).

Our estimate (3.6 mya, 95% HPD: 1.8–6.0) for the timing of colonization of the Indian Ocean islands by Otus coincides with estimates of at least seven other lineages of birds (Table 2), suggesting that the Indian Ocean islands avifauna was highly enriched at that time. The period associated with these multiple independent colonizations corresponds with the emergence of the volcanic islands of the Comoros archipelago (Mohéli 5 mya, Anjouan 11.5–3.9 mya, Mayotte 11.5–7.7 mya and Grande Comore 0.5 mya; 5051, thus providing possible stepping stones for dispersal between Africa and Madagascar. This possibility is also highlighted by the fact that several species or populations with African biogeographic affinities (Streptopelia capicola, Turtur tympanistria, Turdus bewsheri) 535455, colonized the Comoros islands but the colonization of Madagascar has not yet been achieved.

All faunal exchanges that unambiguously involve Madagascar and Indo-Malaya, or Madagascar and the Seychelles occurred during the last 3.5 mya [1349, this study] (note that the biogeographic history of the Philepittinae, Vangidae and the Sturnidae genus Hartlaubia are still uncertain or ambiguous). Warren et al. 48 hypothesised for members of the genus Zosterops that the colonizations could have been favored by dramatic sea-level shifts that occurred during the last 2.5 mya, which would have allowed the emergence of currently submerged land-masses between the Seychelles and Madagascar 565758, implying a 'stepping-stone' model. Our divergence dates estimates are slightly older than those of Warren et al. 49, although confidence intervals are largely overlapping. This biogeographic hypothesis fits the three unambiguous described cases involving Madagascar and Indo-Malaya, or Madagascar and the Seychelles (Hypsipetes, Otus and the Zosterops borbonicus lineage).

Subfossil remains of three extinct small owl species have been described from the Mascarene Islands (La Réunion, Mauritius, Rodrigues). Based on certain osteological features, these three species have been included in their own genus, Mascarenotus, which has been suggested to be derived from Otus 59. The relationships of Mascarenotus with respect to the other owl lineages still needs clarification, as this genus could represent a recent and derived off-shot of the western Indian Ocean lineage, possibly a first off-shot of the colonization process from the Seychelles to more westerly regional islands or even an unrelated lineage of owls.

Whereas the continental biogeographic affinities of certain western Indian Ocean island bird taxa seem largely resolved, the timing of colonizations or faunal affinities among volcanic islands are less well understood and the only aspect that is emerging is the absence of a common pattern, whatever the initial geographic origin. Indeed, all the studies that have been conducted so far indicate explosive diversification and a lack of resolution among the Comoros islands species. These data indicate that once the initial colonization was successful on any of the islands in this archipelago, dispersion and then diversification between nearby islands occurred quickly and randomly. The lack of biogeographic structure at the archipelago scale may thus be partly explained by the geographical arrangement of islands and the short inter-islands distances (maximum distance between two islands in the Comoros archipelago is 90 km) that probably favored colonization by alternative routes.

One final factor that could hide common patterns of diversification is unequal rates of extinction and recolonization across lineages. Indeed, not all lineages would face the same risk of extinction on islands as their characteristics (ecological requirements, population size) often considerably differ. For example, middle-sized birds, like owls (70–120 g), may be more prone to extinction on islands than small birds, like sunbirds (12–15 g) 60. Yet, even if uneven extinctions rates occurred amongst these lineages, it can be concluded that the overall diversification pattern in the Indian Ocean islands is star-like.

We obtained tissue samples from all western Indian Ocean Otus taxa, as well as samples from several Indo-Malayan and Afrotropical species (Table 3), focusing on as many super-species complexes as possible (sensu 22). We obtained tissues for seven of these super-species complexes. We were unable to obtain samples of the distinctive O. rufescens and O. sagittatus, as well as representatives of four super-species with localised and distant distributions relative to the western Indian Ocean (brooki/angelinae from Sumatra/Java/Borneo; mantananensis/magicus from the Lesser Sundas/Philippines/Mollucas; collari/manadensis/beccarii from Sangihe/Sulawesi/Biak and enganensis/alius/umbra from islands off Sumatra). Since the monograph of Marks et al. 22, one further Otus species, O. thilohoffmani has been described from Sri Lanka 61. This species is only known in museum collections by the type specimen, deposited in the National Museum Colombo (Sri Lanka), and based on morphology, has been suggested to be related to either O. rufescens or O. spilocephalus 61. As a consequence, we did not have access to a tissue sample of this newly described species. With the exception of O. magicus, considered by some authors to include O. insularis because of similarities in vocalizations 24, none of the species we were unable to sample have been considered closely related to the Indian Ocean taxa 22. We included two individuals per color morph for O. rutilus (sensu 23). Representatives of the Strigidae genera Aegolius, Athene and Glaucidium (Surniinae), Asio (Asioninae), and Bubo, Strix, New World 'Otus' (Megascops) and the African White-faced Owl ('Otus' Ptilopsis leucotis) (Striginae) were included as proximate outgroups. We rooted our trees using sequences from a representative of the Tytonidae (Barn Owl Tyto alba), which has been recovered as the sister-group of the Strigidae in molecular and morphological analyses 6263.

Total DNA was extracted from frozen, EDTA or alcohol preserved tissues (liver, blood, feathers, muscle) using a CTAB-based protocol 64 with an overnight Proteinase K (0,1 mg.ml-1) digestion. ND2, ATP6 and ND3 were amplified and sequenced using primer pairs L5219/H6313 65, L9245/H9947 66 and L10755/H11151 67, respectively. A 900 bp portion of cytochrome-b was amplified with the primer pairs L14967-H15487 and L15424-H15916 23). Myoglobin intron-2 and TGFB2 intron-5 were amplified with primers Myo2/Myo3F 6869 and tgf5/tgf6 70, respectively. The amplification and sequencing protocol were standard 23.

Molecular phylogenies were estimated using parsimony (P) and model-based approaches (maximum likelihood [ML], and Bayesian inferences [BI]), as implemented in PHYML v2.4 71 and MRBAYES 3.1 727374. Parsimony analyses were conducted with PAUP v4.0b10 75 using the heuristic tree bisection and reconnection branch-swapping (TBR) algorithm with 100 random addition replicates. Likelihood models were estimated with MRMODELTEST 2.0 76 using the Akaike Information Criterion 77. The selected models are listed in Table 4. Clade support in the individual gene trees for the ML and MP analyses was assessed by 1000 non-parametric bootstrap replicates 78. The six gene regions sequenced differ considerably in their properties and substitution dynamics, as inferred from the parameters of the models (Table 4). Consequently, analyses of concatenated data set were only performed using a mixed-model strategy. The relevance of partitioning the data set by gene and/or codon position was assessed with the Bayes Factor (B_F) 7980. Fourteen partitions (myoglobin intron-2, TGFB2 intron-5, first, second and third codon position of each of the four mitochondrial genes) were considered for the Bayesian concatenated analyses according to the functional properties of the markers. Bayesian analyses for the concatenated data set were performed allowing base frequencies, rate matrix, shape parameter and proportion of invariable sites to vary between the partitions (using the unlink and prset commands). Between four and six incrementally heated Metropolis-coupled MCMC chains were run for 15 million generations with trees sampled every 100 generations. The first 2*10⁶ generations (20000 trees) were discarded ('burn-in' period) and the posterior probabilities were estimated from the remaining sampled generations. The default temperature for chain heating (T = 0.2) resulted in not satisfactorily mixing among chains for the concatenated data set; we therefore lowered the temperature to T = 0.05, which resulted in swap frequencies between chains within the 20–70% interval. Two independent Bayesian runs initiated from random starting trees were performed for each data set, and the log-likelihood values and posterior probabilities were checked to ascertain that the chains had reached convergence. We also checked that the Potential Scale Reduction Factor (PSRF) approached 1.0 for all parameters and that the average standard deviation of split frequencies converged towards zero. We detected significant incongruence between the individual gene trees by comparing the topologies and nodal support obtained under different analytical methods (ML, BI). Criteria for incongruence were set at 70% for the bootstrap values 81, and at 0.95 for posterior probabilities 72.

Owls have a very rich fossil record 2282, yet, the taxonomic history of most of these taxa remains controversial and, to our knowledge, no cladistic analyses including fossil and modern taxa has been conducted to date. This hampers the use of most of these fossils as calibration points in our analyses. The family Strigidae is usually divided into three subfamilies or tribes: Striginae, Surniinae and Asioninae 21. These three subfamilies are defined by a combination of shared characters [82 fide 83]. The least inclusive of these three subfamilies, the Asioninae, consists of three genera and nine species (two of the genera, Pseudoscops and Nesasio are endemic to Jamaica and the Solomon Islands, respectively, whereas Asio is widely distributed). The Asionae differ osteologically from other owls by having: 1) the anterior rim of the internal trochlea not protruding more anteriorly than the anterior rim of the external trochlea, 2) the external calcaneal ridge bent posteriorly, 3) bony loop broad and 4) tubercle for Musculus tibialis antiquus displaced externally 83. The most ancient fossil having this combination of characters Intulula tinnipara, has been dated from the Early Miocene (23.7–16.4 mya) 84. We therefore used this date as a minimum age for the split between Asio otus (the member of the Asioninae we sampled) and its closest relative. As a second calibration point, we used the split between O. pauliani (endemic to Grande Comore, a volcanic island) and its closest relative at 0.5 mya. This date corresponds to the emergence of Grande Comore 51, and is thus the oldest possible age for the colonization of the O. pauliani lineage. We used these two calibration points in combination.

We used BEAST V1.4.6 858687 to estimate the divergence dates within the genus Otus. We assigned the best fitting model, as estimated by MRMODELTEST2, to each of the six loci. We used an exponential distribution for the fossil calibration bound 88. We set the lower bound of the exponential distribution to 16.4 mya, which correspond to the lowest bound of the Early Miocene epoch and the exponential mean to 2.5, so that the 95% distribution probability fell within the 23.7–16.4 mya interval, corresponding to the Early Miocene, which is the epoch for the first Asioninae fossil. For the geological calibration point (emergence of Grande Comore), we used a normal distribution with the mean and standard deviation set to 0.5 mya and 0.1 mya, respectively. We assumed a Yule Speciation Process for the tree prior and an Uncorrelated Lognormal distribution for the molecular clock model 86. We used default prior distributions for all other parameters and ran MCMC chains for 75 million generations, as the effective sample size for some parameter estimates was not large enough using the default length (10 million generations).