The specimens, of which DNA sequences were determined in the present study, are listed in Table 1, and their collection sites are mapped in Figure 1. Most solemyid and thyasirid bivalves were collected by submersibles such as “Kaiko 7000 II”, “Hyper- Dolphin 3000”, and “Shinkai 6500” operated by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC). Acharax japonica from Nabeta Bay, Axinopsis

rubiginosa from off Kawarago, solemyid bivalves from off Ashizuri Cape, and Thyasira sp. from off Inatori were collected by dredging. All samples were frozen and preserved at −80˚C or in 100% ethanol and deposited at JAMSTEC. The specimens, of which DNA sequences were quoted from the DNA Data Bank of Japan (DDBJ), are listed in Table 2. Almost all thyasirid specimens were very small and often damaged during collection. Thus, we gave priority to molecular analyses using whole bodies than to morphological identification and measurements such as counting the number of ctenidial demibranchs.

Total DNA was prepared from the soft tissue using a DNeasy® Tissue Kit (Qiagen GmbH, Hilden, Germany) according to the manufacturer’s protocol.

To amplify the partial fragments of the 18S ribosomal RNA (18S rRNA) and cytochrome c oxidase subunit I (COI) genes, PCR was performed in reaction solutions containing template DNA and KOD Dash (Toyobo Co., Ltd., Osaka, Japan) under the following condition: 1) 30 cycles of denaturation at 94˚C for 30 s, annealing at 45˚C for 5 s, and extension at 74˚C for 10 s. The primers used in the present study are described in Table 3. When PCR amplification under this condition was not successful, PCR was

C, for Solemyidae and Thyasiridae; S, for Solemyidae; T, for Thyasiridae.

performed under the following modified conditions: 2) initial denaturation at 94˚C for 2 min, 5 cycles of denaturation at 94˚C for 30 s, annealing at 48˚C for 1.5 min, and extension at 72˚C for 1 min, followed by 35 cycles of denaturation at 93˚C for 30 s, annealing at 51˚C for 1.5 min, and extension at 72˚C for 1 min, and final extension at 72˚C for 7 min. Alternatively, 3) first PCR was performed under the 1) or 2) condition, and the second PCR was performed under the 1) or 2) condition using primers different from those used in the first PCR. Only for two thyasirid bivalves from off Inatori (INT-1 and INT-2), the first PCR was performed with 1F and 9R primers for 18S rRNA and with LCO1490 and HCO2198 primers for COI under the following condition: initial denaturation at 94˚C for3 min, 35 cycles of denaturation at 95˚C for 45 s, annealing at 55˚C for 3 min, and extension at 72˚C for 1.5 min, and final extension at 72˚C for 7 min; the second PCR was performed with Th1F and Th1R, Th2F and Th2R, and 5F and 9Rn primers for 18S rRNA and with CS2 and CA2 primers for COI under the 2) condition. PCR products were purified using a QIAquick® PCR purification Kit (Qiagen GmbH, Hilden, Germany).

Direct sequencing of the double-stranded PCR product was performed using an ABI PRISM BigDye® Terminator v1.1 Cycle Sequencing Kit (Applied Biosystems Inc., CA, USA) and the primers used for PCR on Model 377 and 377XL DNA sequencers (Applied Biosystems Inc., CA, USA) according to the manufacturer’s directions. Alternatively, direct sequencing was performed using a GenomeLab™DTCSQuick Start Kit on a CEQ™ 2000XL DNA Analysis System (Beckman Coulter Inc., CA, USA) according to the manufacturer’s directions. DNA sequences were aligned with DNASIS (Hitachi Software Engineering) and MEGA 6.0 [31] . All sequences obtained in the present study were registered in DDBJ (accession numbers LC186952-LC187063).

We constructed three trees based on only 18S rRNA sequences, only COI sequences, and concatenated 18S rRNA + COI sequences for Solemyidae and Thyasiridae, respectively. Trees were constructed by the neighbor-joining (NJ) and maximum parsimony (MP) methods using MEGA 6.0 [31] and the PAUP*4.0 beta 10 software [32] , respectively. Genetic distances were computed using the Kimura’s two-parameter method [33] . The reliability of trees was evaluated by producing 1000 bootstrap replicates. The majority-rule consensus MP tree was constructed by conducting a heuristic search based on the 1000 bootstrap replicates with an unweighted ts/tv ratio. The Bayesian tree was constructed using the MrBayes version 3.1 software [34] based on the model evaluated by the MrModel test 2.2 [35] . The best models were SYM + G for 18S rRNA, GTR + G for COI, and GTR + I + G for 18S rRNA + COI in Solemyidae. On the other hand, the best models were SYM + I + G for 18S rRNA, GTR + G for COI, and GTR + I + G for 18S rRNA + COI in Thyasiridae. The Monte Carlo Markov chain (MCMC) length was 5 million generations, and we sampled the chain after every 100 generations. MCMC convergence was assessed by calculating the potential scale reduction factor, and the first 25,000 generations were discarded. The outgroup species Acila castrensis (Bivalvia, Nuculidae) for Solemyidae and Myrtea spinifera (Bivalvia, Lucinidae) for Thyasiridae were used.

In the NJ tree based on 18S rRNA sequences (1300 bp, 253 variable sites, and 119 informative sites), Acharax formed a paraphyletic group composed of three clades, Acharax 1, Acharax 2, and Acharax 3. Moreover, Solemya formed a clade (Figure 2). In the NJ tree based on COI sequences (400 bp, 256 variable sites, and 178 informative sites), Acharax and Solemya formed clades, respectively, with an only exception of Acharax sp. Lau 1 (Figure 3). In the NJ tree based on concatenated 18S rRNA + COI sequences (1700 bp, 464 variable sites, and 284 informative sites), Acharax and Solemya formed clades, respectively, and the Acharax genera were divided into three clades: Subgroups 1, 2, and 3 (Figure 4). The tree showed that the Solemya clade and Subgroup 3 in the Acharax clade diverged in the order of shallow-sea residents, whale-bone residents, and deep-sea vent/seep residents, although this was not well supported by MP bootstrap values and Bayesian posterior probabilities. Subgroups 1 and 2 were constituted

by only deep-sea specimens. Acharax sp. Lau 1 was included in Acharax 1 of the 18S rRNA tree (Figure 2) and in Acharax Subgroup 1 of the 18S rRNA + COI trees (Figure 4), but in the Solemya clade of the COI tree (Figure 3). Solemya sp. SWC2 was included in the Solemya clade of the COI tree (Figure 3), but diverged basally from other solemyid mussels in the 18S rRNA and 18S rRNA + COI trees (Figure 2 and Figure 4). We constructed the three trees, 18S rRNA, COI, and 18S rRNA + COI. The 18S rRNA tree had more taxa than the two other trees, because more data have been registered in DDBJ. The 18S rRNA + COI tree seems the most reliable, because it was shown that larger sequence data provided more reliable tree [36] . The COI tree was consistent with the 18S rRNA + COI tree in that Solemya and Acharax formed each clades, whereas the 18S rRNA tree was consistent with the 18S rRNA + COI tree in division of Acharax into the three subgroups and phylogenetic positions of Acharax sp. Lau 1 and Solemya sp. SWC2.

In NJ trees based on 18S rRNA (793 bp, 231 variable sites, and 118 informative sites), COI (317 bp, 200 variable sites, and 145 informative sites), and concatenated 18S rRNA + COI (1110 bp, 400 variable sites, and 237 informative sites) sequences, Thyasira formed a paraphyletic group (Figures 5-7). The genera, Thyasira and Maorithyas, included specimens that have two demibranchs and symbiotic bacteria. The other genera, Adontorhina, Axinopsis, Axinulus, Leptaxinus, and Mendicula, formed a clade including specimens that have one demibranch and no symbiotic bacteria. Thyasirid bivalves did not diverge in the order of shallow-sea residents, whale-bone residents, and deep-sea vent/ seep residents. Thyasira kaireiae in the Japan Trench and Thyasira sp. off Hatsushima formed a clade as shown by Cluster A in Figure 5, Cluster B in Figure 6, and Cluster C in Figure 7. Thyasira sarsi was included in the Thyasira cluster of the 18S rRNA tree, but diverged basally from other thyasirid bivalves in the COI tree and 18S rRNA + COI

trees. Similarly to the above Solemyidae, the 18S rRNA tree included more data from DDBJ. In Thyasiridae, the three trees, 18S rRNA, COI, and 18S rRNA + COI, were generally consistent, although the phylogenetic position of Thyasira sarsi in the 18S rRNA tree was different from that in the COI and 18S rRNA + COI trees.

Miyazaki et al. [7] indicated that the “evolutionary stepping stone hypothesis” was supported in mytilid mussels, because they diverged in the order of shallow-sea residents, whale-bone residents, and deep-sea vent/seep residents. Moreover, the transition of symbiotic systems in mytilid mussels also supported the hypothesis [7] . Furthermore, Lorion et al. [8] revealed, by investigating many mytilids obtained from organic falls, that the evolutionary process as indicated by the “evolutionary stepping stone hypothesis” had occurred not only once in Mytilidae, but also several times in parallel.

As in the mytilid mussels, splitting in the order of shallow-sea residents, whale-bone residents, and deep-sea vent/seep residents was shown in the Cluster X of the COI tree (Figure 3) and the Subgroup 3 and the Solemya clade of the 18S rRNA + COI tree (Figure 4). This suggested that a part of Acharax and Solemya adapted in parallel to deep-sea environments in the process indicated by the “evolutionary stepping stone hypothesis.”

Acharax sp. Lau 1 and Solemya sp. SWC2 presented markedly divergent phylogenetic positions between the 18S rRNA and COI trees, although we used the same specimen in each taxon for sequencing. The present study cannot explain the discrepancies of their phylogenetic positions between the trees.

Thyasira bivalves did not diverge in the order of shallow-sea residents, whale-bone

residents, and deep-sea vent/seep residents (Figures 5-7) and it suggested that the “evolutionary stepping stone hypothesis” was not supported in this group. However, to evaluate the “evolutionary stepping stone hypothesis” in Thyasiridae, more whale-bone thyasirids have to be investigated, because we used only one whale-bone specimen.

The paraphyletic group composed of Thyasira and Maorithyas included specimens which have two demibranchs and symbiotic bacteria, whereas the clade composed of other genera, Adontorhina, Axinopsis, Axinulus, Leptaxinus, and Mendicula, included specimens which have one demibranch and no symbiotic bacteria. Taking the tree topologies, we assume parsimoniously that the ancestor of Thyasiridae has two demibranchs and symbiotic bacteria, and that the latter genera derived from the former genera. Two demibranchs may be advantageous for symbiosis by increasing the gill surface area where chemoautotrophic bacteria dwell and absorb hydrogen sulfide.

Our phylogenetic analysis showed that T. kaireiae in the Japan Trench (5345 m depth) and Thyasira sp. off Hatsushima (855 - 1173 m depth) were very closely related with each other and might be the same species. If that is the case, this species can be only bivalves that inhabit deep sea with a range of over 4000 m depth.

Thyasira sarsi also indicated a discrepancy in phylogenetic positions between the 18S rRNA and COI trees. Thyasira sp. Fiji was closely related to M. hadalis in the 18S rRNA tree (Figure 5). We cannot determine whether this specimen was misidentified as Thyasira, because we did not have this specimen, and only the 18S rRNA sequence was available in the database.

The “evolutionary stepping stone hypothesis” was supported by two lineages of Solemyidae. However, we could not draw explicit conclusions whether this hypothesis was refuted in the other lineages of Solemyidae and Thyasiridae owing to the lack of whale- bone specimens, especially in Thyasiridae. If the “evolutionary stepping stone hypothesis” is not supported, a new hypothesis is needed to explain their invasion and settlement in deep sea. Therefore, we propose the “Antarctica-origin hypothesis”. In this hypothesis, we assume that benthoses on the narrow continental shelf of the Antarctica are ejected from there and sunk into deep sea by expansion of the ice shelf, and survivors in the deep-sea environments expand their habitats from the Antarctic to worldwide deep sea. Shallow-sea residents around the Antarctica have been tolerable to low water temperature and all Solemyidae and some shallow-water Thyasiridae have already acquired symbiosis. Thus, symbiotic Solemyidae and Thyasiridae around the Antarctica need to acquire only tolerance to high hydrostatic pressure to invade deep- sea environments. The expansion of deep-sea organisms from the Antarctic deep sea to worldwide deep sea is supported by some studies. Held [37] showed that Serolidae (Isopoda) invaded deep-sea environments from the Antarctic region. Embryos of shallow-water urchins around the Antarctica had tolerance to high hydrostatic pressure [38] . Bivalvia, Gastropoda, Amphipoda, and Decapoda around the Antarctica had an ability to live in broader depth than those in the Atlantic [39] .