Evolutionary patterns and repeated adaptive strategies of deep-sea anemones

Actiniaria, Deep-sea adaptation, Evolutionary pattern, Genome, Gene 40 loss, Phylogenomics 41 42 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint

43 Life in the deep sea, Earth’s largest yet least explored biome, endures perpetual 44 darkness, crushing hydrostatic pressure, near-freezing temperatures, and severely 45 limited food availability 1-3. Over the past decade, high-throughput sequencing and 46 deep-sea sampling have yielded major genomic insights into deep-sea animals, 47 including fishes and crustaceans 4,5. Yet, these studies focus on relatively derived 48 lineages and leave the earliest stages of animal evolution under extreme conditions 49 largely untested. As stemward representatives of Eumetazoan, sea anemones 50 (Cnidaria: Anthozoa, Actiniaria) thrive from tide pools to the 10,000-m-deep hadal 51 trenches 6-9; their simple, skeleton-less anatomy and well-documented plasticity make 52 them uniquely tractable for disentangling cause-and-effect in adaptive change 10. 53 Consequently, investigating how actiniarians cope with deep-sea darkness, pressure, 54 and cold is poised to bridge crucial gaps in our understanding of deep-sea adaptations 55 in early-diverging multicellular animals, thereby offering clearer insights into the 56 fundamental genomic principles that sustain life at our planet’s extremes. 57 Despite their ubiquity, two fundamental questions remain unresolved. First, 58 where did sea anemones originate—shallow reefs or the abyss? The long-standing 59 assumption of a shallow-water origin for most deep-sea fauna has rarely been put to a 60 rigorous test. Isolated counter-examples, such as Campoy et al. 11 on cold-water 61 corals, show that deep origins are possible. A morphology-plus-five-genes study by 62 Rodríguez et al. 12 revealed striking convergence between deep- and shallow-dwelling 63 actiniarians, yet the ancestral habitat of the order and the macro-evolutionary pattern 64 of its bathymetric radiation remain unresolved, hampered by a sparse fossil record and 65 extensive extinction 12-15. Second, regardless of directionality, which genomic 66 mechanisms allow actiniarian lineages to traverse thousands of meters of depth? 67 Single-species genomes hint at confusing adaptive outcomes, including both 68 reinforcement and breakdown of circadian systems, equivocal signals of positive 69 selection, and idiosyncratic gene-family expansions 16-19, underscoring the absence of 70 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint a coherent, order-wide framework. 71 To resolve these issues, we implemented a multi-pronged strategy integrating: (i) 72 broad, depth-balanced genomic sampling, generating new genomes (n = 13) and 73 transcriptomes (n = 15) from 15 species collected between 432 m and 6,000 m (Fig. 1; 74 Supplementary Table 1), and merging these with all publicly available actiniarian 75 datasets on genetics, taxonomy, distribution, and ecology; (ii) a phylogenomics-based 76 test that infers ancestral habitat by contrasting convergent gene-loss patterns, a signal 77 resilient to extinction bias; (iii) genome anatomy pinpoints genetic signatures 78 associated with repeated adaptation, and (iv) functional assays that verify the 79 physiological impact of deep-sea-specific variants. This comprehensive approach 80 allowed us to revise actiniarian deep relationships, demonstrate a shallow-water origin 81 followed by repeated descents, uncover systematic loss of meiotic and light-related 82 genes that accompany a shift toward clonal reproduction, and validate protein-level 83 adaptations that maintain respiration and membrane fluidity under high pressure and 84 near-freezing temperatures, thereby revealing a unifying “loss-optimization-85 innovation” balance strategy of deep-sea adaptation. 86 87

AND DISCUSSION 88 Repetitive DNA drives genome expansion in deep-sea anemones 89 This study provides a comprehensive genomic resource for 15 deep-sea anemone 90 species (Fig. 1; Supplementary Tables 1-3). Using single-molecule real-time 91 sequencing integrated with the BGI-seq platform, we generated high-quality genomes 92 for seven representative species spanning the three major deep-sea superfamilies: 93 Actinostoloidea, Metridioidea, and Actinernoidea (Supplementary Table 4). The 94 resulting genome sizes ranged from 405.3 Mb to 1.52 Gb, consistent with k-mer-95 based estimates (Supplementary Table 4; Supplementary Fig. 1A). Assessment using 96 BUSCO (metazoa_odb10) indicated high assembly completeness, ranging from 97 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 71.4% to 95.7% (Supplementary Table 4). These genomes were annotated with 98 27,110–32,245 predicted protein-coding genes, exhibiting gene set completeness 99 (BUSCO protein recovery) ranging from 80% to 94.1% (Supplementary Table 5). 100 Notably, these gene counts are comparable to those found in shallow-water anemone 101 species (Supplementary Tables 5-6; Supplementary Fig. 1A). These quality metrics 102 confirm that the generated genome assemblies and annotations are robust and suitable 103 for downstream comparative and evolutionary analyses. 104 To further broaden our sampling, we performed deep whole-genome sequencing 105 (161× to 542× coverage) for an additional six deep-sea species and generated 106 transcriptomes for all 15 species to improve annotation and enable expression analysis 107 (Supplementary Tables 2-4). We integrated these data with all publicly available 108 actiniarian genomic (n = 13) and transcriptomic (n = 25) datasets (Supplementary 109 Table 6), achieving comprehensive representation across all recognized superfamilies. 110 This comparative analysis revealed a significant trend towards larger genome sizes in 111 deep-sea species compared to their shallow-water congeners (Supplementary Fig. 112 1A), with genome size exhibiting a strong positive correlation with habitat depth 113 within the deep-sea cohort (p-value = 3.6e-06; Supplementary Fig. 1B). This genome 114 size expansion is primarily driven by the recurrent amplification of repetitive 115 elements, rather than a proliferation of protein-coding genes (Supplementary Table 7; 116 Supplementary Figs. 1A, C, D). This finding points towards the dynamic evolution of 117 genome structure, particularly involving repetitive DNA, accompanying the 118 diversification of anemones into the deep sea (Supplementary Fig. 1E). 119 120 New superfamily phylogenetic relationship 121 Phylogenomic analysis of 1,849 high-quality single-copy orthologous genes from 41 122 species robustly recovered the monophyly of each of the five currently recognized 123 superfamilies and supported the established subordinal classification dividing 124 Actiniaria into Enthemonae and Anenthemonae (Fig. 2A; Supplementary Fig. 2). The 125 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint topology derived from concatenated maximum likelihood analysis was largely 126 congruent with that inferred using coalescent-based methods, with notable exceptions 127 primarily among members of Actinernoidea (Fig. 2A; Supplementary Figs. 2, 3A). 128 Both Quartet Concordance and Quartet Sampling analyses revealed significant gene 129 tree incongruence and strongly conflicting signals across the phylogeny (Fig. 2A). 130 Under the multispecies coalescent (MSC) model, 5.4% of quartets rejected the 131 incomplete lineage sorting (ILS) model, even under a stringent significance threshold 132 of α = 1e-6, suggesting that both ILS and interspecific gene flow have likely shaped 133 the evolutionary history of sea anemones (Fig. 2B). Despite this underlying 134 complexity, the coalescent-based topology represents the best-supported hypothesis of 135 relationships (Fig. 2C; Supplementary Fig. 3) and is hereafter defined as the species 136 tree. 137 While our species tree supports the traditional Enthemonae/Anenthemonae 138 subdivision, it significantly challenges previous hypotheses regarding the 139 relationships among the three superfamilies within Enthemonae (Fig. 2D). Contrary to 140 the prevailing view placing Actinostoloidea as the sibling group to a clade uniting 141 Actinioidea and Metridioidea 12, our analyses strongly support Metridioidea as the 142 earliest diverging lineage within Enthemonae (Fig. 2D). Likelihood mapping analysis 143 demonstrates that this topology is overwhelmingly supported by the underlying gene 144 set (Fig. 2E), indicating that the conflict between previous perspective and our results 145 likely stems from discordant signals between mitochondrial and nuclear datasets, 146 rather than a lack of phylogenetic signal. 147 Inferring recent demographic histories using pairwise sequentially Markovian 148 coalescent (PSMC) analysis revealed that nearly all examined sea anemones 149 experienced substantial population declines during the Pliocene-Pleistocene transition 150 (Supplementary Table 8; Supplementary Fig. 4A). Consistent with our previous 151 observations 19, the onset of these declines generally occurred earlier in deep-sea 152 species compared to their shallow-water counterparts (Supplementary Fig. 4B). This 153 period is characterized by major marine extinctions, particularly affecting large-154 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint bodied fauna, such as mammals, seabirds, and sharks 20,21. It is particularly 155 noteworthy that this environmental perturbation appears to have impacted anemone 156 populations in the relatively stable deep-sea environment preferentially or at an earlier 157 stage. 158 159 Interrogation of ancestral origin status 160 The nested phylogenetic structure, with deep- and shallow-water lineages 161 intermingled (Fig. 2A; Supplementary Fig. 2), strongly implies multiple independent 162 transitions of sea anemones between these environments, indicative of repeated 163 evolution. However, determining the predominant direction of these transitions – 164 whether primarily shallow-to-deep or deep-to-shallow – requires resolving the 165 ancestral habitat state of Actiniaria. Given the deep divergence times, significant 166 lineage loss, and paucity of fossil evidence for sea anemones 12-14, traditional 167 phylogenetic character state reconstruction methods are unlikely to yield robust 168 conclusions. We therefore employed an alternative rationale: the probability of 169 multiple independent lineages repeatedly losing the same gene is considerably higher 170 than the probability of them repeatedly gaining the same gene de novo. By comparing 171 the extent of convergent gene loss under alternative ancestral state scenarios, we can 172 infer the most likely ancestral habitat and the primary direction of habitat transitions 173 (Fig. 3A). 174 Applying an optimized method for detecting repeated gene loss tailored to our 175 dataset (Fig. 3B), we tested two competing hypotheses. Under the “Shallow-sea 176 Ancestor (ANC)” hypothesis, we identified 44 gene families repeatedly lost across 177 multiple deep-sea lineages; critically, 29 of these losses were deemed robust, lacking 178 detectable paralogous backups in the respective genomes (Fig. 3B; Supplementary 179 Table 9). Conversely, under the “Deep-sea ANC” hypothesis, only 7 gene families 180 were repeatedly lost across shallow-water lineages, with merely 2 qualifying as robust 181 losses (the remainder having paralogous backups) (Fig. 3B; Supplementary Table 10). 182 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint This striking asymmetry provides compelling evidence favoring the “Shallow-sea 183 ANC” hypothesis, suggesting that sea anemones originated in shallow waters and 184 subsequently underwent multiple, independent colonizations of the deep sea. 185 Further examination of the genes repeatedly lost under the “Shallow-sea ANC” 186 scenario revealed that they encompass nearly the entire molecular toolkit for the 187 circadian clock pathway (Supplementary Table 9). The wholesale, repeated 188 dismantling of such a deeply conserved and fundamental biological module 22-26 189 across disparate deep-sea lineages provides powerful, independent support for a 190 shallow-water origin followed by repeated invasions of the deep, lightless 191 environment. 192 193 Repeated collapse of essential genes in deep-sea lineages 194 Our hypothesis testing revealed repeated losses of key functional genes across 195 independent deep-sea lineages (Supplementary Table 9), which reflected a strategic 196 streamlining, jettisoning functions unnecessary in the deep sea’s stable, resource-197 scarce environment. To gain a systematic understanding of the functional landscape 198 shaped by gene loss, we performed a comprehensive survey across all species in our 199 dataset, identifying not only repeatedly lost genes but also lineage-specific losses 200 within relevant pathways or functional categories. This analysis targeted genes 201 operating within the same pathways as those identified in the repeated loss screen, as 202 well as genes with analogous functions. The results uncovered variable but significant 203 patterns of gene loss within deep-sea anemones impacting crucial biological 204 processes, including circadian rhythms, sensory perception, photoprotection, nutrition, 205 and sexual reproduction. 206 207 Circadian rhythm machinery is largely dismantled 208 Before assessing circadian rhythm loss in deep-sea anemones, we first 209 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint characterized the pathway architecture of the circadian rhythm in Actiniaria. Our 210 analyses confirm the presence of at least seven core circadian components: Cry1, 211 Cry2, Cry-dash, Clock1, Clock2, Cycle, and Dec. Notably, the Per gene appears 212 restricted to Nephrozoa 26-28 (Supplementary Fig. 5). This composition suggests that 213 the sea anemone circadian clock operates via two feedback loops regulated by Cry 214 and Dec transcription factors, respectively (Fig. 4A) 26,29. 215 Genome-wide surveys revealed that shallow-water lineages consistently retain 216 the full complement of these circadian genes (Fig. 4B). In stark contrast, nearly all 217 deep-sea species examined, with the specific exception of those within Clade 1 of the 218 superfamily Metridioidea, have lost the entire circadian pathway toolkit (Fig. 4B). 219 This wholesale loss reflects the absence of predictable light-dark and temperature 220 cycles in most deep-sea habitats. Intriguingly, although species in Metridioidea Clade 221 1 exhibit some reduction in circadian gene copy number compared to shallow 222 relatives, they retain both the Cry- and Dec-based feedback loops, and all constituent 223 genes remain actively transcribed (Fig. 4B; Supplementary Fig. 6), indicating 224 functional rhythm preservation. The four Clade 1 species in our study inhabit depths 225 between 1,566–3,527 m (Fig. 1), well within the aphotic, thermally stable zone 30. 226 Therefore, the retention of circadian machinery in Clade 1 appears to be a lineage-227 specific trait, independent of ambient light or temperature cues at their collection 228 depths. Given that all known members of this clade exhibit pelagic dispersal phases or 229 lifestyles influenced by ocean currents31-34, their retained rhythmicity might be linked 230 to this unique ecological trait. 231 232 Sensory modifications for a lightless, stable world 233 We next investigated sensory system modifications in deep-sea anemones. 234 Repeated loss analyses had already implicated genes involved in thermosensation (Til) 235 and photosensation (Cnidop2, Cnidop3). Our broader genomic survey confirms the 236 specific loss of Til, a key gene for detecting temperature fluctuations 35, across all 237 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint examined deep-sea lineages (Fig. 4C). The opsin gene family, responsible for light 238 detection, comprises five members in sea anemones: Cnidop1, Cnidop2, Cnidop3, 239 Cnidop4, and Cnidop5. Among these, Cnidop1, Cnidop2, and Cnidop3 show marked 240 contractions or complete loss in multiple deep-sea lineages (Fig. 4C). Consistent with 241 compromised photosensitivity, in situ observations during sample collection failed to 242 detect any behavioral response of deep-sea anemones to the lights of our remotely 243 operated vehicle (Supplementary Videos 1-6). We also specifically examined the 244 repertoire of Transient Receptor Potential (TRP) channels, a conserved metazoan 245 channel family involved in diverse sensory modalities including vision, olfaction, 246 mechanosensation, and nociception 36-38. Anemones possess seven TRP subfamilies 247 (Trpa, Trpc, Trpm, Trpml, Trpv, Trpvl, and Trpp), but unlike opsins, the gene content 248 within these subfamilies appears remarkably stable across both shallow and deep 249 lineages, with no evidence for major lineage-specific expansions or contractions 250 (Supplementary Fig. 7). 251 Overall, adaptation of sea anemones to the deep sea involved specific, albeit 252 somewhat limited, modifications to the sensory toolkit, primarily entailing the 253 reduction or complete loss of light and temperature fluctuation sensing capabilities. 254 255 Loss of photoprotection and shifts in nutrition 256 In the perpetual darkness of the deep sea, photoprotective mechanisms essential 257 for surface life become redundant. Accordingly, we observed the complete loss of the 258 Phr (Photolyases) gene, crucial for repairing UV-induced DNA damage 39,40, in all 259 examined deep-sea anemones (Fig. 4C). Concurrently, genes encoding fluorescent 260 proteins (FPs), which can serve photoprotective roles in cnidarians 41-43, are 261 conspicuously absent from all deep-sea lineages (Fig. 4C). While the lack of vibrant 262 coloration in deep-sea anemones is partly due to the absence of photosynthetic 263 symbionts (see below), the loss of FPs represents a direct genomic contribution to this 264 phenotype. 265 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint The phototactic behavior common in shallow-water anemones is often linked to 266 the photosynthetic needs of their algal symbionts 41,44-46. Given the absence of light, 267 do deep-sea anemones lack algal symbionts? To address this question, we screened 268 our whole-genome sequencing reads against a comprehensive database of symbiont 269 Internal Transcribed Spacer 2 (ITS2) tags 47-49. This yielded zero hits for ITS2 270 sequences in all deep-sea species (Fig. 4D), strongly suggesting that endosymbiosis 271 with photosynthetic algae is indeed absent in these lineages. 272 Symbiotic algae in shallow-water anemones are well known to provide essential 273 nutrients to their hosts, promoting growth and reproduction 50-54; their absence implies 274 fundamental shifts in the nutritional ecology of deep-sea anemones. Supporting this, 275 we found that genes encoding two toxins, Tx60a and Tx60b, both localized to the 276 predatory tentacles 55, are lost in the vast majority of deep-sea lineages while being 277 retained in all shallow-water relatives (Fig. 4C, E; Supplementary Fig. 8). As sea 278 anemone toxin repertoires are known to correlate strongly with prey composition 56-58, 279 the loss of these specific toxins likely reflects alterations in diet or feeding strategy in 280 the deep sea. Furthermore, the gene Ox2r, implicated in appetite and feeding 281 regulation 59,60, is also convergently lost across deep-sea lineages (Fig. 4C). 282 Collectively, these findings point towards profound shifts in the feeding and 283 nutritional strategies of anemones upon colonizing the deep sea. 284 285 Erosion of sexual reproduction pathways 286 For sea anemones capable of both sexual and asexual reproduction, nutritional 287 and environmental conditions play a crucial role in determining which reproductive 288 strategy is adopted 61,62. The findings above revealed a series of responses by sea 289 anemones to environmental and nutritional changes following their colonization of the 290 deep sea. To further investigate whether these changes have influenced the 291 reproductive strategies of anemones, we first examined the gene Vasa, which is 292 specifically expressed in germline progenitor cells and serves as a marker gene for 293 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint pedal laceration during asexual reproduction in anemones 63,64 (Fig. 4F). The results 294 showed that all investigated anemones retained both Vasa1 and Vasa2 (Supplementary 295 Fig. 9). 296 Next, we analyzed the meiotic toolkit and revealed significant gene degradation 297 in deep-sea lineages within the superfamilies Actinostoloidea, Metridioidea, and 298 Actinernoidea (Fig. 4C; Supplementary Table 11). Key genes essential for distinct 299 meiotic stages show evidence of loss or pseudogenization, including those critical for 300 meiotic initiation (Meiosin; Ythdc2, Fig. 4F), homologous chromosome pairing 301 (Spo11), sister chromatid cohesion (Rad21), Meiosis I progression (Mpk1), and 302 Meiosis II completion (Mlh3). These losses cripple gamete production, curtailing 303 sexual reproduction 65-71. Loss events were particularly pronounced within the 304 Actinernoidea lineage (Fig. 4C). Histological evidence confirms this shift: species 305 with degraded meiotic genes lack gonads, their tissue regions occupied by cnidom 306 (nematocyst clusters), unlike those with intact meiotic genes (Fig. 4G). Thus, some 307 deep-sea anemones have streamlined their biology, favoring energy-conserving 308 asexual propagation. 309 310 Protein optimization and expansion facilitate adaptability 311 Adaptation to the deep sea is not solely a subtractive process of gene loss; it also 312 involves constructive changes, namely the molecular optimization of existing proteins 313 and the strategic expansion of specific gene families. We investigated both 314 phenomena to obtain a complete picture of adaptive strategies. 315 316 Optimized proteins counteract deep-sea challenges 317 The high hydrostatic pressure and low temperatures characteristic of the deep sea 318 pose severe challenges to protein stability and enzymatic activity 4,72,73. Consequently, 319 adaptive modifications at the amino acid level are expected. However, the vast 320 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint evolutionary timescale covered by our study (~434 Ma; Supplementary Fig. 10) 321

in near-saturation of substitutions within many gene lineages (Supplementary 322 Fig. 11), precluding the reliable application of standard selection pressure analyses 323 (e.g., dN/dS) to detect positive selection or rapid evolution. To circumvent this 324 limitation, we implemented a simplified analytical pipeline focused on identifying 325 deep-sea lineage-specific amino acid variants within functionally conserved protein 326 regions, leveraging information entropy metrics (Supplementary Fig. 12). 327 A total of 169 genes were identified with lineage-specific amino acid 328 substitutions present in all deep-sea lineages (Supplementary Table 12), termed Deep-329 sea Specific Genes (DSGs). DSGs are defined by amino acid substitutions that are 330 conserved within each deep-sea lineage but not necessarily conserved across different 331 lineages. These DSGs are involved in fundamental cellular processes, including 332 metabolism (e.g., Ndufa12), transcription (e.g., Elob), and protein folding (e.g., 333 Hsp90b) (Fig. 5A-B; Supplementary Fig. 13), implicating systemic physiological 334 adjustments to the deep-sea environment. Metabolic functions were particularly 335 enriched, accounting for 23.6% of identified DSGs (Fig. 5A). Notably, all five 336 complexes of the mitochondrial respiratory chain contained constituent genes 337 identified as DSGs (Fig. 5C), underscoring the importance of energy metabolism 338 adaptation. 339 To functionally validate the adaptive significance of these deep-sea specific 340 variants, we experimentally assessed enzyme kinetics for five key respiratory chain 341 DSGs (Fig. 5D). For each gene, enzymes carrying the shallow-water-type amino acid 342 variants exhibited higher activity at 30°C, whereas enzymes with deep-sea-specific 343 amino acid variants consistently performed better at 4°C (Fig. 5E-F; raw data in 344 Supplementary Table 13). This provides compelling functional evidence that these 345 specific amino acid changes are critical determinants of enzyme activity under deep-346 sea relevant temperatures (2–4°C). Given that high hydrostatic pressure often exerts 347 effects on biomolecules analogous to those of low temperature 73-76, these results 348 strongly suggest direct protein-level adaptation to the combined physico-chemical 349 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint challenges of the deep sea. 350 Among the DSGs, we also identified two fatty acid desaturase genes, Sld1 and 351 Degs1 (Fig. 5B; Supplementary Figs. 14-15), responsible for introducing double 352 bonds into fatty acids 77-80, a critical function for maintaining membrane fluidity in 353 cold and high-pressure environments 1,19,74,81. We functionally assessed the Sld1 354 variants using a Yeast spot assay. Yeast strains carrying the deep-sea-type Sld1 355 exhibited significantly enhanced growth at 4°C compared to wild-type yeast or those 356 carrying the shallow-water-type Sld1 (Fig. 5G). This further validates the adaptive 357 benefit conferred by deep-sea specific amino acid substitutions in optimizing protein 358 function. 359 Collectively, the identification of numerous DSGs across diverse functional 360 pathways signifies widespread protein tuning as a key adaptive strategy in deep-sea 361 anemones. These findings also affirm the efficacy of our analytical approach in 362 pinpointing functionally relevant amino acid substitutions across deep evolutionary 363 divergences where traditional methods fail. 364 365 Functional compensation via gene family expansion 366 Beyond optimizing existing proteins and jettisoning non-essential functions, 367 deep-sea anemones also employ gene family expansion as an adaptive mechanism. 368 We identified 42 gene families significantly expanded in deep-sea lineages compared 369 to their shallow-water counterparts, encompassing functions related to 370 neurotransmission, membranes, DNA repair, etc (Fig. 6A). Notably, these expansions 371 often result in total transcript levels for these functions in deep-sea species that are 372 comparable to, or even exceed, those observed in shallow-water counterparts (Fig. 373 6A; Supplementary Fig. 16), suggesting a compensatory role. Intriguingly, the 374 expanded neuronal gene families are heavily biased towards components associated 375 with excitatory signaling (Fig. 6A). This includes adrenergic receptors (e.g., Adrb1, 376 Adrb2), dopamine receptors (e.g., Drd4), muscarinic acetylcholine receptors (e.g., 377 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint Machr), nicotinic acetylcholine receptors (e.g., Chrna10, Chrnb2), neuropeptide 378 receptors (e.g., Npffr2, Sifar), kynurenine synthesis involved in neurotransmitter 379 modulation (e.g., Ido2), and neurotransmitter transporters (e.g., Slc6a11, Slc32a1, and 380 Slc5a7). This pronounced expansion of excitatory signaling components may 381 represent an adaptive counterbalance to the inherently ‘suppressive’ nature of the 382 extreme deep-sea environment on neuronal activity and overall metabolic rates. 383 Furthermore, reinforcing the critical importance of lipid metabolism, the fatty 384 acid desaturase Fads6 not only underwent recurrent gene duplication events across 385 different deep-sea lineages but also accumulated deep-sea specific amino acid variants 386 and exhibited signatures of convergent amino acid evolution (Fig. 6B). This highlights 387 how multiple adaptive mechanisms – gene duplication and protein sequence evolution 388 – can converge on the same pathway to meet the physiological demands of deep-sea 389 life. 390 391 Macroevolutionary patterns of actiniarian diversification across depth 392 To investigate the macroevolutionary patterns of sea anemone diversification across 393 different oceanic depths, we constructed a high-resolution phylogeny incorporating 394 267 actiniarian species (Supplementary Table 14), using the well-supported species 395 tree (Fig. 2A) as a backbone constraint (Fig. 7A). This expanded phylogeny revealed 396 an even more pronounced nested structure of deep-sea and shallow-water clades, 397 underscoring the frequency of habitat transitions between these two major marine 398 realms throughout actiniarian history. This tree also highlighted a tendency for extant 399 deep-sea species to occupy relatively older phylogenetic branches (Fig. 7A). 400 To further dissect the depth-temporal dynamics of anemone evolution, we 401 performed ancestral state reconstruction of habitat depth across the phylogeny. The 402

suggest substantial jumps in both time and depth between early nodes, a 403 pattern corroborated by branch length distributions that strongly implies significant 404 historical extinction or sampling gaps among early-diverging lineages (Fig. 7B, 405 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint Supplementary Fig. 17). A disparity-through-time (DTT) analysis 82, quantifying the 406 variance in habitat depth among contemporaneous lineages relative to the variance 407 within subclades, revealed that observed disparity significantly exceeded that 408 expected under a null model of Brownian motion evolution for a prolonged period, 409 from the estimated crown age (~434 Ma) until approximately 140 Ma (Fig. 7C). Such 410 elevated early disparity could arise from either intense early convergent evolution or, 411 more likely, an artifact created by substantial, biased extinction events. 412 The evaluation of the evolutionary direction of sea anemones revealed that, after 413 approximately 150 Ma, the proportion curves of colonization into deep and shallow 414 waters stabilized, with the frequencies of the two directions becoming roughly equal 415 and remaining constant overall (Fig. 7D). Similarly, analyses of the colonization rates 416 into deep and shallow waters indicated that the overall rates stabilized around 130 Ma 417 (Fig. 7E). While certain species demonstrated the capacity for rapid, large-magnitude 418 shifts in depth habitat, the average rates of colonization into deeper versus shallower 419 waters appear remarkably balanced and constant over the last ~130 million years. This 420

was robust across three different metrics used to quantify evolutionary 421 rates (Fig. 7E; Supplementary Fig. 18). 422 In summary, over the last ~130 million years, sea anemones have stably migrated 423 between deeper and shallower regions at nearly identical frequencies and rates (Fig. 424 7D-E). To further explore habitat transitions, we used SIMMAP 83 to reconstruct the 425 history of transitions between discrete states defined as deep (≥200 m) and shallow 426 (<200 m). The analysis overwhelmingly favored the All-Rates-Different (ARD) 427 model allowing asymmetric rates (Supplementary Table 15). Paradoxically, the 428 ancestral state reconstruction under this best-fit model inferred a deep-sea origin for 429 Actiniaria (Supplementary Fig. 19), directly contradicting the shallow-water origin 430 strongly supported by our independent convergent gene loss analysis (Fig. 3). This 431 apparent conflict is readily reconciled by the extensive loss of early shallow-water 432 lineages inferred from our other analyses (Fig. 7B-C; Supplementary Fig. 17). Such 433 biased extinction would mislead standard phylogenetic ancestral state reconstruction 434 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint methods, creating an artifactual signal for a deep-sea origin, while simultaneously 435 explaining the observation that many extant deep-sea species reside on older 436 phylogenetic branches. The convergent gene loss analysis (Fig. 3), being less 437 susceptible to such sampling biases, likely provides a more accurate picture of the 438 group’s ultimate shallow-water origins. 439 440

441 By integrating genomic analyses of 15 deep-sea sea anemone species with 442 phylogenomic and comparative genomic approaches, this study systematically reveals 443 the evolutionary patterns of Actiniaria from shallow waters into the deep sea and 444 delineates the multi-faceted adaptive strategies underpinning this transition: 445 First, employing a new analytical framework based on convergent gene loss, we 446 provide compelling evidence that sea anemones originated in shallow waters and 447 subsequently undertook multiple, independent invasions of the deep sea. This finding 448 challenges the assumption that deep-sea lineages are ancient relics and offers a new 449 paradigm for investigating the origins of deep-sea fauna. Intriguingly, while extant 450 deep-sea anemone species often occupy older phylogenetic branches, our analyses 451 suggest this is likely an artifact of extensive extinction among early shallow-water 452 lineages – a clear case of “survivor bias”. Furthermore, we find that the overall 453 efficiency of migration of sea anemones towards deeper and shallower habitats has 454 been remarkably symmetrical over the last ~130 million years. This underscores the 455 need for caution when reconstructing ancestral states solely from extant taxa and 456 highlights the pervasive influence of extinction history on perceived 457 macroevolutionary patterns. 458 Second, our research uncovers the diverse molecular mechanisms facilitating sea 459 anemone adaptation to the formidable deep-sea environment. At the genomic level, we 460 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint document significant functional remodeling characteriz ed by the systematic loss of 461 genes associated with circadian rhythms, sensory perception (particularly light and 462 temperature fluctuation), and photoprotection, alongside profound shifts in nutritional 463 and reproductive strategies. These losses represent adaptive streamlining in response to 464 the darkness, thermal stability, and energy limitations of the deep sea. Concurrently, at 465 the gene level, specific amino acid substitutions optimize the function of key proteins, 466 such as those involved in energy metabolism , enhancing their activity under high -467 pressure, low -temperature conditions. Complementing these changes, selective 468 expansions of gene families related to neurotransmission and membrane function likely 469 provide crucial compensatory mechanisms to counteract t he suppressive effects of the 470 deep-sea environment. This balanced strategy – shedding costly or unnecessary 471 functions via gene loss while reinforcing essential processes through protein 472 optimization and selective gene expansion – may represent a general pa radigm for 473 adaptation to extreme environments. 474 In conclusion, this study significantly advances our understanding of the 475 evolution and adaptation of a key deep-sea metazoan group. Moreover, it provides a 476 valuable conceptual framework and methodological reference point for future 477 investigations into the adaptation of diverse life forms to Earth’s largest, yet least 478 explored, biome. 479 480

481 Sample Information 482 Fifteen deep-sea anemone specimens were collected between 2018 and 2019 during 483 scientific expeditions utilizing the manned submersibles Shenhai Yongshi and 484 Jiaolong. These specimens were retrieved from depths ranging from 432 to 6,000 485 meters in the Xisha Trough of the South China Sea and the Yap Trench, including its 486 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint adjacent waters (Supplementary Table 1). Robotic arms were employed for precise 487 and careful collection, ensuring the specimens remained in pristine condition. Upon 488 retrieval to the research vessel, specimens were rinsed with 2 μm-filtered seawater to 489 remove surface sediment, photographed, and labeled. They were subsequently 490 preserved either at -80°C or in ethanol for downstream sequencing analyses. All 491 specimens are curated by the Deep-Sea Science and Engineering Research Institute, 492 Chinese Academy of Sciences, and are securely stored in the institute’s specimen 493 repository. 494 495 Genome sequencing and assembly 496 Genomic DNA was extracted from all deep-sea anemone specimens using the 497 QIAGEN® Genomic DNA Extraction Kit (Cat#13323, Qiagen) according to the 498 manufacturer’s standard protocol. The purity and quality of the extracted DNA were 499 evaluated using a NanoDrop™ One UV-Vis spectrophotometer (Thermo Fisher 500 Scientific, USA) and 1% agarose gel electrophoresis, respectively. DNA 501 concentrations were precisely quantified with a Qubit® 3.0 Fluorometer (Invitrogen, 502 USA). Based on DNA quality assessments, Oxford Nanopore sequencing was 503 performed on four deep-sea sea anemone specimens: Actinostolidae (Unidentified 504 genus, UG) sp2, Hormathiidae (NG) sp1, Hormathiidae (NG) sp2, and Actinernus 505 sp3, using the PromethION platforms. In parallel, PacBio® HiFi sequencing was 506 conducted on three additional specimens: Actinoscyphiidae (UG) sp1, 507 Actinoscyphiidae (UG) sp2, and Actinoscyphiidae (UG) sp3. Additionally, all 508 specimens were subjected to Illumina sequencing on the NovaSeq 6000 platform to 509 generate 150 bp paired-end reads. Illumina reads were filtered using fastp v0.19.6 84 510 with default parameters before downstream analyses. 511 The genome sizes of the 15 sea anemone specimens were first estimated based 512 on quality-controlled Illumina reads using SOAPec v2.01 85. For genomes sequenced 513 via Oxford Nanopore technology, de novo assembly was performed using wtdbg2 514 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint v2.4.1 86 with default parameters, followed by three rounds of polishing with 515 Nextpolish v1.3.1 87 to generate genome drafts. Genomes obtained through PacBio 516 HiFi sequencing were assembled with hifiasm v0.19.8 88 under standard parameters. 517 Potential haplotype redundancies in the genome drafts were resolved using 518 Purge_dups v1.0.1 89 with default settings. For specimens with only Illumina paired-519 end reads, genome drafts were assembled using Platanus v1.2.4 90 with default 520 parameters, and scaffold continuity was improved where applicable using RagTag 521 v2.1.0 91 with its scaffold algorithm. The completeness of all genome assemblies was 522 assessed using BUSCO v5.4.3 with the “metazoa_odb10” database 92. 523 The mitochondrial genomes of all sea anemone specimens were assembled and 524 annotated using MitoFinder v1.4.1 93 with the parameter “-o 5” based on quality-525 controlled Illumina paired-end reads. Additionally, the web service MITOS v2 94 was 526 employed to provide supplementary annotations for the mitochondrial genomes. 527 528 Transcriptome sequencing and assembly 529 We collected tissues representing as many sea anemone cell types as possible (e.g., 530 tentacle, column muscle, gonad, mesenterial filament, etc.). Total RNA was extracted 531 from each tissue using TRIzol (Thermo Fisher Scientific, USA) and then pooled in 532 equal amounts. After quality control of the RNA, libraries were constructed using the 533 NEBNext Ultra RNA Library Prep Kit for Illumina (NEB, USA) and sent to 534 Nextomics (Wuhan, China) for sequencing on the Illumina HiSeq 2000 platform. On 535 average, 6 Gb of 150 bp paired-end reads were generated for each species. 536 All RNA reads were quality-controlled using fastp v0.19.6 84 with default 537 parameters. Transcriptome assemblies were generated using rnaSPAdes v3.12.0 95 538 with standard parameters. Subsequently, open reading frames (ORFs) were predicted 539 from all assembled transcripts using TransDecoder v5.5.0 540 (https://github.com/TransDecoder/TransDecoder). 541 542 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint Genome annotation 543 Repetitive elements were identified using a combination of de novo and homology-544 based approaches. First, de novo libraries were constructed for each species using 545 RepeatModeler v2.0.1 96, and repetitive sequences were identified with RepeatMasker 546 v4.0.7 97. For the homology-based approach, repetitive elements were predicted by 547 searching against Repbase using both RepeatMasker v4.0.7 97 and RepeatProteinMask 548 v1.36 98,99. Tandem repeats were identified using Tandem Repeats Finder v4.07 100 549 with the parameters “2 7 7 80 10 50 500 -d -h -ngs”. 550 Gene model annotation refer to previous studies74,81,101,102 using a combination of 551 ab initio prediction, homology-based protein prediction, and transcriptome-based 552 prediction methods: First, ab initio gene predictions were performed for the deep-sea 553 sea anemones using Augustus v3.2.1 103, with the gene training set of Nematostella 554 vectensis (GCF_000209225.1). Subsequently, genome information for seven 555 Anthozoa species (N. vectensis [GCF_000209225.1], Actinia tenebrosa 556 [GCA_009602425.1], Exaiptasia diaphana [GCF_001417965.1], Actinostola sp. 557 cb2023 [GCA_033675265.1], Acropora digitifera [GCF_000222465.1], Orbicella 558 faveolata [GCF_002042975.1], and Stylophora pistillata [GCF_002571385.1]) was 559 retrieved from NCBI. The protein sets of these five species were mapped to the deep-560 sea anemone genomes using BLAT v. 35 104 with default parameters, and gene models 561 were predicted using GeneWise v2.4.1 105 with default parameters. Finally, transcripts 562 with complete ORFs were mapped to the deep-sea sea anemone genomes using BLAT 563 v. 35 104 and further processed using PASA (Program to Assemble Spliced 564 Alignments) v2.5.2 106 to refine gene structures. All prediction results were integrated 565 using EvidenceModeler v1.1.1 107 with the following weighting scheme: 566 TRANSCRIPT:10, PROTEIN:10, and ABINITIO PREDICTION:1, resulting in the 567 final non-redundant gene set. Functional annotation of all gene models was performed 568 with InterProScan v5.39-77.0 108. 569 570 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint Phylogenetic analyses 571 To investigate the phylogenetic positions of deep-sea and shallow-water anemones, 572 we retrieved genome and transcriptome data for 26 published sea anemones from 573 public databases (Supplementary Table 6). Combined with the 15 deep-sea anemones 574 analyzed in this study, a total of 41 species were included, covering all known 575 superfamilies within Actiniaria. Orbicella faveolata, which belongs to Scleractinia, 576 was used as the outgroup for phylogenetic analysis. 577 Based on the whole-genome dataset from the aforementioned species, 1,849 578 orthologous gene sets were identified through Reciprocal Best Hit (RBH) analysis. 579 These orthologous gene sets’ CDS sequences were aligned using MAFFT v7.407 109 580 with default parameters, and low-quality alignment regions were trimmed with trimAl 581 v1.5.0 110 under parameter “-automated1”. Phylogenetic trees were constructed using 582 IQ-TREE v2.2.2.4 111 with the parameter settings “-m MFP -alrt 1000” for the 583 trimmed sequences, including both gene trees of each molecular marker and a tree of 584 all concatenated markers. Additionally, species tree analysis was performed using 585 ASTRAL v5.5.9 112 with default parameters. 586 For mitochondrial genes, 13 protein-coding genes’ CDS sequences were 587 concatenated into a supergene, and phylogenetic analysis was conducted using the 588 same method described above for gene tree construction. 589 Divergence times for sea anemones were estimated using the MCMCtree 590 program in the PAML package v4.9h 113, based on three time calibration points: the 591 most recent common ancestor (MRCA) of Metridium senile and Anemonia viridis 592 (estimated at ~369–504 Ma), the MRCA of Anemonia viridis and Aiptasia (estimated 593 at ~334–339 Ma), and the MRCA of Orbicella faveolata and Metridium senile 594 (estimated at ~437–539 Ma). These calibration times were obtained from the website 595 TimeTree and confirmed through additional studies 17,19. 596 597 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint Quartet Sampling 598 Based on the species tree, a Quartet Sampling (QS) analysis was performed on all 599 gene trees involved in the ASTRAL v5.5.9 112 analysis using the quartet_sampling.py 600 script (https://www.github.com/fephyfofum/quartetsampling) 114 to evaluate the 601 consistency and branch support of each node in the species tree. The parameter 602 settings were as follows: the number of repetitions was set to 1,000 (-reps 1,000), the 603 log-likelihood threshold was kept at the default value (-lnlike 2), and the minimum 604 overlap of sampled loci for all taxa in a quartet was set to 20,000 (-min-overlap 605 20,000). Using the resampled quartet counts, we calculated the following metrics for 606 each internal branch of the focal tree: Quartet Concordance (QC; consistency of 607 quartet information), Quartet Differential (QD; presence of secondary evolutionary 608 histories), Quartet Informativeness (QI; informativeness of the quartet), and Quartet 609 Fidelity (QF; reliability of individual taxa in the tree). These metrics were used to 610 assess the confidence, consistency, informativeness of internal nodes, and the 611 reliability of each terminal branch. 612 613 Dispute genealogy diagnoses 614 This study carefully addressed all conflicting topologies, which may have arisen due 615 to data selection, algorithmic differences, or other biological processes introducing 616 phylogenetic discordance. 617 For controversial branches in the phylogenetic tree, we used the DiscoVista v1.0 618 115 software to evaluate topological bias and complexity. The specific procedure was 619 as follows: First, we set the parameters “-k1 -m 5” to count the quartet frequencies of 620 disputed branches. Next, in the support assessment of conflicting topologies, we 621 considered 14 monophyly hypotheses (taking into account all discordant topologies 622 between gene trees and species tree, as well as between species tree and previously 623 published phylogenetic trees), including: (1) Actinostoloidea；(2) Actinernoidea；(3) 624 Actinernidae (UG) sp/Actinernus sp2/Actinernus sp3/Actinernus sp4；(4) 625 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint Metridioidea；(5) Actinostoloidea/Actinioidea/Metridioidea；(6) Actinioidea；(7) 626 Edwardsioidea；(8) Edwardsioidea/Actinernoidea；(9) Actinernus sp2/Actinernus 627 sp4/Actinernus sp3；(10) Actinostoloidea/Actinioidea；(11) Actinernus 628 sp4/Actinernus sp3；(12) Actinernus sp3/Actinernus sp4/Actinernidae (UG) sp；(13) 629 Actinioidea/Metridioidea；(14) Actinernus sp4/Actinernidae (UG) sp. Support 630 exceeding 75% was considered strong. 631 In addition, this study employed the concordance factors algorithm in IQ-TREE 632 v2.2.2.4 111 to assess the support for gene sets within the phylogenetic tree topology 633 116. The analysis was conducted with the following parameter settings: -lmap ALL -m 634 GTR+F+R5 -n 0. Among these, GTR+F+R5 was determined as the best-fit model by 635 IQ-TREE v2.2.2.4 111. 636 Gene flow and incomplete lineage sorting (ILS) are two major factors 637 contributing to phylogenetic conflicts 117-120. To comprehensively evaluate the impact 638 of these two factors on the phylogeny inferred in this study, we used the T1 model 639 (representing ILS) in the MSCquartets 121 algorithm to examine all gene trees in the 640 dataset. Specifically, the gene trees of the 1,849 orthologous gene sets were used as 641 input data, and the quartet frequencies (quartet concordance consistency factors, 642 qcCF) for each node in these trees were calculated using MSCquartets 121. Under the 643 T1 model, the function quartetTreeTest was applied to test all quartet counts at four 644 different rejection thresholds (α): 0.01, 0.001, 1e-04, and 1e-06. 645 646 Demographic history 647 This study used the PSMC model 122 to estimate the population history dynamics of 648 deep-sea and shallow-water anemones. First, heterozygous sites were extracted from 649 BAM files generated by mapping NGS reads, using SAMtools v1.10-76-g65c8721 123 650 with the parameters “mpileup -q 20 -Q 20”. Subsequently, the base substitution rate of 651 the species was estimated using the penalized likelihood method implemented in r8s 652 v1.81124, based on the divergence time calibration and branch lengths from the species 653 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint tree topology (Supplementary Table 8). Finally, the PSMC analysis was conducted 654 with the following parameters: -t 15 -r 5 -p “4+25*2+4+6”. The generation time for 655 all species was set to 1 year 17,19. 656 657 The repeated loss of gene families 658 This study proposed a suitable workflow to identify gene families that are commonly 659 lost among multiple lineages, tailored to the characteristics of the data. Detailed 660 protocols are shown in Fig. 3B. 661 Step 1: Gene group 662 This study selected the protein dataset of the following sea anemone species: 663 Actinoscyphiidae (UG) sp1, Actinoscyphiidae (UG) sp2, Actinoscyphiidae (UG) sp3, 664 Hormathiidae (NG) sp1, Hormathiidae (NG) sp2, Actinostolidae (UG) sp2, Actinernus 665 sp3, Actinia tenebrosa, Exaiptasia diaphana, Paraphelliactis xishaensis, Nematostella 666 vectensis, Stichodactyla mertensii, Anemonia viridis, Heteractis crispa, Stichodactyla 667 helianthus, Diadumene lineata, Heteractis magnifica, Phymanthus crucifer, and 668 Entacmaea quadricolor. Gene family clustering analysis was conducted using 669 OrthoFinder v2.3.1 125 with the following parameters: -M msa -T iqtree -s species.tre -670 y -S diamond, where “species.tre” represents the species tree topology of these taxa. 671 The analysis ultimately identified 21,791 N0-level orthogroups. 672 673 Step 2: Initial filtering 674 To preserve as many potentially lost gene families as possible, relatively lenient 675 filtering criteria were applied in this step. The specific conditions were adjusted based 676 on the analysis, as detailed in Fig. 3B. This process resulted in the identification of 677 candidate lost orthogroups. 678 679 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint Step 3: Gene check in all species 680 In this step, the protein sequences of candidate orthogroups were used as queries 681 to manually identify potential orthologs in the genomes of all species, using a 682 combination of BLAT v. 35 104 and GeneWise v2.4.1 105. Results with lengths less 683 than 30% of the query were filtered out. Subsequently, all potential orthologs were 684 further confirmed for homology using RAxML v8.2.13 126 in combination with 685 OrthoFinder v2.3.1 125. All software analyses were conducted with default parameters. 686 687 Step 4: Further filtering 688 To prevent some results from being mistakenly excluded due to poor genome 689 assembly quality, relatively lenient filtering criteria were again applied in this step. 690 The specific conditions were adjusted based on the analysis, as detailed in Fig. 3B. 691 This step preliminarily identified the lost orthologs. 692 693 Step 5: Final inspection and filtering 694 This step relied on transcriptomes and NGS reads to further identify missing or 695 retained genes that may have been overlooked due to poor genome assembly quality 696 (in six deep-sea species). A similar approach can be found in Xu et al. 81 and Wang et 697 al. 74. 698 For transcriptome analysis: Transcripts with complete ORFs were analyzed using 699 TransDecoder v5.5.0 (https://github.com/TransDecoder/TransDecoder) to generate a 700 protein set. The orthologs identified in the previous step were used as queries to 701 search the protein set with Diamond v0.9.24.125 127 in combination with RAxML 702 v8.2.13 126 to evaluate the loss or retention of genes within species. 703 For NGS reads analysis: Conserved regions of the identified orthologs were 704 determined, and their positions in the gene-retained genomes were annotated based on 705 GFF files. Subsequently, NGS reads from the six deep-sea species were mapped to 706 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint these gene-retained genomes using BW A v0.7.12-r1039 (https://github.com/lh3/bwa) 707 128. The presence or absence of a gene in a species was assessed by examining the 708 read coverage in the conserved regions. 709 Finally, the results of both analyses were integrated and filtered based on specific 710 parameters in Fig. 3B to obtain the final set of orthologs 711 712 Step 6: Robust loss checking 713 The final orthologs were used as queries to manually examine the genomes of all 714 25 species using a combination of BLAT v. 35 104 and GeneWise v2.4.1 105. This 715 process assessed whether functionally similar paralogs were retained. For orthologs 716 where no similar paralogs were identified, we considered the corresponding function 717 to be completely lost, and these genes were designated as robust loss genes. 718 719 Investigation of symbionts in sea anemones 720 Based on the previous reports 47-49, we retrieved the most comprehensive database of 721 symbiotic algae associated with cnidarians to date. From this database, we extracted 722 all ITS2 sequences classified as Clade A–I endosymbionts. Using BWA v0.7.12-r1039 723 (https://github.com/lh3/bwa) 128, NGS reads from both deep-sea and shallow-water 724 anemones were mapped to the ITS2 database. Finally, the number of successfully 725 mapped reads was quantified to evaluate whether the sea anemones harbored 726 symbiotic algae. 727 728 Histological staining and in situ hybridization 729 To compare the histological characteristics of gonads between deep-sea and shallow-730 water anemones, this study selected gonadal samples from shallow-water species 731 Exaiptasia diaphana and Actinia tenebrosa, as well as deep-sea species 732 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint Actinoscyphiidae (UG) sp1, Hormathiidae (NG) sp1, Hormathiidae (NG) sp3, and 733 Actinernus sp4, based on sample availability. For shallow-water anemones, 734 conventional paraffin sectioning techniques were used for histological observation. 735 Since deep-sea anemones were preserved as frozen samples, cryosectioning 736 techniques were employed to maximize sample integrity. The preparation of sections 737 and hematoxylin-eosin (HE) staining followed previously published protocols 129 and 738 were appropriately optimized based on the preservation conditions of the samples 739 (e.g., sample fixation, section thickness, etc.). 740 For the in situ hybridization experiment, tissue samples of Exaiptasia diaphana 741 were rinsed in PBS and then fixed in in situ hybridization fixative at 4°C for 12 hours. 742 Subsequently, cubic tissue blocks (approximately 1 cm per side) were dissected from 743 the columnar region containing the gonads. The tissue blocks were dehydrated 744 sequentially in 15% and 30% sucrose solutions. 745 The dehydrated tissues were cryosectioned longitudinally, and the sections were 746 air-dried before being fixed in 4% paraformaldehyde for 10 minutes. The sections 747 were washed three times with PBS (pH 7.4) for 5 minutes each. After air cooling, 748 proteinase K (20 μg/mL) was added, and the sections were digested at 40°C, followed 749 by three washes in PBS. 750 Pre-hybridization was performed at 40°C for 1 hour, followed by probe 751 hybridization at 40°C overnight. After hybridization, the sections were washed 752 sequentially with 2× SSC, 1× SSC, and 0.5× SSC. Next, 60 μL of pre-heated branched 753 probe hybridization solution was applied, and the sections were hybridized at 40°C 754 for 45 minutes, followed by washing. Finally, a signal probe hybridization solution 755 (1:200 dilution) was added, and the sections were incubated at 40°C for 3 hours, with 756 washing steps performed as described above. 757 After probe hybridization, the sections were counterstained with DAPI staining 758 solution in the dark for 8 minutes to label nuclei. The sections were mounted and 759 observed under a Nikon upright fluorescence microscope. Images were captured using 760 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint the following excitation and emission wavelengths: DAPI (359 nm/457 nm, blue 761 fluorescence); iF488-Tyramide (491 nm/516 nm, green fluorescence); iF546-762 Tyramide (541 nm/557 nm, red fluorescence). 763 764 Deep-sea specific genes 765 In investigating amino acid-level adaptations in proteins, traditional methods based on 766 selection pressure analyses to identify positive selection or rapid evolutionary events 767 are not suitable for the long evolutionary span of sea anemones involved in this study. 768 Therefore, we introduce an analysis pipeline based on information entropy to detect 769 deep-sea lineage-specific amino acid variations in conserved regions of genes. The 770 specific protocols are as follows: 771 Amino acid site alignment: We used MACSE v2.06 130 software to perform codon 772 alignment of the aforementioned orthologous genes in sea anemones. The parameters 773 were set as “-prog alignSequences -gc_def 1”. 774 Assessment of amino acid site conservation: The conservation level of each amino 775 acid site was evaluated using an information entropy-based algorithm. The formula is 776 as follows: 777 H(x)=−∑[P(x)⋅log2(P(x))] 778 In the formula, P(x) represents the frequency of a specific amino acid occurring at 779 a given site. A smaller information entropy value indicates that the site is more 780 conserved. For sites where amino acid deletions (-) occur, if the deletion proportion 781 exceeds 60%, the entropy value of the site is marked as “x”, and the site is excluded 782 from subsequent analyses. 783 Detection of deep-sea-specific amino acid variation sites: Based on the aligned 784 sequence file, we used an in-house script to identify amino acid variation sites specific 785 to each deep-sea lineage (the four deep-sea lineages shown in Fig. 2A) compared to 786 all shallow-water lineages. The criteria for detection were as follows: the test deep-sea 787 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint lineage must share a common amino acid, while all shallow-water species must share 788 another common amino acid. 789 Evaluation of upstream and downstream conservation of variation sites : The 790 conservation of the regions surrounding the variation sites (within a range of 10 amino 791 acids upstream and downstream) was assessed using the Theta-w value. The formula is 792 as follows: 793 Theta-w = (AA value) / (counted AA number) 794 In the formula, the AA value represents the weighted sum of H(x) values for amino 795 acid sites within the statistical region. The weighting rules are as follows: 796  If H(x) < 0.01, the weight is 1. 797  If 0.01 ≤ H(x) ≤ 0.5, the weight is 0.5. 798  If 0.5 0.6. Based 800 on these results, we further selected genes that exhibited specific amino acid 801 variations in all deep-sea lineages. It is important to note that the specific amino acid 802 variations were not required to occur at the same sites across different lineages. The 803 genes identified through this process were defined as Deep-Sea Specific Genes 804 (DSGs). 805 To understand the functional characteristics of DSGs and their potential roles in 806 deep-sea adaptation, we first performed Gene Ontology (GO) annotation for all DSGs 807 and categorized them according to functional categories. Subsequently, we used the 808 KOBAS platform 131 (http://bioinfo.org/kobas/) to conduct Kyoto Encyclopedia of 809 Genes and Genomes (KEGG) pathway annotation and enrichment analysis for these 810 genes. Enrichment results with FDR-corrected p-values less than 0.05 were 811 considered significant. Next, based on the pathway classification standards provided 812 by the KEGG Pathway Database (https://www.kegg.jp/kegg/pathway.html), the 813 annotated genes were categorized and statistically analyzed. 814 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 815 Preparation of yeast expression strains and enzyme activity assay 816 Plasmid construction 817 In this study, yeast expression plasmids were constructed for both wild-type and 818 mutant forms of Ndufa12, Sdhd, Cytb, Atp6v1a, and Atp6v0c, as well as Sld1 (deep-819 type) and Sld1 (shallow-type). Ndufa12, Sdhd, and Atp6v0c were derived from the 820 cold-adapted Hormathiidae (NG) sp2 (sample ID: HK2SQW55). The mutant forms 821 included Phe22 mutated to Trp (Ndufa12), Gly81 mutated to Ala (Sdhd), and Ser20 822 mutated to Ala (Atp6v0c). Cytb was derived from the cold-adapted Actinoscyphiidae 823 (UG) sp3 (sample ID: HK2SQW43), with three mutations: Ile304Leu, Phe333Leu, 824 and His375Arg. Atp6v1a was derived from the cold-adapted Actinernus sp3 (sample 825 ID: HK2SQW54) with one mutation, Lys477Thr. Sld1 (deep-type) was derived from 826 the cold-adapted Hormathiidae (NG) sp2 (sample ID: HK2SQW55), while the cold-827 intolerant Sld1 gene was derived from the cold-intolerant species Diadumene lineata 828 (GenBank: GCA_918843875.1). All sea anemone samples, except for Diadumene 829 lineata, are stored at the Institute of Deep-Sea Science and Engineering, Chinese 830 Academy of Sciences. The DNA sequences of the wild-type genes were codon-831 optimized and synthesized by Sangon Biotech (Shanghai, China). These sequences 832 were cloned into the pRS416 vector, flanked by the TEF1 promoter and CYC1 833 terminator. The mutant constructs were generated by introducing mutations into the 834 wild-type sequences via PCR. All plasmids were further validated by sequencing. 835 836 Yeast transformation 837 A single fresh colony was inoculated into the appropriate medium and cultured 838 overnight at 30°C. The culture was then diluted to an OD600 of 0.1 and transferred to 839 5 mL of fresh medium, followed by incubation at 30°C until the OD600 reached 840 approximately 0.6. Cells were collected and washed with 1 mL sterile ddH2O and 500 841 μL of 0.1 M LiOAc, then resuspended in 100 μL of 0.1 M LiOAc. Next, 20 μL of the 842 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint cell suspension was mixed with 80 μL of transformation buffer (containing 58.6 μL 843 50% PEG3350, 7.7 μL 1 M LiOAc, 9.0 μL DMSO, and 4.7 μL ssDNA) and the 844 plasmid for gene expression. The mixture was gently pipetted to mix thoroughly and 845 incubated at 30°C for 35 minutes, followed by a 15-minute heat shock at 42°C. The 846 cells were centrifuged to remove the supernatant, resuspended in 200 μL sterile 847 ddH2O, and spread onto the selective solid medium. The plates were incubated at 848 30°C for 3 days. 849 850 Respiratory chain complex activity assay 851 A fresh single colony was inoculated into 5 mL of liquid medium and cultured 852 overnight at 30°C. The culture was then transferred to 100 mL of liquid medium and 853 diluted to an OD600 of 0.1, followed by incubation at 30°C until the OD600 reached 854 0.8. Cells were then collected and resuspended in 1 mL of 1× PBS buffer (P1020, 855 Solarbio), followed by centrifugation at 5000×g for 1 minute at room temperature. 856 This process was repeated twice. Mitochondria were extracted using a yeast 857 mitochondrial extraction kit (EX2900, Solarbio). The isolated mitochondria were 858 disrupted by sonication at a power of 200 W, using 5-second pulses with 10-second 859 intervals, repeated 15 times. Protein quantification of the respiratory chain complexes 860 obtained from the lysed mitochondria was performed using a BCA protein assay kit 861 (23225, Thermo Fisher). The activities of three electron transport chain complex 862 proteins NDUFA12, SDHD, and CYTB were measured using microassay methods 863 according to the manufacturer’s instructions. NDUFA12: Mitochondrial respiratory 864 chain complex I activity assay kit (BC0515, Solarbio). SDHD: Mitochondrial 865 respiratory chain complex II activity assay kit (BC3235, Solarbio). CYTB: 866 Mitochondrial respiratory chain complex III activity assay kit (BC3245, Solarbio). 867 Protein activity assays were conducted under two temperature conditions, 30°C and 868 4°C, as specified by the experimental design. Finally, enzymatic activities of 869 mitochondrial electron transport chain complexes were measured at the corresponding 870 temperature using a multifunctional microplate reader. Each group included three 871 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint technical replicates. The complex activity values were normalized as follows: Relative 872 enzymatic activity = measured value / mean value of three technical replicates of the 873 wild type. 874 875 Vacuolar enzyme activity assay 876 A fresh single colony was inoculated into the appropriate medium and cultured 877 overnight at 30°C. The culture was then diluted to an OD600 of 0.1 and transferred to 878 5 mL of fresh medium, followed by further incubation at 30°C until the OD600 879 reached approximately 0.8. Two tubes of 1 mL culture were collected and centrifuged 880 to remove the supernatant. The cells were washed with 1 mL of sterile PBS buffer, 881 centrifuged again to remove the supernatant, and this process was repeated twice. The 882 cells were then resuspended in 800 μL of PBS buffer. Yeast cells were then stained 883 using LysoSensor Green DND-189 (L7535, Thermo Fisher). A 4 μL aliquot of the dye 884 was added to the resuspended cells, mixed thoroughly, and incubated separately at 885 4°C and 30°C for 6 hours. After incubation, the cells were centrifuged to remove the 886 supernatant and washed three times with 1 mL of sterile PBS buffer. The cells were 887 then resuspended in 1 mL of sterile PBS and immediately observed under a Nikon 888 AXR laser scanning confocal microscope for fluorescence imaging. The green 889 fluorescence protein (GFP) was excited using a 488 nm argon laser, and the emission 890 spectrum was collected between 499 and 542 nm. Image data acquisition and 891 quantitative analysis were performed using Fiji software to ensure image quality and 892 reproducibility. 893 894 Yeast spot assay 895 A single fresh colony was inoculated into SC-URA liquid medium and cultured 896 overnight at 30°C. The culture was then adjusted to an OD600 of 1.0. Yeast strains 897 were either directly spotted or subjected to 10-fold serial dilutions prior to spotting 898 onto SC-URA solid medium. The plates were incubated at 30°C and 4°C for several 899 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint days, after which photographs were captured. 900 901 Gene family expansion 902 This analysis involved 7 deep-sea anemones and 10 shallow-water anemones. Whole-903 genome proteins of all species were clustered using OrthoFinder v2.3.1 125. Based on 904 the clustering results at the N0 level, we scored the gene family expansions for each 905 deep-sea anemone relative to the 10 shallow-water anemones using the Z-score. 906

with a Z-score > 1.96 were considered significantly expanded. Gene families 907 that were significantly expanded in at least 6 out of the 7 deep-sea anemones were 908 defined as deep-sea-expanded gene families. These results were subsequently 909 validated in Metridiidae (UG) sp, Hormathiidae (NG) sp3, and Hormathiidae (NG) 910 sp4. The validation method followed Step 3 in the “The repeated loss of gene 911 families” section. Next, these expanded gene families were categorized based on their 912 functions using GO and KEGG annotations. 913 914 Transcriptome expression analysis 915 To evaluate gene transcription levels, this study used HISAT2 v2.2.1 132 to align 916 quality-controlled RNA-seq reads to the reference genome. Subsequently, based on 917 the gene annotation information of the species, the StringTie v2.1.1 133 software was 918 used to calculate paired reads counts for each gene and compute fragments per 919 kilobase of exon per million mapped reads (FPKM) for subsequent transcript 920 quantification analysis. 921 In addition, this study also considered normalizing transcriptional values using 922 internal controls. Based on previous studies 134-139, we retrieved 19 candidate genes 923 that could serve as internal control genes, including 18S rRNA, 28S rRNA, 40S 924 rRNA, 60S rRNA, and Actin, among others (Supplementary Table 16). Based on the 925 expression stability of these genes across species (Supplementary Fig. 20), we 926 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint selected 40S_3, 40S_4, and 40S_9 (gene symbols were represented in Supplementary 927 Table 16) as internal controls for comparative transcriptomic analyses. 928 929 Phylogeny of 267 sea anemones 930 To construct a higher-resolution phylogenetic tree of Actiniaria in order to better serve 931 macroevolutionary analyses, this study retrieved all available Actiniaria data from 932 public databases. After data cleaning, 267 species were retained for subsequent 933 analyses (Supplementary Table 17). Orbicella faveolata was selected as the outgroup 934 for phylogenetic analysis. Using the time-calibrated species tree obtained above as a 935 backbone, this study inferred the phylogenetic relationships of 267 Actiniaria species 936 based on concatenated sequences of 12S rRNA, 16S rRNA, Cox3, Cox1, 18S rRNA, 937 and 28S rRNA. Phylogenetic analysis was conducted using IQ-TREE v2.2.2.4 111 with 938 the parameter “-m MFP -alrt 1000”. Divergence times for each node were 939 subsequently estimated using the MCMCTree program in PAML v4.9h 113. 940 941 Ancestral status assessment 942 To evaluate the evolutionary patterns of sea anemones in the “time-depth” dimension, 943 we collected depth distribution data for the aforementioned 267 species. After 944 verifying the distribution data, the median depth was calculated for each species and 945 used as its depth parameter (Supplementary Table 14). 946 Based on the phylogenetic relationships and depth distribution data of these 947 species, we used the fossilBM 140 program to infer the distribution states at all nodes 948 in the evolutionary tree under the BMVT model. This algorithm is considered to 949 better accommodate data missingness in large-scale phylogenetic studies 140-142. The 950 analysis was conducted with 2,000,000 Markov chain Monte Carlo (MCMC) 951 iterations, sampling every 2,000 iterations. The first 200,000 generations (i.e., the first 952 10%) were discarded as burn-in. The remaining samples were then aggregated for 953 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint further analysis. 954 Finally, using the high-resolution time-calibrated tree of sea anemones, we 955 conducted a disparity-through-time (DTT) analysis 82 in the R package geiger 143 to 956 evaluate the evolutionary patterns of sea anemones along the “depth” axis. 957 Additionally, the analysis included 10,000 random simulations to assess the expected 958 evolutionary trajectory of depth distribution under unconstrained conditions. 959 960 Evolutionary trend assessment 961 First, we defined colonization events as follows: when the depth of the parent node is 962 shallower than that of the child node, it is considered a shallow-to-deep colonization 963 event; conversely, when the parent node’s depth is deeper than the child node’s, it is 964 considered a deep-to-shallow colonization event. Colonization events are assumed to 965 occur with a uniform probability over the time interval between the parent and child 966 nodes. All evaluations were conducted within a 10 Ma non-overlapping sliding 967 window (step size = 10 Ma). 968 Based on the 900 trees collected from the BMVT model, we calculated the 969 proportion of “deep-to-shallow” and “shallow-to-deep” events within each sliding 970 window for each tree. Finally, the results from all 900 trees were aggregated, and the 971 95% confidence intervals were calculated. 972 We also evaluated the mobility rates of “deep-to-shallow” and “shallow-to-deep” 973 colonization events, specifically the average depth change per unit of time for sea 974 anemones. Three statistical metrics were used in this study to reflect the evolutionary 975 rates: 976 R=(D1-D2) / (|t1-t2|) 144 977 V=(ln(D1) - ln(D2)) / (|t1-t2|) 144,145 978 H=(ln(D1) - ln(D2)) / (S*|t1-t2|) 144,146 979 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint R: Evolutionary rate, calculated as the average depth change per unit time; V: 980 Darwinian rate, defined as the rate of change in the natural logarithm of depth; H: 981 Haldane rate, incorporating the variance of depth changes among sampled trees; D1: 982 Average depth (m) of the parent node; D2: Average depth (m) of the child node; t1: 983 Time (Ma) of the parent node; t2: Time (Ma) of the child node; S: Standard deviation 984 of ln(D1) and ln(D2) across all 900 sampled trees for the same node. 985 986 SIMMAP analysis 987 We defined the status of all sea anemones using 200 m as the boundary between deep 988 and shallow seas 1,19,74. Subsequently, we used the simmap function (via the ace 989 function in the R package) to evaluate the fit of three models: ER (equal-rates model), 990 SYM (symmetrical model), and ARD (all-rates-different model) 83 and reconstructed 991 the inhabitant state of all nodes. 992 993 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint ACKNOWLEDGMENTS 994 We acknowledge the crews of the research vessel Tansuo 1 and the pilots of the HOV 995 Shenhai yongshi. We would like to express our sincere gratitude to Professor Zhiyong 996 Yue from Xi'an International University for his assistance in experimental exploration. 997 We are also indebted to our collaborators, Baosheng Wu, Huishan Yue, Xueli Gao, 998 Zhenhong Pan, Wenshi Xue, and Chenyi Yang, for their instrumental assistance with 999 bioinformatics and experimental analyses. This project was supported by the National 1000 Key R&D Program of China (2022YFC3400300, 2023YFC2809300, 1001 2016YFC0304905), by the Fundamental Research Funds for the Central Universities, 1002 Northwestern Polytechnical University (D5000220464), the National Natural Science 1003 Foundation of China (32100367, 32370452), the 1000 Talent Project of Shaanxi 1004 Province to Kun Wang and Qiang Qiu, and Shaanxi Postdoctoral Research Project 1005 (2023BSHGZZHQYXMZZ53). 1006 1007 AUTHOR CONTRIBUTIONS 1008 C.F., H.Z., Q.Q., and K.W. conceived the study. H.Z. and Y .Z. collected the sea 1009 anemone. Y .Z. performed the morphological validation experiments of the sea 1010 anemone samples. P.X. performed the genome assembly and genome annotation. P.X., 1011 C.L., W.X., C.Z., M.H., Ye.L., Yu.L., J.Z., T.Q., Y .Y ., and P.L. performed the data 1012 analysis. X.W., P.X., H.S., Y .Z., and L.Q. performed the yeast gene editing 1013 experiments. C.F., K.W., P.X., and X.W. wrote the manuscript. All authors provided 1014 comments and approved the manuscript for submission and publication. 1015 1016 1017 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint COMPETING INTERESTS 1018 The authors declare no competing interests. 1019 1020 DATA AVAILABILITY 1021 All sequencing reads and genome assemblies have been deposited into CNGB 1022 Sequence Archive (CNSA, https://db.cngb.org/cnsa/) 147 with accession number 1023 CNP0007216. The CNSA sample accession numbers for all assembled genomes are 1024 documented in Supplementary Table 4. All the scripts and code of the analysis of this 1025 study are available at https://github.com/Peidong-Xin/deep-seaAnemone. 1026 1027 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint

1028 1 Jamieson, A. The hadal zone: life in the deepest oceans. (Cambridge University Press, 2015). 1029 2 Tyler, P. Ecosystems of the deep oceans. (Elsevier Science, 2003). 1030 3 Higgs, N. D., Gates, A. R. & Jones, D. O. Fish food in the deep sea: revisiting the role of large 1031 food-falls. PloS one 9, e96016 (2014). 1032 4 Xu, H. et al. Evolution and genetic adaptation of fishes to the deep sea. Cell 188, 1393-1408. 1033 e1313 (2025). 1034 5 Zhang, H. et al. The amphipod genome reveals population dynamics and adaptations to hadal 1035 environment. Cell 188, 1378-1392. e1318 (2025). 1036 6 Shi, M., Qi, L. & He, L.-S. Comparative Analysis of the Mitochondrial Genome of 1037 Galatheanthemum sp. MT-2020 (Actiniaria Galatheanthemidae) From a Depth of 9,462 m at the 1038 Mariana Trench. Frontiers in Genetics 13, 854009 (2022). 1039 7 Wolff, T. in Deep Sea Research and Oceanographic Abstracts. 983-1003 (Elsevier). 1040 8 González-Muñoz, R. et al. Sea anemones (Cnidaria: Actiniaria, Corallimorpharia, Ceriantharia, 1041 Zoanthidea) from marine shallow-water environments in Venezuela: new records and an updated 1042 inventory. Marine Biodiversity Records 9, 1-35 (2016). 1043 9 Fautin, D. G., Malarky, L. & Soberón, J. Latitudinal diversity of sea anemones (Cnidaria: 1044 Actiniaria). The Biological Bulletin 224, 89-98 (2013). 1045 10 Reft, A. J. & Daly, M. Morphology, distribution, and evolution of apical structure of nematocysts 1046 in hexacorallia. Journal of morphology 273, 121-136 (2012). 1047 11 Campoy, A. N., Rivadeneira, M. M., Hernández, C. E., Meade, A. & Venditti, C. Deep-sea origin 1048 and depth colonization associated with phenotypic innovations in scleractinian corals. Nature 1049 Communications 14, 7458 (2023). 1050 12 Rodríguez, E. et al. Hidden among sea anemones: the first comprehensive phylogenetic 1051 reconstruction of the order Actiniaria (Cnidaria, Anthozoa, Hexacorallia) reveals a novel group of 1052 hexacorals. PloS one 9, e96998 (2014). 1053 13 Barroso, F. R. G. et al. Insights into the lifestyle and preservation of Arenactinia ipuensis n. gen. et 1054 n. sp.(Anthozoa, Actiniaria) from the Early Silurian (Ipu Formation, Parnaíba Basin, Brazil). 1055 Earth History and Biodiversity, 100017 (2025). 1056 14 McMahon, S., Tarhan, L. G. & Briggs, D. E. Decay of the sea anemone Metridium (Actiniaria): 1057 implications for the preservation of cnidarian polyps and other soft-bodied diploblast-grade 1058 animals. Palaios 32, 388-395 (2017). 1059 15 Seilacher, A., Reif, W.-E. & Westphal, F. Sedimentological, ecological and temporal patterns of 1060 fossil Lagerstätten. Philosophical Transactions of the Royal Society of London. B, Biological 1061 Sciences 311, 5-24 (1985). 1062 16 Zhang, H., Zhou, Y . & Yang, Z. Genetic adaptations of marine invertebrates to hydrothermal vent 1063 habitats. Trends in Genetics (2024). 1064 17 Zhou, Y . et al. Genetic adaptations of sea anemone to hydrothermal environment. Science 1065 Advances 9, eadh0474 (2023). 1066 18 Law, S. T. S. et al. The genome of the deep-sea anemone Actinernus sp. contains a mega-array of 1067 ANTP-class homeobox genes. Proceedings of the Royal Society B 290, 20231563 (2023). 1068 19 Feng, C. et al. The genome of a new anemone species (Actiniaria: Hormathiidae) provides 1069 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint insights into deep-sea adaptation. Deep Sea Research Part I: Oceanographic Research Papers 1070 170, 103492 (2021). 1071 20 Pimiento, C. et al. The Pliocene marine megafauna extinction and its impact on functional 1072 diversity. Nature ecology & evolution 1, 1100-1106 (2017). 1073 21 Melott, A. L. & Bambach, R. K. Analysis of periodicity of extinction using the 2012 geological 1074 timescale. Paleobiology 40, 177-196 (2014). 1075 22 Bell-Pedersen, D. et al. Circadian rhythms from multiple oscillators: lessons from diverse 1076 organisms. Nature Reviews Genetics 6, 544-556 (2005). 1077 23 Müller, W. E., Schröder, H. C., Pisignano, D., Markl, J. S. & Wang, X. Metazoan circadian 1078 rhythm: toward an understanding of a light-based zeitgeber in sponges. Integrative and 1079 comparative biology 53, 103-117 (2013). 1080 24 Reppert, S. M. & Weaver, D. R. Coordination of circadian timing in mammals. Nature 418, 935-1081 941 (2002). 1082 25 Panda, S., Hogenesch, J. B. & Kay, S. A. Circadian rhythms from flies to human. Nature 417, 329-1083 335 (2002). 1084 26 Reitzel, A. M., Tarrant, A. M. & Levy, O. Circadian clocks in the cnidaria: environmental 1085 entrainment, molecular regulation, and organismal outputs. Integrative and Comparative Biology 1086 53, 118-130 (2013). 1087 27 Simionato, E. et al. Origin and diversification of the basic helix-loop-helix gene family in 1088 metazoans: insights from comparative genomics. BMC evolutionary biology 7, 1-18 (2007). 1089 28 Stanton, D., Justin, H. S. & Reitzel, A. M. Step in time: conservation of circadian clock genes in 1090 animal evolution. Integrative and Comparative Biology 62, 1503-1518 (2022). 1091 29 Kawamoto, T. et al. A novel autofeedback loop of Dec1 transcription involved in circadian rhythm 1092 regulation. Biochemical and biophysical research communications 313, 117-124 (2004). 1093 30 Pinet, P. R. Invitation to oceanography. (Jones & Bartlett Learning, 2006). 1094 31 Izumi, T., Ise, Y ., Yanagi, K., Shibata, D. & Ueshima, R. First detailed record of symbiosis 1095 between a sea anemone and homoscleromorph sponge, with a description of Tempuractis rinkai 1096 gen. et sp. nov.(Cnidaria: Anthozoa: Actiniaria: Edwardsiidae). Zoological Science 35, 188-198 1097 (2018). 1098 32 Beaulieu, S. E. Colonization of habitat islands in the deep sea: recruitment to glass sponge stalks. 1099 Deep Sea Research Part I: Oceanographic Research Papers 48, 1121-1137 (2001). 1100 33 Beaulieu, S. Life on glass houses: sponge stalk communities in the deep sea. Marine Biology 138, 1101 803-817 (2001). 1102 34 Young, C. M. in Marine Hard Bottom Communities: Patterns, Dynamics, Diversity, and Change 1103 39-60 (Springer, 2009). 1104 35 Berterame, N. M., Bertagnoli, S., Codazzi, V ., Porro, D. & Branduardi, P. Temperature-induced 1105 lipocalin (TIL): a shield against stress-inducing environmental shocks in Saccharomyces 1106 cerevisiae. FEMS Yeast Research 17, fox056 (2017). 1107 36 Himmel, N. J. & Cox, D. N. Transient receptor potential channels: current perspectives on 1108 evolution, structure, function and nomenclature. Proceedings of the Royal Society B 287, 1109 20201309 (2020). 1110 37 Peng, G., Shi, X. & Kadowaki, T. Evolution of TRP channels inferred by their classification in 1111 diverse animal species. Molecular phylogenetics and evolution 84, 145-157 (2015). 1112 38 Venkatachalam, K. & Montell, C. TRP channels. Annu. Rev. Biochem. 76, 387-417 (2007). 1113 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 39 Thiagarajan, V ., Byrdin, M., Eker, A. P., Müller, P. & Brettel, K. Kinetics of cyclobutane thymine 1114 dimer splitting by DNA photolyase directly monitored in the UV . Proceedings of the National 1115 Academy of Sciences 108, 9402-9407 (2011). 1116 40 Sancar, A. Structure and function of DNA photolyase and cryptochrome blue-light photoreceptors. 1117 Chemical reviews 103, 2203-2238 (2003). 1118 41 Salih, A., Larkum, A., Cox, G., Kühl, M. & Hoegh-Guldberg, O. Fluorescent pigments in corals 1119 are photoprotective. Nature 408, 850-853 (2000). 1120 42 Anthony, K. R. & Fabricius, K. E. Shifting roles of heterotrophy and autotrophy in coral 1121 energetics under varying turbidity. Journal of experimental marine biology and ecology 252, 221-1122 253 (2000). 1123 43 Clarke, D. N. et al. Fluorescent proteins generate a genetic color polymorphism and counteract 1124 oxidative stress in intertidal sea anemones. Proceedings of the National Academy of Sciences 121, 1125 e2317017121 (2024). 1126 44 Strumpen, N. F., Rädecker, N., Pogoreutz, C., Meibom, A. & V oolstra, C. R. High light quantity 1127 suppresses locomotion in symbiotic Aiptasia. Symbiosis 86, 293-304 (2022). 1128 45 Pearse, V . B. Modification of sea anemone behavior by symbiotic zooxanthellae: phototaxis. The 1129 Biological Bulletin 147, 630-640 (1974). 1130 46 RIEMANN‐ZÜRNECK, K. How sessile are sea anemones? A review of free‐living forms in 1131 the Actiniaria Cnidaria: Anthozoa. Marine Ecology 19, 247-261 (1998). 1132 47 Franklin, E. C., Stat, M., Pochon, X., Putnam, H. M. & Gates, R. D. GeoSymbio: a hybrid, 1133 cloud‐based web application of global geospatial bioinformatics and ecoinformatics for 1134 Symbiodinium–host symbioses. Molecular Ecology Resources 12, 369-373 (2012). 1135 48 Shi, T. et al. Untangling ITS2 genotypes of algal symbionts in zooxanthellate corals. Molecular 1136 Ecology Resources 21, 137-152 (2021). 1137 49 Tonk, L., Bongaerts, P., Sampayo, E. M. & Hoegh-Guldberg, O. SymbioGBR: a web-based 1138 database of Symbiodinium associated with cnidarian hosts on the Great Barrier Reef. BMC 1139 ecology 13, 1-9 (2013). 1140 50 Cui, G. et al. Molecular insights into the Darwin paradox of coral reefs from the sea anemone 1141 Aiptasia. Science Advances 9, eadf7108 (2023). 1142 51 Muscatine, L. & Porter, J. W. Reef corals: mutualistic symbioses adapted to nutrient-poor 1143 environments. Bioscience 27, 454-460 (1977). 1144 52 Carlisle, J. F., Murphy, G. K. & Roark, A. M. Body size and symbiotic status influence gonad 1145 development in Aiptasia pallida anemones. Symbiosis 71, 121-127 (2017). 1146 53 Clayton Jr, W. S. Pedal laceration by the anemone Aiptasia pallida. Marine ecology progress 1147 series. Oldendorf 21, 75-80 (1985). 1148 54 Szmant, A. & Gassman, N. The effects of prolonged “bleaching” on the tissue biomass and 1149 reproduction of the reef coral Montastrea annularis. Coral reefs 8, 217-224 (1990). 1150 55 Guruacharya, A. Genomic investigation of cnidarian toxin evolution. (2017). 1151 56 Sachkova, M. Y . et al. The birth and death of toxins with distinct functions: a case study in the sea 1152 anemone Nematostella. Molecular Biology and Evolution 36, 2001-2012 (2019). 1153 57 Columbus-Shenkar, Y . Y . et al. Dynamics of venom composition across a complex life cycle. Elife 1154 7, e35014 (2018). 1155 58 Jaimes-Becerra, A. et al. “Beyond primary sequence”—proteomic data reveal complex toxins in 1156 cnidarian venoms. Integrative and Comparative Biology 59, 777-785 (2019). 1157 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 59 de Lecea, L. et al. The hypocretins: hypothalamus-specific peptides with neuroexcitatory activity. 1158 Proceedings of the National Academy of Sciences 95, 322-327 (1998). 1159 60 Wong, K. K., Ng, S. Y ., Lee, L. T., Ng, H. K. & Chow, B. K. Orexins and their receptors from fish 1160 to mammals: a comparative approach. General and comparative endocrinology 171, 124-130 1161 (2011). 1162 61 Shick, J. M., Hoffmann, R. J. & Lamb, A. N. Asexual reproduction, population structure, and 1163 genotype-environment interactions in sea anemones. American Zoologist 19, 699-713 (1979). 1164 62 Chia, F.-S. in Coelenterate ecology and behavior 261-270 (Springer, 1976). 1165 63 Miramón-Puértolas, P., Pascual-Carreras, E. & Steinmetz, P. R. A population of Vasa2 and Piwi1 1166 expressing cells generates germ cells and neurons in a sea anemone. Nature Communications 15, 1167 8765 (2024). 1168 64 Shao, P.-J., Chiu, Y .-L., Tsai, P.-H. & Shikina, S. Possible germline progenitor cells in extra-1169 gonadal tissues of the sea anemone, Exaiptasia diaphana. Frontiers in Marine Science 10, 1170 1278022 (2023). 1171 65 Ishiguro, K.-i. et al. MEIOSIN directs the switch from mitosis to meiosis in mammalian germ 1172 cells. Developmental cell 52, 429-445. e410 (2020). 1173 66 Li, L. et al. The XRN1-regulated RNA helicase activity of YTHDC2 ensures mouse fertility 1174 independently of m6A recognition. Molecular Cell 82, 1678-1690. e1612 (2022). 1175 67 Munro, C., Cadis, H., Pagnotta, S., Houliston, E. & Huynh, J.-R. Conserved meiotic mechanisms 1176 in the cnidarian Clytia hemisphaerica revealed by Spo11 knockout. Science Advances 9, eadd2873 1177 (2023). 1178 68 Challa, K. et al. Meiosis-specific prophase-like pathway controls cleavage-independent release of 1179 cohesin by Wapl phosphorylation. PLoS genetics 15, e1007851 (2019). 1180 69 Lee, M.-H. et al. Multiple functions and dynamic activation of MPK-1 extracellular signal-1181 regulated kinase signaling in Caenorhabditis elegans germline development. Genetics 177, 2039-1182 2062 (2007). 1183 70 Rogacheva, M. V . et al. Mlh1-Mlh3, a meiotic crossover and DNA mismatch repair factor, is a 1184 Msh2-Msh3-stimulated endonuclease. Journal of Biological Chemistry 289, 5664-5673 (2014). 1185 71 Santucci-Darmanin, S. et al. The DNA mismatch-repair MLH3 protein interacts with MSH4 in 1186 meiotic cells, supporting a role for this MutL homolog in mammalian meiotic recombination. 1187 Human molecular genetics 11, 1697-1706 (2002). 1188 72 Gross, M. & Jaenicke, R. Proteins under pressure: the influence of high hydrostatic pressure on 1189 structure, function and assembly of proteins and protein complexes. European Journal of 1190 Biochemistry 221, 617-630 (1994). 1191 73 Somero, G. N. Adaptations to high hydrostatic pressure. Annual review of physiology 54, 557-577 1192 (1992). 1193 74 Wang, K. et al. Morphology and genome of a snailfish from the Mariana Trench provide insights 1194 into deep-sea adaptation. Nature ecology & evolution 3, 823-833 (2019). 1195 75 Yancey, P. H. Adaptations to hydrostatic pressure in protein structure and organic osmolytes in 1196 deep-sea animals. High Pressure Bioscience and Biotechnology 1, 90-95 (2007). 1197 76 Siebenaller, J. & Somero, G. N. Pressure-adaptive differences in lactate dehydrogenases of 1198 congeneric fishes living at different depths. Science 201, 255-257 (1978). 1199 77 Chen, M., Markham, J. E. & Cahoon, E. B. Sphingolipid Δ8 unsaturation is important for 1200 glucosylceramide biosynthesis and low‐temperature performance in Arabidopsis. The plant 1201 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint journal 69, 769-781 (2012). 1202 78 Mashima, R., Okuyama, T. & Ohira, M. Biosynthesis of long chain base in sphingolipids in 1203 animals, plants and fungi. Future science OA 6, FSO434 (2020). 1204 79 Endo, K., Akiyama, T., Kobayashi, S. & Okada, M. Degenerative spermatocyte, a novel gene 1205 encoding a transmembrane protein required for the initiation of meiosis in Drosophila 1206 spermatogenesis. Molecular and General Genetics MGG 253, 157-165 (1996). 1207 80 Ternes, P., Franke, S., Zä hringer, U., Sperling, P. & Heinz, E. Identification and characterization of 1208 a sphingolipid Δ4-desaturase family. Journal of Biological Chemistry 277, 25512-25518 (2002). 1209 81 Xu, W. et al. Chromosome-level genome assembly of hadal snailfish reveals mechanisms of deep-1210 sea adaptation in vertebrates. Elife 12, RP87198 (2023). 1211 82 Murrell, D. J. A global envelope test to detect non‐random bursts of trait evolution. Methods in 1212 Ecology and Evolution 9, 1739-1748 (2018). 1213 83 Bollback, J. P. SIMMAP: stochastic character mapping of discrete traits on phylogenies. BMC 1214 bioinformatics 7, 1-7 (2006). 1215 84 Chen, S., Zhou, Y ., Chen, Y . & Gu, J. fastp: an ultra-fast all-in-one FASTQ preprocessor. 1216 Bioinformatics 34, i884-i890 (2018). 1217 85 Luo, R. et al. SOAPdenovo2: an empirically improved memory-efficient short-read de novo 1218 assembler. Gigascience 1, 2047-2217X-2041-2018 (2012). 1219 86 Ruan, J. & Li, H. Fast and accurate long-read assembly with wtdbg2. Nature methods 17, 155-158 1220 (2020). 1221 87 Hu, J., Fan, J., Sun, Z. & Liu, S. NextPolish: a fast and efficient genome polishing tool for long-1222 read assembly. Bioinformatics 36, 2253-2255 (2020). 1223 88 Cheng, H. et al. Haplotype-resolved assembly of diploid genomes without parental data. Nature 1224 biotechnology 40, 1332-1335 (2022). 1225 89 Guan, D. et al. Identifying and removing haplotypic duplication in primary genome assemblies. 1226 Bioinformatics 36, 2896-2898 (2020). 1227 90 Kajitani, R. et al. Efficient de novo assembly of highly heterozygous genomes from whole-1228 genome shotgun short reads. Genome research 24, 1384-1395 (2014). 1229 91 Alonge, M. et al. Automated assembly scaffolding elevates a new tomato system for high-1230 throughput genome editing. BioRxiv, 2021.2011. 2018.469135 (2021). 1231 92 Simão, F. A., Waterhouse, R. M., Ioannidis, P., Kriventseva, E. V . & Zdobnov, E. M. BUSCO: 1232 assessing genome assembly and annotation completeness with single-copy orthologs. 1233 Bioinformatics 31, 3210-3212 (2015). 1234 93 Allio, R. et al. MitoFinder: Efficient automated large‐scale extraction of mitogenomic data in 1235 target enrichment phylogenomics. Molecular ecology resources 20, 892-905 (2020). 1236 94 Bernt, M. et al. MITOS: improved de novo metazoan mitochondrial genome annotation. 1237 Molecular phylogenetics and evolution 69, 313-319 (2013). 1238 95 Bushmanova, E., Antipov, D., Lapidus, A. & Prjibelski, A. D. rnaSPAdes: a de novo transcriptome 1239 assembler and its application to RNA-Seq data. GigaScience 8, giz100 (2019). 1240 96 Saha, S., Bridges, S., Magbanua, Z. V . & Peterson, D. G. Empirical comparison of ab initio repeat 1241 finding programs. Nucleic acids research 36, 2284-2294 (2008). 1242 97 Chen, N. Using Repeat Masker to identify repetitive elements in genomic sequences. Current 1243 protocols in bioinformatics 5, 4.10. 11-14.10. 14 (2004). 1244 98 Tempel, S. Using and understanding RepeatMasker. Mobile genetic elements: protocols and 1245 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint genomic applications, 29-51 (2012). 1246 99 Bao, W., Kojima, K. K. & Kohany, O. Repbase Update, a database of repetitive elements in 1247 eukaryotic genomes. Mobile Dna 6, 1-6 (2015). 1248 100 Benson, G. Tandem repeats finder: a program to analyze DNA sequences. Nucleic acids research 1249 27, 573-580 (1999). 1250 101 Li, M. et al. Genomes of Two Flying Squid Species Provide Novel Insights into Adaptations of 1251 Cephalopods to Pelagic Life. Genomics, Proteomics & Bioinformatics 20, 1053-1065 (2022). 1252 102 Hu, M. et al. A chromosome-level genome of the striated frogfish (Antennarius striatus). Scientific 1253 Data 11, 654 (2024). 1254 103 Stanke, M., Diekhans, M., Baertsch, R. & Haussler, D. Using native and syntenically mapped 1255 cDNA alignments to improve de novo gene finding. Bioinformatics 24, 637-644 (2008). 1256 104 Kent, W. J. BLAT—the BLAST-like alignment tool. Genome research 12, 656-664 (2002). 1257 105 Birney, E., Clamp, M. & Durbin, R. GeneWise and genomewise. Genome research 14, 988-995 1258 (2004). 1259 106 Haas, B. J., Zeng, Q., Pearson, M. D., Cuomo, C. A. & Wortman, J. R. Approaches to fungal 1260 genome annotation. Mycology 2, 118-141 (2011). 1261 107 Haas, B. J. et al. Automated eukaryotic gene structure annotation using EVidenceModeler and the 1262 Program to Assemble Spliced Alignments. Genome biology 9, 1-22 (2008). 1263 108 Jones, P. et al. InterProScan 5: genome-scale protein function classification. Bioinformatics 30, 1264 1236-1240 (2014). 1265 109 Katoh, K., Misawa, K., Kuma, K. i. & Miyata, T. MAFFT: a novel method for rapid multiple 1266 sequence alignment based on fast Fourier transform. Nucleic acids research 30, 3059-3066 1267 (2002). 1268 110 Capella-Gutiérrez, S., Silla-Martínez, J. M. & Gabaldón, T. trimAl: a tool for automated alignment 1269 trimming in large-scale phylogenetic analyses. Bioinformatics 25, 1972-1973 (2009). 1270 111 Minh, B. Q. et al. IQ-TREE 2: new models and efficient methods for phylogenetic inference in the 1271 genomic era. Molecular biology and evolution 37, 1530-1534 (2020). 1272 112 Zhang, C., Rabiee, M., Sayyari, E. & Mirarab, S. ASTRAL-III: polynomial time species tree 1273 reconstruction from partially resolved gene trees. BMC bioinformatics 19, 15-30 (2018). 1274 113 Yang, Z. PAML 4: phylogenetic analysis by maximum likelihood. Molecular biology and 1275 evolution 24, 1586-1591 (2007). 1276 114 Pease, J. B., Brown, J. W., Walker, J. F., Hinchliff, C. E. & Smith, S. A. Quartet Sampling 1277 distinguishes lack of support from conflicting support in the green plant tree of life. American 1278 journal of botany 105, 385-403 (2018). 1279 115 Sayyari, E., Whitfield, J. B. & Mirarab, S. DiscoVista: interpretable visualizations of gene tree 1280 discordance. Molecular Phylogenetics and Evolution 122, 110-115 (2018). 1281 116 Minh, B. Q., Hahn, M. W. & Lanfear, R. New methods to calculate concordance factors for 1282 phylogenomic datasets. Molecular biology and evolution 37, 2727-2733 (2020). 1283 117 Galtier, N. & Daubin, V . Dealing with incongruence in phylogenomic analyses. Philosophical 1284 Transactions of the Royal Society B: Biological Sciences 363, 4023-4029 (2008). 1285 118 Edwards, S. V . Is a new and general theory of molecular systematics emerging? Evolution 63, 1-19 1286 (2009). 1287 119 Xu, B. & Yang, Z. Challenges in species tree estimation under the multispecies coalescent model. 1288 Genetics 204, 1353-1368 (2016). 1289 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 120 Wen, D. & Nakhleh, L. Coestimating reticulate phylogenies and gene trees from multilocus 1290 sequence data. Systematic biology 67, 439-457 (2018). 1291 121 Rhodes, J. A., Baños, H., Mitchell, J. D. & Allman, E. S. MSCquartets 1.0: quartet methods for 1292 species trees and networks under the multispecies coalescent model in R. Bioinformatics 37, 1766-1293 1768 (2021). 1294 122 Li, H. & Durbin, R. Inference of human population history from individual whole-genome 1295 sequences. Nature 475, 493-496 (2011). 1296 123 Li, H. et al. The sequence alignment/map format and SAMtools. bioinformatics 25, 2078-2079 1297 (2009). 1298 124 Sanderson, M. J. r8s: inferring absolute rates of molecular evolution and divergence times in the 1299 absence of a molecular clock. Bioinformatics 19, 301-302 (2003). 1300 125 Emms, D. M. & Kelly, S. OrthoFinder: solving fundamental biases in whole genome comparisons 1301 dramatically improves orthogroup inference accuracy. Genome biology 16, 1-14 (2015). 1302 126 Stamatakis, A. RAxML version 8: a tool for phylogenetic analysis and post-analysis of large 1303 phylogenies. Bioinformatics 30, 1312-1313 (2014). 1304 127 Buchfink, B., Xie, C. & Huson, D. H. Fast and sensitive protein alignment using DIAMOND. 1305 Nature methods 12, 59-60 (2015). 1306 128 Li, H. Aligning sequence reads, clone sequences and assembly contigs with BWA-MEM. arXiv 1307 preprint arXiv:1303.3997 (2013). 1308 129 Li, Y . et al. Origin and stepwise evolution of vertebrate lungs. Nature Ecology & Evolution, 1-20 1309 (2025). 1310 130 Ranwez, V ., Douzery, E. J., Cambon, C., Chantret, N. & Delsuc, F. MACSE v2: toolkit for the 1311 alignment of coding sequences accounting for frameshifts and stop codons. Molecular biology and 1312 evolution 35, 2582-2584 (2018). 1313 131 Bu, D. et al. KOBAS-i: intelligent prioritization and exploratory visualization of biological 1314 functions for gene enrichment analysis. Nucleic acids research 49, W317-W325 (2021). 1315 132 Kim, D., Paggi, J. M., Park, C., Bennett, C. & Salzberg, S. L. Graph-based genome alignment and 1316 genotyping with HISAT2 and HISAT-genotype. Nature biotechnology 37, 907-915 (2019). 1317 133 Pertea, M., Kim, D., Pertea, G. M., Leek, J. T. & Salzberg, S. L. Transcript-level expression 1318 analysis of RNA-seq experiments with HISAT, StringTie and Ballgown. Nature protocols 11, 1319 1650-1667 (2016). 1320 134 Gornik, S. G. et al. Photoreceptor diversification accompanies the evolution of Anthozoa. 1321 Molecular biology and evolution 38, 1744-1760 (2021). 1322 135 Bernardet, C., Tambutté, E., Techer, N., Tambutté, S. & Venn, A. Ion transporter gene expression 1323 is linked to the thermal sensitivity of calcification in the reef coral Stylophora pistillata. Scientific 1324 Reports 9, 18676 (2019). 1325 136 Dhar, A. K., Bowers, R. M., Licon, K. S., Veazey, G. & Read, B. Validation of reference genes for 1326 quantitative measurement of immune gene expression in shrimp. Molecular Immunology 46, 1327 1688-1695 (2009). 1328 137 Zhang, S. et al. Identification and validation of reference genes for normalization of gene 1329 expression analysis using qRT-PCR in Helicoverpa armigera (Lepidoptera: Noctuidae). Gene 555, 1330 393-402 (2015). 1331 138 Kosakyan, A. et al. Selection of suitable reference genes for gene expression studies in 1332 myxosporean (Myxozoa, Cnidaria) parasites. Scientific reports 9, 15073 (2019). 1333 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 139 Lehnert, E. M. et al. Extensive differences in gene expression between symbiotic and 1334 aposymbiotic cnidarians. G3: Genes, Genomes, Genetics 4, 277-295 (2014). 1335 140 Silvestro, D. et al. Early arrival and climatically-linked geographic expansion of New World 1336 monkeys from tiny African ancestors. Systematic Biology 68, 78-92 (2019). 1337 141 Farina, B. M., Faurby, S. & Silvestro, D. Dollo meets Bergmann: morphological evolution in 1338 secondary aquatic mammals. Proceedings of the Royal Society B 290, 20231099 (2023). 1339 142 Rolland, J. et al. The impact of endothermy on the climatic niche evolution and the distribution of 1340 vertebrate diversity. Nature ecology & evolution 2, 459-464 (2018). 1341 143 Harmon, L. J., Weir, J. T., Brock, C. D., Glor, R. E. & Challenger, W. GEIGER: investigating 1342 evolutionary radiations. Bioinformatics 24, 129-131 (2008). 1343 144 Hendry, A. P. & Kinnison, M. T. Perspective: the pace of modern life: measuring rates of 1344 contemporary microevolution. Evolution 53, 1637-1653 (1999). 1345 145 Darwin, C. On the origin of species by means of natural selection, or, the preservation of favoured 1346 races in the struggle for life. Harvard Botany Libraries--Biodiversity Heritage Library digitization 1347 project (1859). 1348 146 Haldane, J. B. S. in Selected Genetic Papers of JBS Haldane (Routledge Revivals) 127-132 1349 (Routledge, 2022). 1350 147 Wang, W. et al. The China national GeneBank sequence archive (CNSA) 2024 update. 1351 Horticulture Research, uhaf036 (2025). 1352 1353 1354 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint FIGURES 1355 1356 Fig. 1 | Sampling and sequencing information for deep-sea anemones. 1357 The color of the dots on the left side of the figure, representing different depths, 1358 corresponds to the color of the survey regions shown on the map in the upper right 1359 corner. The submersible depicted in the figure is the “Shenhai Yongshi” manned 1360 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint submersible, which served as the primary executor of the missions in this study. 1361 1362 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 1363 Fig. 2 | Phylogenomic analyses of deep-sea and shallow-water anemones. 1364 (A) The species tree (left) was inferred from 1,849 single-copy orthologous genes, 1365 while the mitochondrial gene tree (right) was reconstructed using 13 mitochondrial 1366 protein-coding genes. Taxa in red indicate deep-sea lineages. The values at each node 1367 represent the QC (Quartet Concordance) / QD (Quartet Differential) / QI (Quartet 1368 Informativeness) scores for internal branches of the species tree. QC > 0.2 represents 1369 strong support, while QC < 0 represents conflicting support. 1370 (B) Multispecies coalescent (MSC) analysis of the species tree inferred from 1371 ASTRAL. Each simplex plot represents the distribution of quartet concordance 1372 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint factors (qcCFs) under the T1 model (ILS) at a given significance level (α). Red 1373 triangles within the plot denote quartets that significantly reject ILS on the species 1374 tree. 1375 (C) Gene tree compatibility. Proportion of focal splits in gene trees that are strongly 1376 (or weakly) supported (or rejected) for the dispute genealogies. Abbreviations: Acsp: 1377 Actinernidae (UG) sp, Asp2: Actinernus sp2, Asp3: Actinernus sp3, Asp4: Actinernus 1378 sp4. 1379 (D) Phylogenetic framework of the five superfamilies within Actiniaria. The 1380 "previous scenario" refers to the superfamily relationships proposed by Rodríguez et 1381 al.12. Cnidom types a–j correspond to: a, spirocysts; b, basitrichs; c, holotrichs; d, 1382 microbasic p-mastigophores; e, atrichs; f, gracile spirocysts; g, microbasic b- and p-1383 mastigophores; h, macrobasic p-mastigophores; i, robust and gracile spirocysts; j, p-1384 amastigophores. 1385 (E) Likelihood mapping statistics showing the proportion of quartets supporting 1386 alternative placements of the three Enthemonae superfamilies. Quartets falling into 1387 the three triangle corners are considered informative. The red-labeled values denote 1388 the topology supported by this study and the corresponding proportion of support. 1389 1390 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 1391 Fig. 3 | Hypothesis testing of evolutionary direction in sea anemones. 1392 (A) Schematic illustration of alternative hypotheses: the “Shallow-sea ANC” scenario 1393 assumes a shallow-water ancestral state, while the “Deep-sea ANC” scenario posits a 1394 deep-sea origin. The rationale of testing is that the probability of losing the same gene 1395 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint independently across multiple lineages is substantially higher than that of independent 1396 gene gain. Therefore, by assessing patterns of shared gene loss among lineages, we 1397 can infer the directionality of repeated habitat shifts and reconstruct the ancestral 1398 distribution of sea anemones. 1399 (B) Analytical framework and results under the two alternative hypotheses, based on 1400 patterns of recurrent gene loss. 1401 1402 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 1403 Fig. 4 | Repeated evolution of gene loss in deep-sea anemones. 1404 (A) A schematic diagram of the gene regulatory network for the circadian rhythm in 1405 sea anemones. 1406 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint (B) The number (circle size) and loss (red ‘X’) of circadian rhythm pathway genes in 1407 sea anemones. The yellow-highlighted Clade 1 corresponds to the red clade 1 in Fig. 1408 2A. 1409 (C) The number and loss (red ‘X’) of genes related to functions of sensory, 1410 photoprotection, feeding, and reproduction in sea anemones. The “FP” refers to 1411 fluorescent proteins. The “Venom” refers to two toxin genes, Tx60a and Tx60b. 1412 (D) Assessment of symbiotic algae in sea anemones. The x-axis represents species, 1413 and the y-axis shows the effective retrieval rate of NGS reads against to ITS2 1414 database. “X” indicates species known to lack endosymbiosis. A blue background 1415 represents deep-sea lineages, while a gray background represents shallow-water 1416 lineages. 1417 (E) Diagram of the expression locations of the venom genes and Fluorescent proteins 1418 in sea anemones. 1419 (F) In situ hybridization results for the Vasa2 and Ythdc2 genes. The sampled region 1420 is the columnar tissue of the gonadal area in Exaiptasia diaphana (refer to Fig. 4E). 1421 (G) Histological HE staining of the gonadal area in deep-sea and shallow-water 1422 anemones. Enlarged views of arrow-marked regions are shown in insets. The 1423

color of the species names indicates their distribution: gray for shallow-1424 sea species and blue for deep-sea species. The results show that no gonadal tissue was 1425 observed in Actinernus sp4 and Hormathiidae (NG) sp3. 1426 1427 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 1428 Fig. 5 | Functional analyses of deep-sea specific genes (DSGs). 1429 (A and B) Functional classification of 169 DSGs based on GO annotations (A) and 1430 KEGG annotations (B). Bar charts represent the number of genes in each category. 1431 (C) Schematic diagram of the electron transport chain, highlighting DSGs (in red) 1432 within each complex. 1433 (D) The mutation sites of five DSGs validated for functionality. Amino acid 1434 substitutions specific to deep-sea sea anemones are marked in red. 1435 (E) Functional activity validation of Ndufa12, Sdhd, and Cytb genes under 4°C and 1436 30°C conditions. The analyses were performed using Saccharomyces cerevisiae 1437 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint (yeast). Mitochondria were extracted from yeast, and the activity of the corresponding 1438 genes was tested in vitro. 1439 (F) Functional activity validation of different genotypes of Atp6v1a and Atp6v0c 1440 under 4°C and 30°C conditions using the yeast experimental system. The fluorescence 1441 in the figure comes from LysoSensor Green DND-189 dye. Brighter fluorescence 1442 intensity indicates higher enzyme activity. 1443 (G) Growth of yeast strains carrying deep-sea and shallow-water variants of the Sld1 1444 gene, as well as wild-type yeast, under 4°C and 30°C conditions. 1445 1446 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 1447 Fig. 6 | Repeated evolution of gene expansion in deep-sea anemones. 1448 (A) The Sankey diagram at the top illustrates the functional classification of expanded 1449 genes. The middle section shows the copy numbers of expanded genes in deep-sea 1450 and shallow-water anemones. Gene numbers are displayed using heatmap colors and 1451 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint are labeled in the figure. “X” indicates a gene number of 0. The bottom section shows 1452 the expression of expanded genes. The pie charts represent the proportion of 1453 transcribed gene copies relative to the total copy number. If all copies are expressed, 1454 the pie chart has a 360° arc. The color of the pie chart represents normalized 1455 expression levels (FPKM). 1456 (B) The duplication and amino acid variations of the fatty acid desaturase gene Fads6. 1457 1458 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint 1459 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint Fig. 7 | Macroevolutionary analyses of sea anemones across depth. 1460 (A) Phylogenetic tree of 267 sea anemones. Shades of blue represent species ’ habitat 1461 depth, with darker tones indicating greater depth. Black blocks denote deep-sea taxa 1462 based on 200 m and 1,000 m depth criterion, respectively. 1463 (B) Ancestral depth reconstruction of sea anemones using the BMVT model. The y -1464 axis indicates habitat depth (m), and the x-axis represents divergence time (Ma). White 1465 lines show results from 900 sampling replicates; the red line represents the mean depth 1466 across replicates. 1467 (C) Disparity-through-time (DTT) analysis of sea anemone habitat depth. The grey area 1468 denotes the 95% confidence interval of depth disparity from 10,000 Brownian motion 1469 simulations; the dashed line indicates the median of simulated values. The red arrow 1470 marks the end of significant deviation between observed and simulated values. 1471 (D and E) Evolutionary trend analyses of sea anemone depth transitions, using 10 Ma 1472 non-overlapping windows. (D) shows the 95% confidence interval of the proportion of 1473 transitions toward deeper or shallower habitats across 900 s imulations. (E) illustrates 1474 the evolutionary rates of depth transitions toward deeper or shallower environments; 1475 the red line indicates the mean value. 1476 1477 preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for thisthis version posted September 20, 2025. ; https://doi.org/10.1101/2025.09.17.676462doi: bioRxiv preprint