The genome of Aechmea fasciata provides insights into the evolution of tank epiphytic habits and ethylene-induced flowering

Aechmea fasciata is one of the most popular bromeliads and bears a water-impounding tank with a vase-like rosette. The tank habit is a key innovation that has promoted diversity among bromeliads. To reveal the genomic basis of tank habit formation and ethylene-induced flowering, we sequenced the genome of A. fasciata and assembled 352 Mb of sequences into 24 chromosomes. Comparative genomic analysis showed that the chromosomes experienced at least two fissions and two fusions from the ancestral genome of A. fasciata and Ananas comosus. The gibberellin receptor gene GID1C-like was duplicated by a segmental duplication event. This duplication may affect GA signalling and promote rosette expansion, which may permit water-impounding tank formation. During ethylene-induced flowering, AfFTL2 expression is induced and targets the EIN3 binding site ‘ATGTAC’ by AfEIL1-like. The data provided here will serve as an important resource for studying the evolution and mechanisms underlying flowering time regulation in bromeliads.

A echmea fasciata, also called the urn plant or silver vase, is a common house plant and popular bromeliad. Bromeliaceae contains more than 75 genera and more than 3500 species and is one of the largest families of flowering plants found in the neotropics [1][2][3] . The chromosome number of bromeliads varies little, with most species of Bromelioideae, Puyoideae, Tillandsioideae and Pitcairnioideae being diploid at 2n = 50 and some Tillandsioideae being diploid at 2n = 48 4,5 .One of the most diverse clades from Bromeliaceae, core bromelioids often use CAM photosynthesis and have absorptive foliar trichomes and 'tanks' formed by closely overlapping bases of rosette leaves to impound water and detritus, allowing these plants to absorb water and nutrients on epiphytic perches and rocks and adapt to arid regions 1,[6][7][8] .
The tank habit, CAM photosynthesis, absorptive leaf trichomes, epiphytism and avian pollination are thought to be the key innovations that have allowed bromeliads to spread into the treetops of rainforests, semiarid and arid regions and microsites and become highly diversified 2,9,10 . The tank habit and CAM photosynthesis arose several times independently within the family; for Bromelioideae, CAM photosynthesis arose at the base of Bromelioideae-Puyoideae ca. 10.7 Mya in the Andes/central Chile; 11,12 later, the tank habit coincided with epiphytism, which arose in Bromelioideae ca. 5.6 Mya in the Atlantic Forest region 8 . The tank epiphytic bromelioids had the highest net diversification in Bromeliaceae because CAM photosynthesis was associated with a twofold increase in speciation rate in the tank habit that was five times lower than that in tank-less habits 9 . Whereas Bromelioideae contains 33 genera and ca. 800 species, tank-less Ananas contains 7 species, but tank-habit Aechmea contains 260 species 13 . However, the genomic basis of these functional traits is poorly understood.
Ethylene-induced flowering is a remarkable feature of Bromeliaceae species and is exploited worldwide to promote flowering synchronization 14 . The flowering synchronization is of critical importance in Ananas comosus var. comosus and other ornamental bromeliads cultivation due to the occurrence of natural flowering out of season can cause serious scheduling problems for growers. Although ethylene promotes flower induction in Bromeliaceae species, ethylene commonly delay flowering in many plant species, including rice 15 , pharbitis 16 and Arabidopsis 17 . Ethylene signalling modulates GA-DELLA signalling pathways that delay flowering by repressing the expression of flowering integrator genes such as FLOWERING LOCUS T (FT) and SUPPRESSOR OF CONSTANS OVEREXPRESSION 1 (SOC1) and the floral meristem identity gene LFY 18,19 . As an exceptional case, ethylene signalling activated the expression of FT in Ananas comosus var. comosus 20 and A. fasciata 21 . But the molecular mechanism underlying ethylene-induced flowering in bromeliads remains unexplored.
The publishing of the genome of the famous fruit crop A. comosus var. comosus and A. comosus var. bracteatus provides the possibility to study the genomic basis of divergence between A. fasciata and A. comosus and the rise of tank epiphytic habits 22,23 . Here, we report the sequencing of A. fasciata. Ancestral genome construction and comparative genome analysis show the evolution of chromosome numbers and the loss of orthologous sets. In A. fasciata, the gibberellin receptor gene GID1C-like was duplicated; along with the insertion of 14 and 27 amino acids and multiple nonsynonymous mutations in the duplicated gene pairs relative to AcGID1C-like due to a segmental duplication event followed by mutation, it may affect GA signalling and promote rosette expansion, which allow waterimpounding tank formation. Through analysis of the evolution of the FT family, coupling ChIP-seq, ChIP-qPCR and Y1H, we hypothesized that AfEIL1 can directly bind upstream of AfFTL2 proteins and that AfFTL2 may be the key gene of ethyleneinduced flowering. The resulting genomic information will be helpful for studying the evolution and mechanisms of flowering time regulation in bromeliads.

Results
Genome assembly and annotation. K-mer analysis of the A. fasciata genome estimated that the genome size was 359 Mb, with 1.15% heterozygosity ( Supplementary Fig. 1). To overcome the impact of its high heterozygosity, we generated 40 Gb PacBio reads, 60 Gb Nanopore reads, and 158 Gb Illumina short reads for genome assembly. Using Nextdenovo (https://github.com/ Nextomics/NextDenovo) and Nextpolish 24 , we constructed 528.98 Mb scaffold sequences, with an N50 of 4.68 Mb (Supplementary Table 1). After purging redundant sequences, 352 Mb sequences were used for Hi-C mapping ( Supplementary Fig. 2). Then, we constructed 24 chromosomes with 347 Mb genome sequences, which was 96.66% of the estimated genome size (Fig. 1).
We processed RNA-seq data from roots, flowers, central leaves, mature leaves and central leaves under ethylene treatment to provide transcriptional evidence supporting the annotation and to obtain reliable gene structure annotation. Using the Marker2 pipeline 25 , we predicted the genes of the A. fasciata genome. Gene prediction produced 26,126 protein-coding genes (Supplementary  Table 3). We also identified 1473 pseudogenes and predicted 1,239 transcription factors.
We used LTR-Retriver 27 and BUSCO to estimate genome completeness. The LAI value was 13.94, and 98.4% of the BUSCO orthologues were completed, suggesting a highly complete and reference quality draft genome assembly (Supplementary Table 4). The repeat sequences consisted of 61.72% of the genome (Supplementary Table 5). A total of 0.63% consisted of transposon elements (TEs), and 58.15% of the genome consisted of retrotransposons. The LTR family Gypsy accounted for 24.31% of the genome. LTRs are thought to play a major role in the evolution of chromosome structure, specifically in repeat-dense regions such as centromeres. The LTR distribution of each chromosome was analysed, showing that LTRs were enriched near the ends of most chromosomes (18) and that a few (6) were enriched in the centre of the chromosomes ( Supplementary  Fig. 3). The LTR insertion time was analysed using LTR_retriever, suggesting that there was a burst of LTR insertions ca. 0.91 Mya (Fig. 2a).
Ancestral genome construction. We constructed the ancestral genomes of A. fasciata, A. comosus var. comosus and A. comosus var. bracteatus based on homologous genes detected by Orthofinder 28 according to the BLAST results for proteins (Fig. 2d). The ancestral genome contains 24 ancestral chromosomes and 13,837 ordered protogenes, including 13,119 genes of A. comosus var. comosus and 12,311 genes of A. fasciata and 11,445 genes of A. comosus var. bracteatus. The 24 A. fasciata chromosomes had experienced at least two fissions and two fusions from the ancestral chromosomes, and A. comosus var. comosus and A. comosus var. bracteatus had experienced one fission and formed the current 25 chromosomes. Chr17 of the ancestral chromosomes fissioned into chr5 and chr10 in A. comosus var. comosus and into chr3 and chr11 in A. comosus var. bracteatus, while in A. fasciata, the ancestral chromosome chr17 fissioned into three parts, two of which formed chr5 and chr24, and the last part fused with the ancestral chromosomes chr8 and chr21 into the current chromosome chr1 (Fig. 2e, Supplementary  Fig. 4 and Supplementary Fig. 5). Multiple translocations, inversions, and shifts were identified in the three genomes through synteny alignment with the ancestral genome. In particular, we identified three translocations of ancestral chr9, chr14 and chr21 into the 5' of chr15, forming the current chr23 of A. comosus var. bracteatus.
Gene family expansion and contraction. We identified 15 (Fig. 3b); of the gene families, 99.21%, 98.97% and 99.36% had gene family sizes less than 5 (Fig. 3c). There were 8,092 single copy genes in the three genomes. Compared with those of A. comosus var. comosus and A. comosus var. bracteatus, the number of gene families of size more than 1 was decreased significantly in A. fasciata. Using DupGeneFinder 29 , we predicted the duplicated gene pairs of the genome. As shown in Fig. 3a We also calculated the synonymous substitutions per synonymous site (Ks) of the three genomes. There were similar WGD peaks in the three genomes, corresponding to WGD event occurred 100-120 million years ago; 23 and the divergence peaks between A. comosus var. comosus and A. comosus var. bracteatus were significantly different from the peaks between A. fasciata and A. comosus var. comosus (Fig. 2b, c). Using the single-copy genes of A. fasciata, A. comosus var. comosus and A. comosus var. bracteatus and rice, we estimated the divergence time. Analysis showed that A. comosus var. comosus and A. fasciata diverged ca. 5.843 MYA (Fig. 3e), when the core-Bromeliaceae appeared in the Atlantic rainfall forest 8 . According to the estimated divergence time, we used Café 30 to predict gene family expansion and contraction. There were 514, 1143, and 2733 expanded and 2090, 1067, and 3204 contracted gene families in A. fasciata, A. comosus var. comosus and A. comosus var. bracteatus, respectively.

Candidate genes involved in adaptation to new environments.
Of the expanded families, several may be involved in the adaptation of A. comosus to new environments, including metabolism genes Bromelain and transporter-like genes zinc transporter 10 (ZIP10)-like and disease resistance proteins Receptor kinase-like protein Xa21 (XA21)-like, Germin-like protein (GLP) like and serine/threonine-protein kinase 7 long form homologue (MAIL3)like genes (Fig. 3f). GLPs contribute to broad-spectrum diseases, and XA21 promotes plant innate immunity 31,32 . MAIL3 may be involved in the regulation of root and shoot development by maintaining cell division activity 33 .
The transcription factor families RICESLEEPER2-like and terpene synthase 10 (TPS10)-like were expanded in A. fasciata (Supplementary Fig. 6 and Supplementary Fig. 7). RICESLEE-PER2 is essential for normal plant growth and development 34 . There were 8 RICESLEEPER2-like loci in the ancient genome of A. fasciata. Compared with A. comosus and A. comosus var. bracteatus, two clades of Ricesleeper2-like were expanded in A. fasciata due to the dispersed duplication event after speciation with A. comosus var. comosus. TPS10 is involved in defence against oomycete infection and indirect defence against herbivores by producing mixed signalling to attract natural enemies to plants 35,36 . The TPS10-like gene clusters contained 5, 7 and 12 genes in A. comosus var. comosus, A. comosus var. bracteatus and A. fasciata, respectively. One of the loci, a homologue of rna23024 of A. comosus var. comosus, was expanded to form a cluster with 9 genes.
Candidate genes for tank formation. A. fasciata is tank-forming core bromelioids, while A. comosus var. comosus and A. comosus var. bracteatus are tank-less (Fig. 3d). The gibberellin (GA) receptor gene GID1C was duplicated in A. fasciata due to segmental tandem duplication in a region of chr05, which is a conserved syntenic region with that of chr05 of A. comosus var. comosus and chr11 of A. comosus var. bracteatus (Fig. 4a). In this region, the pentatricopeptide repeat-containing proteins At2g30780-like Af03915 and Af03914 and GID1C-like AfGID1CL1 (Af03916) and AfGID1CL2 (Af03917) were tandemly duplicated gene pairs. However, the duplicated GID1C gene pairs had experienced different mutations in sequence. There were 14 and 27 amino acid insertions in 86-130 amino acids of AfGID1CL2 and AfGID1CL1, respectively, and nonsynonymous mutations at positions 10, 139, 202, 226, 229, 249, 278, 355 and 391 relative to AcGID1C-like (Supplementary Fig. 8 and Fig. 4b). Gene expression analysis revealed that AfGID1CL1 and AfGID1CL2 were expressed in all plant organs and higher in the roots, and the expression level of AfGID1CL1 was one-fold higher than that of AfGID1CL2 ( Supplementary Fig. 9). Overexpression of OsGID1C resulted in a GA hypersensitivity phenotype in transgenic rice 37 . Overexpression of GID1C-like in Arabidopsis resulted in the expansion of rosettes 38,39 . The tandem duplication of the GID1C-like gene increased its expression level and may improve GA sensitivity and promote rosette expansion in A. fasciata, which allows the formation of tanks with closely overlapping leaves. In addition, there were 3 and 6 homologues of DELLA in A. fasciata and A. comosus var. comosus, respectively, which were distributed in conserved syntenic blocks among the chromosomes (Fig. 4c). AfSLR1-like and AcSLR1-like have complete DELLA and GRAS domains but unique motifs, such as 'LEQLD', 'MAMAMG' and 'VVHYNP' (Fig. 4d), different from the previously described 'LEQLE','-MAMGM' and 'TVHYNP' of angiosperm DELLA 40 . AfSLN1-like and AcSLN1-like have a 'DGLLA' motif and a unique 'LERLD' motif, which is also different from the described 'LERLE' motif of angiosperm 'DGLLA' 40 . There is a specific 'AAAAEVEEE GEEAAEE' insertion in the GRAS domains of AfSLR1-like and AcSLR1-like ( Supplementary Fig. 10). Moreover, four homologues lacked DELLA domains and partial GRAS domains in A. comosus var. comosus, including AcRGL1-like, AcRTH1-like, AcSLR1-like1 and AcSLR1-like2. In A. fasciata, the homologues of AcRGL1-like and AcRTH1-like became pseudogenes, and homologues of AcRGL1-like were lost.
The FT family and ethylene-induced flowering. As shown in Fig. 5a, there are 7 FT-like genes in A. fasciata, distributed on 5 chromosomes. Compared with those in A. comosus var. comosus, the locations and gene structures of FT-likes in A. fasciata were conserved; FTL1 and FTL4 are tandemly duplicated genes in both A. fasciata and A. comosus var. comosus. The phylogenetic analysis of FTLs and analysis of amino acid residues showed that AfFTL2 and AcFTL2 are highly homologous to Zcn8 41,42 and SbFT8 43 , which act as florigens in Zea mays and Sorghum bicolor, and harbour similar amide residues in segment B, which is the key amide residue for functioning as a florigen (Fig. 5b). FTL1 shares high similarity with HD3a and RFT1 but does not respond to ethylene treatment. FTL3 is a homologue of ZCN26, and Segment B of the FT proteins varies in different FTs. The two lines of 35 S::AfFTL2 plants flowered early, with one of the lines flowering only with three rosette leaves (Fig. 5g). Using RNA-seq, we analysed the expression of FT-like genes in roots, central leaves, mature leaves, flowers and central leaves under ethylene treatment: AfFTL4 and AfFTL7 were not expressed; AfFTL4 was mainly expressed in flowers; only AfFTL6 was expressed in roots; AfFTL2, AfFTL3, AfFTL5 and AfFTL6 were expressed in central leaves; and only AfFTL2 was dramatically induced by ethylene (Fig. 5c). To investigate the mechanism by which ethylene induces AfFTL2 expression, we conducted AfEIL1-like ChIP-seq and identified the core EIN3 binding motif 'ATGTAC' (Fig. 5e). ChIP-qPCR and Y1H assays validated the binding of EIN3 to AfFTL2 0-500 bp upstream of the promoters (Fig. 5d and Supplementary Table 6). We predicted the ethylene-related cis-elements EBS and ERE in the 2000 bp promoters of the genes (Fig. 5f). Both the FTL3 and FTL2 promoters harbour EBS, but AcFTL2 does not have ERE sites, suggesting that EBS is a conserved element upstream of the FTL2 promoters of A. fasciata and A. comosus var. comosus.
The A. fasciata homologues of five main flowering pathway genes in Arabidopsis and rice were identified by protein alignment. The flowering pathway genes were conserved in bromeliads relative to rice (Fig. 6a). The age pathway genes AfmiR156, AfmiR172, AfTOE1 and AfSPL14 were identified to be involved in A. fasciata flowering time 44,45 . However, in contrast to Arabidopsis and rice, there was an EIL1-dependent pathway to induce FTL2 expression and then downstream signalling genes. In addition, ethylene signalling also induced the expression of homologues of genes involved in plant development, including CAL-A, ERF3, ETR3, EXPA10, LEA5, NAKR3, and LSH6, which shared similar expression with AfFTL2 and synchronized flowering (Fig. 6b). LSH6 plays a critical role in synchronizing flowering 46 . NakR3 may play a role in the transport of FT proteins 47 . CAL-A functions as a floral meristem identity gene 48 . ETR3, ERF03, OTU6, and EXPA10 play roles in plant growth and varied developmental processes 15,49,50 .

Discussion
Bromeliaceae arose in the Guayana Shield ca. 100 Mya 1 . The subfamily Bromelioideae is one of the most diverse clades in Bromeliaceae. The ancestral bromelioids were tank-less, coinciding with the formation of epiphytism, and the core-Bromeliodeae tank habit arose in the Atlantic Forest ca. 5.64 Mya 8 . Using rice as an outlier, we estimated that the divergence time between A. fasciata and rice was 104 Mya and that the divergence time between A. fasciata and A. comosus var. comosus was 5.843 Mya, which is close to the time of the tank-forming core-Bromeliodeae arising in the Atlantic Forest 8 . Although A. comosus var. bracteatus is a variety of pineapple, the introgression of A. macrodontes genes into A. comosus led to genetic differentiation between A. comosus var. bracteatus and A. comosus var. comosus 51,52 . This phenomenon allowed the construction of the ancestral genomes of A. comosus, A. comosus var. bracteatus and A. fasciata. Synteny alignment with the ancestral genome showed that A. comosus var. bracteatus experienced specific translocations on chr23. In addition to multiple translocations, inversions and shifts of chromosomes, A. comosus var. comosus and A. comosus var. bracteatus experienced one fusion, and A. fasciata experienced at least two fissions and two fusions, indicating that changes in genomic structure play key roles in tank epiphytism habit formation. Due to the chromosome fusions of ancestry chromosomes, A. fasciata has 48 (2n = 48) chromosomes and was validated by the root tip squash method (Supplementary Fig. 11). We also observed a karyotype with 2n = 50 due to breakage of chromosome 1 near the centromere, which may be the reason why previous reports described a karyotype with 2n = 50 chromosomes 53 . Transposon elements play key roles in the instability of the genome and genome evolution. In A. comosus var. comosus, 44% of the genome assembly consisted of TEs, and 33% consisted of LTRs, but 69% and 52% of the genome was composed of TEs and LTRs, respectively; in A. comosus var. bracteatus, 64.19% and 48.20% of the genome consisted of TEs and LTRs, respectively; and in A. fasciata, only 58.78% and 33.9% of the genome consisted of TEs and LTRs, respectively. According to the estimation of genome size, A. comosus var. comosus, A. comosus var. bracteatus and A. fasciata were 526, 591 and 359 Mb, respectively. Therefore, the main cause for the significantly smaller size of the A. fasciata genome relative to the A. comosus var. comosus and A. comosus var. bracteatus genomes was the lower insertion of TEs and LTRs. The loss of numerous genes, specifically duplicated genes, would be another reason for the significantly smaller size of the A. fasciata genome relative to that of A. comosus var. comosus and A. comosus var. bracteatus. A total of 1894 orthologue sets were lost in A. fasciata, and coupled with the loss of duplicated genes, the number of predicted protein-coding genes was 7,552 less than that of A. comosus var. bracteatus. It is supposed that WGD pairs were lost in the subsequent millions of years. The duplication of a single gene has a short half-life and occurs continuously, playing roles in key adaptations to environments 54 . According to the expansion and contraction analysis by Café 30 , 1975 gene families were contracted, and only 28 gene families were expanded. The expanded gene families may play key roles in the formation of tank-epiphytic habits in A. fasciata.
With the formation of tank habits, bromeliads invaded the treetops of rainforests. Especially in Atlantic forests of Brazil, they presented with extraordinary biodiversity. The tank habit allows bromeliads to impound and absorb water and nutrients through their leaves, not requiring the uptake of water and nutrients by roots, leading to their degeneration into anchorage. Leaves of most bromeliads are organized in stemless rosettes that allow the development of aquatic and terrestrial ecosystems 55,56 . The duplication of GID1C-like genes and contraction of DELLA-like genes in A. fasciata may promote rosette expansion by affecting GA signalling, which allowed the rosettes to overlap closely and impound water. GA is a hormone that regulates various developmental processes, such as stem elongation, leaf expansion, flowering, seed germination and pollen development [57][58][59] . The effect of GA on controlling leaf expansion has been studied widely [60][61][62][63][64] . In the classical GA signalling pathway, GA regulates the gene expression of various pathways by interacting with the receptor GID1 and destabilizing the DELLA protein, which are master regulators of the GA signalling pathway 37,65 . DELLA proteins have conserved eponymous DELLA motifs, followed by LEQLE, TVHYNP and MAMGM motifs in the N-terminus 40 . The GRAS domain is in the C-terminus and connects to these motifs by a poly S/T/V motif, which consists of five subdomains, including LHRI, LHRII, VHIID, PFYRE and SAW 66 . All the motifs and subdomains, except LHRI, were associated with the affinity of DELLA to GA preceptor GID1 40 . The motifs in the N-terminus function in the transactivation of plant growth regulators; 67 the motifs in the C-terminus are essential for the formation of the transactivation complex and sequestration of DELLA-interacting proteins; specifically, the SAW motif plays a key role in the suppressive activity of DELLA 68,69 . DELLA functions as a master regulator in the trade-off between plant growth and various adverse environmental conditions. AfSLR1-like and AcSLR1-like proteins have unique LEQLD, MAMAMG and VVHYNP motifs in the N-terminus and AAAAEVEEE-GEEAAEE insertions in the GRAS domain, which are different from previously reported conserved motifs in DELLA proteins, suggesting that AfSLR1-like and AfSLR1-like may regulate different growth suppressors and DELLA-interacting proteins and play a pivotal role during adaptation to specific environments in bromeliads. In A. comosus var. comosus and A. fasciata, there were 5 and 2 DGLLA genes, respectively, which were DELLArelated proteins with DGLLA motifs instead of DELLA motifs 69 . DGLLA homologues function as back-up GA-insensitive negative regulators of plant growth under specific conditions 70 . The contraction of DGLLA homologues may promote the plant growth in A. fasciata. In conclusion, the duplication of GID1-like genes and contraction of DGLLA-like genes may play key roles in the formation of tank habits. Transposase-like genes of the HAT (hobo, activator, Tam3) family are essential for plant growth in rice and Arabidopsis 71 . Ricesleeper-overexpressing Arabidopsis plants showed delayed inflorescence development, increased rosette leaf number and irregularity, dwarfism and fasciation. Ricesleeperoverexpressing rice plants were also dwarfed 34 . Compared with A. comosus var. comosus and A. comosus var. bracteatus, A. fasciata contained 8 original Ricesleeper2 loci, 2 of which had expanded. The expansion of ricesleeper2-like may also be associated with tank epiphytic habit formation in A. fasciata. In addition, Key innovations could significantly decrease the extinction rate by improving the dispersion ability, helping the plants to invade new regions or increasing defences against herbivores 9 . TPS10 is involved in the indirect defence against lepidopteran larvae by producing a mixture that attracts natural enemies to plants being fed upon 35 . TPS10 is also induced by oomycete infection in Medicago truncatula 36 . The expansion of AfTPS10-like gene clusters improved the defence against herbivores and pathogens and may play a critical role in adaptation to environments. Tank-less bromeliads often inhabit semiarid or arid regions, where they lie at the limit of physiological tolerance, and most other plants fail to survive. The expansion of AcMAIL3like may play a key role in the development of pineapple roots. The expansion of ZIP10-like may help pineapple adapt to infertile environments. FT genes are highly conserved in sequence and function in angiosperms. FT genes regulate plant growth and multiple developmental processes, including flowering, expansion of storage organs, root development, flower development, stomatal movement and so on. In previous studies, AFTL1 was cloned and promoted flowering in Arabidopsis. As homologues of HD3A, FTL1 was expressed in flowers in both pineapple and A. fasciata and may play a role in flower development. FTL3 was induced by ethylene in pineapple but not in A. fasciata. FTL3 is a homologue of SbFT9, is expressed in leaves and may play a role in leaf development. AfFTL6 was expressed in the roots of A. fasciata, indicating that FTL6 may be involved in root development. Of the FTLs, only FTL2 was induced by ethylene and had the same amino acids in segment B as ZCN8 and SbFT8, indicating that FTL2 regulates flowering in pineapple and A. fasciata. Plants need to flower under optimal conditions to ensure successful reproduction. Therefore, plants have evolved a complex mechanism to integrate endogenous and environmental cues to flower at optimal times. Pineapple flowers naturally under cool night temperatures and short days 72 , as does A. fasciata. It is thought that pineapple flowering is induced by a burst of endogenous ethylene controlled by environmental cues 73 . The ethylene synthase inhibitor amino vinylglycine could delay the flowering of pineapple 74 . Silencing AcAcs2, a critical ethylene biosynthesis gene, also delayed pineapple flowering 73 . In A. fasciata, AfEIL1-like could induce the expression of AfFTL2 directly, indicating that there is an EIL1-dependent flowering pathway in pineapple and A. fasciata. Moreover, the homologues of CAL-A, LSH6, ERF3, NAKR3, etc. were induced by ethylene, similar to pineapple, indicating that there is a complex and conserved mechanism regulating flowering in bromeliads. DELLA functions as a flowering repressor and plays essential roles in ethylene delayed flowering by reducing bioactive GA levels and increasing DELLA accumulation 18 . In pineapple, ethylene treatment also reduced the bioactive GA level 75 . Ethylene-induced flowering was independent of DELLA-mediated flowering repression, which may be due to the unique motifs of AfSLR1-like and AcSLR1-like. Further research on the mechanism by which ethylene induces flowering is needed to facilitate plantation management and culture of new varieties.
In summary, we reported the genome sequences of A. fasciata, a popular ornamental tropical plant, using Nanopore, PacBio, Illumina and Hi-C sequencing to reach a high level of completeness and accuracy of the genome. The genome sequences of A. fasciata will facilitate further research on bromeliad evolution and the genomic basis of tank habit formation of tank bromeliads and ethylene-induced flowering. Hi-C libraries were created from tender leaves of A. fasciata at BioMarker Technologies Company as described previously. Briefly, the leaves were fixed with formaldehyde and lysed, and then the cross-linked DNA was digested with HindIII overnight. Sticky ends were biotinylated and proximity-ligated to form chimeric junctions that were enriched and then physically sheared to a size of 500-700 bp. Chimeric fragments representing the original cross-linked long-distance physical interactions were then processed into paired-end sequencing libraries, and 1,001 million 150-bp paired-end reads were produced on the Illumina HiSeq X Ten platform.

Methods
For RNA-seq, total RNA was extracted from central leaves, roots, flowers and central leaves under ethylene treatment for 24 h. After removing genomic DNA using DNase I (Takara), mRNAs were obtained using oligo (dT) beads and broken into short fragments, followed by cDNA synthesis. Paired-end sequencing was conducted on the HiSeq X Ten platform (Illumina, CA, USA).
Genome assembly and annotation. Using GenomeScope 76 for k-mer analysis, the genome size of A. fasciata was estimated. PacBio reads were first self-corrected with the parameter corOutCoverage = 100. The corrected reads, along with NanoPore long reads, were imported for assembly by Nextdenovo v2.3.1 (https://github.com/ Nextomics/NextDenovo) and NextPolish (https://github.com/Nextomics/ NextPolish) using the default parameters. Then, the redundant sequences of the polished contig sequences were eliminated by Purge Haplotigs 77 . Chromosomal assembly was performed based on proximity-guided assembly using ALLHIC 78 .
A de novo repeat library of the genome was customized using RepeatModeler 79 , which can automatically execute two de novo repeat finding programs, RECON (version 1.08) and RepeatScout (version 1.0.5). Consensus transposable element (TE) sequences generated above were imported to RepeatMasker (version 4.05) to identify and cluster repetitive elements. Unknown TEs were further classified using TEclass (version 2.1.3). To identify tandem repeats within the genome, the Tandem Repeat Finder (TRF) package (version 4.07) was used with the modified parameters of '1 1 2 80 5 200 2,000 -d -h' to find high-order repeats.
To further investigate LTRs, we applied the LTR_retriever pipeline 27 , which can integrate results from public programs such as LTR_FINDER and LTRharvest and efficiently remove false positives from the initial predictions. The predicted LTRs were further classified into intact and nonintact LTRs, and the insertion time was estimated as T = K/2μ (K is the divergence rate, and μ is the neutral mutation rate; the default is 1.38 × 10 − 8 in LTR_retriever) using the scripts implemented in the LTR_retriever package. Genome completeness was assessed with BUSCO v5.2.2 26 .
Gene annotations were processed with MAKER2 25 . Two rounds of MAKER2 processing were used to achieve high-quality gene annotation. The RNA-seq reads were imported to Trinity, genome-guided and de novo assembled with default parameters. Then, the combined reads were imported to the PASA pipeline 80 . The PASA assembled transcripts were used to train the ab initio predictors SNAP 81 , GENEMARK 82 and AUGUSTUS 83 . After that, MAKER2 processed the first gene annotation. After filtering the values of predicted proteins by MAKER2, the ab initio predictors were trained again. BRAKER 84 were used for predicting genes using aligned RNA-seq reads as input. Then, using the transcripts processed by Stringtie 85 as input and gene models predicted by BRAKER, the MAKER2 pipeline was run again. Using the GenblastA 86 pipeline and Genewise 87 , we also predicted pseudogenes. PlantTFDB 88 was used for transcription factor prediction. The A. comosus var. bracteatus genomic data were downloaded from EBI-ENA (PRJEB33121) 89 . The A. comosus var. comosus genomic data were downloaded from NCBI (ASM154086v1) 90 .
Genome structure and evolution. Orthofinder 28 was used for the identification of putative paralogous and orthologous genes from A. fasciata and other species. Because the coding genes have one more transcript, we retained the longest transcripts. Based on homology and collinearity, we investigated the paleohistory of A. fasciata, A. comosus var. comosus and A. comosus var. bracteatus. After alignment of coding genes using BLAST, we identified putative protogenes (pPGs). Then, the pPGs, ordered by location on the chromosome, were imported to MGRA2 91 . After analysis of gene gain/loss, the ancestral genome was reconstructed with an exhaustive set of ordered protogenes (oPGs). Then, the oPGs were used for collinearity analysis by MCscanX 92 .
ChIP-seq and ChIP-qPCR. Chromatin immunoprecipitation assays were performed as previously described 93 . Briefly, 3 g of sample was washed twice in cold PBS buffer, and proteins were cross-linked to DNA by incubating the samples with formaldehyde at a final concentration of 1%. Afterwards, samples were lysed, and chromatin was obtained on ice. Chromatins were sonicated to obtain soluble sheared chromatin (average DNA length of 200-500 bp). One part of the soluble chromatin was stored at -20°C for input DNA, and the remainder was used for immunoprecipitation by the antibodies AfEIL1-like and normal rabbit IgG (CST, 2729). Immunoprecipitated DNA was amplified by PCR using specific primers. All primers are listed in Supplementary Table 6.
To construct ChIP-Seq libraries, the resulting ChIP DNA above was used to generate a sequencing library according to the Illumina ChIP-Seq manufacturer's instructions. Then, an Illumina Genome Analyser II (Illumina, San Diego, CA) was used for sequencing according to the manufacturer's instructions.
Y1H assay. The yeast one-hybrid assay was performed as previously described 45 . The AfEIL1-like sequences were inserted into the pGADT7 vector (Clontech, USA) to construct pGANDT7-AfEIL1-like vectors. For Y1H cDNA library screening, the 200-bp promoter fragment AfFTL2 was cloned into the destination vector pGADT7. The growth of the transformants on SD-His-Ura-Trp medium was considered an indicator of AfEIL1-like binding to the corresponding DNA fragments.
Transgenic plants. The coding sequence of AfFTL2 was cloned into the binary vector Cam35S under the control of the CaMV35S promoter. The confirmed construct was transformed into A. fasciata using agrobacterium strain GV3101 with the leaf base dip method. Transgenic plants were verified by PCR to detect exogenous genes and qRT-PCR to detect the expression of AfFTL2.
Statistics and reproducibility. Statistical analyses were performed using the s software R (version 3.6.3). The significance level was typically set to 0.05 for all the statistical analyses. The RNA-seq, ChIP-seq and ChIP-qPCR experiments were performed with three biological replicates and the merged tissues of three plants were sampled for each treatment.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The Aechmea fasciata genome sequences and raw sequence data from RNA-seq and genome sequencing have been deposited under BioProject accession number PRJNA748557 94 . The AfEIL1-like ChIP-seq data were deposited under GEO accession number GSE205799 95 .