AmarylOmicBase: A unified transcriptome database to accelerate gene discovery in Amaryllidoideae species

The Amaryllidoideae subfamily (Amaryllidaceae) has great medicinal importance due to the production of structurally diverse alkaloids with potent pharmaceutical activities. These specialized metabolites include anti-cholinesterase (galanthamine [1] ), antiviral (lycorine, haemanthamine, and pancracine, among others [2] ), and antitumor (lycorine, narciclasine, and haemantamine, among others [3] ) properties. Understanding the biosynthesis and regulation of Amaryllidaceae alkaloids is therefore of significant biological and biotechnological interest, particularly for metabolic engineering and sustainable production in heterologous systems. However, progress in elucidating their pathways has been hindered by the absence of reference genomes for Amaryllidoideae species. As a result, transcriptome mining has served as a powerful foundation enabling the identification and functional characterization of numerous biosynthetic enzymes and pathway reconstitution efforts. For instance, the first transcriptome assemblies of this subfamily led to the identification and characterization of norbelladine 4’- O-methyltransferase (N4OMT), noroxomaritidine/norcraugsodine reductase (NR) and the phenol-coupling cytochrome P450 96T (CYP96T) in Narcissus aff . pseudonarcissus , Galanthus elwesii and Galanthus sp. [4-6] . Later, the assembly of N. pseudonarcissus King Alfred’s transcriptome led to the identification and characterization of norbelladine synthase (NBS, Singh and Desgagné-Penix [7] ). Transcriptome mining of Lycoris aurea [8] a n d Leucojum aestivum [9] allowed the characterization of 2 cytochrome P450s in the phenylpropanoid pathway, namely cinnamate 4-hydroxylase (C4H or CYP73A) and p -coumaroyl 3Vi9@-hydroxylase (C3’H or CYP98A) [10,11] . Finally, two recent studies successfully reconstituted the Amaryllidaceae alkaloid network in Nicotiana benthamiana through transcriptome mining of L. aestivum [12] and Narcissus sp. Tête-à-Tête [13] further demonstrating the power of transcriptomic data for pathway elucidation. Despite these advances, transcriptome-based discovery remains technically challenging. Datasets are dispersed across repositories, generated using different sequencing platforms and assembly strategies, and vary in annotation quality and expression quantification. This .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint fragmentation creates barriers to reproducibility, comparative analysis, and systematic exploration of enzyme diversity — particularly relevant given that the alkaloid profiles differ markedly among Amaryllidoideae species [14] . To address these challenges, we combined the transcriptomic datasets from 39 studies, spanning 13 genera and 29 species of Amaryllidoideae, into a searchable transcriptome resource. AmarylOmicBase includes assemblies, functional annotations and expression quantification. As a proof of concept, we mined this database to identify candidate enzymes catalyzing 20 reactions in the Amaryllidaceae alkaloid biosynthetic pathway. Res ults and Discu ssion Ama ryllidoideae tr ansc riptom e as se mblies Genome analysis is a crucial tool for identifying genes involved in the production of specialized metabolites. Although Amaryllidoideae present ecological, medical, and ornamental importance, the size and complexity of their nuclear genomes are deterrents to genome sequencing and assembly initiatives [15] . Hence, biosynthetic pathway elucidation in this sub- family has relied on transcriptome assembly and expression analysis. Here, we compiled a database of transcriptome assemblies (previously published or de novo assembled in this study), transcriptome annotation, and expression quantification for 29 Amaryllidoideae species. Most assemblies constructed using Trinity (the current ones, previous ones [9,16,17] , and those from Wang, et al. [18] ) generated a high number of gene and transcripts ( Figure 1 ), with the biggest assemblies (in terms of number of transcripts) being that of Lycoris radiata , with 1,461,347 trinity “genes” and 2,003,744 trinity “transcripts”. The smallest assemblies were those constructed only with long-read sequencing data ( Lycoris aurea (PB) and Narcissus Tête-à-Tête), followed by those from the One Thousand Plant Transcriptomes Initiative [19], One Thousand Plant Transcriptomes Initiative [20] (“1kP”, which reported a single isoform for each gene). Short-read assemblies had lower proportions of “genes” predicted to encode proteins ( Supplem entar y Fi gure S1 ). While 91.7% of L. aurea (PB) genes contained .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint coding sequences, this number dropped to 51.9% for Galanthus elwesii and 32.4% for Narcissus papyraceus . The assembly with the lowest proportion of protein-encoding genes was Crinum × powellii, with only 10.3%. Ama ryllidoideae tr ansc riptom es a ss embl y quality Assembly completeness evaluated with BUSCOs from the liliopsida_odb10 dataset shows that the most complete assembly (single + duplicated) was Clivia miniata , followed by others constructed with Trinity ( Figure 2 ) . The assemblies of L. aurea showed different proportions of complete BUSCOs between them, with the assembly combining both short- and long-reads being more complete. This can be explained by the addition of other replicates that were not sequenced with long-reads by Liu, et al. [21] , as well as samples from different studies. Conversely, the assemblies obtained from the 1kP dataset exhibited the highest proportions of missing BUSCOs, with three species each lacking over 60% of the expected BUSCOs: Traubia modesta (74.3%), Zephyranthes treatiae (73.7%), and Amaryllis belladonna (67.2%). To verify whether this issue was due to the assembly method used, we reassembled these datasets using the pipeline developed in this study. The improvements were relatively modest, with reductions in missing BUSCOs ranging from just 0.7% in A. belladonna to 8.3% in Phycella aff. cyrtanthoides and Z. treatiae (Supplementary Fig. S2). We also evaluated completeness using DOGMA with the monocots reference set (Supplementary Fig. S3). All assemblies generated in this study showed a completeness score greater than 60%. The assembly of the L. aurea transcriptome using only long reads had the lowest score in this group (60.3%). Meanwhile, similar to BUSCO’s result, C. miniata’s was the most complete assembly, with a score of 71.4%. By contrast, the assembly of T. modesta ’s transcriptome showed the lowest overall completeness score (18.6%). The low completeness of the transcriptomes from the 1kP dataset may be due to the effort required to obtain transcriptomes of many species simultaneously, which resulted in limiting sampling depth or tissues representation. However, this problem is not observed in other datasets that also have a single replicate, such as those published by Wang, et al. [18] . .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Another estimation of assembly quality is the percentage of the reads that map back to the assembly. Here, we used the pseudoaligner kallisto for expression estimation ( Figure 3 ). When grouped, the assemblies from the 1kP dataset had the highest proportion of mappable reads. This suggests that the incompleteness of their transcriptomes may be due to low sequencing depth, rather than issues with the assembly itself. Individually, the assembly with the highest proportion of mappable reads was that of N. papyraceus [17] . On the other hand, Galanthus sp. assembly [5] showed the lowest proportion, with only 39.7% ± 5.2% reads mapping back to the transcriptome. The low completeness detected with both DOGMA and BUSCO for the two assemblies of Galanthus species ( Figure 2 ), along with the low mapping rate of their raw reads, indicates that either these datasets should be reanalyzed or may require resequencing for reliable pathway mining. Functional annotation In terms of functional annotation, the two assemblies generated with long reads showed the highest proportion of genes annotated using any of the protein-based databases ( Figure 4 ). The most annotated assembly was that of L. aurea generated solely with long-reads ( L. aurea (PB), 90.7% of the genes annotated), followed by Narcissus sp. ‘Tête-à-Tête’ assembly with 80.7% of its genes annotated [13] . Nevertheless, when considering only genes with a predicted coding sequence, all assemblies had most of their genes annotated ( Supplementa ry Fig. S 4 ). For reference, assembly of the Hippeastrum vittatum transcriptome had the lowest proportion of protein-coding genes annotated, at 79.3%. Assemblies were also annotated with Infernal for a homology-based search against the Rfam database ( Supple menta ry Fig. S5 ), which includes non-coding and cis-regulatory RNA. The proportion of genes annotated with Infernal in each assembly was low, ranging between 0.06% for Zephyranthes candida and 1% for L. aurea (PB). Proof of co ncept: mining the Am aryll idaceae alkaloid biosynt hetic pathw ay Amaryllidaceae alkaloid precursors pathway enzymes .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Given the importance of Amaryllidaceae alkaloid, many efforts were done to elucidate the biosynthetic pathway in different species [4-7,11-13,22,23] . To demonstrate the utility of AmarylOmicBase for pathway discovery, we queried the database for the enzymes known to be involved in the biosynthesis of these alkaloids. Upstream of norbelladine, seven enzymes were searched ( Figure 5 , BLAST output presented in Supple menta ry Table S1 ), but none were found in all 29 species ( Supplementa ry Tab le S2 ). The products of tyrosine decarboxylase (TYDC) and phenylalanine ammonia-lyase (PAL) are precursors of all Amaryllidaceae alkaloids; however, the transcripts of these two enzymes were not found in several species. In the case of TYDC , it was missing from four of the six transcriptomes from 1kP. In particular, no match was found in Amaryllis belladonna for either TYDC2 from Arabidopsis thaliana or TYDC from Oryza sativa (Uniprot: TYDC2_ARATH and TYDC_ORYSJ, respectively). In the remaining three species, matches were detected, but were not considered due to insufficient coverage of the bait sequences. As for PAL , eight species lacked this enzyme. Seven of them had matches close to the thresholds set in this study. However, no match to either PAL1 or PAL2 from Narcissus aff. pseudonarcissus was found in Galanthus sp. Regarding C4H, both Narcissus species lacking transcripts encoding for this enzyme ( N. tazetta and N. viridiflorus ) had matches with high sequence identity (up to 95%) that covered 79% of the baits ( Suppleme ntary Ta ble S2 ). In other species, some matches were also observed with sequence identity greater than 90%, but query coverage was only around 50%, or as low as 25% in Galanthus sp. 4-coumarate-CoA ligase ( 4CL) and hydroxycinnamoyltransferase ( HCT) were both absent in the transcriptome of A. belladonna ; additionally, 4CL had two matches in T. modesta with high sequence identity but query coverage just below the threshold of 80%. HCT had matches in Z. treatiae that did not pass the thresholds set here. C3’H was not found in A. belladonna , nor in Galanthus spp. and N. viridiflorus ; while there were matches with high query coverage, all had sequence identity below 40%; this was also the case of caffeoylshikimate esterase ( CSE), absent in A. belladonna , R. pratensis , T. modesta, and Z. treatiae. The lack of detection of transcripts encoding many crucial enzymes for amino acid metabolism, as well as phenylpropanoid and alkaloid biosynthesis, in the 1kP and the Galanthus spp. .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint assemblies is possibly due to the stringency of our criteria and issues with the transcriptomic data itself, given the incompleteness of these transcriptomes ( Figure 2 a n d Supple me ntary Fig. S3), rather than a genuine absence of these enzymes in these species. For instance, all species had transcripts annotated as TYDC and almost all, with the exception of Galanthus sp., had transcripts annotated as PAL . However, most of them have only partial open reading frames, with predicted peptide sequences small compared to the characterized enzymes used as baits (see annotation reports, 10.5281/zenodo.17307476). Amaryllidaceae alkaloid pathway enzymes The first committed step of the biosynthesis of these alkaloids, the synthesis of norbelladine, is catalyzed by NBS and NR [23] . While the latter was found in all 29 species, no match was found for NBS i n N. viridiflorus ’ transcriptome ( Figure 5 , Supple m entar y Table S3 ). Additionally, NBS was not found in the long-read assembly of L. aurea , though it was present in the hybrid assembly of this species. The only species lacking N4 OMT was R. pratensis . In this species, the closest match to this enzyme shared only 43% sequence identity with the baits ( Supplem entar y Table S4). CYP96T, a key enzyme in the formation of most of these alkaloids, was not detected in 7 species: A. belladonna, P. aff. cyrthantoides, R. pratensis , and Z. treatiae [20] ; Galanthus sp. [5] ; H. striatum and Z. carinata [18] , which could be due to the incompleteness of these transcriptomes . In the galanthamine biosynthesis pathway, Tocopherol N-methyltransferase ( TocoNMT ) was not detected only in Galanthus sp., while the transcript of the aldo-keto reductase 1 ( AKR1 ) was absent in A. belladonna, Galanthus sp., and N. viridiflorus , all of which had matches falling below the coverage threshold ( Supplem entary Table S 3 ). In contrast, the transcript for the recently characterized coclaurine N-methyltransferase-like (CNMT-like) enzyme, responsible for converting norgalanthamine into galanthamine [22] , was undetected in ten species ( Supple menta ry Table S4 ). Among these, G. elwesii and H. vittatum had each one match with high sequence identity but only around 70% coverage; other seven species ( Hippeastrum sp., Lycoris chinensis , L. radiata , N. tazetta , T. modesta , Z. candida and Z. treatiae ) showed low- .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint quality matches; finally, Galanthus sp. had no matches. The lack of detection of CNMT-like transcripts in L. radiata and Hippeastrum sp. underscores the importance of careful evaluation of in silico data. Although CNMT-like was previously identified in these assemblies and cloned from L. radiata [22] , the predicted peptide sequences in both assemblies are partial: Hisp_TRINITY_DN525241_c0_g1_i1.p1, from Hippeastrum sp., is a predicted peptide of 125 amino acids that lacks both start and stop codons; Lyrad_TRINITY_DN1952052_c0_g1_i1.p1, from L. radiata , has 104 amino acids and lacks a stop codon (see annotation reports, 10.5281/zenodo.17307476). Due to the thresholds applied in this study, such fragmented sequences were excluded; however, their annotation enabled the characterization of another enzyme in the galanthamine biosynthesis pathway [22] . The only enzyme-encoding transcript found ubiquitously (downstream of NR) was the short- chain dehydrogenase/reductase ( SDR1 ), which may produce norpluviine. On the other hand, the transcript encoding for SDR2 and the one for a 2-oxoglutarate-dependent dioxygenase ( ODD), in the haemanthamine biosynthesis pathway [13] , were absent from Z. treatiae and T. modesta , respectively. However, matches to these enzymes can be found in these species with less stringent criteria of query coverage. Meanwhile, CYP71DW1, which catalyzes the formation of a methylenedioxy bridge in Narcissus sp. Tête-à-Tête pathway [13] , was not detected in either T. modesta or Z. treatiae . Finally, the transcript encoding the OMT, which may catalyze the last step in haemanthamine synthesis, was not found in six species of this database, namely A. belladonna , Galanthus sp., Hippeastrum sp., R. pratensis , T. modesta, and Z. treatiae. Conservative thresholds: “missing” versus reliable matches The apparent absence of transcripts in some species’ assemblies must be interpreted cautiously. Missing or fragmented transcripts may result from low-quality raw data, incomplete assemblies, or limited sampling of developmental stages, tissues, or environmental conditions. This is evident in species such as A. belladonna , Galanthus sp., G. elwesii , N. viridiflorus , P. aff. cyrthantoides , R. pratensis , T. modesta, and Z. treatiae , which showed low BUSCO and DOGMA scores. .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint To ensure consistency and comparability across species, BLAST results were filtered to keep only matches with ≥50% sequence identity and ≥80% query coverage, thresholds previously used to identify genes involved in benzylisoquinoline alkaloid biosynthesis [24] . These criteria directly influence which transcripts are considered present. For example, C4H matches in N. tazetta and N. viridiflorus were excluded due to 79% query coverage, just below our threshold. Similarly, transcripts with homology to TYDC were excluded in several species, though they may still be biologically relevant. As shown by Liyanage et al. (2025b), such sequences can be valuable for cloning and functional characterization. Relaxing the coverage threshold to ≥70% led to the inclusion of matches to PAL in five additional species, but did not affect the number of species with transcripts encoding TYDC, HCT, C3’H, CSE, NBS, N4 OMT, TocoNMT, C NMT, SDR2, and CYP71DW1. Although lowering the threshold can increase the number of apparent matches, it also raises the likelihood of false positives, particularly within large enzyme families such as cytochrome P450s and methyltransferases. Due to the variability in transcriptome completeness across species, a conservative threshold was applied to prioritize specificity over sensitivity. This approach reduces the chance of spurious hits from fragmented assemblies and improves confidence in the homologs identified.

The Amaryllidoideae subfamily is a valuable source of bioactive metabolites, yet the absence of

genomes and the fragmented nature of available transcriptomic resources have limited systematic investigation of their biosynthesis. AmarylOmicBase addresses this challenge by consolidating 39 studies spanning 29 Amaryllidoideae species into a standardized, searchable resource with assemblies, annotations, and expression quantification matrices, accompanied by reproducible scripts. By harmonizing disparate datasets, it enables rapid fishing of pathway candidates, cross-species comparison of isoenzymes, and targeted validation. For example, our proof-of-concept mining recovered transcripts encoding for enzymes that catalyze 20 reactions in the Amaryllidaceae alkaloid pathway. Using conservative .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint thresholds to filter blast results, we identified transcripts with homology to all known enzymes involved in the biosynthesis of these alkaloids in twelve species included in the database (both Crinum spp., Clivia miniata , L. aestivum , Lycoris aurea , L. incarnata , L. longituba , L. sprengeri , N. papyraceus , N. aff. pseudonarcissus , N. pseudonarcissus , and Narcissus cv. Tête-à-Tête). The database also exposes gaps due to sampling, sequencing depth, or assembly choices, guiding where new experiments or resequencing will have the highest payoff. Beyond alkaloids, the resource supports broader questions in specialized metabolism, from precursor supply to tailoring reactions and regulatory nodes. AmarylOmicBase is openly available (Zenodo 10.5281/zenodo.17307476; GitHub KarenGoncalves/AmarylOmicBase) and designed for iterative updates, making it a practical starting point for both discovery and benchmarking across the subfamily. Metho ds Selecte d datas ets an d ass e mbly pipe line The database was constructed by downloading either raw read files (fastq) or assembly files (fasta) that were publicly available, following the pipeline summarized in Figure 6 . When multiple datasets were available for the same species, all raw data were joined for the assembly ( Supple menta ry Table S5 ). Raw data were downloaded from the Sequencing Read Archive (SRA) using sra-toolkit 3.0.9, then quality control was performed using fastp 0.23.4 [25] (mean quality = 20; unqualified percent limit = 30; cut-front window-size = 5; cut-right window-size = 4; cut-right mean quality = 15; length required = 50). Samples from the bioprojects PRJNA306273, PRJNA306697, and PRJNA301357 [4,5] were sequenced with paired-end reads of 50 bp. Thus, the minimum length after trimming was set to 30 bp. The Lycoris radiata assembly from Park, et al. [26] (SRR8799443) was not included in the database because no raw data were available. Assemblies were carried out with Trinity 2.14.0 [27] , default parameters. Trinity cannot process sample files containing both paired-end and single-end datasets. Therefore, single-end reads were concatenated with forward reads into a single file, and the script was modified .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint accordingly. Next, assemblies were simplified by removing redundant transcripts with CD-HIT- EST 4.8.1 ( –c 0.97, –l 200). In the case of Lycoris aurea , two long-read datasets were available (PRJNA1106235, PacBio Sequel sequencing by Liu, et al. [21] ), with samples from the bioprojects PRJNA637967 and PRJNA1065507. The two long-read datasets were obtained in their original BAM format from SRA and assembled using isoseq from smrtlink-sequel2 13.1.0.221970 (default settings). The short-read datasets were assembled using Trinity, with the full-length, non-chimeric reads from Iso-Seq in the –long_reads option. Both isoseq (PB) and Trinity-hybrid (TH) assemblies were then processed with CD-HIT-EST as described above. Annotation and read mapping, as described below, were performed for both assemblies. Expression quantification was performed, regardless of whether the datasets were assembled in this study or not, using kallisto 0.46.1 [28] , with the support scripts align_and_estimate_abundance.pl and abundance_estimates_to_matrix.pl from Trinity. For datasets assembled in this study, the assemblies were cleaned up by removing Trinity isoforms with a total read count = 0. This pipeline for the assembly and clean-up was applied to the 1kP datasets for quality control. Quality control The following steps were carried out for all species datasets. The assembly quality was assessed with BUSCO 5.7.0 [29] , using the liliopsida_odb10 dataset. Additionally, transcriptomes were analyzed with DOGMA 1.0.2 (DOmain-based General Measure for transcriptome and proteome quality Assessment, reference set for monocots), based on the analyses from RADIANT 1.1.5 with the database radiant_db_pfam36. Annotation pip eline Next, the Trinotate 4.0.0 pipeline (http://trinotate.github.io/) (Bryant et al., 2017) was used, with parameters for each software set following Trinotate suggestions. Coding sequences were .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint predicted with TransDecoder 5.5.0 (http://transdecoder.github.io/). This initial prediction of coding sequences was annotated with BLASTp (against Uniprot SwissProt database release 2024_04; blast+ 2.13 [30] , evalue=10-5, maximum number of targets=1), hmmscan (against Pfam-A database version 37.0; HMMER 3.2.2 [31,32] , default settings). Afterward, the outputs of these two analyses were used to improve coding sequence predictions with TransDecoder.Predict, using the option “ --single_best_only”. This final prediction was used for annotation with eggnog-mapper 2.1.12 [33] and prediction of signal peptides as well as transmembrane domains (using SignalP6 [34,35] and TmHMM 2.0c [36] , respectively), with default settings for all of them. The transcriptome assemblies were also analyzed with Infernal 1.1.4 [37] for identification of sequences with secondary RNA structure, using Trinotate’s " --run 'infernal'" command. These results were compiled into a report using the Trinotate report. Identification of A maryllidacea e alka loid biosynthe sis candidat es In the search for genes involved in the biosynthesis of Amaryllidaceae alkaloids and their precursors, we created a blast database for the predicted proteins (species names were removed from protein IDs from 1kP assemblies for the creation of the database). Enzymes upstream of norbelladine synthase (NBS) were found in Uniprot SwissProt and used as baits for BLASTp. Given that several enzymes downstream of NBS have been recently characterized, their sequences are not currently available in UniProt. Thus, their mRNA sequences were used as baits for BLASTx. Protein or mRNA sequences used as baits are provided in Supplementary Data S1 and S2, respectively. For BLASTp and BLASTx, the parameters were set to e-value = 10 -10 (database of 2 510 506 sequences, with predicted proteins of all assemblies), minimum sequence identity = 50%, and minimum query coverage = 80% [24,38] . Additionally, results were filtered to retain only the “best” protein sequence from each Trinity “gene”, based on sequence completeness, length, and prediction score. .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Acknowl edg em ents The authors would like to thank the research community working on molecular aspects of Amaryllidoideae for their datasets. This research was enabled in part by support provided by the Digital Research Alliance of Canada (alliancecan.ca). This research was funded by Canada Research Chair Tier 1 on plant specialized metabolism Award No CRC-2023-00353 to I.D-P. Thanks are extended to the Canadian taxpayers and the Canadian government for supporting the Canada Research Chairs Program. Author Cont ributions K. C. Gonçalves dos Santos: Conceptualization, data curation, formal analysis, investigation, methodology, Writing – original draft, Writing – review & editing; N. Merindol: Conceptualization, Validation, Writing – original draft, Writing – review & editing; I. Desgagné- Penix: Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing – review & editing. Data availability stateme nt All assemblies, predicted protein sequences, annotation results, expression quantification and Trinotate reports were published in Zenodo under the DOI: 10.5281/zenodo.17307476. Scripts for read quality control, assembly, annotation, and report are available on the GitHub repository KarenGoncalves/AmarylOmicBase. Conflict of interest The authors declare no conflicts of interest. .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Short leg end s for Suppo rting Info rm ation Supple mentar y Fi gure S1. Propo rtion o f pr otein-encodin g genes (in rel ation to the tot al nu mbe r of “genes”) in each transcripto me asse mbl y. On the rig ht, the asse mbly o rigin is indic ated: “Ass embled de novo”: data asse mbled in thi s study;“1kP”: One Thousand Pl ant Tr ansc ript o mes Ini tia tive [19], One Thousand Plant Trans crip to me s Initi ative [20] ; “Desgagné-P e nix tea m”: Tousignan t, et al. [9], Koi ra la, e t al. [16], Hot chandani, e t al. [17], Koira la, et a l. [39] '; “Kilgor e e t al., 2014, 2 016”:Kilgor e, e t al. [4], Kilg o re, et a l. [5] ; Meht a et al. 2024 Meht a, e t al. [13] ; “Wang et al. 2024”: W ang, et al. [18] . The Lycoris aurea (PB) as se mbly inc ludes only th e l ong-re ads published by Liu, et a l. [21] . This assembly wa s then used for T rinity in hybrid m ode, using all othe r sho rt- read da ta ava ilabl e fo r that spec ies. Supple mentar y Fi gu re S2. Comp ariso n of missin g BUSC O (Benchm arking Unive rs al Single-C opy Orth ol ogs, lili opsida _odb10 datab ase) of spec ies s equenced by the One Thousand Pl ant Trans c ri pto mes Init iat ive [19], One Thousand P lant Tr ansc ript o mes Initi ative [20] , betwe en orig inal ass embl ies (S OAP (1kP) ) an d Trini ty-ass embl ed da ta. Supple mentar y Fi gu re S3. DOGMA an al y sis result for transcripto me assem blies of diff erent species and gener ated b y se ver al grou ps. On th e r ight, t he ass embly o rigin i s indi cat ed: “Asse mbl ed de novo ”: da ta asse mbled in this s tudy;“1kP”: One Thousand P lant T ran scr ipt om es Init iat ive [19], One Thous and Plan t Trans crip to me s Initi ative [20] ; “Desgagné-Penix te am”: Tousign ant, et a l. [9], Koi ra la, et al. [16], Ho tchandani, e t a l. [17], K oi ral a, et al. [39] '; “Kilg o re et al., 2014, 2016”:Kilgor e, et al. [4], Kilg o re, e t al. [5] ; Mehta e t al. 2024 Mehta, e t al. [13] ; “Wang et al. 2024”: Wang, et al. [18] . The Lycoris aurea (PB) as se mbly i ncludes only the long- reads published by Li u, et al. [21] . This a sse mbly was then us ed f or Tr inity in hybrid m ode, usi ng all o ther sh or t-r ead da ta ava ilabl e f o r th at spec ies. Supple mentar y F igu re 4. Propo rtion of p rotein-codin g ge nes annot ated with prote in-b ased data bases (EggNOG, Pfam and Unip ro t S wiss Pr ot ). If mul tipl e peptides we re p redi cted f o r a gen e, th e an nota ti on of any of tho se peptide s equence s wa s c onside red. On the right, the a sse mbly or igin is indica ted: “As sembl ed de novo ”: d ata asse mbled in thi s study;“1kP”: One Thousand Pl ant Tr ansc ript o mes Ini tia tive [19], One Thousand Plant Trans crip to me s Initi ative [20] ; “Desgagné-P e nix tea m”: Tousignan t, et al. [9], Koi ra la, e t al. [16], Hot chandani, e t al. [17], Koira la, et a l. [39] '; “Kilgor e e t al., 2014, 2 016”:Kilgor e, e t al. [4], Kilg o re, et a l. [5] ; Meht a et al. 2024 Meht a, e t al. [13] ; “Wang et al. 2024”: W ang, et al. [18] . The Lycoris aurea (PB) as se mbly inc ludes only th e l ong-re ads published by Liu, et a l. [21] . This assembly wa s then used for T rinity in hybrid m ode, using all othe r sho rt- read da ta ava ilabl e fo r that spec ies. Supple mentar y Fi gure 5. Pro portion of gen es with at le ast one “tra nscript” an notate d with Infe rna l usin g the Rf a m d atab ase . On the right, the asse mbl y or igin is indic at ed: “Asse mbled de novo” : data as se mbled in this study;“1kP”: One Thous and Plan t Tr ansc ript om es Ini tia tive [19], One Th ousand P lant T ran scr ipt om es Ini tia tive [20] ; “Desgagné-Penix t ea m”: Tousignan t, et a l. [9], Koi ral a, et a l. [16], Hot chandani, e t al. [17], Koi r ala, e t al. [39] '; “Kilgor e et al., 2014, 2016”:Kilgore, e t al. [4], Kilgo re , et al . [5] ; Mehta et al. 2024 Mehta , et al. [13] ; “Wa ng et al. 2024”: Wang, et al. [18] . The Lycoris aurea (PB) as se mbly inclu des only the long- re ads published by Liu, et al. [21] . This assembly was then used fo r T rini ty in hyb rid mod e, us ing al l o the r sh or t- read dat a avail able f o r that spe cies. Supple mentar y Ta ble S1. Nu mbe r of bl as t hits of pr ot eins involv ed in A ma ryllida c eae alkal oid bio synthesi s upstre am of N orbe lladin e Synthas e in ea ch specie s in the Am aryl lido idea e pr ote o me databas e. Only hits w ith evalue= 50%, and query cove r age >= 80% we re c onsid er ed. Supple mentar y T abl e S2. BLASTp re sult o f pro teins involv ed in Am aryl lida cea e alka lo id biosynthes is upst rea m of Norb ell adine Synthas e ag ainst the Am ary llid oide ae p ro te o me d atab ase. Supple mentar y Ta ble S3. Nu mbe r of bl as t hits of pr ot eins involv ed in A ma ryllida c eae alkal oid bio synthesi s downst re am of No rbe lladine Synthas e in ea ch speci es in the Am aryl lido idea e pr ote o me databas e. Only hits wi th evalue= 50%, and query cove r age >= 80% we re c onsid er ed. .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Supple mentar y T ab le S4. BLASTx r esult of p ro teins involv ed in Ama ryllida ce ae a lkal oid biosynthe sis do wnst re am of N o rbell adine Synthas e agains t th e A ma ryll idoide ae p r ote o me d atab ase. Supple mentar y Ta ble S5. D atasets inc luded in the d atab ase, w ith accession nu mbers fo r ra w d ata (Bi oPr oje ct o r individual SRA run IDs , when the pr ojec t c onta ined da ta f o r mult iple spec ies) and, w hen avai lable , a cc ess ion numbers o r nam es of as se mbly fi les. Supple mentar y D ata S1. P r ot ein sequen ce s of b aits invo lved in Am ary llida ce ae a lkal oid pa thway upst re am of Norb ell adine synthase. Supple mentar y D ata S2. RNA sequen ces of genes involv ed in A ma ryllida ce ae a lkal oid pathway d ownst re am of Norb ell adine synthase. .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Figure le gen ds Fi gure 1. Nu mber of genes, gene isof or ms (transcripts), and predicted protei ns in each assembl y. On the r ight, the a sse mbly o rigin is indica ted: “1kP”: On e Thous and Pl ant Tr ansc ript o mes Initi ative [19], One Thous and Pl ant Trans crip to me s Initi ative [20] ; “Desgagné-P e nix tea m”: Tousignan t, et al. [9], Koi ra la, e t al. [16], Hot chandani, e t al. [17], Koira la, e t al. [39] '; “Kilgo re et al. 2014, 2016”: Kilgor e, e t al. [4], Kilg or e, e t al. [5] ; “Meh ta et al. 2024”: Meht a, e t al. [13] ; “Wang et al. 2024”: Wang, et al. [18] ; “Assembled de novo ”: da ta as se mbled in this s tu dy. The Lycoris aurea (PB) asse mbly includes only the l ong reads publis hed by Liu, e t a l. [21] . Fi gure 2. Assemb l y com pleteness e va luated with Bench m ark in g U ni versa l Sin gle- Cop y Ortholo gs (BUS CO), usin g the lilio psid a_odb10 d atab ase. On the ri ght, the as se mbly o rigin is indi ca ted: “1kP”: One Thousand Pl ant Trans crip to me s Initi ative [19], One Thousand P lant Tr ansc rip to mes Initi ative [20] ; “Desgagné-Penix te am”: T ousignan t, et al. [9], Ko ir ala, et al. [16], Ho tch andani, et al. [17], Koir ala , et al. [39] '; “Kilgo re e t al. 2014, 2016”: Kilgore, e t al. [4], Kilgor e, e t a l. [5] ; “Meht a e t a l. 2024”: M ehta, et al. [13] ; “Wang e t a l. 2024”: Wang, et a l. [18] ; “Assembl ed de novo”: da ta ass emb led in this s tudy. The Lycoris aurea (PB) as se mbly in cludes only the l o ng re ads publish ed by Liu, et al. [21] . Fi gure 3. Pseudo ali gn ment pe rcenta ge pe r assem bl y. Ba rs indica te me an pseudo ali gment pe r centag e, do ts indicat e of individual sa mple s. On th e r ig ht, the ass embly o rigin i s indic at ed: “1kP”: One Thous and Plan t Trans crip to me s Initi ative [19], One Thousand P lant Tr ansc rip to mes Initi ative [20] ; “Desgagné-Penix te am”: T ousignan t, et al. [9], Ko ir ala, et al. [16], Ho tch andani, et al. [17], Koir ala , et al. [39] '; “Kilgo re e t al. 2014, 2016”: Kilgore, e t al. [4], Kilgor e, e t a l. [5] ; “Meht a e t a l. 2024”: M ehta, et al. [13] ; “Wang e t a l. 2024”: Wang, et a l. [18] ; “Assembl ed de novo”: da ta ass emb led in this s tudy. The Lycoris aurea (PB) as se mbly in cludes only the l o ng re ads publish ed by Liu, et al. [21] . Fi gure 4. Propo rtion o f “ge nes” annot ated with pr otein-b ased d atab ases (Egg NO G, P fa m a nd Unip rot SwissP rot). If multipl e peptides w er e predi ct ed fo r a ge ne, annot ati on of any of tho se peptide s equences w as c onsid er ed. On the right , the as se mbly o rigin is indica ted: “ 1kP”: One Thousand Plant T rans c ript om es I nitia tive [19], One Thousand Plant Tr ansc ript o mes Ini tia tive [20] ; “Desgagné-Penix te am”: T ousignant, e t al. [9], Koi ra la, et al. [16], Hot chandani, e t al. [17], Koi ral a, et al. [39] '; “Kilgor e et al. 2014, 2016”: Kilgore , et al. [4], Kilgo re , et al. [5] ; “Me hta et al. 2024”: Mehta , et al. [13] ; “Wang et al. 2024”: Wang, et a l. [18] ; “Assembl ed de novo”: data a sse mbled in thi s study. The Lycoris aurea (PB) a sse mbly in cludes only the l ong r eads p ublished by Liu, et al. [21] . Fi gure 5. Enzy mes in the A mar y llid aceae a l kalo id b iosy nthetic p athw ay detected in di f ferent species by m inin g the Am ar ylO m icBase. Cha ra ct er ized seque nces of ea ch enzym e (sh own in read) we re s ea rched agains t the predic ted pr ot e om es o f 29 Ama ryllid oide ae speci es. Numb er in pa ren thesi s indic ate t he number of spe cie s in which the enzy me w as f ound. PAL: phe nylalanine a m mon ia-lya se; C4 H: cinna m at e 4-hydroxyla se; 4C L: 4- cou ma rat e-C oA l igase ; H CT: Hyd roxy ci nnamoyl tr ansfe ra se; C3’H: p - cou ma roy l 3(iCw-hydroxylas e; CSE: caffe oylsh ikim ate es te ras e; TYDC: t yros ine de ca rboxyl ase ; NBS: n o rbell adine syntha se; NR: nor cr augsodin e/n or ox om ar itidine r educt ase ; N4 O MT: no rbel ladine 4’- O - me thyltr ansfe r ase; CYP96T: cyto chr o me P450 96T; Toco N MT: t o cophe r ol N -m eth yltr ansfe ras e; AKR1: aldo-ke to r educt ase 1; CN MT: c ocl aurin e N - methyl tr ansfe ra se-lik e enzym e; SDR1/2: sh or t-chain al coh ol dehydr ogen ase /r educt ase 1/ 2; CYP71DW: cyt och ro me P450 71DW; ODD: 2- oxoglut a rat e-depende nt dioxygen ase; O MT: O -m ethylt ran sfe ras e ; ?: unknown enzym es. Dashed a r ro w indi cat es unknown pa thway. Fi gure 6. A ma ry lO micBase construction p ip eline. When only on e tr ansc ript o mi c study exist ed fo r a spec ies and the tr ansc ript o me ass emb ly was avai lable , the asse mbly was down lo aded, and trans crip ts we re anno ta ted. If multipl e studies ex ist ed for a spe cie s o r no asse mbly was avail able, r aw re ads we re downl o aded fro m the Sequenc e Read A rchive (SRA). R eads w er e tri m med and fi lte red with f astp. R eads w ere then a sse mbled with Trinity and si mplifi ed with CD-HIT-EST. Expr essi on re ads we re pseudo aligned t o the as s embli es with kal list o. The resul t wa s used t o filt er our ass embli es fu rthe r, r em oving Tr inity is of or ms that w er e not exp ress ed. Both our .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint asse mbli es and th ose d ownl oad ed we re p ro ces sed and an alyzed wi th BUS CO and D OGMA fo r e sti ma ti on of co mple tenes s. Tr ansc ript s we re als o ana lyz ed with INFERNA L f or the pred ict ion of s ec ondary RNA s truc tur e. Additiona lly, an initi al coding sequenc e iden tific ati on w as p erf o rm ed with T ransd ec ode r. LongOR Fs and the resul t was used f or h om ol ogy-based annot ati on w ith BLASTp using UniP r ot S wiss Pr ot , and detec ti on of pr ot ein fa mily dom ains, h mms can us ing the Pf am-A da taba se. Resul ts w er e ent er ed int o Tr ansDec ode r. Predi ct, and the i mpr oved open r eading f ra me p redi cti ons w er e used to s ea rch signal pep tides and t rans me mbr a ne dom ains, with Signal P and TmH MM, r espec tive ly, and f or o rth ol o gous gr oup s ea rch w ith EggNOG e m apper. Annota ti on r esult s we re co mpil ed int o a rep or t us ing Tr inot at e. C re at ed in Lucid (luc id.c o). .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Refe renc es 1 Harvey, A. L. The pharmacology of galanthamine and its analogues. Pharmacology and Therapeutics 68, 113-128, doi:10.1016/0163-7258(95)02002-0 (1995). 2 Merindol, N. et al. Amaryllidaceae alkaloids screen unveils potent anticoronaviral compounds and associated structural determinants. ACS Pharmacology & Translational Science 7, 3527-3539, doi:10.1021/acsptsci.4c00424 (2024). 3 Martins de Paiva, M. J. et al. Antitumor evaluation of amaryllidaceae alkaloids on cancer cell lines: A literature review. . Electronic Journal of General Medicine 21, em562, doi:10.29333/ejgm/14040 (2024). 4 Kilgore, M. B. et al. Cloning and characterization of a norbelladine 4'/i1O/i1methyltransferase involved in the biosynthesis of the Alzheimer's drug galanthamine in Narcissus sp. aff. pseudonarcissus . PloS one 9, e103223, doi:10.1371/journal.pone.0103223 (2014). 5 Kilgore, M. B., Holland, C. K., Jez, J. M. & Kutchan, T. M. Identification of a noroxomaritidine reductase with Amaryllidaceae alkaloid biosynthesis related activities. Journal of biological chemistry 291, 16752, doi:10.1074/jbc.M116.717827 (2016). 6 Kilgore, M. B., Augustin, M. M., May, G. D., Crow, J. A. & Kutchan, T. M. CYP96T1 of Narcissus sp. aff. pseudonarcissus catalyzes formation of the Para-Para' C-C phenol couple in the Amaryllidaceae alkaloids. Frontiers in Plant Science 7, doi:10.3389/fpls.2016.00225 (2016). 7 Singh, A. & Desgagné-Penix, I. Transcriptome and metabolome profiling of Narcissus pseudonarcissus ‘King Alfred’ reveal components of Amaryllidaceae alkaloid metabolism. Scientific reports 7, 17356, doi:10.1038/s41598-017-17724-0 (2017). 8 Wang, R. et al. De novo sequence assembly and characterization of Lycoris aurea transcriptome using GS FLX titanium platform of 454 pyrosequencing. PloS one 8, e60449, doi:10.1371/journal.pone.0060449 (2013). 9 Tousignant, L. et al. Transcriptome analysis of Leucojum aestivum and identification of genes involved in norbelladine biosynthesis. Planta 255, 30, doi:10.1007/s00425-021- 03741-x (2022). 10 Li, Y. et al. De novo biosynthesis of p-coumaric acid in E. coli with a trans-cinnamic acid 4-hydroxylase from the amaryllidaceae plant Lycoris aurea . Molecules 23, doi:10.3390/molecules23123185 (2018). 11 Karimzadegan, V. et al. Characterization of cinnamate 4-hydroxylase (CYP73A) and p- coumaroyl 3'-hydroxylase (CYP98A) from Leucojum aestivum , a source of .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Amaryllidaceae alkaloids. Plant Physiology and Biochemistry 210, 108612, doi:10.1016/j.plaphy.2024.108612 (2024). 12 Lamichhane, B. et al. Elucidating the enzyme network driving Amaryllidaceae alkaloids biosynthesis in Leucojum aestivum . Plant Biotechnology Journal 23, 1988-2005, doi:10.1111/pbi.70026 (2025). 13 Mehta, N., Meng, Y., Zare, R., Kamenetsky-Goldstein, R. & Sattely, E. A developmental gradient reveals biosynthetic pathways to eukaryotic toxins in monocot geophytes. Cell 187, 5637, doi:10.1016/j.cell.2024.08.027 (2024). 14 Lin, G. D., Vishwakarma, P., Smith, P. N. & Li, R. W. The occurrence and bioactivities of Amaryllidaceae alkaloids from plants: A taxonomy-guided genera-wide review. Plants (Basel) 14, doi:10.3390/plants14131935 (2025). 15 Liyanage, N. S. et al. Navigating Amaryllidaceae alkaloids: bridging gaps and charting biosynthetic territories. Journal of Experimental Botany 76, 16-34, doi:10.1093/jxb/erae187 (2025). 16 Koirala, M. et al. Auxin and light-mediated regulation of growth, morphogenesis, and alkaloid biosynthesis in Crinum x powellii 'Album' callus. Phytochemistry 216, 113883, doi:10.1016/j.phytochem.2023.113883 (2023). 17 Hotchandani, T., de Villers, J. & Desgagne-Penix, I. Developmental regulation of the expression of amaryllidaceae alkaloid biosynthetic genes in Narcissus papyraceus. Genes 10, doi:10.3390/genes10080594 (2019). 18 Wang, X. X. et al. Phylotranscriptomics reveals the phylogeny of Asparagales and the evolution of allium flavor biosynthesis. Nature Communications 15, 9663, doi:10.1038/s41467-024-53943-6 (2024). 19 One Thousand Plant Transcriptomes Initiative. Updated data for "A Phylogenomic View of Evolutionary Complexity in Green Plants". GigaScience Database 10.5524/100910 (2021). 20 One Thousand Plant Transcriptomes Initiative. One thousand plant transcriptomes and the phylogenomics of green plants. Nature 574, 685, doi:10.1038/s41586-019-1693-2 (2019). 21 Liu, Z. et al. Functional characterization of CYP96T1-like cytochrome P450 from Lycoris aurea catalyzing para-para' and para-ortho' oxidative coupling in Amaryllidaceae alkaloids biosynthesis. Frontiers in Plant Science 15, 1438102, doi:10.3389/fpls.2024.1438102 (2024). .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint 22 Liyanage, N. S. et al. Coclaurine N-methyltransferase-like enzymes drive the final biosynthetic reaction of the anti-Alzheimer's drug galanthamine in Amaryllidaceae. Plant Physiol Biochem 226, 110067, doi:10.1016/j.plaphy.2025.110067 (2025). 23 Majhi, B. B., Gelinas, S. E., Merindol, N., Ricard, S. & Desgagne-Penix, I. Characterization of norbelladine synthase and noroxomaritidine/norcraugsodine reductase reveals a novel catalytic route for the biosynthesis of Amaryllidaceae alkaloids including the Alzheimer's drug galanthamine. Frontiers in Plant Sciences 14, 1231809, doi:10.3389/fpls.2023.1231809 (2023). 24 Pei, L. et al. Genome and transcriptome of Papaver somniferum Chinese landrace CHM indicates that massive genome expansion contributes to high benzylisoquinoline alkaloid biosynthesis. Horticulture research 8, 5, doi:10.1038/s41438-020-00435-5 (2021). 25 Chen, S., Zhou, Y., Chen, Y. & Gu, J. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 34, i890, doi:10.1093/bioinformatics/bty560 (2018). 26 Park, C. H. et al. Transcriptome analysis and metabolic profiling of Lycoris radiata . Biology 8, doi:10.3390/biology8030063 (2019). 27 Grabherr, M. G. et al. Full-length transcriptome assembly from RNA-Seq data without a

genome. Nature Biotechnology 29, 652, doi:10.1038/nbt.1883 (2011). 28 Bray, N. L., Pimentel, H., Melsted, P. & Pachter, L. Near-optimal probabilistic RNA-seq quantification. Nature Biotechnology 34, 525-527, doi:10.1038/nbt.3519 (2016). 29 Simao, F. A., Waterhouse, R. M., Ioannidis, P., Kriventseva, E. V. & Zdobnov, E. M. BUSCO: assessing genome assembly and annotation completeness with single-copy orthologs. Bioinformatics 31, 3212, doi:10.1093/bioinformatics/btv351 (2015). 30 Camacho, C. et al. BLAST+: architecture and applications. BMC Bioinformatics 10, 421, doi:10.1186/1471-2105-10-421 (2009). 31 Eddy, S. R. A probabilistic model of local sequence alignment that simplifies statistical significance estimation. PLoS Computational Biology 4, e1000069, doi:10.1371/journal.pcbi.1000069 (2008). 32 Eddy, S. R. Accelerated profile HMM searches. PLoS Computional Biology 7, e1002195, doi:10.1371/journal.pcbi.1002195 (2011). 33 Cantalapiedra, C. P., Hernandez-Plaza, A., Letunic, I., Bork, P. & Huerta-Cepas, J. eggNOG-mapper v2: Functional annotation, orthology assignments, and domain prediction at the metagenomic scale. Molecular Biology and Evolution 38, 5829, doi:10.1093/molbev/msab293 (2021). .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint 34 Nielsen, H. Predicting secretory proteins with SignalP. Methods in Molecular Biology 1611, 73, doi:10.1007/978-1-4939-7015-5_6 (2017). 35 Teufel, F. et al. SignalP 6.0 predicts all five types of signal peptides using protein language models. Nature Biotechnology 40, 1025, doi:10.1038/s41587-021-01156-3 (2022). 36 Krogh, A., Larsson, B., von Heijne, G. & Sonnhammer, E. L. Predicting transmembrane protein topology with a hidden Markov model: application to complete genomes. Journal of Molecular Biology 305, 580, doi:10.1006/jmbi.2000.4315 (2001). 37 Nawrocki, E. P., Kolbe, D. L. & Eddy, S. R. Infernal 1.0: inference of RNA alignments. Bioinformatics 25, 1337, doi:10.1093/bioinformatics/btp157 (2009). 38 Li, H. L. et al. Haplotype-resolved genome of diploid ginger (Zingiber officinale) and its unique gingerol biosynthetic pathway. Horticulture research 8, 189, doi:10.1038/s41438- 021-00627-7 (2021). 39 Koirala, M. et al. Crinum x powellii transcriptome assembly and predicted peptide sequences. figshare 10.6084/m9.figshare.22726235.v1 (2023). .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Figure s Fig. 1 .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Fig. 2 .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Fig. 3 .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Fig. 4 .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Fig. 5 .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint Fig. 6 .CC-BY-NC 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted November 27, 2025. ; https://doi.org/10.1101/2025.11.24.690262doi: bioRxiv preprint